Cristallografia PDF

Fundamentals of
Crystallography
C. GIACOVAZZO, H. L. MONACO, D. VITERBO

F. SCORDARI, G. GILLI, G. ZANOTTI, M. CATTl
Edited by
C. GIACOVAZZO
Dipartimento Geomineralogico, University of Bari, ltaly
and lstituto di Ricerca per lo Sviluppo delle Metodologie Cristallografiche,
CNR, Bari, ltaly
INTERNATIONAL UNION OF CRYSTALLOGRAPHY

OXFORD UNIVERSITY PRESS
OXFORD
UNIVERSITY PRESS
Great Clarendon Street, Oxford 0x2 6 D P
Oxford University Press is a department of the University of Oxford.
It furthers the University's objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New Yorlz
Athens Auclzland Bangkok Bogota Buenos Aires Calcutta
Cape Town Chennai Dares Salaam Delhi Florence Hong Kong Istanbul
Karachi KualaLumpur Madrid Melbourne MexicoCity Mumbai
Nairobi Paris S%oPaulo Shanghai Singapore Taipei Tokyo Toronto Warsaw
with associated companies in Berlin Ibadan
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
O C. Giacovazzo, H. L. Monaco, D. Viterbo,
F. Scordari, G. Gilli, G. Zanotti, M. Catti, 1992
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
Parts of this work first appeared as Introduzione alla cristallografia moderna by
M. Bolognesi, A. Coda, C. Giacovazzo, G. Gilli, F. Scordari, and D. Viterbo, edited by
C. Giacovazzo, and published by Edizioni Fratelli Laterza, O 1985
First published by Oxford University Press 1992
Reprinted 1994,1995,1998,2000
All rights resewed. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose this same condition on any acquirer
A catalogue record for this book is available from the British Library
Library of Congress Cataloging in Publication Data
Fundamentals of crystallography/C. Giacovazzo . . . [et al.]; edited by C. Giacovazzo.
(International Union of Crystallography texts on crystallography; 2). Includes bibliographical
references.
1. Crystallography. I. Giacovazzo, Carmelo. 11. Series. QD905.2.F86 1992 548-
dc2O 91-27271
ISBN 0 19 855578 4 (Pblz.)
Printed in Great Britain
on acid-free paper by
The Bath Press, Bath
Preface
Crystallography, the science concerned with the study of crystals, is a very

old subject. However, only in this century has it developed into a modern
science, after the discovery of X-rays and their diffraction by crystals. In
recent years crystallography as assumed an increasingly important role in
f
the modern sciences because o its interdisciplinary nature, which has acted
as a bridge between, and often as a stimulus for, various rapidly evolving
disciplines. Indeed, Chemistry, Physics, Earth Sciences, Biology, Mathema-
tics, and Materials Science have all provided stimuli to the development of
new crystallographic interests and techniques. In turn, crystallography has
significantly contributed to the advancement of these sciences. Thus, while
on the one hand crystallography has been enriched, on the other hand
writing a textbook describing all of its aspects has been made more difficult.
Recently, the demand for a compact book that gives a comprehensive
account of the modern crystallographic subjects has increased. This volume
should therefore be a useful and handy textbook for university courses that
cover crystallography, fully or only partially. It should also be useful at the
more advanced level required for doctorate studies as well as for ex-
perienced researchers.
It was with these ideas in mind that I first set out to co-ordinate the
in 1985, of a textbook in Italian (Introduzione alla cris-
tallograjia moderna, Edizioni Fratelli Laterza, Bari) of which Fundamentals
of crystallography is not only an English translation, but a completely
revised and updated version with a new chapter on crystal physics. It was
clear to me that (a) the book had to be written by several authors in order
to take advantage of their specific expertise; (b) the different chapters had
to be carefully harmonized in order to conform them to a unified plan.
It seems to me that these two requirements are even more valid today and
their achievement is entirely due to the creative co-operation of the
co-authors of this book.
Two of the co-authors of the Italian textbook, M. Bolognesi and A.
Coda, were unable to carry out the translation and revision of their
chapters. I wish to express my thanks for their valuable contribution to the
previous edition. In this book their topics are treated by G. Zanotti and
H. L. Monaco. An additional chapter on crystal physics has been written by
M. Catti. I thank the three new authors for entering our team and all the
authors for their enthusiastic participation in this project.
Bari, Italy C.G.

August 1991
Acknowledgements
We have the pleasant task of expressing our gratitude to Dr C. K. Prout and

to Dr D. J. Watkin for encouraging us to write this book and for their useful
advice. We have also to thank many friends and colleagues who have
commented on the manuscript or given technical support. The cover picture
was kindly provided by Professor A. Zecchina and Dr S. Bordiga, who used
the Chem-X software of Chemical Design Ltd, Oxford, UK. Thanks are
also due from C. Giacovazzo to Mr D. Trione for his accurate and rapid
typing of the manuscript, to Mr G. Pellegrino for his aid with the drawings
and to Ms C. Chiarella for her technical support; from H. L. Monaco to Mr
M. Tognolin for the figures, to Mr R. Pavan for the photographs, and to
Mrs A. Migliazza for typing the manuscript; from F. Scordari to Professor
F. Liebau for the critical reading of the silicate section and for helpful
suggestions, Mr Trione is also acknowledged for the typing of the
manuscript; from G. Gilli to Professor V. Bertolasi, Dr V. Ferretti, and Dr
P. Gilli for their help in the critical reading of the manuscript; from M. Catti
to Mr M. Bandera for his aid with the drawings.
Contents
...
List of contributors Xlll 1.D The space-group
rotation matrices
1 Symmetry in crystals 1.E Symmetry groups
Carmelo Giacovazzo l . F Symmetry
generalization
The crystalline state and isometric
References
operations
Symmetry elements
Axes of rotational symmetry
Crystallographic computing
Axes of rototranslation or screw axes Carmelo Giacovazzo
Axes of inversion
Introduction
Axes of rotoreflection
The metric matrix
Reflection planes with translational
component (glide planes) The reciprocal lattice
Lattices Basic transformations
The rational properties of lattices Transformation from triclinic to
orthonormal axes
Crystallographic directions
Rotations in Cartesian systems
Crystallographic planes
Some simple crystallographic calculations
Symmetry restrictions due to the lattice
periodicity and vice versa Torsion angles
Point groups and symmetry classes Best plane through a set of points
Point groups in one and two dimensions Best line through a set of points
The Laue classes Principal axes of a quadratic form
The seven crystal systems Metric considerations on the lattices
The Bravais lattices Niggli reduced cell
Plane lattices Sublattices and superlattices
Space lattices Coincidence-site lattices
The space groups Twins
The plane and line groups Calculation of the structure factor
On the matrix representation of symmetry Calculation of the electron density
operators function
Appendices: 1.A The isometric The method of least squares
transformations Linear least squares
l . B Some combinations of Reliability of the parameter estimates
movements Linear least squares with constraints
1.C Wigner-Seitz cells Non-linear (unconstrained) least squares
viii I Contents
Least-squares refinement of crystal Introduction

structures Thomson scattering
Practical considerations on Compton scattering
crystallographic least squares Interference of scattered waves
Constraints and restraints in Scattering by atomic electrons
crystallographic least squares Scattering by atoms
Alternatives to the method of least The temperature factor
squares
Scattering by a molecule or by a unit cell
Rietveld refinement
Diffraction by a crystal
The basis of the technique
Bragg's law
Some practical aspects of Rietveld
refinement The reflection and the limiting spheres
Analysis of thermal motion Symmetry in reciprocal space
The effect of thermal motion on bond Friedel law
lengths and angles Effects of symmetry operators in the
About the accuracy of the calculated reciprocal space
parameters Determination of the Laue class
Appendices: 2.A Some metric relations Determination of reflections with
between direct and restricted phase values
reciprocal lattices Systematic absences
2.B Some geometrical Unequivocal determination of the space
calculations concerning group
directions and planes Diffraction intensities
2. C Some transformation Anomalous dispersion
matrices The Fourier synthesis and the phase
2.D Reciprocity of F and I problem
lattices Modulated crystal structures
2.E Transformations of Appendices: 3.A Mathematical
crystallographic background
quantities in rectilinear 3.B Scattering and related
spaces topics
2.F Derivation of the 3.C Scattering of X-rays by
normal equations gases, liquids, and
2.G Derivation of the amorphous solids
variance-covariance 3.D About electron density
matrix M, mapping
2.H Derivation of the 3 .E Modulated structures
unbiased estimate of and quasicrystals
Mx
References
2.1 The FFT algorithm and
its crystallographic
applications 4 Experimental methods in X-ray
2.5 Examples of twin laws crystallography
References Hugo L. Monaco
X-ray sources
3 The diffraction of X-rays by crystals Conventional generators
Carmelo Giacovazzo Synchrotron radiation
Contents I ix
Monochromatization, collimation, and Phase determination procedures

focusing of X-rays Completing and refining the structure
Data collection techniques for single Difference Fourier method
crystals Least-squares method
The Weissenberg camera Absolute configuration
The precession camera
Appendices: 5.A Structure factor
The rotation (oscillation) method in probability
macromolecular crystallography distributions
Densitometry 5.B Patterson vector
The single-crystal diffractometer methods
Area detectors 5.C Two examples of
Data collection techniques for deriving phase
polycrystalline materials information from
X-ray diffraction of polycrystalline positivity
materials 5.D Probability formulae
Cameras used for polycrystalline for triplet invariants
materials 5 . E Pseudotranslational
Diffractometers used for polycrystalline symmetry
materials 5.F Magic integers
Uses of powder diffraction 5.G New multisolution
Data reduction techniques
Lorentz correction 5.H Procedures for
Polarization correction completing a partial
model
Absorption corrections
References
Radiation damage corrections
Relative scaling
Appendices: 4.A Determination of the 6 Ionic crystals
number of molecules in Fernando Scordari
the unit cell of a crystal
The structure of the atom
References
Atoms with a single electron
Atoms with more than one electron
5 Solution and refinement of crystal Interactions between ions
structures
Notes on chemical bonds
Davide Viterbo
Ionic crystals
Introduction Lattice energy: the contributions of
Statistical analysis of structure factor attractive and repulsive terms
amplitudes Lattice energy: CFSE contribution
The Patterson function and its use Applications of lattice energy
The heavy atom method calculations
Advanced Patterson methods Ionic radius
Direct methods Maximum filling principle
Introduction Coordination polyhedra
Structure invariants and semi-invariants Radius ratio rule
Probability methods Applications of the concept of ionic
Fixing the origin and the enantiomorph radius
x 1 Contents
Closest packings Ring conformations

Pauling's rules Ring conformation and group theory
Pauling's first rule Computation of puckering coordinates
Pauling's second rule Molecular geometry and the chemical
Pauling's third rule bond
Pauling's fourth rule An overview of bond theories
Pauling's fifth rule The VSEPR theory
Ideal and defect structures Valence bond (VB) theory
MX structures Hybridization. The machinery
MX, and M2X structures Molecular mechanics
MX3 and M,X, structures Molecular hermeneutics: the
A,B,X, structures interpretation of molecular structures
On the classification of silicates Correlative methods in structural
analysis
Liebau's crystallochemical classification
Some three-centre-four-electron linear
Structural formulae
systems
Relationship between classification
Nucleophilic addition to organometallic
parameters and properties of the cations
compounds
Appendices: 6.A Application of the
Nucleophilic addition to the carbonyl
concept of the packing
coefficient (c,) group
A case of conformational rearrangement
6.B Structural inferences
from crystallochemical Resonance assisted hydrogen bonding
parameters (RAHB)
References References
7 Molecules and molecular crystals 465

Gastone Gilli 8 Protein crystallography
Chemistry and X-ray crystallography 465 Giuseppe Zanotti
Crystal and molecular structure Introduction
The growth of structural information Protein crystals
The nature of molecular crystals Principles of protein crystallization
Generalities The solvent content of protein crystals
A more detailed analysis of Preparation of isomorphous heavy-atom
intermolecular forces derivatives
Thermodynamics of molecular crystals How isomorphous are isomorphous
Free and lattice energy of a crystal from derivatives?
atom-atom potentials The solution of the phase problem
Polymorphism and the prediction of The isomorphous replacement method
crystal structures Anomalous scattering: a complementary
Effect of crystal forces on molecular (or alternative) approach to the solution
geometry of the phase problem
Elements of classical stereochemistry The use of anomalous scattering in the
Structure: constitution, configuration, determination of the absolute
and conformation configuration of the macromolecule
Isomerism 486 The treatment of errors
Contents 1 xi
The refinement of heavy-atom References and further reading

parameters
Picking up minor heavy-atom sites: ,
the difference-Fourier synthesis
9 Physical properties of crystals
A third approach to the solution of the
Michele Catti
phase ambiguity: real-space filtering
Rotation and translation functions and Introduction
the molecular replacement method Crystal anisotropy and tensors
Direct methods and the maximum- Tensorial quantities
entropy principle in macromolecular Symmetry of tensorial properties
crystallography Overview of physical properties
The interpretation of electron density Electrical properties of crystals
maps and the refinement of the model
Pyroelectricity
The interpretation of electron density
Dielectric impermeability and optical
maps
properties
Interactive computer graphics and
Elastic properties of crystals
model building
Crystal strain
The refinement of the structure
Inner deformation
Protein structure
Stress tensor
General aspects
Elasticity tensor
Levels of organization of proteins:
secondary structure Examples and applications
Polypeptide chain description Piezoelectricity
Higher levels of organization: tertiary Symmetry properties of the piezoelectric
and quaternary structure, domains, and tensor
subunits Crystal defects
Groups other than amino acids Experimental methods
Thermal parameters and disordered Planar defects
structures Line defects: dislocations
Solvent structure The Burgers circuit
The influence of crystal packing X-ray topography of dislocations
Protein classification Energy of a dislocation
Appendices: 8.A Some formulae for Motion and interaction of dislocations
isomorphous Partial dislocations
replacement and Small-angle grain boundaries
anomalous dispersion
Point defects
8.B Translation functions
Thermal distribution of defects
8.C Macromolecular least-
Diffusion
squares refinement and
the conjugate-gradient Ionic conductivity
algorithm Appendix: 9.A Properties of second-rank
8.D Conventions and tensors
symbols for amino Further reading
acids and peptides Index
Nec possunt oculi naturam noscere rerum: proinde animi vitium hoc oculis adfingere noli.
Lucretius, De Rerum Natura, IV, 386-7.

Contributors
Carmelo Giacovazzo, Dipartimento Geomineralogico, Universith di Bari,

Campus Universitario, 70124 Bari, Italy; Istituto di Ricerca per lo
Sviluppo delle Metodologie Cristallografiche, CNR, 70124 Bari, Italy.
Hugo L. Monaco, Dipartimento di Genetica e Microbiologia, Sezione di
Biologia Molecolare e Biofisica, Universith di Pavia, Via Abbiategrosso
207 27100, Pavia, Italy.
Davide Viterbo, Dipartimento di Chimica, Inorganica, Chimica Fisica e
Chimica dei Materiali, Universith di Torino, Via P. Giuria 7, 10125
Torino, Italy.
Fernando Scordari, Dipartimento Geomineralogico, Universith di Bari,
Campus Universitario, 70124 Bari, Italy.
Gastone Gilli, Dipartimento di Chimica and Centro di Strutturistica
Diffrattometrica, Universith di Ferrara, Via L. Borsari 46, 44100 Ferrara,
Italy.
Giuseppe Zanotti, Dipartimento di Chimica Organica, Universith di
Padova, Via Marzolo 1, 35126 Padova, Italy.
Michele Catti, Dipartimento di Chimica Fisica ed Elettrochimica,
Universith di Milano, Via Golgi 19, 20133 Milano, Italy.
Symmetry in crystals
CARMELO G I A C O V A Z Z O
The crystalline state and isometric operations

Matter is usually classified into three states: gaseous, liquid, and solid.
Gases are composed of almost isolated particles, except for occasional
collisions; they tend to occupy all the available volume, which is subject to
variation following changes in pressure. In liquids the attraction between
nearest-neighbour particles is high enough to keep the particles almost in
contact. As a consequence liquids can only be slightly compressed. The
thermal motion has sufficient energy to move the molecules away from the
attractive field of their neighbours; the particles are not linked together
permanently, thus allowing liquids to flow.
If we reduce the thermal motion of a liquid, the links between molecules
will become more stable. The molecules will then cluster together to form
what is macroscopically observed as a rigid body. They can assume a
random disposition, but an ordered pattern is more likely because it
corresponds to a lower energy state. This ordered disposition of molecules is
called the crystalline state. As a consequence of our increased understand-
ing of the structure of matter, it has become more convenient to classify
matter into the three states: gaseous, liquid, and crystalline.
Can we then conclude that all solid materials are crystalline? For
instance, can common glass and calcite (calcium carbonate present in
nature) both be considered as crystalline? Even though both materials have
high hardness and are transparent to light, glass, but not calcite, breaks in a
completely irregular way. This is due to the fact that glass is formed by long,
randomly disposed macromolecules of silicon dioxide. When it is formed
from the molten state (glass does not possess a definite melting point, but
becomes progressively less fluid) the thermal energy which remains as the
material is cooled does not allow the polymers to assume a regular pattern.
This disordered disposition, characteristic of the liquid state, is therefore
retained when the cooling is completed. Usually glasses are referred to as
overcooled liquids, while non-fluid materials with a very high degree of
disorder are known as amorphous solids.
A distinctive property of the crystalline state is a regular repetition in the
three-dimensional space of an object (as postulated as early as the end of
the eighteenth century by R. J. Haiiy), made of molecules or groups of
molecules, extending over a distance corresponding to thousands of
molecular dimensions. However, a crystal necessarily has a number of
defects at non-zero temperature and/or may contain impurities without
losing its order. Furthermore:
1. Some crystals do not show three-dimensional periodicity because the
2 1 Carmelo Giacovazzo
basic crystal periodicity is modulated by periodic distortions incom-

mensurated with the basic periods (i.e. in incommensurately modulated
structures, IMS). It has, however, been shown (p. 171 and Appendix
3.E) that IMSs are periodic in a suitable (3 + d)-dimensional space.
2. Some polymers only show a bi-dimensional order and most fibrous
materials are ordered only along the fiber axis.
3. Some organic crystals, when conveniently heated, assume a state
intermediate between solid and liquid, which is called the mesomorphic
or liquid crystal state.
These examples indicate that periodicity can be observed to a lesser or
greater extent in crystals, depending on their nature and on the thermo-
dynamic conditions of their formation. It is therefore useful to introduce the
concept of a real crystal to stress the differences from an ideal crystal with
perfect periodicity. Although non-ideality may sometimes be a problem,
more often it is the cause of favourable properties which are widely used in
materials science and in solid state physics.
In this chapter the symmetry rules determining the formation of an ideal
crystalline state are considered (the reader will find a deeper account in
some papers devoted to the subject, or some exhaustive or in the
theoretical sections of the International Tables for Cryst~llography).[~~
In order to understand the periodic and ordered nature of crystals it is
necessary to know the operations by which the repetition of the basic
molecular motif is obtained. An important step is achieved by answering the
following question: given two identical objects, placed in random positions
and orientations, which operations should be performed to superpose one
object onto the other?
The well known coexistence of enantiomeric molecules demands a
second question: given two enantiomorphous (the term enantiomeric will
only be used for molecules) objects, which are the operations required to
superpose the two objects?
An exhaustive answer to the two questions is given by the theory of
isometric transformations, the basic concepts of which are described in
Appendix 1.A, while here only its most useful results will be considered.
Two objects are said to be congruent if to each point of one object
corresponds a point of the other and if the distance between two points of
one object is equal to the distance between the corresponding points of the
other. As a consequence, the corresponding angles will also be equal in
absolute value. In mathematics such a correspondence is called isometric.
The congruence may either be direct or opposite, according to whether
the corresponding angles have the same or opposite signs. If the congruence
is direct, one object can be brought to coincide with the other by a
convenient movement during which it behaves as a rigid body. The
movement may be:
(1) a translation, when all points of the object undergo an equal

displacement in the same direction;
(2) a rotation around an axis; all points on the axis will not change their
position;
Symmetry in crystals ( 3
(3) a rototranslation or screw movement, which may be considered as the

combination (product) of a rotation around the axis and a transl-ation
along the axial direction (the order of the two operations may be
exchanged).
If the congruence is opposite, then one object will be said to be
enantiomorphous with respect to the other. The two objects may be brought
to coincidence by the following operations:
(1) a symmetry operation with respect to a point, known as inversion;
(2) a symmetry operation with respect to a plane, known as reflection;
(3) the product of a rotation around an axis by an inversion with respect to
a point on the axis; the operation is called rotoinversion;
(4) the product of a reflection by a translation parallel to the reflection
plane; the plane is then called a glide plane.
(5) the product of a rotation by a reflection with respect to a plane
perpendicular to the axis; the operation is called rotoreflection.
Symmetry elements
Suppose that the isometric operations described in the preceding section,
not only bring to coincidence a couple of congruent objects, but act on the
entire space. If all the properties of the space remain unchanged after a
given operation has been carried out, the operation will be a symmetry
operation. Symmetry elements are points, axes, or planes with respect to
which symmetry operations are performed. .
In the following these elements will be considered in more detail, while
the description of translation operators will be treated in subsequent
sections.
Axes of rotational symmetry

If all the properties of the space remain unchanged after a rotation of 2nIn
around an axis, this will be called a symmetry axis of order n ; its written
symbol is n. We will be mainly interested (cf. p. 9) in the axes 1, 2, 3, 4, 6.
Axis 1 is trivial, since, after a rotation of 360" around whatever direction
the space properties will always remain the same. The graphic symbols for
the 2, 3, 4, 6 axes (called two-, three-, four-, sixfold axes) are shown in
Table 1.1. In the first column of Fig. 1.1 their effects on the space are
illustrated. In keeping with international notation, an object is represented
by a circle, with a + or - sign next to it indicating whether it is above or
below the page plane. There is no graphic symbol for the 1 axis. Note that a
4 axis is at the same time a 2 axis, and a 6 axis is at the same time a 2 and a
3 axis.
Table 1.1. Graphical symbols for symmetry elements: (a) axes normal to the pfane of
projection; (b) axes 2 and 2, ,parallel to the plane of projection; (c) axes parallel or
inclined to the plane of projection; (d) symmetry pfanes normar to the plane of
projection; (e) symmetry planes parallel to the plane of projection
Fig. 1.1. Arrangements of symmetry-equivalent

objects as an effect of rotation, inversion, and
screw axes.
Symmetry in crystals 1 5
Axes of rototranslation or screw axes

A rototranslational symmetry axis will have an order n and a translational
component t, if all the properties of the space remain unchanged after a
2nln rotation around the axis and the translation by t along the axis. On p.
10 we will see that in crystals only screw axes of order 1, 2, 3, 4, 6 can exist
with appropriate translational components.
Axes of inversion
An inversion axis of order n is present when all the properties of the space
remain unchanged after performing the product of a 2nln rotation around
the axis by an inversion with respect to a point located on the same axis.
The written symbol is fi (read 'minus n' or 'bar n'). As we shall see on p. 9
we will be mainly interested in 1, 2, 3, 4, 6 axes, and their graphic symbols
are given in Table 1.1, while their effects on the space are represented in the
second column of Fig. 1.1. According to international notation, if an object
is represented by a circle, its enantiomorph is depicted by a circle with a
comma inside. When the two enantiomorphous objects fall one on top of
the other in the projection plane of the picture, they are represented by a
single circle divided into two halves, one of which contains a comma. To
each half the appropriate + or - sign is assigned.
We may note that:
(1) the direction of the i axis is irrelevant, since the operation coincides
with an inversion with respect to a point;
(2) the 2 axis is equivalent to a reflection plane perpendicular to it; the

properties of the half-space on one side of the plane are identical to
those of the other half-space after the reflection operation. The written
symbol of this plane is m;
(3) the 3 axis is equivalent to the product of a threefold rotation by an

inversion: i.e. 3 = 31;
(4) the 4 axis is also a 2 axis;

(5) the 6 axis is equivalent to the product of a threefold rotation by a
reflection with respect to a plane normal to it; this will be indicated by
6 = 3/m.
Axes of rotoreflection
A rotoreflection axis of order n is present when all the properties of the
space do not change after performing the product of a 2nln rotation around
an axis by a reflection with respect to a plane normal to it. The written
symbol of this axis is fi. The effects on the space of the 1, 2, 3, 4, 6 axes
coincide with those caused by an inversion axis (generally of a different
order). In particular: i = m, 2 = 1, 3 = 6, 4 = 4, 6 = 3. From now on we will
no longer consider the ii axes but their equivalent inversion axes.
Reflection planes with translational component (glide

planes)
A glide plane operator is present if the properties of the half-space on one
side of the plane are identical to those of the other half-space after the
product of a reflection with respect to the plane by a translation parallel to
the plane. On p. 11 we shall see which are the glide planes found in crystals.
Symmetry operations relating objects referred by direct congruence are
called proper (we will also refer to proper symmetry axes) while those
relating objects referred by opposite congruence are called improper (we
will also refer to improper axes).
Lattices
Translational periodicity in crystals can be conveniently studied by con-
sidering the geometry of the repetion rather than the properties of the motif
which is repeated. If the motif is periodically repeated at intervals a, b, and
/Cl /Cl /Cl ,,Cl /Cl ,,Cl ,Cl
c along three non-coplanar directions, the repetition geometry can be fully
o H , H? ?H ? H ? H? H? H described by a periodic sequence of points, separated by intervals a, b, c
/C1 /Cl /Cl /Cl &C1 7 1 /Cl along the same three directions. This collection of points will be called a
lattice. We will speak of line, plane, and space lattices, depending on
H H H H
whether the periodicity is observed in one direction, in a plane, or in a
three-dimensional space. An example is illustrated in Fig. 1.2(a), where
HOCl is a geometrical motif repeated at intervals a and b. If we replace the
molecule with a point positioned at its centre of gravity, we obtain the
lattice of Fig. 1.2(b). Note that, if instead of placing the lattice point at the
centre of gravity, we locate it on the oxygen atom or on any other point of
the motif, the lattice does not change. Therefore the position of the lattice
with respect to the motif is completely arbitrary.
If any lattice point is chosen as the origin of the lattice, the position of
any other point in Fig. 1.2(b) is uniquely defined by the vector
where u and v are positive or negative integers. The vectors a and b define a
parallelogram which is called the unit cell: a and b are the basis vectors of
the cell. The choice of the vectors a and b is rather arbitrary. In Fig. 1.2(b)
four possible choices are shown; they are all characterized by the property
that each lattice point satisfies relation (1.1) with integer u and v.
Nevertheless we are allowed to choose different types of unit cells, such
as those shown in Fig. 1.2(c), having double or triple area with respect to
those selected in Fig. 1.2(b). In this case each lattice point will still satisfy
(1.1) but u and v are no longer restricted to integer values. For instance, the
point P is related to the origin 0 and to the basis vectors a' and b' through
( 4 v) = (112, 112).
The different types of unit cells are better characterized by determining
the number of lattice points belonging to them, taking into account that the
~ i ~ (a). R~~~~~~~~~ of a graphical motif as an points on sides and on corners are only partially shared by the given cell.
example of a two-dimensional crystal; (b) The cells shown in Fig. 1.2(b) contain only one lattice point, since the
corresponding lattice with some examples Of four points at the corners of each cell belong to it for only 114. These cells
primitive cells; (c) corresponding lattice with
some examples of multiple cells. are called primitive. The cells in Fig. 1.2(c) contain either two or three
points and are called multiple or centred cells. Several kinds of multiple
cells are possible: i.e. double cells, triple cells, etc., depending on whether
they contain two, three, etc. lattice points.
The above considerations can be easily extended to linear and space
lattices. For the latter in particular, given an origin 0 and three basis II
vectors a , b, and c, each node is uniquely defined by the vector
= ua + ub + W C . (1.2)
The three basis vectors define a parallelepiped, called again a unit cell. a
Fig. Notation for a unit cell.
The directions specified by the vectors a , b, and c are the X , Y, Z
crystallographic axes, respectively, while the angles between them are
indicated by a, 0, and y, with a opposing a , opposing b, and y opposing
c (cf. Fig. 1.3). The volume of the unit cell is given by
where the symbol '.' indicates the scalar product and the symbol ' A ' the
vector product. The orientation of the three crystallographic axes is usually
chosen in such a way that an observer located along the positive direction of
c sees a moving towards b by an anti-clockwise rotation. The faces of the
unit cell facing a , b, and c are indicated by A, B, C, respectively. If the
chosen cell is primitive, then the values of u, u, w in (1.2) are bound to be
integer for all the lattice points. If the cell is multiple then u, u, w will have
rational values. To characterize the cell we must recall that a lattice point at
vertex belongs to it only for 1/8th, a point on a edge for 114, and one on a
face for 112.
The rational properties of lattices

Since a lattice point can always be characterized by rational numbers, the
lattice properties related to them are called rational. Directions defined by
two lattice points will be called rational directions, and planes defined by
three lattice points rational planes. Directions and planes of this type are
also called crystallographic directions and planes.
Crystallographic directions
Since crystals are anisotropic, it is necessary to specify in a simple way
directions (or planes) in which specific physical properties are observed.
Two lattice points define a lattice row. In a lattice there are an infinite
number of parallel rows (see Fig. 1.4): they are identical under lattice
translation and in particular they have the same translation period.
A lattice row defines a crystallographic direction. Suppose we have chosen a
primitive unit cell. The two lattice vectors Q , and Q,,,,, ,, ,,
with u, u,
w, and n integer, define two different lattice points, but only one direction.
This property may be used to characterize a direction in a unique way. For
instance, the direction associated with the vector Q9,,,, can be uniquely
defined by the vector Q,,,,,with no common factor among the indices. This
direction will be indicated by the symbol [3 1 21, to be read as 'three, one,
two' and not 'three hundred and twelve'. Fig. 1.4. Lattice rows and planes.
When the cell is not primitive u, v, w, and n will be rational numbers.

Thus Q112,312,-113 and Q512,1512,-5,3 define the same direction. The indices of
the former may therefore be factorized to obtain a common denominator
and no common factor among the numerators: Q, ,-,,,
= Q3,6,6,-,6 +
[3 9 -21 to be read 'three', nine, minus two'.
Crystallographic planes
Three lattice points define a crystallographic plane. Suppose it intersects the
three crystallographic axes X , Y , and Z at the three lattice points ( p , 0, 0 ) )
(0, q, 0 ) and (0, 0 , r ) with integer p, q, r (see Fig. 1.5). Suppose that m is
the least common multiple of p, q , r. Then the equation of the plane is
x'lpa + y'lqb + z'lrc = 1.

If we introduce the fractional coordinates x = x ' l a , y = y ' l b , z = z l / c , the
equation of the plane becomes
xlp + y l q + z l r = 1. (1.3)
Multiplying both sides by m we obtain
hx + ky + lz = m (1.4)
Fig. 1.5. Some lattice planes of the set (236).
where h , k , and 1 are suitable integers, the largest common integer factor of
which will be 1.
We can therefore construct a family of planes parallel to the plane (1.4),
by varying m over all integer numbers from -m to +m. These will also be
crystallographic planes since each of them is bound to pass through at least
three lattice points.
The rational properties of all points being the same, there will be a plane
of the family passing through each lattice point. For the same reason each
lattice plane is identical to any other within the family through a lattice
translation.
Let us now show that (1.4) represents a plane at a distance from the
origin m times the distance of the plane
The intercepts of the plane (1.5) on X, Y , Z will be l l h , l l k and 111

respectively and those of (1.4) m l h , m l k , m l l . It is then clear that the
distance of plane (1.4) from the origin is m times that of plane (1.5). The
first plane of the family intersecting the axes X , Y , and Z at three lattice
points is that characterized by a m value equal to the least common mutiple
of h , k , I. We can therefore conclude that eqn (1.4) defines, as m is varied, a
family of identical and equally spaced crystallographic planes. The three
indices h , k , and 1 define the family and are its Miller indices. To indicate
that a family of lattice planes is defined by a sequence of three integers,
these are included within braces: ( h k I ) . A simple interpretation of the
three indices h, k , and I can be deduced from (1.4) and (1.5). In fact they
indicate that the planes of the family divide a in h parts, b in k parts, and c 0 k. (110) (010)
in 1 parts.
Crystallographic planes parallel to one of the three axes X, Y, or Z are
defined by indices of type (Okl), (hol), or (hkO) respectively. Planes parallel
to faces A, B, and C of the unit cell are of type (hOO), (OkO), and (001)
respectively. Some examples of crystallographic planes are illustrated in Fig.
1.6.
As a numerical example let us consider the plane
(zio)
which can be written as Fig. 1.6. Miller indices for some crystallographic
planes parallel to Z ( Z i s supposed to be normal
to the page).
The first plane of the family with integer intersections on the three axes will
be the 30th (30 being the least common multiple of 10, 15, and 6) and all the
planes of the family can be obtained from the equation lox + 15y 62 = m,+
by varying m over all integers from -m to +m. We observe that if we divide
p, q, and r in eqn (1.6) by their common integer factor we obtain
+ +
x/3 y/2 z/5 = 1, from which
Planes (1.7) and (1.8) belong to the same family. We conclude that a
family of crystallographic planes is always uniquely defined by three indices
h, k, and 1 having the largest common integer factor equal to unity.
Symmetry restrictions due to the lattice

periodicity and vice versa
Suppose that the disposition of the molecules in a crystal is compatible with
an n axis. As a consequence the disposition of lattice points must also be
compatible with the same axis. Without losing generality, we will assume
that n passes through the origin 0 of the lattice. Since each lattice point has
identical rational properties, there will be an n axis passing through each
and every lattice point, parallel to that passing through the origin. In
particular each symmetry axis will lie along a row and will be perpendicular
to a crystallographic plane.
Let T be the period vector of a row passing through 0 and normal to n.
We will then have lattice points (see Fig. 1.7(a)) at T, -T, T', and T". The
vector T' - T" must also be a lattice vector and, being parallel to T, we will
have T' - T" = mT where m is an integer value: in a scalar form
2 cos (2nIn) = m (m integer). (1.9)
Equation (1.9) is only verified for n = 1, 2, 3, 4, 6. It is noteworthy that a
5 axis is not allowed, this being the reason why it is impossible to pave a
room only with pentagonal tiles (see Fig. 1.7(b).
(b)
A unit cell, and therefore a lattice, compatible with an n axis will also be
Fig. 1.7.(a) Lattice points in a plane normal to the
compatible with an ii axis and vice versa. Thus axes I, 3, 3, 4, 6 will also be symmetry axis n passing through 0.(b) Regular
allowed. pentagons cannot fill planar space.
Let us now consider the restrictions imposed by the periodic nature of

crystals on the translational components t of a screw axis. Suppose that this
lies along a row with period vector T. Its rotational component must
correspond to n = 1, 2, 3, 4, 6. If we apply the translational component n
times the resulting displacement will be nt. In order to maintain the
periodicity of the crystal we must have nt = p T , with integer p , or
For instance, for a screw axis of order 4 the allowed translational

components will be (0/4)T, (1/4)T, (2/4)T, (3/4)T, (4/4)T, (5/4)T, . . .; of
these only the first four will be distinct. It follows that:
(1) in (1.10) p can be restricted within 0 s p < n ;
(2) the n-fold axis may be thought as a special screw with t = 0. The nature
of a screw axis is completely defined by the symbol n,. The graphic
symbols are shown in Table 1.1: the effects of screw axes on the
surrounding space are represented in Fig. 1.8. Note that:
If we draw a helicoidal trajectory joining the centres of all the objects
related by a 3, and by a 32 axis, we will obtain, in the first case a
right-handed helix and in the second a left-handed one (the two helices
are enantiomorphous). The same applies to the pairs 4, and 4,, 61 and 6,,
and 6, and 6,.
4, is also a 2 axis, 6, is also a 2 and a 32, 64 is also a 2 and a 3,, and 63 is
Fig. 1.8. Screw axes: arrangement of symmetry- also a 3 and a 2,.
equivalent objects.
Symmetry in crystals I 11
We will now consider the restrictions imposed by the periodicity on the

translation component t of a glide plane. If we apply this operation twice,
the resulting movement must correspond to a translation equal to pT, where
p may be any integer and T any lattice vector on the crystallographic plane
on which the glide lies. Therefore 2t = p T , i.e. t = ( p / 2 ) T .As p varies over
all integer values, the following translations are obtained OT, (1/2)T,
(2/2)T, (3/2)T,. . . of which only the first two are distinct. For p = 0 the
glide plane reduces to a mirror m. We will indicate by a, b, c axial glides
with translational components equal to a / 2 , b / 2 , c / 2 respectively, by n the
+
diagonal glides with translational components ( a b ) / 2 or ( a + c ) / 2 or
+
( b c ) / 2 or ( a + b + c ) / 2 .
In a non-primitive cell the condition 2t = p T still holds, but now T is a
lattice vector with rational components indicated by the symbol d. The
graphic symbols for glide planes are given in Table 1.1.
Point groups a n d s y m m e t r y classes

In crystals more symmetry axes, both proper and improper, with or without
translational components, may coexist. We will consider here only those
combinations of operators which do not imply translations, i.e. the
combinations of proper and improper axes intersecting in a point. These are
called point groups, since the operators form a mathematical group and
leave one point fixed. The set of crystals having the same point group is
called crystal class and its symbol is that of the point group. Often point
group and crystal class are used as synonyms, even if that is not correct in
principle. The total number of crystallographic point groups (for three-
dimensional crystals) is 32, and they were first listed by Hessel in 1830.
The simplest combinations of symmetry operators are those characterized
by the presence of only one axis, which can be a proper axis or an inversion
one. Also, a proper and an inversion axis may be simultaneously present.
The 13 independent combinations of this type are described in Table 1.2.
When along the same axis a proper axis and an inversion axis are
simultaneously present, the symbol n/ri is used. Classes coinciding with
other classes already quoted in the table are enclosed in brackets.
The problem of the coexistence of more than one axis all passing by a
common point was first solved by Euler and is illustrated, with a different
approach, in Appendix 1.B. Here we only give the essential results. Let us
suppose that there are two proper axes I , and l2 intersecting in 0 (see Fig.
1.9). The I , axis will repeat in Q an object originally in P, while 1, will
Table 1.2. Single-axis crystallographic point groups
Proper axis Improper axis Proper and improper

axis
1 = i)
(i[i
-
1
2 = m- 212-= 2jm
3 3 31 (3[3 = 3)
4 4 414 = 4/m
6 6 = 3/m 616 = 6/m
5 + 5 + 3 = 13
t.'
12 ( Carmelo Giacovazzo
11 Table 1.3. For each combination of symmetry axes the minimum angles between axes
are given. For each angle the types of symmetry axes are quoted in parentheses
Combination of cu (ded B (ded Y (ded
.-
P '.--
symmetry axes
2
3
2
2
2
2
90
90
(22)
(2 3)
90 (2 2)
90 (2 3)
90
60
(22)
(2 2)
---0
4 2 2 90 (2 4) 90 (2 4) 45 (2 2)
6 2 2 90 (2 6) 90 (2 6) 30 (2 2)
0 2 3 3 54 44'08" (2 3) 54 44'08" (2 3) 70 31 '44" (3 3)
Fig. 1.9. Arrangement of equivalent objects 4 3 2 35 15'52" (2 3) 45 (2 4) 54 44'08" (4 3)
around two intersecting symmetry axes.
repeat in R the object in Q. P and Q are therefore directly congruent and

this implies the existence of another proper operator which repeats the
object in P directly in R. The only allowed combinations are n22, 233, 432,
532 which in crystals reduce to 222, 322, 422, 622, 233, 432. For these
combinations the smallest angles between the axes are listed in Table 1.3,
while their disposition in the space is shown in Fig. 1.10. Note that the
combination 233 is also consistent with a tetrahedral symmetry and 432 with
a cubic and octahedral symmetry.
Suppose now that in Fig. 1.9 1, is a proper axis while 1, is an inversion
one. Then the objects in P and in Q will be directly congruent, while the
object in R is enantiomorphic with respect to them. Therefore the third
operator relating R to P will be an inversion axis. We may conclude that if
one of the three symmetry operators is an inversion axis also another must
be an inversion one. In Table 1.4 are listed all the point groups
characterized by combinations of type PPP, PII, IPI, IIP (P=proper,
I = improper), while in Table 1.5 the classes with axes at the same time
proper and improper are given. In the two tables the combinations
coinciding with previously considered ones are closed within brackets. The
Fig. 1.10. Arrangement of proper symmetry

axes for six point groups.
Table 1.4. Crystallographic point groups with more than one axis
PPP PI1 I P I IIP
(43 2 ----
432
rnlm
Table 1.5. Crystallographic point groups with more than one axis, each axis being
proper and improper simultaneously
results so far described can be easily derived by recalling that:

If two of the three axes are symmetry equivalent, they can not be one
proper and one improper; for example, the threefold axes in 233 are
symmetry referred by twofold axes, while binary axes in 422 differing by
45" are not symmetry equivalent.
If an even-order axis and a ? axis (or an m plane) coexist, there will also
be an m plane (or a ? axis) normal to the axis and passing through the
intersection point. Conversely, if m and ? coexist, there will also be a 2
axis passing through ? and normal to m.
In Tables 1.2, 1.4, and 1.5 the symbol of each point group does not reveal
all the symmetry elements present: for instance, the complete list of
symmetry elements in the class 2/m33 is 2/m 2/m 2/m?3333. On the other
hand, the symbol 2/m% is too extensive, since only two symmetry operators
are independent. In Table 1.6 are listed the conventional symbols used for
the 32 symmetry classes. It may be noted that crystals with inversion
symmetry operators have an equal number of 'left' and 'right' moieties;
these parts, when considered separately, are one the enantiomorph of the
other.
The conclusions reached so far do not exclude the possibility of crystal-
lizing molecules with a molecular symmetry different from that of the 32
point groups (for instance with a 5 axis). In any case the symmetry of the
crystal will belong to one of them. To help the reader, some molecules and
their point symmetry are shown in Fig. 1.11.
It is very important to understand how the symmetry of the physical
properties of a crystal relates to its point group (this subject is more
extensively described in Chapter 9). Of basic relevance to this is a postulate
E E
E E E
E E E E Ekkbmm
.-2E341mmo E E
E E
E E E
EEE E EkEnmm
. - 2 E 9 3 ~ ~ m wE E
2. The variation of the refractive index of the crystal with the vibration
direction of a plane-polarized light wave is represented by the optical
indicatrix (see p. 607). This is in general a three-axis ellipsoid: thus the
lowest symmetry of the property 'refraction' is 2/m 2lm 2/m, the point
group of the ellipsoid. In crystal classes belonging to tetragonal, trigonal, or
hexagonal systems (see Table 1.6) the shape of the indicatrix is a rotational
ellipsoid (the axis is parallel to the main symmetry axis), and in symmetry
classes belonging to the cubic system the shape of the indicatrix is a sphere.
For example, in the case of tourmaline, with point group 3m, the ellipsoid is
a revolution around the threefold axis, showing a symmetry higher than that
of the point group.
We shall now see how it is possible to guess about the point group of a
crystal through some of its physical properties:
1. The morphology of a crystal tends to conform to its point group
symmetry. From a morphological point of view, a crystal is a solid body
bounded by plane natural surfaces, the faces. The set of symmetry-
equivalent faces constitutes a form: the form is open if it does not enclose
space, otherwise it is closed. A crystal form is named according to the
number of its faces and to their nature. Thus a pedion is a single face, a
pinacoid is a pair of parallel faces, a sphenoid is a pair of faces related by a
diad axis, a prism a set of equivalent faces parallel to a common axis, a
pyramid is a set of planes with equal angles of inclination to a common axis,
etc. The morphology of different samples of the same compound can show
different types of face, with different extensions, and different numbers of
edges, the external form depending not only on the structure but also on the
chemical and physical properties of the environment. For instance, galena
crystals (PbS, point group m3m) tend to assume a cubic, cube-octahedral,
or octahedral habit (Fig. 1.12(a)). Sodium chloride grows as cubic crystals
from neutral aqueous solution and as octahedral from active solutions (in
the latter case cations and anions play a different energetic role). But at the
same temperature crystals will all have constant dihedral angles between
corresponding faces (J. B. L. Rome' de l'Ile, 1736-1790). This property, the
observation of which dates back to N. Steno (1669) and D. Guglielmini
(1688), can be explained easily, following R. J. Haiiy (1743-1822), by
considering that faces coincide with lattice planes and edges with lattice
rows. Accordingly, Miller indices can be used as form symbols, enclosed in
braces: {hkl). The indices of well-developed faces on natural crystals tend
to have small values of h, k, 1, (integers greater than six are rarely
involved). Such faces correspond to lattice planes with a high density of
lattice points per unit area, or equivalently, with large intercepts alh, blk,
cll on the reference axes (Bravais' law). An important extension of this law
is obtained if space group symmetry (see p. 22) is taken into account: screw
axes and glide planes normal to a given crystal face reduce its importance
(Donnay-Harker principle).
The origin within the crystal is usually chosen so that faces (hkl) and
(hit) are parallel faces an opposite sides of the crystal. In Fig. 1.13 some
idealized crystal forms are shown.
(b)
The orientation of the faces is more important than their extension. The
Fig. 1.12. (a) Crystals showing cubic or cube-
orientations can be represented by the set of unit vectors normal to them. octahedral or octahedral habitus, (b) crystal with
This set will tend to assume the point-group symmetry of the given crystal a sixfold symmetry axis.
independently of the morphological aspect of the samples. Thus, each

sample of Fig. 1.12(a) shows an m3m symmetry, and the sample in Fig.
1.12(b) shows a sixfold symmetry if the normals to the faces are considered
instead of their extensions. The morphological analysis of a crystalline
sample may be used to get some, although not conclusive, indication, of its
point-group symmetry.
2. Electrical charges of opposite signs-may appear at the two hands of a
polar axis of a crystal subject to compression, because of the piezoelectric
effect (see p. 619). A polar axis is a rational direction which is not symmetry
equivalent to its opposite direction. It then follows that a polar direction can
only exist in the 21 non-centrosymmetric point groups (the only exception is
the 432 class, where piezoelectricity can not occur). In these groups not all
directions are polar: in particular a direction normal to an even-order axis
or to a mirror plane will never be polar. For instance, in quartz crystals
(SOz, class 32), charges of opposite sign may appear at the opposite hands
Fig. 1.13. Some simple crystal forms: (a)
cinnabar, HgS, class 32; (b) arsenopyritf, FeAsS, of the twofold axes, but not at those of the threefold axis.
class mmm; (c) ilmenite, FeTiO,, class 3; (d)
gypsum, CaSO,, class 2/m. 3. A point group is said to be polar if a polar direction, with no other
symmetry equivalent directions, is allowed. Along this direction a per-
manent electric dipole may be measured, which varies with temperature
(pyroelectric effect, see p. 606). The ten polar classes are: 1, 2, m, mm2, 4,
4mm, 6, 6mm, 3, 3m. Piezo- and pyroelectricity tests are often used to
exclude the presence of an inversion centre. Nevertheless when these effects
are not detectable, no definitive conclusion may be drawn.
4. Ferroelectric crystals show a permanent dipole moment which can be
changed by application of an electric field. Thus they can only belong to one
of the ten polar classes.
5. The symmetry of a crystal containing only one enantiomer of an
optically active molecule must belong to one of the 11 point groups which
do not contain inversion axes.
6. Because of non-linear optical susceptibility, light waves passing
through non-centrosymmetric crystals induce additional waves of frequency
twice the incident frequency. This phenomenon is described by a third-rank
tensor, as the piezoelectric tensor (see p. 608): it occurs in all non-
centrosymmetric groups except 432, and is very efficientL7]for testing the
absence of an inversion centre.
7. Etch figures produced on the crystal faces by chemical attack reveal
the face symmetry (one of the following 10 two-dimensional point groups).
Point groups in one and two dimensions

The derivation of the crystallographic point groups in a two-dimensional
space is much easier than in three dimensions. In fact the reflection with
respect to a plane is substituted by a reflection with respect to a line (the
same letter m will also indicate this operation); and ii axes are not used. The
total number of point groups in the plane is 10, and these are indicated by
the symbols: 1, 2, 3, 4, 6, m, 2mm, 3m, 4mm, 6mm.
The number of crystallographic point groups in one dimension is 2: they
are 1 and m = (I).
The Laue classes

In agreement with Neumann's principle, physical experiments do not
normally reveal the true symmetry of the crystal: some of them, for example
diffraction, show the symmetry one would obtain by adding an inversion
centre to the symmetry elements actually present. In particular this happens
when the measured quantities do not depend on the atomic positions, but
rather on the interatomic vectors, which indeed form a centrosymmetric set.
Point groups differing only by the presence of an inversion centre will not be
differentiated by these experiments. When these groups are collected in
classes they form the 11 Laue classes listed in Table 1.6.
The seven crystal systems

If the crystal periodicity is only compatible with rotation or inversion axes of
order 1, 2, 3, 4, 6, the presence of one of these axes will impose some
restrictions on the geometry of the lattice. It is therefore convenient to
group together the symmetry classes with common features in such a way
that crystals belonging to these classes can be described by unit cells of the
same type. In turn, the cells will be chosen in the most suitable way to show
the symmetry actually present.
Point groups 1 and i have no symmetry axes and therefore no constraint
axes for the unit cell; the ratios a:b:c and the angles a , P, y can assume any
value. Classes 1 and are said to belong to the triclinic system.
Groups 2, m, and 2/m all present a 2 axis. If we assume that this axis
coincides with the b axis of the unit cell, a and c can be chosen on the lattice
plane normal to b. We will then have a = y = 90" and P unrestricted and the
ratios a:b:c also unrestricted. Crystals with symmetry 2, m, and 2/m belong
to the monoclinic system.
Classes 222, mm2, mmm are characterized by the presence of three
mutually orthqgonal twofold rotation or inversion axes. If we assume these
as reference axes, we will obtain a unit cell with angles a = P = y = 90" and
with unrestricted a:b:c ratios. These classes belong to the orthorhombic
system.
For the seven groups with only one fourfold axis
[4,4,4/m, 422,4mm, 42m, 4/mmm] the c axis is chosen as the direction of
the fourfold axis and the a and b axes will be symmetry equivalent, on the
lattice plane normal to c. The cell angles will be a = P = y = 90" and the
ratios a:b:c = 1:l:c. These crystals belong to the tetragonal system.
For the crystals with only one threefold or sixfold axis [3, 3, 32, 3m, 3m,
6, 6, 6/m, 622, 6mm, 62m, 6/mm] the c axis is assumed along the three- or
sixfold axis, while a and b are symmetry equivalent on the plane
perpendicular to c. These point groups are collected together in the trigonal
and hexagonal systems, respectively, both characterized by a unit cell with
angles a = /3 = 90" and y = 120°, and ratios a:b :c = 1:1:c.
Crystals with four threefold axes [23, m3, 432, 43m, m3m] distributed as
the diagonals of a cube can be referred to orthogonal unit axes coinciding
with the cube edges. The presence of the threefold axes ensures that these
directions are symmetry equivalent. The chosen unit cell will have a = P =
y = 90' and ratios a :b:c = 1:1:1. This is called the cubic system.
The Bravais lattices
In the previous section to each crystal system we have associated a primitive
cell compatible with the point groups belonging to the system. Each of these
primitive cells defines a lattice type. There are also other types of lattices,
based on non-primitive cells, which can not be related to the previous ones.
In particular we will consider as different two lattice types which can not be
described by the same unit-cell type.
In this section we shall describe the five possible plane lattices and
fourteen possible space lattices based both on primitive and non-primitive
cells. These are called Bravais lattices, after Auguste Bravais who first listed
them in 1850.
Plane lattices
An oblique cell (see Fig. 1.14(a)) is compatible with the presence of axes 1
or 2 normal to the cell. This cell is primitive and has point group 2.
If the row indicated by m in Fig. 1.14(b) is a reflection line, the cell must
be rectangular. Note that the unit cell is primitive and compatible with the
point groups m and 2mm. Also the lattice illustrated in Fig. 1.14(c) with
a = b and y # 90" is compatible with m. This plane lattice has an oblique
primitive cell. Nevertheless, each of the lattice points has a 2mm symmetry
and therefore the lattice must be compatible with a rectangular system. This
can be seen by choosing the rectangular centred cell defined by the unit
vectors a' and b'. This orthogonal cell is more convenient because a simpler
coordinate system is allowed. It is worth noting that the two lattices shown
in Figs. 1.14(b) and 1.14(c) are of different type even though they are
compatible with the same point groups.
In Fig. 1.14(d) a plane lattice is represented compatible with the presence
of a fourfold axis. The cell is primitive and compatible with the point groups
4 and 4mm.
In Fig. 1.14(e) a plane lattice compatible with the presence of a three- or
a sixfold axis is shown. A unit cell with a rhombus shape and angles of 60"
and 120" (also called hexagonal) may be chosen. A centred rectangular cell
can also be selected, but such a cell is seldom chosen.
(b) m ; 2 m m (c) m;2mm
I I i i I
Fig. 1.14. The five plane lattices and the
corresponding two-dimensional point groups. (d) 4;4mm
Table 1.7. The five plane lattices

- --
Cell Type of cell Point group Lattice parameters

of the net
Oblique P 2 a, b, y
Rectangular P, C 2mm a, b, y = 90"
Square P 4mm a = b, y = 90"
Hexagonal P 6mm a=b,y=120"
The basic features of the five lattices are listed in Table 1.7
Space lattices
In Table 1.8 the most useful types of cells are described. Their fairly limited
number can be explained by the following (or similar) observations:
A cell with two centred faces must be of type F. In fact a cell which is at
the same time A and B, must have lattice points at (0,1/2,1/2) and
(1/2,0, 112). When these two lattice translations are applied one after
the other they will generate a lattice point also at (1/2,1/2,0);
A cell which is at the same time body and face centred can always be
reduced to a conventional centred cell. For instance an I and A cell will
have lattice points at positions (1/2,1/2,1/2) and (0,1/2,1/2): a lattice
point at (1/2,0,0) will then also be present. The lattice can then be
described by a new A cell with axes a ' = a/2, b' = b, and c' = c (Fig.
1.15).
It is worth noting that the positions of the additional lattice points in
Table 1.8 define the minimal translational components which will move an
object into an equivalent one. For instance, in an A-type cell, an object at
+
( x , y, z) is repeated by translation into ( x , y + m/2, z n/2) with m and n
integers: the shortest translation will be (0,1/2,1/2).
Let us now examine the different types of three-dimensional lattices
grouped in the appropriate crystal systems.
J
Fig. 1.15. Reduction of an I- and A-centred cell
Table 1.8. The conventional types of unit cell to an A-centred cell.
Symbol Type Positions of Number

additional of lattice
lattice points points
per cell
P primitive - 1
I body centred (112,1/2, l r 2 ) 2
A A-face centred (0,1/2,1/2) 2
B B-face centred (1/2,0,1/2) 2
C C-face centred (1/2,1/2,0) 2
F All faces centred (112,112, O), (1/2,0,1/2) 2
~0,112,112~ 4
R Rhombohedrally (1/3,2/3,2/3), (2/3,1/3,1/3) 3
centred (de
scription with
'hexagonal axes')
Triclinic lattices
Even though non-primitive cells can always be chosen, the absence of axes
with order greater than one suggests the choice of a conventional primitive
cell with unrestricted a , p, y angles and a:b:c ratios. In fact, any triclinic
lattice can always be referred to such a cell.
Monoclinic lattices
The conventional monoclinic cell has the twofold axis parallel to b, angles
a = y = 90", unrestricted p and a :b:c ratios. A B-centred monoclinic cell
with unit vectors a, b, c is shown in Fig. 1.16(a). If we choose a' = a ,
+
b' = b, c' = ( a c ) / 2 a primitive cell is obtained. Since c' lies on the (a, c)
plane, the new cell will still be monoclinic. Therefore a lattice with a B-type
monoclinic cell can always be reduced to a lattice with a P monoclinic cell.
An I cell with axes a, b, c is illustrated in Fig. 1.16(b). If we choose
a' = a, b' = b, c' = a + c, the corresponding cell becomes an A monoclinic
cell. Therefore a lattice with an I monoclinic cell may always be described
by an A monoclinic cell. Furthermore, since the a and c axes can always be
interchanged, an A cell can be always reduced to a C cell.
An F cell with axes a, b, c is shown in Fig. 1.16(c). When choosing
+
a' = a, b' = b, c' = ( a c ) / 2 a type-C monoclinic cell is obtained. There-
fore, also, a lattice described by an F monoclinic cell can always be
described by a C monoclinic cell.
We will now show that there is a lattice with a C monoclinic cell which is
not amenable to a lattice having a P monoclinic cell. In Fig. 1.16(d) a C cell
with axes a, b, c is illustrated. A primitive cell is obtained by assuming
+
a' = ( a + b ) / 2 , b' = ( - a b ) / 2 , c' = c, but this no longer shows the
features of a monoclinic cell, since y' # 90°, a' = b' # c ' , and the 2 axis lies
along the diagonal of a face. It can then be concluded that there are two
distinct monoclinic lattices, described by P and C cells, and not amenable
one to the other.
Orthorhombic lattices
In the conventional orthorhombic cell the three proper or inversion axes are
parallel to the unit vectors a, b, c, with angles a = /3 = y = 90" and general
a:b:c ratios. With arguments similar to those used for monoclinic lattices,
the reader can easily verify that there are four types of orthorhombic
lattices, P, C, I, and F.
Tetragonal lattices
In the conventional tetragonal cell the fourfold axis is chosen along c with
a = p = y = 90°, a = b, and unrestricted c value. It can be easily verified
that because of the fourfold symmetry an A cell will always be at the same
time a B cell and therefore an F cell. The latter is then amenable to a
tetragonal I cell. A C cell is always amenable to another tetragonal P cell.
Thus only two different tetragonal lattices, P and I, are found.
Fig. 1.16. Monoclinic lattices: (a) reduction of a Cubic lattices

B-centred cell to a P cell; (b) reduction of an In the conventional cubic cell the four threefold axes are chosen to be
I-centred to an A-centred cell; (c) reduction of an
F-centred to a C-centred cell; (d) reduction of a parallel to the principal diagonals of a cube, while the unit vectors a, b, c
C-centred to a P non-monoclinic cell. are parallel to the cube edges. Because of symmetry a type-A (or B or C)
cell is also an F cell. There are three cubic lattices, P, I, and F which are not
amenable one to the other.
Hexagonal lattices
In the conventional hexagonal cell the sixfold axis is parallel to c, with
a = b, unrestricted c, a = /3 = 90") and y = 120". P is the only type of
hexagonal Bravais lattice.
Trigonal lattices
As for the hexagonal cell, in the conventional trigonal cell the threefold axis
is chosen parallel to c, with a = b, unrestricted c, a = /3 = 90°, and y = 120".
Centred cells are easily amenable to the conventional P trigonal cell.
Because of the presence of a threefold axis some lattices can exist which
may be described via a P cell of rhombohedral shape, with unit vectors a R ,
bR, CR such that aR= bR = cR, aR= PR = YR, and the threefold axis along
the UR + bR + CR direction (see Fig. 1.17). Such lattices may also be
described by three triple hexagonal cells with basis vectors UH, bH, CH
defined according to[61
These hexagonal cells are said to be in obverse setting. Three further triple
hexagonal cells, said to be in reverse setting, can be obtained by changing
a H and bH to -aH and -bH. The hexagonal cells in obverse setting have
centring points (see again Fig. 1.17)) at
(O,O, O), I ,I ,I , (113,213,213)
while for reverse setting centring points are at
It is worth noting that a rhombohedral description of a hexagonal P lattice

is always possible. Six triple rhombohedral cells with basis vectors a;, bk,
Fig. 1.17. Rhombohedra1 lattice. The basis of the

rhombohedral cell is labelled a, b,, c,, the
basis of the hexagonal centred cell is labelled
a, b,, c, (numerical fractions are calculated in
terms of the c, axis). (a) Obverse setting; (b) the
same figure as in (a) projected along c,.
ck can be obtained from a H , bH, CH by choosing:

U ~ = U H + C H , bk=b,+ cH, ck=-(aH+bH)+cH
a k = -aH + CH, bk = -bH + CH, ck = a H+ bH + cH
and cyclic permutations of a;, bk, ck. Each triple rhomobohedral cell will
have centring points at (O,0, O), (113,1/3,1/3), (213,213,213).
In conclusion, some trigonal lattices may be described by a hexagonal P
cell, others by a triple hexagonal cell. In the first case the nodes lying on the
different planes normal to the threefold axis will lie exactly one on top of
the other, in the second case lattice planes are translated one with respect to
the other in such a way that the nth plane will superpose on the (n + 3)th
plane (this explains why a rhombohedral lattice is not compatible with a
sixfold axis).
When, for crystals belonging to the hexagonal or trigonal systems, a
hexagonal cell is chosen, then on the plane defined by a and b there will be
a third axis equivalent to them. The family of planes (hkl) (see Fig. 1.18)
divides the positive side of a in h parts and the positive side of b in k parts.
If the third axis (say d) on the (a, b) plane is divided in i parts we can
a introduce an extra index in the symbol of the family, i.e. (hkil). From the
same figure it can be seen that the negative side of d is divided in h k +
Fig. 1.18. Intersections of the set of
crystallographic planes ( h k l ) with the three
parts, and then i = -(h +k). For instance (1 2 -3 5), (3 -5 2 I),
symmetry-equivalent a, b, daxes in trigonal and
(-2 0 2 3) represent three plane families in the new notation. The
hexagonal systems. four-index symbol is useful to display the symmetry, since (hkil), (kihl), and
(ihkl) are symmetry equivalent planes.
Also, lattice directions can be indicated by the four-index notation.
Following pp. 7-8, a direction in the (a, b) plane is defined by a vector
+
(P - 0)= ma nb. If we introduce the third axis d in the plane, we can
+
write (P - 0)= ma + nb Od. Since a decrease (or increase) of the three
coordinates by the same amount j does not change the point P, this may be
represented by the coordinates: u = m - j, v = n - j, i = -j.
+
If we choose j = (m n)/3, then u = (2m - n)/3, v = (2n - m)/3, i =
-(m + n)/3. In conclusion the direction [mnw] may be represented in the
new notation as [uviw], with i = -(u +
v). On the contrary, if a direction is
already represented in the four-index notation [uviw], to pass to the
three-index one, -i should be added to the first three indices in order to
bring to zero the third index, i.e. [u - i v - i w].
A last remark concerns the point symmetry of a lattice. There are seven
three-dimensional lattice point groups, they are called holohedries and are
listed in Table 1.6 (note that 3m is the point symmetry of the rhombohedral
lattice). In two dmensions four holohedries exist: 2, 2mm, 4mm, 6mm.
The 14 Bravais lattices are illustrated in Fig. 1.19 by means of their
conventional unit cells (see Appendix 1.C for a different type of cell). A
detailed description of the metric properties of crystal lattices will be given
in Chapter 2.
The space groups

A crystallographic space group is the set of geometrical symmetry opera-
tions that take a three-dimensional periodic object (say a crystal) into itself.
Triclinic
Cubic
Trigonal
Fig. 1.19. The 14 three-dimensional Bravais

Hexagonal lattices.
The total number of crystallographic space groups is 230. They were first
derived at the end of the last century by the mathematicians Fedorov (1890)
and Schoenflies (1891) and are listed in Table 1.9.
In Fedorov's mathematical treatment each space group is represented by
a set of three equations: such an approach enabled Fedorov to list all the
space groups (he rejected, however, five space groups as impossible: Fdd2,
Fddd, 143d, P4,32, P4132). The Schoenflies approach was most practical and
is described briefly in the following.
On pp. 11-16 we saw that 32 combinations of either simple rotation or
inversion axes are compatible with the periodic nature of crystals. By
combining the 32 point groups with the 14 Bravais lattices (i.e. P, I, F, . . .)
one obtains only 73 (symmorphic) space groups. The others may be
obtained by introducing a further variation: the proper or improper
symmetry axes are replaced by screw axes of the same order and mirror
planes by glide planes. Note, however, that when such combinations have
more than one axis, the restriction that all symmetry elements must
intersect in a point no longer applies (cf. Appendix l.B). As a consequence
of the presence of symmetry elements, several symmetry-equivalent objects
will coexist within the unit cell. We will call the smallest part of the unit cell
which will generate the whole cell when applying to it the symmetry
Table 1.9. The 230 three-dimensional space groups arranged by crystal systems and
point groups. Space groups (and enantiomorphous pairs) that are uniquely deter-
minable from the symmetry of the diffraction pattern and from systematic absences (see
p. 159) are shown in bold-type. Point groups without inversion centres or mirror planes
are emphasized by boxes
Crystal Point Space

system group groups
Triclinic [i3 p1
i P1
Monoclinic P2, P2,, C2

m Pm, PC, Cm, Cc
2/m P2/m, P2,/m, C2/m, P2/c, P2,/c, C2/c
Orthorhombic 12221 P222, P222,, P2,2,2, P2,2,2,, C222,, C222, F222, 1222,
12,2121
mm2 Pmm2, PmcP,, Pcc2, PmaP,, PcaS,, PncZ,, PmnZ,, Pba2,
Pna2,, Pnn2, Cmm2, Cmc2,, Ccc2, Amm2, Abm2, Ama2,
Aba2, Fmm2, Fdd2,lmm2, lba2, h a 2
mmm Pmmm,Pnnn,Pccm,Pban,Pmma,Pnna,Pmna,Pcca,
Pbam, Pccn, Pbcm, Pnnm, Pmmn, Pbcn, Pbca, Pnma,
Cmcm, Cmca, Cmmm, Cccm, Cmma, Ccca, Fmmm,
Fddd, Immm, Ibam, Ibca, lmma
Tetragonal p4, p41, p4,, p4, 14, 14,

4 P4. 14
Cubic (231 P23, F23, 123, P2,3,_12,3 -

~ m 3~, dFm3,, F a , lm3, ~ a 3 I&
,
PG2, Pa&?, F$32, Y . 2 , 1432, P?,32, P4,32, l4?32
P43-m, F43m, 143m, P43n, F43c, 1Gd
m3m Pm_3m, Pn3n, PrnBn, Pn3m, Fm3m, Fm&, F b m , F&c,
lm3m, la3d
operations an asymmetric unit. The asymmetric unit is not usually uniquely

defined and can be chosen with some degree of freedom. It is nevertheless
obvious that when rotation or inversion axes are present, they must lie at
the borders of the asymmetric unit.
According to the international (Hermann-Mauguin) notation, the space-

group symbol consists of a letter indicating the centring type of the
conventional cell, followed by a set of characters indicating the symmetry
elements. Such a set is organized according to the following rules:
1. For triclinic groups: no symmetry directions are needed. Only two space
groups exist: PI and PI.
2. For monoclinic groups: only one symbol is needed, giving the nature of
the unique dyad axis (proper and/or inversion). Two settings are used:
y-axis unique, z-axis unique.
3. For orthorhombic groups: dyads (proper and/or of inversion) are given
along x, y, and z axis in the order. Thus Pca2, means: primitive cell,
glide plane of type c normal to x-axis, glide plane of type a normal to the
y-axis, twofold screw axis along z .
4. For tetragonal groups: first the tetrad (proper and/or of inversion) axis
along z is specified, then the dyad (proper and/or of inversion) along x is
given, and after that the dyad along [I101 is specified. For example,
P4,lnbc denotes a space group with primitive cell, a 4 sub 2 screw axis
along z to which a diagonal glide plane is perpendicular, an axial glide
plane b normal to the x axis, an axial glide plane c normal to [110].
Because of the tetragonal symmetry, there is no need to specify
symmetry along the y-axis.
5. For trigonal and hexagonal groups: the triad or hexad (proper and/or of
inversion) along the z-axis is first given, then the dyad (proper and/or of
inversion) along x and after that the dyad (proper and/or of inversion)
along [1?0] is specified. For example, P6,mc has primitive cell, a sixfold
screw axis 6 sub 3 along z, a reflection plane normal to x and an axial
glide plane c normal to [ ~ I o ] .
6. For cubic groups: dyads or tetrads (proper and/or of inversion) along x ,
followed by triads (proper and/or of inversion) along [ I l l ] and dyads
(proper and/or of inversion) along [110].
We note that:
1. The combination of the Bravais lattices with symmetry elements with no
translational components yields the 73 so-called symmorphic space
groups. Examples are: P222, Cmm2, F23, etc.
2. The 230 space groups include 11 enantiomorphous pairs: P3, (P3,),
P3,12 (P3212), P3,21 (P3,21), P41 (P43), P4J2 (P4322), P4,&2 (P4&?,2),
P6i (P65), P6, (P64), P6,22 (P6522), P6222 (P6422), P4,32 (P4,32). The
( + ) isomer of an optically active molecule crystallizes in one of the two
enantiomorphous space groups, the ( - ) isomer will crystallize in the
other.
3. Biological molecules are enantiomorphous and will then crystallize in

space groups with no inversion centres or mirror planes; there are 65
groups of this type (see Table 1.9).
4. The point group to which the space group belongs is easily obtained from
the space-group symbol by omitting the lattice symbol and by replacing
the screw axes and the glide planes with their corresponding symmorphic
symmetry elements. For instance, the space groups P4Jmmc, P4/ncc,
14,lacd, all belong to the point group 4lmmm.
5. The frequency of the different space groups is not uniform. Organic
compounds tend to crystallize in the space groups that permit close
packing of triaxial ellipsoids.[81According to this view, rotation axes and
reflection planes can be considered as rigid scaffolding which make more
difficult the comfortable accommodation of molecules, while screw axes
and glide planes, when present, make it easier because they shift the
molecules away from each other.
Mighell and Rodgers [9] examined 21 051 organic compounds of known
crystal structure; 95% of them had a symmetry not higher than orthorhom-
bic. In particular 35% belonged to the space group P2,/c, 13.3% to PI,
12.4% to P2,2,2,, 7.6% to P2, and 6.9% to C21c. A more recent study by
~ i l s o n , [ ' ~based
] on a survey of the 54599 substances stored in the
Cambridge Structural Database (in January 1987), confirmed Mighell and
Rodgers' results and suggested a possible model to estimate the number Nsg
of structures in each space group of a given crystal class:
Nsg = Acc exp { -BccE21sg - Ccclmls,)
where A,, is the total number of structures in the crystal class, [2],, is the
number of twofold axes, [m],, the number of reflexion planes in the cell, B,,
and Cc, are parameters characteristic of the crystal class in question. The
same results cannot be applied to inorganic compounds, where ionic bonds
are usually present. Indeed most of the 11641 inorganic compounds
considered by Mighell and Rodgers crystallize in space groups with
orthorhombic or higher symmetry. In order of decreasing frequency we
have: Fm3m, Fd3m, P6Jmmc, P2,/c, ~ m 3 m ~, 3 m C2/m,, C2/c, . . . .
The standard compilation of the plane and of the three-dimensional space
groups is contained in volume A of the International Tables for Crystallog-
raphy. For each space groups the Tables include (see Figs 1.20 and 1.21).
1. At the first line: the short international (Hermann-Mauguin) and the
Schoenflies symbols for the space groups, the point group symbol, the
crystal system.
2. At the second line: the sequential number of the plane or space group,
the full international (Hermann-Mauguin) symbol, the Patterson symmetry
(see Chapter 5, p. 327). Short and full symbols differ only for the
monoclinic space groups and for space groups with point group mmm,
4/mmm, 3m, 6/mmm, m3, m3m. While in the short symbols symmetry
planes are suppressed as much as possible, in the full symbols axes and
planes are listed for each direction.
3. Two types of space group diagrams (as orthogonal projections along a

cell axis) are given: one shows the position of a set of symmetrically
equivalent points, the other illustrates the arrangement of the symmetry
elements. Close to the graphical symbols of a symmetry plane or axis
parallel to the projection plane the 'height' h (as a fraction of the shortest
lattice translation normal to the projection plane) is printed. If h = 0 the
height is omitted. Symmetry elements at h also occur at height h + 112.
4. Information is given about: setting (if necessary), origin, asymmetric

unit, symmetry operations, symmetry generators (see Appendix l.E)
selected to generate all symmetrical equivalent points described in block
'Positions'. The origin of the cell for centrosymmetric space groups is
usually chosen on an inversion centre. A second description is given if
points of high site symmetry not coincident with the inversion centre occur.
For example, for ~ n 3 ntwo descriptions are available, the first with origin at
432, and the second with origin at 3. For non-centrosymmetric space groups
the origin is chosen at a point of highest symmetry (e.g. the origin for ~ 4 2 c
is chosen at 4lc) or at a point which is conveniently placed with respect to
the symmetry elements. For example, on the screw axis in P2,, on the glide
plane in PC, at la2, in P ~ a 2 at
~ ,a point which is surrounded symmetrically
by the three 2, axis in P2,2,2,.
5. The block positions (called also Wyckoff positions) contains the
general position (a set of symmetrically equivalent points, each point of
which is left invariant only by application of an identity operation) and a list
of special positions (a set of symmetrically equivalent points is in special
position if each point is left invariant by at least two symmetry operations of
the space group). The first three block columns give information about
multiplicity (number of equivalent points per unit cell), Wyckoff letter (a
code scheme starting with a at the bottom position and continuing upwards
in alphabetical order), site symmetry (the group of symmetry operations
which leaves invariant the site). The symbol adoptedr9]for describing the
site symmetry displays the same sequence of symmetry directions as the
space group symbol. A dot marks those directions which do not contribute
any element to the site symmetry. To each Wyckoff position a reflection
condition, limiting possible reflections, may be associated. The condition
may be general (it is obeyed irrespective of which Wyckoff positions are
occupied by atoms (see Chapter 3, p. 159) or special (it limits the
contribution to the structure factor of the atoms located at that Wyckoff
position).
6. Symmetry of special projections. Three orthogonal projections for

each space group are listed: for each of them the projection direction, the
Hermann-Mauguin symbol of the resulting plane group, and the relation
between the basis vectors of the plane group and the basis vectors of the
space group, are given, together with the location of the plane group with
respect to the unit cell of the space group.
7. Information about maximal subgroups and minimal supergroups (see
Appendix l.E) is given.
In Figs. 1.20 and 1.21 descriptions of the space groups Pbcn and P4222 are
respectively given as compiled in the International Tables for Crystallog-
raphy. In order to obtain space group diagrams the reader should perform
the following operations:
1. Some or all the symmetry elements are traced as indicated in the
space-group symbol. This is often a trivial task, but in certain cases
special care must be taken. For example, the three twofold screw axes do
not intersect each other in P2,2,2,, but two of them do in P2,2,2 (see
Appendix 1.B) .
Pbcn Orthorhombic
No. 60 P 2,lb 2 / c 2 , / n Patterson symmetry P m m m
Origin at i on I c 1
Asymmetric unit OSxli; 0 OlzS:
Symmetry operations
Fig. 1.20. Representation of the group Pbcn (as

inlnternational Tables for Crystallography).
CONTINUED No. 60 Pbcn
Generatorsselected ( I ) ; ( 1 0 0 ) ; ( 0 1 0 ) ; t(0.0,)); (2); (3); (5)

Positions
Mulliplicity. Coordinates Reflection conditions
Wyckoff letter.
Site Spmetry
General :
8 d 1 (1) x (2) x + ~ , J + : , z + ~ (3) f ,p,i+i ( 4 ) x+:,y+i,r Okl : k = 2 n
(5)x ' , j , ~ ( 6 ) x+:,y+;.5+: ( 7 ) x.J',z++ ( 8 ) x'+:,y+i,z h01: 1 = 2 n
hkO: h +k = 2n
hOO: h = 2n
OkO: k = 2n
001: 1 = 2 n
Special: as above, plus
hkl: h+k=2n
hkl : h + k , 1 = 2 n
hkl : h + k , l = 2 n
Symmetry of special projections

Along [001] c 2 m m Along [ 1001 p 2g m Along [010] p 2g m
al=a bl=b al=ib b'=c a'= bl=a
Origin at 0,O.z Origin at x.O.0 Origin at O,y,O
Maximal non-isomorphic subgroups

1 [2]P2,22,(P2,2,2) 1;2;3;4
[ 2 ] P1 1 2 , / n ( P 2 , l c ) 1; 2; 5 ; 6
[2]P12/cI(P2/c) 1;3;5;7
[2]P2,/b11(P2,/c) 1;4;5;8
[ 2 ] P b c 2 ,( P c a 2 , ) 1 ; 2; 7 ; 8
[ 2 ] P b 2 n( P n c 2 ) 1 ; 3; 6 ; 8
[2]P2,cn(Pna2,) 1;4;6;7
IIa none
IIb none
Maximal isomorphic subgroups of lowest index

IIc [ 3 ] P b c n ( a 1 =3 a ) ; [ 3 ] P b c n ( b 1 =3 b ) ; [ 3 ] P b c n ( c 1 =3 c )
Minimal non-isomorphic supergroups

I none
I1 [2]Abma(Cmca);[2]Bbab(Ccca);[2]Cmcm;[2]1bam;[2]Pbcb(2a1= a ) ( P c c a ) ;
[ 2 ] P r n c a ( 2 b 1 =b ) ( P b c m ) ;[ 2 ] P b mn(2c1= c ) ( P m n a )
Tet rag on al
No. 93 P4222 Patterson symmetry P 4/m m m
Origin at 2 2 2 at 422 1
Asymmetric unit Olxli; OIySI; 0Izl$
Symmetry operations
Fig. 1.21. Representation of the group P4,22 (as

in International Tables for Crystallography).
2. Once conveniently located, the symmetry operators are applied to a
point P in order to obtain the symmetry equivalent points P', P", . . . . If
P', P , . . . , fall outside the unit cell, they should be moved inside by
means of appropriate lattice translations. The first type of diagram is so
obtained.
3. New symmetry elements are then placed in the unit cell so producing the
second type of diagram. .
Some space group diagrams are collected in Fig. 1.22. Two simple crystal
structures are shown in Figs 1.23 and 1.24: symmetry elements are also
located for convenience.
The plane and line groups

There are 17 plane groups, which are listed in Table 1.10. In the symbol g
stays for a glide plane. Any space group in projection will conform to one of
these plane groups. There are two line groups: p l and pm.
A periodic decoration of the plane according to the 17 plane groups is
shown in Fig. 1.25.
CONTINUED No. 93
Generators selected ( I ) ; t (I ,O,O); t (0, I ,0); t (O,O, I); (2); (3); (5)
Positions
Mulliplicily. Coordinates Reflection conditions
Wyckofl kllcr.
S ~ l esymmetry
General:
001: I = 2 n
Special: as above, plus

4 o ..2 x,x,+ f,f,i f,x,+ x,f,+ Okl: I = 2 n
4 n ..2 x,x,$ f.1,: f,x,: x.f,? Okl: I =2n
4 1 .2. x,O,+ f,O.f O.x,O O,f,O hhl : I = 2n
4 h 2.. , , ,L, z
1 f ,:,z+: +,i,T f,$,f+f hkl : 1 = 2 n
2 f 2.22 +,+,+ it+,+ hkl : I = 2 n

2 e 2.22 O,O,+ O,O,+ hkl : I = 2n
hkl : h + k + l = 2 n
2 b 222. f,f,O f,f,i hkl : I = 2 n
Symmetry of special projections

Along [OOI] p 4 m m Along [I001 p2n1tn Along [I101 p 2 m m
a'=a bl=b a'=b bl=c a'= + ( - a f b ) bl=c
Origin at 0,O.z Origin at x,O,O Origin at x,x,$
Maximal non-isomorphic subgroups
I (2]P4,1 1 ( P 4 2 ) 1;2;3;4
[2]P221(P222) l;2;5;6
[ 2 ] P 2 1 2 ( C 2 2 2 ) 1;2;7;8
IIa none
IIb [ 2 ] P 4 , 2 2 ( c 1 = 2 c ) ; [ 2 ] P 4 , 2 2 ( c ' 2= c ) ; ( 2 ] C 4 , 2 2 , ( a 1 = 2 a , b 1 = 2 b ) ( P 4 , 2 , 2 ) ;
[ 2 ] F 4 , 2 2 ( a 1 =2a,b1= 2b,c1= 2 c ) ( 1 4 , 2 2 )
Maximal isomorphic subgroups of lowest index
IIc [ 3 ] P 4 , 2 2 ( c 1 = 3 c ) ; [ 2 ] C 4 , 2 2 (a1= 2 a , b l = 2 b ) ( P 4 2 2 2 )
Minimal non-isomorphic supergroups
I I21P4Jmmc; [ 2 ] P 4 J t n c m ; [ 2 1 P 4 2 / n b c ;[21P42/nnm;[31P4232
I1 ( 2 ] 1 4 2 2 ; [ 2 ] P 4 2 2 ( 2 c 1 =c)
0- 0-
C2
4 -0 0- -0 0- -
P 2/m
TI-
:! I
0+
- 10 o+ +0 0 -
I
0-
o+ -
-
0-
0+ 0"; -
O+
-1 -
4 -0 0-
+O o+
-0 0-
+0 o+
c
1 1--
------ 0 ---0
+ +
Fig. 1.22. Some space group diagrams.
On the matrix representation of symmetry

operators
A symmetry operation acts on the fractional coordinates x,y, z of a point P
to obtain the coordinates (x', y', 2') of a symmetry-equivalent point P':
The R matrix is the rotational component (proper or improper) of the

symmetry operation. As we shall see in Chapter 2 its elements may be 0 ,
$1, -1 and its determinant is f1. T is the matrix of the translational
component of the operation. A list of all the rotation matrices needed to
conventionally describe the 230 space groups are given in Appendix 1.D.
Symmetry in crystals / 33
Fig. 1.23. A P2,2,2, crystal structure (G. Chiari,

D. Viterbo, A. Gaetani Manfredotti, and C.
Guastini (1975). Cryst. Struct. Commun., 4,561)
and its symmetry elements (hydrogen atoms are
not drawn).
Fig. 1.24. A P2,/c crystal structure (M. Calleri, G.

Ferraris, and D. Viterbo (1966). Acta Cryst., 20,
73) and its symmetry elements (hydrogen atoms
are not drawn). Glide planes are emphasized by
the shading.
Table 1.10. The 17 plane groups
Oblique cell P I , ~2
Rectangular cell pm, pg, a n , ~ 2 m m P, ~ V ,~ 2 9 9c2mm
,
Square cell p4, p4n-m~p4gt1-1
Hexagonal cell p3, p3rn1, p31n-1,p6, p6mrn
When applying the symmetry operator C1= (R,, TI) to a point at the end
of a vector r, we obtain X' = CIX = RIX + TI. If we then apply to r' the
Fig. 1.25. A periodic decoration of the plane
according to the 17 crystallographic plane symmetry operator C2, we obtain
groups (drawing by SYMPATI,a computer
program by L. Loreto and M . Tonetti, pixel, 9, X = C2Xf= R2(RlX + TI) + T2= R2RlX + R2Tl + T2.
9-20; Nov 1990).
Since the symmetry operators form a mathematical group, a third symmetry

operator must be present (see also pp. 11-12),
C3 = C2C1= (R2R1, R2T1+ T2), (1.12)
where R2Rl is the rotational component of C3 and (R2T, +T2) is its
translational component. In particular the operator C2= CC will be present
and in general also the CJ operator. Because of (1.12)
C1 = [RJ, (RJ-' + . . . + R + I)T]. (1.13)
Let us now apply this result to the space group P6,. Once we have defined
the R and T matrices corresponding to an anti-clockwise rototranslation of
60" around z, we obtain all the six points equivalent to a point r by applying
to it the operators CJ with j going from 1 to 6. Obviously C6= I and
C6+j= CJ. For this reason we will say that the 6, operator is of order six
(similarly 2 and m are of order two).
If r is transferred to r' by C = (R, T) there will also be an inverse operator
C-' = (R', TI) which will bring r ' back to r. Since we must have C-'C = I,
because of (1.12) we will also have R'R = I and R'T + T' = 0, and therefore
C-1 = (R-1, -R-~T) (1.14)
where R-l is the inverse matrix of R. In the P6, example, C-l = C5. When
all the operators of the group can be generated from only one operator
(indicated as the generator of the group) we will say that the group is
cyclic.
All symmetry operators of a group can be generated from at most three
generators. For instance, the generators of the space group P6,22 are 6, and
one twofold axis. Each of the 12 different operators of the group may be
obtained as C',, j = 1, 2, . . . , 6, say the powers of 61, or as C2, the twofold
axis operator, or as their product. We can then represent the symmetry
operators of P6'22 as the product {C1){C2), where {C) indicates the set
of distinct operators obtained as powers of C. Similarly there are two
generators of the group P222 but three of the group ~ 4 3 mIn . general all the
operations of a space group may be represented by the product {C1){C2){C3).
If only two generators are sufficient, we will set C3 = I, and if only one is
sufficient, then C2 = C3 = I. The list of the generators of all point groups is
given in Appendix 1.E.
So far we have deliberately excluded from our considerations the
translation operations defined by the Bravais lattice type. When we take
them into account, all the space-group operations may be written in "a very
simple way. In fact the set of operations which will transfer a point r in a
given cell into its equivalent points in any cell are:
{ c ~ {c2}
} {c3} (1.15)
+ +
where T = mla m2b m3c is the set of lattice translations.
The theory of symmetry groups will be outlined in Appendix 1.E.
Appendices
1.A The isometric transformations
It is convenient to consider a Cartesian basis (el, e2, e3). Any transforma-
tion which will keep the distances unchanged will be called an isometry or
an isometric mapping or a movement C. It will be a linear transformation,

+ +
in the sense that a point P defined by the positional r x t o r r = xe, ye2
+ +
ze3 is related to a point P', with positional vector r' = x f e l y1e2 zfe3by
the relation
with the extra condition

RR= I or R = R-'.
R indicates the transpose of the matrix R and I is the identity matrix.
We note that X and X' are the matrices of the components of the vectors
r and r' respectively, while T is the matrix of the components of the
+ +
translation vector t = Tlel T,e2 T3e3.
A movement, leaving the distances unchanged, will also maintain the
angles fixed in absolute value. Since the determinant of the product of two
matrices is equal to the product of the two determinants, from (1.A.2) we
have (det R)'= 1, and then det R = f1. We will refer to direct or opposite
movements and to direct or opposite congruence relating an object and its
transform, depending on whether det R is +1 or -1.
Direct movements
Let us separate (l.A.l) into two movements:
X1=X0+T
Xo = RX.
(1.A.3) adds to each position vector a fixed vector and corresponds
therefore to a translation movement. (1.A.4) leaves the origin point
invariant. In order to find the other points left invariant we have to set
Xo = X and obtain
(1.A.5a) will have solutions for X#O only if det (R - I) = 0. Since det (R - I) =
det (R - RR) = det [(I - R)RJ = det (I - R) det R = det (I - R) = -det (R - I),
then this condition is satisfied. Therefore one of the three equations
represented by (1.A.4) must be a linear combination of the other two. The
two independent equations will define a line, which is the locus of the
invariants points; the movement described by (1.A.4) is therefore a
rotation. In conclusion, a direct movement can be considered as the
combination (or, more properly, the product) of a translation with a
rotation around an axis.
If in eqn (1.A.1) is R = I then the movement is a pure translation, if T = 0
the movement is a pure rotation. When the translation is parallel to the
rotation axis the movement will be indicated as rototranslation. An
example of direct movement is the transformation undergone by the points
of a rigid body when it is moved. Another example is the anti-clockwise
rotation around the z axis of an angle 8; this will move r(x, y, z ) into
r1(x', y', z') through the transformation

xl=xcosO-ysinO
y1=xsin8+ycos0
which in matrix notation becomes X' = RX, with
(l.A.5b)
R-' can be obtained by substituting 8 with -8 and it can be immediately

seen that, in agreement with (1.A.2), R = R-l.
We will now show that any direct movement can be carried out by means
of a translation or a rotation or a rototranslation. Let us suppose that an
isometric transformation relates the three non-collinear points A, B, C with
the points A', B', C' respectively. If a is the translation bringing A on A',
then, if also B and C superpose on B' and C', the movement is a
translation; if not, then the complete superposition can be achieved by a
rotation R around an axis 1 passing through A'. If 1 is perpendicular to a,
the movement resulting from the combination of the rotation and transla-
tion operations is still a pure rotation around an axis parallel to 1 (see
Appendix l.B). If a is not perpendicular to 1, then it can be decomposed
into two translational components a1 and a2, one perpendicular and the
other parallel to 1. The product of R times al is a pure rotation around an
axis parallel to 1, which, when composed with a2, results in a rototranslation
movement.
Opposite movements
An opposite movement can be obtained from a direct one by changing the
sign to one or three rows of the R matrix. For instance, when changing the
sign of the third row, we substitute the vector ( x ' , y', 2') with (x', y', -zl),
i.e. the point P' with its symmetry related with respect to a plane at z = 0.
This operation is called a reflection with respect to the plane at z = 0.
Changing the signs of all three rows of the R matrix implies the substitution
of the vector (x', y', z') with (-x', -y ', -2')) i.e. of the point P with its
symmetry related with respect to the origin of the coordinate system. This
operation is called inversion with respect to a point.
We may conclude that each direct movement, followed by a reflection
with respect to a plane or by an inversion with respect to a point yields an
opposite movement. On the other hand an opposite movement may be
obtained as the product of a direct movement by a reflection with respect to
a plane or by an inversion with respect to a point.
1.B. Some combinations of movements

Only those combinations of movements explicitly mentioned in this book
will be considered (for further insight, the reader is referred to the splendid
book by Lockwood and ~acMillan[~I). The stated laws may be interpreted
Fig. 1.B.1. (a) Composition of two reflections;

(b) composition of two rotations about parallel
axes; (c) composition of two rotations, the first
through cuand the second through -a,about
parallel axes; (d) composition of two rotations
about axes passing through a point.
in terms of combinations of symmetry operations if all the space is invariant

with respect to the movements.
1. Composition of two reflections. In Fig. l.B.l(a) the two reflection
planes m, and m, are at a dihedral angle a and intersect along a line, the
trace of which is in 0 . The image of OQ with respect to m, is OQ1 and the
image of OQ, with respect to m, is OQ,. It is possible to superpose OQ to
OQ, by a rotation of 2 a around the axis through 0 . We may conclude that
the product of the two reflections is a rotation of 2 a around 0 ; in symbols
mlm2 = R. The product is not commutative (m,m, # m,m,): in fact by first
reflecting OQ with respect to m, and then reflecting the image with respect
to m, we obtain a 2 a rotation in the opposite direction.
2. The Silvester theorem relative to three parallel rotation axes. The traces
of the three axes on the plane of Fig. l.B.l(b) are A, B, C. The Silvester
theorem states that consecutive anti-clockwise rotations of 2a, 2P, 2y
around A, B, C respectively produce the identity. In fact, because of point
1, the rotations are equivalent to reflection operations with respect to the
three pairs of planes AC-AB, BA-BC, CB-CA respectively and all these
+ +
reflections cancel each other out. Since a P y = n, two successive
rotations of 2 a and 2P around A and B respectively, will be equivalent to a
+
rotation of +(2a 2P) around C. When a = -P, the third axis goes to
infinity and the resulting movement is a translation.
In Fig. l.B.l(c) AC moves to AB by a rotation of a around A; by a -a

rotation around B BA goes to BD. The resulting movement brings AC to
BD and can be achieved by a translation of DA = 2AB sin (a12) perpen-
dicular to the direction of the rotation axes and at an angle of ( n - a)/2
with respect to AB. In symbols: R,RI_,=T. We can then deduce the
following point 3.
3. A rotation and a translation perpendicular to the rotation axis combine
in a resulting rotation movement around an axis parallel to the original axis.
4. The Silvester theorem relative to three rotation axes passing through a
point. In Fig. l.B.l(d) ABC is a spherical triangle with angles a , P, y in A,
B, C respectively. If A, B, C are in a clockwise order, rotations of 2a, 2P,
2y around A, B, C leave the figure unchanged. In fact, because of point 1,
the three rotations correspond to the products of the reflections with respect
to the pairs of planes AOC-AOB, AOB-BOC, BOC-COA respectively,
and these reflections cancel each other out. Since a + p + y > n, then
consecutive rotations of 2 a around A and of 2P around B are equivalent to
+
a rotation of 2y around C, with 2y # (2a 2P).
5. Coexistence of rotation axes passing through a point: the Euler
theorem. We will study this problem using the Silvester theorem treated in
point 4. In Fig. l.B.l(d) let OA and OB be two symmetry rotation axes of
order m and n respectively. The angles a and P are chosen in such a way
that 2 a = 2n/m and 2P = 2nln. Because of the Silvester theorem, anti-
clockwise rotations of 2 a and 2P around OA and OB are equivalent to a 2y
anti-clockwise rotation around OC. C is therefore a symmetry axis of order
p = 2nI2y = nly. The angles of the spherical triangle are then nlm, n l n
and nlp. Since the sum of the angles must be greater than n, the inequality
+ +
l l m l l n l l p > 1 follows. The possible solutions of this inequality are: 1,
2, 2 with 1 integer >1; 2, 3, 3; 4, 3, 2; 5, 3, 2. We can now consider the
different solutions, keeping in mind that the surface of a sphere of radius r is
+ +
4nr2 and that of a spherical triangle is ( a P y - n)r2:
(a) Solution 1, 2, 2. a , p, and y are equal to n12, n12, n l l respectively and
A and B may be chosen on an equatorial circle with C as a pole. The
binary axes are therefore always at 90' with respect to the 1 axis. Values
of 1 different from 2, 3, 4, 6 correspond to non-crystallographic groups
which occur as possible symmetries of molecules or as approximate
local site symmetries in crystals.
(b) Solution 2, 3, 3. a , p, and y are equal to n12, n13, n13 respectively
and the area of the spherical triangle is nr2/6. On the sphere there will
be 24 such triangles. The 24 n / 2 angles meet four at a time at six
vertices and the 48 n13 angles meet six at a time at eight vertices. This
implies the presence of three twofold axes and of four threefold axes.
The three twofold axes will be perpendicular one to the other and can
be assumed as the axes of a reference system. The four threefold axes
run from the centre to the points (1,1, I), (1, - 1, -I), (- 1,1, -I),
(-1, -1,l). The group of rotations is consistent with the symmetry of a
tetrahedron.
(c) Solution 4, 3, 2. a , p, and y are equal to n/2, n13, n14 respectively
and the area of the spherical triangle is nr2/12. On the sphere there will
Fig. 1.8.2. Stereographic projections of the 32

crystallographic point groups.
be 48 such triangles. The 48 n / 2 angles meet four at a time at twelve

vertices, the 48 n / 3 angles meet six at a time at eight vertices and the 48
n / 4 angles meet eight at a time at six vertices. A total of six twofold
axes, of four threefold axes, and of three fourfold axes will be present.
The three fourfold axes-will be perpendicular one to the other and can
be assumed as the axes of a reference system. The threefold axes are
located along the [I, 1,1], [I, - 1, -11, [-I, 1, -11, [-I, -1, 11 direc-
tions, while the twofold axes are on the bisecting lines of the angles
between the fourfold axes. The group of rotations is consistent with the
symmetry of a cube and of an octahedron (see Fig. 1.10).
The mutual disposition of the symmetry elements in the 32 crystal-
lographic point groups is illustrated in Fig. 1.B.2, where the so called
'stereographic projections' are shown. The c axis is normal to the plane,
the a axis points down the page, and the b axis runs horizontally in the
page from left ro right. A stereographic projection is defined as follows
(see Fig. l.B.3(a)). A unit sphere is described around the crystal in C.
A point P (terminal of some symmetry axis) in the $ 2 hemisphere is
defined in the ( x , y ) plane as intersection P' of that plane with the line
connecting the point with the south pole of the unit sphere. If the point
to be projected is in the -2 hemisphere then the north pole is used.
In Fig. l.B.3(b) parts of the stereographic projections for m3m are
magnified in order to make clearer the statements made in the text.
(d) Solution 5, 3, 2. This solution, which is compatible with the symmetry
poie
of the regular icosahedron (20 faces, 12 vertices) and its dual, the
regular pentagon-dodecahedron, (12 faces, 20 vertices), but not with
the periodicity property of crystals, will not be examined.
It is however of particular importance in Crystallography as sym-
metry of viruses molecules and in quasi-crystals.
6. Composition of two glide planes. In Fig. 1.B.4 let S and S' be the
traces of two glide planes forming an angle a and 0 be the trace of their
intersection line. The translational components OA and OB are chosen to
lie on the plane of the drawing and Q is the meeting point of the axes of the
OA and OB segments. X, Y, and Q' are the reflection images of Q with
respect to S, S', and to the point 0, respectively. The product S'S moves Q
to Q' and then back to Q. Since S'S is a direct movement it leaves Q
unchanged and corresponds to a rotation around an axis normal to the plane
of the figure and passing through Q. Since S'S moves first A to 0 and then
to B, the rotation angle AQB = 2a. Note that the two glides are equivalent
to a rotation around an axis not passing along the intersection line of S
and S'.
7. Composition of two twofold axes, with and without translational
component. From point 5 we know that the coexistence of two orthogonal
twofold axes passing by 0 , implies a third binary axis perpendicular to them
and also passing through 0 (see Fig. l.B.5(a)). The reader can easily verify
the following conclusions:
(a) if one of the two axes is 2, (Fig. l.BS(b)), then another 2 axis, at 114
from 0 and intersecting orthogonally the screw axis, will exist;
(b) if two 21 intersect in 0 (Fig. l.B.5(c)), then another 2 axis perpendicu-
lar to them and passing at (114,114) from 0 will be present;
(c) if a pair of mutually perpendicular 2 axes is separated by 114 of a period
(Fig. l.BS(d)), then a 2, axis orthogonally intersecting both axes will
exist; Fig. 1.8.3. (a) Geometry of the stereographic
projection. (b) Angular values occurring in m3m
(d) if a 2 and a 2, axis are separated by 114 of a period (Fig. l.B.5(e)) there stereographic projection.
will then be a new 2, axis normal to both of them and intersecting the
first 2, axis at 114 froin the 2 axis;
(e) if two orthogonal screws are separated by 114 of a period (Fig.
l.B.5(f)), then a third screw axis normal to them and passing at
(114,114) from them will be present.
1.C. Wigner-Seitz cells

The 14 Bravais lattices are compatible with cells which are different from
those conventionally associated to them. A conventional cell is a paral-
lelepiped: as such it may be considered as a particular type of polyhedron.
There are several families of polyhedra with which we can fill up the space Fig. 1.8.4. Composition of two glide planes.
Fig. 1.8.5. Various arrangements of three

orthogonal binary axes. The length of the
graphical symbol for screw axes i n the plane of
the page corresponds t o half repetition period.
by translation. A very important type is the one obtained through the

Dirichlet construction. Each lattice point is connected with a line to its
nearest neighbours. We then trace through the mid-points of the segments
the planes perpendicular to them. These intersecting planes will delimit a
region of the space which is called called the Dirichlet region or Wigner-
Seitz cell. An example in two dimensions is given in Fig. l.C.l(a) and two
three-dimensional examples are illustrated in Fig. 1.C. l(b, c) . The Wigner-
Seitz cell is always primitive and coincides with the Bravais cell if this is
rectangular and primitive. A construction identical to the Wigner-Seitz cell
delimits in the reciprocal space (cf. Chapters 2 and 3) a cell conventionally
known as the first Brillouin zone. There will be 14 first Brillouin zones
corresponding to the 14 Bravais lattices. We recall here that to a lattice I in
direct space corresponds an F lattice in reciprocal space and vice versa (see
Appendix 2.D); then, the first Brillouin zone of an I lattice will look like a
Wigner-Seitz cell of an F lattice and vice versa. The Brillouin zones are
very important in the study of lattice dynamics and in electronic band
theory.
Fig. 1.C.1. Examples of Wigner-Seitz cells.

1.D. The space-group matrices

All the rotation matrices R needed to conventionally describe all 230 space
groups will be listed. They operate according to the relation X' = RX. The
matrices are grouped following the directions along which they operate, and
for the hexagonal and trigonal systems they are preceded by the letter H.
They may be constructed using the following practical criterion: in the first,
second, and third columns are the coordinates of the points symmetry
related to points (1,0, O), (0,1, O), and (0,0,1) respectively.
The determinant of a matrix will have a +1 or a -1 value depending on
whether the represented operation is of rotation or of inversion type. This
number, together with the trace (sum of the diagonal elements) of a matrix,
are characteristic of the symmetry element:
element 1234612346,
trace 3 i 0 1 2 3 1 0 1 2 ,
determinant l l l l l i i i i i .
The matrix of the fi operator may be simply obtained from that of the n
operator by multiplying by the matrix corresponding to the i operator:
i = ( i -1o -1
0
:).
This corresponds to changing the sign of all the elements of the original
matrix. Therefore in the following list we will not give all the 64 matrices
necessary to describe the space groups, but only the 32 matrices cor-
responding to proper symmetry elements.
Direction [0 0 01
Direction [I 0 01
2 (~ 0 0 ° 1) ( 1 )
0H 2 = 0 1 0 4
( l o O); ( l o
~ 0 0 14 3 = O 0 1 .
0)
o o i o o i o I 0' 0 1 0
(yg;)
Direction [0 1 01
2 =( i0 o10 O) ; H 2 =O 1 O1 O) ; 3 =( O
0 O1 i O
); 4 3 = 0 1 0 .
o o i o o i l o o
Direction [0 0 11
i o o 1 1 0 o i o
Direction [ l 1 01
Direction [l 0 11
0 0 1
Direction [0 1 11
Direction [l i 01
Direction [ i 0 11
Direction [0 1 I]
Direction [I 1 11
0 0 1 0 1 0
Direction [I1 11
o i 0 o o i
Direction [l i 11
l o o o i o
Direction [I 1 ?I
3
(
~ 0
0);
0 13
(O O T )
2 ' 1 0 0 .
T o o 0 i 0
Direction [2 1 0]
Direction [ l 2 01.
1.E. Symmetry groups

A group G is a set of elements g,, g2, . . . g,, . . . for which a combination
law is defined, with the following four properties: closure, the combination
of two elements of the group is an element of the group gigj =gk;
associativity, the associative law (gigj)gk= gi(gjgk) is valid; identity, there
is only one element e in the group such that eg = ge = g; inversion, each
element g in the group has one and only one inverse element g-' such that
g-lg = gg-l = e.
Examples of groups are:

(1) the set of all integer numbers (positive, negative, and zero), when the
combination law is the sum. In this case e = 0, g-' = -g;
(2) the rational numbers, excluding zero, when the law is the product:
e = 1, g - l = l/g;
(3) the set of all n x n son-singular matrices under the product law: e is the
diagonal matrix with aii = 1;
+ +
(4) the set of all lattice vectors r,,,,, = ua vb wc with u, v, w positive,
negative, or null integers, when the combination law is the vector sum:
then e = g,,,,, g-l = -g.
The number of different elements of a group is called the order of the
group and can be finite or infinite. If a group also possesses the
commutative property gigj=gigi for any i and j, then G is said to be
Abelian. With reference to the examples given above, (I), (2), (4) are
infinite and Abelian groups, (3) is infinite and non-Abelian.
If g is an element of G, all powers of g must be contained in G. An
integer n may exist for which
then gn+'= g, gn+2= g2, . . , If n is the smallest integer for which (l.E.l) is
satisfied, there will only be n distinct powers of g. Since gJgn-j= gn-jgi = e,
then gn-j is the inverse of gi. The element g is then said to be of order n and
Table 1.E.1. List of generators for non-cyclic point groups. There are 2 1 proper
generators in all
Point group Generators Point group Generators
2/m 4oro1, 2roio1 Jrn

222 2 ~ 1 0 0 1210101
~ 62rn
rnrn2 2[1001' 210011 622
rnrnrn 2110011 ~ 1 0 1 0 1 ~ 6/m
210011
422 ~ [ ~ o o 3I , 1 i o ~
4/ rn 4100111?[0011
4rn rn ~ [ I O O I ~2 [ i i o 1
42rn T r r o o i ~2 [ i i o 1
4Jrnrnrn ?[looit 4 i i o i
210011
32 3 1 o o i 1Triooi
~
3rn 310011~ 21ioo~
the set of all powers of g is a group of order n:

G I = (g, g 2 , . . . , gn-l, gn = e).
A group, such as (1.E.2), in which all the elements are powers of a single
generating element, is called cyclic. All cyclic groups are Abelian, but the
converse is not true. An example of cyclic group of order n is the set of
rotations around a given axis, which are multiple of an angle a = 2nln.
Any point group can be represented as the product of powers of at most
three elements, which are the generators of the group. In Table l . E . l the
list of the generators of the non-cyclic groups is reported. We also note that
the definition of the generators is not unique. For instance in the class 222
we may chose the generators 2[,ooland 2[olo1,or 2,,,] and 2[00,1,or 210101and
~[ooI].
When the physical properties of the group elements are not specified the
group is said to be abstract. From a mathematical point of view all its
properties are determined by its multiplication table. For a group of n
elements this table has the form:
We note that
1. Each element appears once and only once in a given row (or column) of
the table. In order to demonstrate this statement let us consider the ith
row of the table and suppose that there are two different elements gj and
gk, for which gigj = gigk = g,. Then g, would appear twice in the row, but
by multiplying the two equations by g;' we obtain g, = g,, in contrast
with the hypothesis.
2. Each row (column) is different from any other row (column); this
property follows immediately from property 1.
3. For abelian groups the table is symmetric with respect to the diagonal.
Table 1.E.2. The 18 abstract groups corresponding to the 32 crystallographic point

groups
Point group Order of Characteristic

the group relations
1
i,2, m
3
4.4
2/m, mm2, 222
6, 6, 3
32.3m
mmm
4/m
4mm, 422.42m
g/m
3m, 62m, 6mm, 622
23
4/mmm
432.43m
m3
6/mmm
m3m
Groups having the same multiplication table, even though their elements
might have different physical meaning, are called isomorphous. They must
have the same order and may be considered as generated from the same
abstract group. For instance the three point groups 222, 2/m and mm2 are
isomorphous. To show this let us choose g,, g,, g3, g4 in the following way.
group 222: 1, 2, 2, 2;
group 2/m; 1, 2, m, i;
group mm2; 1, m, m, 2.
The multiplication table of the abstract group is
In Table 1.E.2 the 32 crystallographic point groups are grouped into 18

abstract groups and for each of them the defining relationships are listed.
We note that all cyclic groups of the same order are isomorphous.
Subgroups
A set H of elements of the group G satisfying the group conditions is called
a subgroup of G. The subgroup H is proper if there are symmetry
operations of G not contained in H. Examples of subgroups are:
(1) the set of even integers (including zero) under the sum law is a
subgroup of the group of all integers;
(2) the point group 32 has elements g, = 1, g2 = 310011,g3 = 3, = 3-l,
g4 = 2[1m1,g5 = 2[0101,g6 = 2[iio1;H = (gl, g2, g3) is a subgroup of G;
(3) the point groups 1, 2, I,

m are subgroups of the point group 2/m;
(4) 222 is a subgroup of 422;
(5) the set T of all primitive lattice translations is a subgroup of the space
group G.
Conversely G may be considered a supergroup of H.
Subgroup-group-supergroup relationships are very important in the
study of phase transitions and of order-disorder problems.
Cosets
Let H = (h,, h2, . . .) be a subgroup of G and gi an element of G not
contained in H. Then the products
giH = (gihl, g h , . . .) and Hgi = (hlgi, h2gi, . . .)
form a left and a right coset of H respectively. In general they will not be
identical.
Furthermore H can not have any common element with giH or Hgi. In
fact, if for instance, we had gihi = hk, it would follow that gi = hkhjl, i.e.
contradicting the hypothesis, gi would belong to H.
It can be shown that two right (or left) cosets, either have no common
element or are identical one to the other. This allows us to decompose G
with respect to H in the following way:
It follows that the order of a subgroup is a divisor of the order of the group
and if this is a prime number, the only subgroup of G is e and G must also
be cyclic.
The decomposition of the group 2/m into separate left cosets with respect
to the subgroup 2 is:
The number of distinct cosets obtained from the decomposition of a

group with respect to a subgroup is called index of the subgroup. In the
previous example, the index of the subgroup 2 of the group 2/m is two.
A coset is never a group because it does not contain the element e.
Conjugate classes
An element gi is said to be conjugate to an element gj of G if G contains an
element gk such that
gi = gklgjgk. (1.E.5)
If g, is fixed and gk varies within G, then the set of elements gi forms a class
of conjugate elements.
In agreement with relation (1.E.5) the element e forms a class on its own.
Since each element of G can not belong to two different classes, it is
possible to decompose G into the factorized set G = e U T, U T2 U . . . .
A physical or geometrical meaning may be attributed to the classes. In
the transformation (1.E.5), let the element g, be a coordinate transforma-

tion due to a symmetry operator and the element gj a matrix operator
related to another symmetry operation. Since g, is transformed by (1.E.5)
(see eqn (2.E.7)), the operators belonging to the same class are changed
one onto the other by coordinate transformations represented by the
elements of the group. For instance, for the point group 32 three classes
may be set up:
(el, (3,3-'), (2[iool,~ [ o I D2~iiol.I
],
As we will see later, the character is identical for the matrix repre-
sentation of all the elements of the same class.
The following rules may be useful to set up the point-group classes:
the classes are formed by one element only for the point groups up to
orthorhombic; this means that each symmetry operator commutes with
all the others;
in an Abelian group the classes are formed by only one element;
the operators identity, inversion, and reflection with respect to a mirror
plane perpendicular to the principal symmetry axis (the axis with the
highest order), are each in a separate class.
Conjugate subgroups
Let H be a subgroup of G and g an element of G not in H. Then all the
elements g - l ~ g ,form a group. H and g-lHg are conjugate subgroups.
Normal subgroups and factor groups

If the right and left cosets of the subgroup H are the same, i.e. Hg, =g,H
for every i, then g;'Hg, = H. This relation is still valid when g, is in H since
both g,H and g, belong to H. Conversely, if a subgroup is transformed into
itself when applying all the elements of the group, the corresponding left
and right cosets must be equal.
Subgroups which are transformed into themselves by applying all the
elements of the group, are called invariant or normal. They must contain
complete classes.
For instance the subgroup 2 of the point group 2/m is normal since
1(1,2) = (1,2)i, m(1, 2) = (1, 2)m. Furthermore, in the group 32, the
subgroup (1,3,3-') is normal, while (1,2i1,1) is not; the subgroup T of all
the primitive lattice translations is an invariant subgroup of the space group
G.
Let H be a normal subgroup of G with index p, while n is the order of G.
Because of (1.E.4) the order of H is n/p. We observe that for H and all its
distinct cosets, the following multiplication law may be established:
(g,H&H) = giHHgj = giHgj = g,&H
Besides, it is:
(g,H)-'(gi~)= H-'~&H = el
We can now define a new type of group of order p, called a factor group
or quotient group, indicated by the symbol G/H: its elements are cosets of
H. The following multiplication table is for the quotient group (we assume
= e).
rrl
For instance, for the quotient group (2/m)/2, (1.E.6) becomes
As a further example let us consider the elements of the point group

4mm: e, 4, 43, q2 = 2, Z[lool,2[olol,Z[llol, Z[liol.Five classes may be formed:
(e), (21, (4, 43), (Z[iool,Z[oio]), (Z[iio172[iiol). The subgroup H = (e, 2) is
invariant and the factor group with respect to H may be written as: H, 4H,
Z[100]H7Z[llop.
The relat~onbetween G and G / H is a n / p + 1 correspondence, i.e.
G -, H. In detail glH = H -, e, g2H -,f,, . . . , g,H + fp. A correspon-
dence of this type (many + one) is called homomorphism and G / H is said
to be homomorphic with G.
Isomorphism is then a special case of homomorphism for which there is a
one to one correspondence. A homomorphic correspondence allows us to
reduce the study of the multiplication laws of the group G to that of the
multiplication laws of the smaller G / H group.
To display many of the abstract definitions given so far, we consider as
major example, the point group G = 23: say
Readers will find the four classes:
and the ten subgroups
of which only {I), (1, 211,1, 2[olo17

2[,11), and G are invariant subgroups.
Calculate now the factored group of H = (1, 211,1, 210101, 2[ooll): we
multiply it by an element not in H, say 3[1111,and obtain
We take now an element not in H or in 3[llllH, say 3flll], and we get

The group is now exhausted and we may write
On assuming g1 = e, then {e, 3[1111H,3fl11]H) form a group under the

following multiplication table
This group is homomorphic with G itself if we associate the elements in H

to e, the elements in 3[111]H with the 'element' 3[,l11H, the elements in
3elllH with the 'element' 3tl11]~.
Maximal subgroups and minimal supergroups

The number of subgroups of a space group is always infinite. They may bp
the site-symmetry groups (groups without a lattice), or line groups, ribbon
groups, rod groups, plane groups, or space groups. For example, the set T
of all the primitive lattice translations is a subgroup (invariant) of the space
group. Further subgroups may be found by considering the set of all the
translations defined by a superlattice. We are here interested only in
subgroups which are space groups themselves.
Let us first recall the concept of proper subgroup. A subgroup H is called
proper subgroup of a group G if there are symmetry operations of G not
contained in H. Now we define maximal subgroups: a subgroup H of a
space group G is called a maximal subgroup of G if there is no proper
subgroup M of G such that H is a proper subgroup of M. For example,
P112,, P1211, P2,ll are maximal subgroups of P212121, while P1 is not
maximal.
For every H, according to (1.E.4), a right coset decomposition of G
relative to H may be made. The index of the decomposition determines the
degree of 'dilution' of the symmetry in H with respect to that in G. Such a
dilution may be obtained in three different ways:
1. By eliminating some symmetry operators (e.g., from G = P212121 to
H = P2,ll). Subgroups of this kind, are called Translationengleiche or t
subgroups. Since the point group P of G is finite, the number of subgroups,
and therefore, of maximal subgroups, is finite. All the maximal subgroups
of type t for any given G are listed in the Znternational ~ a b l e sas type I
maximal non-isomorphous subgroups (see Figs 1.20 and 1.21).
2. By loss of translational symmetry, i.e. by thipning out the lattice. Such
subgroups are called Klassengleiche or k subgroups and are classified as
type 11. A subset of k subgroups are those belonging to the same space
group G or to its enantiomorphic: their number is infinite and they are
called maximal isomorphous subgroups. Those of lowest index are listed
as IIc in the International Tables. For example, if G = (2222, maximal
isomorphous subgroups of lowest index are C222 with a' = 3a or b' = 3b
and C222 with c' = 2c.
Maximal non-isomorphous subgroups of C222 are also P222, P2J12,
P2,22, P2212, P222,, which have the same conventional cell: for practice1
reasons they are labelled as subgroups of type IIa. Space groups with
primitive cells have no entry in the block IZa. Some further subgroups of
C222 are C222, (with c' = 2c), I222 (with c' = 2c) and I2,2,2, (with c' = 2c).
These subgroups have conventional cells larger than that of C222 and are
denoted as subgroups of type IIb. For k subgroups the point group P of G is
unchanged.
3. By combination of 1 and 2. In this case both the translation group T
and the point group P of G are changed.
A theorem by Hermann states that a maximal subgroup of G is either a t
subgroup or a k subgroup. Thus in the International Tables only I, IIa, IIb,
and IIc subgroups are listed.
Sometimes we are interested to the possible space groups G' of which a
given space group G is a subgroup. G' is called a minimal supergroup of the
group G if G is a maximal subgroup of G'. Of course we will have a
minimal t, or a minimal non-isomorphous k, or a minimal isomorphous k
supergroup G ' of G according to whether G is a maximal t, or a maximal
non-isomorphous k, or a maximal isomorphous k subgroup of G'. The
minimal non-isomorphous supergroups of C222 are:
of type t: Cmmm, Cccm, Cmma, Ccca, P422, P42,2, P4,22, P42,2,
~ 4 r n 2~, 4 ~~ 24 b, 2 ~, 4 n 2 P622,
, P6,22, P6,22,
of type k: F222, P222 (with a ' = a/2, b' = b/2).
Maximal subgroups and minimal supergroups for three-

dimensional crystallographic point groups
With trivial changes the definitions of maximal subgroup and minimal
supergroup given above for space groups may be applied to three-
dimensional crystallographic point groups. For example, it will be easily
seen that Laue symmetry is always a minimal supergroup of index 2 of a
non-centrosymmetric point group.
A scheme showing the subgroup and supergroup relationships for point
groups is illustrated in Fig. 1.E.1. Maximal invariant subgroups are
indicated by full lines: if two or three maximal invariant subgroups exist
with the same symbol then double or triple full lines are used.
A set of maximal conjugate subgroups is referred to by a broken line. For
example, from 3m three conjugate subgroups of type m can be formed.
Thus a dashed line refers 3m to m. Furthermore, from 422 two invariant
subgroups of type 222 with index 2 can be formed (no symmetry operation
of 422 refers one subgroup to the other). Thus in Fig. l.E.l a double solid
line refers 422 to 222.
Limiting groups in two and three dimensions

In two dimensions there are two types of point groups:
For very large values of the order of the rotation axis the two types
approach w and am respectively. From the geometrical point of view w and
wm are identical, and our standard notation will be wm (the situation
Fig. 1.E.1. Group, subgroup, supergroup

relationships for point groups (from
International Tables for Crystallography).
a,# a m occurs when the rotation direction is taken into account: i.e. for a
magnetic field round a disc).
In three dimensions a point group can include continuous rotations about
one or about all axes (this is a consequence of the Euler theorem applied to
such a limiting case). In the first case two groups can be identified, wm and
w 2
--, according to whether there is or not a mirror perpendicular to a, axis
mm
(the symmetry of the two groups can be represented by a circular cone and
w m
by a circular cylinder respectively). The symbol -- represents the case in
mm
which continuous rotation about any axis is allowed (the symmetry is
represented by a sphere).
Representation of a group
If a square matrix d can be associated to each g E G , in such a way that
when gigj = g k also didj = d,, then the matrices form a group D isomor-
phous with G. These matrices form an isomorphous or exact repre-
sentation of the group: the order n of the matrices is the dimension of the
representation. In accordance with this point of view, in Chapter 1 we have
represented the symmetry groups through square matrices of order 3.
Different representations of G may be obtained through a transformation of
the type
When condition (1.E.7) is verified for the two representations

rl(d,, d 2 , . . .) and r2(dI, d;, . . .) then the two representations are said to
be equivalent, since q can be interpreted as a change of coordinate system.
It is often possible to find a new coordinate system for which each matrix d
is transformed into
with dl of order m < n and d2 of order (n - m). If this can not be obtained
by any transformation, then the representation is called irreducible;
otherwise it is called reducible. Sometimes dl and d2 can be further
reduced, and at the end of the process each matrix d j will be transformed
into
q-ld,q = diag [djl), dj2), . . . , d,(")]= d,'
where dji) are themselves matrices.
The matrices dl1), dp), dill, . . . all have the same dimension. Similarly
dl2),di2),di2),. . . have the same dimension. From the rule of the product of
blocked matrices it follows that (dl1),d$'), dS1),. . .) form a'representation of
the group, as well as (dl2),ds2),dS2),. . .), etc.
It can be shown that for finite groups the number of irreducible
representations is equal to the number of classes. For instance an
isomorphous (reducible) representation of the point group 32 is
Also the two-dimensional irreducible representation
and the one-dimensional non-exact (homomorphic) representations exist:
Character tables
The sum of the diagonal elements of a matrix, elsewhere called trace, in
group theory is called character and is indicated by ~ ( g It) is obvious that
~ ( g , )defines the dimensionality of the representation. The complete set of
characters for a given representation is called the character of the
representation. Since the traces of two matrices related by a coordinate

transformation are identical, the characters of two equivalent repre-
sentations will be identical (the converse is also true). Several properties of
a point group may be deduced from the characters of its irreducible
representations alone. It is therefore convenient to set them up in tables
called character tables. Each row of the table refers to a particular
irreducible representation and each column to a given class.
1.F. Symmetry generalization

Only a few intuitive elements of this subject are given, since a full treatment
would exceed the limits of the present book. The reader is referred to
specific texts or
The symmetry groups G:

A space may or may not be periodic in all its m dimensions. The
corresponding symmetry groups are indicated by GT, with m 3 n, where n is
the number of dimensions of the subspace in which the group is periodic. In
this space only symmetry operations transforming the space into itself will
be allowed. For instance in the G: groups, describing objects periodic in one
direction and finite in the other two, at least one line will remain invariant
with respect to all symmetry operations.
The G1 groups
1. Gh groups. In a one-dimensional space (a line), which is non-periodic,
only two symmetry operators are conceivable: 1 and ? (which is the
reflection operator m). The only two (point) groups are therefore 1
and I.
2. G: groups. Besides the 1 and ? operators, they contain the translation
operator. Only two groups of type G: are then possible.
The G2 groups
1. G?, groups. In a 'two-dimensional space (a plane), which is non-periodic,
the only conceivable operators are those of rotation around an axis
perpendicular to the plane and of reflection with respect to a line on the
plane. The number of (point) groups is infinite, but there are only ten
crystallographic groups (see p. 16).
2. ~f (border) groups. In a two-dimensional space, periodic in one
dimension, only the symmetry operators (and their combinations), which
transform that direction into itself, are allowed. We may therefore
consider reflection planes parallel or perpendicular to the invariant t a---a--t:2
direction, glides with translational component parallel to it and two-fold - - - - - - - - - - - t.a 1- - - + - - - I at:2,a
axes. There are seven Gf groups (the symmetries of linear decorations)
which are represented in Fig. 1.F.1. t.m +-+-+-+t:zm
3. G$ groups. There are the 17 plane groups described on the pages 30 and I-It:m
34. Fig. 1.F.l. The seven border groups
The G3 groups
1. G; groups. There describe non-periodic spaces in three dimensions. The
number of (point) groups is infinite (see Appendix l.B), but there are
only 32 crystallographic point groups (see pp. 11-16).
2. G: (rod) groups. Rod groups may be considered as arising from the
combination of one-dimensional translation groups with point groups G;.
They describe three-dimensional objects which are periodic in only one
direction (say z). This must remain invariant with respect to all
symmetry operations. The only allowed operations are therefore n and ri
axes coinciding with z, 2 and 3 axes perpendicular to it, screw axes and
glide planes with a translational component parallel to the invariant
direction.
There are 75 G: crystallographic groups. In Table l.F.l the rod group
symbols are shown alongside the point groups from which they are derived.
The first position in the symbol indicates the axis (n or ri) along z, the
Table 1.F.1. The 75 rod groups
Point group Rod groups
1 1
2 2 21
3 3 31
4 4 41
6 6 61
Im Im Ic
2mm 2mm 2,mc
31-17 3m 3c
4m m 4mm 4,mc
6mm 6m m 6,mc
m m
2/m 2/m 2,/m
4/m 4/m 421m
6/m 61m 6,/m
m2m m2m m2c
2 2 2
--- 2 2 2
--- 2, 2 2
---
mmm mmm mmc
4 2 2
--- 4 2 2
--- 4, 22
---
rnrnrn mmm mmc
6 2 2
--- 6 2 2
--- 6, 2 2
---
mmm mmm mmc
12 12
222 222 2,22
32 32 312
422 422 4,22
622 622 6,22
1 1
3 3
4 4
6 6
- 2 -1 -
2 -2
1-
1-
m m C
- 2 - 2
3-
-2
3- 3-
m m C
4m2 4m2 4c2
6m2 6m2 6c2
second position refers to axes (n or i i ) in the plane (x, y) normal to the

invariant direction, the third position for axes in the (x, y) plane bisecting
the previous ones. Glides in the z direction are denoted by c.
3. G: groups. There are called layer groups and d e s ~ r i b e [ ~ , ~the ,~~,'~]
symmetries of doubly periodic three-dimensional objects. They are
useful in describing the patterns of walls, panels, and, at an atomic level,
in describing structures with layer molecular units.
Let us denote by z the direction perpendicular to the layer plane. It will
be the vertical axis from now on: directions in the plane will be called
horizontal. Rotations can only occur about a vertical axis, and if twofold,
also about a horizontal one. No more than one horizontal reflection plane
can exist, otherwise translational symmetry should also occur along the z
axis. The layer groups can be enumerated according to the five nets quoted
in Table 1.7 for the plane lattices. In Table 1.F.2 the 80 layer groups are
divided in blocks, each block divided by the subsequent by a double line:
each block refers to a specific net (in sequence, parallelogram, rectangular,
centred rectangular, square, hexagonal) and contains the point groups
compatible with the net and the corresponding layer groups.
The number of point groups which can be used is 31: the 32 three-
dimensional point groups minus the five incompatible groups 23, m3, 432,
43m, m3m plus four second settings (this time the z axis is distinguishable
from x and y: so 2 is different from 12, or 2/m from ?2/m, . . .).
The first position in the layer group symbol gives the type of cell, the
second refers to the z direction. The third and fourth positions refer:
(a) for rectangular nets, to x and y respectively;
(b) for a square net, to x (and therefore to y) and to a diagonal direction;
(c) for a hexagonal net, to x (and therefore symmetry related axes) and to
diagonals directions.
As subgroups of the G; groups we may consider the G: groups, which

are obtained by projecting the Gq groups along the axis normal to the
singular plane.
4. G: groups. These are the 230 space groups (see pp. 22-30).
The 64,groups
The three-dimensional Euclidean space may be insufficient to describe the
symmetries of some physical objects. We can therefore introduce one or
more additional continuous variables (e.g. the time, the phase of a wave
function, etc.), thus passing from a three-dimensional space into a space
with dimensions m > 3. In a four-dimensional Euclidean space the sym-
metry groups G: may be constructed from their three-dimensional projec-
tions G:, which are all well known. Thus there are 227 point groups G: and
4895 groups Gi.
The groups of colour symmetry

Groups in which three variables have a geometrical meaning while the
fourth has a different physical meaning and is not continuous, are
Table 1.F.2. The layer groups
Point group Plane groups
1 P1
2 P2
rn prn Pb
Tim p2/m p2/b
1 P1
1rn plm plb
2rnrn p2rnm p2bm p2ba
rn2m prn2rn pm2,a pa2,m pa2a

pn2,m pn2a pb2rn pb2,a
2 2 2
--- 2 2 2 P--l
P--- 222 22,2,
mrnm mmm rnbrn
p2- -2A2 p22
- l2L 222 22,2
brnm bm a P b b i Pbba-
22121 p---
P--- 22,2 P,b2a2 2
nrnm nbm
1rn c l rn
2mm c,mrn
m2m cm2m
2 2 2
--- 2 2 2
C---
mmm mmm
12 c12
222 c222
- 2 - 2
1- cl-
m rn
-
4 P4
4m rn p4mm p4grn
41rn p4m/m p4/n
4 2 2
--- 4 2 2
P--- 2- 2
p -4A 42,2 422
mmrn mmm r n g m P---
r n g m P---
ngm
4 P4
422 ~ 2 2 ~ 4 2 ~ 2
4m2 p4m2 p4g2 p42rn p42,rn
particularly important in crystallography. For instance:

1. The position and orientation of the magnetic moments of the cobalt
atoms in the CoA1204structure (space group ~ d 3 m may ) be described by
means of a two-colour group (see Fig. l.F.2(a)) in which each colour
corresponds to a given polarity of the magnetic moment. Groups of this
type are also called 'groups with antisymmetry' or 'black-white' sym-
metry. More complicated cases require more colours, and the term
colour symmetry is used. Classical groups involve only neutral points.
As a further example let us consider the case of NiO, a material used
-
in the ceramic and electronic industries. At room temperature, NiO is
rhombohedral with edge length a , 2.952 A and a~ =. 60'4': aR ap-
proaches 60" with increasing temperature, and, above 250°C, NiO is
cubic with ac2: 4.177 A. The relation between the two cells is shown in
Fig. l.F.2(b): the same set of lattice points is described by the primitive
rhombohedral unit cell and by the face-centred cubic cell provided that
aR= 60' exactly and a R = a/*. If the cube is compressed (or extended)
along one of the four threefold axes of the cubic unit cell then symmetry
reduces from cubic to rhombohedral (the only threefold axis is the
compression axis). The polymorphism of NiO is due to its magnetic
properties. Each Ni2+ ion has two unpaired spins (the [Ar]3d8 electronic
configuration). At room temperature the spins in NiO form an ordered
antiferromagnetic array: layers of Ni2+ with net spin magnetic moments
all in the same direction alternate with layers of Ni2+ with magnetic
moments all in the opposite direction, as in Fig. l.F.2(c). In these
conditions the threefold axis is unique and the structure is rhom-
bohedral. Above 250°C the antiferromagneti~ordering is lost: the
rhombohedral + cubic transition occurs and NiO displays ordinary
paramagnetism.
2. If we project a G: group, in which a 6, axis is present, on a plane
perpendicular to the axis, we obtain a G; group. But, if we assign a
different colour to each of the six atoms related by the 6, axis, we will
obtain a colour group G;,@)with a clear meaning of the symbols.
In the groups with antisymmetry there will be four types of equivalence
between geometrically related objects: identity, identity after an inversion Fig. 1.F.2. Examples of structure described by
operation, anti-identity (the two objects differ only in the colour), identity an antisymrnetry group: (a) CoAI,O, magnetic
after both an inversion operation and a change in colour. A general rotation structure; (b) geometrical relation between a
face-centered cubic unit cell and a primitive
matrix may be written in the form rhombohedral unit cell; (c) antiferromagnetic
superstructure of NiO (only ~ i ' +ions are
shown).
\ O 0 0 R44/
where R4, = -1 or +1 depending on whether or not the operation changes
the colour.
For the three-dimensional groups with antisymmetry we observe that,
because of the existence of the anti-identity operation 1' (only the colour is
changed), the anti-translation operation t' = t l ' will also exist. New types of
Bravais lattices, such as those given in Fig. 1.F.3, will come out. As an
example, in Fig. l.F.2(c) the quasi-cubic magnetic unit cell of NiO has an
edge length twice that of the chemical unit cell. It may be seen[14]that if the
five Bravais lattices are centred by black and white lattice points (in equal
percentage) then five new plane lattices are obtained. In three dimensions
there are 36 black and white Bravais lattices, including the traditional
uncoloured lattices.
? 9 References
fg
1. Shubnikov, A. V. (1960). Krystallografiya, 5, 489.
2. Shubnikov, A. V. and Belov, N. V. (1964). Coloured symmetry. Pergamon,
Oxford.
3. Bradley, C. J. and Cracknell, A. P. (1972). The mathematical theory of
symmetry in solids. Representation theory for point groups and space groups.
Clarendon Press, Oxford.
4. Lockwood, E. H. and MacMillan, R. H . (1978). Geometric symmetry.
/
,
I
b---- -b
Cambridge University Press.
5. Vainshtein, B. K. (1981). Modern crystallography I: Symmetry of crystals,
methods of structural crystallography. Springer, Berlin.
6. (1983). International tables for crystallography, Vol. A, Space group symmetry.
?' Reidel, Dordrecht.
Fig. 1.F.3. Two antisymmetrical Bravais lattices. 7. Dougherty, J . P. and Kurtz, S. K. (1976). Journal of Applied Crystallography, 9,
145.
8. Kitaigorodskij, A. I. (1955). Organic crystallochemistry. Moscow.
9. Mighell, A. and Rodgers, J. R. (1980). Acta Crystallographica, A36, 321.
10. Wilson, A. J. C. (1988). Acta Crystallographica, A44, 715.
11. Fischer, W., Burzlaff, H, Hellner, E., and Donnay, J. D. M. (1973). Space
groups and lattice complexes, NBS Monograph 134. National Bureau of
Standards, Washington, D.C.
12. Alexander, E. and Herrmann, K. (1929). Zeitschrift fur Kristallographie, 70,
328.
13. Alexander, E. (1929). Zeitschrift fur Kristallographie, 70, 367.
14. Mackay, A. L. (1957). Acta Cryst., 10, 543.
Introduction
In this chapter elements of crystallographic computing are described.
Material is treated in order to answer day-to-day questions and to provide a
basis for reference. Among the various topics, those which are of more
frequent use have been selected: axis transformations, geometric calcula-
tions (bond angles and distances, torsion angles, principal axes of the
quadratic forms, metric considerations on the lattices, structure factors,
Fourier calculations,. . .). The method of least squares and its main
crystallographic applications are treated in greater detail. For practical
reasons some calculations useful in characterization of thermal ellipsoids are
developed in Appendix 3.B .
The following notation is adopted: rl. r2 denotes the scalar product
between the two vectors rl and r2, rl A r2 is their cross product, r will be the
modulus of r. S1S2is the (row by columns) product of two matrices S1and
S2: S is the transposed matrix of S, and S is the determinant of the square
matrix S.
We will also distinguish between coordinate matrices and vectors. For
example, with respect to a coordinate system [0, a, b, c] the vector r will be
written as
where X is the coordinate matrix and A is the matrix which represents the
basis vectors of the rectilinear coordinate system.
The metric matrix

Let [0, a, b, c] be our coordinate system. The scalar product of r, and r2 is
.
rl r2 = (xla + ylb + zlc) . (x2a + y2b + z2c)
=xlx2a2+yly2b2+ zlz2c2+ (x1y2+ x2y1)ab cos y
L IPC+ ( Y ~+zY~ ~ z ~cos
+ ( ~ 1 +~ X2 ~ Z ~ )cos ) ~CY c
or, in matrix notation,
G is the metric matrix, also called the metric tensor: its elements define both
the moduli of a, b, c and the angles between them. The value of its
determinant is
G = a2b2c2(1- cos2 a - cos2j3 - cos2 y + 2 cos a cos P cos y)
(2.2)
which (see Table 2.1 and p. 69) is equal to V2 (square of the volume of the
unit cell). If r, = r2= r then (2.1) becomes
r2 = XGX = x2a2+ y2b2+ z2c2+ 2xyab cos y

+ 2xzac cos j3 + 2yzbc cos a (2.3~)
which gives the modulus square of a vector. We c8n now calculate the
following:
1. The interatomic distance d between two atoms positioned in (xl, y,, z,)
and (x2y2z2)Denoting
A1 = a(xl - x2), A2 = b(yl - y2), A3 = c(zl - 22)
gives
+ + +
d2 = A: A; A$ + 2A1A2COSY + 2A1A3cos P 2A2A3cos a. (2.3b)
2. The angle 8 between two vectors
cos 0 = ~,GX,l(r,r~). (2.4)
3. The cross product r2 A r3:
4. The scalar triple product rl r2 A r3:
is the volume of the unit cell

The following formulae of vector algebra are also recalled for future
convenience:
Some properties of the rotation component R of the symmetry operators

..
Crystallographic computing 1 63
can now be proved:

1. If a, b, c define a primitive cell, the elements of R are integers. Indeed
the relation Xr = RX transforms lattice vectors r into lattice vectors r r .
Both r and r r have integer components, consequently also the elements
of R must be integers.
2. A symmetry operator does not change the moduli of the vectors and the
.
angles between vectors: i.e. r; ri = r, r2. Therefore, according to (2.1),
X;GX; = XlRG~X2= XlGX2, from which
Only the matrices R which satisfy (2.9) can be symmetry rotation

matrices in the coordinates system defined by G.
3. Multiplying the second eqn (2.9) by R gives GR= W G . If only
determinants are taken into consideration then R = f1 arises. Properties
1, 2, 3 can be verified for the rotation matrices quoted in Appendix 1.D.
The reciprocal lattice

Very useful in metric calculations (as well as in diffraction geometry), the
reciprocal lattice was introduced by P. Ewald in 1921. Let a, b, c be the
elementary translations of a space lattice (called here a direct lattice). A
second lattice, reciprocal to the first one, is defined by translations a*, b*,
c*, which satisfy the following two conditions:
Equation (2.10a) suggests that a* is normal to the plane (b, c), b* to the
plane (a, c) and c* to the plane (a, b). The modulus and sense of a*, b*, c*
are fixed by (2. lob).
According to (2.10a) a* may be written as
where p is a constant. The value of p is obtained if the scalar product of

both the sides of (2.11) by a is taken:
a*.a=l=p(b~c.a)=pV
from which p = 1/V. Equation (2.11) and its analogue may then be written
as
or, in terms of moduli,
Equation (2.10) also suggest that the roles of direct and reciprocal space
may be interchanged: i.e. the reciprocal of the reciprocal lattice is the direct
lattice. Therefore
1 1 1
a=,(b*~c*), b = - ( c* A a * ) , c =-
v v* v * (a* A b*). (2.13)
It may be easily verified that the reciprocals of triclinic, monoclinic, . . .

lattices are triclinic, monoclinic, . . . themselves. However, the reciprocal of
an F lattice is an I lattice and vice versa (see Appendix 2.D). In detail:
1 . In monoclinic lattices b* JJ b while a* and c* are in the plane (a, c): then
a* = l / ( a sin p), b = llb, c* = l l ( c sin P),
a*=y*=n/2, p*=n-/3.
2. In rhombic, tetragonal, and cubic lattices a* )I a, b* (1 b, c* 1) c, and
3. In trigonal and hexagonal lattices c* )I c while a* and b* are in the plane

(a, b ) :
For the rhombohedrical basis
a* = p* = y * , cos a* = -COS @ / ( I + COS a ) .

Some relations between direct and reciprocal bases are discussed in
Appendix 2.A: a summary is given in Table 2.1.
Some important properties of the reciprocal lattice are:
1 . The scalar product of two vectors, the first defined with respect to the
reciprocal basis and the second to the direct basis, assumes a very simple
expression:
Table 2.1. Relationships among direct and reciprocal lattice

parameters. Inverse relationships are obtained by interchang-
ing t h e starred by unstarred parameters
a*=-.b c s i n a b*- acsinp . c* - a b sin y

V ' V ' v
sin a* =
v cos a -
* - cos /3 cos y - cos a
abc sin p sin y ' sin p sin y
*~cosa~~sy-cosp
sin p* = cosp-
abc sin a sin y ' sin a sin y
v *~cosacosp-cosy
sin y* = cosy -
abc sin a sin p ' sin a sin p
V = abc(1 - cos2 a - cos2P - cos2 y + 2 cos a cos P cos y)'I2
= abc sin a sin p sin y* = abc sin a sin p* sin y
= abc sin a* sin /3 sin y
v* = 1/v
2. The vector r i = ha* + kb* + lc* is normal to the family of lattice

planes (hkl). In Fig. 2.1 the plane of the family nearest to the origin is
drawn. In accordance with p. 8 the vectors (A-0), (B-0) and (C-0) are
equal to alh, blk and cll respectively. Consequently
(B-A) = blk - alh, (C-A) = cll - a l h , (c-B) = cll - blk.
Therefore
.
r i (B-A) = (ha* + kb* + lc*) (blk - a l h ) = 0,
r ; . (C-A) = r ; . (C-B) = 0 .
Since r i is perpendicular to two lines in the (hkl) plane, it is normal to the
plane.
Fig. 2.1. The plane of the faniily (hkl) nearest to
3. If h, k, 1 have no common factor, then the origin.
where dH is the spacing of the planes (hkl) in the direct lattice. This may be
proved by observing that dH is equal to the length of the normal ON to the
plane ABC from the origin 0. Since r i has the same direction as ON,
r 1
d H = (alh) .-=-
r; ria
4. As well as for direct lattice, a metric matrix G* may be defined for the
reciprocal lattice:
It may be easily verified that
Consequently (see (2.1) and (2.3) for the direct lattice)
from which
+ + +
d H = (h2a*2 k2b*2 1 2 ~ * 22hka*b* cos y*
+ 2hla *c*cos /3* + 2klb *c*cos (2.17b)
is easily obtained.
Specific expressions of d H for the various crystal systems are given in Table
2.2.
Basis transformations
In three-dimensional space the coordinate system defined by the base
vectors a ' , b r , c r may be defined in terms of the base vectors a, b, c by
Table 2.2. The algebraic expressions of 4, for the various crystal systems
Cubic
Tetragonal
Orthorhombic
Hexagonal and
trigonal ( P )
Trigonal (R)
( h2 + k2 + 1') sin2 a + 2(hk + hl + kl)(cos2a-cos a)
a2 1 +2cos3cu-3cos2~
h2 I? 1' 2hlcosP
Monoclinic 2 . 2 +7+2.2--
a sln fi b c sin /3 ac sin2P
Triclinic (1 - cos2 a - cos2 /3 - cos2 y + 2 cos a cos fi cos y ) - l ( $ sin2 a
21h 2hk
+- (cos y cos a - cos p ) + -(cos a c0s.p - COS y )
ca ab
three equations (suppose the transformation leaves the origin invariant)
where mi,are any real numbers. In matrix notation (2.18) is written as
The reverse transformation will be A = M-'A'. The vector r = xa yb zc + +

may be written in the new coordinate system A' as r' = r = x'a' y'b' + +
z'c'. Introducing (2.18) into the r' expression leads to
The reverse transformation is X' = (M)-'X. It should be noted that (2.20)

provides the transformation rule for the components of the vector r while
the vector itself is unaffected (r' = r) by a change of axes.
The metric matrix G' in A' may be calculated by substituting into (2.1) a ' ,
b', c' for a, b, c and then using (2.18). The result is
The unit cell volume defined by A' is V' = a' A .

b' c' therefore, according
Crystallographic computing ( 67
to (2.19) and (2.5), V ' = V M . Thus, for any transformation of axes the unit
cell volume is multiplied by the determinant of the transformation matrix.
The set of matrices conventionally used to pass from centred cells to
primitive ones and vice versa is shown in Table 2.C.1 (however, transfor-
mation matrices are not unique).
Let us now apply (2.20) to derive the transformation rules of a quadratic
form. Let
be a quadratic form defined in the coordinate system A. In accordance with

(2.20) X Q X = X ' M Q M X 'from which
Q' = MQM or Q = M - ~ Q ' ( M ) - ~ . (2.22)
A special linear transformation is that which relates a' = a * , b' = b*,
c' = c* to a, b , c (in this case A' will be replaced by A*). Let us show that in
this case M = G-l = G*. Write the vector r = xu + yb + zc as
(in terms of reciprocal axes
r' = r* = x'a* + y'b* + z'c* = (r' .a)a* + (r' b)b* + (r' c)c*). (2.24)
On assuming in (2.23) r = a *, b *, c*, the following relations
. . .
a* = (a* a*)a + (a* b * ) b + (a* c*)c
. .
b* = (b* a*)a + (b* b*)b + (b* c*)c
.
C* = (c* a*)a + (c* b*)b + (c* c*)c
are respectively obtained, which in matrix notation reduce to

A * = G * A or A = G A * . (2.25)
It is easily seen that relations (2.25) are special cases of (2.19). Similarly:
1. Relations (2.26) are special cases of (2.20)
X=G*X* or X*=GX. (2.26)
They transform components defined in direct space into components
defined in reciprocal space and vice versa. In accordance with (2.1) and
(2.26) we have
.
rl r2 = XlGX2 = XIX; = Xax;. (2.27)
In particular r2 = r*2= X X * is the square modulus of a vector.
2. Equations (2.28) are special cases of (2.22):
Additional transformation rules are given in Appendix 2.E. The reader

is referred to the splendid book by sandsr1]for further insight.
Transformation from triclinic to orthonormal axes

Geometrical calculations are often more easily made in an orthonormal than
in a crystallographic frame.
Let
A= (I) and E= (i)

be the crystallographic and the orthonormal bases respectively. Then,
according to (2.18), E = MA, and inversely A = M-'E.
If we choose (see Fig. 2.2) el along a, e2 normal to a in the (a, b) plane,
e3 normal to el and e2 (and therefore parallel to c*), then the unit vectors
ala, blb, clc are referred to el, e2, e, by means of
/---
where (I,, I,, I,), (m,, m2, m,), (n,, n2, n,) are direction cosines of the unit
a vectors ala, blb, clc in E. Therefore
ClT=Cm?=Cn:=l.
i i i
C From Fig. 2.2 it may be deduced that
Fig. 2.2. Orthonormalization of crystallographic
bases. 1, = 1, 12= 0, 1, = 0, m l= cos y,
m2 = sin y, m3 = 0, nl = cos p.
Since
cos a = C mini = cos cos p + sin sin n2
i
we obtain
n2 = (cos a - cos p cos y)/sin y = -sin p cos a*.
Furthermore, from the relation Ci n: = 1
n, = sin p sin a*= l/(cc*)
is easily obtained. Finally
(fI))=(co:y cos p
0
siny
0
-sin p cos a* ll(c*c)

from which
=(bc:s
c cos p
0
b sin y
0
0
-c sin /3 cos a* l / c *
)(,) = WIJL (2.30)
The matrix M to use in E = MA is therefore

1/a 0
-cos y/(a sin y) l/(b sin y)
a * cos p* b * c o s a * c*
A crystallographic frame may be orthonormalized in infinite ways. For

example, el may be chosen along a* and e2 in the plane (a*, b*): e3 is then
along c. In this case
M=
("d*~;~*
0
c*cosp*
b*siny* - c * s i n p * c o s a
llc
and
l l a * -cot y*/a* a cos /3
~ - l = 0 l/(b*siny*) b c o s a
0 C
Also, we may choose el along a* and e2 along b (then e3 is in the plane

(b, c)). We obtain
l l ( a sy sp*) l l ( b t a tp*) - l l ( b ty sp*) -ll(c s a tp*)

M=( 0 l/b 0 (2.31~)
0 -ll(b t a ) ll(c s a )
and
where sy, cy, ty stand for sin y, cos y, and tan y, etc. . . .
The family of all the possible transformations M which orthonormalize a
given frame A according to E = MA may be obtained from the following
decomposition of the metric matrix G of A:
Such a relation is obtained by requiring that in (2.21) the condition G' = I is

satisfied.
The volume of the unit cell defined by a, b, c may be easily obtained if M
is known. Indeed, if we express a, b, c in terms of el, e2, e3 according to
.
(2.30), and if (2.5) is then used for calculating a b A c, then the result
V =a b A c = det (M-')
is obtained. Furthermore, since det (M-l M-l) = v2= G, the assumption

made on p. 62 according to which G is the square of the volume of the unit
cell is also proved.
Rotations in Cartesian systems

In a right-handed Cartesian coordinate system [0, el, e2, e3] the anti-
clockwise rotation of a vector r through angle al about el or through a2
about e2 or through cu3 about e3 produces a vector r' = R,r where s = x , y , z
and
0
cos a, 0 sin a,
(-sin a,
c a s ia
Rz(a3) = sin a3 cos a3
0 cos a2
0)
Since the matrices are orthogonal, the following relations hold:
R(a) = R-l(cu) = R-l(-a) = R(-a).
Corresponding clockwise rotations are obtained by changing a;. into -a;..
Matrices corresponding to rotatory-reflection operations about el, e2, e3
are obtained by replacing in (2.32) the integer 1 by -1. Matrices
corresponding to rotatory-inversion operations about el, e2, e3 are obtained
by changing the signs of all the elements. The traces of the matrices that
represent proper rotations, rotatory-reflection and rotatory-inversion oper-
+
ations are 1 2 cos a, 2 cos a - 1, -2 cos a - 1 respectively.
It should be noted that rotating r about el, e2, e3 in anti-clockwise mode
is equivalent to rotating el, e2, e3 in clockwise mode. For example if the
framework [0, el, e2, e3] may be superimposed to the new framework
[0,e;, eh, e;] by a clockwise rotation through a, about e3, then E' =
R,(a3)E. According to (2.20), X ' = R;l(a3)x = R,(a,)X, which corresponds
to an anti-clockwise rotation of a vector r in E.
See now some useful applications:
1. Rotation about the unitary vector I in a rectilinear coordinate system.
Given the crystal base A = (a, b, c), an orthonormal base E = (el, ez, e3)
may be chosen (E = M A ) such that el coincides with 1. In accordance with
(2.32) a rotation about 1 in E is represented by R,, and, according to Table
2.E.1, the same rotation in A is represented by R = MR,(M)-'. As an
example we calculate in the hexagonal system the matrix corresponding to
an anti-clockwise rotation through x about a. According to (2.30) and
(2.31) the matrices M and M-I are
from which
1(cos - 1)/2 -c sin x / ( a f l )
R = MR,(M)-~ = cos x -2 c sin Xl(afl)
0 aflsinX/(2c) cos x
Incidentally, it may be noted that the ratio a/c is unconstrained. Thus R
(which is an integer matrix) may correspond to a symmetry axis along a in
the hexagonal system only if x =O, n (onefold or twofold proper or

improper axis).
The reader will easily find that the general rotation matrix about b in the
monoclinic system is
.=( cos x + cot plsin x 0 c sin %/(asin p)

0 1 0
-a sin xl(c sin$) 0 cos x - cot /3 sin x
R will correspond to a symmetry operator if = 0, n.
2. Explicit expression for the rotation-matrix R about the unit vector I in an
orthonormal system. Let I,, I,, I, be the direction cosines of 1 in the
orthogonal system A. A new orthonormal system E = (el, e,, e,) is chosen
so that 1 coincides with el: then, according to (2.29),
where mij are suitable parameters. The matrix corresponding to a rotation

about I in A is then
where sx and cx stand for sin x and cos X.

3. Sequence of rotations about the same axis or about different axes.
The overall effect of a sequence of three successive rotations R,(w),
Ry(q), Rx(q) on the vector r may be so described:
from which
X = RxRyR,X = RX
where
cqcw -cqso
c q s o + sqsqcw c q c o - sqsqsw -sqcq
+
sqsw - cqsqcw s q c o c q s q s o c q c q
and c q stands for cos q , s q for sin I$, etc. . . .

If the three rotations are made in different order, the overall effect
changes. As an example, the reader will easily verify that rotating r through
120" about e3 and then through 90" about el leaves r in r" where
and
4. Eulerian angles. In Fig. 2.3(a) two orthonormal frameworks

[ 0 , el, e2, e3] and [O, el, e;, e;] are shown. The axis OK, called the line of
nodes, is the intersection of the (el, e2) and (e;, e;) planes and is
perpendicular both to e3 and e;. E may be superimposed to E' by three
anti-clockwise rotations in the following order: (a) rotate about e3 by the
angle 81 (OK and el are now identical); (b) rotate through 8, about OK
which will bring e3 into coincidence with e;; (c) rotate about e; by O3 which
then brings el to el and e2 to e;.
Suppose now that r is a vector with coordinates X in the space-fixed
system and coordinates X' in a body-fixed system: then X and X' are
connected by
where cei and soi stand for cos ei and sin Oi respectively.
Fig. 2.3. (a) Eulerian angles. (b) Spherical polar
coordinates. The Eulerian angles can also be used in order to calculate, in any
crystallographic system A, the rotation function corresponding to any
desired rotation. The simplest procedure could be:
(a) Transform coordinates in A (say XA) into coordinates in E (say XE). If
E = MA, then according to (2.20), X, = (M)-lXA. For example, M may
be (2.31a) or (2.31b) or (2.31~).
(b) Transform the Cartesian coordinates into a rotated set of axes. Then XE
transforms in XI, = REuXE.
(c) Return these coordinates into the system A. Then the inverse operation
described in (a) has to be made.
The final coordinates are Xj, = M X =~ MR,,(M)-'X~ so that the desired
rotation function is
the full range of rotation operations is

Since Eulerian angles are difficult to visualize it is preferred to specify a

rotation in terms of spherical polar coordinates (see Fig. 2.3(b)). The
rotation about a given axis is specified by the angle x and the polar
coordinates and q, define the direction of the rotation axis. According to
Fig. 2.3(b) the orthonormal coordinates x , y, z are given by
Since (2.32b) represents the anti-clockwise rotation matrix about the unit
vector 1 in an orthonormal frame we can replace the direction cosines
11, 12,l3 of the rotation axis 1 by
Il = sin I) cos q,, l2 = cos q, l3 = -sin qj sin q,
and so obtain the expression of the rotation matrix Rsp in terms of the
rotation angle x and the spherical polar coordinates q, and q :
CX+ (1 - cx)s2*c2Q) -s*sQ)sx+ (1 -cx)c*s*cQ) -cvsx-- (1 -cx)s2*cQ)sQ)
Since
(
R,, = s * s q s ~ +(1 - c ~ ) c v s v c q ,
c ysx- (1 - cx)s2*cQ)sq,
cx + (1 - c x ) c 2 ~
-s*cQ)sx- (1 - cx)c*s*sq,
S~CQ)SX - (1 - C X ) C I ~ S ~ S Q )
cx + (1 - c~)sz*s2Q)
all rotation operations qre included in
Some simple crystallographic calculations
The reader is referred to Appendix 2.B for calculations concerning

crystallographic directions and planes. Here we limit ourselves to the
description of some calculations which occur very frequently in crystal
structure analysis.
Torsion angles
For a sequence of four atoms A , B, C, D, the torsion angle o(ABCD) is
defined as the angle between the normals to the planes ABC and BCD (see
Fig. 2.4). By conventiod2] w is positive if the sense of rotation from BA to
CD, viewed down BC, is clockwise, otherwise it is negative. Note that
w(ABCD) and w(DCBA) have the same sign; furthermore, the sign of a
torsion angle does not change by rotation or translation, and is reversed by
reflection or inversion. According to the definition (see again Fig. 2.4)
cos o =
(a A .
b ) (b A c) b
-sin w =
(a A b) A (b A c)
ab2c sin a sin y b ab2c sin a sin y
which, owing to (2.7) and (2.8), become
cos a cos y - sin /3 Vb
cos w = , sin o =
sin a sin /3 ab2c sin a sin y ' Fig. 2.4. Definition of the torsion angle w .
If the vectors a * , b * , c* reciprocal to a , b , c are considered, it may easily

be seen from Table 2.1 that w coincides with P*.
Best plane through a set of points

Consider a set of p atoms at positions rl, r2, . . . , r, where rj = AX,.. The best
plane through them is that for which the sum of the squares of the distances
of the atoms from the plane, multiplied by the weights of
the atomic positions, is minimum. Such a plane is characterized by the
minimum of the function (see eqns (2.B.5) and ( 2 . B . 7 ) )
where d is the distance of the plane from the origin of the coordinate
+ +
system, n = n , a + n2b n,c = A N = nFa* nib* ngc* = A*N* is the +
normal to the plane. The weights w, should be taken as being inversely
proportional to the variances of the atomic positions in the direction normal
to the desired plane, but they are often assumed to be unitary.
If the atoms are considered as point masses of weight w j , the least squares
plane coincides with the principal plane of least inertia.
The minimum of Q will be searched with respect to d and n:, n z , n;
under the condition that n is a unit vector. This kind of problem is best
solved by the method of Lagrange multipliers. The function to minimize is
thenL3]
= C w j ( A * x j- d ) 2 - A ( N * G * N *- 1). (2.34)
i
The partial derivative of (2.34) with respect to d gives
from which
d = N * [ ( F w j x j ) ( ? w,)-'1 = N*Xo. (2.35)
Equation (2.35) states that the plane passes through the centroid ro = ao.
Owing to (2.35), eqn (2.34) becomes
where r; = r;. - ro and
Note that N * S N * is the weighted sum of the squares of the distances of the
atoms from the plane. Setting to zero the derivative of (2.36) with respect to
N* (in practice with respect to the components n:, ng, nz) gives
SN* - AG*N* = 0, which may be also written as
(A- hl)N = O (2.37)
where A = SG and N = G*N* (see Table 2.E. 1). Writing (2.37) as AN = AN
and multiplying both sides for N* gives
The eigenvalue A is therefore the weighted sum of the squares of the

distances of the atoms from the plane, N is the corresponding eigenvector.
There are three solutions of eqn (2.37), which in general correspond to
three different A values, say A, >AP 3 A., Each eigenvalue of A gives a
stationary value of Q ' : the three corresponding eigenvectors N, NP, N,
define the principal axis of inertia of the system of atoms when considered
as points of masses w,, . . . , w,. The best plane corresponds to the
eigenvalue A, and coincides with the plane passing through ro and normal to
N, (see eqn (2.B.6)), while A, and N, define the worst plane. Furthermore,
N,*Xi is the distance of the ith atom from the best plane and N i x ; and N/TXI
are useful to prepare a diagram of the projection of the system of atoms on
to the best plane.
The search of the best plane is remarkably simplified if Cartesian systems
are used (then G = G* = I).
Best line through a set of points

This problem is strictly connected to results described in the above
subsection. It may be shown that the least squares line passes through the
centroid ro of the atoms and is normal to the plane corresponding to the
eigenvalue A,.
Principal axes of a quadratic form

Let
be the quadratic form. Finding its principal axes is equivalent to finding the
directions n in which q is stationary. As in calculating the best plane through
a set of points, the problem may be solved via the Lagrange multipliers by
minimizing
The derivative of q' with respect to N brings to
from which
(A - Al)N* = O
where A = QG*, and N* = GN is the general eigenvector the components of
which are referred to the reciprocal axis. The eigenvalue A gives the value of
q in the n direction. Indeed, if (2.38) is written as AN* = AN* and both sides
are multiplied by N we obtain
Substituting the three eigenvalues A,, A,, A, into (2.38a) provides the three
eigenvectors N:, N;, N: which represent the principal axes of q.
If the quadratic form is referred to the reciprocal basis (i.e. q * = H$H =
Pllh: + 2P12hk+ . . . + P3,12) the problem may be solved in the same way
on condition that G* and N* replace G and N respectively. As an example
let us determine the principal axes of an atomic thermal ellipsoid for which
PI1= 0.00906, P12= -0.00049, P13 = -0.00102, PZ2= 0.00401, P23 =
0.00038, P3, = 0.01424. Let the orthohombic unit cell parameters be
a = 8.475, b = 10.742, c = 5.8991$. The function to minimize is
which, derived with respect to N*, leads to the condition
Since N = G*N*, that condition may be written

($G - Al)N = 0 .
The corresponding secular equation is
Expansion of the determinant produces the cubic equation
which has solutions A, = 0.483, AB = 0.448, A, = 0.677. Using the first root
gives
0.1677 -0.0565 -0.0354
(-0.0352
-0.0733
-0.0203
0.0438
0.0132)
0.0125
(I::)(!).
n3,
=
Since the three equations are linearly dependent n,, and n2, can be found in
terms of n,,: n,, = +0.2592n3,, n2, = 0.185n3,. The eigenvector N, will
have unitary modulus (remember that n,,, n,,, n,, are the components of
N, in A) if n3, = 0.1515. Therefore N, = -[0.0393, 0.0280, 0.15151. In an
analogous way N~ = [-0.0139, -0.0866, 0.05871 and N, = [-0.1097,
0.0210, 0.04921. Since
each eigenvalue A fixes the value of q* along the corresponding eigenvector.

If this conclusion is referred to the tensor U* = j3/(2n2), (see eqn (3.B.6)
the following relations follow:

A3
( ~ 2=)-~ (2.38b)
2n2 )
~ mean-square displacement along the ith principal axis.

where ( u ~ is) the
The reader will easily find, for the example above, that the root-mean-
square displacements along the principal axes are 0.156& 0.151 A and
0.185 A respectively.
It would be worthwhile remembering two basic properties of the
eigenvectors:
(1) eigenvectors corresponding to different eigenvalues are orthogonal to
each other;
(2) the matrix M = (GV)-l, where V is the eigenvector matrix
transforms the basis A into a Cartesian coordinate system A' in which the
axes are the eigenvectors of $. Indeed, according to (2.E.8), $ transforms
into $' = VG$GV. Because of (2.38b)
so that
Metric considerations o n the lattices

The results obtained so far can be used to characterize lattices and their
properties.
Niggli reduced cell

A unit cell defines the lattice: conversely any lattice may be described by
means of several types of cell. However, a special cell exists, called ,the
Niggli reduced cell, which uniquely describes the lattice:[41it is primitive and
is built on the shortest three non-coplanar lattice translations (the
~ e l a u n a ~reduction
[~] procedure is also a suitable tool for the identification
of a crystal lattice: readers will find careful description of it in some recent
papers["']).
A unit cell characterized by the three shortest non-coplanar translations is
called a Buerger cell.[92101 Several algorithms can be used to obtain it. The
easiest is: lattice vectors of magnitudes r,,,,, are calculated by (2.3) where
u, v, w vary over the smallest integers set (usually between 3 and -3). The
smallest three non-coplanar translations will be the Buerger cell edges. This
cell is, however, not unique: if it is, then it coincidesiwith the Niggli cell.
For 7 of the 14 Bravais lattices a unique Buerger cell exists,[''] while in a
face centred cubic lattice (see later) two Buerger cells can be found. In
other lattice types up to five types of Buerger cell can be found (the values 4
and 5 occur only in triclinic lattices) according to whether some conditions
on the parameters of the conventional cell are satisfied or not. For example,
the triclinic lattice described by a Buerger cell with
may be described by means of four other Buerger cells having the same a,
b, c values, but with
a = 60°00', P = 86'24', y = 75'31';
a = 120"001, P = 93'36', y = 100'48';
a = 117'57', /3 = 93'36', y = 104'28';
a = 113'58', P = 100°48', y = 104'28'.
It will be shown later that only the first of the five cells is the Niggli cell.
If gij are the elements of the metric matrix, the Niggli cell is defined by
the following conditions:[129131
1. Positive reduced cell (all the angles <90°). Main conditions:
Special conditions:
(b) if gz3= 1/2gZ2then g126 2g13; if g13 = 1/2gll then g12s 2gZ3;if g12=
1/2gll then g13 s 2g23.
2. Negative reduced cell (all the angles 290'). Main conditions:
(b) if k231 = 1/2g22 then g12 = 0; if lg131= 1/2gll then g12= 0; if lglzl =
1/2gii then g13 = 0; if (Ig23l + Ig13l + Ig121) = 1/2(g1, +gZ2) then gll a
2 18131 + 18121.
The main conditions define a cell based on the three shortest non-
coplanar vectors. Conditions (a) break down ambiguity when two cell edges
are equal, conditions (b) define the Niggli cell when there is more than one
symmetrically independent Buerger cell.
As an example of systematic ambiguity let us consider the face-centred
cubic lattice with cubic edge a. If we move to the primitive unit cell by
means of the appropriate matrix quoted in Table 2.C.1 we get
g;1= g h = gj3 = a2/2;g;, = gi3 = gi3 = a2/4.
If we move to the primitive cell by means of the transformation matrix
1(1/2 112 011 1/2 112 011 0 1/2 11211 then we get
Both the primitive cells are Buerger cells but the second violates the
conditions (a): thus the first is the Niggli reduced cell.
Matrices which derive Niggli cells from Buerger cells are given by Santoro
and ~ i ~ h e l l .A[ 'very
~ ~ efficient algorithm to derive the Niggli cell from any
primitive cell is described by Krivy and ~ r u b e r . ~ ' ~ ]
For any Bravais lattice ~ i g g l i [defined
~] the algebraic relations that the gijs
of the reduced cell must satisfy. The type of Bravais lattice may be thus
derived from the Niggli cell just by comparing the found with the expected
relations. For example, in a face-centred cubic lattice the gij of the Niggli
cell must satisfy gll = g22= g33, gI2= g13 = g23 = g11/2.
The use of automatic procedures devoted to identify the Niggli cell may
yield incorrect conclusions as a consequence of errors in the cell parameters
or of rounding errors in the calculations. Some auxiliary procedures recently
suggested by different are less sensitive to these error sources.
The final steps from the Niggli cell to the conventional cell may be
performed by means of suitable transformation matrices.["] It would be
worthwhile recalling that the lattice symmetry determined via the Niggli cell
is only of metric nature, and that may be equal to or larger than the
symmetry of the crystal structure.
Reduced cells may be used:
As a useful step for the correct definition of the space group (see also
Chapter 3). An advisable sequence may the following:[181from the
conventional cell to a primitive cell, and then to the Niggli cell; analysis
of the latttice symmetry, analysis of Laue symmetry and of systematic
extinctions; space group choice.
As an effective tool for the identification and characterization of
crystalline materials['91(as an alternative to powder methods in which the
identification is based on matching diffraction positions and intensities).
An advisable sequence may be: a unit cell is determined, the reduced
cell is derived together with derivative supercells and subcells (derivative
cells are calculated to overcome possible errors made by the experimen-
talist). These cells are checked against a suitable file containing as
complete as possible a file containing crystallographic data (the NBS
Crystal Data File handles data of more than 60 000 materials).
It could be asked now if Niggli cell expresses some geometrical property.
G r ~ b e r [has
~ ~ shown
] that a cell is a Niggli cell if and only if the following
conditions are fulfilled:
+ +
(1) a b c is a minimum when calculated for all primitive cells of the
lattice;
(2) 1x12 - a1 + 1x12 - PI + 1x12 - yI

+
lcos crl + lcos p1 lcos yI - max. for the cells
defined in (1).
lcos cr cos p cos yl
As an example let us consider the triclinic lattice defined by the primitive

cell
The Niggli cell is obtained from the previous one by application of the
matrix 001/100/111:
a = 8.070 A, b = 9.562 A, c = 12.434 A,
a = 100.97", /3 = lO6.54", y = 110.03".
This cell satisfies the geometrical properties suggested by Gruber.
Sublattices and superlattices

The term superlattice is commonly used for a structure closely related to a
parent structure and having cell dimensions superior to that of the parent
structure. The superlattice then belongs to a subgroup of the parent
structure: indeed increasing cell dimensions causes a loss of translational
symmetry. Let
A= (B) and A' = (f:)

be two triplets of non-coplanar vectors defining two primitive cells in two
different lattices L and L' respectively. The cell in L' is related to the cell in
L by
A' = MA. (2.39)
Consider four important cases:
1. The mii elements are integers and M = 1. Then L and L' will coincide.
2. The miis are integers and M > 1. Then A' defines a new lattice L' called
a superlattice of L whose elementary cell is M times larger than the
primitive cell in L.
3. M = Q-I where Q is a matrix with integer elements for which Q > 1. In
this case A = QA' and L is a superlattice of L', or, with equivalent
terminology, L' is a sublattice of L. It should be noted that if L' is a
sublattice of L then L'* is a superlattice of L*.
4. M is a rational matrix. This case is described in the following subsection.
Super- and sublattices are frequently related to important properties of
the crystals. For example,
(1) twinning by merohedry takes place only if a superlattice exists with
symmetry higher than that of the crystal lattice
(2) crystallographic phase transitions often take place between the struc-
tures for which one lattice is a superlattice of the other. The knowledge
of the possible superlattices of a given lattice limits the set of possible
structures of a new phase. That is of particular usefulness when the
phases are simultaneously present in a polycrystalline sample and
Table 2.3.
diffraction peaks having identical positions are generated. In particular,

if an order-disorder transformation occurs, the ordered phase is
characterized by a cell larger than that of the disordered phase;
magnetic structures are based on cells often larger than those of the
corresponding conventional chemical structure (see Figs. 1.F.2 and
1.F.3).
For any value of M there is a finite number of matrices M that produce
distinct super lattice^.[^^] This number quickly gets large with M: in Table 2.3
the unique matrices M are given generating superlattices for M = 2, 3. The
unique matrices generating sublattices for M = 112, 113 are obtained by
applying the same matrices to the reciprocal lattice and then by calculating
the lattices that are reciprocal to the resulting superlattices.
Coincidence-site lattices
Most materials of technological interest are used in their polycrystalline
form. Their mechanical and chemical properties are controlled to a large
extent by the boundary between crystallites. The energy of a polycrystal is
higher than that of a single crystal with the same mass: the additional energy
is stored in the grain boundary areas, and depends on the orientations of the
neighbouring grains. Thus modern treatment of these materials tend to
optimize the size of the grains and quality of the grain boundaries.
The mathematical model of the crystalline interfaces is today based on the
properties of coincidence-site lattice (CSL) and related lattices. Consider
two lattices L and L' with bases A and A'. Without loss of generality it will
be assumed that the two lattices have one lattice point in common, taken as
the origin of the coordinate systems. Let N and N' be two matrices with
integer elements. The lattices L and L' will have a common superlattice if a
lattice point of L (defined by NA) can be found which is also a lattice point
of L' (defined by N'A'): then NA = N'A', or also
where X, = N'-'N is a matrix with rational elements. In this case the CSL is
defined as that superlattice (at 1 or 2 or 3 dimensions) of L and L' which
contains all (and only) the lattice points in-common to L and L'. Note that
several other lattices could be defined having points in common with L and
L' but all of them will be superlattices of the CSL.
To determine the CSL one has to find[22.231 a factorization of Xc of the
type X, = N'-'N with the smallest possible values of N and N'. If No and NI,
satisfy this condition then No and NA indicate the reciprocal fraction of
coincidence points (degree of coincidence) in lattices L and L' respectively,
and the CSL basis will be NoA = N S ' . If N or N' are sufficiently small (a
large fraction of points of one of the two lattices consists of coincidence
sites) and if the boundary coincides with a dense net plane of CSL, then the
boundary energy per unit area will be a minimum.
Analogously, two lattices L and L' will have a common sublattice if two
matrices N and N' (with integer elements) can be found such that
N-IA = N I - I A I or also
A' = XdA,
where Xd= N'N-I is a matrix with rational elements. In this case the
displacement-shift-complete lattice (DSC) is defined as the sublattice with
the largest volume of the primitive cell. All lattices which are sublattices of
both L and L' will be sublattices of DSC.
The DSC may be determined by means of a factorization similar to that
used for X,: again we will look for the smallest possible values of N and N'.
If No and Nh satisfy such a factorization process then the DSC basis will be
NilA = NA-'A', which defines a cell with volume l/No times the volumes of
the cell defined by A and 1/NA times the volume defined by A'.
It may be also shown that:
1. The CSL (DSC) of the reciprocal lattices is the reciprocal lattice of the
DSC (CSL) of the two lattices.[241
2. The coarsest lattice which contains all vectors of the form u + u', where
u and u' are vectors of L and L' respectively, is the DSC lattice.[251With
respect to the energy of the grain boundaries DSC lattices have the same
importance as CSL: indeed translations by DSC vectors do not destroy
the coincidence sites. Such vectors are the geometrically possible Burgers
vectors (energy considerations will dictate the most probable of them) of
dislocations in grain boundaries.
If the two lattices L and L' are congruent then one can be transformed
into the other by means of a rotation: this is called coincidence rotation if L
and L' have a CSL in common. The ratio between the volume of the CSL
unit cell and the volume V of the crystal unit cell is called the multiplicity of
the CSL and is denoted by Z (the analogous ratio for the DSC cell will be
1/Z). The determination of all possible coincidence orientations with low
values of Z is an important premise for the understanding of grain
boundaries. If A' = XA defines one of the required orientations, then, owing
to (2.21),
G' = XGX. (2.40)
Several attempts have been made to find the general solution of (2.40).
Special methods for the solution of this problem were developed for
cubi~,[~h ~ ,e~~' ]a ~ o n a l , and
[ ~ ~r h, ~ m
~ ]b o h e d r a l [lattices.
~~] The problem may
be so stated: determine all the rotation angles 8 about a given lattice axis
[UVW] which generate CSLs. It may be shown that in cubic lattices a CSL is
obtained by a rotation 8 about an axis [uvw] coincident with a lattice
direction if
tan (812) = (u2 + v2 + ~ ~ ) l ' ~ / r n
where m is an integer. A unitary greatest common divisor among m, u, v, w

may be chosen withoyf restricting generality. Denoting by N the number of
odd integers among m, u, v, w, the multiplicity I: is given by
m2+u2+v2+w2
I:=
a
where a = 1 if N = 1, 3, a = 2 if N = 2, a = 4 if N = 4.
An example is illustrated in Fig. 2.5(a). The lattice direction [OOl] has
been chosen as the rotgion axis: if we choose m = 3 then N = 2, a = 2,
I: = 5, 8 = 36'52'. It may easily be seen from the figure that the volume of
the CSL unit cell is f o r m ~ dby the vectors [210], [120], [OOl].
In Fig. 2.5(b) the CSL with I: = 7 is shown for the cubic (111) plane
( 8 = 38.21"). The black and the cross-like points belong to lattices 1 and 2
respectively. The CSL pgints are those on which cross-like and black points
overlap. Black and cross-like points are both part of the DSC lattice, whose
units are evidently smaller than crystal units. If lattice 1 is shifted by a DSC
unit with respect to latticg 2 the pattern shown in the figure is reproduced in
another position.
Twins
Twins are regular aggregates consisting of individual crystals of the same
species joined together i i some definite mutual orientation. There are three
principal types of twin: growth twins (produced by accident as the crystal
grows from its initial nucleus), deformation twins (considered as a means
of relieving the strain in$uced by some applied stress), and transformation
twins (the product of a polymorphic transformation, i.e., when a higher
symmetry crystal is cooled and converts to a lower symmetry structure).
From a geometrical p ~ i n of
t view a twin is characterized by the symmetry
operations which relate one individual to the other individuals in the
composite crystal. The operation is very frequently a rotation through n
about a zone axis (in this case the axis is the twin axis and the twin is a
rotation twin), or a reflection in a lattice plane called the twin plane (the
twin is then a reflectipn'twin).
-, *
i Rotation twins of n/3, n12, 2n/3 also occur Fig. 2.5. (a): 6SL lattice and CSL unit cell for a
but are less common. rotatiop of a ~ ~ blanice i c about I0011. (b): CSL
Obviously diad, tetrad, or hexad axes cannot be considered twin axes (at and ~ s ~ j a t t i c efor
s the cubic (111) plane.
least for rotation through n). If a triad is a twin axis the twinning operation
may be equivalently described as a n/3, n, or 4n/3 rotation about the axis:
conventionally the n rotation is preferred.
A twin is called a contact twin if the two components are joined in a
plane (known as fhe composition plane). In the case of a rotation twin the
composition plane is parallel to the twin axis, in reflection twins the
composition plane is parallel to the twin plane. In interpenetrating twins
the twin c~mponentsintergrow so as to generate an irregular interface
between components.
Multiple twins consist of three or more components. If the twinning
operations relating adjacent components are all identical then the twins are
known as lamellar or polysynthetic twins (the components have a lamellar
form parallel to the composition plane). Polysynthetic twins may be on a
microscopic or macroscopic scale.
Supplementary information on the most common types of twin and some
morphological details are given in Appendix 2.5. Here, we are interested in

the characterization of the twins in terms of lattice geometry. Twins are
usually classified in accordance with the (pre-X-ray) theory developed by
the French crystallographic school (Bravais, Mallard, Friedel). A recent
classification[319321
divides the twin kingdom into two species, TLQS and
TLS. TLQS (twin-lattice quasi-symmetry) twins are characterized by two or
more reciprocal lattices differently oriented, giving rise to double or
multiple diffraction spots: they are generally recognizable simply by optical
observations. TLS (twin-lattice symmetry) twins show a single orientation of
the reciprocal lattice so that they give rise to single diffraction spots: they
are optically indistinguishable, and their presence may be indicated by a
high irreducible residual R in the attempted crystal structure determination.
The twin lattice is the lattice with the smallest cell that is common to both
(or to all) individuals of the twin. Such a lattice will show perfect continuity
at the composition surface of the twin in case of TLS, and will suffer some
deviation for TLQS twins. The slight deviation, expressed in degrees, is
referred to as the twin obliquity w , and is usually less than 6". Occurrence
of TLQS mostly depends on special relationships between cell parameters
while the existence of TLS depend on the symmetry properties of the
lattice.
A further criterion for subdividing twins is the twin index, n:
n = (vol. per node in twin lattice)/(vol. per
node in crystal lattice).
For n > 1 the twin lattice is a superlattice of the crystal lattice (see CSL in
previous subsection); for n = 1 the twin lattice coincides with the crystal
lattice. TLS twins with n > 1 are often found in highly symmetrical lattices,
for example, in minerals.
The twin law specifies the mutual orientation of the two twinned crystals:
it is usually expressed in terms of the rotation necessary to bring one of the
lattices into coincidence with the other. According to ~ r i e d e l [ twinning
~~]
may occur because of the accidental presence in a lattice of a net and a row,
sufficiently dense, which are exactly or approximately perpendicular.
Friedel's necessary (but not sufficient) condition for twinning in triclinic
lattices (specialized conditions can be found for the other lattice types) is
that
g11:g22:g33:g12:g13:g23
are, or approach rational numbers. Friedel's rule may be obtained by
requiring G' = G in (2.40), and may be so expressed:[341'a crystal may twin
if the condition
is satisfied where M is a matrix with rational coefficients'.

Obviously only matrices with M = 1 have to be considered: furthermore
symmetry operators M must also be excluded since they are trivial solutions
of (2.41). If M satisfies (2.41) -M = (-1)M also satisfies (2.41): therefore,
for any twin rotation through a about the direction [uvw]there should exist
a twin operation which is the combination of M and of the inversion of the
lattice. For example, if a = 180°, (-1)M will correspond to a reflection in
the plane (hkl) normal to [uvw].
Twins of special interest are TLS twins with n = 1. They were called by
Friedel twins by merohedry since the crystal symmetry is merohedry of
order n (subgroup of order n ) of the symmetry of its lattice. Accordingly,
merohedrical twins have one or more symmetry operations which are
present in the lattice and not in the crystal. In order to explain their
diffraction behaviour, they may be divided into two classes:[35]
Twins in class I show the same crystal Laue symmetry as the lattice
symmetry. Then the twin operation belongs to the Laue symmetry of the
crystal: in these conditions the set of intensities collected from the twin
coincides, except for anomalous scattering, with that which would be
measured on a single crystal. Structure determination is therefore not
hindered but the determination of the absolute configurations (using the
methods described on p. 97) is impossible.
Twins in class I1 are characterized from the fact that the Laue symmetry
of the crystal is lower than the crystal lattice symmetry. Then at least one
of the twin operations belongs to the lattice symmetry but not to the TWINS
Laue symmetry of the crystal. Twins by hemiedry, tetartohedry, and
ogdohedry can be found: they are made by two, four, and eight crystals
respectively.
A scheme for twin classification is drawn in Fig. 2.6. In Fig. 2.7 some
examples of twinning are collected.[361
In Fig. 2.7(a) the projection along the b axis of a monoclinic lattice with
nyy yy
~~s-twins twins
P = 90" is shown, together with its twinned lattice (the assumed twin
operation is the mirror plane m perpendicular to a , but we could also
choose the mirror plane perpendicular to c ) . The o misfit is intentionally
/\
class I class II
exaggerated. Fig. 2.6. A scheme for the classification o f twins.
TLQS, n=l(oiO"), TLQS, n1

. (w-O")
TLS, n =l,classII
Im
Fig. 2.7. Examples of twins (from reference

1361).
In Fig. 2.7(b) the projection along the b axis of the monoclinic lattice of
1-aspartic acid with a = 7.617, b = 6.982, c = 5.142, /3 = 99.84" is shown: the
lattice is also shown after a two-fold rotation about the a " axis. It is easily
seen that 2a of the original lattice nearly coincides with 2a-c of the rotated
lattice, while a * is the common reciprocal lattice of the two lattices. The
twin lattice unit cell is defined by a ' = 2a, c' = c, b' = b, P' = /3 (but also a
B-centred orthohombic cell may be chosen, four times larger than the
original cell).
In Fig. 2.7(c) the projection of a hexagonal lattice along the c axis is
shown. If the space group of the crystal is supposed to be ~ 3a TLS , twin of
class I1 may be generated by reflection with respect to the plane m drawn in
the figure. The diffraction pattern will show then R3m symmetry.
In Fig. 2.7(d) the classical penetration twin of fluorite (CaF,) is described.
Two cubic lattices, referred by the twin operation (111) mirror plane, are
viewed along the direction [i10]. The twin lattice has a volume three times
the volume of the original cell.
An elegant derivation of twin laws by merohedry has been recently
proposed.[371Denote by H and G the point-group symmetry of the crystal
and the point group of its lattice respectively (G may be obtained by the
process of cell reduction). Since H is a subgroup of G, the coset
decomposition of G with respect to H may be made (see Appendix l.E).
Any system of g, operations (g, E G, g, 4 H) used for the (left) coset
decomposition will lead to the superposition of the lattice onto itself, and
therefore will contain the possible merohedral twin laws for a crystal of
point symmetry H in a lattice of point symmetry G.
For example, a-quartz crystallizes in P3121 with a = 4.913 and c =
5.404 A. The crystal point group is 321 and the metric symmetry is 6/mmm:
thus
H (1; 2,1101;2[im];2[orol;3[ooil;3fhi1)
G {H; 2pm1;2[iiol; 2[nol; 2[ziol;6[m1];6foo11;
1; 40011; z[lio]; 2[1zol;2[z101:3pmI;
3&,; ~[Ool]; ~f0011;2[110];2[100];2[010]).
The coset decomposition is therefore
G = H U (2[,1]H) U ( W U ( ~ [ 0 0 1 , ~ ) .
It may be seen that 2,,,], I, and correspond to the classical twin laws
for DauphinC, Brazil, and combined twinning respectively. The twin-related
reflections are therefore (hkl), (&El), (&El), (hki).
The same procedure, applied to a crystal with point group H = 4/m, will
decompose the metric symmetry group G = 4/mmm into
G = H U (2[010]H).
Hence the twin-related reflections are (hkl) and (&kt).
In the case of hemiedry (twins in class 11, two individuals) two reflections
which are not equivalent by Laue symmetry contribute to a twin
reflection:[381then
IFtn12 = a IFHl2+ (1 - a)IFKI2
+
IFMI' = (1 - a ) lFHI2 a IFKI2
where IF,^^ and 1FtKI2are twin reflection intensities, a is the volyme

fraction of crystal 1, and IFH12#1FKI2are the intensities of the two
overlapping reflections. The algebraic relation between H and K is fixed by
the twinning law. A trial estimate of a may be obtained via suitable
statistical tests on diffraction data:[3g-411a may then be refined as an extra
parameter in adapted crystallographic least-squares programs.
An alternative procedure may be that of 'detwinning' reflection data: i.e.
and IFKI2can be calculated from observed IF,^^ and IF,^^ according to
Previous knowledge of a is a necessary condition for the application of the

above formulae: these cannot be applied if a = 0.5. However, alternative
methods are available for such a specific case (see [36] and references
quoted therein).
Calculation of t h e structure factor

Let m be the order (i.e. the total number of symmetry operators) of the
space group and t the number of symmetry-independent atoms. The
structure factor may be then written as (see notation on pp. 1.52-4)
where
AH = xf
j=1
njfoj(H) x
m
s=l
exp (-H&,H) cos 2n H(R,X, + Ts) = x
j=1
f
A, (2.42a)
BH=
f
j=1
n,foi(H) x
m
s=l
+
exp (-Apis~)sin 2n H(R,X, T,) =
t
j= 1
B,. (2.42b)
Aj and Bj are the contributions of the jth atom and of its symmetry
equivalents to AH and BH respectively, pis is the 3 X 3 temperature factor
matrix for the atom j in symmetry position s, nj is the occupation number of
atom j, defined as mjlm, where mi is the number of different atomic
positions which are symmetry equivalent to the jth atom. Accordingly,
nj = 1 for an atom in a general position, n, < 1 for atoms in special positions
(the use of nj allows that summation over s is always extended from 1 to m,
independently of the atomic site type). If the jth site is only partially
occupied because of some statistical disorder then nj will be proportionally
reduced.
The calculation of (2.42) will be simpler if, for a given H the symmetry
equivalent indices H, = HR,, s = 1, . . . , m are calculated. In this case (see
eqn (3.36)) HR,X,. may be replaced by H,x, and (see eqn (2.E.8)) H~,,H by
As ~ j h .
We will then write eqns (2.42) as
C [ c njfhi(H) exp (- li,p,~,)cos 2 n (HsXj + ATs)]

f m
AH =
j=l s=l
t m
BH= C
j=l
[Cnj$(H)
s=l
exp (-A,~,H,) sin 2 n (H,x, + AT,)]
For each j the maximum number of ujs (and vjs) to calculate is 24. Indeed, if
the space group is centrosymmetric (origin on a centre of symmetry), s may
vary only over the symmetry matrices not referred by the inversion centre;
AH is then multiplied by 2 and BH is settled to zero (origin on a centre of
symmetry). For space groups with centred unit cell s may vary only over the
matrices not referred by non-primitive lattice translations: AH and BH are
then multiplied by the centring order of the cell.
Scattering factors f,,. have been tabulated[421for all elements: their
accuracy depends on the wave functions and on the numerical methods
used. The values of fg,at the actual sin 8/A value may be obtained from the
tables by interpolation. A more usual procedure is to approximate
scattering factors by the sum of one or more Gaussian functions: for
accurate structure factor calculations four Gaussians are used according
to 1431
4
fo(8) = C a, exp [-b, sin2 8/A2] + C.
i=l
It should be noted that only nine parameters have to be stored for each
element.
Calculation of the electron density function

According to (3.46) we have to calculate
p(x, y , r ) = 2/V 2 2' 2' [AHcos 2 n ~ X+ BH sin Z ~ H X ] , (2.44)
where the prime to the summation implies that only half of the reflections
(0, k, 1) have to be considered.
The calculations may be performed in a trivial fashion starting from the
list of symmetry independent Fhkl, generating symmetry equivalents, and
evaluating the sum in (2.44) for every X. The crystal symmetry may be more
conveniently exploited by combining in advance the terms containing the
symmetrical structure factors, thus obtaining an expression valid for that
given symmetry. The summations in (2.44) are then limited to the set of
independent Fhklvalues. For example, in Pmmm
Then
p(x, y) Z) =
1
Vh=0
x
"
[COS2nhxTl(hkz) + sin 2nhxT2(hyz)]
which is again a one-dimensional transform for fixed y and z :

R(xyz) = xh
T(hyz) exp (-2nihx).
Why is this procedure preferable to the direct use of (2.44)? If p is

calculated directly from (2.44) the work to do is of the order n6 (say
nhnknlnxnynz,where nh is written for the number of different h values,
etc. . . .); if the Beevers-Lipson technique is used the work for the three
transforms is of the order n 4 (say nhnknlnz+ nhnknynz+ nhnxnyrtz).Thus for
large values of n the Beevers-Lipson technique makes a considerable
saving. Furthermore, within every transform a single line of Fhk,is involved
and, after its completion, the line is no longer necessary: thus S(hkz) can
replace the corresponding set of Fhklvalues after the first transform, T(hyz)
can replace S(hkz) after the second transform, and R(xyz) can replace
T(hyz) at the end of the process.
A further reduction in the number of operations is obtained if the fast
Fourier transform[441(FFT) algorithm is used. The technique factorizes
one-dimensional into two-dimensional transforms by a procedure quite
similar to that of Beevers and Lipson (see Appendix 2.1). The saving of the
computer time may be so described: a one-dimensional transform with n
values of h and x requires n2 operations, while the FFT technique reduces
this to 2n log, n. The saving is not impressive for n < 16 but becomes large
for larger values of n (as in protein crystallography).
Electron density maps are usually not printed: frequently peak search
routines are used to locate electron density maxima and to print their list in
order of peak intensity. A typical process for locating maxima involves two
stages:
1. The p values observed at 19 grid points (see Fig. 2.8), centred on that
point for which p is largest, are stored. Six of these points are nearest
neighbours, and twelve are next nearest neighbours. The full peak is
expected to be Gaussian in shape, so it could be fitted by the function
p(xyz) = exp (a + bx + cy + dz + ex2 +fy2 + gz2 + hyz + kzx + lxy).
2. By least squares the ten values a, . . . , 1 are obtained which minimize
C19 (In p,,, - In p,,,J2 (nine degrees of freedom are retained). Then the
Fig. 2.8. The 19 grid points used for peak search position of the maximum is obtained from the conditions 6p/6x =
routines. 6p/6y = 6 p / & = 0.
The m e t h o d of least squares

Linear least squares
A very common problem found in physical sciences is the following: given a
set of experimental observations ( 5 ) and a theoretical model which, from
trial values of some parameters, generates a set of calculated f values, find
the values of parameters which give the best fit to the data, estimate their
accuracy, and comment on the adequacy of the assumed model. There are
several approaches for answering this problem. Because of its wide use in
crystallography we will mostly be interested in the method of least squares.
Alternative approaches are briefly mentioned on pp. 108-9.
Suppose that a set of n experimental observations
F-(fljf2,.. ,fn)
is available for which:
(1) h is subjected to some random error e, due to the finite precision of the
measurement process;
(2) 4 is known to linearly depend on a set of m S n parameters
X' (xl, x2, . . . , X,).
Then the observational equations may be written[451as
where A = {aii) is the n x m design matrix (n rows, m columns) of rank m

which is assumed to be known, and E = (el, e2, . . . , en). The condition
n > m emphasizes the fact that the parameters are overdetermined. Our
problem is to obtain satisfactory estimates g of the parameters X and, at the
same time, estimates of the variances of such estimates. We will assume that
errors e, have joint distributions with zero means and finite second
moments: i.e.,
( F ) =F"=Ax (2.48)
where Mf = ((F - F")(F - F")) is the variance-covariance matrix of rank n,

and
where pii are the correlation coefficients.

The case Mf = I (uncorrelated errors, equal variances) was treated by
Gauss who showed that, among the various X, the most satisfactory was
that (say A) for which the residual
s= x v? = w = minimum
where v - i=l
(v,, v2, . . . , vn) is given by

V=F-AX.
This result suggests that, when Mf is a diagonal matrix with elements
(in this case errors on h and fi are statistically independent and pii = 0 for
i # j), the most satisfactory estimate 2 is that for which the weighted
deviance
2 wiv?= x wi(A - fy)2 =

n n
S= = VMT'V = minimum (2.51a)
i=l i=l
where wi = 1/o?.
If Mf is not diagonal then
= \iWV = bM; 'V = minimum. (2.51b)

In most applications Mf is taken to be diagonal.
Relations (2.51) suggest that, for a general variance-covariance matrix,
the most satisfactory solution 8 is given by
S = bMTIV = minimum. (2.52)
In accordance with Appendix 2.F, the relation (2.52) leads to the so
called normal equations
B ~ = D (2.53)
where
B = AM;~A, D = AM;+. (2.54)
B is a m x m symmetrical (bij = bji) square matrix. From (2.53) the
least-squares estimate of X is obtained as
It may be easily seen that 8 is independent of the scale of the covariance

matrix. Indeed suppose that Mf = K,Nf, where Nf (the working matrix) is
known and K, unknown. Substituting K,Nf for Mf in (2.55) eliminates K,.
Reliability of the parameter estimates

It may be shown (see Appendix 2.G) that the variance-covariance matrix
M, relative to the unbiased estimate 8 is given by
M, = ((8 - x ) ( ~ ) X ) ) = B - ~ . (2.56)
If Mf is known within the scale factor K, (Mf = K,Nf) then
where B, is the working matrix calculated via the known Nf matrix.

According to (2.57) M, is completely determined only if K, is known:
luckily (see Appendix 2.H) an unbiased estimate of K, is available from the
least-square treatment:
so that the unbiased estimate of M, is given by
k, is often called goodness of fit and denoted by GofF
Linear least squares with constraints

Assume that the parameters xi are not independent but are constrained to
satisfy the set of b linear equations
OX = Z (2.60)
where Q is a b m matrix of rank b, and Z is a column vector with b

components. In this case the unbiased estimate of X may be obtained by
considering the variation function
- 2X(QX
S, = V M ~ ~ V - Z)
where I is a column vector with b components (the Lagrange multipliers).

The condition 6s" = 0 brings to
2 V M 7 ' 6 ~- 2X(Qx - 2 ) = 2(F - AX)M;'SV - 2XQ6X = 0.
Since 6V = -A6X, we obtain
- X A M ~ ~ AXQ = 0.
FM~A +
Calling the solution X, gives
which, according to (2.55), may be written as

Xa = XB(),.
The value of I may be obtained by post-multiplying both sides of (2.62)
by B -andl ~by requiring that (2.60) is satisfied:
X = (zKGR)(QB-'Q)-'.
Eliminating I in (2.62) gives
X, = 2 + ~ - l Q ( ~ e - l Q ) - l ( z- 0 2 ) (2.63)
The variance-covariance matrix for X, is (B-l is the variance-covariance
for X)
M, = B-1 - B-~Q(QB-~Q)-~QB-~. (2.64)
By a procedure analogous to that described in Appendix 2.H it may be
shown that the expected value of the weighted deviance S, is given by
(3,) = (VN-~V)/K, = (n - rn + b)lK,
which is larger (see Appendix 2.H) than the expected value of the
unconstrained residual
(3) = (VN-'V) = (n - m)lK,.
Non-linear (unconstrained) least squares

Let us suppose that the observations F = (fl, . . . ,f,) do not linearly depend
on the X parameters. Then the residual S will have several local minima
(see Fig. 2.9) and condition (2.51) will not usually provide a satisfactory 4
estimate for the parameters, unless a good approximation S
XO = (x:, xi, . . . , x:) of X is available. To this aim each may be expanded
in a Taylor series about the point XO:
+' 2 (-) 6"

2j,p=l &j6~p
6 . ~ ~ 6. . ~. ~ +. (2.65)
parameters
Fig. 2.9. Examples of local minima for the
residual S.
If X is sufficiently close to X0 second and upper derivatives can be

dropped in (2.65) (Gauss-Newton approximation) to get
where derivatives are calculated in XO,and
In matrix notation
where
is the design matrix (compare (2.66) with (2.47)).

If the Taylor expansion is valid, the problem has been reduced to a linear
one: thus an estimate for AX, say A% will be obtained in the same way as
described in the previous paragraphs, from which more satisfactory values
+
X = X0 A% may be derived. Mf can still be considered as the variance-
covariance matrix of AF if X0 is a good starting model.
Since in (2.66) second and upper derivatives have been ignored, AA will
not be an unbiased estimate of AX. Therefore for non-linear problems an
iterative procedure must be followed, according to which the new para-
+
meters X = X0 A 2 are used as a new starting model for the application of
(2.66): in each cycle the derivatives will have changed so that the design
matrix has to be recalculated. The iterative process will continue until the
changes A% are very small or zero: then it may be concluded that the
least-squares procedure has converged, and that the refinement of the
parameters has been completed.
Least-squares refinement of crystal structures

Least-squares techniques are widely applied in crystallography. Examples
are: refinement of unit cell parameters from diffraction angles (see p. 300);
calculation of least squares planes for molecular fragments (p. 74), thermal
motion analysis (pp. 117-20), calculation of the Wilson plot (p. 322), and
profile analysis in powder methods (pp. 109-17). In the following we will
describe the application of least-squares techniques to crystal structure
refinement, by taking into account two different situations:
1. The Rietveld technique for the analysis of powder diffraction data. The
entire diffraction profile is calculated and compared with the observed
step profile, point by point: the model parameters are then adjusted by
the least-squares method. We will describe this technique on pp. 109-17.
2. A large number of moduli JFHJo are measured. Parameters are then
refined in order to minimize the difference between the IFHl,s (structure
factors moduli calculated from the structural model) and the IFHlos.This
technique is of general use for single-crystal data and of remarkable
usefulness also for powder data; it is described in the following.
To carry out least-squares refinement of a crystal structure let us associate
a weight wHto each IFHI,.Then, in accordance with (2.51), we want to make

minimum the quantity
According to the previous subsection if a satisfactory model X0 is

available, we expand in a Taylor series: then
which, according to (2.51b), may be written as

S = (AF - A AX)W(AF - A AX) = W V = VMY'V.
We have denoted
IFH,lcis the modulus of the ith structure factor calculated in XO; its
derivatives are also calculated in XO.Thus eqn (2.66) is again obtained.
The normal equations may be obtained by settling to zero the derivatives
of S with respect to AX:
a lFHlc
-=o for j = I , . . . , rn.
axi
In matrix form (see eqns (2.53) and (2.54))
BAA=D
or, more explicitly,
where
D= AM;^ AF = MT'AAF = {d,} = {CwH[1FHIo- lFH;llc] lFHl

H
-1.
dxi
Then (compare with (2.55))
A% = BW1D
provides the required solution.
Parameters usually refined are: are overall scale factor (observed inten-
sities are usually on an arbitrary scale); a parameter defining chirality in
non-centrosymmetric space groups; for each atom, up to three coordinates
(fewer if the atom is on a special position); thermal parameters (1 for
isotropic motion, up to 6 for anisotropic); the site occupancy if the atomic
position is statistically occupied because of some structural disorder.
Let us now calculate the explicit expressions of the derivatives for the
various parameters. If xii is the ith parameter of the jth atom (i.e. in
increasing order of i we will refer to n,, xi, y,, zit B, or n,, xi, yi, z,, p j l l ,
p j 2 2 , pi339 pi129 p j 1 3 ) p j 2 3 ) ) then
6A 6 Bi
=cos g,-+ sin g,-
axji sxji
where A , and B, are the contributions of the jth atom to A H and BH
respectively.
Consider the various cases:
1. xji is an atomic coordinate: then
where hi,, i = 1 , 2 , 3 are the components of H, and u, v are defined by eqns

(2.43).
2 . xji is the component UGq of the vibrational tensor U; (see eqn (3.20)).
Then
For isotropic motion (Q denotes here the thermal factor, in order to

avoid confusion with the normal matrix B)
3 . xji is the overall scale factor K . Even if (FHJc

are on the absolute scale
and (FHIOon a relative one, during a least-squares refinement it is the
structural model which has to be refined and not vice versa. Therefore
from which the following estimate for K is obtained
Then the IFHJ,swill be multiplied by 1 / K in order to lead them on the

absolute scale. The IFH(,so rescaled will work as observations in subsequent
cycles of least squares.
Since 6 ( K IFH(,)/6K= IFH(,,the second derivative of S with respect to K
is zero: therefore K is the parameter which converges faster. It should be
noted that, if we erroneously had chosen to minimize S'= ,ZH w H ( K(FHIO -
IFH1c)2,the minimum could be found at K = 0 and with an extremely large

thermal motion.
4. xji is the atomic occupancy factor. Then
5. xi, is a parameter defining chirality in non-centrosymmetry space

groups. An efficient way for determining the absolute configuration is based
on measurement of the intensities of the Bijvoet pairs of reflections (see pp.
165-9) when dispersion effects occur. In this case the symmetry of IFIo in
reciprocal space degrades from Laue to the true point-group symmetry;
indeed (h k 1) and its- symmetry
- - equivalents have an intensity value
different from that -of- (h - k 1) reflection and its equivalents. Thus Bijvoet
pairs (h k I) and ( h k I) can be used for the assignment of chirality.
More often only the asymmetric set of reflections is collected: then
chirality may be fixed by the following procedure: (a) the structure is solved
and refined by using real atomic scattering factors (and preferably
absorption-corrected data); (b) the final values of scale factor, coordinates,
and thermal factors are kept intact to calculate structure factors. By merely
reversing the signs of all the ifl' two R values are obtained (see eqn (5.88)
for the meaning of R), say R: and R;; (c) the ratio T = RJR: is used in
the Hamilton test (the unweighted ratio R-/R+ is usually regarded as an
acceptable approximation). For a given value of T one looks up Hamilton's
tables R (1, N, a ) to estimate a, where N is the number of degrees of
freedom (see p. 103).
Reasons for considering the ratio test over-optimistic were given by
~ o , e r s [ ~who
~ ] suggested a better procedure: a single least-squares variable
q is introduced into every dispersion term (iqfy instead of $7) that
contributes to the structure factor. q is kept zero in the early stages of
refinement: later it should converge to values close to 1 or -1, with a
readily computable precision a(q)/q.
A very efficient alternative procedure[47] considers the crystal as an
inversion twin:
I F ~ ( X=) ~(1~-x) ( ~ ~ 1+' x IF-^^'
where (1 - x) and x are the fractions of the structure and its inverse in the
macroscopic sample. The parameter x is refined: usually it converges in a
few cycles to the final value.
6. xii is the secondary extinction parameter g (see p. 164). While primary
extinction is often negligible in single-crystal studies, the effects of secon-
stw~-extinction are usually corrected by means of the equation
(Z0)corr = Zo(1 - gzo1-l.
Since this equation corrects observed quantities, it cannot be directly used
in least-squares procedures, where it is replaced by[4s501
(1c)corr = & ( I +gZc)-l.
In this case
where L is the Lorentz factor, P a suitable polarization factor, and K the

scale factor.
Suppose now that Nf is known instead of Mf (= K,Nf): then A% will be
insensitive to the scale factor K,, while the matrix M, for the estimated
parameters will strongly depend on it. According to eqn (2.59)
mx = &B;'=-B;' 3
=
H
K1>
($) n-m
where B, = A N ~ ~ from
A , which variance and covariance values for the
parameters may be calculated. In particular the variance of the es'timated
parameters is given by
oii= (bii)- 1 "

C WH(AFH)~
n-m
where bcl is the ith diagonal element of B;'. It is easily seen that increasing
the number of observations reduces the variance of the parameters ((bii)-'
and n are mutually related in an inverse fashion so that the standard
deviation 6 varies asm). If the quality of the data is improved then
WH = 1/ah will increase: since (bii)-' varies inversely with w it follows that
standard deviations of the parameters are proportional to a,.
The standard deviations of the parameters are a useful guide during the
refinement process. If they are sufficiently small, the refinement process
may be considered complete if the ratios (calculated shiftslstandard
deviations) are sufficiently small (e.g. less than 0.2-0.3).
Correlation coefficients between two parameters are given by
pI'. = b../(b..b..)112.
I' 11 I1
pi, can range from 0 to f1: as a rule, pi, s 0.2-0.3 are frequent, pi, = f1
refers to two completely dependent parameters, one of which has to be
eliminated.
As already stressed, owing to the presence of systematic errors very often
the working matrix Nf experimentally available is referred to Mf by an
unpredictable relation much more complex than a scaling factor. In these
cases the technique of multiplying the working variance-covariance matrix
B;' by $I($) in order to obtain M, may be highly questionable. A report[511
of the International Union of Crystallography Subcommittee suggests that
besides indices R or R, as given in eqns (5.3) and (5.85) the goodness of fit
ratio $I($) (see also eqn (5.86)) should be also reported in publications as
a global measure of fit.
Practical considerations on crystallographic least squares

Crystallographic least-squares refinement may not be a trivial task. Suppose
that, for a structure of modest complexity (40 atoms in the asymmetric
unit), n = 4000 reflections have been measured. Generally speaking, an
overall scale factor, 120 positional, and 240 anisotropic thermal parameters
should be refined, for a total of 361 parameters. Then a square matrix B of
order 361, with m = 361 x 36012 = 64 980 distinct elements will arise, each
one constituted by a summation on thousants of terms. The computing time
(Cpu) and the storage (St) needed for a 'structure factor calculation-least-
squares refinement' cycle will comply with the following table:
Step Cpu St
Calculate structure factors nrn n

Calculate derivatives nrn rn
Calculate normal matrix nrn2 rn(m + 1112
Invert matrix m3
Calculate shifts AX rn
In accordance with the above table, computing time and storage rapidly
increase with the complexity of the structure, so that the task soon becomes
prohibitive even for large and fast computers when large-scale problems
(thousands of parameters) are treated. A useful suggestion arises by
observing that the elements on the principal diagonal of B are sums of
squares, so that they are always positive and rather large. On the contrary,
the off-diagonal elements are sums of products which may be positive or
negative; therefore they are generally expected to be smaller than diagonal
elements. Accordingly computer storage and computing time may be
reduced by setting all off-diagonal elements to zero (diagonal-least-squares
approximation). That is equivalent to assuming a complete statistical
independence among the parameters, but this is often unrealistic. For
example:
(1) errors in the thermal parameters generate a systematic error (larger to
high sin 8 / A ) on IF[,,which on its turn produces a bias into the estimate
of the scale factor K;
(2) in oblique coordinate systems, non-negligible correlations between
some coordinates of the same atom may be found. Indeed an error on a
coordinate is compensated by errors on some other coordinates (see
Fig. 2.10);
Fig. 2.10. P(xo, yo) is the true atomic position. If
(3) a high correlation will be found among the site occupancy and an error A, is introduced, the 'best' value for x i s
temperature factors if they are contemporaneously refined. obtained by minimizing, along the line yo +
A y = const, the distance of the atom from the
true position P. That produces A, = -A, cos y.
An alternative to the diagonal approximation is the block-diagonal
approximation; a first block may involve the correlation between the scale
and the overall thermal parameter; the other blocks, one for each atom, are
9 x 9 matrices comprising positional and anisotropic temperature factors (a
4 x 4 matrix for an isotropic atom). The matrix B will then appear as in Fig.
2.11. Larger blocks are sometimes used; e.g. all the atoms in the same
molecule could belong to the same block; or, also, a block for all positional
parameters and a block for all vibrational parameters with the overall scale
factor; or, . . . .
The storage requirement for the block diagonal approximation is certainly
smaller than for full matrix methods. Each refinement cycle is faster, but
convergence is slower: so the complete refinement process requires more
cycles and almost the same computing time. Thus the major advantage of
the method is the lessened storage requirements for problems too large to
be treated by a full matrix. Fig. 2.11. A scheme for block-diagonal
A very large saving of computer time has been recently achieved by approximation.
100 I Carmelo Giacovazzo
application of the fast Fourier transform (FIT) algorithm. FFT based

programs allow full-matrix refinement of protein structures at a reasonable
cost (see Appendix 2.1 for some details).
A critical point in least squares is the occurrence of large correlation
coefficients among parameters: this can give rise to ill-conditioned (deter-
minant close to zero) B matrices and therefore to serious problems for the
calculation of B-' and for the accuracy of AX. That occurs when the
following conditions hold.
1. Two atomic sites nearly overlap because of some structure disorder (or
some error).
2. The symmetry of the structural model is higher than the space-group
symmetry in which refinement is made. As an example, let us suppose that
refinement of two symmetry independent molecules has been unsatisfac-
torily carried out in P2/m. The question arises as to whether or not the
space group is really centrosymmetric. Refinement in P2 is then carried out
on four symmetry independent molecules, the second two obtained from the
first two by application of the inversion centre. Then the new B matrix will
be singular: indeed the imaginary parts of the structure factors are
vanishing, the derivatives of IFI, with respect to pairs of positional
parameters referred by the inversion centre have equal moduli and opposite
signs, the derivatives with respect to corresponding pairs of thermal
parameters have equal moduli and signs. Since the elements of B are sums
of products of derivatives, pairs of lines will be identical or opposite, so that
B is singular. To overcome this difficulty small random shifts are applied to
parameters when refinement moves from the first to the second space
group. However, the matrix still remains ill-conditioned and further
unpredictable shifts can be generated. A more efficient procedure is that
which modifies the normal matrix to limit the parameter shifts (not
equivalent to dumping shifts after the matrix solution).
As a further example, a singularity may occur when two atoms with the
same x and y coordinates but differing by a 50 along the x axis are refined
without using reflections with h odd.
3. Symmetry constraints are not introduced. For example if an atom lies
on 4[rn11,the following constraints will hold: Pll = fizz, PZ3= P13= PI, = 0. If
pll and pzzare erroneously refined as if they were independent, the matrix
B will have two rows and two columns equal and will be singular.
4. In some non-centrosymmetric space groups the origin may float in
some directions. The normal equations matrix will become singular if the
atomic coordinates of one atom along the free directions are not fixed (e.g.
z in Pca2,, or x and z in Pm).
Such a practice was frequently used in the primitive least squares
programs but provides incorrect estimated standard deviations unless the
full variance-covariance matrix is used. The problem may be solved by
constraining (or restraining, see p. 107) the sum of the free coordinates to a
constant value.
An efficient least-squares program should:
1. Provide easy ways for applying symmetry constraints to parameters of
atoms in special positions. These are handled by not refining some
parameters or by equivalencing atomic parameter shifts. For example, in a

hexagonal space group for an atom in special position at (x, h,z) the only
independent positional variables are x and z, so that
Analogous problems occur for thermal parameters: symmetry constraints

on them are described in Appendix 3.B.
2. Allow for a variety of weighting schemes (see p. 369). The weights
play a critical role in any refinement process. In the initial stages of
refinement artificial weights may be used to accelerate convergence (e.g.
w, = 1 for all reflections is commonly used when a partial structure is under
refinement). In the final stages the weights should reflect the accuracy of the
individual observations (w, = 1/02(IFIO)should be deduced from an analysis
of the experiment). Then observations which are believed to be unreliable
get very small weights.
Unfortunately it is not trivial to assess the accuracy of observations and
the correctness of the assumed model. A long list of errors[521can affect the
values of the deviates (IFHlo- IFHlc)and therefore the results of a crystal
structure analysis: experimental accidents (fluctuating power in the incident
beam, crystals moving on their mounts, etc.); statistical fluctuations in
quantum counts; crystal damage by the incident beam; errors in corrections
for absorption, background radiation, thermal diffuse scattering, extinction,
anomalous dispersion, anisotropic and/or anharmonic vibration, disorder,
multiple reflections; deviations from the isolated atom approximation;
incorrect symmetry assignment; correlation among observations.
Different types of errors can have similar effects and may be compensated
by others. For example, absorption and extinction corrections tend to
reduce intensities at low sin 6/Avalues more than at high values: thus errors
in correcting these effects produce, as a partial compensation, errors in the
thermal tensors.
In most cases the effects of each type of error cannot be estimated: then
the standard deviation of individual intensity measurements is assumed as
the standard deviation calculated from counting values plus an additional
term which allows for errors of an undefined nature. It should be stressed
that estimated variances of observations, and therefore weights, should not
be based on counting statistics alone. Indeed variances estimated in this way
are proportional to the intensities themselves: the consequence is that low
intensities will systematically have lower variances than high intensities.
That results in a bias in the estimate variances, and consequently (the
Gauss-Markov theorem states that only if errors are random and uncorre-
lated it may be expected that the least-squares methods give the best
unbiased estimate of the parameters) in a bias in the parameter estimates
which may be of some relevance in high-precision structure determinations.
In these cases it is highly recommended that all symmetry equivalent
reflections are measured, that the averaged intensity is used in least squares,
and that the variance based on counting statistics is combined with that
arising from differences among symmetry-equivalent data.[511
The possible presence of errors may be checked by observing that S is
expected to be distributed like X2 with n - m degrees of freedom. For large
n - m values x2/(n - m) is expected to be close to unity: so, if w,, = l/&
the expectation (see Appendix 2.H) for 3 is ( 3 ) = n - m. Accordingly the

set of observations may be divided in large subsets (any significant criterion
may be used for obtaining subsets: intensity ranges, or intervals of Bragg
angles, or parity of Miller indices, etc.): then the weighted sum of
discrepancies for each subset may be analysed. Each jth sum should be
nearly (n - m)(p,/n), where p j is the number of terms in the jth subset. If
some unexpected behaviour appears then one may guess that some source
of error has not been taken into account, and weight should be suitably
changed.
A practical weighting criterion which takes into account all random
experimental errors for which precise estimates are not available from
experiment may be so described: the lFlorange is subdivided into a number
of intervals each containing approximately the same number of reflections.
For each ith interval the average (IFlo)i and the difference (AF),=
((IFIo- IFJ,))i are calculated. Plotting (AF), versus (IF,[) (see Fig. 2.12)
will show a distribution of points which can be approximated by an
analytical function such as
p(lFl0) = a0 + a1 IF10 + a2 IFIi + a3 IF[:.
. The weighting scheme will be designed in order to make differences

(A&) constant; thus
Fig. 2.12. More efficient empirical weights which make ( A F ) ~constant as a

function of lFlo can be obtained through truncated Chebyshev
polinomials.[531
But the above practices are not able to eliminate systematic errors or
correlation among errors. If a careful check does not suggest any new
strategy the only way for lessening the effects of systematic errors is based
on a weighting scheme based on the deviates (IFo[- I&[). Wang and
~ o b e r t s o n [ modified
~~] least-squares weights by adding to the calculated
variance a term which is a fundion of the disagreement between the
experimental and the expected distribution of the deviates.
Robust-resistant techniques[551also use weights based on the deviates.
Due to some random or systematic errors, or to imperfections in modelling
parameters, or to many other reasons it may occasionally occur that a small
number of data, called outliers, show a very large discrepancy with respect
to the corresponding calculated values. Least squares are very sensitive to
the presence of large residuals which, having a high leverage, may cause
non-negligible distorsions in results. A naive approach, often used in
practice, may be to discard outliers by setting to zero the corresponding
weights. The robust-resistant techniques decrease weights when disagree-
ment increases, so avoiding any discontinuity in the process.
3. Allow for a proper choice of the number of reflections to be included
in the refinement. In order to save computing time, or to avoid many
low-quality data reducing the efficiency of the process, reflections for which
IF1 < qo(lF1) are often excluded from the refinement (q depends also on the
number of available data: q = 2, 3 is a frequent choice. Higher values of q
could bring to a unsatisfactory ratio number of observations/number of
parameters).
However, the omission of observations is equivalent to assigning zero

weights to them. If a((F1) depends on IF1 such a practice preferentially
discards low-intensity reflections so introducing a bias in the process. Often,
that has little effect on the results,[56]but sometimes it discards important
information (very unreliable results can be obtained when a structure with
superstructure effects is refined without using weak reflections).
The problem of identifying meaningful changes in R produced when the
model is altered has been faced by ~ a m i l t o n . [ He ~~~ considered
~~] the
weighted residual (5.88) instead of the conventional residual (5.85), and
proposed the test (known now as Hamilton's test) based on the ratio
T = Rw(1)lRw(2),where Rw(l) and Rw(2) (with Rw(l)> Rw(2)] are obtained
-
from the two different models. In the practice T = R(l)/R(2) is often used
because usually Rw(1)/Rw(2) R(l)/R(2). The test compares T with the
function
If the computed T satisfies T > Rm,,-m,a the hypothesis can be rejected at

the 100a% significance level (a significance level of a = 0.05 is commonly
used). Tables for Frn,,-,,, were computed from Hamilton. For large values
of n - m (very frequent in crystallography). .
where x2 is the well known chi-square function: in these cases
As an example, let us suppose that two least-squares refinements based

on the same number of parameters m and on the same number of
observations n are carried out by two different investigators. The refine-
ments resulted in two different R factors, say R(l) > R(2). The hypothesis
'the m parameters obtained by investigator 1 are consistent with the
experimental data' may be rejected at the cu significance level if T >
Rm,n -m, a.
Sometimes the two models used for refinement involve a different number
of parameters. For example one refinement is carried out where b
parameters are held fixed. Then Hamilton's test may be so modified: the
value of T given by
R (b fixed param., m - b varied param.)
T=
R (all m param. varied)
may be tested against values of
For example, a least-squares refinement was first carried out up to the

R(l) value on assuming that all the atoms have isotropic temperature
factors (ml is the number of parameters). A subsequent refinement was
carried out up to a residual R(2) < R(l) on assuming anisotropic atoms
(m2>m1 is the number of parameters). The hypothesis 'all the atoms
vibrate isotropically' may be rejected at the a significance level if
As a further example, let us suppose that refinement in a centrosymmetric

space group (say P2/m), stops with the residual R(l), while refinement in a
non-centrosymmetric space group (say P2 or Pm) results in the residual
R(2) < R(1). If ml and m2 are the numbers of parameters for the two
refinements, the hypothesis that the correct space group is centrosymmetric
may be rejected at the a significance level if
A last remark on the crystallographic least-squares technique concerns

which function to minimize. In place of the function S defined by eqn (2.67)
the function
is sometime minimized. The normal equations (2.68) still hold if w, is

replaced by wk= wH/(4 IF^^!), 6 IFHlc/6~j (6 IFH1,/G~j)and, in D,
by 2 IFH[,
[IFH10 - lFHlcl [IFHI: - IFHI:].
There is still a strong disagreement[51]over the question of whether S or
S t should be used in refinement. In favour of (FI-refinement it may be
observed[s8] that 6 IFI2/6xi is small if IFI2 is small: thus weak reflections
have too weak a leverage in IFI2-refinement. Arguments in favour of
IF 12-refinement are based on the observation[s93601 that I FI are obtained
from intensity data via a square root. This is a non-linear operation which
may introduce a bias proportional to the variance of IFI2.Furthermore the
formula a(lF()= a ( l ~ 1 ~ ) /IF()
( 2 so widely used in practice is not appropriate
for zero or close to zero IF1 values, for which a(lF1) become infinite or
extremely large (we must, however, mention that methods[613621 for obtain-
ing more reliable IF1 and a((F1) from 1 ~ and 1 ~o(lFI2) values are now
available).
Constraints and restraints in crystallographic least

squares
The accuracy of the derived parameters is generally small (e.g. it would be
found that C-C distances in a phenyl group were in the range 1.2-1.3 A,
which is highly improbable) when at least one of the following conditions
occurs:
1. Low ratio between the number of the observations and the number of
parameters to refine. The ratio is particularly unfavourable for biological
macromolecules. Indeed, owing to their intrinsic flexibility, the ampli-
tudes of the thermal vibrations from the average positions are rather
large (usually between 0.2 and 0.7 A), so limiting the diffraction data to
a resolution of about 2 A. While to a resolution of 0.86 A (typical limit
for small molecules) there are about 30 observations for each coordinate
being refined, at 2.7 A the number of parameters is nearly equal to the
number of the observations.
2. Atoms with very large atomic number and very light atoms coexist in the
unit cell. Then modest errors on the heavy-atom parameters cause strong
errors on the light-atoms parameters.
3. Too high thermal motion, presence of structural disorder, etc. Only poor
and scarce data are then available.
If, however, some prior stereochemical information is available on parts
of the structure its use in the least-squares procedures may increase the
degree of overdetermination of the system and improve the accuracy of the
results. Three methods will be briefly recalled here.
Rigid body refinemenP1

If the structure contains rather rigid molecular fragments of well established
geometry (e.g. benzene rings), the number of parameters to refine may be
considerably reduced. The atoms concerned are initially regularized to some
ideal geometry which is preserved during the refinement.
For each rigid group of n atoms the number of independent positional
parameters reduces from 3n to 6 ((xo, yo, zo) parameters defining the
position of some reference point and w, 3 , q5 for the group orientation).
Thermal parameters are usually considered to be isotropic, and, to keep the
number of parameters small, a single overall thermal parameter may be
assumed for the group.
Least-squares calculations could be performed according to the following
scheme.[641For the jth atom of the group let Xi be the coordinates with
respect to the system A defined by the conventional unit cell, Xi' the
coordinates (in A) with respect to a set of orthogonal internal axes A', Xi"
the coordinates with respect to an orthogonal cell of unit dimensions (say
A ) oriented in some way with respect to A. Assume A" = MA where M is
given by (2.31~)or a similar matrix. A rotation matrix R may be then
defined which aligns A = (ey, e;, e;) on to A' (say A' = RA): according to
the discussion on p. 72 the matrix R may be expressed as the product of
three successive anti-clockwise rotations, say by about e,, by 3 about the
new e2, and by q5 about the new el. Then X," = RX; (see Table 2.E.1) and
The atomic coordinates so obtained may be used in the usual manner for
calculating structure factors. The problem is now to calculate the appropri-
ate contribution for the rigid parameters x,, yo, zo, w, v, q5 to the matrix of
normal equations.
Derivatives with respect to such parameters are calculated from those for
the atomic parameters Xi using the chain rule: i.e.
6 lFHIc -
-- a lFH1c
axo j=1 ax, '
and, for a group temperature factor B,,
The size of the normal matrix is therefore reduced, convergence is

accelerated but there is no remarkable saving in computing time (derivatives
with respect to each atomic parameter have to be calculated for each
structure factor and chain rule implemented).
The success of this method relies on the accuracy of the assumed
geometry: if this is wrong systematic errors on other parameters will be
introduced (e.g. the symmetry of phenyl groups sometime deviates re-
markably from the sixfold symmetry[651).
A technique related to, but not coincident with, rigid body refinement is
that used for putting hydrogen atoms into structures (X-ray data). Some-
times H atoms are geometrically positioned at the end of each round of
LSQ, but more frequently H parameters ride upon those of a reference
atom (riding refinement). In this last case the normal matrix has to be
modified in order to take into account all physical parameters (the common
practice fixes the thermal parameter of H, which is set equal to that of the
reference atom). At the end of calculations H and the reference atom will
have the same positional estimated standard deviations, while the bond
distance between them must have an estimated standard deviation equal to
zero (if the full variance-covariance matrix is used in the calculation).
Constraints via Lagrangian multipliers

In the absence of complete information on a fragment's geometry one may
wish to fix some molecular parameters (e.g. some bond distances and angles
are fixed to given values, a group of atoms is held in coplanar arrangement,
etc.). In such cases Lagrangian multipliers can be used: the function to
minimize is h
where the (generally non-linear equations)
represent the fixed constraints. Usually the available model X0 will not
exactly satisfy (2.69). The problem will be linearized by expanding both the
IFHIcsand the Gis in Taylor series:
where G: = G,(XO) and 6 ~ ~ is 1the6 derivative

~ ~ of G calculated in XO.
The normal equations can be derived by settling to zero the derivatives
(with respect to Ax and A) of S:
The augmented normal equations, expressed in terms of partitioned

matrices, are
where mjq = i(6G:/6xj).

Even if a large variety of stereochemical constraints can be taken into
account by the method, the approach is not very attractive for large-size
problems: indeed the order of the normal matrix is increased.
Use of restraints
Soft, flexible constraints (say restraints) may be imposed to some functions
of the parameters in order to permit only realistic deviations of their values
from fixed standard ones. These functions are used as supplementary
observations, so that the order of the normal equation matrix is neither
increased nor reduced. In this situation the function to m i n i m i ~ e [ ~ ~
is, ~ ' ]
where gq is the function describing the gth restraint, g,, is its standard (or
optimal) value, w, is the weight to associate to the qth restraint. The normal
equations are obtained by expanding S in Taylor series and equalling to zero
its derivatives (with respect to AX):
The effect of restraints is therefore to add contributions to the elements of

the normal equation matrix. Thus parameters highly correlated can be
found, but this is expected since parameters are deliberately correlated by
restraints.
For example, a restraint may be to set an interatomic distance d (bonded
or non-bonded) to a given value taken from the literature. If we recall,
according to p. 62, that
where
and so on.
Another restraint may be: the sum of all the atomic coordinates along a
polar direction can be fixed to its current value in order to keep fixed the
centre of gravity of the molecule and so determine the origin. Several other
types of restraints can be imposed (see Chapter 8, p. 564) and they concern
van der Waals distances, planarity of groups, chirality (a restraint on the
chiral volume about an asymmetric carbon atom may maintain the
conformation in the correct hand), bond and torsion angles, thermal
motion, positional and thermal parameters related by non-crystallographic

symmetry, potential energy functions, etc. Also, the sum of site occupancies
of two or more atoms can be restrained to some fixed value when disorder
problems occur. When the number of restraints is sufficiently large the
equations are overdetermined and refinement will also converge with
low-resolution data.
The approximate nature of the restraints well agrees with our approxi-
mate knowledge of the chemistry involved. A basic difficulty of their use is
to assess an optimal weighting scheme: indeed the first component of S is in
electrons squared while the second may be considered an energy quantity,
where w, plays the role of an effective force constant. A convenient
mechanism is to indicate the relative precisions (i.e. the expected standard
deviations) of the restraints and scale them with respect to the observations.
As an example, let us suppose that the bond distance dj is to be restrained
to a set value. If the weights of X-ray observations are (as usually)
sufficiently accurate but on an arbitrary scale, the normalization factor
&, = CHw ~ ( A F ~ )-~m/ )( ~may be obtained. Then the restraint equation
involving di with estimated standard deviation o(dj) is given the weight
Very small standard deviations make restraints equivalent to constraints.

If refinement provides a bond distance differing from dj by several times
a(di) it should be concluded that experimental data are in conflict with the
fixed restraint.
Applications of constraints and/or restraints in least-squares processes
usually lead to larger residuals than by unconstrained refinement. That does
not necessarily mean that the constrained solution fits less well to the data.
A sensitive criterion may be to compare estimated standard deviations
obtained by constrained and unconstrained refinements. When constraints
are applied both S and n - m increase but oii(see p. 98) may be larger or
smaller according to circumstances. An alternative criterion may be the use
of the Hamilton's test (see p. 103).
Alternatives to the method of least squares

In the above sections some deficiencies of least squares have emerged,
deficiencies which invite the application of alternative methods. In statistical
sciences the main alternative to least squares are maximum-likelihood
methods. Let us suppose that the errors e defined by (2.47) are distributed
according to some probability density function P(e). The likelihood function
L is then given by
Since P is everywhere greater than or equal to zero, its logarithm is a

monotomically increasing function of the argument. Thus the maximum of
L is attained when In L is a maximum. If the error function is Gaussian then
P(ei) = ( 2 ~ o ? ) - exp
~ ' ~[-e?/(2o?)]
where a is the variance of the ith observation. Then
Since the second and third terms are constant, In L has its maximum
when
where wi = l/a:. We have so shown that, if the distribution of the errors is

Gaussian, and observations are weighted by the reciprocal of the variance,
the least-squares and the maximum-likelihood methods provide equivalent
results. Conversely, if errors are not distributed according to a Gaussian
model, the method of least squares does not lead to a maximum-likelihood
estimate.
The above results also explain why least squares are practiclly the only
ones used in crystallographic refinement: they only depend on the variance
of the errors while maximum-likelihood methods require the knowledge of
the distribution of the errors. When this is available maximum-likelihood
methods provide better estimates of the parameters, since such methods can
exploit more information. However, such a situation is very infrequent in
crystallographic refinement, where only rather imperfect estimates of the
variances are available.
Parameters may also be estimated by maximum entropy methods, which,
based on Jaynes' extension of Shannon's theory, require the maximization
of the entropy. But such methods, as well as maximum-likelihood methods,
are far from being competitive with least squares in most of crystallographic
applications.
Rietveld refinement
The basis of the technique
Powder diffraction patterns (see p. 293) may be collected in a step scan
mode: intensity is measured for a given interval of time, and the theta and
two-theta axes are then stepped to the next position. The pattern is then
indexed: i.e. appropriate Miller indices are associated with observed
reflections and simultaneously accurate unit-cell dimensions are calculated.
Because of unavailable experimental errors in the estimates of the
diffraction angles and because of the frequent overlapping of peak inten-
sities, indexing is a rather difficult task for relatively large cell volumes
and/or low-symmetry crystals. Several approaches are today available for
this aim: implemented in computer programs, they often provide more than
one solution, so that proper figures of can be used to distinguish
between bad and good solutions. The reader will find general remarks on
the various approaches, and tests on their efficiency, in a paper by ~hirley[~']
,and in some more recent paper^.[^"^^] This stage of analysis is followed by
the examination of possible systematic absences to suggest a space group.
If a (even imperfect) structural model is available then the intensity yi,
observed at the ith step may be compared with the corresponding intensity
yic calculated via the model. According to ~ i e t v e l d , the

~~~ model
] may be
refined by minimizing by a least-squares process the residual
S = C Wi I~ i-
o yic12
where wi, given by
(wi)-' = 0; = B:~ + o;~,
is a suitable weight. oi, is the standard deviation associated with the peak
(usually based on counting statistics) and a,, is that associated with the
background intensity yib.
yic is the sum of the contributions from neighbouring Bragg reflections
and from the background:
where s is a scale factor, Lk is the Lorentz-polarization factor for the

reflection k, Fk is the structure factor, mk is the multiplicity factor,
Aerk= 28, - 20k where 28, is the calculated positon of the Bragg peak
corrected for the zero-point shift of the detector, and G(Ae,k) is the
reflection profile function.
The parameters to adjust by refinement include unit cell, atomic
positional and thermal parameters, and parameters defining the functions G
and yzb*
The determination of an accurate model for the profile function G(Aock)
is one of the fundamental problems in both single-peak and in Rietveld
analysis. This is particularly true today when high-resolution neutron and
X-ray spectra are available: deficiencies in the profile model are no longer
lost in relatively large instrumental diffraction profiles. The shape of a
diffraction peak depends on several parameters: the radiation source, the
wavelength distribution in the primary beam (e.g. possibly selected by a
monochromator crystal with its specific mosaic spread), the beam charac-
teristics (as influenced by slits and collimator arrangements between the
primary radiation source and monochromator, between monochromator
and sample, and between sample and detector), the detector system.
Accordingly, there are many choices of analytical peak-shape functions. We
quote:
(Gaussian) ;
cli2
2(1 + CIX;k)-l (Lorentzian);
nHk
-(1 + c2xfk)-2
'L 2
(modif. 1 Lorentzian);
~tHk
"'(1 + c3xfk)-1'5 (modif. 2 Lorentzian);

2Hk
with 0 s q G 1 (pseudo-Voigt);
r(B)
r ( p - 0.5)
52
7tk
(1 + 4c4x:,)-p (Pearson VII);
where Co = 4 In 2, C1 = 4, C, = 4 ( - I),
~ C3 = 4 (2213- I), C4= 21f8- 1,
Xik = AOik/Hk. Hk is the full-width at half-maximum (FWHM) of the kth
Bragg reflection, and r is the gamma function.
It is easily seen that the pseudo-Voigt function presents the mixing
parameter rj which gives the per cent Lorentzian character of the profile.
When P = 1, 2, m, Person VII becomes a Lorentzian, modified Lorentzian,
and Gaussian function respectively. Of some use also is the pure Voigt
function which is the convolution of Gaussian and Lorentzian forms.
The FWHM is usually considered to vary with scattering angle according
to
+
(FWHM), = ( U tan2 8 V tan 8 - w)'" (2.71)
for the Gaussian component,[781according to
(FWHM), = X tan 8 + Y/cos 8 (2.72)
for the Lorentzian component.[791 U, V, W, and/or X, Y are variable
parameters in the profile refinement.
Besides analytical functions non-analytical functions arising from an
analysis of resolved peaks may also be usedL8'] to describe peak shape (in
the Rietveld method the peak shape is not the end but a tool of the
method).
There is no well established approach to the background. It is mainly due
to insufficient shielding, to diffuse scattering, to incoherent scattering
(rather high for neutrons), to electronic noise of the detector system. The
background and its variation with angle is usually defined by refinement of
the coefficients of a power series in 28:
where the b, terms are refinable parameters.

All the above described parameters are introduced in the Rietveld
refinement process according to (2.70). The agreement between the
observations and the model is estimated by the following indicators:[s11
(2) The weighted profile R,, = [C wi(yi0- ~ i c ) ~ ~iy?o]l'~*

/C
(3) The Bragg RB= C JZko - Ikcll(CZko). The values I,, are obtained by
partitioning the row data in accordance with the Ik, values of the
component peaks.
(4) The expected R E = [ ( N - P)/(X where N and P are the
number of profile points and refined parameters respectively.
(5) The goodness of fit GofF = C wi(yio- Y , ~ ) ~ /-( N
P ) = (R,,/R,)~, which
should approach the ideal value of unity.
The most meaningful indices for the progress of refinement are R,, and
GofF since they show in the numerator the quantity being minimized . Also
RB is of considerable use since it depends on the fit of structural parameters
more than on the profile parameters.
The use of restraints in the Rietveld method has proved to be of

particular usefulness for complex structures.[80,82,831
Some practical aspects of Rietveld refinement

The generation of instrument profile calibration curves usually precedes
Rietveld refinement. Such an analysis allows one to set up the angular
dependence of the peak shape. A peak profile may be expressed as
where b(28) is the background function, f(28) is a specimen related

function, and w(28) *g(28) is the so-called instrument function. This may
be considered as the convolution of the function g(28) representing the
diffractometer's optics and a second function w(8) representing the wave-
length distribution of the incident radiation.
As a first step, diffraction profiles from a standard specimen (e.g. silicon,
corundum, or quartz standard materials) devoid of crystallite size or strain
broadening effects are collected: in this case f (28) may be represented by a
delta function and (2.73) becomes
If a single wavelength is used (i.e. in synchrotron or neutron experiments) a

single peak may be represented by one or by a combination of the functions
described in the previous subsection. If a Cu X-ray source is used and a2
and a, are not eliminated by some experimental device (i.e. a germanium
monochromator) a compound profile consisting of three split functions has
to be chosen, one for each spectral component. The number of variable
parameters in the profile function may be reduced by fixing a, and a3
intensities relative to a, intensity via a previous profile-fitting experiment.
Analogously, the angles for a2and a3lines may be fixed according to their
wavelengths, and their FWHM can be set equal to that of al.
The correct profile function is in general obtained by trial-and-error
methods from single well behaved profiles of the standard material. The
goodness of the profile fit is measured by
where y,, and y,, are the observed and calculated values at the position 28,.
For good experimental data the R value is expected to be around one per
cent.
The values of the profile parameters may be refined for each single peak
collected from the standard: each peak is separated and analysed in a region
with enough points to allow a good sampling of background on each side of
the peak (a parameter for background may also be refined). In particular,
the values of FWHM obtained from refinement are used for a first estimate
of the U , V, W values in eqn (2.71) and/or of X , Y values in eqn (2.72). It
is worthwhile mentioning that standard specimen data are also used for the
determination of wavelength A (if necessary) and for the zero-point
calibration 28, of the detector scale.
Analysis of the specimen profile f usually follows instrument profile
calibration. Specimen broadening of the profiles is often present: it is

traditionally assumed to be Lorentzian if profile broadening is present.
The profile analysis of the full diffraction pattern is often used to
peak positions and peak heights for each reflection: from them
lattice parameters are calculated and refined in an iterative way. Peaks
generated by overlapping of more reflections are deconvoluted and integr-
ated intensities of individual hkl are calculated, together with their standard
deviations and their correlations (see Appendix 3.A, p. 184).
A specific feature of Rietveld refinement is that the number ot observa-
tions N can be chosen arbitrarily large simply by decreasing the step interval
in 28. The increase of N does not remarkably affect the conventional
Rietveld agreement but decreases the parameter estimated standard devia-
tions approximately in proportion to N'". Such an improvement is often
fictitious. For example, a single Bragg reflection may be measured using 10,
100, or 1000 steps. The observed intensities are not statistically
independentFa7]and, for single peaks, yield structural information nearly
equivalent to a single integrated intensity count (even if the precision
obtained by 1000 steps count may be higher than by 10 or 100 counts). As a
consequence the correlation among observations becomes higher with
N = 10, 100, 1000 and least-squares residuals A, = y,, - y , , are correlated
themselves, i.e. they tend to assume the same sign. In such a situation
Durbin-Watson d-statisti~s['~I may be used to assess the nature of correla-
tion. It is based on the parameter
For no serial correlation a d value close to 2 is expected. For positive

serial correlations adjacent A tend to have the same sign and d will be
smaller than 2, whereas for negative serial correlation (alternating signs of
adjacent A) d is expected to be larger than 2 (and smaller than 4).
The parameter d may be tested against the 0.1 per cent significance point
via the formula[891
where P is the number of least-squares parameters involved in the

calculations. Q < d < 4 - Q if consecutive terms are uncorrelated, if d < Q
or d > 4 - Q consecutive terms tend to have positive or negative serial
correlation. d-statistics suggest that an optimum value of the step width is[901
between one-fifth and one-half of the minimum FWHM of well resolved
peaks.
The success of the Rietveld method is based on some fundamental
requirements:
1. A starting model for atomic positional and thermal parameters should
be available in advance. That can be provided by the following
procedure:[911(a) deconvolution techniques and profile-fitting methods are
used to derive intensities Zk of individual reflections and to resolve
overlapping peaks; (b) integrated intensities are calculated and structure
solved by direct or Patterson methods. The same intensities are then used in
a full-matrix least-squares refinement as for single-crystal structure deter-
mination. Because of peak overlapping, this time error margins for each
observation and correlation among reflections intensities could be processed

to calculate a non-diagonal matrix Mf (see pp. 91-3).
2. The crystallites should be randomly distributed. That is not easy to
achieve in practice.[92,931
Traditionally preferred orientations are minimized
by loose packing of the powder: however, sufficient compression improves
homogeneity and reduces surface roughness effects. If a specimen spinner is
used in the experimental set-up the preferred orientation can be charac-
terized by a few parameters only, thus it may be introduced in a rather
simple way in the fitting schemes and in Rietveld refinement. If the
preferred orientation is not taken into account, errors in the relative
intensities may occur which limit the accuracy of the refinement process.
Some geometries are less sensitive to this problem (i.e. Guinier cameras).
The problem is also minimized in neutron diffraction because of the large
transmission specimen, but it is severe when synchrotron radiation is used
because the number of crystallites correctly oriented to reflect the parallel
beam is much smaller.[941
The most simple way to correct the intensities calculated from the model
structure is to use an empirical correction function depending on @, the
acute angle between the preferred orientation plane and the diffracting
plane k:,
Ik(corr) = IkPk(@).
Pkmay be ~ a u s s i a n ,951
[~~~
or of trigonometric type1961
p is a parameter to refine in the refinement process.

3. The crystallites should be of equal size. Particles much larger than
10 pm should be removed from the sample, as well as very small particles
(<1pm), which are responsible for profile broadening. Diffraction patterns
of materials with single composition but two different distinct distributions
of crystallite size (or of microstrain) could be characterized by profile-
broadening effects with bimodal distribution.[971 In this case unimodal
distributions are inadequate to describe profile functions, and unavoidable
profile misfit errors will still appear at the end of refinement when observed
and calculated patterns are compared.
4. Correction for instrumental aberrations[9s1001should be made.
The instrument function g in eqn (2.73) may be considered as the result of
six specific functions:
The function gl depends on the projected focal spot profile, g2 is due to the
varying displacements of the various parts of the flat specimen surface from
the focusing circle, g3 is due to the axial divergence (as regulated by Soller
slit collimators), g4 arises from specimen transparency (i.e. from the
penetration by the beam of a sample with finite absorption), g5 is defined by

the receiving slit, g6 is due to the possible misalignment of the experimental
set-up. As may be seen from Fig. 2.13, only gl, g5, and g6 broaden the
profile symmetrically, while gZ, g3, g4 broaden it asymmetrically and shift
the peak from the theoretical position. The deleterious effects of the
aberrations can be minimized by the use of appropriate analytical functions
in Rietveld refinement or by a careful design and/or alignment of the f~nalprof~le
N ~ A
instrument.
5. Resolution of the pattern may be improved by mathematical
technique^:[^^^-^^^] i.e. by deconvolution of the pure from the observed
profile, this last containing the effect of instrumental broadening (see Fig. 2.13. Functions defining the instrument
Appendix 3.A, p. 185). function g.
The resolution of the pattern as well as the signal-to-noise ratio can be
appreciably improved with carefully designed diffraction experiments. When
a non-conventional diffraction apparatus is used one of the first choices to
be made is fixing the wavelength. Complex structures will produce peaks
closely spaced and frequently overlapping: a long wavelength improves
separation between peaks but reduces the number of accessible reflections
which cannot then offset the large number of structural parameters to
refine. A short wavelength increases the number of accessible peaks but
produces severe overlap of them. Thus a useful compromise has to be
chosen between the two requirements.[lo4]When higher resolution is needed
for conventional X-ray sources it is advisable to use the Kg doublet (but the
total experiment time will increase).
In order to compare the performances of different experimental arrange-
ments, in Fig. 2.14 typical FWHM (instrument-only contribution) are
plotted against the diffraction angle for modern neutron diffractometer (N),
for a conventional Bragg-Brentano with (XCS) and without (XCW)
diffracted beam Soller slit (Cu K, radiation and a diffracted-beam curved
graphite monochromator) limiting vertical beam divergence to less than the
standard value of 5", and for a synchrotron powder diffractometer (S). The
use of incident beam monochromators can further improve performances of
conventional divergent beam diffractometry.[lo5]Indeed, the K,, component
may be removed, so halving the number of lines in the pattern. As a
consequence, the resolution of the remaining lines is improved, the
-e
(I)
D
N
......
/.
/
H
/'
'...., ........
........ . .
XCW .... .. ,
-.-. -. -.-. .....
0.2 - -.-.-.-.-.-.-.-a
- .-:-.-./
XCS .-.-.-.
_._,-._._._._ .~-~-._.~._._._._._._._._._._._._)_._-._.c._._._._._._._._._._._._)_._-._.c._._._._._._._._._._._._)_._-
.o -S I
Fig. 2.14. FWHM for different experimental

arrangements.
background is kept essentially constant (indeed KO lines are removed at

source, and also structures arising from the filter absorption edge are
eliminated).
The best resolution is attained by synchrotron radiation: the probability
of measuring single peaks is larger than for any other radiation source. It is
not infrequent[lo6]that synchrotron data present so large a percentage of
unique I,, values that structure determination may be carried on by
techniques identical to those used for single-crystal data (for rather large
structures specialized solution techniques have still to be applied[lo7]).
Neutrons have smaller resolution, but a better signal-to-noise ratio at high
sin 8/A values. Because of the narrow range of neutron scattering lengths,
refinement is facilitated when some heavy atoms are present; however, the
structure solution may be difficult. This situation is just complementary to
that occurring for X-ray data, where location of some dominant atoms may
be easy but subsequent accurate location of very light atoms may be
difficult. It is therefore not infrequent that the Rietveld method benefits by
a combined use of neutron and X-ray data. When some heavy atoms are
present in the structure or high-angle X-ray data are not attainable,
refinement with neutron data may be used to confirm results obtained with
X-ray data[79,'08,lo9](however, improvement in precision may be obtained if
refinement is carried out on the combined X-ray and neutron data sets[llol).
As an example in Fig. 2.15(a,b) observed (points), calculated (bold line),
and difference profiles generated by powder samples of a-CrP04 are
shown.[79] Synchrotron intensity data show a very good instrumental
resolution (FWHM = 0.07" at 28 = 50") but a remarkable intensity decay
with sin BIA: the attained maximum sin BIA value is 0.35 A-', correspond-
ing to 97 independent reflections. Neutron data have smaller resolution
(FWHM = 0.6" at 28 = 50") but no serious intensity decay with sin 8/A: the
maximum sin 8/A value is 0.40 A-l, corresponding to 127 independent
reflections. Since the neutron scattering amplitudes for Cr, P, and 0 are
0.352, 0.53, and 0.577 respectively (see Appendix 3B, p. 198) neutron
refinement is expected to be more sensitive than X-ray to oxygen positions.
The Rietveld procedure has been applied to a very wide range of

inorganic and organic materials[llll and scientific interest in this area is
increasing (the present annual publication rate is approximately 100). The
limits of the method depend on progress in experimental aspects (sample
preparation, higher resolution, better treatment of profiles, etc.) and on the
available prior information which may be introduced into refinement. Such
information is often generated by complementary experimental techniques.
For example, intensities of overlapping reflections of LaMo50, were broken
up by considering electron diffraction patterns:[1091 14 symmetry-
independent atoms were then positioned and their isotropic thermal motion
determined.
As a further example we recall the application of the Rietveld method to
polymers.[82~112~1'31Electron diffraction is a generally applied technique
when chain-folded crystals are of minimal thickness and crystal deformation
negligible. If such conditions are not satisfied quantitative kinematic
treatment leads to models of low accuracy and the Rietveld method may be
applied in order to discriminate unsatisfactory structural models. Owing to
the smallness of crystallites and to the greater disorder present in the
(a) a-CrP04 NSLS Cap I 1 l a r y Oat a 1. 323 A

1 1000
10 20 30 40 50 60
20 (deg.)
-+- Fig. 2.15. Powder diffraction profiles for a-CrPO,

25 50 75 100 125 from: (a) the synchrotron X-ray experiment; (b)
neutron experiment. Reflection positions are
28 (deg.) marked.
polymers the width of peaks is rather large for polymers so that overlapping
events in their pattern are more frequent. In spite of that, discrimination
among various structural models may often be accomplished provided prior
information (size and shape of rigid molecular fragments) is adequately
introduced into refinement.
Analysis of thermal motion

In biological macromolecules the root-mean-square displacements of atoms
from their mean positions are of the order of tenths of an angstrom. On the
other hand the allowed variations for covalently bonded distances are no
greater than some hundredths of an angstrom. The apparent contradiction
may be overcome by assuming a high correlation among the thermal
motions of bonded atoms. In other words, if an atom at a certain instant is
out from its equilibrium position, some other atoms coherently move in
such a way that bond distances are chemically meaningful.
The above conclusion is valid also for small molecules. Indeed more or
less rigid groups of atoms can be found (e.g. rings, condensed rings, etc.)
for which the internal molecular motion may be negligible and the thermal
motion may be described in terms of a rigid body model.[1141161
In the crystallographic least-squares procedures (see p. 94) any correla-
tion among fi tensors of different atoms is usually ignored. However, if
thermal ellipsoids correctly represent the thermal motion some a posteriori
correlations among them can be found. For example, since bond stretching
vibrations have a much smaller amplitude than other sources (i.e. bond
bending or torsional vibrations) the mean square displacement of pairs of
bonded atoms should be approximately equal in the bond direction. Or
also, in long-chain molecules the thermal motion normal to the chain should
be greater than at right angles to it; or, terminal atoms such as the 0 atoms
in carbonyl groups or H atoms in methyl groups have generally greater
thermal motion than the atoms to which they are bonded. Thus if the crystal
contains more-or-less rigid groups of atoms, it makes sense to analyse
thermal motion in terms of translational and librational oscillatipns of these
groups. In the following, such an analysis will be described on assuming a
Cartesian coordinate system.
In accordance with Appendix l . A the most general motion of a rigid
body is a screw rotation. If the axis of rotation is correctly oriented but
incorrectly positioned, the rotation and the translation component parallel
to the axis do not vary, but additional translation components perpendicular
to the rotation axis are introduced. Thus the most general motion of a rigid
body may be considered as the combination of a rotation with a suitable
translation. These operations do not commute in general; luckily an
adequate treatment of anisotropic thermal motion, accurate to the level of
quadratic approximation, can be used upon infinitesimal rotations which do
commute.
To illustrate this representation of rotation, consider an atom at X =
(xl, x2, x3) in a Cartesian coordinate system E. According to p. 71, in such
a system, a rotation through x about the unitary vector 1 (I1, I,, 1, will be its
x
direction cosines), is represented by relation (2.32b). If is sufficiently small
(2.32b) may be written as
X' = RX, where R= I + XP and P = ( 2

\-12 11
!;A./
0
(2.75)
The change of variable q = ~1 brings to

X'=(I+K)X=X+KX=X+AQ
where
It may be concluded that the most general infinitesimal displacement of

an atom in a molecular crystal from its equilibrium position is
u=t+q/\r=t+(rl-r)=t+Aq. (2.76)
The variance-covariance matrix for the displacement u is
u = ( U C )= ( t i ) + ( ~ ( 9 9 ) +~ () ~ ( q q +) ((t9)A)
=T+A~+AS+SA (2.77)
S + D provided
AD+DA=O
is satisfied. Indeed, if S in (2.77) is replaced by S +D
is obtained from which (2.79) arises. Since A = -A we will have D = kl

where k is an arbitrary constant. It may be concluded that the values of Uij
calculated from T, L, S do not vary if D is.added to S. Vice versa, only the
differences SI1- SZ2,SZ2- S33, S33- Sll may be obtained from the observed
5 .Since their sum is zero the number of determinable components of S is
eight: usually it is preferred to have the trace of S equal to zero.
Often thermal motion analysis produces an excellent agreement between
observed and calculated U,? (e.g. ((Au~,)') lJ2= (1-4) X A'): sometimes
worse agreement occurs (e.g. of the order of low2A2 or more). Disagree-
ment may be due to internal molecular motion of the atoms as well as to
errors in the observed Uij, or also to the inadequacy of the approximation
(2.75).
The effect of thermal motion on bond lengths and

angles
From a crystal structure analysis atomic coordinates are obtained which
represent the centroid of a distribution of electron density arising from the
combined effect of the atomic structure and of the thermal vibration.
Usually interatomic distances are calculated as separations between a pair of
atomic positions (distances from mean positions), but it has become clear
that, if the thermal motion is not negligible, the most suitable
measure of an interatomic distance is the mean separation (usually longer
than inter-mean distances), the average being calculated by taking into
account the joint distribution of the thermal motion of the two atoms. Such
complex information is seldom available, but simplified models of the
vibrating system do provide useful estimates of the mean separation.
Let dorepresent the separation of the mean positions of two atoms A and
B, and d = do+ s be the instantaneous separation. For convenience consider
a cylindrical coordinate system with axis z in the direction of do. If v and w
are the axial and radial vectorial components of s respectively then
+ + +
d = do+ v w from which d = [(do v)' ~ ~ 1 " ~ .
Taylor-Maclaurin expansion of d(v, w) yields
from which
Express now the value of (w2) in terms of quantities which are

experimentally accessible for three simplified models of thermal motion. It
may be noted that w is the difference between the projected instantaneous
displacements w6 and wA of the two atoms B and A (w = w~ - wA) so that
(w2) = (w;) - 2(wgw*) + (wi).
We will examine three simplified models:

1. The motion of the two atoms is not correlated (as a first approxima-
tion, that occurs for non-bonded atoms). In this case (w2) = ( w i ) + ( w;) ;
therefore
( 4 = do + ( ( ~ +3( & ) ) l ( W . (2.81)
2. The atom B has all of the translational motion of the atom A plus an
additional motion uncorrelated with the instantaneous position of A (riding
motion). Such a model well fits the case in which the riding atom B is
strongly linked only to the atom A and the mass of B is considerably smaller
than the mass of A (eventually summed to neighbouring atoms strongly
linked to A, as occurs for example if A belongs to a rigid group). Then
( 4 =do + ( ( ~ -3( w W ( 2 d o ) . (2.82)
Express now ( w i ) and ( w i ) in terms of the vibrational parameters
usually obtainable from least-squares refinements: the corresponding ex-
pressions could be introduced into (2.81) and (2.82) in order to obtain (d).
If the thermal motion of the atom A is assumed to be isotropic then,
according to (3.19), (wk) = 2 ~ , / 8 n ~ (w:)
, = 2 ~ ~ / 8 where
n ~ , BA and BB
are the thermal factors estimates arising from least-squares calculations for
A and B atoms respectively. Then (2.81) and (2.82) become
and
respectively. If the thermal motion is assumed to be anisotropic, then a

suitable coordinate system in the plane normal to do may be chosen, and
(see eqn (2.38b)) we can write
(w:) = ( ~ 2 +) (w?ly) = (A1A+A2A)l(2n2).
According to (3 .B. 10a)
where A3,1(2n2) is the mean-square displacement of the atom A along do.

On introducing (3.B.10~)and (3.B.9a) we obtain
where Xo represents now the direct lattice components of the vector do.
3. The rigid body model. Let n be the unit vector about which a group of
atoms oscillates with mean square amplitude (w2). A small rotation dw
about n will produce over the interatomic vector d the variation 6d = (n A
d) d o .
Since 6d is perpendicular to d, we can write
-
(w2) = d2(w2)sin2 3 dt(w2) sin2 3
where 3 is the angle between do and n. In accordance with (2.80)
(d) =do + (dow2sin2 3 ) / 2 = do(l + w2sin2 312).
Shortening of interatomic distances will be more clearly deduced from

Fig. 2.16. The libration axis n is normal to the page through 0 , d is the
- -
distance from P to the axis. The radial displacement is approximately given
by d - d cos o d(1- cos o ) d((u2/2).
As an example, a librational mean-square amplitude of 10' may lead to
Fig. 2.16. Shortening of interatomic distances
due to libration motion. shortening of interatomic distances of up to 0.025 A.
Thermal motion is expected to produce distortion also in apparent angles.
To study this type of effect the joint distribution of three correlated thermal
motions has to be taken into account.[1161
About the accuracy o f the calculated parameters

At the end of a structure analysis both positional and vibrational parameters
are available, together with variances of the parameters and covariances
among them. This supplementary information is of primary importance in
assessing the accuracy of the parameters themselves. The variance of the
parameter f i n terms of variances of and covariances among other directly
determined parameters p is
If the pjs are uncorrelated then (2.83) reduces to
As an example, let us calculate the standard deviations of the unit cell

volume V and of the angle a*when the standard deviations of a, b, c, a , (p,
y are known. On applying (2.84) to the expression of V in Table 2.1 we
obtain
where
A, = o(a)/a, B1 = sin a(cos a - cos (p cos y)a(a),
A, = a(b)/b, B2 = sin (p(cos (p - cos a cos y)a((p),
A3 = a(c)/c, B3 = sin y(cos y - cos a cos (p)a(y).
On applying (2.84) to the expression of cos a* in Table 2.1 we obtain
a2((a) = {a2(a)sin2 a + a2((p)(sin(p cos y + cos a*cos (p sin y)'
+
+ 02(y)(cos(p sin y cos a*sin (p cos ~ ) ~ ) l ( sa*
i n sin (p sin Y ) ~ .
Apply now (2.84) to the relation (2.3b) in order to derive a2(d2),where d
is the distance between two uncorrelated atoms positioned in (xl, yl, 2,) and
(x,, y2, z2) respectively. Then
1
a2(d) = 7 {(A1+ A2 cos y + A3 cos ( p ) 2 [ ~ T+~ a2(a2(x1)
: + a2(x2))]
d
+ (A, cos y + A, + A3 cos a ) 2 [ ~ : ~+: b2(a2(yl)+ 02(y2))]
+ (A, cos + A, cos a + A3)2[A3: + c2(c?(z1) + a2(z2))]
+ (AlA2a(y) sin y)' + (AlA3a((p)sin (p)' + (A2A3a(a)sin a)2).
(2.85a)
Crystallographic computing ( 123
(2.85a) is simpler if the errors on the unit cell parameters can be neglected.
-
An additional simplification is obtained if the errors are isotropic (i.e.
a2&, = bZa;, = c20:, = a: and a2a;, = b2at2= c2afZ a:) and the axes are
orthogonal. Then (2.85a) reduces to
For pairs of symmetry-related atoms (2.85a) cannot be applied. Thus, for

a pair of atoms referred by an inversion centre, any error in one position is
completely correlated with the error in the other and
where a is measured in the direction of a line joining the atomic centres. It

may be observed that 20 (standard deviation for the distance of two
completely correlated atoms) is larger than (standard deviation for the
distance of two uncorrelated atoms: see eqn (2.85b) for al = a2).
In general any pair of atoms in a structure are neither completely
independent nor completely correlated, and the general expression (2.83)
has to be invoked. An estimate of the covariance between the positional
parameters pi and pj is usually obtained from the structure least-squares
refinement (as the ijth element of the matrix M, defined on p. 92). The
covariances among the unit cell parameters are usually available from the
least-squares refinement of the unit cell parameters.
By an analogous technique the standard deviation^["^^^^] of bond angles,
torsion angles, and least-squares planes may be obtained. To assist the
reader the simplified expressions of the standard deviations for bond angles
8 and torsion angles o are quoted:
a&BcD) = 2
dABsin2 (ABC)
+ d& sin2(BCD)
- dABcos (ABC)
dABsin (ABC)
dBc - dABcos (ABC)
- 2 cos o cot (BCD)(
,, sin (ABC)
- dcD cos (BCD)
cD sin (BCD)
- 2 cos w cot (ABC)(

~ B C- d~~cos (BCD))]
cD sin (BCD)
If in the first expression dAB= dBCand a common a2is assumed then
a: = 2a2(2 - cos 8)/d2.
- -
If in the second expression dAB dBc do, equal bond angles (ABC =
BCD = cp), and a common a2are assumed then
4u2
a 2 ( o )= [I - cos cp(1- cos cp)(l - cos o)].
d2 sin2 cp
Two simple examples on the use of the variances of the parameters in

crystal structure analysis may be described:
1. It is often required to take a decision whether a bond lengtb or angle
differ significantly from some others. For example, suppose that two
independent C-C bonds dl and d2 are found which differ by 0.02A,
while the isotropic standard deviations for the atomic positions are
a, = a, = a, = a, = 0.003 A. According to (2.85b) a(d) = 0.0042 A while
the standard deviaton associated with 6 = dl - d2 is
It is therefore found that the observed value of 6 is A = 3.39 times larger

than a,. If 6 is assumed to be distributed around zero according to a
normal distribution with standard deviation a,, then a statistical criterion
may be used to decide if d l and d2 are significantly different. The
conclusive inference is that the two bonds differ.
2. We are required to decide if a sample of n observed bond distances di is
drawn from a common population or not. If the di are assumed to be
normally distributed about the mean (d), the quantity
is evaluated and compared with the tables of the xi,, distribution.
Appendices
2.A Some metric relations between direct and
reciprocal lattices
Let us prove that V* = 1/V. By definition
Owing to (2.8)
1 1
V* =-(b A c ) . [(c a A b)a] = -
v3 v '
Derive now the values of a*, p*, y* from direct lattice parameters.
According to the first eqn (2.13)
from which, owing to second and third relations (2.12b),
sin a* =
v
abc sin p sin y
is obtained. Expressions for sin p* and sin y* quoted in Table 2.1 are
obtained by cyclic permutation of the parameters. Derive now the expres-
sion of cos a*.By definition

b*-c*
cos a*=-- -
1 1
b*c* b*c* V2
-.-
(C A a ) . (a A b).
Using (2.7) and the above derived expression of sin a* gives

cos a*= (cos p cos y - cos a)/(sin p sin y).
The expressions for cos P* and cos y* quoted in Table 2.1 may be obtained
by cyclic permutation of the parameters.
2.B Some geometrical calculations concerning

directions and planes
All crystallographic planes parallel to a given direction are said to belong to
the same zone. According to p. 7 any vector ru = ua + vb wc defines the +
crystallographic direction [uvw] which is also the symbol of the zone. We
derive now the following results:
1. Angle 4 between two planes H, = (hlklll) and H2 = (h2k212):
* *
, 4
. cos , = rHl ' d H l d H 2 r.irli 2
r ~ l r ~ 2
where r:, r i , may be calculated by means of eqn (2.17a).

2. The family of planes H = (hkl) belongs to the zone [uvw] if
Indeed if (2.B.1) is verified then r i is normal to r, so that the planes (hkl)

are parallel to ru.
3. Symbol of the zone [uvw] defined by two planes HI and Hz. Because of
(2.B.1)
4. The planes HI, H2, H3 shall belong to the same zone if r i , , r i , , r i , are
coplanar. Then they shall define a cell in the reciprocal lattice whose volume
is zero:
5. The condition that the point P(x, y, z) lies in a plane (the nth of the set
from the origin) of the family (hkl). We require that the projection of
+ +
r = xu yb zc onto the direction of r i be equal to n times the interplanar
spacing d H :
r - r i / r : = ndH= n l r i
from which
.
r r i = hx + ky + lz = HX = n. (2.B.2)
126 1 Carmelo Giacsvazzo
6. The condition that the point P(h, k, 1) of the reciprocal lattice lies in
the nth (starting from the origin) plane of the set of planes (uvw) of the
reciprocal lattice. It is the same problem as in 5 above, but transferred in
the reciprocal lattice. The required condition is therefore
Note that the lattice vector ru = ua + vb + wc defines the direction [uvw]

orthogonal to the planes (uvw) of the reciprocal lattice. Therefore eqn
(2.B.3) defines the reciprocal lattice points which lie on a plane normal to
the direction [uvw] defined in the direct space. As it will be seen on p. 247,
these points give rise to the nth layer line in a rotation photograph when the
crystal is rotated about the axis [uvw].
7. Condition for the existence of a lattice plane perpendicular to the
direction [uvw]. Such a plane will exist if a triplet of integer numbers h, k, 1
can be found for which
ri A r u = (ha* + kb* +lc*) A (ua + v b + wc) =O.
In cubic lattices h, k, 1 coincide with u, v, w respectively (in other words the
plane (uvw) is perpendicular to the direction [uvw]. Indeed ru= 0
owing to the fact that
= O u a * ~ v b + v b * ~ u .a.=. = 0.
a * ~ a = b * ~ b = c * ~ c and
8. Line determined by two points: (r - rl) A (r - r2) = 0, where rl and r2
are the positional vectors of the two points.
9. Line parallel to the direction n and going through the point ro:(r -
ro) A n = 0.
10. Line through the point Po and perpendicular to two unitary vectors u
and v:(r - ro) A (u A v) = 0.
11. Projection of the vector r, on to the unitary vector n (from now on we
will denote by N and N* the matrices of the components of n with respect to
direct and reciprocal bases respectively):
12. Plane determined by the three points Pi, P2, P3 whose positional
vectors are P l , P2, p3:(r -PI) ' (p2 -pi) A (p3 -PI) = 0.
In terms of coordinates
("'
X~-XI
Y -Y1
y2-yl
"I)
22-21 =O. (2.B.4)
x3 - X1 Y3 - Yl 23 - 21
Equation (2.B.4) derives from (2.5): indeed the unit cell volume defined by
the vectors
i =P i , r 2 = p 2-pl, r3 =P3
is vanishing.
13. Plane normal to the unitary vector n:r n = d, where d is the distance
of the plane from the origin. It may also be written
r . n = ~ * ~ = d . (2.B.5)
14. Plane normal to the unitary vector n and through the point defined by
ro:
( r - r , ) . n = N*(x-x,) =o. (2.B.6)
15. Distance from P1 to the plane N*X - d = 0:
D = N*Xl - d. (2.B.7)
If the plane is defined by means of the equation r rn = d', where rn is a
general vector, before applying (2.B.7) n = rnln and d = d'lm have to be
calculated.
16. Projection of the vector rl on to the plane N*X - d = 0:
P(rl (1 n*) = rl - ( N * x , ) ~ *
17. Principal axes of the symmetry operator R. A vector r along an axis
of R should satisfy the eigenvalue equation Rr = Ar.
To give an example, let R = R, in a Cartesian system. The secular
equation will be
cos8-A -sin8
det (R, - Al) = det
= (1
(+ sin 8
0
cos 6' - A
0 1-A
- 2~ cos 8)(1 - A)= 0.
A = 1 is one root. If sin 8 # 0 the other eigenvalues are A, = eie, A3 = e-ie.

The corresponding eigenvalues are [O, 0,1], [l/fl, -i/fi, 01, [I/$&
i l f i , 01 respectively.
The above result may be generalized to a rectilinear coordinate system:
there will always be a real eigenvalue the eigenvector of which corresponds
to the direction of the symmetry axis. For a proper rotation axis the
eigenvalue is + 1, for an improper axis it is -1. If sin 6' = 0 all the
eigenvalues are real. If two eigenvalues are equal, any linear combination of
the corresponding eigenvectors is an eigenvector itself. For example, the
matrix 11 i 010 i 010 0 11 represents the twofold symmetry axis 211001in a
hexagonal framework. The eigenvectors corresponding to the doubly
degenerate eigenvalues -1 are normal to [loo] . Furthermore, the identity
operator has three eigenvalues 1, 1, 1: the eigenvectors are any three
unitary vectors orthogonal to each other.
2.C Some transformation matrices

Transformation matrices conventionally used to generate primitive from
centred cells are shown in Table 2.C.1.
2.D Reciprocity of F and I lattices

Let AF = (aF, bF, cF), the basis vectors of a face-centred lattice, and
AP=(aP, bp, cp), the basis vectors of a primitive unit cell for the same
lattice. Then Ap = MAF, where M is one of matrices shown in Table 2.C.1:
more explicitly
Table 2.C.1. Transformation matrices M conventionally used to generate centered from

primitive lattices and vice versa, according to the relationships A' = M A
According to Table 2.E. 1 A: = (M)-IA:, from which
or, in matrix notation, A: = QA:, where Q coincides (see Table 2.C.1) with
the matrix which transforms an I cell into a P cell. It may be concluded that
the reciprocal of an F lattice is an I lattice whose cell is defined by the vector
2a:, 2b:, 2c:.
If we index the reciprocal lattice with respect to A,* and A: we obtain
where all of h, = 2h1, k , = 2k1, I , = 21, will either assume even values (when
h,, k,, 1, are integer numbers) or odd values (when hI, k,, 1, are of type m/2,
n/2, p / 2 with integer values of m, n, p). These are just the conditions for
systematic extinctions in an F lattice.
2.E Transformations of crystallographic quantities in

rectilinear spaces
In this paragraph we describe the transformation properties of crystal-
lographic quantities not studied on pp. 65-7.
Let us first prove that the basis vector transformation (2.18) (from A to
A') transforms the space group symmetry operator C, = (R,, T,) into
C; = (Rk, T;), where
R; = (M)-~R,M, T; = (MI-IT,. (2.E.1)
+
The relation X, = RpX T, is transformed, according to (2.20), into
+ +
MX; = R,MX' T, from which X; = (M)-~R,MX' (M)-'T, which coin-
cides with (2.E.1). Note that the trace of R is invariant under transforma-
tions such as (2.E.l).
A particular case of (2.E.1) occurs when A' =A*. Then
R,*=GR,G*, T,*=GT,. (2.E.2)
If the second of the equations (2.9) is introduced into the first equation
(2.E.2) the following result is obtained
R,* = (R,)-~GR;~R,G-~
= (R,)-l. (2.E.3)
In conclusion, if the operator R is a symmetry element in the direct space,
(R)-' is a symmetry element in the reciprocal space. The list of the
symmetry operators in the reciprocal space may be obtained without
inverting the various matrices. Indeed, if R is a symmetry element in the
direct space, group properties guarantee that R-' is also an element of the
direct space symmetry group: therefore R is a symmetry operator of the
reciprocal space symmetry group. Consequently, if the set of matrices R
operate in direct space, the set R operate in reciprocal space (however, R,
and R, may pertain to different symmetry elements).
In an orthonormal system a* = a, b* = b, c* = c and G = I. Therefore any
symmetry operator C = (R, T) will have identical expression both in direct
and in the reciprocal space. Furthermore R = R-' holds, as already obtained
in Appendix 1.A.
It is often useful to know the transformation rules valid in reciprocal
space when the basis vector transformation A' = MA is performed in direct
space. We describe some of them:
1. A* e A 1 * : according to (2.25), A' = GIA'* = MGA*. Owing to (2.21),
that gives
2. X*@X1*: by analogy to point (1) it may be obtained
Note that relations (2.E.5) are the transformation rules of the Miller
indices (hkl).
3. G * e G 1 * : let us introduce the first eqn (2.16) in the first eqn (2.21),
which thus becomes GI*-' = MG*-IM. On post-multiplying both sides of
this equation by GI* we obtain I = MG*-~MG'*from which
4. C * e C 1 * : introduce the first eqn (2.E.2) in the first eqn (2.E.1). We

obtain G'-'RYG' = (M)-'G-~R,*GM, which, introduced into (2.E.6)
gives
RA* = MR,*M-' and TA* = MT,* (2.E.7)
Note that (2.E.7) represents the transformation rules for the reciprocal
space symmetry operators generated by a basis vector change in direct
space.
5. Q * e Q 1 * .According to (2.E.5) it will be
x*Q*x*= x l * ( ~ ) - l Q * ~ - l x ' *
Table 2.E.1. Transformation relationships. In the table M is the
matrix transforming A - (a, b, c) into A' = ( a ' ,b ' , c'). G and G'
- -
are the metric matrices of A and A' respectively, G* and GI* are
the metric matrices of A* = ( a * , b*, c*) and A'* ( a r * ,b ' * , c ' * )
respectively. C (R,T) is a symmetry operator (R is its rotational
part, T its translational part): C, C ' , C*, C'* are symmetry
operators defined in A, A', A*, A'* respectively. Q and Q* are
the quadratic forms of A and A*.
from which
Qr* = (M)-'Q*M-', Q* = MQ'*M. (2.E.8)
The transformation rules obtained in this paragraph and on pp. 65-7 are
collected in Table 2.E.1.
A particular basis transformation is that corresponding to a symmetry
operation. In accordance with eqns (2.19) and (2.20) a transformation R
acting on the coordinates is equivalent to the transformation M = (R)-'
acting on the basis vectors. In this case the metric matrix will not vary:
indeed, according to (2.21), G' = (R)'GR-~ which is identical (because of
(2.9)) to G. Find now the relationships existing between the matrices 8' and
p defining the anisotropic temperature factors of the atoms related by a
symmetry operator. In accordance with (2.E.8)
As an example, the reader will easily verify that the relationships existing in
the cubic system along the components of p of two atoms related by the
symmetry axis 311111(see the matrix given in Appendix l.D) are
Pi1 = P331 Pi2 = P137 Pi3 = P23)
2.F Derivation of the normal equations

Because of (2.50) the parameter S in (2.52) may be written as
S = V M ~ ' V = F M ~ ' F - X A M ~ ' F - ~ M ~ ' A X + ~(2.F.1)
M~'AX.
By _applying to (2.F.1) the differential operator 6 we obtain (note that
SX = 6X)
Since X and 6X are vectors, the first two terms on the right-hand side of
(2.F.2) are 1x 1 matrices which are equal (My1 = ~ 7 ' ) . For the same
reason the third and the fourth terms are equal too. Therefore
6s = ~(%)(AM;'AX - AM;~F) =0
from which
(AM,-'A)X = A M ~ ~ F ,
which coincides with (2.53).
2.G Derivation of the variance-covariance matrix M x

Because of (2.55), eqn (2.56) may be written as
M, = ( ( B - ~ A M ~-~A-~F)(B-~AM;~F
F - A-~F);). (2.G.1)
According to (2.54) A-' = B-~AM;~so that (2.G.1) becomes
M, = B-~AM;~((F - F)(F - F))M;'AB-'= B-~AM;~M~M;~AB-~= B-'.
2.H Derivation of the unbiased estimate of M,

Let us calculate (2.F.1) for X = 2 : then
3 = FM;~F - ~ A M F-~(FM;~A
F -~ A M Y ~ A ) ~ . (2.H.1)
Because of (2.54) and (2.55) the eqn (2.H.1) reduces to
Y
S = FM,-'F- ~ B X (F-
= F)M,-I(F- F ) - (X - X)B(X-X)
where X = A-IF.The expected value of 3 will then be
(9) = ((F - F)Myl(F -F)) - ((X - X)B(X - x)).
(F - F ) and ( 2 - X) are random variables with zero means and finite
variances, and with variance-covariance matrices of rank n and m
respectively. In this condition it is possible to show[451
that
7
((F - F)M,-'(F - F ) ) = n, ((X - x)M;~(% - X)) = m.
Therefore
I f Mf is known within a scale factor (Mf = K,Nf) then
which coincides with (2.58).
2.1 The FFT algorithm and its crystallographic

applications
Suppose that we want to calculate the one-dimensional transform
p(x) = F(h ') exp (2nihrx) (2.1.1)
h'
where -N/2 s h ' < N/2. Subdivide the x axis in N parts so that
Denoting h' = h - N/2 gives

exp (2nih1x)= exp (2nihjlN) exp (-nij) = (-1y exp (2nihjlN)
where 0 G h < N. Then (2.A.21) becomes
N- 1
p(j) = (- 1y' 2 F(h) exp (2nihjlN)
h =O
where wN = exp (2nilN).

Because of the transformation h ' + h the negative indices are removed:
furthermore, the number of Fourier coefficients is chosen equal to the
number of points in which p has to be calculated (the Fh array may
eventually be completed with a number of zeros).
-
If N is not a prime number, say N = r s (of particular interest and
simplicity is the case N = 2")) then the expression hls ho will generate all+
the integers h in the range [0, (N - I)]:
Similarly, the expression jlr + jo will generate all the integers j in the range
[O, (N - 111
j=jlr+jO ( = O 1 .- 1 ; jo=O, 1 , . . . , r - 1 ) .
Then
W# = W , h o i ~ W f ~Ni ~ W h ~ i ~
and
s-1 r-1
p(j) = p(jo, jl) = (-1y 2 2 F(hl, ho)w,hoJ1wfljOh"io
ho=O h1=0
WN
where F(hl, ho) represents the Fh factor obtained by combination of hl and

ho in eqn (2.1.2).
As in the Beevers-Lipson procedure the expression (2.1.3) may be
calculated in two
Step 1: calculate
for jo = 0, 1, . . . , r - 1 and ho = 0, 1, . . . , s - 1. T(jo, h,) values can

replace F(hl, ho) values in the memory of a computer since these last are
no longer necessary,
Step 2: calculate
forjo=O, 1 , . . . , r - 1 and jl=O, 1 , . . . , s - 1 .

While the evaluation of (2.1.3) by the classical method requires N2
+
operations, the FFT method involves r(s r) operations in the first step and
+
s(s r) operation in the second step. If N is sufficiently large that
corresponds to a large saving of time.
Besides calculating electron density maps, the FFT method is also used
for the calculation of the structure factors when the number of atoms is very
large and computing time has to be saved. Calculations are organized into
two steps:
1. Step 1: All the atoms (in the asymmetric unit) which contribute to the
electron density are selected. For each of them the electron density is
nothing else but the Fourier transform of the atom scattering curve
corrected for thermal motion. Usually each atom is represented by a
single Gaussian function (for low resolution data) or as the sum of two or
three Gaussian components. The overall model electron density map is
sampled on a grid not too fine (that would greatly increase cost and
computer storage) and not too coarse (in order to avoid too rough an
approximation of the electron density).
2. Step 2: The fourier inversion of the map is made, which provides
structure factor magnitudes and phases.
The speed of steps 1 and 2 depends on the number of grid points in which
the density has been sampled and on the number of terms in the Gaussian
approximation to the atomic scattering curve. Too coarse a sampling grid
would produce structure factors which are the sum of the desired ones and
those with indices spaced away by multiples of N. Ten ~ i c k [ ' proved
~ ~ ] that
the number of grids may be taken small by artificially increasing the B value
of the atoms.
Besides structure factor calculations most of the steps involved in LSQ
structure refinement can be performed by the FFT algorithm:[125~1261 i.e. the
calculation of the gradient vector D and of the normal matrix B (see p. 92).
Indeed, in accordance with p. 145, the following properties of the Fourier
transform hold:
Real space Reciprocal space
6dr)
6 FT -* (-2nik)F(H)
6Y
etc.
These more recent developments have made possible the use of a

full-matrix least-squares for macromolecule structure at a reasonable cost.
2.J Examples of t w i n laws

Some examples of twin laws are described here, with special references to
minerals. From the figures the reader will conclude that morphology
exhibited by twins often shows re-entrant angles, which are not displayed by
single crystals.
A
Cubic system
Twinning according to the spinel law is commonly found in crystals of the
class 43m, which often exhibit forms (111). Twinning occurs with (111) as
a twin plane (see Fig. 2.J.l(a)). The same law applies to the penetration
twin shown in Fig. 2.J.l(b), where the two cubes are rotated about [ I l l ]
(classical mineral: fluorite). An interpenetrant rotation twin about [ l l i ] is
shown in Fig. 2.J.l(c).
(a) (b)
Crystals of pyrite, point group m3, frequently obey the iron cross law:
they twin with (110) as a twin plane.
In Fig. 2.J.l(d) the classical pentagonal dodecahedron is shown, with
form e = (210): twinned crystals display the form shown in Fig. 2.J.l(e).
Tetragonal system
Cassiterite (SnO,, point group 4/mmm), presents the so called elbow twin,
twin plane {101),
. .
shown in Fig. 2.J.2. The elbow twin is also found in rutile
\Iiiil ( ~ i 0 2 and
) zircon (ZrSiO,), which are often polysynthetic.
(c) / Calcopyrite crystals (CuFeS,, point group 42m) are commonly tetra-
hedral, with (112) as the dominant form. Twinning occurs on (1121,
contact, lamellar, or penetration.
Hexagonal and trigonal systems

Twinning is rare in minerals of the hexagonal system and frequent in
minerals of the trigonal system.
The twin plane (and the composition plane) in calcite (CaCO,, point
group 3m) is often (0001) (see Fig. 2.J.3(a)), more frequently (1102) (see
Fig. 2.J.1. Spinel law. (a) Twin plane (1 11); (b) Fig. 2.J.3(b)). This last twin is readily provoked by mechanical deformation
penetration twin; (c) interpenetrant rotation (for example, the pressure of a razor blade on a rhombohedron of calcite).
twin; (d) classical pentagonal dodecahedron of
pyrite; (e) iron cross law.
The twinning is due to the ability of structural layers parallel to (1102) to
slide past each other.
Quartz (SiO,, point group 32) twins in several ways. The most important
are listed below.
1. Brazil twins, with (1120) as the twin plane. This combines right- and
left-handed crystals in a penetration twin (often with plane composition
surface). A left-handed single crystal is shown in Fig. 2.J.4(a), a
right-handed crystal is shown in Fig. 2.J.4(b), and the Brazil twin is
displayed in Fig. 2.J.4(c). Such types of twin are useless for electrical
work, and may be detected in polarized light (the plane of polarization is
rotated in opposite directions by the two parts of the twin).
2. Dauphine twins, with c as twin axis. Two right- or two left-handed
individuals are combined, separated by a very irregular surface. Then the
faces (10i1) of one individual and the faces (0111) of the other will
Fig. 2.J.2. Elbow twin.
coincide (see Fig. 2.J.4(d)).
3. Japan twins, contact twins with (1122) as twin and composition plane
(see Fig. 2.J.4(e)). The c axes intersect at 84'33', one pair of faces
(1010) is common to both pairs of twins.
Orthorhombic system
Aragonite (CaCO,, point group mmm, polymorphous with calcite) often
twins on (110) faces. In Fig. 2.J.5(a) a single crystal of aragonite in shown,
in Fig. 2.J.5(b) a twinned crystal is displayed, in Fig. 2.J.5(c) the structural

scheme of the twin is given.
Twinning is often repeated and produces interpenetrant triplets of
pseudo-hexagonal shape (Fig. 2.J.5(d)).
Staurolite (A14FeSi2010(OH),, point group mmm) forms cruciform
penetration twins: if the twin plane is (032) then the two individuals form a
nearly right-angled cross (Fig. 2.J.6(a)); an oblique cross (angle =60°) is
formed by twinning on (232) (see Fig. 2.J.6(b)).
Monoclinic system
Orthoclase crystals (KAlSi,08, point group 2/m) often twin according to the
Carlsbad, Baveno, or Manebach law.
In the Carlsbad law c is the twin axis and (010) the composition plane
(see Fig. 2.J.7(a)). The twin and composition plane is (021) in Baveno
twins (see Fig. 2.J.7(b)), and (001) in Manebach twins (see Fig. 2.J.7(c)).
While Carlsbad twins are penetrating twins, Baveno and Manebach are
usually contact twins.
Triclinic system
In twins of plagioclase feldspars [(Ca, Na)(Al, Si)A1Si208, point group i] (b)
the albite law is often satisfied: twin plane (010) , type of twinning usually Fig. 2.5.3. Some examples of calcite twins. (a)
multiple and polysynthetic (see Fig. 2.J.8(a,b,c)). Also common is the Plane {OOO1); (b) twin plane {lio2).
Fig. 2.J.4. Quartz twins. (a) Left-handed quartz;

(b)right-handed quartz; (c) Brazil twin: the
twinning plane is indicated by a broken line; (d)
Dauphin6 twin: interpenetration of two left-hand
crystals.
Fig. 2.J.5. Aragonite twinning. (a) Single crystal;

(b) twinned crystal; (c) structural scheme of the
twin; (d) triplet.
1- -1
(c) (4
(c) Fig. 2.J.8. Plagioclase feldspars. (a) Single
Fig. 2.J.6. Staurolite twinning. (a) Twin after Fig. 2.J.7. Orthoclase twins. (a) Carlsbad twin; crystal; (b) twin by Albite law; (c) polysynthetic
{032}; (b) twin after {232). (b) Baveno twin; (c) Manebach twin. albite twin; (d) twin by pericline law.
pericline law, with twin axis [OlO]. The composition plane is the so-called
'rhombic section', a plane parallel to b whose orientation does not
correspond to rational indices; see Fig. 2.J.8(d).
References
1. Sands, D. E. (1982). Vectors and tensors in crystallography. Addison-Wesley,
Reading.
2. Klyne, W. and Prelog, V. (1960). Experientia, 16, 521.
3. Schomaker, V., Waser, J., Marsh, R. E., and Bergman, G. (1958). Acta
Crystallographica, 8, 600.
4. Niggli, P. (1982). Handbuch der Experimentalphysik, Vol. 7, Part 1. Academ-
ische Verlagsgesellschaft, Leipzig.
5. Delaunay, B. N. (1983). Zeitschrift fur Kristallographie, 84, 109.
6. Katayama, C. (1986). Journal of Applied Crystallography, 19, 69.
7. Burzlaff, H , and Zimmermann, H. (1984). Zeitschrift fur Kristallographie, 170,
241.
8. Burzlaff, H, and Zimmermann, H. (1984). Zeitschrift fur Kristallographie, 170,
247.
9. Buerger, M. J. (1957). Zeitschrift fiir Kristallographie, 109, 42.
10. Buerger, M. J. (1960). Zeitschrift fur Kristallographie, 1l3, 52.
11. Gruber, B. (1973). Acta Crystallographica, A29, 433.
12. Santoro, A. and Mighell, A. D. (1970). Acta Crystallographica, A26, 124.
13. Krivy, I. and Gruber, B. (1976). Acta Crystallographica, A32, 297.
14. Clegg , W. (1981). Acta Crystallographica, A37, 913.
15. Ferraris, G. and Ivaldi, G. (1983). Acta Crystallographica, A39, 595.
16. Andrews, L. C. and Bernstein, H. J. (1988). Acta Crystallographica, A44,
1009.
17. Azaroff, L. 17. and Buerger, M. J. (1958). The powder method in x-ray
crystallography. McGraw-Hill, New York.
18. Mighell, A. D. and Rodgers, J. R. (1980). Acta Crystallographica, A36, 321.
19. Mighell, A. D. and Himes, V. L. (1986). Acta Crystallographica, A42, 101.
20. Gruber, B. (1989). Acta Crystallographica, A45, 123.
21. Santoro, A. and Mighell, A. D. (1972). Acta Crystallographica, A28, 284.
22. Fortes, M. A. (1983). Acta Crystallographica, A39, 351.
23. Grimmer, H. (1976). Acta Crystallographica, A32, 783.
24. Grimmer, H. (1974). Scripta Metallurgica, 8, 1221.
25. Bolmann, W. (1970). Crystal defects and crystalline interfaces. Springer, Berlin.
26. Ranganathan, S. (1966). Acta Crystallographica, A21, 197.
29. Bonnet, R., Cousineau, E., and Warrington, D. M. (1981). Acta
Crystallographica, A37, 184.
31. Donnay, G. and Donnay, J. D. H. (1974). Canadian Mineralogist, 12, 422.
32. Le Page, Y., Donnay, J. D. H., and Donnay, G. (1984). Acta
33. Friedel, G. (1926). Legons de cristallographie. Paris: Berger-Levrault, Paris.
(Reprinted 1964. Blanchard, Paris.)
34. Santoro, A. (1974). Acta Crystallographica, A30, 224.
35. Catti, M. and Ferraris, G. (1976). Acta Crystallographica, A32, 163.
36. Van der Sluis, P. (1989). Thesis. University of Utrecht.
37. Flack, H. D. (1987). Acta Crystallographica, A43, 564.
38. Britton, D. (1972). Acta Crystallographica, A28, 296.
39. Stanley, E. (1972). Acta Crystallographica, 5, 191.

40. Rees, D. C. (1982). Acta Crystallographica, A38, 201.
41. Yeates, T. 0 . (1988). Acta Crystallographica, A44, 142.
42. International tables for x-ray crystallography (1962), vol. 111. Kynock Press,
Birmingham.
43. Cromer, D. T. and Mann, J. B. (1968). Acta Crystallographica, A24, 321.
44. Cooley, J. W. and Tukey, J. W. (1965). Mathematics of Computation, 19, 297.
45. Hamilton, W. C. (1964). Statistics in physical science. Ronald Press Company,
New York.
46. Rogers, D. (1981). Acta Crystallographica, A37, 734.
47. Berardinelli, G. and Flack, H. D. (1985). Acta Crystallographica, A41, 500.
48. Larson, A. C. (1970). In Crystallographic computing, pp. 291-4. Munksgaard,
Copenhagen.
49. Zachariasen, W. H. (1963). Acta Crystallographica, 16, 1139.
51. Report of the International Union of Crystallography. Subcommittee on
Statistical Descriptors (1989). Acta Crystallographica, A45, 63-75.
52. Rollett, J. S. (1988). Error analysis. In Crystallographic computing 4. Tech-
niques and new technologies (ed. N. W. Isaacs and M. R. Taylor), pp. 149-66.
Oxford University Press and International Union of Crystallography, Oxford.
53. Carruthers, J. R. and Watkin, D . J. (1979). Acta Crystallographica, A35, 698.
54. Wang, H. and Robertson, B. E. (1985). In Structure and statistics in
crystallography (ed. A. J . C. Wilson), pp. 125-35. Adenine Press, Guiderland,
NY.
55. Prince, E. (1982). Mathematical techniques in crystallography and material
sciences. Springer, New York.
56. Seiler, P., Schweizer, W. B., and Dunitz, J. D. (1984). Acta Crystallographica,
B40, 319.
57. Hamilton, W. C. (1965). Acta Crystallographica, 18, 502.
58. Prince, E. and Nicholson, W. L. (1985). In Structure and statistics in
crystallography (ed. A. J . C. Wilson), pp. 183-95. Adenine Press, Guiderland,
NY.
61. French, S. and Wilson, K. (1978). Acta Crystallographica, A34, 517.
62. Gonschorek, W. (1985). Acta Crystallographica, A41, 189.
63. Scheringer , C. (1963). Acta Crystallographica, 16, 546.
64. Doedens, R. J. (1970). In Crystallographic computing (ed. by F. R. Ahmed),
pp. 198-204. Munksgaard, Copenhagen.
65. Domenicano, A. and Vaciago, A. (1975). Acta Crystallographica, B31, 2553.
66. Waser, J. (1963). Acta Crystallographica, 16, 1091.
67. Watkin, D. (1988). In Crystallographic computing 4. Techniques and new
technologies (ed. N. W. Isaacs and M. R. Taylor), pp. 111-29. Oxford Science
Publications and International Union of Crystallography, Oxford.
68. Wolff, P. M. de (1968). Journal of Applied Crystallography, 1, 108.
69. Taupin, D. (1988). Journal of Applied Crystallography, 21, 485.
70. Wu, E. (1988). Journal of Applied Crystallography, 21, 530.
71. Shirley, R. (1978) In Computing in crystallography (ed. H . Schenk, 0.
Olthof-Hazekamp, H. van Koningsveld, and G. C. Bassi), pp. 221-34. Delft
University Press.
72. Pawley, G. S. (1981). Journal of Applied Crystallography, 14, 357.
73. Louer, D. and Vargas, R. (1982). Journal of Applied Crystallography, 15, 542.
74. Werner, P. E., Ericksson, L., and Westdahl, M. (1985). Journal of Applied
Crystallography, 18, 367.
75. Paszkowicz, W. (1987). Journal of Applied Crystallography, 20, 166.
76. Taupin, D . (1989). Journal of Applied Crystallography, 22, 455.

77. Rietveld, H. M. (1969). Journal of Applied Crystallography, 2, 65.
78. Caglioti, G., Paoletti, A., and Ricci, F. P. (1958). Nuclear Instruments and
Methods, 3, 223.
79. Attfield, J. P., Cheetham, A. K., Cox, D. E., and Sleight, A. W. (1988).
Journal of Applied Crystallography, 21, 452.
80. Baerlocker, C. (1984). Proceedings of the 6th International Zeolite Conference,
Reno, USA, July 1983, p. 823. Butterworth, Guildford.
81. Young, R. A. and Wiles, D . B. (1982). Journal of Applied Crystallography, 15,
430.
82. Immirzi, A. (1980). Acta Crystallographica, B36, 2378.
85. Jansen, E., Schaefer, W., and Will, G. (1988). Journal of Applied
86. Howard, S. A. and Snyder, R. L. (1989). Journal of Applied Crystallography,
22, 238.
87. Sakata, M. and Cooper, M. J. (1979). Acta Crystallographica, 12, 554.
88. Durbin, J. and Watson, G. S. (1971). Biometrika, 58, 1.
89. Theil, H. and Nagar, A. L. (1961). Journal of the American Statistical
Association, 56, 793.
90. Hill, R. J. and Madsen, I. C. (1987). Powder Diffraction, 2, 146.
91. Will, G., Jansen, E., and Schafer, W. (1983). POWLS-80: a program for
calculation and refinement of powder diffraction data. Report Juel-1967. KFA
Jiilich.
92. Ahtee, M., Nurmela, M., Suortti, P., and Jawinen, M. (1989). Journal of
Applied Crystallography, 22, 261.
93. Dahms, M. and Bunge, H. J. (1989). Journal of Applied Crystallography, 22,
439.
94. Parrish, W., Hart, M., and Huang, T. C. (1986). Journal of Applied
Crystallography, , 19, 92.
95. Will, G., Parrish, W., and Huang, T. C. (1983). Journal of Applied
96. Dollase, W. A. (1986). Journal of Applied Crystallography, 19, 267.
97. Young, R. A. and Sakthivel, A. (1988). Journal of Applied Crystallography,
21, 416.
98. Wilson, A. J. C. (1963). Mathematical theory of x-ray powder diffractometry.
Centrex, Eindhoven.
99. Prince, E. (1983). Journal of Applied Crystallography, 16, 508.
100. Klug, P. H. and Alexander, L. E. (1974). X-ray diffraction procedures. Wiley,
New York.
101. Kauppinen, J. K., Moffatt, D. J., Mantsch, H. H., and Cameron, D. G.
(1981). Applied Spectroscopy, 35, 271.
102. Toraya, H. (1988). Journal of Applied Crystallography, 21, 192.
103. Wiedemann, K. E., Unnam, J., and Clark, R. K. (1987). Powder Diffraction 2,
3, 130.
104. Christensen, A. N., Lehmann, M. S., and Nielsen, M. (1985). Australian
Journal of Physics, 38, 497.
105. Louer, D , and Langford, J. I. (1988). Journal of Applied Crystallography, 21,
430.
106. Cheetham, A. K. (1986). High resolution powder diffraction, Materials Science
Forum, Vol. 9 (ed. C. R. A. Catlow), pp. 103-12. Transtech Publications
Limited, Switzerland.
107. Cascarano, G., Favia, L., and Giacovazzo, C. (1990). XI.' IUCr Congress,
Bordeaux, abs. 02.08.22, page C77.
108. Lehmann, M. S., Christensen, A. N., Fjellvig, H., Feidenhans, R., and
Nielsen, M. (1987). Journal of Applied Crystallography, 20, 123.
109. Hibble, S. J., Cheetham, A. K., Bogle, A. R. L., Wakerley, H. R., and Cox,
D. E. (1988). Journal of the American Chemical Society, 110, 3295.
110. Maichle, J. K., Ihringer, J., and Prandl, W. (1988). Journal of Applied
111. Cheetham, A. K. and Taylor, J. C. (1977). Journal of Solid State Chemistry,
21, 253.
112. Meille, S. V., Briickner, S., and Lando, J. B. (1989). Polymer, 30, 786.
113. Briickner, S., Meille, S. V., Porzio, W., and Ricci, G. (1988).
Makromolekulare Chemie, 189, 2145.
114. Cruickshank, D. W. J. (1956). Acta Crystallographica, 9, 754.
115. Schomaker, V. and Trueblood, K. N. (1968). Acta Crystallographica, B24, 63.
116. Johnson, C. K. (1970). In Crystallographic computing (ed. F. R. Ahmed), pp.
207-19. Munksgaard, Copenhagen.
117. Busing, W. R. and Levy, H. A. (1964). Acta Crystallographica, 17, 142.
118. Cruickshank, D. W. J. and Robertson, A. P. (1953). Acta Crystallographica, 6,
698.
119. Stanford, R. H. Jr and Waser, J. (1972). Acta Crystallographica, A28, 213.
120. Nardelli, M. (1983). Computers and Chemistry, 7, 3, 95.
121. Shmueli, V. (1974). Acta Crystallographica A30, 848.
122. Ten Eick, L. F. (1973). Acta Crystallographica, A29, 183.
123. Ten Eick, L. F. (1977). Acta Crystallographica, A33, 486.
124. Immirzi, A. (1976). In Crystallographic computing techniques (ed. F. R.
Ahmed), pp. 399-412. Munksgaard, Copenhagen.
125. Aganval, R. (1978). Acta Crystallographica, AM, 791.
126. Jack, A. and Levitt, M. (1978). Acta Crystallographica, AM, 931.
The diffraction of X-rays
by crystals
Introduction
Crystal structure analysis is usually based on diffraction phenomena caused
by the interaction of matter with X-rays, electrons, or neutrons. Although
the theory of diffraction is the same for all types of radiation, we shall
consider X-ray scattering with particular interest: some references to
electron and neutron scattering are made in Appendix 3.B pp. 195 and 198.
The most important properties of X-rays were described by Rontgen in
1896. However, with equipment in common use in optics at that time he
could not measure any effect of interference, reflection, or refraction.
Several years later Sommerfeld measured an X-ray wavelength of about
0.4 A. In 1912 M. von Laue, starting from an article by Ewald, a student of
Sommerfeld, suggested the use of crystals as natural lattices for diffraction.
This experiment was successfully performed by Friedrich and Knipping,
both students of Rontgen. In 1913 W. L. Bragg and M. von Laue used
X-ray diffraction patterns for deducing the structure of NaC1, KC1, KBr,
and KI. In such a way, in only few years, the electromagnetic nature of
X-rays and their usefulness in the determination of crystal structure was
indisputably demonstrated.
Let us recall some properties of electromagnetic radiation:
1. Electromagnetic radiation propagates in a vacuum with the velocity c

equal to 300 000 km s-'.
2. The vectors electric field E and magnetic field H are both orthogonal to
the direction of wave propagation. In addition, they are mutually
orthogonal and vary sinusoidally with time.
3. The range of X-ray wavelengths is placed between the ultraviolet region
and the region of y-rays emitted by radioactive substances. The interval
of wavelengths of particular usefulness in crystallography ranges between
0.4 and 2.5 A.
4. The refractive index of X-rays is very near to unity: for A = 2 A and for
high-density substances the difference from unity is of the order lop4,
being lo-' for most cases. For this reason the X-rays cannot be focused
by means of suitable lenses like ordinary light or electrons. Thus, if
X-rays are used, we cannot talk about a direct observation of objects by
means of instruments equivalent to optical or electron microscopes.
The interaction of X-rays with matter essentially occurs by means of two

processes:
1. Some photons of the incident beam are deflected without a loss of
energy. They constitute the scattered radiation with exactly the same
wavelength as the incident radiation. Other photons are scattered with a
small loss of energy; they constitute the Compton radiation (see pp.
185-6) with wavelength slightly greater than the wavelength of the incident
radiation.
2. The photons can be absorbed by the atoms of a target and will increase
its temperature. Discontinuities in curves of absorption are caused by the
photoelectric effect. In this case the photon energy is used to remove one of
inner-shell electrons of the absorber element. This atom can come back to
its minimal energy state by emission of a photon X whose wavelength is
characteristic of the atom (fluorescent radiation). Absorption phenomena
will be discussed in Chapter 4.
The present chapter is mainly devoted to the so-called kinematic theory
of diffraction (only few brief remarks on some aspects of dynamic theory of
diffraction are given on pp. 162-5 and Appendix 3.B). Major emphasis is
given to diffraction by perfect crystals, but basic elements of diffraction by
gases, liquids, and amorphous solids are described in Appendix 3.C.
Diffraction theory is described by making use of Fourier transforms.
Readers not acquainted with such a mathematical concept are urged to
consult Appendix 3.A.
Thomson scattering
Q
,
Let suppose that (see Fig. 3.l(a)) a free material particle with electric
charge e and mass m is at the origin 0 of our coordinates system and that a
plane monochromatic electromagnetic wave with frequency v and electric
vector Ei propagates along the x axis in positive direction. Its electric field is
described by equation
Ei = Eoiexp 2niv(t - X/C)
where EOiis the amplitude of the wave and Ei is the value of the field at
position x at time t. The field exerts on the particle a periodic force F = eEi
and therefore the particle will undergo oscillatory motion with acceleration
a = F l m = eEi/m and frequency v. In accordance with classical theory of
electromagnetism a charged particle in accelerated motion is a source of
electromagnetic radiation: its field at r is proportional to acceleration and
lies in the plane (Ei, r). Let us orient the axes y and z of our coordinates
system in such a way that the observation point Q defined by vector r is in
the plane ( x , y). At the point Q we will measure the electric field Ed due to
scattered radiation
exp [2niv(t - rlc) - ia].
Ed= EOd
Thomson showed that (see also pp. 165-6)
Fig. 3.1. (a) A free charged particle is in 0:a
plane monochromatic electromagnetic wave
propagates along the x axis. (b)Surface element
1
Eod= - Eoi(e2/mc2)sin 9
at scattering angle 28. r
The diffraction of X-rays by crystals 1 143
where q, is the angle between the direction of acceleration of electron and

the direction of our observation. The term sin q, is a polarization term: we
'see' only the component of vibration parallel to the observer and normal to
the direction of propagation. a is the phase lag with which the charge
re-emits the incident radiation. The decrease of Ed with r is caused by the
scattering of radiation in all directions.
In terms of intensity eqn (3.1) becomes
e4
ZeTh = Zi 7 sin2 q,
mrc
where I,,, is the intensity of scattered radiation and Zi is the intensity of
incident radiation. This simple result excludes neutrons from the category of
X-ray scatterers because they do not have electric charge, and makes
negligible the contribution to scattering by protons whose factor (elm)2 is
about 18372 times less than that of electrons. Therefore, from now on and
according to tradition, the symbol e will represent the electron charge.
If the primary beam is completely polarized: (a) with Ei along the z axis,
then I,,, = 4e4/(m2r 2c4); (b) with Ei along the y axis, then I,,, =
~~e~cos228/(m2r2c4),where 28 is the angle between the primary beam and
the direction of observation. In general, the computation can be executed
by decomposing the primary beam into two beams whose electric vectors
are perpendicular and parallel respectively to the plane containing the
primary beam and the scattered radiation being observed. If K1 and K2 are
parts of these two beams in percentage we obtain
e4
ZeTh = 4
mrc
+
(K1 K2 C O S 28).
~
If the primary beam is not polarized, then K1 = K2 = 112 and
where P = (1 + cos228)/2 is called the polarization factor (see also p. 303).

It suggests that the radiation scattered in the direction of the incident beam
is maximum while it is minimum in the direction perpendicular to the
primary beam.
Equation (3.3) gives the intensity scattered into a unit solid angle at angle
28. If we want to obtain the total scattered power P we have to integrate
(3.3) from 0 to n (see Fig. 3.l(b)).
2nr2 sin 28 d(28)
where (2nr sin28) r(d(28)) is the surface element at angle 28. The total
scattering 'cross-section' PI4 is equal to 6.7 x cm2/electron, which is a
very small quantity. It may be calculated that the total fraction of incident
radiation scattered by one 'crystal' composed only of free electrons and
having dimensions less that 1mm is less than 2 per cent.
The scattered radiation will be partially polarized even if the incident
radiation is not. Thus, if the beam is scattered first by a crystal (monochro-
mator) and then by the sample the polarization of the beam will be
different. The scattering is coherent, according to Thomson, because there
is a well defined phase relation between the incident radiation and the
scattered one: for electrons a = n.
Unfortunately it is very difficult to verify by experiment the Thomson
formula since it is almost impossible to have a scatterer composed
exclusively of free electrons. One could suppose that scatterers composed of
light elements with electrons weakly bound to the nucleus is a good
approximation to the ideal Thomson scatterer. But experiments with light
elements have revealed a completely different effect, the Compton effect.
Compton scattering
The process can be described in terms of elastic collision between a photon
and a free electron. The incident photon is deflected by a collision from its
original direction and transfers a part of its energy to the electron.
Consequently there is a difference in wavelength between the incident
radiation and the scattered one which can be calculated by means of the
relation (see also Appendix 3.B, p. 185)
Ah (A) = 0.024 (1 - cos 28). (3.4)
The following properties emerge from eqn (3.4): Ah does not depend on
the wavelength of incident radiation; the maximum value of AA (Ah =
0.048) is reached for 28 = n (backscattering) which is small but significant
for wavelengths of about 1A. Besides, Ah = 0 for 28 = 0.
Compton scattering is incoherent; it causes a variation in wavelength but
does not involve a phase relation between the incident and the scattered
radiation. It is impossible to calculate interference effects for Compton
radiation.
Interference o f scattered waves

Here we shall not be interested in wave propagation processes, but only in
diffraction patterns produced by the interaction between waves and matter.
These patterns are constant in time since they are produced by the system of
atoms, which can be considered stationary. This fact permits us to omit the
time from the wave equations.
In Fig. 3.2 two scattering centres are at 0 and at 0'. If a plane wave
excites them they become sources of secondary spherical waves which
mutually interfere. Let so be the unit vector associated with the direction of
propagation of the primary X-ray beam. The phase difference between the
wave scattered by 0' in the direction defined by the unit vector s and that
scattered by 0 in the same direction is
Fig. 3.2. Point scatterers are in 0 and O',so and

s are unit vectors. Therefore A 0 = - r 9 so, BO = where
r s.
If A is much greater than r there will be no phase difference between the

scattered waves and consequently no appreciable interference phenomena
will occur. Since interatomic bond distances lie between 1 and 4 A no
interference effect could be observed by using visible-light waves.
The modulus of r* can be easily derived from Fig. 3.2:
r* = 2 sin elil (3.6)
where 28 is the angle between the direction of incident X-rays and the
direction of observation. If we outline two planes normal to r* passing
through 0 and 0' (OQ in Fig. 3.2 is the trace of the plane passing through
0 ) we can consider interference as a consequence of specular reflection with
respect to these planes.
If A, is the amplitude of the wave scattered by the material point 0 (its
phase is assumed to be zero) the wave scattered by 0' is described by
A,, exp (2nir* r). If there are N point scatterers along the path of the
incident plane wave we have
F(r*) = x Aj
N
j=1
exp (2nir* . I;.) (3.7a)
where Aj is the amplitude of the wave scattered by the jth scatterer.

The Thomson formula plays an essential role in all calculations to obtain
the absolute values of scattering. In our case it is more convenient to
express the intensity I scattered by a given object (for example, an atom) in
terms of intensity I,,, scattered by a free electron. The ratio I/IeTh is f2,
where f is the scattering factor of the object. Vice versa, for obtaining the
observed experimental intensity it is sufficient to multiply f by I,,. To give
an example, let us imagine a certain number of electrons concentrated at 0'
which undergo Thomson scattering. In this case forexpresses the number of
electrons.
According to the convention stated above eqn (3.7a) becomes
N
F(r*) = x h exp (2nir* ' I ; . ) .
j=1
(3.7b)
If the scattering centres constitute a continuum, the element of volume dr

will contain a number of electrons equal to p(r) d r where p(r) is their
density. The wave scattered on the element dr is given, in amplitude and
.
phase, by p(r) dr exp (2nir* r) and the total amplitude of the scattered
wave will be
F(r*) = 1
v
p(r) exp (2nir* r) dr = T[p(r)] (3.8)
where T represents the Fourier transform operator.

In crystallography the space of the r* vectors is called reciprocal space.
Equation (3.8) constitutes an important result: the amplitude of the
scattered wave can be considered as the Fourier transform (see Appendix
3.A, p. 175) of the density of the elementary scatterers. If these are
electrons, the amplitude of the scattered wave is the Fourier transform of
the electron density. From the theory of Fourier transforms we also know
that
p(r) = lv%
F(r*) exp (-2nir* .r) dr* = T-'[~(r*)]. (3.9)
Therefore, knowledge of the amplitudes of the scattered waves (in

modulus and phase) unequivocally defines p(r).
Scattering by atomic electrons

The processes of Thomson and Compton scattering are an example of
wave-particle duality and they seem to be mutually incompatible.
In fact both processes are simultaneously present and they are precisely
described by modern quantum mechanics. In common practice the scat-
terers are atomic electrons: they can occupy different energetic states
corresponding to a discontinuous set of negative energies and to a
continuous band of positive energies. If, after interaction with the radiation,
the electron conserves its original state the photon conserves entirely its
proper energy (conditions for coherent scattering). If the electron changes
its state, a portion of the energy of the incident photon is converted into
potential energy of an excited atom (conditions for incoherent scattering).
Quantum-mechanical calculations indicate that the processes of coherent
+
and incoherent scattering are simultaneously present and that I,,, Zincoe =
The coherent intensity I,,, can be calculated on the basis of the following
observations. An atomic electron can be represented by its distribution
function pe(r) = 1v(r)l2, where v ( r ) is the wave function which satisfies the
Schrodinger equation. The volume dv contains p, dv electrons and scatters
an elementary wave which will interfere with the others emitted from all the
elements of volume constituting the electron cloud. In accordance with
p. 145 the electron scattering factor will be
fe(r*) = pe(r) exp (2nir* r) dr (3.10)

S
where S is the region of space in which the probability of finding the

electron is different from zero. If we assume that p,(r) has spherical
symmetry (what is in fact justifiable for s electrons, less so for p, d, etc.
electrons) then eqn (3.10) can be written (see eqn (3.A.33)) as:
where Ue(r) = 4nr2pe(r) is the radial distribution of the electron and

r* = 2 sin 8/A. For instance, there are two 1s electrons, two 2s electrons,
and two 2p electrons in carbon atom. In radial approximation the 2s and 2p
electrons have an equivalent distribution. For carbon the Slater formulae
give
with cl = 10.77 A-l, c2 = 6.15 A-'. Then, eqn (3.11) gives
respectively.
Equation (3.12) are illustrated in Fig. 3.3(a) and eqns (3.13) in Fig.
3.3(b). In accordance with eqn (3.10) the electron scattering factor is equal
to 1 when r * = 0. Moreover, the scattering of 1s electrons, whose
distribution is very sharp, is more efficient at higher values of r*. If the
distribution of 1s electrons could really be considered point-like their
scattering factor would be constant with varying r* (see Appendix 3.A, p.
177 for the transform of a Dirac delta function).
According to the premise of this section the intensity of the Compton
radiation of an atomic electron will be
lincoe = ZeTh(l -f z)
where ZeTh is given by eqn (3.2) or eqn (3.3). The intensity of the Compton
radiation has the same order of magnitude as the radiation scattered
coherently.
Scattering by atoms
Let q,(r), . . . , qz(r) be the wave functions of Z atomic electrons: then
pejdv = I%(r)12dv is the probability of finding the jth electron in the
volume dv. If every function qj(r) can be considered independent of the
others, then pa(r) dv = (Cf=, pej) dv is the probability of finding an electron
i (b) r* t
1
A-'I
2
in the volume dv. The Fourier transform of ,d -, r ,) is called the atomic Fig. 3.3. (a) Radial distribution for I s and 2s
electrons of a C atom as defined by Slater
scattering factor and will be denoted by fa. functions. (b) ScatteRing factors for 1s and 2s
Generally the function pa(r) does not have spherical symmetry. In most electrons.
crystallographic applications the deviations from it, for instance because of
covalent bonds, are neglected in first approximation. If we assume that pa is
spherically symmetric and, without loss of generality, that the centre of the
atom is at the origin, we will have
fa(r*) = 6"
ua
sin (2nrr*)
*2, dr=C.tq
j=1
where Ua(r) = 4m2pa(r) is the radial distribution function for the atom. The
pa function is known with considerable accuracy for practically all neutral
atoms and ions: for lighter atoms via Hartree-Fock methods, and for
heavier atoms via the Thomas-Fermi approximation. In Fig. 3.4(a) the fa
functions for some atoms are shown. Each curve reaches its maximum
value, equal to Z, at sin O / A = 0 and decreases with increasing sin OIA.
According to the previous paragraph most of radiation scattered at high
values of sin O / A is due to electrons of inner shells of the electron cloud
(core). Conversely scattering of valence electrons is efficient only at low
sin O / A values. f, can thus be considered the sum of core and valence
electron scattering:
fa =fcore +fvalence.
In Fig. 3.4(b) fCore and fva,e,c, of a nitrogen atom are shown as function of
sin O / A .
As a consequence of eqn (3.14) the intensity of the radiation coherently
scattered from an atom can be obtained by summing the amplitudes relative
-hs to the electrons taken individually:
\,=I /
The Compton radiation scattered from an electron is incoherent with

respect to that scattered from another electron: its intensity is obtained by
summing the individual intensities relative to every single electron:
,--
sin O/n Since fe = 1 for sin B I A = 0 there is no Compton radiation in,the direction of
(4 the primary beam. Nevertheless it is appreciable at high values of sin BIA.
When we consider the diffraction phenomenon from one crystal the
intensity coherently diffracted will be proportional to the square of the
vectorial sum of the amplitudes scattered from the single atoms while the
intensity of the Compton radiation will be once more the sum of the single
intensities. As a consequence of the very high number of atoms which
contribute to diffraction, Compton scattering can generally be ignored: its
presence is detectable as background radiation, easily recognizable in
crystals composed of light atoms.
0 05
(b) The temperature factor
Fig. 3.4. (a) Scattering factors for S, ~ a +0.
, (b)
core and valence scattering for nitrogen atom.
In a crystal structure an atom is bound to others by bond forces of various
types. Their arrangement corresponds to an energy minimum. If the atoms
are disturbed they will tend to return to the positions of minimal energy:
they will oscillate around such positions gaining thermal energy.
The oscillations will modify the electron density function of each atom
and consequently their capacity to scatter. Here we will suppose that the
thermal motion of an atom is independent of that of the others. This is not
completely true since the chemical bonds introduce strong correlations
between the thermal motions of various atoms (see pp. 117-20 and
Appendix 3.B, p. 186).
The time-scale of a scattering experiment is much longer than periods of
thermal vibration of atoms. Therefore the description of thermal motion of
an atom requires only the knowledge of the time-averged distribution of its
position with respect to that of equilibrium. If we suppose that the position
of equilibrium is at the origin, that p ( r r ) is the probability of finding the
centre of one atom at r', and that pa(r - r') is the electron density at r when
the centre of the atom is at r', then we can write
where pat(r) is the electron density corresponding to the thermically

agitated atom. Notice that the rigid body vibration assumption has been
made; i.e., the electron density is assumed to accompany the nucleus during
thermal vibration.
In accordance with Appendix 3.A, p. 181), pat is the convolution of two
functions and its Fourier transform (see eqn (3.A.38)) is
where
q(r*) = 1
S'
.
p ( r r ) exp (2nir* r') dr' (3.17)
the Fourier transform of p(rl), is known as the Debye-Waller factor.

The function p ( r r ) depends on few parameters; it is inversely dependent
on atomic mass and on chemical bond forces, and directly dependent on
temperature. p ( r l ) is in general anisotropic. If assumed isotropic, the
thermal motion of the atom will have spherical symmetry and could be
described by a Gaussian function in any system of reference:
p ( r r )=p ( r l ) - ( ~ J T ) - ~ " u -exp
~ ' ~[-(rt2/2U)] (3.18)
where r r is measured in A and U = (rI2) is the square mean shift of the
atom with respect to the position of equilibrium. The corresponding Fourier
transform is (see eqn (3.A.25))
q(r*) = exp ( - 2 ~ ~ U r *=~exp
) (-8n2u sin2 8/A2)
= exp (-B sin2 8/A2) (3.19)
where
B = 8n2U (A2).
The factor B is usually known in the literature as the atomic temperature
factor.
The dependence of B on the absolute temperature T has been studied by
Debye who obtained a formula valid for materials composed of only one
chemical element. From X-ray diffraction structure analysis it is possible to
conclude schematically that the order of value of fi is in many inorganic
-
crystals between 0.05 and 0.20 A (B lying between 0.20 and 3.16 A2) but
can also reach 0.5 81 (B 20 A2) for some organic crystals. The conse-
quence of this is to make the electron density of the atom more diffuse and
therefore to reduce the capacity for scattering with increasing values of
sin 8/A.
In general an atom will not be free to vibrate equally in all directions. If
we assume that the probability p ( r r ) has a three-dimensional Gaussian
distribution the surfaces of equal probability will be ellipsoids called
vibrational or thermal, centred on the mean position occupied by the atom.
Now eqn (3.19) will be substituted (see Appendix 3.B, pp. 186 and 188)
by the anisotropic temperature factor (3.20) which represents a vibrational
ellipsoid in reciprocal space defined by six parameters UTl, U12, Uz3, UF2,
uT3, ul3:
+
q(r*) = exp [-2n2(UT1x*' Ug2y*2 U3*3~*22UT2x*y* + +
+ 2UT3x*z* + 2Uz3y*z*)]. (3.20)
The six parameters UG (five more than the unique parameter U necessary
to characterize the isotropic thermal motion) define the orientation of the
thermal ellipsoid with respect to the crystallographic axes and the lengths of
the three ellipsoid axes. In order to describe graphically a crystal molecule
and its thermal motion each atom is usually represented by an ellipsoid,

centred on the mean position of the atom, and surrounding the space within
which the atomic displacement falls within the given ellipsoid with a
probability of 0.5 (see Fig. 3.5).
Scattering by a molecule or by a unit cell

Fig. 3.5. Let pj(r) be the electron density of the jth atom when it is thermically
agitated, isolated, and localized at the origin. If the atom is at position q its
electron density will be pj(r - 9). If we neglect the effects of redistribution
of the outer electrons because of chemical bonds, the electron density
relative to an N-atom molecule or to one unit cell containing N atoms is
N
P M ~=
> C pj(r - 5).
j=1
(3.21)
The amplitude of the scattered wave is
FM(r*)= I2
S j=1
N
pj(r - q) exp (2nir* r) dr
= 21
j=l S
pj(Rj) exp [2nir* (q + R,)] d ~ ,
N
=
j=l
J;(r*) exp (2nir* q), .
where J;(r*) is the atomic scattering factor of the jth atom (thermal motion
included; in the previous section indicated by fa,). The fact that in eqn (3.21)
we have neglected the redistribution of the outer electrons leads to
negligible errors for FM(r*), except in case of small r* and for light atoms,
where the number of outer electrons represents a consistent fraction of Z.
p,(r), as defined by (3.21), is the electron density of a promolecule, or,
in other words, of an assembly of spherically averaged free atoms
thermically agitated and superimposed on the molecular geometry. Such a
model is unsatisfactory if one is interested in the deformation of the electron
density consequent to bond formation. In a real molecule the electron
density is generated by superposition of molecular space orbitals V iwith
occupation ni:
Since pmolecule
can be decomposed into atomic fragments, a finite set of
appropriately chosen basis functions can be used to represent each jth
atomic fragment (see Appendix 3.D). Then
where Ap models the effects of bonding and of molecular environment (in

particular, pseudoatoms may become aspherical and carry a net charge).
By Fourier transform of Ap the deformation scattering is obtained:
Since the core deformation scattering is negligible AF practically coin-

cides with deformation scattering of the valence shells.
Diffraction b y a crystal
One three-dimensional infinite lattice can be represented (see Appendix
3.A, p. 174) by the lattice function
where 6 is the Dirac delta function and r,,,,, = ua + vb + wc (with u, v, w

being integers) is the generic lattice vector. Let us suppose that p,(r)
describes the electron density in the unit cell of an infinite three-dimensional
crystal. The electron density function for the whole crystal (see Appendix
3.A, p. 183) is the convolution of the L(r) function with pM(r):
As a consequence of eqns (3.A.35), (3.A.30), and (3.22) the amplitude of

the wave scattered by the whole crystal is
+
where V is the volume of the unit cell and r i = ha* + kb* Ic* is the
generic lattice vector of the reciprocal lattice (see pp. 63-5).
If the scatterer object is non-periodic (atom, molecule, etc.) the ampli-
tude of the scattered wave FM(r*) can be non-zero for any value of r*. On
the contrary, if the scatterer object is periodic (crystal) we observe a
non-zero amplitude only when r* coincides with a reciprocal lattice point:
The function F,(r*) can be represented by means of a pseudo-lattice: each

of its points has the position coinciding with the corresponding point of the
reciprocal lattice but has a specific 'weight' FM(H)/V. For a given node the
diffraction intensity I,, will be' function of the square of its weight.
Let us multiply eqn (3.25) scalarly by a , b, c and introduce the definition
(3.5) of r*: we obtain
The directions s which satisfy eqns (3.26) are called diffraction directions
and relations (3.26) are the Laue conditions.
Finiteness of the crystal may be taken into account by introducing the
form function @(r): @(r) = 1 inside the crystal, @(r) = 0 outside the crystal.
In this case we can write
and, because of eqn (3.A.35), the amplitude of the diffracted wave is
where
D(r*) =
S
O(r) exp (2nir* r) dr = exp (2nir' .r) dr
and 52 is the volume of the crystal. Because of eqn (3.A.40) the relation
(3.27) becomes
+
= ,
1
FM(H) x
53
D (r* - r i ) .
If we compare eqns (3.28) and (3.24) we notice that, going from an

infinite crystal to a finite one, the point-like function corresponding to each
node of the reciprocal lattice is substituted by the distribution function D
which is non-zero in a domain whose form and dimensions depend on the
form and dimensions of the crystal. The distribution D is identical for all
nodes.
For example, let suppose that the crystal is a parallelepiped with faces Al,
A,, A,: then
D(r * -- ,-Ad2 ,-Ad2 ,-A312
+ +
exp [2ni(x * x y *y z *z)] dx dy dz.
-A1/2 -A2/2 -A3/,
If we integrate this function over separate variables, it becomes, in

accordance with Appendix 3.A, p. 174
sin (nAlx*) sin (nA,y *) sin (nA3z*)
D(r*) = (3.29)
nx * JCY * nz *
Each of the factors in eqn (3.29) is studied in Appendix 3.A and shown in
Fig. 3.A.1 (p. 174). We deduce:
1. The maximum value of D(r*) is equal to A,A,A,, i.e. to the volume S2
of the crystal;
2. The width of a principal maximum in a certain direction is inversely
proportional to the dimension of the crystal in that direction. Thus,
because of the finiteness of the crystals each node of the reciprocal lattice
is in practice a spatial domain with dimensions equal to A;'. In Fig. 3.6
some examples of finite lattices with the corresponding reciprocal lattices
are shown.
When we consider the diffraction by a crystal the function FM(H)bears the
name of structure factor of vectorial index H (or indexes h, k, 1 if we make
reference to the components of r;) and it is indicated as:
FH=
j=l
x fi
N
exp (2nirG . q )
where N is the number of atoms in the unit cell. In accordance with

Fig. 3.6. Direct and reciprocal lattices for: (a) a

one-dimensional lattice; (b) a two-dimensional
lattice in the form of a rectangle; (c) a two-
dimensional lattice in the form of a circle; (d) a
cubic crystal in the form of a sphere; (e) a cubic
crystal in the form of a cube; (f) a crystal in the
form of a parallelepiped (from Kitaigorodskii, A.
1. (1951). The theory o f crystal structure analysis,
Consultants Bureau, New York).
p. 64 we write
FH= xf j
N
j=1
exp (2niAxj) = A H iBH + (3.30a)
where
AH= x
N
j=1
f j cos ~ J G H X ~ BH
, =
N
j=1
fi sin 2nHXj. (3.30b)
According to the notation introduced in Chapter 2, we have indicated the

vector as rlT, and the transpose matrix of its components with respect to the
reciprocal coordinates system as H = (hkl). In the same way I;. is the jth
positional vector and the transpose matrix of its components with respect to
the direct coordinates system is xi = [xi yj zj]. In a more explicit form (3.30a)
may be written
imag. axis
Fhkl
=
j=1
xf j exp 2ni(hxj + kyj + lzj).
I
In different notation (see Fig. 3.7)

FH= IFH(exp (iq,) where qH= arctan (BH/A.HI. (3.31) '
qHis the phase of the structure factor FH.

If we want to point out in eqn (3.30a) the effect of thermal agitation of
the atoms we write, in accordance with p. 149 and Appendix 3.B
N real axis
FH= fOj exp (2niHxj - 8 n 2 4sin2 8/A2) Fig. 3.7. F, is represented in the Gauss pla_nefor
j=1 a crystal structure with N = 5. It is cui = 2zHX,.
N
FH= &, exp (2niAXj - 2n2AU?H)
j=l
depending on the type of the thermal motion (isotropic or anisotropic) of

the atoms. f,, is the scattering factor of the jth atom considered at rest. Let
us note explicitly that the value of FH, in modulus and phase, depends on
the atomic positions i.e. on the crystal structure.
Details of the structure factors calculation from a known structural model
are given on pp. 87-8 and Appendix 2.1.
Bragg's l a w
A qualitatively simple method for obtaining the conditions for diffraction
was described in 1912 by W. L. Bragg who considered the diffraction as the
consequence of contemporaneous reflections of the X-ray beam by various
lattice planes belonging to the same family (physically, from the atoms lying
on these planes). Let 8 be (see Fig. 3.8) the angle between the primary
beam and the family of lattice planes with indices h, k, 1 (having no integer
common factor larger than unity). The difference in 'path' between the
+
waves scattered in D and B is equal to AB BC = 2d sin 6. If it is multiple
of A then the two waves combine themselves with maximum positive
interference:
Fig. 3.8. Reflection of X-rays from two lattice
planes belonging to the family H = (h, k, I). dis
the interplanar spacing.
2dHsin 8 = nA,
Since the X-rays penetrate deeply in the crystal a large number of lattice
planes will reflect the primary beam: the reflected waves will interfere
destructively if eqn (3.32) is not verified. Equation (3.32) is the Bragg
equation and the angle for which it is verified is the Bragg angle: for
n = 1, 2, . . . we obtain reflections (or diffraction effects) of first order,
second order, etc., relative to the same family of lattice planes H.
The point of view can be further simplified by observing that the family of
fictitious lattice planes with indices h' = nh, k' = nk, I' = nl has interplanar
spacing d H = j dHIn. NOWeqn (3.32) can be written as
/ 2(dH/n)Sin 6 = 2dH,sin 8 = (3.33)

where h', k', I' are no longer obliged to have only the unitary factor in
common.
In practice, an effect of diffraction of nth order due to a reflection from
lattice planes H can be interpreted as reflection of first order from the family
of fictitious lattice planes H ' = n H.
It is easy to see now that eqn (3.33) is equivalent to eqn (3.25). Indeed, if
I 0 we consider only the moduli of eqn (3.25) we will have, because of eqns
(2.14) and (3.6),
r* = 2 sin 8/A = lld,.
-/ The reflection and t h e limiting spheres

Let us outline (see Fig. 3.9) a sphere of radius l/A in such a way that the
Fig. 3.9. Reflection and limiting spheres. primary beam passes along the diameter 10. Put the origin of the reciprocal
.
lattice at 0. When the vector rtT, is on the surface of the sphere then the
corresponding direct lattice planes will lie parallely to IP and will make an
angle 6 with the primary beam. The relation
OP=r~=l/dH=IOsin6=2sinB/A
holds, which coincides with Bragg's equation. Therefore: the necessary and
sufficient condition for the Bragg equation to be verified for the family of
planes (hkl) is that the lattice point defined by the vector ri lies on the
surface of the sphere called the reflection or Ewald sphere. AP is the
direction of diffracted waves (it makes an angle of 28 with the primary
beam): therefore we can suppose that the crystal is at A.
For X-rays and neutrons A - (0.5-2) A, which is comparable with the
dimensions of the unit cell (-10A): the sphere then has appreciable
curvature with respect to the planes of the reciprocal lattice. If the primary
beam is monochromatic and the crystal casually oriented, no point of the
reciprocal lattice should be in contact with the surface of the Ewald sphere
except the (000) point which represents scattering in the direction of the
primary beam. It will be seen in Chapter 4 that the experimental techniques
aim to bring as many nodes of the reciprocal lattice as possible into contact
-
with the surface of the reflection sphere.
In electron diffraction A 0.05 A: therefore the curvature of the Ewald
sphere is small with respect to the planes of the reciprocal lattice. A very
high number of lattice points can simultaneously be in contact with the
surface of the sphere: for instance, all the points belonging to a plane of the
reciprocal lattice passing through 0.
If r i > 2/A (then dH< A/2) we will not be able to observe the reflection H.
This condition defines the so-called limiting sphere, with centre 0 and
radius 2/A: only the lattice points inside the limiting sphere will be able to
diffract. Vice versa if A > 2a,,,, where a,,, is the largest period of the unit
cell, then the diameter of the Ewald sphere will be smaller than rg,, (the
smallest period of the reciprocal lattice). Under these conditions no node
could intercept the surface of the reflection sphere. That is the reason why
we can never obtain diffraction of visible light (wavelength -5000 A) from
crystals.
The wavelength determines the amount of information available from an
experiment. In ideal conditions the wavelength should be short enough to
leave out of the limiting sphere only the lattice points with diffraction
intensities close to zero due to the decrease of atomic scattering factors.
Symmetry in reciprocal space

There are some relationships among structure factors relative to different
indexes H which are originated either from the physics of diffraction and
from the crystal symmetry. Let us give the relevant rules.
Friedel law
+
In accordance with eqn (3.30) we write FH= A H iBH.Then it will also be:
F-, = AH- iBH and consequently
Q)-H = -Q)n (3.34)
The value of V - H is opposite to the value of rpH.Since the intensities IHand

I-, depend on and IF-^^^ respectively, we have
From that the Friedel law is deduced, according to which the diffraction
intensities associated to the vectors H and -H of the reciprocal space are
equal. Since these intensities appear to be related by a centre of symmetry,
usually, although imperfectly, it is said that the diffraction by itself
introduces a centre of symmetry.
Effects of symmetry operators in the reciprocal space

Let us suppose that the symmetry operator C = (R, T) exists in direct space.
+
Then the points r;. and r,! whose coordinates are related by X,! = RXj T are
symmetry equivalent. We wonder which kind of relationships brings in the
reciprocal space the presence of the operator C. Since
xfi
F"R exp ( ~ J T ~ H=T )
N
;=l
.
exp ( ~ J T ~ H R Xexp
; ) (2niHT)
= x fi
N
j=1
+XT); = FH
exp ~ J T ~ H ( R
we can write
F", = FHexp ( - ~ J T ~ H T ) . (3.35)
Sometimes it is convenient to split eqn (3.35) into two relations:
From (3.36) it is concluded that intensities ZH and I,qRare equal while their
phases are related by eqn (3.37). The most relevant consequences of eqn
(3.35) are described in the following.
Determination of the Laue class

The Laue class of a crystal (see p. 17) may be determined by means of the
qualitative examination of the diffraction intensities.
Let Pl[(x, y, z)] be the space group (symmetry equivalent positions are
shown in square brackets). The only symmetry operator is the identity, and
its use in eqn (3.36) does not give us any useful result. Anyhow, because of
the Friedel law = IFHI holds. If the space group is PI
[(x, y, z), (2, jj, Z)] we can use R2 in eqn (3.36) and consequently obtain
F-, = FH. In this case the Friedel law does not add any additional
observable relationship. In conclusion, the diffraction intensities from
crystals with symmetry P1 and PI will both show a centre of symmetry or, in
other words, they will show the symmetry of the Laue class I.
As a further example let P2 [(x, y, z), (x,y, 2)] be the space group. By
introducing R, in eqn (3.36) and by applying the Friedel law we obtain the
following relationship (symmetry equivalent reflections):
If the space group was Pm [(x, y, z), (x, y, z)], by using R2 and by
applying the Friedel law we would obtain eqn (3.38) again. If the space
group was P2/m eqn (3.38) would be obtained again only by using matrices
R2, R3, and R, in eqn (3.36). This time the Friedel law does not add any
additional relationship to those obtained from eqn (3.36). We can conclude
that the symmetry of the diffraction intensities from crystals belonging to
space groups P2, Pm and P2/m is that of the Laue class 2/m.
The reader will easily verify that the crystals belonging to groups P222,
Pmm2, and P2/m 2/m 2/m show intensity symmetry of the 2/m 2/m 2/m
Laue class:
IFhkrl = I F h d = = IFiid
=IF---l=IF-
hkl hkll - IFhill = IFhkil
and those belonging to space groups P4, ~ 4 P4/m

, show the intensity
symmetry
of the Laue class 411x1.
Determination of reflections with restricted phase values

Let us suppose that for a given set of reflections the relationship HR = -H is
satisfied. If we apply (3.37) to this set we will obtain 2qH= 2nHT + 2nn
from which
qH= XHT + nn. (3.39)
Equation (3.39) restricts the phase qHto two values, XHT or X(HT+ 1).
These reflections are called reflections with restricted phase values, or less
properly, 'centrosymmetric' .
If the space group is centrosymmetric the inversion operator
will exist. In this case every reflection is a restricted phase reflection and will
assume the values XHT or ~ ( H +T 1). If the origin is assumed on the centre
of symmetry then T = 0 and the permitted phase values are 0 and n. Then
according to eqn (3.30b), FHwill be a real positive number for q, equal to
0, and a negative one for qHequal to n. For this reason we usually talk in
centrosymmetric space groups about the sign of the structure factor instead
of about the phase.
In Fig. 3.10 FHis represented in the complex plane for a centrosymmetric
structure of six atoms. Since for each atom at q another symmetry
equivalent atom exists at -q, the contribution of every couple to FHwill
have to be real.
As an example of a non-centrosymmetric space group let us examine
P2A2,, [(x, y, z), (i -x, J , 4 + z), ( i + x, 4 - y, 2), (ii,+ y, - z)] where
the reflections (hkO), (Okl), (h01) satisfy the relation HR = -H for R = R2,
R3, R4 respectively. By introducing T = T2 in eqn (3.39) we obtain
imag. axis
Fig. 3.10. F, is represented in the Gauss plane

for a centrosymmeJric crystal structure with
N = 6. It is q = ZRHX,.
qhko = ( n h l 2 ) + nn. Thus qhkowill have phase 0 or n if h is even, phase

fn12 if h is odd. By introducing T = T3 in eqn (3.39) we obtain
qok[ = (nk12) + n n : i.e. q o k l will have phase 0 or n if k is even, fA12 if k is
odd. In the same way, by introducing T = T, in eqn (3.39) we obtain
qhol= (n112) + n n : i.e. q h 0 i will have phase 0 or n if 1 is even, fn12 if 1 is
odd.
Phase restrictions to ( ~ ~ 154~, ~ 1or 4 )to (3x14, 7 ~ 1 4or
) to ( n / 6 , 7 n / 6 )or
. . . , can be found for some space groups. The allowed phase values depend
Table 3.1. Restricted phase reflections for the 32 crystal classes
Point group Sets of restricted phase reflections

p p p p
1 None
1 All
m (0, k, 0 )
2 (h,O, I )
2/m All
mm2 [ ( h ,k, 0 ) masks (h, 0, O), (0, k, 0)l
222 Three principal zones only
mmm All
4 (h, k,O)
4 (h, k, 0 ) ; (0,0, I )
4/ m All
422 (h, k,O); {h,O, I ) ; { h , h, I )
42m [(h, k, 01, { h , h, 011; [ { h , 0, I ) ,
(O,O, 111
[(h, k, 01, { h , 0,0), { h , h, O)]
All
None
All
{ h, O,fi,O)
{ h , 0, h, I )
All
( h , k,O)
(O,O, 1 )
All
[ { h , h, I ) , { h , h, 0 ) , (O,O, / ) I
[(h, k, 01, { h , h, 0), { h , 0, O)]
(h,k,O); (h,O,I); ( h , h , I )
All
{ h, k, 0 )
All
( { h , k. O), { h ,h, 011
{ h , k,O}; { h , h, I }
All
on the translational component of the symmetry element and on its location

with respect to the cell origin.
A different point of view may also be used: the existence of reflections
with restricted phase is due to the presence of symmetry elements which at
least in projection simulate the centre of inversion. For instance, the
projection of a structure in P2,2,2, along the c axis is centrosymmetric:
correspondingly, the reflections belonging to the zone (hkO), being insensi-
tive to the z coordinate, have restricted phase values. The reader will easily
verify that zones orthogonal to axes 2 , 4 , 4, and 6 (and, of course, to the
corresponding screw axis) have symmetry restricted phases, as well as zones
parallel to the axes 2 , 4, 6. In Table 3.1 the sets of restricted phase
reflections are given for the 32 crystal classes.
Systematic absences
Let us look for the class of reflections for which HR = H and let us apply eqn
(3.35). This relation would be violated for those reflections for which AT is
not an integer number unless IFHI = 0 . From this fact the rule follows:
reflections for which HR = H and AT is not integer will have diffraction
intensity zero or, as usually said, will be systematically absent or extinct. Let
us give a few examples.
+
In the space group P2, [ ( x , y , z ) , ( 3 , y 1, Z ) ] the reflections (OkO) satisfy
the condition HR, = H. If k is odd, HT, is semi-integer. Thus, the reflections
(OkO) with k # 2n are systematically absent.
In the space group P4, [ ( x , y, z ) , (2, P, 1 + z ) , (JJx, +
, +z ) , ( y ,3, 2 z ) ]
only the reflections (001) satisfy the condition HR, = H for j = 2, 3, 4 . Since
HT, = 112, AT3= 114, HT4= 3114, the only condition for systematic absence
is I # 4n, with n integer.
In the space group PC [ ( x , y , z ) , ( x , j , z + +)I the reflections (h01) satisfy
the condition HR, = H. Since HT, = 112 the reflections (h01) with 1 # 2n will
be systematically absent.
Note that the presence of a glide plane imposes conditions for systematic
absences to bidimensional reflections. In particular, glide planes opposite to
a, b, and c impose conditions on classes (Okl), (hol), and (hkO) respec-
tively. The conditions will be h = 2n, k = 2n, 1 = 2n for glide planes of type
a, b, or c respectively. The reader can easily check the data listed in Table
3.2.
Let us apply now the same considerations to the symmetry operators
centring the cell. If the cell is of type A, B, C, I, symmetry operators will
exist whose rotational matrix is always the identity while the translational
matrix is:
respectively. If we use these operators in eqn (3.35) we obtain: the relation

HR = H is satisfied for any reflection; the systematic absences, of three-
dimensional type, are those described in Table 3.2.
A cell of type F is simultaneously A-, B-, and C-centred, so the respective
conditions for systematic absences must be simultaneously valid. Conse-
Table 3.2. Systematic absences
Symmetry elements Set of reflections Conditions
Lattice none
h+k+l=2n
h+ k=2n
k+l=2n
hkl h+l=2n
i
h+ k=2n
k+l=2n
h+l=2n
-h+ k+I=3n
h-k+I=3n
a
Glide-plane 11 ( 0 0 1 ) b hku
n
d
Glide-plane 11 ( 1 0 0 ) b
c Okl
n
d
C
Glide-plane 11 (1 1 0 ) b hhl
n
d
Screw-axis 11 b 2 1 ~ 4 ~
4 1 ~ 4 ~ Oku
quently only the reflections for which h, k, and 1 are all even or all odd will
be present.
The same criteria lead us to establish the conditions for systematic
absences (-h + +
k 1# 3n for obverse setting and h - k 1# 3n for reverse +
setting) for a hexagonal cell with rhombohedra1 lattice.
Rules for systematic absences may be also derived by using the explicit
algebraic form of the structure factor. Suppose, for example, that the space
group contains a c-glide plane perpendicular to the b axis (then ( x , y, z) and
(x, j , z + 4) will be symmetry equivalent points). The structure factor is
then
Nl2
Fhkl=
j=1
f j exp 2ni(hx, + kyj + lz,)
NI2
+ f j exp 2ni[hxj - ky, + l(z, + ;)I.
,=I
For 1 even ~ 1 2
+
Fhkl= 2 x fi exp [2ni(hxj lz,)] cos 2nkyj;
j=1
for 1 odd
Ni2
Fhkl= 2 x fi exp [2ni(hxj
j=l
+ lz,)] sin 2nky,.
It is easily seen that Fhol= 0 for 1 odd, in accordance with our previous
results.
Unequivocal determination of the space group

Determination of the space group is an important first step in any crystal
structure analysis. From the qualitative examination of the diffraction
intensities (which leads to the identification of the Laue class) and from the
analysis of the systematic extinctions 58 space groups are unequivocally
determinable (without making any distinction between enantiomorph
couples). The symbols of these groups are printed in bold character in Table
1.9. As an example, suppose that relations (3.38) and systematic absences
(OkO) for k = odd and (h01) for I = odd are observed. Then the space group
P2Jc is unequivocally deduced since:
(1) the cell has to be primitive (no systematic absences involving three-
dimensional reflections are observed);
(2) the Laue-class is 2/m (from (3.38));
(3) a twofold screw axis along b and a c-glide plane normal to b do exist
according to systematic absences.
Sometimes the identification of space group extinctions is disturbed by the
presence of one or a few reflections seemingly violating one of the
+
extinction rules (e.g. only one (hkO) reflection with h k =odd has a
meaningful non-zero intensity). One way to produce such forbidden
reflections is the Renninger effect, described in Appendix 3.B, p. 191, which
occurs when two or more reciprocal lattice points happen to intersect the
reflection sphere simultaneously. One way to avoid the Renninger effect is
to rotate the crystal about the reflecting plane's normal and repeat
measurements.
Space groups which are not unequivocally determined could, with few
exceptions, be unambiguously identified if their point group is known. The
methods which can give us information about point groups have been
schematically described in Chapter 1 (p. 15) (crystal habit, piezoelectricity,
pyroelectricity, etc.). Let us add to them the anomalous dispersion effects
(see p. 165) and the statistical methods (see Chapter 5, p. 322) based on the
expectation that the distribution of the intensities is influenced by the
presence of symmetry elements and, in particular, it is different for
centrosymmetric and non-centrosymmetric space groups.
Diffraction intensities
The theory so far described is called kinematic: basically it calculates
interference effects between the elementary waves scattered inside the
volume of the crystal. However, it neglects two important phenomena:

when the incident wave propagates inside the crystal its intensity decreases
gradually because a part of its energy is transferred to the scattered beam or
it is absorbed; the diffracted waves interfere with each other and with the
incident beam.
The theory which takes into account all these phenomena and analyses
the wave field set up as a whole is called the dynamic theory of diffraction.
It was initiated by ~wald:[']later on Laue showed that Ewald's theory is
equivalent to analysing the propagation of any electromagnetic field through
a medium having a periodically varying complex dielectric constant. The
description of the dynamical theory is out of the scope of this book. The
reader is referred to specialist books or review articles[2331 for exhaustive
information. Only a dynamical effect of particular importance for crystal
structure analysis, the Renninger effect, will be here described (in Appendix
3.B, p. 191) in any detail. Other dynamical interactions will be described on
the basis of the kinematical theory properly modified by Darwin and other
authors.
Dynamic effects develop gradually in a crystal: it may be shown that for
sufficiently small thicknesses the incident beam is not weakened con-
siderably, the diffracted waves are not yet so strong as to give rise to
remarkable interference effects with the incident beam and the effects of
absorption are negligible. Under these conditions (theoretically, thicknesses
<10-~-10-~cm) the kinematic theory is a fairly accurate approximation to
dynamic theory. However, in practice, corresponding equations proved to
be valid even for crystals having dimensions of several tenths of a
millimetre. This is due to the real crystal structures.
A simplified model of real crystal was proposed by Darwin: [4,51 it can be
ideally schematized like a mosaic of crystalline blocks with dimensions of
about 10-'cm, tilted very slightly to each other for angles of the order of
fractions of one minute of arc: each block is separated by faults and cracks
from other blocks. The interference between the waves only occurs inside
every single block, whose dimensions satisfy the theoretical conditions of
applicability of the kinematic theory. Because of the loss of coherence
between the waves diffracted from different blocks, the diffracted intensity
from the whole crystal is equal to the sum over the intensities diffracted
from every single block.
Real crystals, however, differ by ideal ones also because they may contain
a large variety of defects, which are convenient to classify into the following
groups (see Chapter 9): transient defects, having lifetimes measured in
microseconds (e.g. phonons, which are elastic waves propagating through
the crystal and inducing atomic displacements); point defects, which can be
missing (called vacancies), interstitial, or vicarious atoms; line defects,
extending along straight or curved lines (e.g. dislocations); plane defects,
extending along planes or curved surfaces (e.g. small angle boundaries,
stacking faults); volume defects, extending throughout small volumes in the
crystal (e.g. inclusions, precipitates, voids).
The importance of defects with respect to diffraction intensities depend
on their nature and on their density. For example, a single point defect does
not produce detectable effects on diffraction maxima but a large number of
them, as in the case of order-disorder transitions, strongly affects
diffraction. When agglomerated they can form voids or cracks in a crystal,
or, clustering along certain planes, they form two-dimensional precipitates.

Furthermore, individual dislocations have little effect on diffraction but they
may array themselves to form small-angle boundaries separating two
relatively perfect regions tilted relative to each other by about one minute
of arc (mosaic blocks).
We can therefore expect that any type of defect which disturbs the crystal
periodicity by lattice distortion or substitution or shift of atoms from the
equilibrium position will produce some effect on diffracted intensities:
among others, it will cause a more rapid statistical decrease of diffraction
intensities with sin BIA. In particular, the lattice distortions cause variations
in the unit cell dimensions and therefore modify the form and volume of the
reciprocal lattice points (variations of the spacing dH bring variations in the
modulus r i , variations in the orientation of the planes H cause modifications
in the orientation of r:).
On the basis of these premises it is nonsense to affirm that a crystal is in
the 'exact' Bragg position for a given family of lattice planes. Indeed,
because of the finite size of the crystal, its mosaic structure, defects, and
lattice distortions, etc., each node of the reciprocal lattice will have a finite
volume and will be in contact with the surface of the reflection sphere for a
finite angle interval. In addition the surface of the reflection sphere itself has
in practice to be substituted by a solid domain. Indeed, the incident X-ray
beam does have an inevitable divergence and an imperfect monochromati-
city. As a consequence (see Fig. 3.11) the spherical surface of radius 1/31is
replaced by a family of spherical surfaces whose efficiency relative to
diffraction depends on the distribution of the intensities as a function of the
angle divergence and of the wavelength of the incident beam.
According to the above remarks quantity of practical interest is the
integrated intensity and not the maximum intensity of the diffraction peak.
Experimental arrangements normally used to measure diffraction intensities
change the orientation of the crystal (see Chapter 4) so as to compel
reciprocal lattice points to cross progressively the Ewald sphere while
continuously recording the intensity of the diffracted beam. Thus the total
diffracted energy during a fixed time is measured. Equivalently, the same
total energy may be measured by integrating the diffracted intensity over a
suitable angular range around the ideal Bragg angle. According to eqns
(3.3) and (3.28) the integrated intensity is given by
IH= klk210LPTE (3.41)
where I, is the intensity of the incident beam, k1 = e4/(m2c4) takes into
account the universal constants existing in eqn (3.3). k2 = 3 1 3 S 2 / ~ 2is a
Fig. 3.11. (a) Incident radiation with non-

vanishing divergence. ( b ) Non-monochromatic
incident radiation.
constant for a given diffraction experiment (Q is the volume of the crystal,

V is the volume of the unit cell), P is the polarization factor, defined on
p. 143, T is the transmission factor and depends on the capacity of the
crystal to absorb the X-rays (see Chapter 4, p. 304), L is the Lorentz factor
and depends on the diffraction technique (Chapter 4, p. 301). E is the
extinction coefficient. It depends on the mosaic structure of the crystal and
has two components. The most important one, called secondary extinction,
takes into account the fact that the lattice planes first encountered by the
primary beam will reflect a significant fraction of the primary intensity so
that deeper planes receive less primary radiation. That causes a weakening
of the diffracted intensity, mainly observable for high-intensity reflections at
low sin O/h values in sufficiently perfect crystals. If the mosaic blocks are
misoriented (as they usually are) then they do not diffract together and
shielding of deeper planes is consequently reduced. Secondary extinction is
equivalent to an increase of the linear absorption coefficient: thus it is
negligible for sufficiently small crystals. Reflections affected by secondary
extinction can be recognized in the final stages of the crystal structure
refinement when for some high-intensity reflection JF,,,J < 1 F J . A method
for inclusion of secondary extinction in least-squares methods is recalled in
Chapter 2, p. 97.
The second component of the extinction coefficient, called primary
extinction, takes into account the loss of intensity due to dynamic effects
inside every single block. This phenomenon can be understood intuitively
by means of Fig. 3.12. At the Bragg angle every incident wave can suffer
multiple reflections from different lattice planes: after an odd number of
reflections the direction will be the same as the diffracted beam: after an
even number of reflections the direction will be the same as the primary
beam. Each scattering causes a phase lag of h/4. Thus, the unscattered
radiation having direction So in Fig. 3.12 is joined by doubly scattered
Fig. 3.12. Multiple reflections from a family of
lattice planes.
radiation (with much smaller intensity) with a phase lag of n:consequently
destructive interference will result. The same consideration holds for waves
propagating along the direction of the diffracted beam: the result is that
both primary and diffracted beams are weakened because of dynamical
effects.
A theory describing the mutual transfer of intensity between incident and
diffracted beams was proposed by zachariasen.[(jl If absorption is neglected
the intensity I, of the beam in the incident direction and the intensity I of
the beam in the diffracted direction should be related by:
where t, and t are lengths in the direction of the primary and diffracted
beams respectively, a is the diffracted power per unit distance and intensity.
The equations have to be solved subject to the boundary conditions: I,
should be equal to the intensity of the primary beam when to = 0 and I = 0
when t = 0. The sum of the two equations is zero, which is the condition for
the conservation of energy. Zachariasen's theory has been modified by
other author^:[^-^] the introduction of an extinction correction parameter in
least-squares analysis may or may not have, according to circumstances, an

appreciable influence on the accuracy of the structural parameters included
in the refinement. Indeed prior information on the mosaic structure is not
usually available and therefore the corrections which have to be made are
not easy to calculate a priori. An experimental (often efficient) way to
reduce extinction consists of rapid cooling of the crystal by means of
immersion in liquid air: this reduces the dimensions of the mosaic grains.
Anomalous dispersion
It is well known that electrons are bound to the nucleus by forces which
depend on the atomic field strength and on the quantum state of the
electron. Therefore they have to be considered as oscillators with natural
frequencies. If the frequency of the primary beam is near to some of these
natural frequencies resonance will take place. The scattering under these
conditions is called anomalous and can be analytically expressed by
substitution of the atomic scattering factor fa defined earlier by a complex
quantity
Af' and f" are called the real and imaginary dispersion corrections. In order
to have a simple insight into the problem (a rigorous quantum-mechanical
treatment was carried out by Honl) we recall that the classical differential
equation describing the motion of a particle of mass m and charge e in an
alternating field intensity Eoiexp (iwt) is
where 0 / 2 n is the frequency of the incident wave, wo is the natural angular

frequency of the vibrating particle, g dxldt expresses a damping force
proportional to the velocity. The steady-state solution of the above
equations is
x(t) = Xoexp (iwt)
where
If the displacement x(t) is multiplied by e the polarizability moment [ex(t)]

of each dipole is obtained; the electrical susceptibility of a collection of Z
uncoupled dipoles is then
ZeXo 2e2 1
x = = - m wi - w2 + igo
Eoi
which is a complex function of the frequency of the incident radiation. The
electric field produced by the dipole oscillator at a distance r >>Xohas a
magnitude (we neglect the polarization factor and the phase shift due to the
travelling in r of the scattered wave) which is w2/(rc2) times its dipole
moment:
02ex(t) w2e2 exp (iwt)
Ed = EOdexp (iot) = -= -Eoi
rc2 mrc2 08 - o2 igw' +
If the electron is unrestrained and undamped then g = wo = 0 and
-eL
Ed = (E&h = 7 EOiexp (iot)
mrc
mrc2 Eoiexp [i(ot + n)]

=
which well agrees with eqn (3.1) suggested by Thomson: n is the phase lag
between the scattered and the incident radiation.
Since g << o , when o >> o, the expression of Ed is not very different from
that of a free electron. Therefore Thomson scattering is only applicable
when o >> w,.
We define now the scattering factor for an electron as the ratio
While the imaginary term is always positive, the real term is negative when
o < wo and positive when o > 01,. From the quantum-theory point of view,
the frequency wo coincides with that of a photon with just sufficient energy
to eject the electron from the atom. Such an energy corresponds to the
wavelength A, = 2nc/wo corresponding to the absorption edge. Thus it may
be expected that a remarkable deviation from Thomson scattering will arise
when the primary beam wavelength is close to an absorption edge of the
atom being considered.
An important question is whether Af' and f " vary with diffraction angle.
Existing theoretical treatments suggest changes of some per cent with
sin e/A but no rigorous experimental checks have been made so far:
therefore in most of the routine applications Af' and f " are considered to be
constant.
For most substances at most X-ray wavelengths from conventional
sources dispersion corrections are rather small. Calculated values for CrK,
(A = 2.291 A), CuK, (A = 1.542 A), and MoK, (A = 0.7107 A) are listed in
the International tables for x-ray crystallography, Vol. 111. In some special
cases ordinary X-ray sources can also generate relevant dispersion effects.
For example, holmium has the L3 absorption edge (-1.5368A) very close
to CuK, radiation: in this case the holmium scattering factor is not the same
for K,, and K,, wavelengths. The following dispersion corrections are
cal~ulated:[~~1
CuK,,(A= 1.5406 A): Af = -15.41 3.70
-
fl'-
CuK,,(A = 1.5444 A): Aft = - 14.09 fu 3.72

Furthermore, the holmium L2 absorption edge (211.3905 A) is very close to
the CuKBwavelength (A = 1.3922 A), so giving rise to
Af'= -11.88, f"~8.75.
Fig. 3.13. Anomalous scattering terms A f ' and

P'for: (a) gadolinium near the L3 edge; (b)
samarium near the L, edge.
If synchrotron radiation is used, its intense continuous spectrum may be

chosen to high precision in order to provoke exceptionally large anomalous
scattering. Very large effects have been measured["] for rare-earth elements
in the trivalent state near L3 absorption edges (corresponding wavelengths
are of crystallographic interest because edges span from 2.26A for
lanthanum to 1.34 A for lutetium). In Fig. 3.13 we show the anomalous
scattering terms Af' and f" for gadolinium and samarium near the L3
absorption edge: spectacular effects as large as -30 electrons/atom could be
measured.
Since anomalous dispersion may induce substantial variation of the
diffracted intensities depending on the wavelength used, anomalous scatter-
ing is an important tool for solving crystal structures (see Chapter 8).
Several recent works suggest an important role for multiple-wavelength
methods: the power of such methods depends on the distances between the
working points representing f i n the complex plane.[l21As an example, we
+
plot in Figs. 3.14 (a) and (b) the complex scattering factor Af' if" near the
L3 edge for gadolinium and samarium respectively.
Now we will give only a few elementary ideas about the effects of
anomalous dispersion: we postpone the methodological aspects for crystal
structure analysis until Chapter 8. Let suppose that a non-centrosymmetric
crystal contains N atoms in the unit cell from which P are anomalous
scatterers and the remaining Q = N - P atoms are normal scatterers. Then
(b)
where + and - indicate that the magnitudes are calculated for the vectors
H and - H respectively. The subscripts P and Q indicate that the structure ~ ~ ; ~ ; ~ ~ L ~ ~ ~ ~
factors are calculated only with the contribution of P or Q atoms edge; (b)samarium nearthe ne edge.
imag imag.
axis axis
1 real axis
Fig. 3.15. (a) Relation between F, and F-, when

anomalous dispersion is present; (b) relation
between F, and F!, when anomalous
dispersion is present.
respectively:
Fb+ = x f,!
P
j=1
exp 2niHXj;
The vectors F+ and F- are described in Fig. 3.15(a). In Fig. 3.15(b) F +

and (F-)* are shown: the latter is the complex conjugate of F- and they are
symmetric with respect to the real axis:
The difference AZ = J F +- ~IF-(^

~ is known as the Bijvoet difference[13]
and can be easily calculated by means of Fig. 3.15(b):
from which
AZ = 4 (F'JIF:( cos q.
Furthermore
In general, as we can see, IFHI= I F-,1 is no longer valid, i.e. the Friedel
law is not satisfied in the presence of anomalous dispersion. The value of AZ
depends on the collinearity of F; and F&. If they are collinear then
(F+I= IF-1: but this happens by mere chance. AZ is a maximum when Fp
and FQare approximately at the right angles.
The Friedel law is satisfied if: the structure is centrosymmetric-in this
case IF+[and IF-I are always equal; the reflection is centrosymmetric even
if the structure is non-centrosymmetric; the crystal is constituted of only one
chemical element which is the anomalous scatterer.
As a last observation it should be mentioned that besides X-ray, neutron
and gamma-ray anomalous dispersion are also very useful in crystal
structure analysis. Neutron anomalous dispersion techniques employ
nuclear isotopes with resonances in the range of thermal-neutron energies

(see p. 198 for further details).
Gamma-rays are elastically scattered by the electrons of the atoms in the
crystal. An elastic resonant scattering by the nucleus (Mossbauer effect) also
occurs-the transition involves energies comparable with those employed in
conventional X-ray diffraction. Since both the processes are coherent,
scattering by resonant nuclei (there are no other nuclear scattering
contributions) and by electrons can occur simultaneously and interfere with
each other.
An ideal nucleus for gamma-ray resonance is j7Fe; its 14.4 keV resonance
corresponds to a wavelength of 0.86 A. A widely used experimental set-up
includes a radioactive source which emits the 14.4 keV radiation in the
decay of the parent isotope j7Co. This produces a resonance effect in 57Fe
(this atom may naturally be present in the crystal or implanted by
techniques such as those used for isomorphous replacement) which is
superimposed on the various Bragg reflections upon the gamma radiation
elastically scattered by all the atoms in the crystal, iron atoms included.
Since 5 7 ~ has
e a natural abundance of about 2.2 per cent, isotopic
enrichment techniques must be applied in order to provide a sufficiently
large resonant scattering.
The frequency of the incident radiation may be modified by moving the
radiation source, at low velocity (some mm s-l), towards or away from the
crystal (linear Doppler effect).
The anomalous scattering amplitudes increase dramatically in going from
X-ray to neutron to gamma-ray. Conversely, the intensities of the radiation
sources decrease dramatically. This is the most severe drawback for the use
of gamma rays: relatively large signals are produced by relatively very weak
radiation sources.
The Fourier synthesis and the phase problem

If the structure factors are known in modulus and phase the atomic
positions are unequivocally determinable. Indeed, according to eqn
(3.A.16) the electron density is the inverse Fourier transform of F(r*):
p(r) = F(r*) exp (-2nir* r) dr*
1
=- Fhklexp [-2ni(hx + ky + lz)]. (3.45)
Vh,k,l=-2
x = [x, y, z] are the fractional coordinates of the point defined by the vector
r. The atomic positions will correspond to the maxima of p(r).
If in eqn (3.45) we sum up the contributions of H and -H we will have
FHexp (-2niHX) + F - H exp ( 2 n i ~ x )

= (AH+ iBH)eXp ( - ~ J G ~ H +
X )(AH- iBH)eXp ( ~ G ~ H X )
= 2[AHcos 2 n ~ + x BHsin 2nHXI
from which
x [A,,, cos 2n(hx + ky + l z ) + Bhklsin 2n(hx + ky + lz)] (3.46)

is obtained. The right-hand side of (3.46) is explicitly real and is a sum over
a half of the available reflections.
The mathematical operation represented by the synthesis (3.46) can be
interpreted as the second step in the process of formation of an image in
optics. The first step consists of the scattering of the incident radiation
which gives rise to the diffracted rays with amplitudes FH.In the second step
the diffracted beams are focused by means of lenses, and by interfering with
each other they create the image of the object. There. are no physical
focusing lenses for X-rays but they can be substituted by a mathematical
lens (exactly, by Fourier synthesis (3.46)).
Because of the decrease of the atomic scattering factors the diffraction
intensities (and consequently IFH\)weaken 'on average' with the increase of
sin B/A, and can be considered zero for values above a given (sin B/A),,, =
1/(2dmin).Since the reflections at high values of sin B/A, give the fine details
of the structure (small variations of the atomic coordinates can produce big
changes in high-angle structure factors) the quantity dm, is adopted as a
measure of the natural resolution of the diffraction experiment. dmin
depends on different factors such as: the chemical composition of the crystal
(heavy atoms are good scatterers even at high values of sin B/A), the
chemical stability under the experimental conditions of temperature and
pressure, the radiation used (the resolution improves when we pass from
electrons to X-rays and neutrons), the temperature of the experiment.
Roughly speaking, for X-rays dm,, can reach the limit of 0.5 A in inorganic
crystals, 0.7-1.5 A in organic crystals, and 1.0-3 A in protein crystals.
Because of the limit of natural resolution or of an artificially introduced
limit (for instance in order to save time and calculations) the electron
density function will be affected by errors of termination of series. The
effect can be mathematically evaluated by calculating the function pr(r)
available via the function
@(r*) is the form function: @(r*)= 1 inside the available reflection sphere,
@(r*)= 0 outside this sphere. According to eqn (3.A.35) we have
If r * is the radius of the available reflection sphere, according to

Appendix 3.A, p. 181, T[@(r*)] is a function with a maximum at r = 0 and
the subsidiary maxima of weight decreasing with l/rL. The effect of the
convolution (3.47) is qualitatively represented in Fig. 3.16. In particular,
even if p(r) is positive everywhere, pl(r) can be negative in more or less
extended regions: the atomic peaks can be broad and surrounded by a series
of negative and positive ripples of gradually decreasing amplitude.
The number of reflections used in practice in eqn (3.46) varies from
Fig. 3.16. (a) Electron density p ( r ) ; (b) electron
density obtained via a Fourier synthesis with
several tens or hundreds, for unit cells of small dimensions, to several tens
series termination errors. of thousands for macromolecules.
In order to reveal the atomic positions, pr(r) is sampled upon a

three-dimensional grid whose spacings along each of the unit-cell axes have
to be fixed with some care. If the grid is too coarse the interpolation
between grid points to find the maximum of the electron density may be
uncertain, if the grid is too fine a great deal of computing may be
unnecessary. In absence of symmetry, at a resolution dm,, there are
measurable (only N,/2 independent) reflections and the number of grid

points is Np = v / h 3 where A is the grid spacing (say 0.2-0.4 A in the three
- -
directions). In practical cases N, and N, are rather large: for instance, for
V = 1000 A3, dmin= 0.8 A , A = 0.25 A, we have N, 8180 and Np 64 000.
If symmetry is present the amount of calculation is smaller. The number
of independent reflections to be measured is roughly N,/(tm) where z is the
centring order of the cell and m the multiplicity factor of the Laue class (this
is not strictly exact as the multiplicity factor refers to general reflections of
type (hkl) and may be different and less than m for certain zones of
reflections). Furthermore, it will be sufficient to sample p upon the grid
points lying inside the asymmetric unit for reconstructing the whole content
of the cell.
For instance, let P2/m be the space group with a = 7.8 A, b = 16.2 A and
c = 8.1 A and /3 = 93'. If we divide a and c into 33 and b into 66 intervals
the grid spacing will have a sufficient and almost identical resolution in all
three directions. The number of grid points lying inside the asymmetric unit
(114 of the unit cell) is now 33 x 33 x 17 = 18 513.
Very often the volume of the unit cell is much larger than 103A3
(V > lo6hi3 is not infrequent for macromolecules). Thus even with the use
of high-speed computers, the calculation of p is a fairly arduous task
involving time-consuming procedures. Different algorithms are used to
make calculations faster. The most convenient are the Beevers-Lipson
technique and the fast Fourier transform algorithm by Cooley and Tookey
(see Chapter 2, pp. 88-90, and Appendix 2.1).
Unfortunately, it is not possible to apply eqn (3.47) only on the basis of
information obtained directly from X-ray diffraction. Indeed, according to
eqn (3.41), only the moduli IFHIcan be obtained from diffraction intensities
because the corresponding phase information is lost. This is the so-called
crystallographic phase problem: how to identify the atomic positions
starting only from the moduli IFH(.A general solution to the problem has
not been found, but there are methods we can successfully apply (see
Chapters 5 and 8).
Modulated crystal structures

So far our attention has been devoted to condensed-matter systems with
perfect three-dimensional space group symmetry. In recent years numerous
systems have been found which can be considered as perfect crystals with
periodic distortion (from some basic structure) of the atomic positions
(displacive modulation) and/or of the occupation probability of atoms
(density modulation). The presence of periodic distortions is revealed from

the existence of the so-called satellite reflections, generally weak, which are
regularly distributed around the so-called main reflections. If satellite
reflections can be labelled by rational indices in terms of the reciprocal
lattice L* of the basic structure, then the distortion periods are commensur-
ate with the translation periods of the basic structure and a new lattice Lr*
can be introduced which contains L*. In such a case the main and the
satellite reflections are called substructure and superstructure reflections
respectively (see page 80).
For some crystals the positions of the satellite reflections vary con-
tinuously with respect to L*, depending on the temperature. In these cases
we have to admit that satellite reflections can assume non-rational indices,
or also, that the periodic distortions are incommensurable with the
translation periods of the basic lattice. The result is the so-called
incommensurately modulated structure (IMS).
In order to prove the existence of satellite reflections we briefly examine
the simple case of a one-dimensionally displacively modulated structure
when the displacement vector field is a harmonic function.[76]
In a perfect crystal the jth atom in the unit cell defined by the lattice
+
vector r,, = ua + vb + wc is located at r,,, = ry ru. In a harmonically
modulated crystal it will be
, ry
r U= + ru + g, sin [2nK . (ry + ru) - @,I,
where ry is now the average position of the jth atom, and gj and 4, are its
displacement wave amplitude and phase respectively. K is the modulation
vector, which may be expressed in the reciprocal space by K = k,a* +
+
k2b* k,c*. The contribution of the jth atom to the structure factor is now
given by
fi(r*)C exp 2nir* {ry + ru + g, sin [2nK . (r? + ru) - @,I)

U
=fi(r*) exp (2nir* r ? ) x exp (2nir* r,,)

u
.
x exp {2nir* .g, sin [2nK (r? + ru) - @,I) (3.48)
where fi(r*) is the usual atomic scattering factor. To the last exponential
term in (3.48) the Jacobi expansion
exp (iz sin a ) = x

+m
m=-m
exp (-ima)J-,(z)
may be applied. Jm(see Fig. 3.17) is the Bessel function of the first kind of
order m, satisfying J-,(z) = (-l)"Jm(z). Then (3.48) reduces to
.
fi(r*) exp (2nir* r y ) x exp (2nir* ru)
u
.
x x exp {im[@,- 2nK
m
(r; + r,)]}J-,(2nr* .g,)
=fi(r*) exp (2nir* r y ) x J_,(2nr*
m
.g,)
Fig. 3.17. The Bessel functions J,,(z),n = x exp {im(@,- 2 n K . ry)}x exp [2niru . (r* - mK)].
0,1,2,3. u
Provided the number of cells in the crystal is large enough the sum over u
+
leads to (1/V) 6 ( H - r* + mK) where H = ha* kb* + lc*. Consequently
the reflections occur for r* = H' = H + mK: for m = O we have main
reflections (H' = H), for m # 0 satellites are defined. We also see that four
indices are now needed for the identification of a diffraction effect. The
structure factor of the reflections H' = (h, k, 1, n) may then be written as
N
=
j=l
.
A(H1)exp (2nir* r y ) ~ , ( 2 n ~ 'gj)(-l)m
X exp (-2niK r;) exp (im@j). (3.49).
According to Fig. 3.17 the average intensity of satellite reflections rapidly
decreases with m.
The above formalism may be extended to one-dimensionally density
modulated structures and also to multi-dimensional (harmonic or not)
modulations. The above results also suggest that the reciprocal lattice of an
IMS is aperiodic in the three-dimensional space, and that the symmetry
group of an IMS cannot be a three-dimensional space group. We will show
in Appendix 3.E (main references will also be given) that such a reciprocal
lattice may be transformed into a periodic lattice provided a higher-
dimensional space is taken into consideration. Thus the symmetry in the
three-dimensional space can also be a poor residue of the full symmetry in
the higher-dimensional space.
Appendices
3.A Mathematical background

The Dirac delta function
In a three-dimensional space the Dirac delta function 6(r - ro) has the
following properties
6 = 0 for r f ro, 6 = w for r = r o , 6(r-ro) d r = 1 (3.A.1)
where S indicates the integration space. Thus the delta function corresponds
to an infinitely sharp line of unit weight located at ro. It is easily seen that, if
ro = xoa + yob + zoc, then
6 (r - rO)= 6 (X - x0) 6 (y - yo) 6 (2 - 20). (3.A.2)
6(x -xo) may be considered as the limit of different analytical functions.
For example, as the limit for a + 0 of the Gaussian function
Of particular usefulness will be the relation
where x* is a real variable. It easily seen that (3.A.4) satisfies the properties
Fig. 3.A.l. The function Y = (sin 2 n g x ) / ( n x )is

plotted for g = 1,3. Clearly Y ( - x ) = Y ( x ) .
(3.A.1): indeed its right-hand side may be written as

sin [2ng(x - xO)]
exp [2nix* ( x - xo)]dx * = lim
n(x-xo) '
The function sin [2ng(x - xO)]I[n(x - xO)]takes the maximum value 2g at

x = x o (see Fig. 3.A.1), oscillates with period l / g , and has decreasing
subsidiary maxima with increasing x : the value of its integral from -w to
+ w is unitary for any value of g. Therefore the limit for g+ w of
sin [2ng(x- x O ) ] / [ n (-x xO)]satisfies all the properties of a delta function.
Consequently we can also write:
sin [2ng(x - xo)]
6 ( x - xo) = lim
.
-
g n(x-x0) .
In a three-dimensional space (3.A.4) becomes
d(r - 4 ) = I,*
exp [2nirS ( r - ro)]dr* (3.A.6)
where S* indicates the r* space. Two important properties of the delta

function are:
6 ( r - ro)= 6(ro- r )
(3.A.7)
f ( r ) - ro) =f (r0) - ro).
Indeed, for r # ro, left- and right-hand members of (3.A.7) are both
vanishing, for r = ro both are infinite. From (3.A.7)
is derived. Consequently
The lattice function L

Delta functions can be used to represent lattice functions. For example, in a
one-dimensional space a lattice with period a may be represented by
+m
L(x) = 6 ( x - x,)
,=-o(r
where x, = nu and n is an integer value. L ( x ) vanishes everywhere except at

the points nu. Analogously a three-dimensional lattice defined by unit

vectors a, b, c may be represented by
+ +
where r,,,,, = ua ub wc and u, v, w are integer values.
Accordingly, in a three-dimensional space:
(1) a periodic array of points along the z axis with positions z, = nc may be
represented as
+=
f'~(r>= w ) ~ ( Y I
n=-m
C
- 2,); (3.A.12)
(2) a series of lines in the (x, z) plane, parallel to x and separated by c may
be represented by
(3) a series of planes parallel to the (x, y) plane and separated by c is

represented by
+m
P3(r) = C
,=-m
S(z - z,). (3.A.14)
The Fourier transform

The Fourier transform of the function p(r) is given (for practical reasons we
follow the convention of including 2n in the exponent) by
r
p(r) exp (2nir* r) dr. (3.A.15)
The vector r* may be considered as a vector in 'Fourier transform space',

while we could conventionally say that r is a vector in 'direct space'.
We show now that
p(r) = i* F(r*) exp (-2nir' .r) dr*. (3.A.16)
Because of (3.A.15) the right-hand side of (3.A.16) becomes
p(r') (l* .
exp [ 2 d r * (r' - r)] dr*) dr',
which, in turn, because of (3.A.6), reduces to
h p(r') d(rf - r) dr' = p(r).
Relations (3.A.15) and (3.A.16) may be written as

F(r*) = T[P(~)], (3.A.17)
p(r) = T-'[F(r*)] (3.A.18)
respectively: we will also say that p is the inverse transform of F. Obviously
T-lT[p(r)l= p(r)
but
?T[p(r)l = T[F(r*)I = ~ ( - 4
F(r*) is a complex function: by denoting
A(r*) = 1S
.
p(r) cos (2nir* r) dr
B(r*) = 1S
p(r) sin (2nir* r) dr
then
We calculate now the Fourier transform of p(-r). Since
I, P(-r) exp (2nir' r) dr =

I,
p(r) exp [2ni(-r* . r)] dr = A(r*) - iB(r*),
it may be concluded that
where F-(r*) is the complex conjugate of F(r*).

If p(r) is symmetric with respect to the origin, say p(r) = p(-r), because
of (3.A. 17) and (3.A.21) it will result F(r*) = FV(r*), or, in other words,
F(r*) will be a real function (B(r*) = 0). Vice versa, if F(r*) = F-(r*) then
p(r) is symmetric with respect to the origin.
If p(r) = -p(-r) then F(r*) = -FW(r*),so that F(r*) is pure imaginary
(then A(r*) = 0).
Some examples of Fourier transform

In a one-dimensional space (3.A.15) becomes
Let us consider some examples for p(x).

1. Gaussian function :
Since
+m
exp [-$hu2 + itu] du =

then
T[p(x)] = F(x*) = exp ( - 2 ~ c ~ a ~ x * ~ ) . (3.A.25)
It should be noted that the larger the 'width' of p(x), the smaller is that of
F(x*) (see Fig. 3.A.2(a))
2. Exponential function : p(x) = exp (-g Ix 1). Its Fourier transform is the
Cauchy function (see Fig. 3.A.2(b))
Fig. 3.A.2. Examples of Fourier transform.
3. Rectangular aperture:
p(x) = c for -g < x < g, otherwise p(x) = 0.
Then
sin (2ngx *)
~ ( x *=) cl-y exp (2nix*x) dx = c
nx *
which is plotted in Fig. 3.A.1.
4. Dirac delta function: p(x) = 6(x). Then (see Fig. 3.A.2(c))
+m
F(x*) = 6(x) exp (2nix*x) dx = 1.

- CD
If p(x) = 6(x - a ) then F(x*) = exp (2niax*).
5. One-dimensional finite lattice: pp = C;E-, 6(x - nu), where pp

represents a set of N = 2p + 1 equally spaced delta functions. Then
T[pp(x)] = x
P
n=-p
exp (2ninax*) =
P
n=-p
cos (2nnax*)
- l
2 sin (nax *) .=-,

2 2 cos (2nnax *) sin (nax *)
1
-,sin [n(2n + l)ax*] - sin [n(2n - l)ax*]
-
-
2 sin (nax *) =
1
- {sin [n(2p + l)ax*]- sin [ n ( 2 p - l)ax*]
2 sin (nax *)
+ sin [n(2p - l)ax*]- sin [n(2p - 3)ax*]
+ sin [n(-2p + l)ax*]- sin [n(-2p - l ) a x * ] )
- sin [(2p + 1)nax*I - sin Nnax *
- -
sin nu* sin m x * '
The function
sin N n y
f(y)==
is plotted in Fig. 3.A.3 for N = 6 , 7. When y = h ( h is an integer value)

numerator and denominator of the right-hand member of (3.A.27) both
vanish. Then the value of f ( y ) (as determined from its limit for y +h ) is
equal to N if N is odd; if N is even, then f ( y ) is equal to N if h is even, to
-N if h is odd. Between each pair of main maxima there are N - 2
subsidiary peaks. Each main peak has width equal to 2 / N ( 1 / N is that of the
subsidiary peaks), therefore it becomes sharper as N increases. Furthermore
the ratio between the amplitude of a main peak with respect that of a
subsidiary one increases with N.
What we noted for f ( y ) may be easily applied to (3.A.26), whose
principal maxima are at x* = hla.
6. One-dimensional infinite lattice :
According to 5 above the Fourier transform of p will be
sin Nnax *
F ( x * ) = lim
N-= sin nax* '
The function F ( x * ) will present infinitely sharp lines at x* = h l a of weight
l l a . Indeed
I&
Whichever the value of
+& sin N m x *
sin nax *
E',
1
&*=-lim
a I& +'sin N n y
-
sin n y dye
when N+m the value of the integral is unity.
Fig. 3.A.3. The function f ( y ) = s i n N n y l s i n n y for

N = 6,7.
Consequently we could write
or, in words, the Fourier transform of a one-dimensional lattice with period

a is a one-dimensional lattice with period l l a represented by delta functions
with weight l/a.
7. Three-dimensionalfinite lattice:
Its Fourier transform is
F(r*) = x x x
P1
u=-p1
P2
u=-p2 w=-p3
P3
.
exp (2nir* ru,u,w)
9 . .x
P2 P3
= exp (2niur' a ) exp (2nivr' b). 2 exp (2niwr* .e)

- sin Nina . r* sin N2nb r* sin N3nc. r*
sin n a . r* sin n b r* sin JGC r* '
+ +
where Nl = (2p1 1), N2 = (2p2 I), N3 = (2p3 + 1). In accordance with
point 5, each of the terms in the right-hand side of (3.A.29) is maximum
when r* satisfies
with integer values of h, k, 1. It is easily seen that the solution of the above
three equations is given by
r: = h a * + kb* + lc*,
where
and V = a b A c. The vectors a*, b*, c* are nothing else but the basic
vectors of the reciprocal lattice defined in 3 2.3. When Nl, N2, N3 are
sufficiently large then F(r*) has appreciable values only in the reciprocal
lattice points defined by the triple of integers H = (h, k, 1).
8. Three-dimensional infinite lattice:
Its Fourier transform is the limit of (3.A.29) for Nl, N2, N3 tending to
infinity:
sinNlna . r * sinN2nb . r * sinN3ncar*
F(r*) = lim
N,,N~,N+- sin n a . r* sin n b r*. sin JGC r* '
According to points 6 and 7 F(r*) represents a three-dimensional lattice by
an array of delta functions the weight of which may be calculated by
integrating F(r*) on a domain dV* about a specific lattice point H. Since

dV* = d h a * . dkb* ~ d l c =
* V* d h d k d l
because of the point 6
sin N,nh
~ ( r *d)v * = V* lim
N,.N~.NP= I- ',
'l
sin nh
sin N2nk sin N,nl d l = V * = -1
x dhl
E2
-,, sin nk dkkl_3

E3
sin n~ v
arises. In conclusion, the Fourier transform of a lattice in direct space
(represented by the function L(r)) is the function L(r*)IV:
which represents a lattice again (called the reciprocal lattice) in the Fourier
transform space.
9. Fourier transform of a one-dimensional periodic array of points along
the z axis, as defined by (3.A.12). Then
10. Fourier transform of a lattice plane lying on the plane z = 0 and with
translation constants a and b. Then
p(r) = b(x - nu) 6(y - nb) S(z)
and
Fourier transform of spherically symmmetric functions

If p(r) is spherically symmetric we can represent r and r* in spherical polar
coordinates (r, 0, cp) and (r*, %*, cp*) defined by (see Fig. 3.A.4)
/
,I.
. .
'
x = r sin p, cos 8 y = r sin g, sin 8 z = r cos p,
with r > 0, 0 < cp 6 n, 0 s 8 < 2n. Analogous transformations could be
written for r*. Without loss of generality we can choose z along the r*
direction: then r r* = rr* cos cp. Furthermore, for each point with coordin-
ates (r, 8, cp) another point will exist, equivalent to the first, with
coordinates (r, n + 8, n - cp). The contribution of both the points to the
y*
+
integral (3.A.15) will be exp (2nirr* cos cp) exp [2nirr* cos ( n - p,)] =
2 cos (2nrr * cos 9). Thus (3 .A.15) reduces to
3~ n 2n
F r ) p(r)cos(2nrr*coscp)r2sincpdrdp,d%
Fig. 3.A.4. Polar and Cartesian coordinates.
where r2 sin rp is the Jacobian of the transformation (from Cartesian to polar

coordinates) Integration over rp and 8 gives
F(r*) = 1
m
0
4nr2p(r)
sin 2nrr *
2nrr *
dr = Q U(r) sin2nrr2nrr** dr (3.A.33)
where U(r) = 4nr2p(r) is the radial distribution function. Thus F(r*) is also
spherically symmetric and its value at r* = 0 is given by
F(0) = U(r) dr.

0
As an example, let p(r) be a spherically symmetric function equal to 1 for

r S R and vanishing elsewhere. Its Fourier transform
F(r*) = 6 4s2
sin 2nrr*
2nrr *
2
+
r sin 2nrr* dr = $ n ~ ~ r p ( y )
where rp(y) = 3(sin y - y cos y)Iy3 and y = 2nr*R. The function q ( y ) is

plotted in Fig. 3.A.5. The main maximum occurs at y = 0. Intensities of the
subsidiary maxima decrease when y increases. The distance between Fig.3.A.5. Thefunction p,y ,=B(sin y -
consecutive maxima is inversely related to R. ycos y ) / y 3 is plotted as a function of y = 2nr'R.
Convolutions
The convolution (or folding) of two functions p(r) and g(r) (it will be
denoted by p(r) *g(r)) is defined by the integral
where S is the r space. Note that in (3.A.34) the integrand is a function of

both u and r while the integral is only a function of u. The relation between
p(r) and g(r) is symmetrical in forming their convolution. That may be
proved first by replacing in (3.A.34) r by R = u - r and after R by r: then
Convoluting two functions very often has the effect of 'broadening' the
one by the other. As an example, the convolution of two Gaussian functions
+
N(a,, al) and N(a,, a;?)is the Gaussian function N((a: 0;)ll2, a , a,). +
The convolution operation appears in many scientific areas, and is
involved in the interpretation of most experimental measurements. For
example, when the intensity of a spectral line is measured by scanning it
with a detector having a finite slit as input aperture, or when a beam of light
passes through a ground-glass screen and is broadened out into a diffuse
beam. Suppose in the second example that p(8) is the angular distribution
of the incident beam and g(B) is the angular distribution which could be
obtained if the incident beam was perfectly collimated. For any given p(8)
the angular distribution of the transmitted beam is given by:
That may be explained by observing that the component of the
transmitted beam emerging at angle 0 due to the light component incident
at angle 8 ' (and therefore deviated through the angle 8 - 8') has intensity
p(01)g(8- 0'). If interference effects are absent the total intensity in the
direction 0 is the integral of p(O1)g(8 - 8') over all values of 0'.
A very important theorem for crystallographers is the following:
T [ P ( ~*) g(r)l = T[p(r)l T[g(r)l. (3.A.35)
The left-hand side of (3.A.35) may be written
p(r)g(u - r) exp [2ni(u r*)] dr du (3.A.36)
where r and u vary in S and S" respectively. Denoting u = r + r' gives

T[p(r) * g(r)] = 11
S' S
p(r)g(rt)exp [2ni(r + r') . r*] dr dr'
= I, .
p(r) exp [2ni(r. r*)] d r l g(rt) exp [2ni(rt r)] dr'
S'
In an analogous way the reader will prove that
If g(r) = p(-r) (3. A.34) will represent the autoconvolution of p(r) with
itself inverted with respect to the origin: in crystallography it has a special
significance, the 'Patterson function', and will be denoted by P(u). It is
the transform of which, according to (3.A.3.9, is given by

+
T[P(u)] = [A(r*) iB(r*)][A(r*)- iB(r*)] = (F(r*)I2. (3.A.39)
It is now easily seen that the Fourier transform of an autoconvolution is
always a real function: therefore, in accordance with the conclusions of
p. 176, P(u) will always be centrosymmetric even if p(r) is not.
Convolutions involving delta functions

Because of (3.A.8) and (3.A.34)
If r and u are assumed to belong to the same space, choosing the same
coordinate system transforms the above relation onto
6 (r - ro)* p (r) = p (r - ro). (3.A.40)
We see that the convolution of p(r) with 6(r - r,) is equivalent to a shift
of the origin by r , (see Fig. 3.A.6(a)).
Suppose now that f (x) is a function defined between 0 and a. Because of
(3.A.10)
~(x,Y) 0 0 0
Fig. 3.A.6. Convolutions of the function fwith:
(a)the Dirac 6 ( x - a ) delta function; (b) a one-
w 0 0 0
dimensional lattice. In (c) the convolution of the
function f(x, y ) with a two-dimensional lattice is
shown.
where p(x) is a periodic function defined between -a and +w, equal to

f(x) for 0 s x s a and with period a. Indeed each term in the summation
corresponds to a function equal to f(x) but shifted by na (see Fig.
3.A.6(b)). It may be concluded that each function p(x), periodic with
period a, may be considered as the convolution of the function f (x) = p(x)
defined between 0 and a, with an array of delta functions located at the
lattice point positions.
This result may be generalized to a three-dimensional space. Indeed
where: f (r) is a function defined for 0 s ' x s a, 0 s y s b, 0 s z s c; p(r) is a

function defined for x, y, z between -w and + w , periodic with periods a, b,
c, and equal to f ( r ) when O s x s a , O S y s b , O s z s c .
In conclusion, a three-dimensional periodic function p(r) may be con-
sidered as the convolution of a function defined in an elementary cell with a
three-dimensional lattice, this last represented by the lattice function L.
Some properties of convolutions

Let p(x) and g(x) be two one-dimensional distributions defined in the
interval (-a,+w) and C(u) their convolution. Consider the characteristic
equation of C(u):
+=
C(U)exp (itu) du =
+m +m
L (L
p(x)g(u - x) exp (itu) dx du.
)
Table 3.A.1. Properties of distributions under the

convolution operation
Changing the variable u into q = u - x gives

+rn +"
C(u) exp (iru) du = p(x) exp (itx) dxl-m g(q) exp (itq) dq
(3.A.42)
which implies that the characteristic function of a distribution obtained by
convolution is equal to the product of the characteristic functions of the
constituent distributions. From (3.A.42) several properties of great import
follow.
Setting t = 0 in (3.A.42) gives
where A represents the area under the distribution.

Taking the derivative of (3.A.42) with respect to t and setting t = 0 yields
Thus, the mean of the convolution is equal to the sum of the means of the
constituent distributions.
By extending the procedure Table 3.A.1 may be obtained. The following
notation has been used:
are the central moments of order p for the convolution distribution (similar
expression can be derived for the constituent distributions). Accordingly
po= 1, 1-11 = 0, p, coincides with the variance a', while yl = p3/a3 and
y,= [(p4/a4)- 31 are the skewness and the excess parameters for any
distribution.
Deconvolution of spectra
Often it occurs that an experimentally measured function C may be
considered as the convolution of the functions p and g. If p is known in
advance then it may be of some interest to obtain g. That frequently occurs
in spectroscopy or in powder diffraction, where a spectrum is often
constituted by overlapping peaks and it is wanted to deconvolute from such
a spectrum a given lineshape function. Effects of such self-deconvolution
are:['"16] the component lines are more clearly distinguished, their location,
area, etc. are more correctly defined, the signal to noise ratio is increased.
Let us consider C as the convolution of the lineshape function p and of
the ideal spectrum g. Then p may be decon~oluted[l~~ from C by taking the
Fourier transform of
and by calculating
T k l = T[C]/Tbl. (3.A.43)
g is finally obtained as inverse Fourier transform of the right-hand side of
(3.A.43):
An example of what can be achieved is shown in Fig. 3.A.7. In (a) a

diffractogram of quartz measured between 67" and 69" is shown, in (b) a
possible Fourier self-deconvoluted spectrum is drawn. The widths of
deconvoluted lines are narrower, peak centroids shift as a result of changes
in degree of overlap: since integrated intensities of peaks are retained (see
p. 184) more careful estimates of integrated intensities of single peaks may
now be accomplished.
The practical use of (3.A.44) is not always straightforward.[183191 The
limited instrumental resolution and the random errors 'associated with
experimental data are a difficult problem. In particular, random errors are
usually amplified by the process so that the right-hand side of (3.A.44)
could coincide with a function g' substantially different from g. Filtering
operations or supplementary considerations are then introduced to reduce 28
spurious features. Fig. 3.A.7.
3.B Scattering and related topics

Compton scattering
A photon of energy hv (v = CIA) moving along the direction so is scattered
by a completely free electron located in 0 , initially at rest. Let s be the
+
direction of the scattered photon, 28 the scattering angle, hv' = hc/(A dA)
its energy and mv2/2 the recoil electron energy. Conservation of energy
(neglecting any relativistic effect) requires
which may be approximated to

hc
- dA = $mu2
A2
Conservation of momentum requires
which, on assuming A + dA = A, may be reduced to

h
imv = - sin 8. (3.B.3)
A
Finally, v may be eliminated from (3.B.1) and (3.B.3) so as to obtain
The anisotropic temperature factor

Thermal agitation will cause the atoms to fluctuate about their equilibrium
positions. Let us suppose that the force field around an atom located at the
origin of the coordinate system (its equilibrium position) is anisotropic and
harmonic. Then the mean-square vibrational amplitude is different for
different directions and equal in the two ways of the same direction. In any
rectilinear coordinate system defined by A = [a, b, c] let p ( r l ) be the
probability of finding the atomic centre moved in r t = x t a + y l b + z'c =
X'A. In general
p(xl, y', z t ) = ( 2 ~ r ) - ~ / ~ (U)-1'2
d e t exp [-~(X'U-'x')], (3.B.5)
where U is the variance-covariance matrix (it represents a symmetrical
tensor)
u=
i
(x'" (xtyt) (xtzt)
1
(x'y') ( y f 2 ) (y'z') = ( x t X ' ) .
(xtzt) (ytz') (zt2)
The Fourier transform of p(X1) gives
+
q(X*) = exp (-2n2X*u*x*) = exp [ - ~ ? G ~ ( U T ~2U&.x*y*
X*~ + . . .)I
= exp (-X*/3X*) = exp - ( p , , ~ +
*~2P12x*y*+ . . .)I
where U* is the variance-covariance matrix expressed in reciprocal coordi-
nates and
Usually crystallographic least-squares calculations (see p. 94) provide estim-

ates of the tensor fi from which U* is easily derived via eqn (3.8.6). Then
the atomic scattering factor, calculated for r* = r & and corrected for
thermal motion is (see p. 153)
fo exp (-HPH) =fo exp [-2n2(Au*H)]
where fo refers to the atom at rest.
In order to obtain U from U* we only need to apply eqn (2.28). We
obtain U = GU*G from which
In accordance with a well known property of tensors, the mean-squares

vibrational amplitude (u2) in the direction defined by the unit vector
n = NA = N*A*, is given by
(u2)(,) = NUN = N*u*N*. (3.B.8)
Let us now express n as a ratio between a vector r = XA = X*A* parallel

to n and its modulus: n = rlr. Then eqn (3.B.8) may be written as
or also
From (3.B.9) the following relations arise:
The following exercises may be useful to clarify some aspects of the

subject:
Calculate the Pijs corresponding to the isotropic temperature factor Biso
defined by (3.19).
If in (3.19) sin2 8/A2= r*'/4 = 1/(4dh) is replaced by (2.17b) then
$ = BisoG* /4
is obtained.
If $ is the anisotropic tensor in the coordinate system A, calculate its
expression in A' = MA.
According to (2.A. 15) $' = (M)-~$M-,.
If $ is the thermal tensor of a given atom, calculate the tensor $' of the
atom referred to the first one by the symmetry operation C = (R, T).
In accordance with (2.A. 16) $' = R$R.
Calculate the principal axes of the ellipsoid describing the anisotropic
temperature factor $ and their orientations with respect to the crystal-
lographic axes.
See p. 76.
Calculate the isotropic temperature factor equivalent[201 to a given
anisotropic temperature factor (often the anisotropic parameters are
deposited and equivalent isotropic are published). The equivalent
isotropic motion is defined as that one which gives rise to the same value
of (u2): in our case ( u ~= () u ' ~) ~ ~~~ . ~ ~ ~
Let us suppose that eigenvalues A,, A,, and A3 of the anisotropic tensor
$ have been calculated. Then, according to (2.38b) we can write
1
(u2)aniso= $(A1 + A2 + A3) (3.B.lOa)
while, for an isotropic motion (see eqn (3.19))

Isotropic and anisotropic motion will be then equivalent if
Let us now observe that, under a transformation of axes, the matrix

$G transforms, because of eqns (2.E.8) and (2.21), according to
in other words, PG transforms in the same way as a rotation matrix of a

symmetry operator (see eqn (2.E.1)). Therefore the trace of $G is
invariant under the transformation and, according to p. 75, it is
+
constantly equal to A, + A, A,.
At the end we can replace eqn (3.B.lOb) by the most simple equation
1 1
(u2)equiv = 2 Tr (PO) =
6~ i,j
The reader will easily derive from (3.B.lOd) specific formulae for
specific crystallographic systems: e.g. for cubic, tetragonal, and ortho-
rhombic systems
(u2)equiv= (Plla2+ P22b2 + P 3 3 ~ ~ ) 1 ( 6 ~ d ~ ) ,
for hexagonal and trigonal systems (hexagonal setting)
etc. Readers are referred to Schomaker and ~ a r s h [ ~for

l ] the estimated
standard deviation of (u2)
In reports of crystal structure determination thermal motion is usually
presented in direct space by means of a probability ellipsoid. Each ellipsoid
is centred on the mean position of the atom: the surface has constant
probability density obtained, in accordance with eqn (3.B.9, on assuming
jpu-lx'= ~2
where C is a chosen constant. The probability P that a thermal displacement
falls within the ellipsoids may be calculated as
P(c)=~ p(X')dXr.
ellipsoid
The most often used value is C = 1.5282: then[''] the ellipsoid encloses 50
per cent of the trivariate Gaussian probability density.
Symmetry restrictions on the anisotropic temperature factors

Coefficients Pii as provided by crystallographic least-squares procedures (see
p. 94) are affected by various error sources: for example, experimental
errors on diffraction intensity measurements, inadequacy of the atomic
scattering model, wrong or unapplied absorption correction (see Chapter 4),
existence of non-harmonic force fields around the atoms, etc. Thus the
thermal parameters estimated by least squares could be physically unrealis-
tic; for example the variance-covariance matrix U should not result
positive-definite. Necessary and sufficient conditions for positiveness are
(det U) > 0; Uii > 0; and UiiUi,> 0 for i, j = 1, 2, 3.
If the above conditions are violated then atomic vibrations could be

described by imaginary ellipsoids or by paraboloids or by hyperboloids, with
the obvious meaning that experimental data are not of sufficient accuracy to
justify the use of anisotropic temperature factors.
If an atom lies on a special site of a space group the components of the
thermal tensor are restricted according to the site symmetry. Indeed the
thermal ellipsoids must remain invariant after application of the n symmetry
operations which leave invariant the site. In accordance with point 3 in the
previous section the relations
arise, from which the $ restrictions can be derived. Such restrictions can
also be obtained in an easier way by using Wigner's theorem according to
which the symmetry of the $ restrictions is displayed by the matrix
Since no atomic position is left invariant after the application of glide

planes or screw axes, only point symmetries have to be taken into account.
Furthermore, ellipsoids are centrosymmetric, thus no additional $
restriction arises if an inversion centre is added to the site symmetry, or if a
mirror plane is replaced by a twofold axis perpendicular to the plane. Thus
it is not difficult to understand that ellipsoid symmetry is one of the
following three cases:
(1) spherical symmetry, for atomic sites with symmetry 23 and its
supergroups;
(2) ellipsoid of revolution, for site symmetries 3, 4, 4, and 'non-cubic'
supergroups;
(3) general ellipsoid. For site symmetry 222 and its orthorhombic super-
groups the orientations of the ellipsoid axes is fixed by symmetry; for
site symmetry 2, m, or 2/m only the orientation of one ellipsoid axis is
fixed by symmetry.
In Table 3.B.1 all the conventional special positions are collected in seven
b l o ~ k s . [ ~Each
~ , ~block
~ ] shows the possible point groups to which each site
may belong, the 'minimal' site symmetry characterizing each position and a
cross-reference number identifying the type of $ restrictions described in
Table 3.B.2. The letter h, when present, warns that an hexagonal setting is
involved and the letter w suggests that the rule holds for hexagonal and
non-hexagonal settings. To make Table 3.B.1 clearer we note that all the
special positions in the first block belong to one of the point groups m3m,
43m, 432, m3, 23. According to previous observations, we now neglect
whether symmetry axes are proper or improper; thus the 'minimal'
symmetry 311111 210011 is associated with the sites. Conversely, under the same
convention, all sites with symmetry 2[,,] 211001i or 2[rn11Z[lool or 2[ooll2 k 0 ~ ~ l
or ?[,ll 211001 or 2[,,] 2[loo1are all represented in block five of Table 3.B.1 by
the site with 'minimal' symmetry 2[,,] 2[1001
To make Table 3.B.2 clearer we observe that case No. 23 may be
rewritten as:
P11, P33 = P22' P121 P13 = -P121 P23.
Table 3.B.1. Site symmetry table giving key for the 29 types
of symmetry p-restrictions
m3m, 43m, 432, rn3,23 2/m, m, 2

3[1111 2 [ ~ ~ 1 1 l W2001] I6
0101 17
6/mmm, 6m2, 6mm, 622, 6/m, 1001 18
6, 6, 3m, 32, 3, 3 W 0 1 19
h 310011 9 w 21iio1 20
101, 21
4/mmrn, 42m, 4mm, 422 I 22
4ff-r~4,4 20111 23
4l0011 2 2 0 24
h 1001 25
4[~1~1 3 Il 2210] 26
h 1201 27
411~~1 4 tl 0101 28
3rn, 3m, 32, 3, 3
%iil 5
3[iii1 6 i, 1
3[1i11 7 1 29
3[iii1 8
rnrnrn, rnm2, 222
2[00,1 2[1001 10
w 2[001, 2[1101 11
210101 2[1011 12
2~001 ~[OIII 13
h 210011211001 14
h ~ [ O O ~210101
I 15
Table 3.B.2. Symmetry fi-restrictions.
Cross- Pi1 P22 P33 Pi2 Pi3 P23 Cross- Pi1 Pz, P33 Pi2 Pi3 P2,
refer- refer-
ence ence
No. No.
1 A A A O 0 0 16 A B C D 0 0
2 A A C O 0 0 17 A B C O E 0
3 A B A O 0 0 18 A B C O 0 F
4 A B B O 0 0 19 A A C D E -E
5 A A A D D D 20 A A C D E E
6 A A A D - D - D 21 A B A D E -D
7 A A A D -D D 22 A B A D E D
8 A A A D D -D 2 3 A B B D -D F
9 A A C A / 2 0 0 24 A B B D D F
10 A B C O 0 0 25 A B C B/2 F/2 F
11 A A C D 0 0 26 A B C A/2 0 F
12 A B A O E 0 27 A B C B/2 E 0
13 A B B O 0 F 28 A B C A/2 E E/2
14 A B C B/2 0 0 29 A B C D E F
15 A B C A/2 0 0
The Renninger effect and experimental phase determination

by means of multiple diffraction experiments
Let us suppose that the two nodes HI and Hz are simultaneously at the
surface of the reflection sphere: according to eqn (3.25) it will have to be
By subtracting the second equation from the first we obtain

(sl - s2)lA = r& - r& = r&-H2. (3.B.12)
Equation (3.B.12) sets up that the diffracted beam Hz can act as an
incident beam for the reflection H1 - HZ which is thus in the diffraction
position (see Fig. 3.B.l(a)). Therefore, to the beam diffracted in direction
sl by lattice planes H1 a twice-reflected beam overlaps, first from the planes
Hz and then from the planes HI - Hz (see Fig. 3.B.l(b)). In the same way
the beam diffracted in direction s2 from the planes Hz will overlap with a
twice-reflected beam, first from the planes H1 and then from the planes
Hz - HI.
In general, when diffraction data are collected for crystal structure
solution, the multiple diffraction (Renninger) effect can be considered as a
disturbance. As already noted, it can be the reason for apparent violation of
the systematic extinctions (but the shape of their diffraction peaks is
different from that of the normal reflections and therefore easily recog-
nizable). On the other hand, the three-beam diffraction effect can be treated
in accordance with dynamic diffraction theory and used for the experimental
solution of the phase problem. This point of view is extremely promising,
and a short account is given here.
First ~ i p s c o m b [investigated
~~] the possibility of using multiple diffraction
intensities for phase determination, but conclusive results were not ob-
tained. In spite of many subsequent investigations, only in more recent
years[2s301has it become clear that dynamical theory applied to three-beam
diffraction is a powerful tool for the experimental solution of the phase
problem even for common (say imperfect) crystals. General solutions for
the three-beam case are now available which provide a satisfactory
interpretation of experimental results. The reader is referred to a recent
book[311for a review of the various contributions and a description of the
Fig. 3.8.1. Renninger effect described in: (a)
reciprocal space; (b)direct space.
Detector
I - scan
Fig. 3.8.2. *-scan experiment described in: (a)

direct space; (b) reciprocal space. (b) Ewald sphere
theoretical background: here a short account of the results is given in

accordance with Hummer and ill^.[^^]
The interference between the wave directly diffracted from H1 planes
(primary reflection) and the 'Renninger Umweg wave' generated from Hz
and HI - H, scatterings produces a wave whose intensity depends on the
triplet invariant phase
Without anomalous dispersion effects it is equal to
Such an intensity can be measured by a q-scan experiment (see Fig. 3.B.2):

the integrated intensity IH,(3)is monitored while the crystal is rotated about
the vector r:, and scanned through the three-beam position. The integrated
3-scan profile may be approximately calculated (except for a range very
close to the three-beam setting) according to (see Fig. 3.B.3)
where the Ds are the amplitudes of the excited wavefields in the crystal (the
intensities are given by 1= 1 D 12). In eqn (3.B. 13) DL, is the two-beam
amplitude of the reflection HI, D, is the amplitude of the Umweg wave
defined by
Du = R W ) F H ~ F H , - H ~ .
Ciare suitable parameters.
lR(+)l and A(3) are the modulus and phase of the complex resonance[331
term R ( q ) = )R(q)l exp (iA(+)] which governs the amplitude and the
resonance phase shift of the Umweg wave. Typical resonance curves for
IR(q)1 and A(q) are given in Fig. 3.B.4. IR(q)I is highest near the
three-beam position. A(q) varies from zero to n when 3 is scanned through
the three-beam position: it is less than n/2 when H, is inside the Ewald
sphere; it is between n/2 and n when it is outside.
Equation (3.B.13) and (3.B.14) may be used to interpret the main
features of typical q-scan profiles (we assume that r;, crosses the Ewald
Fig. 3.8.3. Diagram of the interference between
the unperturbed two-beam amplitude Do and
the Umweg resonance term D,. -
Suppose Q, 0: for
. < 0) to outside ( 3 > 0)).
sphere from inside (. 3
6
<<O it is @j+ A($) = 0, interfering waves are
essentially in phase, but R ( 3 ) = 0. Then IH1(3)=I&.When we get near the

three-beam position ID,(qj)l rapidly increases while A(3) approaches 7612:
then the intensity ZHl(3)>GI. Just after scanning through the three-beam
+
position IR(q)( is still large but cos ( a , A 3 ) is negative: then ZH1(3)drops
below the two-beam value G I . For larger values of 3 ZH1(3)will approach
I$. The type of asymmetry is reversed for triplet phase a3= 180".
-
It is easy now to foresee that for triplet phases with a, 7612 the q-scan
profile will show a nearly symmetrical decrease of the intensity about I$ 0, -
- -
+
-
where cos ( a , A(3)) = -1. If a, = -n/2 an increase in intensity will
+
occur about 3 0, where cos ( a , A(3)) 1. According to this method,
in - - 0 out
- -
expected profiles for a, +45" (or Q3= -45') should present characteris-
tics between those for a, = 0 and for Q3= 90" (or a3 -90"). Ideal q-scan
profiles for a, 0, n, f7612, kn14, have been recently secured[341and are
v
shown in Fig. 3.B.5. Profiles for other quadrants can be obtained by

observing that, according to (3.B.14), IHl(v) for a given a3is equal to
IH1(-3) for the phase (180 - Q3). Therefore 3-scan profiles for Q3 and
(180 - a , ) are related by a mirror line through 3 = 0.
Therefore one is able to define ideal profiles that satisfy the condition
AI(3) = Gl
where AI(3) is defined as
AI(3) = m i 1 ( 3 ) + G1(-3)). in -- 0 - out
I & ( 3 ) and I&(+) are the q-scan profiles for the positive ( + a 3 ) and the
v
Fig. 3.8.4. Schematic drawings for the
negative triplet phase (-a3) respectively. Thus the ideal profiles shown in amplitude of the resonance term Rand for its
Fig. 3.B.5 are marked by the condition AZ(tp)lG1= 1. phase factor in function of tp. tp is assumed to be
zero at the ideal three-beam position.
Ideal 3-scan profiles can only be obtained if the dominant process in he
three-beam interaction is due to the interference effect. That occurs when
IFH21and IFHl-H21 are about twice as strong as IFHl/.If IFH2[ is small and
IFHZ-HII is large enough (or vice versa as the influence of (FH21
and IFH2-H11is
symmetric) then intensity is removed from the H1 reflection and coupled
into the H, reflection via the coupling Hz- HI. This loss of intensity is not
compensated by the scattering power from H2 into H1 reflection. The result
is the so-called aufhellung effect, that is a strong depletion of the HI
intensity. If IFH,Iis very small and IFH,(is large then the intensity of HI is
increased at the cost of the H2 reflection intensity. This effect is called
umweganregung by ~ e n n i n g e r . [ ~ ~ ]
Unweganregung and aufhellung effects can be evaluated[361by comparing
the 3-scan profiles for 0, and - a 3 . Then profiles can be separated into two
parts: the symmetric AI curve, which represents the phase-independent
umweganregung or aufhellung effect; the ideal 3-scan profiles, which
contain the phase information.
Some concluding remarks may be useful:
1. Exact 3-scan can be difficult for conventional four-circle
diffra~tometers.[~~]A special six-circle d i f f r a ~ t o m e t e rproved
~ ~ ~ ] to be useful:
two circles (8, Y) for the detector and four circles for the crystal motion (see
Fig. 3.B.6). When the vector H1 is aligned into the T) axis, the 3 scan is
performed by rotating only about the axis: the detector circles (8, Y) may
be moved to observe the reflections H1or H2.
2. Since dynamical three-beam interaction has a very small angular
range, both the divergence and the spectral width of the primary beam
should be small enough. Synchrotron radiation seems therefore the most
suitable source for q-scan experiments. But also properly modified X-ray
equipments based on rotating-anode generators can play an important role.
3. Measuring with the necessary statistical accuracy a single 3-scan
profile is a time-consuming process (say, half an hour when synchrotron
radiation is used, some hours for less intense sources). In order to avoid the
influence of long-range intensity variations, the q-scan profile is usually
obtained as the sum of many fast scans.
4. Finding three beam points is not very easy for large structures (more
than three reciprocal lattice points can simultaneously lie on the Ewald
sphere). It is safe to calculate in advance the most suitable 11, angles for each
reflection HI and choose for measurements three-beam cases which are
separated from all the others by an angular distance greater than 0.1".
Fig. 3.8.6. Non-conventional six-circle
5. As a rule, more reliable phase estimates can be obtained if FH,, FH2, diffractometer,
FH1-H2 have comparable magnitudes. If polarized radiation is used, the
polarization factor may be exploited to attain such a rule.
6. The absolute configuration of a non-centrosymmetric structure may be
determined by accurately measuring one (or more) triplet phase, having a
value near fnI2.
Electron diffraction
The diffraction of electrons (e-diffraction) was demonstrated by Davisson
and Germer in 1927. The electron beam is produced in an electron gun by a
'hair-pin' filament with diameter of about 10 pm, or in a heated pointed
filament of 1-2 pm size (see Chapter 9). Electromagnetic lenses restrict
divergence to or lop4 rad, but also divergence of 10-6rad may be
achieved for special purposes. Electrons are accelerated through a potential
difference of V volts. The following relation is valid:
Two energetical intervals are commonly used: we will talk about high-
energy diffraction (HEED) when V -50-120kV (with the advent of
high-voltage electron microscopes this range needs to be extended to
1MeV) and A =0.05 A, and low-energy diffraction (LEED) when V = 10-
300V and A=4-1 A. Electrons are strongly absorbed by matter and
therefore e-diffraction in transmission is applicable to very thin layers of
matter (10p7-10-5 cm). The scattering is caused by the interaction of the
electrons with the electrostatic field q(r) of the atoms. q(r) is the sum of
the field caused by the nucleus and the field caused by the electron cloud.
Thus the interaction of electrons with matter may be divided into three
processes: (a) no interaction-the electron passes straight through the
specimen; (b) elastic scattering-the electrons are scattered by the Coulom-
bic potential due to the nucleus. Since the proton mass is much larger than
that of electron, no loss of energy occurs: such a scattering is the most
important in electron microscopy. (c) Inelastic scattering-electrons of the
primary beam interact with atomic electrons, and are scattered having
suffered a loss of energy. In a microscope such electrons are focused at
different positions and produce an effect called chromatic aberration

(causing blurring of the image).
q(r) is related to the electron density by Poisson's equation
V"@) = -4nbn(r) - p e w )
where p,(r) is the charge density due to the atomic nucleus and p, is the
electron density function (as defined for X-ray scattering). While the
nucleus with the charge Ze can be considered as a point, the electrons
occupy a finite volume: therefore, as for X-rays (see p. 146), the e-scattering
will have a geometric component which will take into account the
distribution of the electrons around the nucleus. By supposing a spherical
symmetry a simple relation may be obtained between the atomic scattering
factor for X-rays, f,,, and that for electrons, fa,. Let us introduce into
Poisson's equation the inverse Fourier transform
v2(] fae(r*)exp (-2nir* . r) dr*)
I
= 4n fax(r*)exp (-2niP .r) dr*
I
- 4n Z exp (-2nir*
Since the left-hand side is equal to

.r) dr*
I I -2nir*12fae(r*) exp (-2nir' r) dr*,
we obtain the Mott formula

fae(r*)= n-l(Z -fax(r*))lr*'.
+
For sin 8 / A 0 the value of fa, is indeterminate, but a sensible value may be
obtained by the boundary condition
It is still preferred to tabulate (see International tables for x-ray

crystallography, Vols. 3 and 4)
where fa, is in electrons and f:, in (fte occurs in the first Born
approximation for electron scattering by atoms).
If we compare numerically eqn (3.B.15) with the atomic scattering factor
for the X-rays, we observe that:
1. e-scattering is much more efficient than X-scattering (see Fig. 3.B.7).
Consequently diffraction effects are easily detected even from volumes
much smaller than those required for X-rays (in practice, starting from
thicknesses of 100 A for simple structures constituted of heavy atoms).
sin @/A
Fig. 3.8.7. Typical scattering curves for: ( I ) 2. The curves fa, are less sensitive (see Table 3.B.3) to the atomic number
electrons; (2) X-rays; (3) neutrons. Z than fa, (on the average fae(0) ~ 2 " ~ Therefore
). the positions of the
Table 3.B.3. Some atomic scattering amplitudes for electrons

at sin 811 = 0 based on the rest mass of the electron (for
electrons of velocity v, multiply fa, by [ I - ( v / c ) ~ ] - " ~and
)
nuclear scattering for neutrons for specific nuclei
Element fte(0) (lo-' cm) Specific b (lo-'' cm)
light atoms in presence of heavy atoms (e.g. hydrogen atoms in an

organic compound) may be more easily determined by e-diffraction.
3. As for X-rays, the atomic positions can be fixed by means of the
three-dimensional Fourier synthesis whose maxima correspond to the
positions of the atomic nuclei.
4. The electrons can exist in two spin states and therefore can be polarized.
Usually internal fields of crystal structures are not strong enough to do
that.
In HEED techniques diffraction angles are very small. There are a lot of
reflections having 20 between 0" and 2". Under these conditions sin 8 in the
Bragg law can be approximated by 0. Besides, the radius of the Ewald
sphere is very large and its surface is in practice reduced to a plane: thus, all
the nodes of a reciprocal plane can simultaneously be in Bragg position.
Tens or hundreds diffracted beams can be simultaneously observed on a
fluorescent screen, or collected on a photographic plate for measuring the
intensities, or transformed into images by means of adequate electronic
optics.
With A = 0.05 A, one might think it would be easy to obtain resolved
images of the atoms (in some cases they have been effectively obtained).
The aberrations of magnetic lenses, particularly the spherical aberration are
the principal obstacle: electrons widely scattered are focused at positions
different from those to which are focused electrons travelling close to the
lens axis. The result is that a point object is spread over a length Ar in the
image plane. The theoretical resolution of an electron microscope is
approximately 2 A at 100 kV and less than 2 at higher potentials. Because
of these characteristics, HEED microscopy is particularly favourable for
studying superstructures, order-disorder phenomena, grain boundaries, and
generally the real structure of crystals.
LEED techniques are commonly used for studying crystal surfaces
(electrons can penetrate into a crystal only for a very small thickness), in
order to study processes of electronic and ionic emission, catalysis,
nucleation of new phases, oxidation, etc. The function q ( r ) terminates at
the crystal surface: thus in principle the structure of outer layers can be
different from that of internal layers. Interpretation of the spectra is not
easy because of multiple scattering of the electrons. Additional information
is provided by means of Auger spectroscopy of scattered electrons.
Neutron scattering
A neutron is a heavy particle with spin 4 and magnetic moment of 1.9132
nuclear magnetons. Its wave properties were shown in 1936 by Halban and
Preiswerk and by Mitchell and Powers. Neutron diffraction experiments
(n-diffraction) require high fluxes provided nowadays by modern reactors.
They produce fast neutrons whose energy is reduced by collisions in a
moderator of heavy water or graphite. So retarded neutrons are called
thermal and their speed obeys a Maxwell distribution: the corresponding
spectrum is white (see Fig. 3.B.8(a)). A monochromator (usually single
crystals of Ge, Cu, Zn, Pb) selects the desired wavelength A.
Neutrons can also be produced in a pulsed manner by spallation, at a
repetition rate between 24 Hz and 50 Hz. High-energy protons (-1 GeV),
in short pulses at the appropriate pulse frequency, strike a target such as
uranium or tungsten releasing several tens of neutrons per proton (-25
neutrons for 238U).The pulsed neutron flux is only present for a very short
time (the burst lasts around 0.4 ys): heat-removal is then easy and high
fluxes are allowed (higher than those produced by reactors).
High-energy neutrons are slowed down to thermal energies by appropri-
ate moderators. Water, polyethylene, liquid hydrogen, or liquid methane
are frequently used: the choice is determined by the scattering experiment.
During the thermalization process neutrons undergo a large number of
collisions which cause pulse broadening. The width of the pulse leaving the
moderator is roughly proportional to the wavelength (see Fig. 3.B.8(b)) so
that the fractional wavelength resolution is nearly constant. A characteristic
neutron spectrum is shown in Fig. 3.B.8(c): high intensities at short
wavelength (A < 1A) is a very significant characteristic of pulsed neutron
sources.
The neutron-atom interaction comprises interaction with the nucleus and
interaction of the magnetic momentum associated with the spin of the
neutron with the magnetic momentum of the atom. This effect mainly
occurs for atoms with incompletely occupied outer electron shells (for
instance, transition elements).
The neutron-nucleus interaction is governed by very short range nuclear
forces (-10-l3 cm). Since the nuclear radius is of the order 10-l5 cm, i.e. of
several orders of magnitude less than the wavelength associated with the
incident neutrons, the nucleus will behave like a point and its scattering
factor bo will be isotropic and not dependent on sin B/A (see Fig. 3.B.7). By
convention, the scattering amplitude is assumed positive where there is a
phase change of 180" between incident and scattered waves: it has the
dimension of length and is measured in units of 10-12 cm. When the neutron
Fig. 3.0.8. (a) Spectrum from a nuclear reactor. is very close to the nucleus a metastable system, nucleus + neutron, is
The shaded wavelengths are selected by a
monochromator. (b) A schematic time- created which decays by re-emitting the neutron. For appropriate energy a
dependence of two pulses of neutrons leaving resonance effect can occur: then the scattering factor assumes the form
the moderator with different energies (A, > A,). A b = bo - Ab'. Since Ab' can be greater than b, it is possible to have
next pulse starts when the number of neutrons
(N,) is small enough. (c) Neutron flux negative scattering factors for some nuclei (for instance, 'H, 48Ti, 62Ni,
distribution on a pulsed neutron machine. 5 5 ~ n )for
: them the scattering is out of phase by 180" with respect to the
nuclei with b > 0. A few nuclei also present an imaginary wavelength-

dependent term Ab" (see pp. 165-9 for a more extensive treatment of
anomalous scattering).
The angular momentum of the nucleus is another factor which influences
the capacity of scattering. If I is the angular momentum, then it can be
combined with the neutron spin in a parallel or in an antiparallel fashion
+
yielding the total spin of either J = I 4 or J = I - 4, and corresponding
scattering factors b+ and b-. According to quantum mechanics, there are
+
2J 1 orientations in space compatible with one spin of value J. The
+ + +
compound nucleus will then have [2(I+ 4) 11 [2(I - i) 11= 2(2I+ 1)
+
possible states. One fraction of these, w+ = [2(I + 4) 1]/[2(2I+ I)] = (I +
+
1)/(2I 1) corresponds to states of parallel spins while the other one,
+
w- = I/(2I I), corresponds to states of antiparallel spins. In conclusion, a
single isotope with I > 0 can be described as a random mixture of nuclei
with atomic fractions w+ and w- and scattering amplitudes b+ and b-. Then
+
b = w+b+ w-b- will be the coherent scattered amplitude. Incoherent
scattering will also occur, with square amplitude given by [(w+b: +
+
W-b:) - (w+b+ W-b-)'I which contributes to a uniform background with
no diffraction effects. This is particularly important for hydrogen, for which
- -
I = 4, b+ 1.04 x 10-l2 cm, b- -4.7 x 10-l2 cm. Then w+ = 0.75, w- =
-
0.25, b -0.39, and most of the scattering is incoherent background.
In general, one chemical element exists in a crystal structure in different
isotopic forms which are randomly distributed over all sites occupied by the
element. The scattering factor of the element will be a weighted average
over various states of spin of every isotope and over various isotopes. This
disorder generates a pervasive coherent and incoherent background in the
diffraction spectrum. If wn is the relative abundance of the nth isotope then
the effective scattering factor will be b = C wnbn and the total diffuse
background scattered intensity will be 1,wnbE - b2.
Atoms which possess a magnetic momentum because of the presence of
unpaired electrons interact with the magnetic momentum of the neutron
giving additional neutron scattering. Interaction occurs in a finite domain
around the nucleus: therefore, as for X-rays and electrons, magnetic
scattering will decline with sin6lA.
Let us consider the case of identical atoms with all spins parallel
(ferromagnetic materials) and spins alternating parallel and antiparallel
(antiferromagnetic materials). The magnetic scattering amplitude of each
atom is given by
-
p = (e2ySfmaglmc2) 0. 54Sfmag(10-l2 cm)
where y is the neutron magnetic moment in nuclear magnetons, e and m are
the charge and the mass of the electron, c the velocity of light, S the
electron spin quantum of the scattering atom, fmag the atomic scattering
form factor given by the Fourier transform of the distribution of electrons
having unpaired spin and normalized so as to be fmag = 1 for 6 = 0. p is of
the same order as nuclear scattering.
Nuclear and magnetic scattering are additive for unpolarized neutron
beams: then
IF 1' = IF, , :1 + sin2 cv I FmagI2
where a is the angle between the unit vector in the direction of the spin
200 / Carmelo Giacovazzo
orientation in the sample and the vector r * . For a sample containing a single
magnetic domain sin2 a is equal for all atoms. For the multi-domain samples
it will be necessary to average over the various spin orientations. For
ferromagnetic and ferrimagnetic samples external fields for orienting the
spins can be applied so as to obtain sin2 a = 0 or sin2 a = 1. Measurements
of IFI2 for both cases separate magnetic from nuclear scattering and allow
the study of magnetic structures. Sometimes the magnetic cell coincides with
the 'chemical' cell so that the magnetic contribution to the intensities is
added to nuclear contribution. In other cases the magnetic structure has a
cell which is a multiple of the chemical cell causing additional purely
magnetic reflections. Magnetic symmetry can be described by means of
space groups of antisymmetry or by means of colour groups (see Appendix
l . F and Fig. 1.F.2(b)) associated with classical three-dimensional space
groups.
Monochromatic beams of polarized neutrons are easily obtained from
unpolarized beams by using suitable single-crystal monochromators (for
example, a Co-Fe alloy). For a given polarization, either constructive or
destructive interference between nuclear and magnetic scattering amplitudes
can occur. Thus the technique may detect a weak magnetic scattering even
when it is accompanied by a strong nuclear scattering.
Among the most important characteristics and applications of neutron
diffraction we quote:
1. The interaction of neutrons with the matter is weaker than that of
f,-
-
X-rays or electrons (see Table 3.B.3). Generally speaking, scattering
amplitudes are fa, (10-12-10-11) cm for X-rays, fa, -
lo-' cm for electrons,
10-l2 cm for neutrons. Therefore high neutron fluxes and crystals with
dimensions of several millimetres are needed for measuring appreciable
scattered intensities.
2. The b values vary non-monotonically with the atomic numbe; Z:
isotopes of the same element can have very different values of b. This allows
us to distinguish between atoms having very close values of Z (but very
different values of b : e.g. b,, = -0.36, b,, = 0.96, b,, = 0.25) and to
localize the positions of light atoms in the presence of heavy atoms.
Neutrons are particularly useful for localizing hydrogen atoms. Usually they
are partially or completely substituted by deuterium with a value b > 0 and
with negligible incoherent scattering.
3. The three-dimensional Fourier synthesis with coefficients F,,,, gives the

positions of the nuclei (they do not necessarily coincide with the atomic
positions found by X-ray diffraction). The peak heights are proportional to
the b values of the corresponding nuclei: if b is negative, the peak is
negative too. Fourier syntheses with coefficients F,, give the spin density
distribution of the magnetic atoms.
4. Since b does not depend on sin 8/A nuclear scattering decreases only
because of the temperature effect. Thus reflections with high values of
sin 8lA can be collected giving atomic positions and thermal parameters
with accuracy higher than from X-rays. In many cases X- and n-diffraction
experiments are both performed in such a way that accurate maps of the
electron density can be obtained.
The diffraction of X-rays by crystals ( 201
5. The energy of a neutron with wavelength of 1A is of the order of

0.1 eV, which is comparable with the energy of the modes of thermal
vibration of the crystal. This causes inelastic scattering but also allows for
the possibility of studying energy changes with good accuracy.
6. Some additional information about neutron diffraction experiments
with pulsed neutron sources should be useful. In diffraction techniques using
monochromatic radiation, intensities are measured by moving a detector to
different scattering angles OH chosen according to Bragg's law A. =
2dHsinOH. At spallation neutron sources, where the neutron beam is
pulsed, the discrimination among neutrons with different wavelengths may
be accomplished by their time of arrival at the detector (time of flight): thus
diffraction effects can be measured at a fixed scattering angle $0:
AH = 2dHsin OO.
From de Broglie's relation mu = h/A (m and v are mass and velocity of the
neutron respectively) the following relation arises:
mv = m(Llt) = hlA,
+
where L = L, L,. Here L, is the flight path from moderator to sample and
L2 is that from sample to detector. Then
tH= 505.555LdHsin O0
where L is measured in meters, A and dH in A, t in ys. Thus time of flight
depends linearly on both the flight path and on the wavelength. In addition
the equations suggest that resolution improves with increasing flight path.
Area detectors (see Chapter 4) are the best choice for time-of-flight
techniques. Indeed, even if the source is on for brief pulses the full
spectrum may be used (in monochromatic techniques coupled with a reactor
the source is on all the time but a small portion of the spectrum is used).
3.C Scattering of X-ray by gases, liquids, and

amorphous solids
Introduction
In this appendix we will describe the scattering of X-rays from gases,
Iiquids, and amorphous solids. No attempt will be made to deal exhaustively
with the great amount of work made in the field: only the basic principles of
the phenomena will be described. For further information the reader is
referred to specialist papers on the subject or to some enlightening
books.[3w1]
First ~ e b y e [ ~(but
' ] see also ~ h r e n f e s t ~realized
~ ~ ] ) that no arrangement of
molecules of finite size could be considered completely random. The most
random array which can be imagined is that of a monoatomic gas: but even
in this case the finite size of the atoms prevents their approaching one
another within a distance smaller than a given threshold. According to
Debye, two types of effects could be found on the diffraction patterns:

(1) the effects due to the more or less fixed spatial arrangements of the
atoms in the molecules (even if molecules have random positions and
orientations), called 'internal interference effects';
(2) the effects due to the mutual configuration of the molecules, which may
be not arbitrary since it is conditioned by their dimensions and their
packing, called 'external interference effects'.
The relative importance of the two types of effects is a matter of degree.
Indeed, for monoatomic gases external interference effects are dominant,
while for polyatomic gases internal interference effects are not negligible.
Again in liquids (or amorphous solids) the existence of average interatomic
distances produces well marked external interference effects: if the liquid is
composed of groups of atoms the internal interference effects are again not
neglibible.
In the following sections some formulae will be developed which refer to
the coherent scattered radiation. However, in experiments the total, say
both coherent and incoherent radiation, is usually collected; in the case of
amorphous bodies the incoherent radiation may be an important fraction of
the total radiation. According to the discussion on p. 147, the incoherent
intensity relative to one atom is given by
k
where k varies over the individual electrons of the atom. If we sum (3.C.1)
over all the atoms in the assemblage the total incoherent scattering is
obtained, which is a smooth function slowly increasing with sin B l A . Such a
function has to be subtracted from the total scattering pattern: only after
that can the coherently scattered intensity be used to deduce structural
parameters.
Diffraction from a finite statistically homogeneous object

Usually the motion of atoms or molecules in gases and liquids is extremely
rapid compared with the duration of a diffraction experiment. In these
conditions the phase value of F(r*) has no practical interest while its
modulus can provide useful information on the statistical distribution of the
atoms. Indeed the measured diffraction intensity is the weighted sum of
contributions arising from the various configurations of the atoms. A similar
conclusion holds for amorphous materials, for which trying to calculate the
atomic positions has no practical meaning while it is possible to estimate
interatomic distances (on which diffraction intensities depend).
Suppose that the object we want to investigate is finite and statistically
homogeneous, of volume a. In accordance with eqn (3.27)
and the Patterson function becomes
P(u) = I
R
p d r ) p d r + u)@(r)@(r+ u) dr. (3.C.2)
The product @(r) @(r + u) is always vanishing except when both the points
+
r and r u are inside the object: in this case @(r)@(r+ u) = 1. The space ' \
\
I
domain in which this condition is verified is a function of u and will be \
denoted by Qv(u), where 0 6 v(u) 6 1. It is readily seen from Fig. 3.C.1, \
that Rv(u) is the volume which belongs to the object and to the object
shifted by u. Furthermore, v(u) is centrosymmetric (v(u) = v(-u)), it
decreases with increasing u and takes its maximum for u = 0 (v(0) = 1).
+& , 1
I
I
I
Accordingly I
where
RY(U)=
I" +
@(r)@(r u) d r = @(r)* @(-r).
For homogeneous objects (3.C.3) may be written as

(3.C.4) Fig. 3.C.1. The dashed area has volume Rv(u)
where P,(u) is the Patterson function for the unit volume.

According to (3.A.37)
The Fourier transform of Qv(u), according to eqns (3.C.4), (3.A.37), and

(3.A.21) is
T[Qv(u)] = j Qv(u) exp (2nir" u) du = ID(r*)12 (3.C.6)
so that
Since @(r) is real, 1D(r*)l2 is centrosymmetric: its maximum value is

1 D (0).1' According to (3.27) D (0) = 52 so that
Furthermore, according to (3.C.6)
Q* ID(r*)l2dr* = L .
Pv(u) d u b *exp (2nir* u) dr*
From (3.C.8) and (3.C.9) it is easily deduced that 1D(r*)12rapidly decreases

with (r*l, and is appreciably different from zero in a domain near the origin NU -
of the reciprocal space (when the diffracting object is a crystal, (D(r*)I2 f
repeats itself around each lattice point: the larger is 52, the smaller the A
-
domain).
Replace in (3.C.9) the decreasing function ID(r*)I2 by a constant which is
equal to its maximum value 52' in a volume w defined by S2'w = 52 (see Fig.
3.C.2 for a one-dimensional example). Then an approximate estimate of the
-
domain volume is obtained, say w = 1/52, together with its approximate
dimension rz ( 1 / ~ )=~l/ro,
diffracting object.
' ~ where ro is the average dimension of the Fig,3.C.2. Thetypicalform
function of r*.
of , D ( r X , as
,~a
r*
In terms of diffraction angles the width of the diffraction peak near the
- -
origin of the reciprocal space may be obtained from the Bragg relation
(2/r,*) sin 8, (2Ir:) 8, 2roeo= A from which
For Cu K, radiation and an object with ro = 0.1 pm it is 8, = 4.4 x On

assuming that diffraction around the central peak may be observed down to
8 = low3,0.1 pm constitutes the upper limit for the dimension of the object.
Diffraction from a finite statistically homogeneous object with

equal atoms
Suppose that the object is composed of N atoms of a single type, and Nl is
the number of atoms contained in the unit volume (of course fluctuations
from this value are possible due to thermal motion or to structural features
of amorphous solids). Then v, = l/Nl is the average volume available for
each atom. Equations (3.C.5) and (3.C.7) may be written in terms of
autocorrelation functions of atomic density rather than of electron density.
Let us define
P(r) = Pa(r) * qa(r)
where pa(r) is the electron density relative to one atom and qa(r) is the
probability of finding an atom in r. Then
P(u) = Pa(r) * Pa(-r) * qa(r) * qa(-r)
and
IF(r*)I" .
T[P(u)] = f '(r*) T[Pa(u)] (3.C.11)
where P,(u) is just the autocorrelation function of the atomic density.
Let Pal(u) be the function Pa(u) calculated for the unit volume. Then,
according to (3.C.5)
Pa(u) = Qv(u)Pal(u)
and (3. C. 11) becomes
Let us now express Pal(u) in a more convenient form. On assuming the

origin on an atomic site, the probability of finding a second atom in the
volume dv at the extremity of the vector u is dp(u) = dvlv, only for a
completely random atomic distribution. In general we could write
where p(u) is a distribution function defining the configuration of the atoms.

In general p(u) is not easily found. A very simple model for which p(u) may
be rigorously calculated was proposed by Zernicke and ~ r i n s : [it~ concerns
~]
the one-dimensional arrangement on a segment of length L of N rigid
spherical particles of diameter d. Figure 3.C.3 shows p(u) as a function of
four values of the particle concentration dll, where 1 = LIN is the average
length allotted to each particle. It is easily seen that always p(u) = 0 when
u < d (the spheres are impenetrable). At very small concentrations (quasi-
zero volume particles which do not exert any influence on each other) p(u)
-
v--=- Fig. 3.C.3. The distribution p ( u ) for one-
dimensional arrangements of objects of length d
(d) as a function of the compactness.
is similar to the atomic distribution of the perfect gas. For higher

concentrations the model represents distributions similar to those found in
-
real gases, liquids, and amorphous solids. p(u) will oscillate about unity at
short distances from the origin, while p(u) 1 at long distances. Oscillations
of p(u) are larger for higher concentrations. When dl1 becomes maximum
the close packing of the particles gives rise to a perfect lattice of period d:
then very sharp maxima of p(u) will occur at u = nd (full short-range and
long-range order).
Let us now denote by z(u) the probability of finding an atom in the
element of volume dv located at the extremity of the vector u from an atom
located at the origin. Then
It may be noted that l / v l is not only the asymptotic value of z(u) but also
its mean value. Indeed, integrating z(u) on the unit volume gives
from (3.C. 12) the relation
follows. Write now (3.C.14) by emphasizing the oscillating part of p(u):

Then the Fourier transform of z(u) is
and (3.C. 16) becomes (Nl = NlQ)

N N 1
F(r*)12= f2(r*)60*
v1
~ ( r * ) ~ ' +f2(r*)(l + -T[p(u)
Vl
- l]) * lD(r*)12.
The function ID(r*)I2 can be considered very broad with' respect to the
Dirac delta function but very sharp with respect to T[p(u) - 1] (which in
disorder structures is a function slowly varying with r*). According to
(3.C.9) the integral of the ID(r*)I2 peak is Q: therefore
In terms of the normalized intensity I ~ ( r * ) l 'eqn (3.C.18) becomes
[p(u) - 11exp (2niu r*) du
The first term in (3.C.19) corresponds to the peak at the origin of the
reciprocal space. It is detectable only at very small angles and it is
distinguishable from the primary beam (see the previous section) provided
the diffracting object is very small (say, <l pm). It does not depend on the
internal structure of the object but only on its external shape. We will
discuss such a peak on p. 213.
The second term depends exclusively on the statistical distribution of the
atoms: if this is perfectly uniform p(u) = 1 and I E ( ~ * ) [1.~ =Thus the
variations of IE(r*)I2about its average contain information about the atomic
distribution in the object. Such a distribution may be determined by
inversion of (3. C. 19):
which is the most general formula describing the diffraction from a

statistically homogeneous object composed of atoms of one kind.
Diffraction from an isotropic statistically homogeneous object

If the diffracting object is statistically homogeneous and isotropic (that
occurs, for example, in gases, most liquids, and in finely dispersed crystal
powders) the function I E ( ~ * and
) [ ~p(u) will also be isotropic. Then using
results in Appendix 3.A, p. 181, and neglecting the first term on the
right-hand side of eqn (3.C.19) give

sin 2nr * u
du
2nr*u
where 4nu2(p(u) - 1) is the radial atomic distribution function: the

product of such a distribution for du gives the number of atoms situated at a
distance between u and u + du from the atom assumed to be at the origin.
Inverting (3.C.21) gives
r*(I~(r*)l'- l)(sin 2nr*u) dr*. (3.C.22)

U
Equation (3.C.22) shows that from diffraction experiments the average

number of neighbouring atoms can be deduced as a function of the distance
from a given atom chosen to be at the origin.
Equations (3.C.21) and (3.C.22) also hold when the object is made by
particles constituted by groups of atoms: then v, is the average volume
available for each particle.
The Debye formula

A simple way to describe diffraction from an isotropic statistically homoge-
neous object composed by a collection of N identical groups of atoms with
known geometry but random orientation and position was proposed by
Debye. Let
P
4 (r*) = C J(r*) exp (2zir* 5 )
]=I
be the structure factor of the nth group of atoms (each group composed of p
atoms). Then
N
and
Since the observed intensity will be the average with respect to all the
possible mutual configurations of the atomic groups
~ ,ql - q,. In accordance with p. 180 we obtain

where u ~ , ,=
sin 2nr * uii
sin 2nr*uii
2nr*uij
) (3.C.23)
In terms of normalized intensity (3.C.23) becomes
(1E(r*)I2) = 1 + 5
i+j=l
YiY,
sin 2nr * uij
2xr*uij
where Y = f /N(Cf=l (f?)"'. In order to give some examples, for a molecule
composed of two atoms at distance 1 (3.C.23) becomes
For a tetrahedral molecule (i.e. SiC1,) (3.C.23) becomes
where 1 is the edge of the tetrahedron and d its radius.

In favourable conditions estimates for interatomic distances can be readily
According to p. 174, the
obtained from the intensity diffraction pattern.[433451
value of the interference term sin2nr*x/(2nr*x) is unity for r* = 0 and
becomes zero for r*x = n/2: subsidiary maxima occur when
or, in other terms, when
The position of the mth maximum for large values of m will approach more
and more closely to
x sin B/A = (0.125 + 0.5m)
but maxima will become weak and ill defined.
It is easily seen that the maxima in the IE(r*)I2 curve will coincide with
those given by (3.C.27). In particular, from the first maximum located in
(sin O/A), the interatomic distance
x = 0.615/(sin (3.C.28)
is obtained.
It is easily seen that Debye maxima will correspond to those provided by
(3.C.21).
Diffraction by gases
For a perfect gas (consisting of identical atoms with negligible volume and
exercising no action on each other) any atom can occupy any position with
the same probability. Therefore p(u) = 1, (3.C.14) reduces to
and (3.C.21) reduces to

I ~ ( r * ) l ' =1
or, in terms of IF(r*)I2, to
IF(r*)12= ~ f ' ( r * ) .
If the central peak is also taken into consideration, the observed intensity is
of type described in Fig. 3.C.4. For r* = 0 the value of ( ~ ( r * )isl ~~~f
(replace JD(r*)I2by 1D(0)1' = Q2 and v, by S2/N in (3.C.18)). As soon as r*
is out of the peak at the origin the value of IF(r*)I2 is equal to ~ f ' .
If the gas is a mixture of perfect monoatomic gases the scattering is
practically given by
If the gas is composed of identical molecules formed by atoms with

negligible volume and arranged as in a perfect gas then features will appear
in the scattering curve which are influenced by the molecular structure. A
typical diffraction pattern provided by chlorine diatomic molecules is shown I
sin 011
in Fig. 3.C.5 (curve (a)). The maxima in the experimental curve are not well
Fig. 3.C.4. Schematic diagram of the scattered
defined (they are represented by roughly horizontal portions of the curve) intensity from a monoatomic gas.
because of the continuous decay o f f with sin @/A. The maxima should be
better emphasized by plotting JE(r*)I2(curve (b)).
For real gases the intrinsic volume of the atoms is no longer negligible.
Since atoms are impenetrable, an atomic distribution function p(u) such as
that described in Fig. 3.C.3(a) may be chosen, where d is the atomic
diameter. Then (3.C.21) becomes
where v o = 4 ~ ~ ( d / 2 ) is 3 volume of one atom, C is the concentration

~ /the
factor (ratio between the effective volume occupied by the atoms and the Fig. 3.C.5. Schematic diagram of the diffraction
gas volume) and (see p. 181) q ( y ) = 3 (sin y - y cos y)ly3, where y = 2nr*d. pattern for the diatomic molecule CI,.
Since q(0) = 1 it is IE(0)I2= 1 - 8C: thus the asymptotic value of JE(r*)I2
+
for r * 0 is smaller than unity by a quantity proportional to C. For C > 118
(e.g. for high-pressure gases) IE(O)(' < 0, which is not acceptable: for such
cases the diffraction by gases closely resembles that by liquids or amorphous
solids, for which supplementary considerations are needed (see p. 212).
Since electron scattering is much more intense than X-ray scattering (see
p. 195) electron diffraction is preferable for gases. Very small quantities of
gas can be used with shorter exposure times, and peaks can be measured at
greater values of sin 81A.
Gas electron diffraction a p p a r a t ~ s [ ~
usually
~ , ~ ~consists
] of an electron gun,
a sample injection system, and the detector: the last is often a photographic
plate. Typical electron accelerating voltages, highly stabilized, are between
30 and 80 kV, with exposure times of a few minutes. In order to compensate
the fast decay of the scattered intensity with sin @/A a rotating sector is
introduced into the path of the scattered electrons: it is a metallic disc
designed in such a way that the intensity reaching the photographic plate is
larger at larger scattering angles. In these conditions the diffraction pattern
will emphasize high-angle interference effects, whose intensity will now be
Fig. 3.C.6. Electron diffraction patterns from

C,H,-NO, gas (working temperature about
353 K, two camera di~tances).'~''
sufficient to be measured by a microphotometer. After the background

elimination, the scattering optical density distribution is converted into
electron distribution.
In Fig. 3.C.6 two gas-phase electron diffraction patterns[481of nitroben-
zene C6H,-NO2 are shown (corresponding to two different positions of the
plate). The experiment was made in order to determine, in combination
with ab initio molecular orbital calculations and X-ray single-crystal
structure analysis of several derivatives, the molecular structure of C6H5-
NO2 in the planar and orthogonal conformation. From data in Fig. 3.C.6
the one-dimensional spectrum in Fig. 3.C.7 may be derived. In the figure
the curve sM(s) is given where M(s) = (I, - Ib)lZb, s = 4n sin 8/A, I, is the
total observed radiation, Zb is the background intensity. The corresponding
radial distribution curve p ( u ) is shown in Fig. 3.C.8: the positions of the
most important distances are marked by vertical bars whose height is
proportional to the weight of the distances. The geometrical parameters of
the theoretical model (the molecule as a whole was supposed to have a
binary symmetry, while a local mm2 symmetry was assumed for the benzene
ring and the nitro group), together with vibration amplitudes, were refined
by a least-squares process applied to molecular intensities: the good final
agreement may be deduced from the small differences between the E and T
curves.
I I ! I I Fig. 3.C.7. Molecular intensity curves (E,

I I I
experimental; T, theoretical) for C,H,-NO, (gas
0 5 10 fb 20 25 30 35 electron diffraction, two camera distances). Also
s (Am' shown are the difference
- 3-
A
v- - -- - -
V V
A n -
V
v - -
v - --
I I I I I I
I I Fig. 3.C.8. Radial distribution curves for CH
, -,
0 1 2 4 5 6 7 NO, (E, experimental; T, theoretical). Also
r(A) shown is the difference curve (E-T).'~'
Diffraction by liquids and amorphous bodies

In liquids or amorphous bodies each molecule may be considered in contact
with a certain number of neighbours. Thus both internal and external
interference effects will dominate the diffraction pattern. Let us first
consider an atomic liquid: if atoms are assumed to be hard spheres which do
not exert any force on each other the distribution p(u) may be cal~ulated.[~~1
In Fig. 3.C.9(a) p(u) is plotted for two values of the concentration
parameter C (ratio between the effective volume occupied by the spheres
and the total volume of the liquid). The corresponding I E(r*)I2 functions
may be derived by application of (3.C.21) and are shown in Fig. 3.C.9(b).
Their main features are:
(1) for small values of r* it is 1 E(r*)I2 < 1;

(2) a large maximum occurs at about r* =0.95/d. Its intensity increases
with C while its position does not vary;
(3) for lage values of r* 1E(r*)I2 tends to unity as for perfect gases.
In spite of the purely geometrical nature of the model the diffraction pattern
of monoatomic liquids closely satisfies the model predictions. In Fig. 3.C.10
the IE(r*)I2 curve of liquid mercury is shown: its Fourier transform (eqn
(3.C.22)) shows a maximum a little over 3 A.
A further example concerns the diffraction pattern of water. As it is well
1/d r* known, the molecules are strongly polar and V-shaped with 0 - H distances
(b) of about 1 A, and nearly tetrahedral HOH angles (109"). For our purposes,
Fig. 3.12.9. (a) Distribution functions P ( u )for they may be geometrically represented by spheres of about 2.8 A diameter.
hard spheres of diameter dfor two values of the The water scattering curves[s01 at 1.5 "C and 83 "C are shown in Fig.
concentration parameter C. (b) I E ( ~ * )curves
I~
corresponding to the two distribution shown in 3.C.ll(a); the corresponding radial distribution curves are given in Fig.
(a). 3.C.ll(b). The main maximum for the radial distribution occurs at a radius
of about 2.8 A while a second broader maximum occurs at about 4.5 A,
nearly vanishing at high temperature. Such results suggested to Bernal and
~ o w l e r [ ~that
l ] the arrangement of H 2 0 molecules in water may be
described as a broken-down ice structure in which each molecule tries to
bind four neighbours: since bonds are continually breaking and re-forming
at any instant each molecule is bonded to fewer than four neighbours.
As a last example, examine now Fig. 3.C.12, where the diffraction
patterns by vitreous silica, a cristobalite crystal powder, and silica gel are
shown. The main maxima of all the three curves nearly overlap: but
cristobalite shows numerous sharp maxima while only one maximum occurs
in vitreous silica. Furthermore, vitreous silica intensity decreases with
sin 8/A as we have just seen for liquids, while silica gel shows increasing
intensity toward small sin OlA. The radial distribution function for vitreous
shows a first peak at r = 1.62 A and a second one, at 2.65 A: that
indicates that the tetrahedral coordination in the crystalline state persists in
the vitreous state (but here the orientations of the tetrahedral groups are
randomly distributed). The vitreous state is essentially homogeneous
(diffraction intensity decreases at small sin $/A as for liquids) while silica gel
- is made from very small discrete particles (10-100 A) with voids among
0.5 1 2 r * them (diffraction intensity increases at small scattering angles: see the next
Fig. 3.C.10. I E ( ~ * ) /curves
' for liquid mercury. section).
Small-angle scattering
Small-angle scattering is a technique for studying structural features or
inhomogeneities of colloidal dimensions.[401For wavelengths of about 1A
the typical angular domain of the technique ranges up to one or two
degrees.
We have already seen (p. 204) that when diffraction occurs from a finite
statistically homogeneous object of volume S2 a central peak in the intensity
curve may be measured which becomes broader as the object size decreases,
and does not depend on the internal structure of the object. Its intensity
distribution varies according to (3.C.18):
Since diffraction is considered at very small angles f (r*) will coincide with
the number of electrons per atom. Thus the above equation reduces to
We consider now a system constituted by N particles with electron density p

randomly dispersed in a homogeneous medium of electron density po
(solvent). Particles are supposed to be separated from each other widely
enough, thus they will generate independent contributions to the diffracted
intensity. The system may also be thought of as a medium of constant p,
and macroscopic volume S2 to which N particles are summed with density
Ap = p - p,. If S2 is sufficiently large the central peak due to the
homogeneous medium is totally unobservable, while the observed intensity
distribution may be ascribed to electron inhomogeneties only. In this case
the mean scattering power for particle is
where IDP(r*)l2is the average value of )DP(r*)l2over all the possible particle
orientations. Different shapes of particles will give rise to different shapes of
Fig. 3.C.11. (a) 1 f ( r * ) l 2water scattering curves at
the scattering function in reciprocal space: ideal scattering intensities can 1 .5"C and 83 "C; (b) the corresponding
therefore be calculated for spherical, cylindrical, flat, ellipsoidal, etc., distributions 4nu2p(u).
particles. The results are all rather similar, particularly in the central range,
but remarkable differences occur at larger angles. It may then be expected
that in the central part a universal approximation for all particle shapes
must exist
Let us assume the origin 0 in the centre of gravity of a particle of volume
S i l ~ c aGel
I .
.....- ....
Vit. SIO, - # - - ' . . ..::. . .:.;:. . .: : . .
..
. .. .. .... .... .......,..:..: ,
. . . :.
.. .
.......
Fig. 3.C.12. Diffraction patterns for silica gel,
vitreous SiO,, and for cristobalite.
up and let us estimate
Dp(r*)= l exp (2nir* r) dv. (3.C.32)
At very small angles we can expand the exponential function in (3.C.32) in

power series: it will suffice to expand cos 2nr* . r according to cos 0
1- 0'12. In Cartesian coordinates
-
.
cos 2nr* r = 1 - + y2P2+ z2y2)2,
where a; p, y stand for the cosines between r* and Cartesian axes. When
integration is performed the contribution of mixed products vanishes by
hypothesis (the centre of gravity has been located at the origin) and we have
Dp(r*) = v, - 2n2r*2 ((u21 x2 dx + 82j

y2 dy + y21 z2 dr).
Since we are interested to calculate IDP(r*)l2, say the average of IDP(r*)l2

over all particle orientations, we have to rotate Cartesian axes about the
origin. We can then replace (u2, P2, y2 by ( a 2 ) = ( p 2 ) = (y2) = 5 SO as to
obtain
where r2 stands for (x2+ y2 + z2) and
R = ( l v ) ( r2 du)
UP
is the radius of gyration (= the mean square distance) of the particle with
respect to its centre of gravity.
According to (3.C.31) the scattering power per particle will be
where n, = ( p - p,)vp. By taking the logarithm to the base 10
is obtained. If lg ~ ~ , ( r * ) lis' plotted against ( 2 0 ) ~a curve will be obtained

which, at small values of 0, is close to a straight line with slope
(u = -5.715R;lA2, from which the gyration radius R, = 0.418~- is
obtained.
In Fig. 3.C.13 the Guinier plot for a low-concentration solution of

haemocyanin from Astacus leptodactylus is shown.[54]The reader will easily
derive from the figure (CuK, radiation used) a gyration radius of %
R,= 62 A.
The radius of gyration is connected to the geometrical parameters of ,
some simple homogeneous bodies in the following way:[s5]
sphere of radius r:
I
R i = 3r2/5; 1 2 3 4 5 6
10512tv2
hollow sphere (limiting radii r, and r2): Fig. 3.C.13. A Guinier plot.
ellipsoid (semi-axes a, b, c):

~ 2=, (a2 + b2 + c2)/5;
prism with edge lengths a, b, c:
R: = (a2 + b2 + c2)/12;
elliptic cylinder (height h ; .semi-axes a, b):
hollow cylinder (height h, radii r,, r2):
Other important parameters can be evaluated directly from the scattering

data: among others the molecular weight (from the scattering intensity at
zero anglesrs6]),the total surface of the particles for unit mass (from the
region where the small-angle scattering tends to zero[s73s8]), and the volume
of the particle. This latter may be obtained by integrating (3.C.31) over all
the reciprocal space (by extrapolation of r*2~ ~ , ( r *at) lvery
~ small r*):
According to (3.C.9) the above relation reduces to
from which up may be derived.

So far we have considered particles dispersed in a homogeneous solvent
and widely separated from each other. If the concentration of the particles
increases the mutual interference will enter into play and the scattered
intensity is no longer the sum of the individual particle scatterings. As
described in on p. 206 a probability function p(u) can be introduced: then
4m2p(u) du will represent the number of particles with distance u lying in
+
the interval (u, u du) from a particle at the origin. Then the scattering
power per particle will be (see eqn (3.C.21))
2 "
( r * ) = ( ~ ~ ( r * ) ~ )-( l + u[p(u) - 11sin (2nr*u) du
r*v1 0
where (IFp(r*)12)is the squared modulus of the structure factor of a particle
averaged over all orientations. From the above equation the function p ( u )
may be derived by inverse Fourier transform.
3.D About electron density mapping

The accurate determination of charge density distribution is of basic
importance in several scientific areas. For example, for the study of
(1) electronic structure of metals and alloys;
(2) metal-ligand interaction;
(3) variation of solid state properties (i.e. conductivity) with temperature.
Spatial partitioning of charge density can also answer some specific
questions, such as the location of lone-pair maxima, the net charge of a
particular atom, the excess charge accumulated in a covalent bond, and so
on.
Such results can be attained only if the selection of the compounds, the
process of measuring diffraction data, and their subsequent treatment is
carried out with great care. Among the various desiderata we mention the
following.
1. Compounds with a small valence-to-core ratio are less suitable to charge
density studies. If we assume that the number of valence electrons per
unit volume is approximately constant for different compounds, the
valence-to-core ratio is fixed by the number of core electrons per unit
volume. Stevens and ~ o ~ ~ e introduced
n s [ ~ ~ the
] criterion
1 core
as a possible criterion to estimate the suitability of a system for charge

density studies (V is the volume of the unit cell). S varies from 3-5 for
the first-row-atom organic crystals to 0.1-0.3 for metals and alloys.
Crystals with S < 1 require extremely precise diffraction data and careful
work.
2. Centrosymmetrical crystals are often preferred (in acentric crystals the
phases of the structure factors cannot be uniquely determined).
3. Collection of diffraction data has to be carried out at low temperature.
Then thermal diffuse scattering can be neglected (a posteriori corrections
are seldom accurate), high resolution data become available (then
accurate structural parameters can be calculated), and electron charge
density can be easily deconvoluted from the atomic thermal vibrations.
4. Sophisticated measuring programmes should be used, in order to
perform optimal scanning mode and profile analysis, decide
differentiated measuring times (in order to obtain a set of structure
factors with equal variance), prevent multiple scattering, and perform
careful correction for extinction, absorption, anomalous dispersion, etc.
5. When possible, neutron data should also be collected in order to obtain
nuclear positional and thermal parameters which are unbiased by
asphericity in the atomic charge density.
1
Yq,, = 3 sin2 6 cos 241 = 3(s: - s:)
Y;,, = 6 sin2 6 sin @ cos @ = 6sxsy
Yq,, = 3 sin 6 cos 6 cos 4 = 3sxs, quadrupoles.
Y;,, = 3 sin 6 cos 6 sin 4 = 3sys,
Y;,, = 6 cos2 6 - 2 = 6s; - 2
I
Y',,, = 15 sin3 6 cos 341 = 15(s: - 3s,2)sx
Y;,, = 15 sin3 6 sin 341 = (45s: - 1 5 s ; ) ~ ~
Y;,, = 15 sin2 6 cos 6 cos 241 = 15($ - s;)s,
Y:, = 30 sin2 6 cos 8 sin cos @ = 30s,sys,
$J octapoles.
Yg,,= 1.5 sin 6(5cos2 6 - 1)cos @ = 1.5(5s;- l)s,
Y:,, = 1.5 sin 6(5 cos2 6 - 1) sin @ = 1.5(5s: - l)sy
Y;,, = 10 cos3 6 - 6 cos 6 = (10s; - 6)s,
In Fig. 3.D.1 some electron density plots in special sections of direct
space are shown.[611
In most organic molecules the expansion (3.D.2) may be stopped at
octapoles: then the deformation charge density is described by a linear
combination of 16 terms, for which 16 population parameters have to be
estimated.
In some formulations[621the free-atom valence shell is modified by an
expansion-contraction radial parametrization in order to take into account
the fact that an atom will expand or contract when it becomes more
negative or more positive. Such a feature may be incorporated into a
perturbed valence density p:,,ence(r) given by
where pvaIenceis, as in eqn (3.D.1), the free-atom ground state density,

PValenc,is the valence shell population, and k3 is a normalization factor. If
k > 1 the atom is contracted relative to the free one [indeed p l ( r ) is
proportional to p(kr) with kr > r].
The atomic scattering amplitude has the same type of multipole expansion
as the atomic charge density. By Fourier transformation of (3.D.2) one
obtains
where 6*, #* are angular components of the Bragg vector,
and ji(x) is the spherical Bessel function of order 1.

From (3.D.3) the calculated structure factors Fcalcare easily obtained.
Then (full or block-diagonal) least-squares refinement can be carried out: as
well as the usual positional and thermal parameters for each atom, the
population parameters of the deformation terms may also be refined. Such a
formalism should improve (with respect to traditional crystallographic
refinements) the agreement between observed and calculated structure
factors.
Suitable Fourier syntheses can be calculated and analysed in order to

reveal the effects of interatomic bonding on the charge clouds. Fourier maps
of large usefulness are
1
Ap(r) = y x
H
( 4 ( H ) - F2(H)) exp (-2niH. r)
where
6 = F c a ~ cmultipole,
, F2 = F c a ~ cfree
, atoms. AP is then the deformation map,
i.e. the difference between the atomic densities represented by spherical
harmonics and those represented by free atoms;
F' = Fobs,F2 = Fcalc,rnultipole Ap is then the residual density, e.g. the part
of electron density not accounted for by the multipole model;
4 = Fobs, F2 = Fcalc,
free atoms,under the condition that F2 is calculated
with parameters from high-order (HO) refinement (high-order reflec-
tions are unaffected by chemical deformation of valence orbitals and
locate atoms carefully). Ap is then denoted as a (X - XHO)deforma-
tion map;
Fl = Fobs,F2 = Fcalc,free atoms,with fi calculated as in (3) but for core
electrons only. Ap is the (X - X,,) valence map;
4= Fobs, F2= Fcalc,
free atoms,with F2 calculated with parameters from
neutrons or a joint high-order X-ray and neutron data. Then Ap is the
+
X - N or X - (XHO N) deformation map;
4 = Fobs, = Fca~c,~reeatorns,with F2 calculated as in (5) from core
+
electrons only. Ap is then a X - (XHO N) valence map.
The estimation of the charge density model obtained at the end of
calculations is different according to whether least squares or Fourier
methods have been used[671.However, direct and reciprocal fitting are
nearly equivalent: indeed, if we assume in (3.D.4) that Fl = Fobs,F2= Fcalc,
AF = Fobs(H)- Fcalc(H),and we calchlated the difference Patterson
I 1
x
Ap(r)Ap(r + u) dr = - I A F cos
V H
~ ~2nH u,
then for u = 0 we obtained
Thus least-squares parameters which give the best fit between observed
and calculated structure factors are expected to give rise to the lowest
variance in Ap.
Direct evaluation of atomic or molecular charges, of dipoles and higher
moments, etc, can be obtained from the estimated population parameters or
directly from the electron d e n ~ i t y . [ For
~ ~ ,example,
~~] the net charge on atom
j is
when the integration is made over the (not always easily defined) atomic
volume. Also, the dipole moment of a molecule may be calculated as
where the integration is over the molecular volume.
3.E Modulated structures and quasicrystals

Embedding of modulated structures in higher-dimensional
space
In Chapter 3, pp. 171-3, it has been suggested that incommensurately
modulated structures (IMS) can be better described in higher-dimensional
spaces. From now on S, and L, will denote n-dimensional space and lattice
respectively, while SX and L,* will be their reciprocals. We will show that
the reciprocal lattice L: of an IMS, even if aperiodic in ST, may be
+
transformed into a (3 d)-dimensional periodic lattice (say LT+d) by
embedding the IMS in a metrical (3 + d)-dimensional space S3+,.
Let us suppose that the reciprocal vector H of a main or of a satellite
reflection may be written in S: as
where
and kf, k:, k: are the a * , b*, c* components of Ki. One of them should be
irrational for incommensurate modulation. Since (3 + d) indices hi are
necessary to label a single diffraction effect, we will speak of d-dimensional
periodic modulation or of d-dimensional modulated structure. In Fig. 3.E.1
a section of the three-dimensional diffraction pattern of a one-dimensional
modulated structure is sketched. The main reflections are marked by the Fig. 3.E.l. One-dimensional modulated
largest spots. structure: sketch of a section of the three-
A L:+d lattice may now be with basis vectors dimensional diffraction pattern, showing main
and satellite reflections.
bl = a * , b2 = b*, b3 = c*, b3+i= Ki + ei (i = 1, . . . , d) (3.E.2)
where the eis are unit vectors perpendicular to S g Then in S:+, a reciprocal
vector may be written as
0- a* -Q
W W M W W
ns a
v w L,
According to hypotheses, H is the projection of H' in ST. Consequently el

the whole three-dimensional diffraction pattern LT is the projection in ST of 0 0
+
a (3 d)-dimensional reciprocal lattice L,*+,. In Fig. 3.E.2 such a projection
is shown schematically for a one-dimensional modulation: Sz is represented
by a horizontal line, el is a unit vector perpendicular to ST, main and
Fig. 3.E.2. From SX to S:. Main and satellite
satellite reflections are labelled by M and W respectively. reflections are denoted by M and W
+
In S3+dthe (3 d)-dimensional direct lattice L3+dis defined by the set of respectively.
vectors ai reciprocal to bi according to the conditions ai . bj = 4,.

We have
where a, b, c are reciprocal to a * , b * , c * . The reciprocity condition is easily

verified: for example,
If a position is defined in S3 by
it will be defined in S3+, by
where
Indeed, the first three coordinates in S3 and in S3+dcoincide, owing to the

fact that L3 is the projection of L3+, along ei; any point r in S3 satisfies the
relations
In accordance with (3.E.3)
so that
x3+, = k{xl + k h 2 + k&x3
is obtained, which coincides with (3.E.4).
If in S3+, the new coordinates
are introduced, it may be concluded that S3 is defined by

tj=O for j = l , . . . , d.
The electron density p ' in S3+dis periodic in (3 + d)-dimensions and may
be calculated in the usual way:
1
p(xl, . . . , ~ 3 + d=) - 2 FH,exp
hl,...,h~+,~ j=1
where V is the volume of the unit cell in S3.

According to reciprocity rule (the Fourier transform of a projection
corresponds to a section and vice versa) the electron density distribution p
in S3 is the section of p ' with S3. Therefore in S3+dthe shape of the atoms is
a hypersurface which extends in the extra dimensions. In Fig. 3.E.3, a
schematic representation of p ' for a one-dimensional modulation is
described.
A scattering formalism for a large variety of modulated structures is now
a ~ a i l a b l e : [ ~ substitutional
~-~~I and/or displacive modulations of each of the
atoms in the crystal may be taken into account, together with the
translational or rotational displacement of a molecule or one of its segments
in molecular crystals.
Quasicrystals
In a famous paper by Shechtman, Blech, Gratias, and cahn17'] electron
diffraction patterns of a rapidly solidified A1-Mn alloy were shown which
suprisingly displayed five-fold symmetry. Sharpness of the spots suggested
long-range translational order, but the presence of the five-fold symmetry
violated the sacred rules of crystallography. In particular, by rotation of the
specimen, five-fold axes (in six directions), three-fold axes (in 10 directions)
and two-fold axes (in 15 directions) could be revealed: the subsequent Fig. 3.E.3. Schematic representation of the
ascertainment of the existence of an inversion centre fixed, for this Al-Mn density p' i n S,. S, is the horizontal line: its
phase, the icosahedral point group m55. Somewhat later a large number of bulging parts represent real atoms. (a) Perfect
crystal; (b)density modulation; (c) displacive
alloys with 'forbidden' symmetries were found: such kinds of materials modulation.
(providing electron diffraction patterns displaying sharp peaks and forbid-
den symmetry such as icosahedral, octogonal, decagonal, dodecagonal, etc.)
are called quasicrystals. A huge amount of theoretical and experimental
publications are now available: for a more comprehensive treatment of the
subject and for relevant literature the reader is referred to three excellent
review~.[~+'~]
A useful premise to quasicrystals (as well as to the IMSs) is the definition
of quasiperiodic functions. It is well known that a periodic function
+
f (x) =f (x 1) may be uniformly approximated by finite sums of functions
exp (2ninx). Accordingly, in a p -dimensional space
f '(XI, . . . , x,) = q(nl, . . . , n,) exp (2ni(nlxl + . . . + n,x,)
nl, ..., np
is a periodic function. However, its 'projection' on one-dimensional space
f (x) = C...,
nl, np
q h , . . . , n,) exp (2ni(nlv1 + . . . + npvP)x) (3.E.6)
obtained by fixing in f ' xj to v,x is not periodic if some of the vjs are
irrational. Functions which may be uniformly approximated by finite sums
+
of functions exp [2ni(n,v1x + n,v2x2 . . . + n,v,x)] are quasiperiodic func-
tions. We take three simple examples.
1. Let f '(x,, x,) = A 1 sin 2nx1 + A 2 sin 2nx2.
If we assume X, = ax1 where cu is an irrational number then
+
f (x) = A, sin 2nx1 A2 sin 2ncux,
is not periodic.
2, f (x) is the superposition of a periodic sequence of large open circles,
schematically represented by C, 6(x -nu), and a periodic sequence of
small solid circles represented by C, G(x - rnaz12). Here, z =
2 cos (1615) = (1 + f i ) / 2 = 1.618034. . . is the golden mean, n and m are
any integers. Since a and az/2 are incommensurate numbers the
structure is not periodic (see Fig. 3.E.4, first line) but the diffraction
pattern will display delta peaks owing to the fact that the order is
perfectly maintained at long distances. The existence of delta peaks may
be demonstrated by examining periodic approximations off (x). Since
successive approximations to t are
to which periodic structures of increasing periods a12, a, 3a, 5a can be

associated. It is easily seen from Fig. 3.E.4 that better periodic
approximations of the aperiodic sequence can be obtained by higher
order approximants.
3. In Fig. 3.E.5(a), a square lattice with a unit cell with edge length
(1 + z2)'I2 is drawn. An irrational direction is drawn, at an angle of
arctan(l1z) with the cell edge. All the lattice points contained in a band
+
parallel to such a direction and with width (1 z) are projected on it
(strip-projection method), giving rise to a non-periodic pattern (the
numerical sequence is 0, 1, 2.618, 3.618, 5.236, 6.854, 7.854, 9.472,
10.472, etc.). All intervals in the projection will have length either 1 or z
(L and S respectively), in the sequence SLSLLSLSLLSLLSLS.. .
(Fibonacci sequence). The chain may also be written
x, = n ( 3 - z) - ( z - 1) frac(nz),
Fig. 3.E.4. An aperiodic long-range ordered

0 0 *o o* 0. 0 0 *o* o* '63
structure and its periodic approximants. I 5a I
Fig. 3.E.5. (a) Projection of a band i n a two-

dimensional lattice t o obtain a non-periodic
ordered sequence (Fibonacci sequence). (b)The
Fibonacci sequence results from a cut of the
disconnected 'line atoms' of the two-
dimensional crystal with the real space S,.
where frac (x) is the fractional part of x. Since there is a weighted

average lattice with constant a = (3 - z) the sequence may be considered
as a modulated structure with modulation factor g(x) = ( t - 1) frac (x),
which may be embedded in a two-dimensional space.
The above three examples suggest that aperiodic sequences in one
dimension can be obtained by projection of periodic sequences in two
dimensions. The procedure for embedding a one-dimensional aperiodic
sequence in a two-dimensional space may be performed in accordance with
the previous section. For the last example we can choose al .=(z, -1) and
az = (1, z) (see Fig. 3.ES(b)): then the reciprocal average lattice is also
square with basis vectors
In order to obtain the quasiperiodic one-dimensional sequence on S1 (see

again Fig. 3.E.5(b)) one has to attach parallel line elements in the lattice
vertices. Thus the real one-dimensional quasicrystal results from a cut of the
disconnected 'line atoms' of the two-dimensional crystal with the physical
space S,.
The procedure may be generalized to the n-dimensional case: the
dimension of the space S,, in which a lattice with translational periodicity is
obtained is determined by the number of n rationally independent recipro-
cal basis vectors which are necessary to index the diffraction pattern. The
real aperiodic crystal is again the cut (in the direct space) of the
n-dimensional crystal with the physical space.
Well known two-dimensional examples of quasiperiodic structures are
Penrose ti ling^.[^^,^^] TWOexamples are shown in Fig. 3.E.6: in (a) tiling of
the plane is achieved by putting together two rhombic units in accordance
with some matching rules (without them the plane should be covered in a
periodic way). The pattern shows a five-fold rotation point. In (b) 'kites'
and 'darts' are used: no pentagonal symmetry is shown. ~ a c k a y [ first ~~]
showed that their Fourier transform satisfies a five-fold symmetry.
Octagonal, decagonal, and dodecagonal two-dimensional quasicrystals are
also known; all of them can be embedded in a periodic five-dimensional
space. The three-dimensional icosahedral lattice mentioned at the beginning
of this section may be embedded in a six-dimensional space.
The characteristics of the quasicrystals do not coincide with those of the
incommensurately modulated structures. While the latter show main and
satellite reflections, an average structure, and crystallographic point sym-
metry, the quasicrystals show one kind of reflection only, no average
structure, and non-crystallographic point symmetry.
References
1. Ewald, P. P. (1917). Annalen der Physik, 54, 519.
2. Laue, M. von (1931). Ergebnisse der exakten Naturwissenschaften, 10, 133.
3. Pinsker, 2. G. (1978). Dynamical scattering of x-rays in crystals. Springer,
Berlin.
4. Darwin, C. G. (1914). Philosophical Magazine, 27, 315.
5. Darwin, C. G. (1922). Philosophical Magazine, 43, 800.
7. Becker, P. J. and Coppens, P. (1974). Acta Crystallographica, A30, 129.
Fig. 3.E.6. (a) A two-dimensional quasilattice 8. Kato, N. (1976). Acta Crystallographica, A32, 453.
showing one five-fold rotation point (plane
symmetry 5mm). Basic oblate and prolate
9. Becker, P. J. (1982). In Computational crystallography, pp. 462-9. Clarendon,
rhombi with their matching rules are shown: Oxford.
similarly arrowed edges must fit. (b) Penrose 10. Chapuis, G., Templeton, D. H., and Templeton, L. K. (1985). Acta
tiling with kites and darts (it does not show Crystallographica, A41, 274.
five-fold symmetry).
11. Templeton, L. K., Templeton, D. H., Phizackerley, R. P., and Hodgson, K. 0.
(1982). Acta Crystallographica, A38, 74.
12. Phillips, J. C. and Hodgson, K. 0. (1980). Acta Crystallographica, A36, 856.
13. Bijvoet, J. M. (1949). Proceedings of the K. Nederlandse Akademie van
Wetenschappen, B52, 313.
14. Phillips, D. L. (1962). Journal of the Association for Computing Machinery, 9 ,
84.
15. Twomey, S. (1963). Journal of the Association for Computing Machinery, 10,97.
16. Kennett, T. J., Brewster, P. M., Prestwich, W. V., and Robertson, A. (1978).
Nuclear Instruments and Methods, 153, 125.
17. Kauppinen, J. K., Moffatt, D . J., Mantsch, H. H . , and Cameron, D. G. (1981).

Applied Spectroscopy, 35, 3, 271.
18. Wiedemann, K. E., Unnam, J., and Clark, R. K. (1987). Powder Diffraction, 2,
3, 130.
19. Toraya, H . (1988). Journal of Applied Crystallography, 21, 192.
20. Hamilton, W. C. (1959). Acta Crystallographica, 12, 609.
21. Schomaker, V. and Marsh, R. E. (1983). Acta Crystallographica, A39, 819.
22. Johnson, C. K. and Levy, H. A. (1974). In International tables for x-ray
crystallography, Vol. IV, pp. 311-36. Kynock Press, Birmingham.
23. Peterse, W. J. A. M. and Palm, J. H. (1966). Acta Crystallographica, 20, 147.
24. Lipscomb, W. N. (1949). Acta Crystallographica, 2, 193.
25. Colella, R. (1974). Acta Crystallographica, A30, 413.
26. Post, B. (1977). Physics Review Letters, 39, 760.
27. Chapman, L. D., Yoder, D. R., and Collella, R. (1981). Physics Review Letters,
46, 1578.
28. Chang, S. L. (1982). Acta Crystallographica, A38, 516.
29. Hummer, K. and Billy, H. (1982). Acta Crystallographica, A38, 841.
30. Juretscke, H. J. (1984). Acta Crystallographica, A42, 127.
31. Chang, S. L. (1984). Multiple diffraction of x-rays in crystals. Springer, Berlin.
32. Hummer, K. and Billy, H. (1986). Acta Crystallographica, A42, 127.
33. Ewald, P. P. (1917). Annalen der Physik, 54, 519.
34. Weckert, E. and Hummer, K. (1990). Acta Crystallographica, A46, 387.
35. Renninger, M. (1937). Zeitschrift fur Physik, 106, 141.
36. Hummer, K., Weckert, E., and Bondza, H. (1989). Acta Crystallographica,
A45, 182.
37. Mo, F., Hauback, B. C., and Thorkildsen, G. (1988). Acta Chemica
Scandinavica. Series A, 42, 130.
38. Guinier, A. (1963). X-ray diffraction. Freeman, San Francisco.
39. Hosemann, R. and Bacchi S. N. (1962). Direct analysis of diffraction matter.
North-Holland, Amsterdam.
40. Glatter, 0 . and Kratky, 0. (1982). Small angle x-ray scattering. Academic,
London.
41. Guinier, A. and Fournet, G. (1955). Small-angle scattering of x-rays. Wiley,
New York.
42. Debye, P. (1915). Annalen der Physik, 46, 809.
43. Ehrenfest, P. (1915). Proceedings of the Amsterdam Akademie, 17, 1132.
44. Zernicke, F. and Prins, J. A. (1927) Zeitschrift fur Physik, 41, 184.
45. Fournet, G. (1951). Transactions of the Faraday Society, 11, 121.
46. Hildebrandt, R. L. (1975). Molecular structure by diffraction methods, The
Chemical Society, special periodic report, No. 3 (ed. G. A. Sim and L. E.
Sutton).
47. Oberhammer, H. (1976). Molecular structure by diffraction methods, The
Chemical Society, special perodic report, No. 4 (ed. G. A. Sim and L. E.
Sutton).
48. Domenicano, A., Schultz, G., Hargittay, I., Colapietro, M., Portalone, G.,
George, P., and Bock, C. W. (1989). Structural Chemistry, 1, 107.
49. Kirkwood, J. G., Maun, E. K., and Alder, B. J. (1952). Journal of Chemical
Physics, 85, 777.
50. Morgan, I. and Warren, B. E. (1938). Journal of Chemical Physics, 6, 666.
51. Bernal, I. D. and Fowler, R. H. (1933). Journal of Chemical Physics, 1, 515.
52. Warren, B. E., Krutter, H., and Morningstar, 0. (1936). Journal of the
American Ceramic Society, 19, 202.
53. Guinier, A. (1939). Annales d e Physique, 12, 161.
54. Pilz, I., Goral, K., Hoyaerts, M., Lontie, R., and Witters, R. (1980). European
Journal of Biochemistry, 105, 539.
55. Mittelbach, P. (1964). Acta Physica Austriaca, 19, 53.

56. Kratky, 0. and Porod, G. (1953). In Die Physik der Hochpolymere, Vol. 11.
(ed. H. A . Stuart). Springer, Berlin.
57. Porod, G. (1951). Kolloid-Zeitschrift 124, 83.
58. Luzzati, V. (1960). Acta Crystallographica, 13, 939.
59. Stevens, E. D. and Coppens, P. (1976). Acta Crystallographica, A32, 915.
60. Hirshfeld, F. L. (1971). Acta Crystallographica, B27, 769.
61. Stewart, R. F. (1976). Acta Crystallographica, A32, 565.
62. Hansen, N. K. and Coppens, P. (1978). Acta Crystallographica, A 3 4 909.
63. Kurki-Suonio, K. (1977). Israel Journal of Chemistry, 16, 115.
64. Hehre, W. J., Stewart, R. F., and Pople, J. A. (1969). Journal of Chemical
Physics, 51, 2657.
65. Stewart, R. F. (1973). Journal of Chemical Physics, 58, 4430.
66. Epstein, J., Ruble, J. R., and Craven, B. M. (1982). Acta Crystallographica,
B38, 140-9.
67. Rees, B. (1977). Israel Journal of Chemistry, 16, 180.
68. Hirshfeld, F. L. (1977). Israel Journal of Chemistry, 16, 198.
69. Coppens, P. (1975). Physical Review Letters, 35, 98.
70. de Wolff, P. M. (1974). Acta Crystallographica, A30, 777.
71. Janner, A. and Janssen, T. (1977). Physical Review, B15, 643.
72. Janner, A. and Janssen, T. (1980). Acta Crystallographica, A36, 399.
73. Janner, A. and Janssen, T. (1980). Acta Crystallographica, A36, 408.
74. de Wolff, P. M., Janssen, T., and Janner, A. (1981). Acta Crystallographica,
A37, 625.
75. Yamamoto, A. (1982). Acta Crystallographica, A38, 87.
76. Petricek, V., Coppens, P., and Becker, P. (1985). Acta Crystallographica, A41,
478.
77. Petricek, V. and Coppens, P. (1988). Acta Crystallographica, A 4 4 1051.
78. Shechtman, D., Blech, I., Gratias, D., and Cahn, J. W. (1984). Physical Review
Letters, 53, 1951.
79. Janot, Ch. and Dubois, J. M. (1988). Journal of Physics F : Metal Physics, 18,
2303.
80. Janssen, T. (1988). Physics Reports, 168, 55.
81. Steurer, W. (1990). Zeitschrift fur Kristallographie, 190, 179.
82. Duneau, M. and Katz, A. (1985). Physical Review Letters, 54, 2688.
83. Janssen, T. (1986). Acta Crystallographica, A42, 261.
84. Penrose, R. (1974). Bulletin of the Institute of Applied Mathematics, 10, 266.
85. Gardner, M. (1977). Scientijic American, 236, 110.
86. Mackay, A. L. (1982). Physica, A114, 609.
Experimental methods
in X-ray crystallography
H U G O L. M O N A C O
Introduction
This chapter discusses the experimental methods used to study the
diffraction of X-rays by crystalline materials. Although, as seen in Appendix
3.B (pp. 195 and 198), electrons and neutrons are also diffracted by crystals,
we will concentrate our attention on X-ray diffraction. We begin discussing
how X-rays are produced and how one can define the beam of radiation that
will interact with the crystalline sample. The specimens that will receive our
attention are single crystals and polycrystalline materials, that is aggregates
of a very large number of very small crystals. We discuss the methods used
to record the diffraction pattern and to measure the intensities of the X-rays
scattered by these two types of specimen in separate sections. The ultimate
goal of extracting structure factor amplitudes from diffracted intensities
requires the application of a series of correction factors. This process, called
data reduction, is discussed in the final section of the chapter.
X-ray sources
Conventional generators
All the standard laboratory sources normally used for X-ray diffraction
experiments generate radiation using the same physical principles but can
vary substantially in their construction details. The two types of conven-
tional generators that are used in conjunction with the data recording
devices discussed on pp. 245 and 287 are sealed-tube and rotating-anode
generators. Most of the techniques used for diffraction data collection
require monochromatic radiation. Due to the way in which radiation is
produced in the conventional generators, only a discrete number of possible
wavelengths can be selected for experimental use. This limited choice and
the difference in intensity are two of the major differences between this type
of radiation and that generated by synchrotrons.
The origin of X-rays in the conventional sources

X-rays are produced when a beam of electrons, accelerated by a high
voltage, strikes a metal target and is therefore rapidly decelerated by
collision with the metal atoms. Most of the electrons do not lose their
230 1 Hugo L. Monaco
energy in a single collision but do it gradually through multiple events. The

result is the production of a continuous spectrum of X-rays called white
radiation. If all of the energy carried by an electron is transformed into
radiation, the energy of an X-ray photon is
Emax= hvma,= eV
where h is Planck's constant, vmaxthe photon frequency and the subscript
max indicates that this is the maximum possible energy, e is the charge of
the electron, and V the accelerating potential.
If we substitute the frequency in terms of the wavelength
and
where the potential V has to be measured in volts and the minimum

wavelength is obtained in angstroms. This equation shows that there is a
minimum value for the wavelength of the X-rays that can be obtained by
this process which is a function of the voltage accelerating the electrons.
Total conversion of the electron energy into radiation is not a highly
probable event and therefore the radiation of the highest intensity is
obtained at a somewhat longer wavelength. Figure 4.1 shows some
continuous X-ray spectra as a function of the accelerating voltage. Notice
0 0.2 0.4 0.6 0.8 1.0
that as the voltage is increased both the minimum wavelength and the
~avelength(i)
position of the intensity maximum shift to the left. The intensity maximum
Fig. 4.1. X-ray white radiation spectra as a
function of the accelerating voltage.
is found at a wavelength which is approximately 1.5 times Amin.
The total X-ray intensity generated per second in this way, a quantity
which is proportional to the areas under the curves, can be calculated using
the equation
where A is a proportionality constant, i the electrical current, a measure of

the number of electrons which are generating X-rays, Z is the atomic
number of the target, V the accelerating potential, and n a constant with a
value of about 2.
When the energy of the electrons striking the target is higher than a
certain threshold value, a second type of spectrum, discontinuous and with
very sharp lines, appears superimposed on the white radiation curves just
described. This second spectrum is called the characteristic spectrum
because its peaks are found at precisely defined wavelengths which depend
on the material constituting the target. The characteristic spectrum of
copper is shown in Fig. 4.2. The electrons with an energy above the
threshold potential are capable of ionizing the target atoms by ejecting an
electron from one of the inner shells. When that happens another electron
from a higher atomic energy level can move in to fill the vacancy created
and, since the new level has a lower energy, emit the energy difference as a
A
0.2 0.5 1.0 1.5 2.0 characteristic X-ray photon whose wavelength depends on the difference in
Wavelength (ii) energy between the two levels involved.
Fig. 4.2. Characteristic spectrum of copper The characteristic lines of this type of spectrum are called K, L, and M
superimposed on the white radiation spectrum.
Notice the ratio of the relative intensities of the and correspond to transitions from higher energy orbitals to the K, L, and
K, and Kg lines. M orbitals, that is to the orbitals of principal quantum numbers n = 1, 2,
Experimental methods in X-ray crystallography 1 231
and 3. When the two orbitals involved in the transition are adjacent the line
is called a,if they are separated by another shell, the line is called P. Thus,
the Cu K, line is produced by a copper target in which the atoms lost an
electron in the orbital of n = 1 and the vacancy was filled by an electron of
the orbital n = 2. The X-ray photon energy is the difference between these
two energy levels. Since for every principal quantum number n there are n
energy levels corresponding to the possible values of the quantum number 1
(from 0 to n - I), the a and /3 lines are actually split into multiple lines that
are very close to one another because the difference between these energy
levels is small. Still, X-ray radiation corresponding to all the possible energy
differences is not observed because some energy transitions are forbidden
by the selection rules. Thus, although Fig. 4.2 has a scale on the abscissa
which cannot show it, the Cu K, line is actually split into a doublet, the Km1
and K,, lines, of very similar wavelengths and which are, for this reason,
not easily separable.
The frequency of the characteristic line corresponding to a given
transition is related to the atomic number of the element that gave rise to it,
Z, by Moseley's law
Y = C(Z - a)2 (4.3)
where the constant C depends on the atomic energy levels involved in the
transition and the constant a takes into account the interactions with other
electrons. Thus, in a plot of Y " ~as a function of 2 for a given transition the
points corresponding to different target elements lie in a straight line and
different lines are obtained for the K,,, K,,, KB1, etc., transitions. The
characteristic frequency is higher the higher the atomic number and so the
Mo K, line ( 2 = 42) has a higher frequency and therefore a higher energy
than the Cu K, line ( 2 = 29). A full list of the wavelengths of the
characteristic lines of the elements which are used in X-ray diffraction
studies can be found in the International tables for x-ray crystallography.['1
Here, we will just point out that the two most frequently used lines are the
Cu K, line, 12 = 1.5418 A and the Mo K, line, 12 = 0.7107 A. Both are
doublets of slightly different wavelengths as pointed out before.
The intensity of a characteristic K line can be calculated using the
equation:
ZK = Bi(V - VK)'.' (4.4)
where B is a constant, i the electrical current, and VK the excitation
potential of the K series, a quantity which is proportional to the energy
required to remove a K electron from the target atom. It can be shown[21
that the ratio zK/zwis a maximum if the accelerating potential is chosen to
be V = 4VK. If this condition is fulfilled, the K, line is about 90 times more
intense than the white radiation of equal wavelength (I,). The K,, line is
approximately twice as intense as the K,, line and the ratio K,/KB depends
on Z but it averages 5 (see ~ieck'']).The data collection methods that use
monochromatic radiation discussed later all use K, radiation and therefore
require the elimination of the Kg component of the spectrum which is
always present. The methods used to achieve this are discussed on p. 241.
Sealed-tube and rotating-anode generators

A conventional generator consists of a high-voltage power supply with
electronic controls, connected to either a sealed tube or to the cathode of a
rotating-anode generator. In the second case there must also be a way of

making, keeping, and monitoring the high vacuum required for X-ray
generation. Just how high the voltage to be applied must be, can be
estimated from the considerations of the previous paragraphs and taking an
excitation voltage given by ~ i e c k . ' ' ] This value for the copper K energy
level is 8.981 kV so if one wishes to apply a voltage of 4 times V, about
36 kV have to be applied. For molybdenum even higher voltages are
required.
The sketch of a sealed X-ray tube is shown in Fig. 4.3(a). It consists of a
cathode with a filament that emits the electrons that are accelerated under
vacuum by the high voltage applied and hit the fixed anode made of the
metal whose characteristic spectrum has a K, line of a wavelength which is
adequate for the diffraction experiment to be performed. The high vacuum
is necessary because the presence of gas molecules in the tube decreases the
efficiency of the X-ray generating process by collisions with the electrons in
the beam.
The process of X-ray production discussed in the previous section is
highly inefficient: only about 0.1 per cent of the power applied is
transformed into X-rays, the rest being dissipated as heat. In order to avoid
melting of the target, it therefore becomes necessary to cool it, which is
done by circulating cold water as shown in the figure. It is the efficiency of
FIL
the cooling system that ultimately determines the maximum power that can
be applied to the tube. The problem of heat dissipation also dictates the
choice of the focal area of the beam of electrons on the anode. This area is
always chosen to be rectangular with the sides in a ratio of at least six to
one. In this way, when one looks at the focused electron beam on the
target, which determines the surface that produces X-rays, in the direction
of the longest side of the rectangle one can see, by choosing the appropriate
WATER
angle, a small square focus with a side equal to the smallest rectangle side.
At the same time, the real area that dissipates heat is several times that of a
square of that size. The X-rays generated come out of the tube through four
CATHODE
I
beryllium windows: the two that are shown in the figure, that are found in
FOCUSING
\
CUP
I FILAMENT
the direction of the longest axis of the rectangular focus, and another two in
a direction perpendicular to it. The first two windows are used for
single-crystal work, the others may be used when a linear focus is needed.
In rotating-anode generators, the target area seen by the electron beam is
continuously renovated because the anode is rotated. In this way, much
higher powers per unit of focal area can be applied to the unit and
consequently higher X-ray intensities can be produced. Figure 4.3(b) is a
sketch of a rotating anode chamber showing the two X-ray windows used
for single-crystal work, the filament, placed with its longest axis parallel to
-1
WATER
1 R OA
TAX TI SI O N
the direction of the windows, the focusing cup, used to focus the electron
beam on the target, and the anode which rotates about an axis parallel to
the direction of the windows.
In a rotating-anode generator, the chamber has to be kept under vacuum
for the reasons stated before, water has to be circulated to cool the anode as
is done in the case of a sealed tube, and of course the anode has to be
Fig. 4.3. (a) A sealed X-ray tube. (b)Sketch of a rotated at a certain speed. The system is therefore mechanically much more
rotating-anode chamber. The path followed by complicated than a tube and it requires a fairly frequent and demanding
the water that cools the anode is not shown in
the figure. The anode is cylindrical and rotates maintenance work. Figure 4.4 is a photograph of a modern rotating anode
about the axis shown. X-ray generator. ~ h i l l i p s ' ~gives
] a detailed discussion of the advantages of
Fig. 4.4. A rotating-anode X-ray generator.

(Photograph courtesy of Enraf Nonius.)
rotating-anode generators together with an exhaustive description of the
practical problems encountered in their operation.
Choice of the type of radiation

As discussed above, the K, lines are those of highest intensity and therefore
they are the ones that are normally used for standard X-ray diffraction
work. Thus, in practice, the choice of the wavelength of the X-rays
generated by conventional equipment is limited to the values of the
characteristic K, lines of the metals commonly used as targets. Two of these
metals are most frequently used: copper and molybdenum; their K,
transitions generate radiation with a wavelength of about 1.5 and 0.7A
respectively. As will be seen later, for a given crystal-to-detector distance
and unit cell, the diffracted beams will be more separated on the detector if
a longer wavelength is used. Thus, copper is used for macromolecular work
in which one usually was large unit cell parameters and for the structure
determination of organic molecules, which do not contain atoms that absorb
this radiation strongly. Absorption of the radiation by the sample is
therefore also a primary consideration in the selection of the wavelength to
be used.
Another consideration is the maximum resolution of the reflections that
will be recorded. We saw on p. 155 that there is a limiting sphere of radius
2 / A that limits the volume of reciprocal space accessible to diffraction

experiments. This sphere obviously has a different size for the two types of
radiation discussed and therefore molybdenum radiation may be required to
record data to a resolution not accessible to copper radiation.
Finally we mention the detection efficiency of the method that will be
used to record the diffracted intensities. Film is a better detector for copper
than for molybdenum radiation and some area detectors have been
optimized for use with copper radiation. On the other hand, diffractometer
counters have a very high counting efficiency for the Mo K, radiation which
explains why this type of radiation is so widely used for small-molecule
single-crystal X-ray diffraction work.
Synchrotron radiation
X-rays, as well as other types of electromagnetic radiation, can also be
generated by sources known as synchrotron radiation facilities. In these
installations either electrons or positrons are accelerated at relativistic
velocities along orbits of very large radii, several metres or even hundreds
of metres. These sources are, by necessity, very complex and those that
produce suitable X-rays are limited, located mainly in the United States and
Europe. However, since, as we will see, the X-rays they produce are in
many ways much better than those generated by conventional sources, their
use has grown steadily in the crystallographic community. As a result, more
beam time has been made available to crystallographers and more syn-
chrotron sources are planned for construction in different countries. Among
those, the 6 GeV storage ring to be built in the USA[~]and the European
Synchrotron Radiation ~ a c i l i t ~ designed
,[~] specifically to produce the best
possible X-rays, promise to be of special importance is the development of
this field.
From the extensive literature that exists in this ever expanding field we
recommend two very elementary description^,[^^'] an introductory
textbook,['] and a more advanced treatise in several volumes.[g]In this last
treatise chapters 1, 2, and 11 of volume 1 are specially relevant to our
discussion; volume 3 of the series is totally devoted to X-ray methods.
Generation of X-rays in a synchrotron radiation source

It is well known that charged particles moving under the influence of an
accelerating field emit electromagnetic radiation. The energy of this
radiation is dependent on the velocity of the particle. If the velocity is like
that of the electrons moving in an antenna, emission takes place in the
radio-frequency range but when the charged particles, electrons, or posi-
trons, move with a speed approaching that of light, the spectrum extends
into higher energy regions covering the X-ray range. An additional
consequence of relativistic effects is a distortion of the angular distribution
Fig. 4.5. Outline of a typical synchrotron of the radiation. When the charged particles move at velocities approaching
radiation facility. The electrons or positrons are that of light, radiation is emitted in a very narrow cone parallel to the
accelerated in the linear accelerator (linac) and
then in the booster t o be finally injected into the
instantaneous velocity. In a synchrotron source the charged particles are
storage ring where they are kept circulating for made to move in closed trajectories, often circular or elliptical, and so the
periods of several hours. The beam lines, not radiation is generated in cones tangent to the path followed by the particles
shown i n the figure, are tangential t o the particle
trajectory. Notice the scale at the bottom of the (see for example chapter 1 of ~ o c h [ ~ ] ) .
picture corresponding t o 0-50 m . Figure 4.5 shows the essential elements of a synchrotron radiation source.
Notice the scale at the bottom which gives an idea of the size of this type of
facility. The basic element of the installation from which radiation is
generated is the storage ring, a toroidal cavity in which the charged particles
are kept circulating under vacuum. An extremely high vacuum is required
or else the particles are lost by collision with the atoms present in the cavity.
Prior to injection into the storage ring, the particles must be accelerated, for
example, first by a linear accelerator and then by a booster as shown in the
figure. The other two elements that are essential to the operation of the ring
are the so-called lattice, that is the set of magnets which force the particles
to follow a closed trajectory as well as performing other functions, and the
radio-frequency cavity system which restores to the particles the energy they
lose as synchrotron radiation. The beam lines, not shown in the figure, are
tangential to the storage ring.
It can be shown that a charged particle moving along a circular orbit emits
as electromagnetic radiation the following power[8*9]
where P is the energy emitted per unit time, e is the particle charge, c the
speed of light, E the energy of the particle, mo its mass at rest, and R the
bending radius of the orbit. This equation explains why high-energy
particles are required and also why only electrons or positrons are used. The
power emitted by heavier particles such as protons is too low to be of
significant importance.
The quantity y, the ratio of the total energy to the rest energy of the
particle, is of considerable importance since it is approximately related to
the opening angle of the cone of radiation by
where AI+!J is the emission angle in radians. A more exact relationship

between these two parameters is given in reference [lo].
The total power emitted by the ring is the power emitted by a particle in
one revolution multiplied by the number of particles and divided by the
time it takes them to complete a revolution.
The total power can be shown to be equal to[8391
P,,, = 2 6 . 6 ~ ~ ~ i (4.7)
where the energy of the particles is measured in GeV, the magnetic field B
in tesla, the current i in amperes and the power is obtained in kW. The
power is thus seen to be directly proportional to the current in the storage
ring.
The radiation emitted at a modern storage ring comes from two sources:
the bending magnets and the insertion devices. The first is the radiation we
have discussed so far, the second is generated by special devices called
wigglers and undulators which are briefly discussed below.
An important property of the radiation generated by bending magnets is
its wide spectral distribution which can be described quantitatively in terms
of the spectral flux N which is the number of photons emitted per unit time
interval into a relative band width AA/A into an angle element d e in the
plane of the electron orbit and integrated in the vertical plane. It can be
shown that the flux of radiation normalized to the ring current generated by
bending magnets is equal toLs1

N(hv) = 1.256 x lo7 y Gl(y) photons s-' mrad-I m ~ - '
(0.1% band width) (4.8)
where the factor Gl(y) is an energy-dependent function that can be found
tabulated for example in the book by ~ a r ~ a r i t o n d o .The
[ ~ ] variable y is
defined as follows
where E, is the critical energy associated with a magnetic field B which in

keV is given by
where E is measured in GeV and R, the radius of the bending magnet in

metres. An alternative expression for E , is
with B measured in tesla.

A widely used parameter related to the critical photon energy is the
critical wavelength A,
E, = hv, = he/& = 12.4/Ac (4.11)
where E, is measured in keV and A, in A.
The spectra of total emitted flux for several synchrotron radiation sources
is shown in Fig. 4.6 taken from ~aterlik.["] The critical wavelength, A,, is
useful as an indicator of the suitability of a source for X-ray experiments.
As can be seen in the figure although it does not correspond to the
maximum flux it is close enough to give a good idea of the position of this
maximum. It can be shown that ideally the wavelength used at a
synchrotron source should fall in the range 0.25ACto 4.0Ac. Table 4.1, taken
from the literature,15] gives E,, A,, and other relevant parameters for the
WAVELENGTH (A)
124 12.4 124 0 12
'-
I I I I 1 I
Fig. 4.6. Spectral distribution of several storage

rings; a =ADONE (Frascati. Italy) (1.5 GeV,
105 mA); b = DCI (Orsay, France) (1.8 GeV,
250 mA); c = SRS (Daresbury, UK) (2.0 GeV,
685 mA); d = NSLS (Brookhaven, USA) (2.5 GeV,
500 mA); e = DORIS (Hamburg, Germany)
(5.0 Gev, 100 mA); f = ESRF (Grenoble, France)
(5.0 GeV, 565 mA). The critical wavelengths are
indicated by arrows, notice their position
relative t o the maxima. Recall that the photon
energy of the Cu K, line corresponds to about
8 keV and that of the M o K, line t o about 17 keV.
HOR = horizontal plane. (Figure taken from
~aterlik."") PHOTON ENERGY ( K e V )
Table 4.1. Relevant parameters of the Synchrotron Radiation Sources i n operation in 1987 (taken from reference 5)
Name Location E(GeV) R(m) i(mA) EAkeV) 1,I.J.) Emittance Insertion

( x 10' m rad) (devices)
Group I, E, s 0.06 keV

N-100 Karkhov, USSR
Surf II Washington, USA
Tantalus I Wisconsin, USA
Group II, E, 0.06-2 keV
Sor ring Tokyo, Japan

Siberia I Moscow, USSR
Max Lund, Sweden
COSY Berlin, Germany
Teras Tsukuba, Japan
ACO Orsay, France
NSLS Brookhaven, USA
UV SOR Okazaki, Japan
Fian C-60 Moscow, USSR
Vepp-2M Novosibirsk, USSR
Hesyrl Hefei, China
Superaco Orsay, France
SPRL Stanford, USA
Bessy Berlin, Germany
Aladdin Wisconsin, USA
INS-ES Tokyo, Japan
Pakhara Moscow, USSR
Sirius Tomsk, USSR
Adone Frascati, Italy
Group Ill, E, 2-30 keV
DCI Orsay, France
SRS Daresbury, UK
Vepp-3 Novosibirsk, USSR
Photon Factory Tsukuba, Japan
NSLS Brookhaven, USA
Bonn Bonn, Germany
Siberia II Moscow, USSR
BEPC Beijing, China
Elsa Bonn, Germany
Arus Erevan, USSR
Spear Stanford, USA
Doris II Hamburg, Germany
Tristan ACC Tsukuba, Japan
Group IV, E, :
,30 keV
CESR Ithaca, USA

Vepp-4 Novosibirsk, USSR
Petra Hamburg, Germany
PEP Stanford, USA
Tristan Tsukuba, Japan
synchrotron sources in operation in 1987. If the spectral flux is transformed

into power, i.e. the number of photons is multiplied by the energy of a
photon hv, it can be shown that half the total power is irradiated below and
half above the critical value A,.
Other parameters of interest are the source size and the divergence of the
irradiated beam. The charged particle beam in the storage ring has a
Gaussian profile characterized by the parameter ox, in the horizontal and o,
in the vertical plane. The full width at half maximum of the beam can then
be calculated as 2.350 and the source area F is estimated as
Similarly the angular distribution is characterized by the parameters ox,

and a,, and the solid angle of emission is then estimated as
Although a,, a,. and a,, a,, vary along the orbit their variations are
correlated and it is thus useful to define another parameter, the emittance,
which at special symmetry positions is found to be
The emittance is instead a constant along the charged particle path and it is
thus another important parameter characteristic of an installation. The
emittances of the synchrotron sources in operation in 1987 are also shown in
Table 4.1.
Another useful function, which is often used to compare the potential
performance of two sources, is the spectral brightness, also called spectral
brillian~e,'~]
defined as the number of photons emitted per unit area of the
source at point x, z over a 0.1 per cent relative band width per unit solid
angle dS2 and unit time in the direction defined by the angles 3 (defined by
the instantaneous velocity of the charged particle and the projection of the
direction of observation onto the vertical plane) and 19 (defined by the
projection onto the horizontal plane instead). It can be shown that
the spectral brightness b is equal to (see ~ a r ~ a r i t o n d ochapter
[ ~ ] 2)
where N is the spectral flux. If one defines the central brightness b, which is
the brightness for x = z = q = 0 it is obvious that
From this equation it can be seen that the brightness can be increased by
increasing the flux or by decreasing the as. Decreasing the a s can be
accomplished by reducing the emittance of the storage ring (eqn (4.12)).
The emittance of a ring is thus a fundamental parameter to be taken into
account in comparing the expected performance from two different sources.
A radiation spectrum quite different from that produced by bending
magnets can be obtained by the use of insertion devices. These are a series
of periodically spaced magnets of alternating polarity which are inserted in a
straight region of the ring and which do not alter the ideal closed orbit of
the particles in the storage ring. Most insertion devices create a sinusoidal
magnetic field which forces the charged particles to oscillate around the
mean orbit. According to their characteristics they are called wigglers or
undulators.
The parameter that has to be examined to determine whether an insertion
device is a wiggler or an undulator is the K parameter
where a is the maximum deflection angle of the electron or positron

trajectory, Bo is the oscillating magnetic field measured in tesla and A, is the
period of the magnetic array measured in centimetres. Recalling that l l y is
approximately equal to the natural opening angle AI), K becomes the ratio
between the maximum deflection angle of the electron trajectory along the
insertion device, a , and AI).
When the parameter K >> 1 the device is called a wiggler. The effect of a
wiggler on the emitted spectrum is to shift the critical energy E, to higher
values and to increase the intensity of the radiation by a factor proportional
to the number of periods of the magnetic array. No interference effects are
observed and so the emitted spectrum is qualitatively very similar to that
obtained from bending magnets.
If the parameter K << 1 the device is called an undulator and interference
occurs between radiation which is emitted by the same electron at different
points in its path through the device. As a result, radiation is emitted as a
series of relatively sharp peaks whose wavelengths are given by the
equation[l01
where 8 is the angle at which radiation is emitted and j is the harmonic

number. As is evident from this equation, the wavelength of the peaks can
be shifted by changing the parameters that appear in parentheses, 8 the
angle of observation and a which, as we have seen, is related to the
magnetic field of the device Bo. The intensity which is produced by an
undulator with N poles at 8 = 0 is amplified by a factor proportional
to N2.[lo]
Comparison of synchrotron and conventionally generated

X-rays
We have seen on p. 233 that in practice only the characteristic lines of
copper and molybdenum are sufficiently intense for use in crystallographic
X-ray work and thus with conventional sources one is limited to only a
discrete number of possible wavelengths. Synchrotron radiation does not
have this limitation as should be clear from Fig. 4.6. The spectrum emitted
is very wide and in fact extends in both directions far beyond the range
spanned by the Cu K, and Mo K, lines. The possibility of selecting the
wavelength for X-ray diffraction work has not been fully exploited although
it is clear that it has far-reaching consequences. It can be used for example
to solve the phase problem in macromolecular work (see Chapter 8, p. 544)
by measuring anomalous dispersion effects (see for example Hendrikson et
al.[ll]). Furthermore, crystal absorption and radiation damage (see pp. 304
and 308) are wavelength dependent and thus can be minimized by
optimixing the X-ray wavelength used.[12]
Probably the best known property of synchrotron radiation is its high
brightness, resulting from the small cross-section of the charged particle
beam and the high degree of collimation of the radiation.
A detailed comparison of the brightness and other properties of X-rays
produced by synchrotron sources and sealed-tube and rotating-anode
generators can be found in ~ o n s e . [ ' ~
In] Fig. 4.7, taken from ~ i s e n b e r g e r , ~ ~ ]
Fig. 4.7. The brightness of x-ray sources as a

function of the year in which they became
available. The points correspond to the
following sources: A = X-ray tube; B = rotating.
anode generator; C = Stanford Synchrotron
Radiation Laboratory (SSRL) bending magnet;
D = Brookhaven National Synchrotron Light
Source (NSLS) bending magnet; E = SSRL 54
pole wiggler; F = 6 GeV source with undulator. 1900 1940 1980
(Figure taken from ~ i s e n b e r ~ e r . ' ~ ' ) Year
the brightness of X-ray sources is plotted as a function of the year in which

they became available. In the figure one can see that radiation from bending
magnets, which, as we have seen, are the least intense of the possible
synchrotron sources, can be lo5 times more intense than that from a
rotating-anode generator. As a result, data collection at a synchrotron
source is faster, produces higher-resolution data, and permits single-crystal
work with specimens that can be only a few pm3 in volume.[141A sample of
such a small size can only be treated as a polycrystalline material with a
conventional source.
An important property of synchrotron radiation is that it is completely
linearly polarized in the plane of the orbit of the charged particles and
elliptically above and below this plane.[g1X-rays generated from conven-
tional sources are totally non-polarized. As a result diffracted intensities
have to be corrected for this polarization effect in different ways as we will
see later (p. 303).
We mentioned before that the energy emitted by the moving charged
particles is replenished by a radio-frequency cavity. Only the particles with
an adequate phase relation with respect to the radio-frequency field can
keep a stable orbit and they end up gathered in bunches which have a length
that is dependent on the radio frequency used. The number of bunches of
charged particles circulating around the orbit is an integer and can be
regulated by varying the parameters of operation. Thus, synchrotron
radiation is, in fact, produced at a given point in the orbit as a flash when a
bunch passes through that point. As a result, synchrotron radiation has a
well defined time structure, that is, pulses of radiation are emitted at
perfectly defined time intervals. For example the European Synchrotron

Radiation Facility is planned to produce pulses lasting from 65 to 140 ps and
the pulses will be separated by a minimum of 3 ns.['] This property of
synchrotron radiation, of clear importance in time-resolved studies, has
received so far very little attention in X-ray diffraction work.
Monochromatization, collimation, and focusing of X-rays

We have seen that conventional X-ray sources generate the discrete lines of
the characteristic spectrum of the anode superimposed on the white
radiation continuum and that radiation is emitted in every possible
direction, and that although synchrotron sources emit a spatially confined
narrow beam, the wavelength of this radiation spans a very wide continuous
spectrum. All of the data collection methods discussed later require that a
narrow pencil of X-rays strike the specimen under examination and, in
addition, most of them also require that the energy of the radiation be
limited to a wavelength band as narrow as possible. Ideally the radiation
should consist of photons of only a single wavelength. Here we discuss the
methods used to select a narrow wavelength out of the spectrum generated
by the source and how to define a narrow parallel beam of X-rays to be used
for diffraction experiments in conjunction with the data collection devices
discussed later.
Filters
One way to select a wavelength interval out of the spectrum generated by
the source is by filtering the radiation through a material that selectively
absorbs the unwanted radiation while letting through most of the photons of
the wavelength that will be used for the diffraction experiment. The
absorption of X-rays by a material follows Beer's law:
where I is the transmitted intensity, I,, the incident intensity, x the distance
travelled by the X-rays in the material, i.e. the thickness of the filter, and p
is the linear absorption coefficient which depends on the substance, its
density, and the wavelength of the X-rays. Since p depends on the density
of the material, the quantity that is usually tabulated is pm = p l p , the mass
absorption coefficient, a characteristic of the substance that depends only on
the wavelength considered. Complete tables of p , as a function of the
wavelength for different materials used as filters can be found in Koch and
~ a c ~ i l l a v r ~and ~ of pm versus A, for nickel is shown in Fig. 4.8
, [ ~a ' plot
along with the radiation spectrum generated by a copper anode. In the
figure it can be seen that the curve of pm versus A, shows two continuous
branches separated by a sharp discontinuity, called the absorption edge. If
the filter is a pure element, the continuous parts of the curve follow
approximately the equation
pm = kz3A3 (4.17)
where k is a constant with different values for the two branches of the curve
and Z is the atomic number of the element. This equation shows why harder
X-rays, i.e. X-rays with a shorter wavelength, are absorbed less than those
Fig. 4.8. The broken line represents the variation WHITE RADIATION
of the mass absorption coefficient p, as a
function of the wavelength for nickel. The
continuous line is the X-ray spectrum generated
by a copper anode. Notice that the absorption
edge of nickel falls in between the K, and the KB
characteristic lines of copper.
with a longer wavelength. The presence of the absorption edge in the curve
is explained by the fact that the photons at the edge have the wavelength
corresponding to the energy necessary to eject an electron from the K
orbital of the atoms of the filter. Thus, when this energy is reached massive
absorption of radiation occurs with photoionization of the filter and
production of fluorescent radiation.
Similarly to the displacement of the position of the characteristic lines
with Z, absorption edges move to shorter wavelength as the atomic number
of the element increases. A common single filter is chosen so that its
absorption edge falls in between the K, and the Kg peaks of the anode that
has been used to generate the X-rays. In this way, the unwanted radiation
of highest intensity, i.e. the unavoidable Kg peak, can be greatly attenuated
without reducing too much the intensity of the K, peak that will be used for
the experiment. Figure 4.8 shows that a nickel filter has its absorption edge
at the wavelength necessary to very strongly absorb the copper Kg peak.
Since copper has Z = 29 and nickel Z = 28 there is a difference of one
between the atomic numbers of the target anode used to generate the
X-rays and the filter with an absorption edge falling in between its K, and
Kg peaks. This is generally true for every element with Z s 70 and for the
elements of the second long row of the periodic table it is also true that both
the elements of Z - 1 and Z - 2 can be used to absorb the Kg peak of the
anode with atomic number Z. Thus both Nb ( Z = 41) and Zr (2 = 40) can
be used as filters for Mo (2 = 42) radiation.
The relative intensities of the K, and the Kg peaks depend not only on
the absorption coefficient of the filter but also on its thickness. Roberts and
parrish[16' give a table of the appropriate filter thicknesses necessary to
produce K,/KB ratios of 100 and 500 for different elements used as targets
and filters. The same table gives also the percentage of K, peak lost by
filtering which can vary between about 40 and 70 per cent.
A variation of the simple filter technique is the Ross balanced-filter
method.[16]In this method two filters are used: one with its absorption edge
at slightly shorter and the other at slightly longer wavelength than the K,
peak selected. The thickness of the filters is chosen so that the radiation is
absorbed to the same extent except in the interval in between the two
absorption edges. With this technique two measurements are made with
either one of the two filters in position and the measured intensity is then
taken to be the difference between the two values obtained.
Crystal monochromators
An alternative and more selective way to produce a beam of X-rays with a
narrow wavelength distribution is by using a single-crystal monochromator.
Bragg's equation (3.32) shows that when radiation of different wave-
lengths impinges upon a crystal, diffracted beams are observed at scattering
angles 13that depend on the wavelength of the radiation A. Thus, selecting a
given diffraction angle 8 is equivalent to choosing a particular wavelength
out of the spectrum incident on the crystal.
The simplest type of crystal monochromator consists of a single crystal
with one face parallel to a major set of crystal planes mounted so that its
orientation with respect to the X-ray beam can be properly adjusted. The
most important properties of a crystal monochromator are:
(1) the crystal used must be mechanically strong and should be stable in the
X-ray beam;
(2) the interplanar distance should be in the appropriate range to allow the
selection of the desired wavelength at a reasonable scattering angle;
(3) the presence of one or more strong diffracted intensities that can be
chosen so that the intensity loss of the beam, which is always
appreciable, may be reduced as much as possible; and
(4) the mosaicity of the crystal, which determines the divergence of the
diffracted beam and the resolution of the crystal, should be mall.['^,^^]
The reflection chosen should also have a scattering angle as small as possible
in order to minimize the loss of intensity due to the polarization factor (see
p. 303). Roberts and parrish[16]give a table with the important properties of
crystals commonly used as monochromators.
In a variation of this simple type of flat monochromator, the crystal
surface is cut so that it forms an angle with the set of planes that diffract the
radiation. In this way, the diffracted beam has a smaller width and as a
result more photons are concentrated in a smaller cross section of the
bearn.[l6] By properly curving their surface, crystal monochromators can be
used to focus the X-ray beam in a very small area.[lglThe curvature of the
surface can be produced by simply bending the crystal, in which case the
diffracting planes should ideally be tangential to the curved surface. If the
monochromator is bent in the shape of a cylinder of elliptical section with
the source in one of the foci, the reflected radiation will concentrate on the
other focus of the ellipse. A further variation consists of not only bending
the crystal but also in grinding its surface so that the radius of curvature of
the diffracting planes of the crystal is different from that of its surface. The
advantage of this type of monochromator is that it does not suffer from
some optical aberrations present in singly bent crystals.[161Curved crystal
monochromators are frequently used to select the wavelength of syn-
chrotron radiation. In addition to the requirements stated before, the
244 I Hugo L. Monaco
crystals should have in this case a very small thermal expansion and a large
thermal conductivity because the power applied is much larger than in the
case of conventional sources.[lsl
Another type of monochromator of wide application in synchrotron
sources is the double-crystal monochromator in which the incident X-ray
beam is diffracted twice by two similar crystals. This type of monochromator
can be constructed with different geometries designed to improve the
resolution and/or to keep the X-ray beam in the original direction. A
discussion of this type of crystal monochromator can be found in
~ a r ~ a r i t o n d o . L Crystal
'~] monochromators are more selective than filters
and, in the case of conventional generators, are capable of resolving the Kal
and K,, doublet which cannot be separated by any filtering method.
Collimators
The function of collimators is to define a narrow cylindrical beam of X-rays
that ideally should be as parallel as possible.
A simple pinhole collimator is shown sketched in Fig. 4.9. It consists of a
cylinder with two apertures defining the beam and a third guard aperture
which does not affect the beam size defined by the other two but eliminates
the radiation scattered by the defining aperture furthest from the X-ray
source. These apertures are commonly circular, although slits can be used
instead in which case square or rectangular beams can be defined.
Cylindrical beam pinhole collimators are typically used with conventional
sources to define a beam of radiation that is monochromatized by either a
filter or a crystal monochromator. Such collimators never produce an ideally
parallel X-ray beam but, in addition to the parallel X-rays, they also
produce convergent and divergent X-rays as shown in the figure. A
conventional X-ray source, when viewed at the appropriate take-off angle is
seen as a square. If 1 is the distance between the two defining apertures S1
and S2 and d is the diameter of the collimator the maximum angle of
divergence of the beam, y, can be calculated as shown in the figure as
dl2
tan y/2 = -= dl1
112
and since the angle is very small
where y is calculated in radians.

If we substitute in this equation d and 1 for two reasonable values: 50 and
0.5 mm we can get an estimate for the maximum divergence y
y = 2 x 0.5150 = 2 x lop2radians.
~ u x l e ~and
[ ~more
~ ] recently Arndt and sweet[l7Ihave extensively discussed
the conditions that can be varied in order to optimize collimation. The
I I
Fig. 4.9. Pinhole collimator showing the angle of
greatest divergence y, S, and S, are the two
Id
apertures defining the beam which are I I
separated by the distance I, and have a diameter
of d; S, is the guard aperture. -
variables that can be adjusted are the crystal and X-ray focus size, the
crystal to focus distance, and the crystal to detector distance. Depending
also on the reflection to reflection resolution necessary for the experiment,
different conditions are found which maximize the signal to noise ratio given
the restrictions imposed by the experiment.
Mirrors
X-rays can be reflected by mirrors when the angle of incidence is smaller
than a certain critical angle 8, which is a function of the wavelength A and
can be calculated by the equation[17,191
where Z is the atomic number, p the density, and A the atomic weight of
the reflecting material. If the angle of incidence is chosen to be within ten
per cent of 8, for the Cu K, radiation, the X-rays having a wavelength
shorter than the corresponding A, in particular the fairly intense Kp peak,
will not be reflected and therefore they will be eliminated from the beam.
Thus, by properiy choosing the glancing angle of the X-rays on the mirror
the radiation can be partially monochromatized. The values of the critical
angles 8, depend on the reflecting material as shown by the equation given
above and they are, in general, very small: 14' for glass and 23' for nickel
mirrors, calculated for a wavelength corresponding to the Cu K, radiation.
A table of this property and other parameters of interest of mirrors can be
found in witz.[lgl
If the reflecting surface of the mirror is curved, ideally in the shape of an
elliptical cylinder with the source in one focus, the reflected radiation will
converge to the other focus of the ellipse and a very intense X-ray beam will
be obtained at that point.
This principle is used in the design of a very powerful device that is used
to focus and partially monochromatize an X-ray beam and which uses two
curved mirrors with perpendicular axes of curvature.[211This double-mirror
system has been used for X-ray diffraction work on virus crystals which,
having very large unit cell parameters, pose particularly serious problems
for the spatial separation of the very close diffracted beams.[221Mirrors are
also very extensively used in the beamlines of synchrotrons. In addition to
focusing and partial monochromatization they perform several other func-
tions: splitting of a beam into two, magnification or demagnification of the
source, and change in the polarization of the radiation.[l8I The function to
be performed determines their geometry and so their surface can be flat or
curved and the mirror can be bent or segmented, that is constituted by
several small flat pieces which are easier to produce than large curved single
mirrors.
Data collection techniques for single crystals

We saw in Chapter 3 (p. 155) that the condition for X-ray diffraction to
occur in a crystal is that the scattering vector r* should coincide with a node
of the reciprocal lattice associated to the crystalline lattice defined in
Chapter 2 (p. 63). The sphere of reflection, also called Ewald sphere, was
defined (on p. 155) and it was shown that diffraction is observed whenever a
reciprocal lattice node lies on this sphere. The direction of the diffracted
beam is determined by the vector joining the centre of the Ewald sphere
and the reciprocal lattice node lying on its surface. We saw that the
characteristics of the crystal and the conditions of the experiment cause
these nodes of reciprocal space to have a volume. We will make use of all
these concepts in our discussion of the data collection devices that will
follow.
If the scattering experiment is performed with radiation of a single
wavelength there will be a single Ewald sphere of radius 1/A and the
probability that a stationary reciprocal lattice node may by chance be on its
surface will be fairly low. Furthermore, and as pointed out on p. 163,
diffraction from only a cross-section of the node is not acceptable since it is
the entire volume that should give rise to the diffracted intensity from which
the structure factor amplitude is to be derived. In addition, in order to solve
a structure, one needs all of the diffracted intensities that can be measured
corresponding to the nodes found within a sphere of radius D* = l/R,,,,
the inverse of the resolution of the structure. Broadly speaking there are
two ways to tackle these problems: the first is to use polychromatic
radiation, i.e. to have a series of Ewald spheres corresponding to different
wavelengths; the second method is to move thc reciprocal lattice nodes, i.e.
the crystal, so that all the nodes from which one wishes to measure the
diffraction cross the Ewald sphere completely. The first method is histori-
cally the oldest, it is called the Laue method, and it was with it that
diffraction from crystals was discovered.[23]
In Fig. 4.10, the shaded area represents the volume of reciprocal space
containing all the nodes that will produce diffraction when the stationary
specimen is hit with radiation of a wavelength in the interval between A,,
and A,;,. In the Laue method, different cross-sections of a node are excited
by radiation of slightly different wavelengths and the result is an intensity
integrated over the wavelength rather than over the volume sweeping
through a single Ewald sphere. If the white radiation spectrum resulting
from a conventional generator is used to produce Laue diffraction, the
practical applications of the method are rather limited.[24,251 It is for this
reason, and also because the diffraction pattern produced is rather difficult
Fig. 4.10. The Laue method. A stationary

specimen is irradiated with X-rays covering the
wavelength interval A,, - A,,. The shaded
area represents the volume of reciprocal space
that produces diffraction.
to interpret, that the method fell into disuse until the advent of synchrotron
radiation and the development of very powerful computers. The use of the
Laue method with synchrotron radiati~n"~] provides an extremely fast and
efficient method to record diffraction data. The applications are already
important in the field of small-molecule~27~281 as well as macromolecular
crystallography[29~301where it has opened up the possibility of performing
time-resolved studies in the crystalline state.L3']
All the other data collection methods discussed in this chapter use
monochromatic radiation and therefore require a more or less complicated
mechanism to move the crystal as the diffracted radiation is measured.
The Weissenberg camera

Although the Weissenberg ame era,[^'-^^] introduced in 1924, has nowadays
been largely superseded by other data collection methods, historically it has
played a central role, since for many years it was the standard instrument
used to quantitatively measure diffracted intensities. Currently, it is still
used for space-group and unit cell determination of small-molecule crystals,
that is for the preliminary characterization of crystals whose diffracted
intensities are then measured with the diffractometer.
The cylindrical film rotation camera

The simplest way to produce a regular motion of the reciprocal lattice is to
rotate the crystal about a direct lattice axis oriented normal to the incident
beam direction: then, as seen on p. 126, there is a family of equidistant
reciprocal lattice planes perpendicular to this axis of rotation. Whenever a
reciprocal lattice point intersects the Ewald sphere a diffracted beam passes
through the point as shown in Figs. 4.11(a) and (b). Such points all lie on
equidistant circles of decreasing radii as we move away from the centre of
the Ewald sphere. If a cylindrical cassette containing a piece of film is
placed so that the cylinder axis coincides with the crystal rotation axis, as
shown in Fig. 4.11(c), the circles will project undistorted onto the film and
when the film is unrolled they will appear as straight lines (layer lines) that
contain the diffraction spots.
Figure 4.12 shows a rotating-crystal photograph. From it one can
calculate quite easily the spacing between the equidistant planes of
reciprocal space and then use this value to compute one of the unit cell
parameters of the crystal.
If the distance on the film between the layer line corresponding to the nth
reciprocal lattice plane and the 0 layer line is df, r, is the radius of the
cylindrical cassette and dH is the constant distance between two consecutive
reciprocal lattice planes in the family then
tan an= df/rf

and
ndH
sin an= 7
A-
as shown in Fig. 4.11(c). Recall that the radius of the Ewald sphere is A-'.
248 ( Hugo L. Monaco
rotation
-
axis
Fig. 4.11. The cylindrical film rotation method.

(a) The intersection of a family of equidistant
reciprocal lattice planes and the Ewald sphere.
The points lying on the circles represented will
produce diffraction. (b) Projection i n the
direction of the incident X-ray beam. The
rotation axis is horizontal and is found i n the
plane of the figure. The circles shown i n (a) are
now projected as parallel vertical lines. (c) A
cylindrical film is placed with its axis coincident
with the rotation axis. The intersections of the
reciprocal lattice planes and the Ewald sphere
produce on the film a series of parallel lines.
Thus the interplanar spacing in reciprocal space is

sin a;, 1 df
dH=-=-sintanP1-.
nh nh rf
Using this value, the unit cell parameter corresponding to the real axis
coincident with the rotation axis can be calculated.
Incidentally, notice that because of this relationship equally spaced planes
in reciprocal space do not produce equally spaced layer lines on the film.
Zero-level Weissenberg photographs

Although in a rotation photograph the index corresponding to each layer
line is trivial to assign, the method cannot be used to easily map the
reciprocal lattice. This is because all the reciprocal lattice points present in
one plane are projected by this geometry on to one layer line and thus the
information concerning the other two indices is not easy to extract.
This difficulty can be overcome by the use of the Weissenberg camera,
which is essentially a cylindrical film rotation camera with two differences.
Experimental methods in X-ray crystallography ( 249
Fig. 4.12. A typical rotation photograph showing

a family of parallel lines due to the set of parallel
reciprocal lattice planes. Notice that the lines are
not equally spaced.
The first is the use of a layer line screen that blocks all the diffracted
radiation with the exception of that due to one selected reciprocal lattice
plane at a time. The second difference is the coupling of the rotation motion
to a displacement of the film along the cylinder axis. In this way, spots
belonging to the same reciprocal lattice plane that cross the Ewald sphere at
different times and which would end up recorded on the same layer line are
recorded instead in different positions on the film. Thus, a single reciprocal
lattice plane is mapped on to the film plane.
Figure 4.13(a) shows the Ewald sphere projected on to a reciprocal lattice
plane that would produce one layer line in a normal cylindrical film rotation
camera. The view is in the direction of the rotation axis and each of the
lines represented corresponds to a reciprocal lattice row of a given index. In
this representation the origin of reciprocal space is found at point 0, the
intersection of the incident X-ray beam with the Ewald sphere. The line
tangent to this point has one index equal to zero and corresponds to one of
the reciprocal lattice axes. In Fig. 4.13(b) the reciprocal lattice point P is
crossing the Ewald sphere and therefore it is in diffracting position. Its
coordinates, shown on the unrolled film on the right, are x and z ; the first is
proportional to the angle 2 8 as can be seen in the same figure, the second is
proportional to the rotation angle o since, as already pointed out, crystal
rotation and film translation along the cylinder axis are coupled,
If r, is again the cylindrical cassette radius one can write
film
Fig. 4.13. Zero-level Weissenberg method. (a)

When a layer line screen selects the zero-level
plane, a series of equidistant lines will contain
all the reciprocal lattice nodes that can produce
diffraction on the film. (b) When the crystal is
rotated an angle w = Othe point P belonging to film film
one of the reciprocal lattice axes is in diffracting
position. Its coordinates on the film are xand z.
and
Normally rf is chosen so that C1has the value of 2" mm-l. Thus measuring
the x coordinate of a reflection in millimetres one can automatically
calculate the corresponding value of 20.
The coupling parameters of the film movement to the crystal rotation are
chosen so that the constant C2 in
is also made equal to 2 and thus, the two angles w and 20 can be measured
on the film on the same scale.
Figure 4.13(b) shows that a normal to the zero-level reciprocal lattice row
passing through A bisects the angle 28. Thus, for this particular level, 8 is
equal to w because the sides of the two angles are perpendicular and one
can write
and
x = 2C2zlC,
which is the equation of a straight line of slope 2 if the two constants C , and
CZare chosen to be equal. Thus the reciprocal lattice axis passing through 0
will produce on the film a series of spots that will be found on a straight line
which will normally have a slope of 2.
When the angle o reaches the value of 90" the zero-level line will be
found in the direction of the X-ray beam and the line traced by the spots
will have reached the point where the film is cut to let the X-rays through.
Immediately thereafter the spots will be recorded on the other side of the
film interruption, that is they will begin to be recorded on the bottom half of
the film. When w equals 180" the zero-level line is found again tangent to
the Ewald sphere but it has been flipped over. The spots recorded after that
will change the sign of the only index which is varying along the line. Figure
4.14 shows the Ewald construction and the film appearance at the beginning
of the rotation cycle and after a rotation angle of slightly more than 180".
Fig. 4.14. (a) Five reciprocal lattice nodes have

crossed the Ewald sphere and produced the five
spots lying on a straight line on the film. (b)
After a rotation of 180" plus the w of (a)the film
shows this pattern. Notice that after a 180"
rotation the pattern is perfectly symmetrical
about the point where w = 90" on the projection
of the rotation axis onto the film plane.
Fig. 4.15. For non-central reciprocal lattice lines

28 is not equal to 2w and the plot of xversus zis
not a straight line but the curve shown in the
figure. film
If the second reciprocal lattice axis, which has to be found in the plane
selected for recording, forms with this axis an angle of a*,when o equals
a* this second axis will be found tangent to the sphere of reflection and
after that will begin to produce a second straight line parallel to that traced
by the first axis and separated from it a distance w = a*.Thus, reciprocal
lattice axes are identified in the photograph by the straight lines they
produce and the angle between them is simply the o value that separates
the lines on the film.
Figure 4.15 shows that for reciprocal lattice lines not passing through the
origin of reciprocal space, 8 is not equal to o and therefore the relationship
between x and z is no longer the equation of a straight line. It is instead a
curve of the type shown on the right-hand side of Fig. 4.15. Each of the
layer lines that do not pass through 0 will produce a curve similar to the
one shown in the figure and therefore the film will show a family of
non-intersecting curves or festoons that will be found on both sides of the
line crossing the centre of the film, at increasing distances from the centre.
Each festoon corresponds to one reciprocal space line and therefore the
reflections found on it will have one index in common. On every film there
will be two festoon families, one for each of the two reciprocal lattice axes
found in the plane that is being examined. All non-central reflections are
found at the intersections of two festoons, one from each of the two families
found on the film.
Figure 4.16 is a picture of the camera and Fig. 4.17 is a typical zero-level
Weissenberg photograph. After the plane axes and the festoons have been
identified, it is not difficult to index a photograph like this by inspection.
Upper-level Weissenberg photographs

Figure 4.18(a), a projection of the Ewald sphere on to the plane defined by
the incident X-ray beam and a normal rotation axis, shows that for a
non-zero-level photograph there is an area of the reciprocal lattice plane
that cannot cross the Ewald sphere. This area becomes larger the further
the plane is from the zero level.
Figure 4.18(b) shows the solution which is adopted to circumvent this
problem. The rotation axis is no longer chosen to be normal to the X-ray
Fig. 4.16. A modern Weissenberg camera.

Fig. 4.17. Zero-level Weissenberg photograph.

Notice the straight lines of points produced by
the zero-level lines and the festoons due to the
non-zero-level lines.
r o t a t o n axis beam, it is instead tilted so that its intersection with the reciprocal lattice
plane falls on the Ewald sphere. As a consequence, the angle made by the
incident and diffracted X-rays with a normal to the rotation axis going
through the centre of the Ewald sphere are equal (see Fig. 4.18(b)) and the
method is called the equi-inclination method. A detailed description of this
method is found in ~ u e r ~ e r . ' ~ ~ ]
Since the main uses of the Weissenberg camera are nowadays space-group
and unit cell parameter determination and no longer quantitative intensity
measurements, upper-level Weissenberg photographs are seldom recorded.
The precession camera

The precession camera, introduced by Martin Buerger in 1942, is still a very
popular instrument, especially among macromolecular crystallographers,
because it produces an undistorted picture of a reciprocal lattice plane.[35]A
precession photograph is quite easily indexed by inspection, it shows very
GIs/
rotation
diffracted
beam clearly the reciprocal lattice plane symmetry, and can be used to quickly and
easily calculate the unit cell parameters. This is so because the distance
between the spots on the film is simply a reciprocal lattice distance scaled by
the X-ray wavelength and the crystal to film distance.
The precession motion

In the precession method a crystal is first oriented with one of its axes
parallel to the X-ray beam, that is to say with a family of reciprocal lattice
planes perpendicular to the beam as shown in Fig. 4.19(a). This figure
shows the zero-level reciprocal lattice plane tangent to the Ewald sphere at
the point 0. In the precession camera the crystal is rotated in such a way
Fig. 4.18. Upper-level Weissenberg that the normal to the plane family precesses about the X-ray beam, in
photographs. (a) If the incident X-ray beam is other words it revolves about the beam keeping a constant angle p, the
normal to the rotation axis, there will be a blind
area for upper-level planes. (b) The equi- precession angle. During this revolution the normal to the planes describes
inclination method. a cone whose axis is the X-ray beam. At the same time the film that will
collect the reflections is made to move in the same way so that it is always
parallel to the reciprocal lattice plane that will be photographed and keeps
with it a constant distance.
When the precession angle p is different from 0 , the zero-level reciprocal
lattice plane is no longer tangent to the Ewald sphere; it intersects the
sphere defining with it a circle. All the reciprocal lattice points which fall on
this circle produce a diffracted beam that will be recorded on the film. This
situation is represented in Fig. 4.19(b) which shows on the right-hand side
the projection of this circle onto the film. This projection is also a circle
because the film is always parallel to the reciprocal lattice plane and the
distance between the two planes is constant.
When the normal to the reciprocal lattice planes has described half of its
revolution movement about the X-ray beam we find the situation shown in
Fig. 4.19(c).
After a full cycle has been completed we are back to the situation
represented in Fig. 4.19(b). Figure 4.20 shows the film with the projections
of the intersection of the reciprocal lattice plane and the Ewald sphere in
four situations; two are those shown in Figs. 4.19(b) and (c), the other two
correspond to the positions in which the projection of the normal on to the
plane of the figure coincides with the X-ray beam. These situations are
FILM
I I
E w a l d sphere F I LM
(b)
Fig. 4.19. Precession motion. (a) The crystal is

first or~entedwith a family of reciprocal lattice
X-rays planes perpendicular to the X-ray beam. On the
right-hand side we view the f ~ l min the direction
of the normal to the planes. (b) The normal to
the reciprocal lattice planes forms an angle y
with the X-ray beam. The intersection of the
zero-level plane with the Ewald sphere projects
Ewald sphere onto the parallel film as a circle centred on N,
FILM and tangent to the film centere 0,.(c) After half
a precession cycle we find the situation shown
here. The circle on the film has now moved from
(c ) above to below the film origin.
described by the same Figs 4.19(b) and (c) if we imagine that we are viewing
the Ewald sphere not from a side but from the top or the bottom.
When a full revolution has been completed the intersection of the
reciprocal lattice plane and the Ewald sphere has described a circle whose
radius is the diameter of the intersection and equal to
2 sin plA. (4.22)
All the points in the reciprocal lattice plane which are found within this
radius have passed through the Ewald sphere and therefore have produced
a diffracted beam which has been recorded on the film. Since the precession
angle is usually not larger than 30°, the maximum radius of this circle is
normally A-'.
In Fig. 4.19(b) the reciprocal lattice point R is inside the Ewald sphere, in
Fig. 4.19(c) it is outside. When a full cycle has been completed and we are
back to the situation shown in Fig. 4.19(b), the point is again inside the
Ewald sphere. Thus in a full precession cycle, every reciprocal lattice point
that will produce a signal crosses the Ewald sphere twice, moving each time
in opposite directions.
Isolating a zero-level reciprocal lattice plane

In Figs. 4.19(b) and (c) we have shown the precession motion indicating
only a zero-level reciprocal lattice plane. In fact, as can be easily understood
reconsidering Fig. 4.19(a), it is a family of reciprocal lattice planes that
Fig. 4.20. Movement of the projection of the always intersects the Ewald sphere generating a set of concentric circles of
intersection of the reciprocal lattice plane and
the Ewald sphere onto the film as one
variable radii. In order to record only the zero-level plane, the diffracted
precession cycle is completed. Only four beams originating from all the other planes have to be stopped using a layer
extreme situations are represented. The circle line screen.
defined by the projection will contain all the
reflections collected by the camera. Figure 4.21 shows the cones generated by the zero- and first-level
reciprocal lattice planes and the screen used to stop the beams diffracted
from the first level. The screen is simply a metal plate with an annular
aperture that will let through only the diffracted radiation coming from the
selected reciprocal lattice plane. As the crystal and the film move, the
screen also moves in such a way that it is always parallel to the film and
therefore parallel to the reciprocal lattice planes. In the precession camera,
the movement of crystal, film and screen are always coupled to satisfy this
condition.
The position of the screen, that is the crystal to screen distance, depends
on the radius of the annulus. If s is the crystal to screen distance and r, the
annular radius
Fig. 4.21. Diffraction cones generated by the

zero and first reciprocal lanice planes. The screen
screen is set at a distance from the crystal s so
that only the diffracted beams originated from \
the zero level are let through and therefore \
\
produce a signal on the film. \
and
as shown in Fig. 4.21.
Upper-level precession photographs

Let us consider an upper-level reciprocal lattice plane whose normal forms a
precession angle p with the direction of the X-ray beam as shown in Fig.
4.22.
The upper level origin 0, represented in the figure lies a distance nd*
from the intersection of the X-ray beam with the Ewald sphere. In order to
produce an undistorted photograph of the reciprocal lattice plane, the film
centre has to be' moved forward from 0: to 0:.
From the construction in Fig. 4.22 we can write
Dlnd* =flA-'
where D is the distance the film has to be moved forward, nd* the distance
between the upper-level plane and the zero-level plane, f is the crystal to
film distance, and A-l the radius of the Ewald sphere.
Thus we can calculate D
for any upper-level plane and crystal to film distance.

Figure 4.23 is the construction used to calculate the crystal to screen
distance for an upper-level precession photograph.
In this case we can write
cot v = s / r S
Also the distance between the upper-level layer and the zero-level plane is
nd* = l/A(cos p - cos v).
Fig. 4.22. Upper-level precession geometry. 0;

is the new centre of the film that will permit the
collection of an undistorted upper level; nd* is
the distance between the upper and the zero-
level plane, f i s the crystal to film distance.
Thus
cos v = cos p - nd*A
and finally
s = r, cot cos-'(cos p - nd*A) (4.25)
which can be used to calculate the crystal to screen distance for any
upper-level precession photograph. Incidentally, we notice that if n is made
equal to 0 this equation reduces to (4.23), the expression derived for a
zero-level precession photograph.
Re-examining Fig. 4.22, we notice that there is an area of the reciprocal
lattice plane that will never pass through the Ewald sphere. The projection
of this area onto the plane of the figure is the line that goes from the point
0, to Q,. The blind region circle has a radius, rb, that can be calculated
from Fig. 4.22
rb = l/A(sin v - sin p).
On the film the radius of the blind region, rB, is

rBla = f /Ap1
from which
rB= f (sin v - sin p) (4.26)
where f is the crystal to film distance.

We can use Fig. 4.23 to calculate the radius of the limiting circle for any
upper level. If r, is the maximum radius and f the crystal to film distance
rM/f = (sin v + sin p)(l/A)/A-I
r, = f (sin v + sin p). (4.27)
Fig. 4.23. Setting the layer line screen to select

an upper-level plane.
Calculating unit cell parameters from a precession

photograph
Figure 4.24 is a picture of a precession camera and Fig. 4.25 a zero-level
precession photograph of a protein crystal that belongs to the space group
P3221. Figure 4.25 is an h01 zone, i.e. the axes shown are a* and c * . In
order to determine the unit cell parameters one has to know the wavelength
of the X-ray radiation, the crystal to film distance, and the distance between
two consecutive reflections along the two axes. The precision of the
parameter determination, can be increased if a larger distance is measured,
i.e. the distance between several spots, if possible using one of the many
optical devices available to improve the quality of the measurement. If df is
the distance between two spots on the film, as shown in Fig. 4.26
and
a* = df lhf (4.28)
where f is the crystal to film distance and a* the reciprocal lattice parameter
from which the unit cell parameter can be easily calculated.
The rotation (oscillation) method in macromolecular

crystallography
In our discussion of the Weissenberg and precession cameras we have seen
that in both methods only one reciprocal lattice plane is recorded per film.
As a result, indexing of the photographs is easy but the drawback in both
cases is that most of the diffracted radiation is not recorded and therefore it
Fig. 4.24. The precession camera. (Photog1

courtesy of Huber Diffraktionstechnik.)
Fig. 4.25. Zero-level precession photograph of a

protein crystal showing the t w o reciprocal lattice
axes. From this picture it is quite easy t o
calculate the t w o unit cell parameters as
described i n the text. Notice the holes punched
on the picture used to roughly define its centre
during densitometry.
is wasted. If the crystals do not decay in time, that is if the intensity of the
reflections does not change as the crystal is exposed to the X-ray, this is not
too a serious a problem; with sufficient time available all the reciprocal
lattice region of interest can, in theory, be explored one layer at a time. But
if radiation damage is a problem, as is the case of macromolecular crystals,
both the precession and Weissenberg methods are extremely inefficient. For
Fig. 4.26. Determining the unit cell parameters

from a precession photograph.
example, the recording of a complete data set of a protein crystal with the
precession camera usually requires one crystal per photograph, i.e. one
crystal per reciprocal lattice plane. Although in the early days of protein
crystallography this was the way that data were collected, the search for a
more efficient way to record diffracted intensities led in the early 1970s to
the reintroduction of the screenless flat-film rotation (oscillation) method
for macromolecular data collection. The flat-film rotation method had been
used since the very beginning of X-ray crystallography[36]but it had been
abandoned due to the difficulties in indexing and quantitatively measuring
the reflection intensities on the film. This situation was changed by the
introduction of computer-controlled microdensitometers which assured that
films could be conveniently scanned and reliable intensities could be
extracted from them. A new type of flat-film rotation camera with eight
cassettes that are used for successive exposures, the Arndt-Wonacott
ame era,[^',^*] was built, and in a relatively short time the rotation method
became one of the major, if not the major, method for macromolecular
crystal data collection.
Geometry of the rotation method

In the rotation method a crystal is made to rotate or oscillate, that is it is
moved back and forth, through a small angle about one axis perpendicular
to the X-ray beam and the diffracted intensities are recorded on a flat film
(that can also be V-shaped) contained in a casette perpendicular to the
incident X-rays (see Fig. 4.27). Small rotation angles are required to avoid
reflection overlap on the film. An asymmetric unit of reciprocal space is
covered by successively exposing different cassettes.
When we discussed the Weissenberg camera, we saw that reciprocal
lattice planes perpendicular to the rotation axis generate a series of layer
lines on a cylindrical film. Let us now see what happens if we have a flat film
perpendicular to the X-ray beam as shown in Fig. 4.27. In describing a
crystal rotation, it is easier, and therefore often preferred, to leave the
reciprocal lattice fixed and move instead the Ewald sphere. When the
crystal is rotated a small angle A@, the two circles that result from the
intersection of the Ewald sphere at the beginning and end of the rotation
range and each reciprocal lattice plane, define two crescent-shaped lunes
Fig. 4.27. The rotation motion. The motion is

shown leaving the reciprocal lattice planes fixed
and rotating the Ewald sphere about the point 0.
After the spindle has rotated an angle A@, the
intersections of the Ewald sphere with each
reciprocal lattice plane define two lunes per
plane. A pair of lunes is shown on the right-hand
side. The film is placed flat and normal to the
X-ray beam. During the rotation of the crystal it
is kept stationary.
262 / Hugo L. Monaco
r o t at ion
axis
Fig. 4.28. Projection of the intersections of the

Ewald sphere and three reciprocal lattice planes
on the film. The innermost pair of lunes
corresponds to the plane with index one
because the plane with index zero is tangential
to the Ewald sphere.
that contain all the reciprocal lattice points that will produce diffraction on
the flat film perpendicular to the X-ray beam. Thus a rotation photograph
contains reflections coming from all the reciprocal layers that intersect the
Ewald sphere as shown in Fig. 4.28. The reflections contained in each of the
lune pairs come from the same reciprocal lattice plane and therefore have
one index in common.
Since nodes in reciprocal space have a volume, a reflection is not
completely recorded on the film until the entire volume has passed through
the Ewald sphere. Any reflection whose reciprocal lattice node has not
completely passed through the Ewald sphere is called a partially recorded
reflection or, more improperly, half spot, regardless of the percentage of the
volume that has passed through the sphere of reflection. Since the rotation
range in macromolecular crystallography is usually quite small, as we will
see, a substantial number of reflections on a rotation film are partially
recorded reflections. These have to be properly identified and dealt with
during film processing. One of the reasons why the rotation methud has
f~lm
been so successful is that it has been found that reflections partially
recorded on different films can be added together[381to yield a reliable value
for the total diffracted intensity.
Figure 4.29 shows the idealized shape of the partially recorded reflections,
which is different if they are recorded at the beginning or the end of the
rotation range. In one case the missing part of the reflection has been
recorded in the previous film, in the other it will be recorded on the next.
Re-examining Fig. 4.28 we notice that the area of reciprocal space that
crosses the Ewald sphere becomes a series of points along the projection of
the rotation axis. Thus, no matter how small their reflecting range,
reflections found along this line will always be partially recorded.
~ o n a c o t t [has
~ ~ ]examined the factors that limit the maximum rotation
I I
range allowed without having reflection overlap from different reciprocal
Fig. 4.29. The idealized shape of partially
recorded reflections. The spots labelled A are
lattice planes on the film. The expression for the maximum rotation range
recorded at one end of the rotation range and 1s:
those labelled B at the other end. The missing
parts of reflections A will be found on the A@rnax Ir*IIR&,x - A (4.29)
previous rotation photograph and those of spots
B on the following one. where A@,,, is the maximum allowed rotation range in radians, R:,, the
maximum resolution in reciprocal space of the data that will be collected, r*

is the pertinent reciprocal lattice vector, and A is the reflecting range of the
crystal.
As an example let us calculate A@,,, for a maximum resolution of 2.5 A
when the relevant reciprocal lattice vector is 1/100 A and A is 0.4"
A@,, 6 2.51100 x 57.3 - 0.4 = 1"
where we have multiplied by 57.3 to convert radians into degrees.
In Fig. 4.30 the Ewald sphere is represented projected onto the plane
determined by the X-ray beam and the crystal rotation axis. A limiting
sphere of smaller radius represents the maximum resolution of the data that
will be collected in the rotation experiment. The maximum rotation range of
the crystal, represented by a 360" rotation of the Ewald sphere about the
rotation axis, will cover the volume of reciprocal space limited by the toroid
generated by the rotation of the sphere but will leave out the shaded volume
shown projected onto the plane of the figure, a blind zone inaccessible to
the rotation geometry. From the figure it should be apparent that the
volume of the blind region increases with the resolution of the data to be
collected. A plot of the percentage of reciprocal space which cannot be
measured in a rotation experiment as a function of the resolution desired
~ ~ Idata of up to 5 A resolution, it is
has been calculated by ~ o n a c o t t . [ For
only 0.72 per cent but it increases to 4.58 per cent for 2.0 A resolution. In
order to record the data generated by the reciprocal lattice nodes present in
the blind region, the crystal has to be rotated about a different axis.
Figure 4.31 shows the total volume of reciprocal space swept out as the
crystal is rotated a total angle of 30°, 90°, and 180". From the figure it should
be evident that even if the rotation axis has no symmetry it is never
necessary to rotate a crystal much more than 180" in order to collect all of
reciprocal space with the exception of the blind region, not represented in
the figure. The volume which has not been covered by a 180" rotation has
been dotted in Fig. 4.31(c) and its size depends on the resolution desired as
shown in the same figure. From the figure it should also be clear that, in
rotation
axis
Fig. 4.30. Rotation geometry showing the Ewald

sphere projected on to the plane defined by the
X-ray beam and the rotation axis. A limiting
sphere centred o n 0 shows the resolution
-
desired or attainable. The shaded reaion
film / represents the volume of reciprocal space
inaccessibleto the rotation movement.
-
X - rays limltlng sphere
for a g l v e n
r e s o l u t on
Ewald '
sphere
Fig. 4.31. Volume of reciprocal space explored

by a total rotation range of 30" (a), 90"(b), and
180" (c). In each case a limiting sphere
corresponding t o a chosen resolution defines
the volume of reciprocal space of interest.
order to minimize the total rotation range, it is desirable to rotate the

crystal about the crystallographic axis of highest symmetry. Thus if a
hexagonal crystal is rotated about the c axis, a 30" total rotation range will
suffice to cover a reciprocal space asymmetric unit in which the only missing
volume will be the blind region.
Symmetry is, however, not the only consideration in the choice of the
rotation axis. Another factor, which is equally important, is the size of the
unit cell parameters. We have seen that the maximum rotation interval per
photograph is dependent on the unit cell parameters. Ideally, one would
like to rotate about the axis with the largest unit cell parameter, which in
this way does not become a limiting factor in the choice of the maximum
Fig. 4.32. The crystal, laboratory, and film

coordinate systems. The origin of crystal and
laboratory systems is at the intersection of the
X-ray beam with the Ewald sphere. The origin of
the film system at the intersection of the X-ray
beam with the film.
rotation range per photograph. Still, there may be practical reasons that
partially or totally limit the freedom of choice of the rotation axis. An
example is crystal morphology. A crystal shaped as a very thin plate with its
highest symmetry axis perpendicular to the plate cannot be easily mounted
with that axis parallel to the spindle.
Calculation of the film coordinates for a reflection

In order to calculate the film coordinates for a reflection, four different sets
of coordinate systems have to be introduced. The first is the crystallographic
reciprocal lattice with its unit vectors a * , b*, and c* and the angles a*,P*,
and y*. In this coordinate system, a reciprocal lattice point is specified by
the vector r* (see p. 65).
r* = ha* + kb* +LC*.
The second coordinate system in an orthogonal system called the crystal
coordinate system which is linked to the crystal, that is it rotates just as a*,
b*, and c* and differs from it in that it is always orthogonal (see Fig. 4.32).
In this new coordinate system, the coordinates of a reciprocal lattice
point, (x,, yo, 2,) can be calculated as follows
xo = ha: + kb: + lc:
yo = ha: + kb: + LC,*
z0 = ha: + kb: + LC;
where a%is the projection of a* onto the xo axis, etc.
The third coordinate system is the laboratory orthogonal coordinate
system which remains fixed as the crystal is rotated. This system is defined
so that z is parallel to the rotation axis and x is parallel to the X-ray beam
but points in the opposite direction. If at the beginning of the rotation
experiment the crystal and laboratory coordinate systems coincide, after the
crystal has rotated a certain angle @, the coordinates of a point in the
laboratory system will be
as can be seen in Fig. 4.32. If the angle @ at which a given reciprocal lattice
point crosses the Ewald sphere is known, since xo, yo, and 2, are only
functions of the reflection indices and the unit cell parameters, x, y, and z,
the reflection coordinates in the laboratory system, can be calculated.
The fourth coordinate system is the projection of the laboratory coordin-
ate system on to the film plane. In order to convert from the laboratory to
the film coordinate system we only need to know the crystal to film distance.
Figure 4.32 shows the relationship between the crystal, laboratory, and film
coordinate systems.
Thus we can calculate the film coordinates for a reflection if we know the
angle Q, at which this reflection crossed the Ewald sphere.
Let us consider the reciprocal lattice point P which crosses the Ewald
sphere at a rotation angle @. Figure 4.33 shows that P is on the Ewald
sphere when the centre of the sphere has moved from A to A' that is from
(Ap', 0, 0) to (A-' cos @, A-' sin @, 0).
Since P lies on the Ewald sphere, its coordinates must satisfy the equation
of a sphere
(xo - Apl cos @)2+ (yo - A-I sin + 2; = A-2

and
xo cos @ +yo sin cP = ~A(x:+ y: + 2;) (4.32)
which can be solved to find the value of @.1391

Once @ for a reflection is known its coordinates in the laboratory system
and its position on the film can be calculated in the way that has been
outlined.
-
X-rays
Fig. 4.33. When the crystal rotates an angle Q,

point P lies on the surface of the Ewald sphere
and the centre of the sphere has moved from A
to A'. The crystal coordinate system is only
shown before the rotation, that is when it is
coincident with the laboratory system. film
Indexing a rotation photograph

Figure 4.34 is a photograph of an Arndt-Wonacott rotation camera.[371
There are other commercial versions of rotation cameras, some of which,
with the possibility to expose only one cassette at a time, which are very
demanding on the operator unless very long exposure times per film are
used. Figure 4.35 is a rotation photograph of a protein crystal. When data
collection with the rotation method is started on a protein crystal, the space
group has already been determined using the precession method and the
unit cell parameters are known. Also precession photographs of the
zero-level projections and some upper levels are usually available. The
crystal that produced Fig. 4.35 is orthorhombic, space group P212121,and it
was mounted with the c* axis on the spindle. Although a knowledge of the
crystal setting and indexing of one or more rotation pictures is not essential
to process the data,[401we will briefly discuss how Fig. 4.35 can be indexed.
The picture is the second of a series collected with the c* axis mounted on
the spindle. It was exposed at a crystal to film distance of 7.5 cm and A@ is
in this case 3", the unit cell parameters are a = 46.1, b = 49.1, and
c = 76.1 A. Since in this case there is no higher symmetry axis, the choice of
the rotation axis was dictated only by the size of the unit cell parameters.
Partially recorded reflections can be identified by examining the spot shape
(see Fig. 4.29) or, more easily when the spot shape is not very regular as in
this case, by placing the photograph on top of the previous and the
following one. Partially recorded reflections are those that appear recorded
on more than one film.
Fig. 4.34. The Arndt-Wonacott rotation camera.

Fig. 4.35. Rotation photograph of a protein

crystal. Notice the fiducial marks that are used to
determine the centre of the film during
densitometry. For a discussion of the indexing
procedure see text.
The rotation picture recorded before the one shown in Fig. 4.35 was
exposed with the X-rays in approximately the direction of the b axis. In fact
in Fig. 4.35 we can see the small circle closer to the beam stop that
corresponds to the h01 plane.
The k index of the reflections is the easiest to identify; for the smallest
circle it is 0 and for the concentric lunes it is successively 1, 2, 3, etc.
(compare with Fig. 4.28). In order to find h and I we have to find the c* and
a * axes on the film, then indexing is done simply by counting spots. The c*
axis is horizontal because it coincides with the rotation axis, its intersection
with a * can be found by locating a * , here we are aided by systematic
extinctions. The first index h is the most difficult to determine but the
position of the c* axis, which corresponds to h = 0 can be found looking at
the picture exposed before Fig. 4.35 which shows symmetry about the
rotation axis. The axis of symmetry is the c* axis.
~ e r n a l ' proposed
~~] the use of specially designed charts to more easily
index rotation photographs. These charts are nowadays not very widely
used.
Densitometry
The data collection cameras that we have seen in this chapter normally use
film as a detector for the radiation diffracted by the crystals. In Chapter 3
(p. 161) we have seen that the quantity that is proportional to the structure
factor amplitude is the integrated intensity of the diffracted X-rays.
Therefore, in order to be able to quantitatively use the data recorded with
these methods, the first step is the extraction of the relative integrated
intensities from the film. This is currently done using an instrument called a
densitometer or film scanner, hence the name densitometry applied to the

procedure followed to measure the relative diffracted intensities on film.
Optical density and integrated intensity

The quantity that is measured by the densitometer is the optical density of
the spots produced by the diffracted beams on the film. There is a
relationship between optical density and number of X-ray photons striking
the film that we will explore more carefully in the next section but which can
be assumed, for the time being, to be linear up to X-ray optical densities of
at least 1.0. The optical density of the spots is determined by shining visible
light on the film and measuring the ratio of the intensity transmitted to that
of incident light and it is defined as
D = log I,,/I, (4.33)
where I, is the incident light intensity, I, the transmitted light intensity, and
D is the optical density.
Modern instruments are capable of measuring optical densities of very
small areas, as small as 12.5 x 12.5 ,um2 and are therefore called
microdensitometers.
The total optical density of a spot on a film is the sum of the optical
density of a certain number of squares, called pixels, covering the spot area
minus the same number of terms with an optical density equal to that of the
film background.
D= CD~-C DB.
spot
area
Figure 4.36 shows the computer output from a scanning program with a
Fig. 4.36. Computer output showing a reflection

recorded on film and scanned with a 25 pm
raster. The optical densities from 0 to 2
correspond to numbers that range from 0 to 9 in
the figure.
diffraction spot in which the optical densities from 0 to 2 correspond to

numbers that range from 0 to 9. This last interval was chosen to make the
spot representation clearer, although normally these numbers do not range
from 0 to 9 but from 0 to 255.
The sampling area has to be chosen bearing in mind that because of film
graininess statistically more significant results are obtained with larger areas.
On the other hand, the optical density should not vary significantly within
the sampling area; otherwise one is averaging transmitted light rather than
optical density and the final optical density will be in error because of the
logarithmic relationship between the two quantities. This last effect is called
the Wooster effect.[411
Film as an X-ray detector

When an X-ray photon strikes a film, it excites a silver halide particle which,
when the film is developed, will produce a black silver grain on the film.
Here we will follow Vonk and Pijpers in their derivation of a relationship
between the number of photons striking the film and the optical density they
produce. [421
If the film contains a number of excitable particles no per unit area and n
have been excited by the X-rays, as d E photons strike the film, the number
of excited particles will increase by dn and
where (no - n)ln, is the fraction of particles that have not been excited, a is
the fraction of radiation absorbed by excited and unexcited particles, and rn
is the number of particles excited per photon, normally taken to be equal to
1. Solving the differential equation
n = no[l - exp (-maElno)]
using as boundary conditions n = 0 for E = 0 and n = no for very large
values of E.
As more and more particles are excited on the film, the fraction of light
transmitted, when the optical density is measured, will decrease. Assuming
that the fraction of light transmitted by the film is proportional to the area
that has not been excited one can obtain
dIt/Io= -fIt dnlIoK
where It/Io is proportional to the unexcited area, K is the proportionality
constant, f is the surface covered by a grain, and dn the increment in excited
particles as before.
Solving the differential equation we obtain
ItlIo= exp (-fnlK),
2.3 log ItlIo= -fnlK,
and
D =fn12.3K.
The maximum optical density that can be measured, Dm,,, will cor-
respond to n = n,. If, in the expression of n as a function of E, we substitute
n and no in terms of D and Dm,, we obtain
where So = mafI2.3K.
This equation can be simplified[421
by two successive series expansions to
D I E = So(1- D /2Dmax). (4.35)
From this equation we see that for large values of Dm,, and small optical
densities the relationship between optical density and exposure is linear.
The exposure E, that is the number of photons exciting the film, is
proportional to exposure time when a film is exposed to successively higher
X-ray doses by increasing the time that it is hit by a constant X-ray beam.
The parameter Dm,, is characteristic of the film and can be determined
plotting D I E as a function of D. Some typical values have been determined
by Vonk and ~ i j ~ e r sand[ ~ more
~ ] recently by ~ l d e r [for
~ ~very
] widely used
types of X-ray film. The value of Dm,,, representing the optical density of
the film when all the silver halide granules on it have been excited, is usually
well beyond the optical density range that can be measured with the
densitometer .
The ratio D I E is called the film speed at density D, So is the initial speed,
and the function in parentheses shows how this value decreases as exposure
progresses.
Arndt et have derived a simple approximate expression for the
fractional standard deviation of an optical density measurement on film. If
aDis the standard deviation
where ND is the total number of quanta producing the optical density D.

This equation shows that the fractional standard deviation is not dependent
on the film speed.
All radiation counters are assumed to be approximately Poissonian
detectors, i.e. the statistics of counting rates can be analysed theoretically
by a Poisson distribution. If film were an ideal Poisson detector the factor
1.35 would have to be 1.0 instead. Thus the fractional standard deviation of
film is higher by a factor of 1.35 than the equivalent parameter in more ideal
Poisson detectors, like for example those used for diffractometers.
Microdensitometers
There are currently three different types of microdensitometers in use: the
rotating-drum, cathode ray tube and flat-bed microden~itometer.[~~~ The
first kind is by far the most widely used by X-rays crystallographers and it is
for that reason that it will be briefly described here.
Rotating-drum densitometers have been described by brah hams son[^^]
and ~ u o n g [ ~and
' ] their performance in the scanning of precession photo-
graphs has been analysed by Nockolds and ~ r e t s i n g e r [and
~ ~ ]by Matthews
et a1. [491
In a rotating-drum scanner, the film is mounted on a cylindrical drum that
has a rectangular aperture for the film and which rotates about the cylinder
axis. The rotation speed is variable, and depending on the instrument can
be as high as 12 revolutions per second. A beam of visible light is passed
through the film and its intensity is measured by a detector. Source and
detector are stationary during one revolution of the drum and are
automatically stepped along the cylinder axis until the entire film is covered.
The incident light intensity I,, is measured as the beam goes through air.
The raster size is variable, it can be 12.5, 25, 50, 100, or 200 ym or more
and the transmitted light is measured at intervals equal to these values.
Thus if a raster size of say 100 ym has been chosen a strip 100 ym wide will
be read in one revolution at 100pm intervals. After that, source and
detector will be advanced 100 ym and another strip will be read until the
entire film is covered by 100 x 100 ym pixels. The instrument is interfaced
to a computer and the process is totally computer controlled.
When the light beam goes through air, the instrument gives an optical
density of 0. The maximum integer reading of 255(2'- 1) can be made to
correspond to an optical density of either 2 or 3. The optical density values
thus measured can be stored on magnetic tape or directly in the computer
connected to the densitometer. There are two strategies used by computer
programs in processing densitometer data.[48,491 In the first, the entire film is
read as described above and the data are stored for subsequent computer
processing to obtain the integrated intensities. In the second approach,
integrated intensities are obtained on-line as scanning of the film proceeds.
Both strategies have their advantages and disadvantages and have been used
extensively to scan all types of diffraction films.
The equations relating optical density on the film to the total number of
photons that have caused it ((4.34) and (4.35)) show that there is a
maximum value of optical density that can be measured for a given type of
film and that the optical density is a linear function of the exposure only for
relatively low values of D. We will briefly discuss how these two limitations
are handled experimentally.
The problem of a limited dynamic range of the film, i.e. of a limited
optical density range that can be measured, can be solved by placing more
than one film in the cassette that records the X-ray diffraction pattern. Since
a substantial fraction of the radiation is absorbed by the X-ray film, those
reflections which are too strong to be measured on the first film will
normally fall within measuring range on the second or third. After the
integrated intensities have been calculated, all the films in the pack can be
scaled together.
The non-linearity of the film response can be handled by constructing
experimentally a table that relates an optical density produced by the
scanner to a given exposure. Since one is interested in relative integrated
intensities, the table can be constructed by exposing the film to the same
X-ray beam during different times and measuring the optical density of the
spots under the conditions that will be used for data scanning. In this
experiment the exposure times are proportional to the number of photons
hitting the X-ray film.
An alternative approach proposed by Matthews et a1.[49]is to assume a
parabolic relationship between integrated intensity and optical density.
where A and B are two constants; A is arbitrarily set to one and B is

determined from the scaling factor between successive films in the pack. In
this second approach the non-linearity of the film response is handled in the
film scaling procedure.
Although the simplest approach to determine the integrated intensity is to
simply subtract the background from the area covering the spot, using the
technique called profile fitting, smaller estimated standard deviations can be
~ b t a i n e d .In
~ ~this
~ ] method, a model profile for the reflection is constructed,
that is a model intensity distribution in two dimensions is determined by
averaging the measured profiles of a certain number of reflections on the
film. It is then assumed that all the reflections have this standard profile and
the measured data are fitted to it.
The single-crystal diffractometer

An alternative to the data collection methods using film that we have
described so far is the single-crystal d i f f r a c t ~ m e t e r . [ ~Nowadays,
~,~~] it is with
this instrument, that almost all of the data required to solve small molecule
structures are collected and the diffractometer can also be used in
macromolecular crystallography. Its most important advantage is better
precision since film has a much higher level of background. Its only
disadvantage is that it measures one reflection at a time, a limitation that is
not very important for non-demanding crystals but can become very severe
if radiation damage of the sample is a problem.
DWractometer geometry
A single-crystal diffractometer consists of an X-ray source, an X-ray
detector, a goniostat that orients the crystal so that a chosen X-ray
diffracted beam can be received by the detector, and a computer that
controls goniostat and detector movements and performs the mathematical
operations required to position the crystal and detector in the desired
orientations.
The detector is usually of the scintillation counter type in which X-rays
excite a fluorescent material and thus generate visible light which is
measured in an appropriate way. Xenon filled proportional counter detec-
tors are also used, particularly with Cu K, radiation. This type of X-ray
detector will be further discussed on p. 281.
Both the molecular excitation of the fluorescent material and the gas
ionization in the counter detector are events that can be triggered by the
arrival of only one photon. It can be shown experimentally[s3]that if an
X-ray beam is measured several times with a diffractometer detector, the
different intensity values obtained follow a Poisson distribution. For such a
distribution, the estimated standard deviation is
0=~ 1 ' 2
where N is the number of counts. The fractional standard deviation is thus

= ~ 1 / 2=/N-1'2.
~
We saw on p. 271 that for film methods the fractional standard deviation
is larger by a factor of 1.35. Thus, diffractometer measurements can be said
to be intrinsically more precise than those obtained from film methods.
Modern diffractometers use the equatorial geometry in which the diffracted
beams are always measured in a horizontal plane defined by the incident
X-rays and the rotation of the detector about an axis passing through the
crystal. The detector can only move on this plane and it forms an angle 28
with the incident beam as shown in Fig. 4.37. In the figure, point P is in
diffracting position because it is on the Ewald sphere and produces a
scattered beam that can be detected because it is on the equatorial plane. In
order to observe diffraction, all the reciprocal lattice nodes are brought in
turn to some point on the circle defined by the intersection of the sphere of
reflection and the equatorial plane. At the same time, the detector is moved
to the appropriate 28 angle so that it can receive the diffracted beam.
The most widely used type of goniostat uses the Eulerian cradle which
gives rise to the four-circle diffractometer shown schematically in Fig. 4.38.
x
The cradle is constituted by the circle which carries the goniometer head
Fig. 4.37. The diffractometer equatorial
with the crystal. The instrument has a main axis that is normal to the
geometry. The detector, rotating about the equatorial plane and therefore to the incident and diffracted X-ray beams
instrument main axis, defines a plane that and passes through the crystal. Rotation of the cradle about the main axis
contains the incident beam. Reflections will
always be measured on this plane. defines the angle o,rotation about the spindle axis of the goniometer head
defines the angle Q, in exactly the same way as in the rotation camera. The
angle x is defined by the spindle of the goniometer head and the main
instrument axis. The four circles of the diffractometer are thus the Q, and x
circles about which the crystal can be rotated, the o circle, defined by the
rotation of the cradle, and the 28 circle described by the rotation of the
detector about the main axis.
Both the 28 and o rotations are about the main axis but the first moves
the detector and the second the cradle. In a three-circle diffractometer the
degree of freedom that is missing is the rotation o, in other words the
cradle is fixed with its x plane perpendicular to the incident X-ray beam.
The angle 28 is 0 when the detector is positioned in the direction of the
x
incident X-ray beam, is 0 if the spindle axis is parallel to the main axis, w
is 0 when the x circle is perpendicular to the incident X-ray beam, and the
zero position of Q, is arbitrary and can be defined with respect to the crystal
orientation. If the angle x is 0 the o and Q, rotations coincide.
In general, only two rotations are required to bring a reciprocal lattice
node to the intersection of the Ewald sphere and the equatorial plane. If we
tor
Fig. 4.38. The four-circle diffractometer. The x

circle carries the goniometer head with the
crystal and together with it constitutes the
Eulerian cradle. The cradle can rotate about the
main instrument axis describing the angle w .
The detector moves independently and forms an
angle 20 with the incident X-ray beam.
Experimen$al methods in X-ray crystallography 1 275
Fig. 4.39. Bringing a reciprocal lattice node to

the intersection of the Ewald sphere and the
equatorial plane. (a) The crystal can be first
rotated about @ until the node lies on the
sphere. (b) A x rotation brings the point on the
sphere of reflection to the equatorial plane
where it can be measured.
start with x = 0, a rotation about @ or w can bring the reciprocal lattice

node on the Ewald sphere as shown in Fig. 4.39(a) which is a projection
looking down the instrument main axis. Then, it can be brought to the
equatorial plane by a rotation about the x axis as shown in Fig. 4.39(b).
Measurement of the scattered beam can then be accomplished by placing
the detector at the appropriate 28 angle. However, the direction of the
scattered beam thus defined may be such that detection may be difficult or
even impossible because of the interference of the x circle, collimator, efc.
It is for this reason that, although in principle rotation of the crystal about
two axes plus the detector movement are sufficient to satisfy the diffraction
condition, the extra degree of freedom provided by the w circle can prove
very valuable.
Quite often, the so-called bisecting geometry arrangement is chosen in
which the x plane is moved so that it is made to contain the vector r*.
x
Under these conditions, the x axis, perpendicular to the plane, bisects the
angle formed by the incident and scattered X-rays as shown in Fig. 4.40.
An alternative to the Eulerian cradle, that is used in a widely diffused
instrument, is the kappa goniostat. In this type of geometry the x circle does
not exist and the goniometer head carrying the crystal is mounted on an arm
that can rotate about an axis, the K axis, which forms an angle of 50" with Fig. 4.40. The bisecting geometry arrangement.
The reciprocal lattice vector r* lies in the x plane
the main instrument axis as shown in Fig. 4.41. The advantage of this and the x axis bisects the angle formed by the
system is that it is less cumbersome than the Eulerian cradle and thus incident and diffracted beam.
9 rotations that would cause collisions or would produce diffracted beams that
would be blocked by the x circle in the conventional four-circle
diffractometer are still practicable.
Determination of the crystal orientation and unit cell

parameters: the orientation matrix
Before one can predict the values of the angles cP, X , w, and 28 for which
scattering will be observed for a given reflection of a crystal, it is necessary
to determine the crystal orientation with respect to the Eulerian cradle. In
Fig. 4.41. The K geometry goniostat.
this section we follow the International tables for x-ray crystallography,[54]
and define several coordinate systems similarly to what was done on p. 265
for the rotation method.
Let us define the diffraction orthonormal coordinate system, AD, with
basis vectors a D , bD, cD such that a, and bD are both in the equatorial
plane; a, points towards the X-ray source for 28 = 0, bD is parallel to the
diffraction vector for positive values of 28 and cD is parallel to the
instrument main axis and has a direction such that the three unitary vectors
a D , b,, and cD form a right-handed system. Accordingly, in Fig. 4.42, the
vector cD is normal to the figure plane and points away from the observer.
A second orthonormal coordinate system, A,, the crystal coordinate
system, is defined. with respect to the goniometer head and therefore
relative to the crystal. Its basis uectors are called aG, bG, and CG and it is
coincident with the diffraction system when all the diffractometer angles are
equal to 0. When one or more of the diffractometer angles differ from 0 the
two coordinate systems are related by a rotation matrix F such that
where (see p. 72)
cos cD cos co - sin cD sin w cos x

-sin cD cos w - cos cD sin w cos x
sin x sin w
-sin x cos w cos x

n p Let
be the coordinate matrix of the vector r* with respect to the reciprocal

coordinate system
Fig. 4.42. The diffraction orthonormal system as

seen in the projection on t o the equatorial plane.
a, and b, are the t w o basis vectors that are
found o n t h i s plane.
and let
be the coordinate matrices of r* with respect to the systems AG and AD

respectively. Then, according to p. 66 and since (F)-' = F
XG = FXD.
If the angles w, X, and Q, are such that the diffraction condition is
satisfied, remembering the way in which the diffraction coordinate system
AD was defined (see Fig. 4.42), then the vector coordinates in the diffraction
system are
and its coordinates in the crystal system AG

cos @ sin o + sin Q, cos o cos x
+
-sin Q, sin o cos Q, cos o cos x (4.37)
-sin x cos o
If the bisecting geometry arrangement was chosen, then as was pointed
out the diffraction vector r* is in the x plane; in other words w is 0 and the
vector coordinates in the crystal fixed system are:
sin Q, cos x
(
XG= lr*J cosQ,cosx
-sinx ). (4.38)
Thus if we know the coordinates of a reciprocal lattice node in the fixed

crystal system we can determine without any ambiguity from eqn (4.38) the
values of the angles Q,, X, and o that will bring the node to a position that
satisfies the diffraction condition. How do we then calculate the coordinates
of a reciprocal lattice node in the fixed crystal system? In order to do this we
need to introduce the orientation matrix U defined by
and (see. p. 66)

XG = UH or H = (u)-'x, (4.40)
where U is the transposed orientation matrix.
Thus if we know the orientation matrix we can calculate the fixed crystal
coordinates of any reciprocal lattice node and therefore we can determine
the Q,, X, and w values for which diffraction will be observed for the node.
Let us now see how the orientation matrix can be determined in the general
case in which the crystal unit cell parameters are unknown. If the angles X ,
Q,, and w corresponding to the diffraction condition of three general
reflections, HI, Hz, and HJ, are measured, the coordinates of the three
reciprocal lattice points in the fixed crystal system, XG1, XG2, and XG3can
be calculated since they depend only on the angles and the magnitude of the
vectors Ir*l (eqn (4.38)).
Multiplying both sides of the first eqn (4.39) by H and introducing the
second eqn (4.40) we obtain
HA* = W A G = X ~ U - ~ U=AXGAG.
~ (4.41)
Now we can write for the three reflections, 1, 2, and 3
HIA* = X G l ~ G
If one defines the matrices
Equation (4.42) can be written in the more compact form
Recalling eqn (4.39)
and finally,
U = H ~ l x. ~ .
Since HM and XM are known, one can calculate U, the orientation matrix
from (4.43).
If the unit cell parameters are known the angles corresponding to only
two reflections are sufficient to calculate u.['~]
In the more general case we have discussed, the orientation matrix yields
also the unit cell parameters of the crystal
A* = UAG
and
since AG is orthonormal (see Chapter 2, p. 68).

Now the matrix A*A* is nothing else than the metric matrix G* of the
reciprocal lattice defined by (2.15) and from it one can determine the unit
cell parameters of the crystal.
Measurement of the integrated intensities

In order to obtain the integrated intensity of a reflection, the entire volume
of the node must be made to sweep through the Ewald sphere while the
detector counts the diffracted photons.
There are basically two scanning modes that are very widely used (but see
also ~ ~ c k o f f and ] ) are called the o scan and the 28 or o - 28 scan.
[ ~ ~they
In the o scan the detector is left stationary while the crystal and thus the
chosen reciprocal lattice node is made to cross the Ewald sphere by a
rotation of A o about the main axis. This type of scanning mode is

illustrated in Fig. 4.43(a). From the figure it can be seen that the section of
the reciprocal lattice node sampled lies on an arc passing through the node
and centred at the origin of the reciprocal lattice. In the 26 or 0 - 26
scanning mode the crystal is moved in the same way but the detector follows
the o rotation at twice the angular speed of the crystal so that in the end
A26 = 2Ao. This scanning mode is illustrated in Fig. 4.43(b). In this case,
the section of the node sampled lies in the direction of the vector r*. The
advantages and disadvantages of these two scanning modes are extensively
discussed by ~ y c k o f.[f"I
The simplest way to obtain the net integrated intensity is by using the
Fig. 4.43. Diffractometer scanning modes. (a)

The w scanning mode. (b) The w-2%scanning
mode.
so-called background-peak-background (BPB) method in which the inte-

grated intensity of the reflection is calculated by simply subtracting the
background resulting from the average of measurements to the left and to
the right of the peak from the total peak counts. This method requires an
accurate definition of the scanning width Am. In this simple case, the
standard deviation of the measured intensity can be estimated by
where I, is the intensity of the peak and IB that of the background.

The recording of the complete reflection profile by carrying out step scans
is clearly a more sophisticated alternative but it is much more demanding in
terms of data storage space. If step scans are made, profile analysis of the
peaks further improves the quality of the data.[551In this method peak
profiles are fitted to a shape that was previously determined and is
periodically updated as data collection proceeds.
The problem of defining the peak width in a step-scan-measured
reflection is discussed by Lehmann and ~ a r s e n . [The
~ ~ ]criterion used is the
function a l l , calculated for all possible peak widths, which becomes a
minimum when the appropriate width is found.
Figure 4.44 is a photograph of a modern four-circle-diffractometer.
Once the appropriate A and X-ray source operating conditions have been
chosen, the main steps that have to be taken in order to record a data set
Fig. 4.44. A modern four-circle diffractometer. 1
(Photograph courtesy of Siemens Analytical X-
ray Instruments.)
from a crystal with a diffractometer can be summarized by the following

check-list:
1. Mount the goniometer head with the specimen and centre the crystal
optically.
2. Search for the reflections that will be used to determine the orientation
matrix and unit cell parameters of the crystal. At this stage it is easier to
look for reflections at fairly low 28 values.
3. Control the quality of the crystal by examining the profile of some
reflections. If at this point one suspects that a better crystal is available,
it is wise to select it and repeat steps (1) through (3).
4. Once a satisfactory number of reflections has been found, determine a
preliminary orientation matrix and the unit cell parameters.
5. Using the results of (4) find reflections at high 20 values that will yield a
better orientation matrix and more precise unit cell parameters. If
required, at this point one is in a position to determine very accurate unit
cell parameters by using an adequate number of Friedel pairs (see p.
155) at high 2 8 values and least-square refinement methods (p. 90).
6. If possible, transform the unit cell that has been found into another of
higher symmetry. Check for systematic absences.
7. Choose the scan mode, range, and speed by examining an adequate *
number of reflections carefully.
8. Select a subset of the reflections to periodically check the orientation of
the crystal and another to monitor radiation damage (see p. 308).
9. Decide which reflections will be collected (one or more asymmetric
units), in which order and 20 range.
10.Start data collection.
11.When data collection is finished, before removing the crystal, remember
that an absorption correction is essential, and most probably record the
data necessary for an empirical absorption correction (see p. 306).
Area detectors
Due to its high precision, the diffractometer is the ideal data collection
instrument for small-molecule crystals but it suffers, as we said in the
previous section, from the drawback that it collects only one reflection at a
time. When data have to be collected from macromolecular crystals which
have very large unit cells and which therefore require the recording of many
reflections and which in addition have, in general, a more or less serious
radiation decay problem, the diffractometer is an inadequate data collection
device. On the other hand, the rotation method described earlier (p. 259),
that is with the reflections recorded on film and with a choice of the rotation
range A@ made in order to minimize the number of films exposed and the
fraction of partially recorded reflections, has an intrinsically lower precision.
This is due to two main reasons; the first is that, as we have seen, film is a
poorer detector than diffractometer counters, the second is that during the
film exposure the signal is recorded by the film only during a fraction of the
total exposure time. If the reflecting range of a reflection of the crystal is A
and the rotation range selected A@ this fraction is A/A@. Typically A is no
more than a few tenths of a degree whereas A@ can be one degree or more.
In other words, in the rotation method the signal is recorded on the film
during a time equal to tAlA@ where t is the total exposure time whereas the
background is recorded instead during the total time t.[579581One could, in
principle, improve this situation by simply reducing A@, so that A is
spanned by several rotation photographs, and then measure the integrated
intensities only on those films in which the reflection is found. There are
many reasons why this is not done when working with film but this is instead
perfectly feasible when the detection is done by the devices called area
detectors or X-ray position-sensitive detectors.[591
Area detectors were designed to combine the photon counting efficiency
of the diffractometer with the ability to record a large fraction of the
reflections which simultaneously cross the Ewald sphere, which is the main
advantage of the rotation method. Area detectors are thus probably the best
choice for the data collection of macromolecules, and although they have
not yet found many important applications in small-molecule crystal-
lography they will probably turn out to be very useful in cases when
radiation damage is a problem. However, it should be pointed out that the
devices that are currently available commercially have been optimized for
the detection of copper radiation 2nd are, in most cases, less efficient in the
detection of the higher-energy molybdenum radiation which is very often
used in small-molecule work.[601
Principles of operation of area detectors
The most common area detectors types that are currently used in macro-
molecular work and that include the current commercially available
instruments belong to two groups: multiwire proportional counters and
television area detectors. To these two groups there has been a recent
addition: the imaging plate, which is based on entirely different physical
principles and, because it is more recent, has not yet been as thoroughly
tested as the other two groups. We will briefly discuss the X-ray detection
mechanisms of these three types of detector.
Multiwire proportional counters are gas filled chambers that contain three
parallel planar electrodes; an anode sandwiched in between two cathodes.
The anode and at least one of the cathodes are arrays of parallel wires
which are perpendicular among them.[61,621 The gas filling the chamber is
usually a xenon-carbon dioxide mixture (see also Mokulskaya et u I . ~ ~ ~ ] ) .
When an X-ray photon is absorbed by a Xe molecule an inner shell electron
is emitted with a kinetic energy that is most of the energy of the absorbed
photon and which is sufficient to produce the ionization of many more
molecules. It has been calculatedf621that an X-ray photon of the Cu K,
wavelength has enough energy to induce on average the formation of 320
ion pairs. The free electrons and the positive ions move in opposite
directions, the former in the direction of the anode where they produce an
ionization avalanche with the formation of several orders of magnitude of
new ion pairs.[621These ion pairs move again in opposite directions under
the influence of the electric field and in so doing generate the electrical
signal which is measured in the detector and which is localized in the region
where the initial photon hit the counter. The function of the carbon dioxide
molecules is to absorb the ultraviolet photons which are generated in the
avalanche process and which could produce the photoemission of electrons
and thus start the whole process in another region of the counter.
In television area detectors the X-ray radiation is converted into visible
light by a fluorescent phosphor. These visible photons, after suitable
intensification, are detected by the photocathode of a standard high-
sensitivity television camera tube which is linked to a computer.[-71 The
area detector phosphor, that is the fluorescent material that transforms
X-rays into visible light, is either polycrystalline gadolinium oxysulphide or
zinc sulphide and it produces between 250 and 500 visible photons per X-ray
photon.[661In spite of this gain in photon numbers, the sensitivity of the
camera is not enough to measure them and so an increase in the signal is
required to make it detectable. This enhancement is achieved by an image
intensifier in which the photons produced by the first phosphor generate a
certain number of electrons from a photocathode which are then accelerated
and strike a second phosphor that is optically coupled to the camera. The
photon gain of these intensifiers is of the order of either 100 or 1000.[~~]
The
television camera tube consists of a photoemissive cathode with an
intensifier that accelerates the electrons generated by the light producing a
charge image which is scanned by an electron beam used to measure the
signal. Since the photons arriving in 40 ms, which is the period necessary to
scan the image, are not enough to give good counting statistics, a certain
number of these images have to be added before the statistics become
satisfactory.
The imaging plate is essentially a storage phosphor. This means that the
X-ray photons produce on the plate a latent image that is then excited by
stimulation with a He-Ne laser producing light at 633 nm. The light thus
generated has a wavelength of 390 nm and is irradiated from the plate areas
which were previously hit by the X-ray photons. This phenomenon is called
photostimulated luminescence.[681The radiation energy of the X-ray pho-
tons can be stored by the phosphor for fairly long periods; it has been found
that the photostimulated luminescence is reduced to one half of its initial
value after approximately ten hours. The photostimulable material covering
the plate is Ba F Br:Eu2+ crystals. When the plate is hit by X-ray photons
some of the Eu2+ ions are ionized to Eu3+ ions and the electrons that are
freed are trapped in Br vacancies introduced in the crystal that are called F
centres. Subsequent excitation of these centres by the laser liberates again
these electrons that return to the Eu3+ ions which thus become excited Eu2+
ions. An electronic transition in these ions generates the luminescence with
an intensity proportional to that of the original The storage
phosphor is read by an image reader which releases the stored information
by means of the laser and collects the emitted radiation and channels it into
a photomultiplier tube which converts the radiation into an electrical signal.
The plate can be used repeatedly, since exposure to visible radiation
restores it to its initial condition. The two other elements of the detection
system are an image processor and an image writer that can be used to
imprint the plate image onto photographic film to produce a permanent
The characteristics of the image reader turn out to be crucial for
the performance of the entire system and the best precision could not be
obtained until an adequate read-out instrument was built.['']
The performance of the three types of detector mentioned here has been
analysed in several of the references given in this section and in particular
by ~ r n d t . [Multiwire
~~] proportional counters and television detectors are
already fairly widely spread in many laboratories and the data they have
produced have been used to solve several new protein structure^.^^^-^^] The
special characteristic of the imaging plate is its very wide dynamic range
which appears to make it the ideal detector to be used with the very intense
synchrotron radiation.
Data collection with an area detector

In addition to the area detector, the data collection system requires some
means of orienting and moving the crystal. Multiwire proportional counters
and television detectors use either a modified rotation (oscillation) camera
with a vertical spindle, a Eulerian cradle, or another type of three-circle
diffractometer or a K goniostat. Figure 4.45 is a photograph of a television
area detector which uses a K goniostat. The detector is mounted on a
horizontal arm and can be moved about an axis passing through the crystal
in a way totally analogous to a standard diffractometer counter and, in
addition, it can be moved back and forth providing another degree of
freedom-the crystal to detector distance D. The angle formed by the X-ray
beam and a normal to the centre of the detector is called 8, and, as shown
in Fig. 4.46, limits the resolution of the data to be collected. Following
Xuong et a1.[741we can write Bragg's "aw as follows:
R,,, = A12 sin 8,,,
R,, = A/2 sin Om,,
Fig. 4 A television area detector with a K

gonic . (Photograph courtesy of Enraf
Nonil
detector
Fig. 4.46. Projection of the Ewald sphere and the

area detector on to the plane defined by the
incident X-ray beam and a normal to the
detector passing through the crystal. The
volume of reciprocal space accessible is limited
by the two spheres of radii l/R,,, and l/Rmi,
which in turn are a function of Dand 8,.
where R,,, is the maximum and R,, the minimum resolution of the
reflections that can be measured by the detector in the position correspond-
ing to the selected values of 8, and D. The angles &,, and emin are defined
in the figure and are a function of the detector size, the angle 8, and the
crystal to detector distance D. From the figure it can be seen that
28,,, = 8, + tan-' (a/2D)
28,, = 8, - tan-'(a12D)
where a is the detector width.
Equations (4.45) can be used to calculate the value of 8, to be selected
for data collection to a particular resolution once that D, the crystal to
detector distance, has been chosen.
D must be selected according to the characteristics of the detector, the
wavelength of the radiation, and the unit cell parameters of the crystal. The
first two parameters do not normally change for different experiments
performed at a given installation and therefore D can usually be calculated
with a very simple formula in which the only variable is the maximum unit
cell parameter of the crystal. For example in Howard et a1. ,[73] the crystal to
detector distance for one type of multiwire proportional counter is
calculated in centimetres, for Cu K, radiation by the following equation:
where a,, is the longest unit cell parameter of the crystal measured in
ingstroms. The equivalent equation for another type of multiwire propor-
tional counter[741is:
Thus, data from the same crystal would have to be collected at very
different Ds by the two detectors which, although based on the same
general principles, differ in their construction details. Once D has been
determined, knowing the detector width a, one can calculate the 8, required
to collect data to the resolution desired.
Each of the electronic pictures generated by the detector is called a frame
and the individual elements of the picture are called pixels. The reflection
size on the picture, the space between reflections, and, in general, the
spatial resolution of the detector are expressed by the number and size of
the pixels.
The camera or the goniostat and the detector are controlled by a
computer which is, in general, connected to another computer which
receives from it the frames that are then used to calculate the integrated
intensities (see for example Blum et u Z . [ ~ ~ ~ ) .
Two methods have been proposed to make the reciprocal lattice nodes of
the crystal cross the Ewald sphere: the rotation (oscillation) method and the
stationary picture method. In both cases the detector does not move while
data collection proceeds. In the first method the crystal is rotated about an
axis which is often the vertical axis in pretty much the same way as when
film is used. Many of the considerations discussed earlier (p. 259) are thus
applicable to this technique in which data are collected in a series of
consecutive rotation (oscillation) frames. In addition to the detection
method there are basically two fundamental differences between the two
techniques. The first is that the detector is not always perpendicular to the
X-ray beam but can form with it an angle 8, as pointed out before.
Obviously this fact has to be taken into account in the prediction of the
detector coordinates of the reflections collected. The second major
difference is in the choice of A@, the rotation range, which in this case is
selected so that each reflection appears in several frames.[733751 AS pointed
out before this strategy improves the signal to noise ratio since reflections
are integrated only in the frames in which they appear.
In the electronic stationary picture method,[762771 the crystal is also rotated
about an axis but the frame is recorded with the crystal held stationary. The
reflection intensity is thus extracted from a series of still electronic pictures
at slightly different values of a.The A@ between frames is of the order of
0.06" and subframes are sampled at distances of 0.01" in order to better scan
reflections that in some cases can be very sharp.[771This second data
collection strategy is less widely used than the rotation (oscillation) method.
It is worth noticing that these strategies of data collection are the only
ones described so far that truly sample the entire volume of a reciprocal
lattice node. With film methods what one sees is a projection of the entire
volume on to the film plane, whereas with the diffractometer one looks at a
reflection profile on a single plane that can be chosen to cut the node
volume in different ways as seen earlier (p. 278). Thus, area detector data
are the only ones that can be profile fitted in three dimensions, a possibility
that ought to further improve their quality.
In most cases the crystal is more or less accurately aligned before data
collection can begin so that x,, and y,, the coordinates of a reflection on the
detector, and @, the rotation angle at which the node crosses the Ewald
sphere for all the reflections to be collected can be predicted[761(see also p.
265). However, a full data set collected with an area detector contains a
very large fraction, if not all, of the reciprocal lattice nodes to a given
resolution and, since the crystal orientation can be obtained automatically
by efficient computer programs,[78]it is also possible not to orient the crystal
before data collection begins and find the orientation afterwards, during
frame processing.[731A strategy that can be used to cover a section of
reciprocal space with an area detector, which is obviously applicable when
the crystal orientation is known before data collection begins, is discussed
by Xuong et al.[741
Data collection techniques for polycrystalline

materials
X-ray diffraction of polycrystalline materials
An ideal polycrystalline material or powder is an ensemble of a very large
number of randomly oriented crystallites. Figure 4.47 shows the effect that
this random orientation has on the diffraction of a specimen assumed to
contain only one reciprocal lattice node. The most remarkable difference
with the single-crystal case is that we must now think of the scattering
vectors not as lying on discrete nodes of reciprocal space but on the surfaces
of spheres whose radii are the reciprocal lattice vectors r&, the distances
from the single-crystal reciprocal lattice nodes to the origin of reciprocal
space. Thus, with these specimens, diffraction is observed when the
scattering vectors lie at the intersection of the Ewald sphere and a series of
concentric spheres centred at the reciprocal lattice origin. So, rather than
having one point on the Ewald sphere, which together with the position of
the sample A fixes the direction of the diffracted beam, we now have a
series of circles. In strict analogy with the single-crystal case, these circles
and the sample define a series of concentric cones with apex in A. The
entire surface of these cones gives rise to diffraction.
A simple way to record the diffraction pattern of a polycrystalline
material is by placing a film perpendicular to the incident X-ray beam. The
diffraction cones will, in this case, give rise to a series of concentric rings.
Alternatively, a narrow strip of film can be placed on a cylinder centred at
the sample. In this case, the cones will generate concentric arcs, which are
segments of the rings, on the strip. A final possibility is to reduce the strip
to a line, that is to simply record the position and the intensity of the Fig. 4.47. fa) If the specimen is an aggregate of
randomly oriented crystallites, the vector r* is
diffracted radiation on any plane that contains the incident X-ray beam. In found in all the possible orientations with
this last case one only measures the radius of the cone and the diffracted respect to the X-ray beam. These orientations
define a sphere of radius r*. (b) The intersection
intensity at a single position. If the sample can be considered perfectly of the sphere of radius r* with the Ewald sphere
isotropic this single measurement is sufficient to completely characterize the is a clrcle that together w ~ t hpoint A defines a
diffraction pattern. The parameters reported are 28, that is the angle made diffraction cone of all the possible directions in
which diffraction is observed.
by any vector with origin in A and lying on the diffraction cone surface and
the incident X-ray beam, and the relative intensity of the radiation along
any direction on the cone. If the orientation of the crystallite in the
specimen is not perfectly random, the pattern obtained will not be isotropic.
It may even present spots corresponding to single reciprocal lattice nodes.
In that case the powder sample can be appropriately rotated so that each
crystallite adopts many different orientations in the course of data collec-
tion, thus generating a more homogeneous diffraction pattern. The result is
equivalent to having a sample with many more possible crystallite orienta-
tions and therefore closer to isotropy. Rotation of the specimen is a
standard practice in data recording of polycrystalline materials. Exceptions
Fig. 4.48. The four reciprocal lattice nodes

represented in the figure are found at the same
distance from the origin of reciprocal space 0
and therefore they will all contribute t o the
intensity of the line of the cone corresponding to
the distance r r .
are the cases in which the preferred orientations and other properties of the
crystallites need to be studied.
Another important feature which distinguishes powder diffraction is that
the intensity of the diffracted radiation on the cone surfaces can arise from
the contributions of more than one single-crystal reciprocal lattice node.
Figure 4.48 shows in projection that this can happen both as a result of
chance and crystal symmetry. A powder diffraction maximum, measured
along any direction on the cone surface, is thus said to have a certain
multiplicity that will be higher the higher the symmetry of the crystallites
under examination.
When the diffraction experiment is performed with monochromatic
radiation, that is when there only a single Ewald sphere, there is only one
diffraction cone corresponding to each sphere of a given radius rT in
reciprocal space. In other words, the angle 201 corresponds unambiguously
to the sphere of radius rT, 28, to that of radius r t , etc., and we have only
one possibility if we want to measure the diffraction that arises from the
sphere of radius r:: to have some means of detecting radiation at an angle
28, with the incident X-ray beam. It is, however, possible to shine on the
specimen X-rays with a wavelength variable within a certain range. The
experiment is exactly equivalent to the Laue method used for single crystals.
In this case, there will be many Ewald spheres, one for each wavelength,
and each will generate a diffraction cone with a given sphere of radius rT.
Figure 4.49 shows the Ewald spheres corresponding to the two values
limiting the wavelength interval of the radiation used. In the figure it can be
seen that the diffraction due to the sphere of radius rT can be measured at
many different values of the angle 28,. For different acceptable choices of
28, there will be diffraction produced by radiation of different wavelengths.
The methods which use polychromatic incident radiation and analyse the
energy or wavelength of the scattered radiation at a fixed scattering angle
are called energy dispersive methods in powder diffraction. They obviously
require a detector that will discriminate the energy of the arriving scattered
radiation and have some advantages that make them the best choice in
certain situations.[791Just like the Laue method they are best practised with
a synchrotron source which can furnish, as we have seen, radiation of
adequate intensity in a rather extended energy interval. For the remainder
of this chapter we will assume that we are dealing with monochromatic
X-rays. The methods which use them are the most widely diffused in
Fig. 4.49. The two Ewald spheres limiting the

wavelength range of the polychromatic radiation
used define with the sphere of radius r: two
limiting diffraction cones. All the cones in
between correspond to r: for different
wavelengths. The shaded region in the figure
shows the range of 0 values that can be used to
measure the diffraction in an energy dispersive
experiment.
standard laboratories. From the rich literature that covers the diffraction of
polycrystalline materials in depth we recommend two books.[s0,811
Cameras used for polycrystalline materials

We have already mentioned that a very simple way to record the diffraction
pattern generated by a powder is by simply placing a film perpendicular to
the X-ray beam, tacitly assuming that the specimen was positioned in
between the source and the film. The cameras that use this geometry are
called transmission cameras and in them the film is usually kept stationary
during data collection. Although these cameras offer the advantage of
recording the entire circle resulting from the projection of the diffraction
cone onto the film plane, they suffer from a serious disadvantage: they are
limited to maximum 28 values which for the standard specimen to film
distances and film sizes are about 45".
If the powder diffraction at very high 28 angles needs to be studied, the
flat film can still be placed perpendicular to the X-ray beam but in between
the source and the specimen instead. The cameras that use this geometry
are called back-reflection cameras. In this case the problem is the opposite,
i.e. only the diffraction at 28, higher than a minimum value can be
recorded.
Figure 4.50 shows the sample and X-ray beam and the position of the film
in the transmission and back-reflection cameras. It should be clear from our
discussion that they are useful only in those cases in which the diffraction
pattern of the polycrystalline material needs to be recorded at fairly low or
rather high 2 8 values.
A geometry which allows the recording of the diffraction pattern of a
polycrystalline material at both high and low 28 values is that used by the
Debye-Scherrer camera,[82]probably the most popular diffraction camera
for powders. In the Debye-Scherrer camera, the recording film is a strip
which is placed on a cylindrical drum centred on the sample. Figure 4.51(a)
is a photograph of the camera and Fig. 4.51(b) shows the appearance of the
strip after recording a typical diffraction pattern. As seen in the figure, the
Fig. 4.50. Position of the specimen and the film

in transmission (T) and back reflection (BR)
cameras.
Fig. 4.51. (a) The Debye-Scherrer camera. (b)

Sketch of a diffraction pattern recorded with the
Debye-Scherrer camera using the Straumanis
method of film mounting. Notice the doublets at 8 = o0 @=90°
high diffraction angles. (b)
pattern shows a series of arches resulting from the projection of the '\
diffraction cones on to the cylindrical surface. The big advantage of the
camera is that it records the entire pattern generated at all possible values of
28; its main disadvantage is that it does not record the entire projection of
the diffraction cone but only a segment. Since, as we have seen, in most
cases the diffraction pattern is isotropic, and one is therefore only interested
in the position of the arches and their relative intensities, this limitation is
not very severe.
In addition to the cylinder that holds the film strip in place, the main body
of the camera has a collimator that serves to define the incident X-ray beam
and a beam trap that stops it after it has travelled through the specimen.
Although one can place the film so that the cut in the cylindrical surface is
made to coincide with the collimator or beam trap, punching a hole for the
other, and both ways of mounting the film have been used, a third
alternative is usually preferred. In the so-called Straumanis method of film
mounting two holes are punched in the film strip positioned at about one
quarter and three quarters of the total film length. One of the two holes is
then used for the collimator and the other for the beam trap. The advantage
of the Straumanis method of film mounting is that it provides accurate
measurements for the positions of the arches that will then be translated
into Bragg spacings, dH, for both high and low values of 28. As seen in Fig.
4.51(b), the arches centred on one of the two holes punched are present as
doublets. They correspond to the K,, and K,, lines of conventional
generators which are normally not resolved by X-ray filters but are clearly
separated after diffraction by powder samples at high 2 8 values. That the
doublets correspond to high 2 8 values can be seen by differentiating Bragg's
law:
2d, sin 8 = A,
2dHcos 8 d 8 = dil, il(sin cos 8 A8 = Ail,
and
A8 = A-I tan 8 Ail. (4.46)
In the case of Cu K, radiation, the doublet is separated by 0.0038 A, if we
take il = 1.5418, A8 = 0.0240" for 8 = 10" and A8 = 0.8009" for 8 = 80"
instead. Thus, the presence of double lines centred on one of the punched
holes serves to unambiguously identify it as that corresponding to the
collimator making it unnecessary to mark the strip. It is the diffraction
pattern recorded that tells us which hole corresponds to 8 =90°. An
important advantage of this method of film mounting is that the positions
8 = 0" and 8 = 90" can be very precisely determined by taking the averages
of the arch positions corresponding to several diffraction cones on the film. S'
From the position of the symmetrical arches, one can easily calculate the X-RAYS-
corresponding 8 values since as seen in Fig. 4.52, if S is the distance
between the arches due to a diffraction cone and R is the radius of the
cylinder
SI2nR = 48/360°
Fig. 4.52. Projection of the drum of the Debye-
for the arches centred on the beam trap ( 8 = 0") and Scherrer camera on t o its axis. The specimen is
in the centre of the circle, S a n d S' are the
S1/2nR = (360" - 48)/360° distances between the symmetrical arches
corresponding t o one diffraction cone, and R i s
for those centred on the collimator ( 8 = 90"). the radius of the camera.
From the first equation we find that
If the radius of the camera is chosen so that its value in millimetres is an

exact multiple of l80/1d, a simple relationship will exist between the distance
S measured in millimeters and the scattering angle 8 measured in degrees.
Two values of R that have been very extensively used are 57.26mm and
114.6mm. A camera that has a larger cylindrical radius will resolve the
diffraction lines better since for a given A 0 there will be a larger
corresponding AS. The price to be paid is longer exposure times and higher
levels of background, since the diffracted X-rays must travel a longer path
which is normally through air which with its scattering raises the background
level.
After the 0 values of a given diffraction cone have been determined, the
corresponding interplanar spacings, d H , can be easily calculated by simple
application of Bragg's law. In the Debye-Scherrer camera the sample is
usually present in the form of a small cylinder placed in coincidence with the
camera axis and it is slowly rotated about this axis during data collection in
order to minimize the effect of the possible preferred orientation of the
crystallites.
The Debye-Scherrer camera produces rather broad diffraction lines
which for some applications may be a serious drawback since some of the
lines on the pattern may be found overlapping on the film. If one attempts
to separate the lines by using a collimator of smaller diameter or a camera
of larger radius the exposure time may become unacceptably long. It is for
those cases that the focusing or parafocusing cameras have been devised.
Parafocusing cameras, of which there are many variants used for different
applications, can be broadly classified into two groups: the Seemann-
and the ~ u i n i e r [ ' ~types.
]
The principle on which parafocusing cameras is based is shown in Fig.
4.53(a) for the Seemann-Bohlin and in Fig. 4.53(b) for the Guinier
geometry. In both cases the sample is ideally an arc of a circle which is
irradiated by reflection by a divergent beam in the case of the Seemann-
Bohlin camera and by transmission by a convergent beam in the case of the
Guinier camera. Since an arc of a circle subtends equal angles at every point
on the circle, the diffracted radiation will converge in both cases on the
parafocusing circle of the camera thus producing a very narrow line on the
film.
The convergent beam necessary for the Guinier camera can be produced
by a bent or bent and ground crystal monochromator of the type we have
discussed on p. 243.
As seen in the figure, both cameras restrict the range of 26 that can be
recorded. This range depends on the details of camera construction but one
can roughly say, and from the figure it should be appreciated, that
Seeman-Bohlin cameras record the diffraction lines only above a certain
minimum value of 26 whereas Guinier cameras do it up to a certain
maximum value.
Parafocusing cameras offer the advantage of producing very sharp
diffraction lines in times comparable with those required for standard
exposures in the Debye-Scherrer camera.
\
DIFFRACTED
J BEAM
INCIDENT
BEAM
CRYSTAL MONOCHROMATOR
DIFFRACTED
Fig. 4.53. (a) The Seeman-Bohlin parafocusing

geometry. (b) The Guinier parafocusing
geometry. Notice the different way in which the
BEAM t w o geometries restrict the 20 range available
for measurements.
The final type of camera that we wish to mention here is a modification of

the Debye-Scherrer camera that can be used to produce a powder pattern
when the specimen is a single crystal. In the Gandolfi camera[861the crystal
is mounted on a spindle inclined at 45' with respect to the camera axis and is
rotated about both this and the camera axis. If after a complete revolution
about one of the axes there has not been an integral number of rotations
about the other, the effect of the crystal motion will be that of randomizing
its orientation as exposure proceeds. If necessary the crystal can be
mounted in more than one orientation during a single exposure. The final
result is a diffraction pattern that closely resembles that produced by a true
polycrystalline material.
Diffractometers used for polycrystalline materials

The powder cameras described in the previous section all use film to record
the intensity of the diffracted X-rays. Just as in the case of single crystals, an
alternative to this measuring method is offered by quantum counter
detectors. The instruments that use this type of detector to measure the
position, that is Bragg angle, and relative intensity of the diffraction pattern
produced by a polycrystalline material are called powder diffractometers.
Powder diffractometers are thus characterized by a counter which is no

different from those used for single-crystal work: it is normally of the
scintillation type, although gas ionization counters are also used. Whatever
the detector, the measuring strategy is the same as that of the single-crystal
diffractorneter: the relative intensity and 26 angle of each diffraction cone
generated by the polycrystalline material are measured one at a time. As we
have seen before, in this way of recording the data lies the only weakness of
the diffractometer when compared with the diffraction cameras that
simultaneously record all the reflections in the accessible 26 range. For most
applications, this limitation is not very severe in the case of polycrystalline
materials and is more than counterbalanced by the greater precision offered
by these instruments.
Since the 26 angle and relative intensity of an isotropic diffraction cone
can be measured in any plane that passes through the cone apex and bisects
a projection of the cone, the powder diffractometer is simpler than the
single-crystal diffractometers we have described. The detector simply
moves, tracing a circle centred on the specimen, in a plane that contains the
incident X-ray beam, and which is normally in the vertical position, to
increasingly higher values of 26. When a diffraction line is found by the
detector, its relative intensity is recorded as a function of 28, as measured in
the plane on which the detector is moving. There is, in other words, only
one way of scanning a reflection-by varying the angle 28 that the detector
makes with the incident X-ray beam. As in the case of the other recording
methods described, the sample may also be rotated about an axis which is
found in the plane in which the detector moves in those cases in which the
preferred orientation of the crystallites in the specimen is a problem. Figure
4.54 is a photograph of a powder diffractometer and Fig. 4.55 shows a
typical output record of a powder diffractometer in which the variable in
ordinates is the number of counts measured plotted by the diffractometer as
a function of the scattering angle 26. Each of the peaks in the figure fully
describes one of the diffraction cones of the specimen.
Fig. 4.54. A, powder diffractorneter.

courtesy of Philips Analytical.)
counts
Fig. 4.55. Diffraction pattern of a polycrystalline

material recorded with a diffractometer. Notice
at higher 2Bvalues the separation of the peaks
into doublets corresponding t o the K,, and K ,,
0 20 lines.
A characteristic that distinguishels the modern powder diffractometer is

the use of a parafocusing arrangement that improves the intensity and
resolution of the diffraction maxima. The most widely used parafocusing
geometry is that found in the instrument called the Bragg-Brentano powder
diffra~tometer.[~']In the Bragg-Brentano diffractometer the specimen,
which is flat, is irradiated by a divergent beam and so in a sense this
geometry can be considered a modified version of the Seemann-Bohlin
arrangement discussed before. As we have seen, if the source is at a
distance r from the specimen, the parafocusing effect is produced on a circle
of radius r to which the specimen is tangent (in the Seemann-Bohlin
arrangement the specimen is an arc of the circle). Since in the Bragg-
Brentano diffractometer we have the additional constraint that the detector
moves around the sample keeping a constant distance R , it is obviously
necessary to ensure that the parafocusing effect is also produced at a
constant distance from the sample. This goal is achieved by moving the flat
specimen so that it is tangent to focusing circles of variable radii for
different values of 26 but which produce the focusing effect at the constant
distance from the specimen R.
Figure 4.56 shows the parafocusing geometry of the Bragg-Brentano
diffractorneter. As seen in the figure the triangle SOD is isosceles since two
of its sides are the distance r. A normal drawn from 0 to the other side R
defines with each of the sides r an angle equal to 6 and we can then write
sin 6 = $ Rl r and r = R 12 sin 6.
This equation shows why if R , the radius of the goniometer circle, is a
constant, r, the radius of the focusing circle must vary with the scattering
angle of the diffracted radiation 26. In practice, the parafocusing geometry
of the Bragg-Brentano diffractometer can be maintained by rotating the
sample S in a range spanning from a few degrees to about 160°, the useful
interval in which the instrument can be used.
DETECTOR
Fig. 4.56. Parafocusing geometry of the Bragg-

Brentano diffractometer.
In our discussion of the parafocusing geometry, it has been tacitly

assumed that divergence of the X-ray beam can only take place in the plane
of the figures. In order to ensure that this is truly so and that no important
divergence occurs out of that plane, X-rays are made to travel both before
reaching the sample and after diffraction through a series of metal plates
parallel to the plane where divergence is allowed. These plates are called
Soller slits and they are an essential part of the optics of the powder
diffractometer.
Although less widespread than the Bragg-Brentano diffractometers,
instruments that use the Seemann-Bohlin parafocusing geometry have also
been designed and marketed.[889891 Their main advantage stems from the fact
that in this arrangement all the reflections focus on a circle of fixed radius.
This makes it possible to use more than one detector (or a curved
position-sensitive detector) to simultaneously record several diffraction
maxima. With this geometry the specimen does not have to be moved since
it is assumed to be constantly an arc of the focusing circle but the specimen
to detector distance requires adjusting as the scattering angle 28 varies.
Probably the most serious limitation of this geometry is that it excludes the
value of 28 = 0" and so the instrument requires calibration using the known
28 position of the maxima of samples used as standards.
Modern powder diffractometers can be used with equal ease in both the
continuous and step-scanning modes. The first mode of operation is faster
and, in general, adequate for most standard applications. In the step-
scanning mode, the angle 28 made by the detector and the incident X-ray
beam is slowly incremented by small A28 and the number of counts is
measured and recorded. Although this operation mode requires more time
and computer memory, it results, in the end, in a much more accurate
diffraction line profile, which may be an essential requirement in some types
of application, like for example in crystallite size or and strain measure-
ments or if the data are going to be used for Rietveld refinement.
Fig. 4.57. A C U N ~position-sensitive

~ detector
used for powder diffraction work. (Photograph
courtesy of INEL.)
Curved position-sensitive detectors based on the same principles dis-

cussed on p. 282 have also been devised for use in powder diffraction
~ o r k . They
~ ~ offer
~ ' ~the advantage of combining the high precision of the
diffractometer with the possibility of simultaneously recording the entire
diffraction pattern generated by a polycrystalline material.
These type of detector are curved gas filled chambers with a metal grid in
the position of the cathode and either a curved wire or a blade which
functions as the anode, in other words they are curved linear detectors.
Thus, in a sense curved position-sensitive detectors are the equivalent of the
area detectors used in single-crystal work adapted for use with polycry-
stalline materials. Figure 4.57 is a photograph of a curved position-sensitive
detector used for powder diffraction work. The simultaneous recording of
the entire diffraction pattern offers the possibility of following processes in
which the relative intensities of the Bragg peaks varies in time as the
experiment proceeds.
Uses of powder diffraction

Probably the best known and most widely used application of powder
diffraction is as an analytical tool for both qualitative and quantitative
analysis of crystalline materials. From our discussion it should be evident
that the Bragg spacings dH and relative intensities of the diffraction cones
generated by a given polycrystalline material are a function not only of the
substance under investigation but also of its crystalline form, which may be
one of the several in which the material may crystallize.
Powder diffraction became a widely used technique for the identification
of polycrystalline unknowns in the 1930s as a result of both the clear
definition of the minimum number of parameters required to preliminarily
identify a specimen as a member of a more or less restricted group of
substances, and also of the compilation of a file containing the diffraction
patterns of a number of known crystalline materials sufficiently large to give
a reasonable probability that an unknown would be found among the
standards present in the This original file has since been expanded to
such an extent that in 1986 it included data for more than 44 000 crystalline
phases.[941The file is currently known as the Powder Diffraction File (PDF)
and it is updated and distributed by an international organization called the
25-58 25-59
J
d 3.55 2.22 2.34 4.96 B&lFS (a-phrse)
1/1 100 35 30 20
Barium N u m i n m Fluoride
hkl
I Rad. CuKo A 1.5418
Cut off I/I
Filter Ni
,
Dia.
Diffractomter I / I c o r .
Ref. Schultz et al;, Acto Chea. Scand., 2 2623-30
019
028
(1972) - 134
223
Sy s. Orthothabic S.G 036
ao 5.156 bo 7.575 co 19.64 A c 119
a p Y Z Dx 029
Ref. +
-
nwS 'V sign
D 4.53 mP Color
Ref. Ibid.
-
Hovedfagsarbeide for den Matematisk
Naturvitenskapelige Embetseksamen,
Universitetet i Trondheia, Trondheim (1969)
is stable below 770°C.
formed at 50 -1. 8 BaFz with A1F3.
FORM M - 2
1
Fig. 4.58. Re~roductionof a card of the
J.C.P.D.S. p&der diffraction data file.
Joint Committee for Powder Diffraction Standards (JCPDS), International
Centre for Diffraction Data.
Figure 4.58 is a reproduction of a card in the PDF. The information
contained in the card should be readily interpretable. Notice that the
relative intensities are expressed as percentages of the strongest line which
is arbitrarily assigned an intensity equal to 100. Not all of the cards contain
all of the information shown in the figure. In particular it may not always be
possible to unambiguously index the lines present in a pattern and therefore
the Miller indices corresponding to a given Bragg spacing may not be
available in the file.
Using the information contained in the PDF it is often possible to match
the diffraction pattern of an unknown to that of one of the known
substances present in the file. This task can be accomplished using both
manual[951and computer method^[^^,^^] with a current tendency in favour of
the latter.
The simultaneous identification of more than one component in a sample
is also possible using the method described above but clearly with a degree
of difficulty that increases with the complexity of the diffraction pattern
generated.
The quantitative analysis of the different crystalline phases present in an
unknown is another important application of powder diffraction. Due to
absorption effects of these specimens the assumption of a direct propor-
tionality between the intensities measured and the amount of a given
crystalline phase present in the sample is not possible. Alexander and
K l ~ g [ have
~ ~ ] derived the equation that relates the intensity of a given
diffraction line due to a component to its weight fraction in the sample for
the case of a flat polycrystalline specimen. If the sample is a uniform
mixture of n components and extinction and microabsorption effects can be
neglected, it can be shown thatrg8]
where IH, is the intensity of a given diffraction maximum H due to

component i, KH,is a proportionality constant which depends on the
component and the diffraction line, xi is the weight fraction of component i
in the sample, pi its density, pi is its mass absorption coefficient (defined on
p. 241), and pM the mass absorption coefficient of the matrix, that is of all
the other components in the sample with the exclusion of component i. This
equation can be specifically applied to three different cases.
In the first case the mass absorption coefficient of component i is identical
to that of the matrix. This not very frequent case arises when the specimen
is a mixture of different crystalline forms of the same compound. In this
case eqn (4.48) reduces to
that is there is a direct proportionality between the intensity of a line H due

to component i and the weight fraction of i in the unknown.
In the second case, pi is not equal to pM but there are only two
components in the sample, 1 and 2, whose identity (and therefore their mass
absorption coefficients) is known. In this case the intensity of line H due to
component 1 is
If I:, is the intensity of the same line for the pure component 1
and
so in this case I,, is not a linear function of x l . Plots of the ratio IHI/fi1as a
function of x 1 can then be either calculated using the tabulated values of ,ul
and p2 or determined experimentally from the intensities measured from
samples of known composition. These curves can then be used to determine
x l for an unknown specimen.
In the general case in which pi is not equal to ,uM and there are more than
two components, the determination of x l requires the addition of an
internal standard. For this case it can be shown that1'']
300 / Hugo L. Monaco
The linear plots, experimentally determined, of the ratio IH,/IKS as a

function of xi can then be used to determine xi for an unknown.
Alternative equations can be derived when the quantitative analysis is
done in the presence of a so-called 'flushing which can be any pure
crystalline compound added to the specimen. It can be shown that
where xi and x f are the weight fractions of the i component and the flushing
agent, Ii and If their diffracted intensities and the two constants ki and kf
their intensities relative to that of a reference substance, normally corun-
dum (a-A1,03). If the flushing agent is chosen to be corundum, this
equation reduces to
where the constant kiis readily available for many substances.

Many other theoretical methods that try to avoid the use of an internal
standard have been proposed and have had a more or less limited amount of
success.[loo]
Another common application of powder diffraction is in the determina-
tion of the unit cell parameters of a crystalline phase that can only be.
obtained as microcrystals. In this case the precision of the measurement
depends on the 8 angle of the diffraction maximum used in the determina-
tion as can be seen rewriting Bragg7slaw
and differentiating
2ddH = -A csc 8 cot 8 d 8
whence
Therefore for a given precision in the angular measurement 8 the relative

error in the measurement of d, decreases for larger values of 8.
A careful analysis of the profiles of the diffraction maxima of a
polycrystalline material can give information on the distribution of sizes of
the crystallites and on the presence of lattice strains. In order to carry out
such an analysis, it is essential to previously extract the pure diffraction
profile from the experimentally determined profiles that are influenced by
the instrument used (see p. 112) and the experimental conditions selected
for the experiment.
If the polycrystalline material has texture, that is if its crystallites do not
easily adopt a random orientation, the intensity of a ring corresponding to a
given diffraction cone will not be homogeneous. In this case, information
about the preferred orientation of the sample constituents can be obtained
by studying the entire diffraction ring in two dimensions.
The final application of powder diffraction that will be mentioned here is
the Rietveld method used for the solution of a crystal structure with a
limited number of parameter^['^^-^^^] and for quantitative phase analysis.[104]
We have already seen in Chapter 2 that in the Rietveld method of

refinement the entire diffraction profile of a peak is calculated using a
starting model for the unit cell content and the model is then refined by
comparing point by point the calculated and measured diffraction profiles.
This method of crystal structure determination has already permitted the
solution of structures of moderate complexity and which could not be solved
before by the conventional single-crystal methods because the substances
yielded crystals of inadequate size.
Data reduction
In Chapter 3 we saw that in the relationship between integrated intensity
and the square of the structure factor amplitude there are several factors
that vary from reflection to reflection. In order to calculate the relative
structure factor amplitudes to be used in the solution of the crystal
structures as described in Chapter 5 one needs first to take these effects into
account. The procedure followed to extract relative structure factor
amplitudes from raw integrated intensities is called data reduction. In data
reduction the different reflection dependent parameters present in eqn
(3.41) are taken into account by multiplying the relative intensities by
suitable correction factors. Here we will neglect E, the extinction coefficient
which was discussed on pp. 97 and 164 and will concentrate on L, P, and T,
the other three factors. The corrections applied are called, as we have seen,
Lorentz, polarization, and absorption corrections respectively. In addition,
we will also discuss the problem of radiation damage of the crystals which is
usually handled before the other corrections are applied. Another problem
that is often encountered, especially in the case of macromolecular crystal
data sets, is that of scaling partial data sets originating from different
crystals which when merged will produce the final total set of relative
integrated intensities. We discuss this problem briefly at the end of the
chapter.
Lorentz correction
We have seen that diffraction arises whenever reciprocal lattice nodes, that
always have a non-negligible volume, cross the sphere of reflection. If a
node is in diffracting position for a longer time, the intensity of the
corresponding reflection will be proportionally higher. This factor would not
be important if the method used to record the integrated intensities ensured
that every reciprocal lattice node were in a diffracting position for exactly
the same time, as it would affect every reflection in the same way and in the
end it would simply scale all the intensities by the same factor. This,
however, is not the case. Depending on the method used to record the
reflection intensity and on the position of the reciprocal lattice node, the
times required for different nodes to cross the Ewald sphere are different.
The Lorentz correction simply takes this factor into account.
The time a node is in diffracting position is dependent on two factors: the
position of the node and the velocity with which it sweeps through the
sphere of reflection. We will derive the form of the Lorentz factor in a very
simple case and then show the form it takes in a more complicated situation.
AXIS
Fig. 4.59. Lorentz correction for a crystal rotated

about an axis normal to the plane defined by the
incident and scattered X-ray beams.
Figure 4.59 shows the Ewald sphere for a diffraction experiment in which
the crystal is rotated about an axis which is normal to the plane defined by
the incident and the diffracted beams. This is for example the case of a
zero-level rotation or Weissenberg photograph or of the equatorial reflec-
tions measured with a diffractometer.
The crystal, and therefore the reciprocal lattice, is assumed to be rotated
at a constant angular velocity o ;if Vn is the linear velocity component of the
reciprocal lattice node along the radius of the sphere of reflection, the
Lorentz factor can be defined as follows
which is indeed proportional to the time during which diffraction takes place
for a given reciprocal lattice node.
The linear velocity of the point P is
and its component along the radius of the Ewald sphere

Vn= Ir*l w cos 8
since 8 is the angle formed by the linear velocity V and the radius of the
sphere of reflection passing through the point P as shown in Fig. 4.59.
Substituting Ir*l in terms of Bragg's law
Ir*l = l l d = 2 sin BlA,
Vn = (wlA)2 sin 8 cos 8,
and
L = (2 sin 8 cos 8)-' = (sin 28)-'. (4.51)
which is the simplest possible form that can be taken by the Lorentz factor.
In a more general case one has to compute Vn according to the geometry
'of the diffraction experiment and then apply the standard equation we have
seen.
If the axis of rotation makes an angle 90"-p with the incident X-ray beam
and the point P is not on a zero-level layer but rather on the nth layer with a
diffraction cone with a semiangle equal to 90-Yit can be shown that[10s21061
L = (cos p cos Y sin y)-l (4.52)
where y is the projection on to the zero layer of the angle 28 between the
incident and the diffracted beam.
If the rotation axis is normal to the X-ray beam and the reflection is on a
zero-level p = 0 and Y = 0. In this situation the projection onto the zero
level of 28, i.e. y, is identical to 28 and the expression for L given by eqn
(4.52) reduces to eqn (4.51).
~ i ~ s odiscusses ~ of the L factor for the different experimen-
n [ ~ ~the~ form
tal arrangements which are used in data collection and gives tables of the
values of L as a function of the parameters which can be selected.
Polarization correction
The polarization correction depends on the state of polarization of the
incident X-ray beam and on the scattering angle of the diffracted beam. In
Chapter 3 we have seen that when a totally non-polarized beam is diffracted
by a crystal, the diffracted intensity is affected by a factor, called the
polarization factor, which in this simple case was shown to be equal to
where 6 is the Bragg angle of the reflection considered and the diffracting
crystal was tacitly assumed to be ideally mosaic. This simple expression for
the polarization correction can be applied whenever the incident X-rays are
not polarized, that is when the radiation is produced by a conventional
source and monochromatized using an appropriate filter. Notice that in
theory this factor can have values ranging between 1.0 and 0.5 depending on
the scattering angle, although in practice this variation is less substantial.
For a data set collected with CuK, radiation between 508, and 28,
resolution it varies between Pso= 0.9995 and P2 = 0.7470.
The more general form of the polarization correction for an incident
beam monochromatized with a crystal is[17,1081
where 8 is the Bragg angle of the reflection produced by the specimen and
6; the angle of the reflection of the monochromator crystal which was used
to select the wavelength. The angle p is the angle between the projection of
the normal to the reflecting plane on to a plane perpendicular to the
incident monochromatized X-rays and the plane of incidence.[lo8]When the
original X-ray beam, the monochromated beam, and the scattered beam all
lie in the same plane this angle is equal to 0 and the polarization factor takes
the simpler form
The exponent n depends on the characteristics of the monochromating

crystal. If the crystal is assumed to be a perfectly mosaic crystal it is equal to
2, whereas if it is assumed to be an ideal crystal it is equal to 1. Real

monochromator crystals are usually an intermediate between these two
extreme cases and one should ideally examine the monochromator used to
decide the appropriate form of the polarization correction that should be
used in the specific case.[105"111]
If the radiation striking the crystal monochromator is the totally polarized
synchrotron radiation, the form of the polarization correction changes
again. An expression for the polarizaton correction to be used with
synchrotron radiation has been derived by Kahn et a1.[112]
where the angle p is defined as above and
Here EL is the amplitude of the optical field in the plane of incidence of the
X-rays and Ek is the component perpendicular to it. In this expression for
the polarization correction, the problem is to obtain an accurate value for
the parameter c' which depends on the set-up of the facility used. This can
be done in two ways; one is by measuring the polalrization ratio of the beam
that will strike the specimen. The second method is by calculating it
theoretically on the basis of the characteristics of the source and of the
crystal used to monochromatize the radiation.
The polarization correction is frequently grouped with the Lorentz
correction in a single factor, the LP correction.
Absorption corrections
As pointed out in Chapter 3, the transmission factor T is related to the
absorption of the incident and diffracted X-ray beams by the crystal. We
have briefly discussed the absorption of X-rays on p. 241 where it was
pointed out that according to Beer's law, absorption reduces the intensity of
an X-ray beam travelling through a given material by an amount which
depends on the material and the length of the path travelled by the
radiation in it. Figure 4.60 shows that, for a given scattered beam, this path
can be very different for different points in the crystal. The path lengths are
dependent, as can be seen in the picture for points 0 , R, and T, on the
location of the point scattering the X-rays, and on the incident and
scattering angle, that is on the reflection considered.
The intensity of the diffracted X-rays is thus reduced, with respect to what
it would be without absorption by the factor
which is valid for every point in the crystal. Here x is the total path length
and y is, as we have seen, the linear absorption coefficient, in this case, of
the crystal.
Fig. 4.60. For a given scattering angle 29, the Equation (4.55) can be used to calculate a very rough estimate of the
path of the incident and scattered beams i n the
crystal depends on the position of the scattering optimum crystal size for a given compound of linear absorption coefficient
point within the crystal. y. In eqn (3.41), the constant K2 included 8, the crystal volume, that we
will take to be proportional to x3. Hence

I o a x 3 and I m x 3 e - ~ .
Since we want to maximize I, the diffracted intensity after absorption
and x = 3 / p .
In order to get T, the transmission factor for an entire crystal, one simply
has to integrate eqn (4.55) over the total crystal volume. If instead of
writing x we decompose the path into p the incident or primary beam path
and q the diffracted or secondary beam path the transmission factor T can
be written as follows:
where the integration is over the entire crystal volume V.

The linear absorption coefficient for the crystal can be calculated from the
mass absorption coefficients of the atoms present in the unit cell. No
structural knowledge is required, only the values of the mass absorption
coefficients of the elements which, as we have seen, can be found in the
International tables for x-ray crystallography. From the values of p , for a
given wavelength, p can be calculated by the following equation:
where gi is the mass fraction of element i present in the unit cell, p; is its
mass absorption coefficient, and p is the crystal density. Recall that p i is a
function of the atomic number of the element and of the wavelength of the
radiation used: it is smaller for lower atomic numbers and for shorter
wavelengths. This explains why absorption corrections become more
important for heavy-element crystals and for radiation of longer wave-
lengths. Sometimes all it takes is a change from copper to molybdenum
radiation to sufficiently reduce the absorption problem in a given crystal
structure determination. In any case it is always instructive to calculate the
value of p for the crystal being examined in order to get an indication of the
severity of the absorption problem.
An analytical evaluation of T according to eqn (4.56) would be, in theory,
the ideal method to use in order to take care of the absorption correction.
The result would depend on the beam path in the crystal which is a function
of the reflection considered, i.e. one would get a different value of T for
every reflection measured. The problem is that the integral in eqn (4.56)
cannot be calculated analytically even in the case of the simplest crystal
shapes. Numerical evaluations have been obtained in the case of spheres or
cylinders, they can be found in ~ i p s o n [ " ~where
] T is given as a function of
pR and 8, R being the sphere or cylinder radius and 8 the scattering angle.
Spheres or cylinders are, however, not very good approximations for the
shape of most real crystals so that if one wants to assume that the specimen
under study is a cylinder or a sphere, it is usually necessary to grind it into
that shape. Mechanical devices exist that can be used to accomplish this.[l14]
This approach is, however, not always possible since there are many crystals
that will not survive the very harsh treatment required to shape them into
an ideal form.
An analytical method that can be used to calculate T for any polyhedral
crystal was proposed by de Meulenaer and ~ o m ~ a . This
[ " ~method
~ divides
an arbitrary crystal volume into smaller polyhedra that are ultimately
subdivided into tetrahedra. The total transmission factor is then calculated
as
where AT is the contribution to the transmission factor of each tetrahedron

and VT its volume. The contribution AT can be calculated analytically and it
depends on the values of px at each vertex of the tetrahedron and on v=. In
this method, the calculations are performed by a computer program which
requires a precise knowledge of the crystal shape which is input as the
equations of the planes defining the faces of the crystal.
In many cases, an accurate description of the crystal shape and size is very
difficult, if not impossible, to obtain. Furthermore, if one is dealing with a
protein crystal, it must be mounted inside a capillary and bathed in its
mother liquor, both of which absorb the X-rays differently from the crystal.
Thus, there are many cases in which analytical methods cannot be applied;
it is for these situations that empirical absorption correction methods have
been proposed. These methods, which attempt to measure T experimentally
are currently the most widely used, mainly because they are quite easy to
implement. The experimental absorption correction methods are reviewed
by ~ a r t e l s . [ " ~Here
] we will only briefly discuss one of them, the very widely
used method of North et a1.[1171
The method is designed to correct intensities measured with the
diffractometer and the crystal is therefore assumed to be totally bathed by
the incident X-ray beam. In Fig. 4.40 for example we have point P in the
diffracting position, that is the point is in the equatorial plane and on the
Ewald sphere. If the crystal is rotated about the vector r* the point P will
still remain in the diffracting position and the variations in intensity
observed may be attributed to absorption effects. Such a rotation is called
an azimuthal or q-scan. In practice what is done is to bring the @ rotation
axis to a direction coincident with that of vector r* which requires the
choice of a reflection with a x value of 90" (see p. 274). Then the azimuthal
angle I# is equivalent to the diffractometer angle @ and a @-scan produces a
curve similar to that shown in Fig. 4.61.
The relative transmission factor for a given value of @ is then given by
T = I(@)IZrnax(@)
where I,,,(@) is the maximum intensity observed as @ is varied over the
360" range. Clearly, in this method, and as seen in Fig. 4.61, the
transmission factor T at @ and 4, + 180" must be identical since the only
difference between these two positions is that the crystal has been flipped
over. In the method of North et al. the transmission factor for a given
reflection is given by the mean
T(H) = t ( T ( @ i n c ) + T(@reJ) (4.58)
Fig. 4.61. Relative transmission factor plotted as

a function of the scanning angle @ = y, for a
0.90 reflection chosen with a x value close to 90". The
0.00 40.00 80.00 I20.00 160.00 200.00 240.00 280.00 320.00 plot can be used to apply an empirical
ANGLE @ZY absorption correction as described in the text.
where Q, and define the orientations of the crystal in which the

incident and reflected beams for the H reflection coincide with or lie in the
same plane as the incident X-ray beam.
An underlying assumption of this method is that the total transmission is
the product of the transmission of the primary and the secondary beams. A
discussion of the applicability of this assumption can be found in Kopfmann
and ~ u b e r . [ " ~If] the transmission curves are measured for symmetry-
equivalent reflections this assumption is not necessary.[1191Other methods
that calculate transmission curves based on the differences observed in
symmetry-equivalent reflections have been proposed.[120,1211 They have the
disadvantage that they require several q-scans, i.e. much more experimen-
tal work to obtain a result that does not always justify it. An example of the
way in which absorption correction effects can be taken into account,
together with other factors, by examining the differences in intensity
observed for symmetry equivalent reflections is presented by Takusa-
ga~a[1221 who discusses together absorption and radiation damage correc-
tions for data collected with an area detector. A final possibility is to take
care of the absorption effects after a preliminary structure has been solved
using the uncorrected data. From the differences between the calculated and
measured structure factors it is then possible to calculate correction curves
that can be applied to the raw data.[lZ3]Whatever the absorption correction
used, the redundancy in the data set can be used to check whether the
correction has improved the data or not. This is normally done by
calculating an R factor (see eqn (4.70)) that should obviously decrease after
the absorption correction is applied.
Although, as we have seen, the theory of X-ray absorption by crystals is
very well established, and in spite of the fact that so many different methods
to determine T have been proposed, absorption still remains one of the
most serious, if not the most serious, source of error in the experimental
determination of relative integrated intensities.
Radiation damage corrections

So far we have tacitly assumed that as data collection proceeds the
integrated intensities measured do not change with time, or at least that
they do not change significantly. Unfortunately, this is very seldom the case
since crystal decomposition triggered by exposure to X-rays is not at all an
uncommon occurrence. Radiation damage can vary remarkably with the
specimen considered, it is known to be substantial in the case of macro-
molecular crystals, but it is also present in many small-molecule crystals as a
survey conducted in 1971 has shown.[1241
The causes that lead to the changes in intensity as a crystal is exposed to
higher and higher X-ray doses have been listed by ~ b r a h a r n s . [ ' ~If~the
]
specimen can be considered to be close to an ideal crystal, exposure to
radiation can increase its mosaicity thereby reducing extinction effects. The
result will be that, after exposure, some integrated intensities will be higher.
More often, there is in real crystals a loss of short- and long-range order
that produces instead a decrease in the measured intensities. Another effect
is chemical damage, for example due to free radicals generated by the
radiation, which can either increase or decrease the intensities. A combina-
tion of these factors is, of course, possible, which explains the difficulties
encountered in proposing a satisfactory model for this phenomenon and the
scarcity of research papers dealing with this subject.
Radiation damage is evidenced by monitoring the intensity, as a function
of data collection time, of a certain number of reflections, ideally sampling
as large as possible a region of reciprocal space. Figure 4.62 is a plot of the
values measured for three reflections of a protein crystal that decayed, on
the average, about 20 per cent during an exposure time of less than two
days. This type of monitoring is usually done when data are collected with
the diffractometer, although it is also possible for the rotation and the
area-detector methods as well.
Fig. 4.62. Typical behaviour of the relative

intensities of three reflections of a protein
crystal monitored with a diffractometer at
different times after exposure begins. EXPOSURE I HRSI
The radiation damage correction factor, R(t, 6), is defined as follows
that is it is the inverse of the factor by which the intensity measured at time t
has to be multiplied to yield the intensity corrected at time zero, the value
that should be used to extract the relative structure factor amplitudes. This
correction factor is a function, not only of the time after data collection
started but also of the scattering angle of the reflection 6.
The value of R can be estimated without resorting to any particular model
for the radiation damage process, for example by fitting the intensities of
each monitored reflection to a polynomial of the form[125]
where t is the exposure time and n is a number that ranges from 1 to 7. The
discrepancy index
is then calculated for each standard reflection as a function of n and used to

decide the best form of the polynomial to be used, i.e. the value of n that
best fits the data measured. If a general trend for the different standard
reflections is identified, then the same form of the polynominal can be
adopted and a weighted average of the coefficients A, of the monitored
reflections can be used to correct all the data collected. In the simple case in
which the decay is found to be linear, i.e. if the best n turns out to be equal
to 1, anisotropy can be readily taken into account.[1251
Sometimes, a correction based on just a few monitored reflections may
not be acceptable. A scheme that attempts to overcome this difficulty has
been proposed by 1bers.[lZ6]If, after the structure has been solved, Sn is the
scale factor for the nth reflection measured at time t that brings the
calculated structure factor amplitude IF,] (see Chapter 5) in scale with the
observed amplitude (FoI, one can define Sn as the following function of
exposure time:
and then determine the constants a, and a, by least-squares refinement

along with the other structural parameters (see Chapter 5).
The problem of intensity variations due to radiation damage was
perceived quite early on in protein crystallography. The first model that
attempted to explain them in the case of myoglobin was proposed by Blake
and Phillips in 1962.[12'] The model describes the diffraction pattern from
the irradiated crystal as due to three different components: an undamaged
fraction A,, a disordered fraction A2 that scatters predominantly at low
values of 6, and a totally amorphous fraction A3 that no longer scatters
coherently. The variation of the intensities as a function of time is then
governed by the equation
where D is a disorder parameter, S = sin 6/12, and the quantities Al(t) and
A2(t) can be calculated by making different simplifying assumptions for the

rate processes involved. Hendricks~n['~~] has proposed a general method
that considers all the possible paths between A,, A,, and A,, and has also
tested the validity of the different possible assumptions in the case of
myoglobin. His conclusion is that for moderate radiation damage all the rate
models work satisfactorily whereas for severe cases none do.
A refinement of the Blake and Phillips model has been recently proposed
by Sygusch and ~ l l a i r e . [ ' ~In~ ]their model another state, A;, with a
dose-dependent rate of formation is introduced. The molecules in this state
are thought to be only superficially perturbed. In addition, anisotropy is also
taken into account.
The usefulness of all the models proposed is that they furnish a form for
the variation of R(t, 8) as a function of a certain number of parameters that
can be determined from the variation of I ( t ) as a function of t for the
monitored reflections. The parameters so determined can then be used to
calculate a correction factor for each reflection present in the data set
collected.
An efficient way to minimize radiation damage and therefore to make the
application of a correction less significant is by collecting the data at low
temperature. Even a modest decrease to say 4 "C has already important
beneficial effects on most crystals but the best results are obtained when
working at liquid nitrogen or, even better, liquid helium temperatures.
Although the experimental problems posed in this case are more difficult to
solve, low temperature data collection is a well established technique for
small molecule crystals['301and shows great promise in macromolecular
work as we11.[131,1321 In addition to greatly reducing radiation damage,
working at low temperature decreases the intensity losses due to thermal
vibrations and thereby improves the resolution of the diffraction pattern
and, by reducing the thermal diffuse scattering, that is the background
radiation resulting from the correlated atomic displacements in the lattice, it
improves the signal-to-noise ratio of the diffracted peaks. In Appendix 3.D
we have already seen that precise electron density determinations of small
molecule crystals require that the data be collected at low temperature.
Relative scaling
If the final total data set of relative integrated intensities has not been
collected under strictly constant conditions but results from merging a
certain number of subsets, each measured under more or less different
conditions, before these sets can be merged together into a single one it is'
necessary to scale them by applying the appropriate relative scale factors.
The subsets may be derived from different crystals, if radiation damage has
made it necessary to stop data collection at a certain stage before the set
was completed, or they may also come from the same crystal. For example
data collection on films with the rotation method requires scaling as there is
no way to ensure that all the data collection parameters, take for example
film developing and fixing times, will be held strictly constant throughout
the data collection process.
Relative scaling of partial data subsets is done on the basis of the
reflections which these subsets have in common. In order to determine the
relative scale factors to be applied to the subsets, it is first necessary to
define the conditions on the data that the scale factors ought to satisfy.
Among the different criteria proposed, that of Hamilton et a1.[1331is
currently the most widely used. In this method one defines
where IH, is the ith observation of reflection H, 1 is the subset in which the
ith reflection is present, Ki the relative scale factor to be applied and VH,the
weight of the ith observation of reflection H.
It can be shown that this condition can be stated in the equivalent
Now if one defines
the relative scale factors are chosen so that the quantity I/J is a minimum and
therefore the condition from which the best value for the intensity of
reflection H, I,, is found is
and IHis
Since in this formulation the residual is not linear in the Gis, the best values
for these parameters are determined using the iterative non-linear least-
squares procedure described in Chapter 2 (p. 94), that is for each iteration
qH,is approximated by
Given that the AGl(i)s are not independent, one of them is arbitrarily set
equal to zero, that is one GI is made constant, and then the other Gls are
corrected until convergence is achieved.
An alternative way of solving the equations of Hamilton et al. was
proposed by Fox and ~ o l m e s . [ ' ~
In~their
] formulation one sets in turn all
the derivatives of q with respect to the Gis to zero, i.e.
6q/6G1= 0 for 1 = 1 . . . L
and then approximates q by the Taylor expansion:
In the special simple case in which the weights can be written as the product
of a term which depends on the reflection and another which depends on
the subset an exact solution for this problem has been found.[1341 It is useful
in scaling the different films present in a pack which have in common many
reflections and differ only in those which are outside the dynamic range of
each film.
Another alternative procedure that avoids the use of iterations to
determine the scale factors has been proposed by ~ a ewho [ defines
~ ~ the ~ ~
residual
A",, = log KIIH,- log K,!IH, (4.68)
and minimizes the quantity
where the subscripts i and j refer to different subsets.

If the number of data points is large there is very little difference in the
relative scale factors found by this method and the classical solution of
Hamilton et al.
When merging together different subsets of data into a final data set, it is
customary to calculate a reliability index for the set that can be defined
either in terms of the structure factor amplitudes or of the intensities. A
very widely used definition for this parameter in terms of the intensities is
that given by eqn (4.61)
N
(4.70)
where I(H)i is the ith measurement of reflection H, (Z(H)) is its mean value
and the summation extends over all the reflections measured more than
once in the set.
In the cases in which R is calculated using independent reflections which
ought to have equal intensities for symmetry reasons, the notation R,,, is
used.
Appendix
4.A Determination of the number of molecules in the

unit cell of a crystal
One of the first fundamental parameters required in a crystal structure
determination is the number of molecules present in the unit cell of the
crystal.
Small-molecule crystals do not usually contain important amounts of
solvent and/or other substances whereas in macromolecular crystals a
substantial fraction of the unit cell volume is occupied by solvent (see
Chapter 8, p. 538). In both cases the number of molecules present in the
unit cell is a function of the crystal density, the molecular weight of the
substance under investigation, and the unit cell volume.
If we can assume that the only species present in the crystal is the
compound under investigation, its mass in the unit cell is
where n is the number of molecules in the unit cell, M, the molecular weight
of the substance, and N is Avogadro's number.
The density of the crystal is
Equation (4.A.1) can also be written
where the unit cell volume has to be measured in A3 and the density in
g cmP3.
Measuring the density of a small-molecule crystal usually poses no serious
problem. There are several methods available and the measurements can be
done with high precision.[136]
If the molecular weight of the compound is known, then, using (4.A.2),
the number of molecules in the unit cell, n, can be easily calculated. Since n
has to be an integer, if the density measurement is very reliable and the
molecular weight is not, eqn (4.A.2) can be used to calculate a more
accurate molecular weight. Alternatively, a precise molecular weight can be
used to yield an accurate density for the crystal using the integer closest to
the n determined experimentally.
In the case of protein crystals, the situation is not so simple and there are
several alternative equations that are equivalent to (4.A.2) We will briefly
discuss one of them.[1371
In a macromolecular crystal (see Chapter 8, p. 536) there is water which
is eliminated when the crystal is dried and water which remains bound to
the macromolecule and there is also salt dissolved in the solvent. If d is the
fractional loss of mass when the crystal is dried, u is the fraction of liquid
which remains in the crystal, s is the mass of salt per unit mass of solvent,
and w is the solvent not accessible to the salt because it is strongly bound to
the macromolecule, the total mass of the unit cell of the crystal is
m = m, + dm + um, + s(m - m, - wm,),
where m is the total mass of the unit cell and m, the mass of the protein in
the unit cell.
From this equation we can obtain
The mass of the unit cell is
where V is the unit cell volume and p, is the crystal density.

The mass of the protein in the unit cell is
where n is the number of molecules in the unit cell, M, the molecular weight
of the protein, and N is Avogadro's number. Thus
n=-NVp, 1 - d - s
Mp l + u - s - s w
Vp, 1 - d - s
n = 0.602 - (4.A.3)
Mp 1 + u - s - S W '
Equation (4.A.3) can be used in much the same way as eqn (4.A.2) but in
this case determining the crystal density is a much more serious experimen-
tal problern.[l3'] In addition, one needs to know u, w, s, and d ; the first two
parameters are usually not determined, they are instead estimated from the
average of known protein crystals; s is the quantity most easily measured
and d is quite difficult to determine and requires the use of several crystals
for better precission.
An alternative to eqn (4.A.3) has been derived by ~ a t t h e w s . " Another
~~]
approach to determine n for macromolecular crystals is discussed in Chapter
8 (p. 538).
References
Rieck, G. D. (1962). In International tables for x-ray crystallography, Vol. 111,
(ed. C. H. MacGillavry and G. D. Rieck), pp. 59-72. Kynock, Birmingham.
Luger, P. (1980). Modern x-ray analysis o n single crystals, Ch. 2. Walter de
Gruyter, Berlin.
Phillips, W. C. (1985). In Methods in enzymology, Vol. 114, (ed. H. W.
Wyckoff, C. H. W. Hirs, and S. N. Timasheff), pp. 300-16. Academic,
Orlando.
Eisenberger, P. (1986). Science, 231, 687-93.
ESRF (1987). Foundation Phase Report. ESRF, Grenoble.
Bienenstock, A. and Winick, H. (1983). Physics Today, 36, 48-58.
Winick, H. (1987). Scientific American, 257, 72-81.
Margaritondo, G. (1988). Introduction to synchrotron radiation. Oxford
University Press, New York.
Koch, E. E. (ed.) (1983). Handbook o n synchrotron radiation. North Holland,
Amsterdam.
Materlik, G. (1982). In Uses of Synchrotron Radiation in Biology (ed. H. B.
Stuhrmann), pp. 1-21. Academic, New York.
Hendrickson, W. A., Smith, J. L., Phizackerley, R. P., and Merritt, E. A.
(1988). Proteins : Structure, Function and Genetics, 4, 77-88.
Arndt, U. W. (1984). Journal of Applied Crystallography, 17, 118-19.
Bonse, U. (1980). In Characterization of crystal defects by x-ray methods (ed.
B. K. Tanner and D. K. Bowen), pp. 298-319. Plenum, New York.
Rieck, W., Euler, H., and Schulz, H. (1988). Acta Crystallographica A44,
1099-101.
Koch, B. and MacGillavry, C. H. (1962). In International tables for x-ray
crystallography, Vol. I11 (ed. C. H. MacGillavry and G. D. Rieck), pp.
157-200. Kynock, Birmingham.
16. Roberts, B. W, and Parrish, W. (1962). In International tables for x-ray
crystallography, Vol. 111, (ed. C. H. MacGillavry and G. D. Rieck), pp.
73-88. Kynock, Birmingham.
17. Arndt, U. W. and Sweet, R. M. (1977). In The Rotation Method in
Crystallography (ed. U. W. Arndt and A. J. Wonacott), pp. 43-63. North

Holland, Amsterdam.
18. Margaritondo, G. (1988). Introduction to synchrotron radiation, pp. 50-81.
Oxford University Press, New York.
19. Witz, J. (1969). Acta Crystallographica, A25, 30-42.
20. Huxley, H. E. (1953). Acta Crystallographica, 6, 457-65.
21. Franks, A. (1955). Proceedings of the Physical Society, B68, 1054-64.
22. Harrison, S. C. (1968). Journal of Applied Crystallography, 1, 84-90.
23. Friedrich, W., Knipping, P., and Laue, M. (1912). Sitzungsberichte der
mathematbch-physicalischen Klasse der Koeniglich Bayerbchen Akademie der
Wissenschaften zu Muenchen (reprinted in Naturwissenschaften (1952) pp.
361-7).
24. Henry, N. F. M., Lipson, H . , and Wooster, W. A. (1951). The interpretation of
x-ray diffraction photographs, pp. 71-86. Macmillan, London.
25. Amor6s, J. L., Buerger, M. J. and Canut de Amor6s, M. (1975). The Laue
method. Academic, New York.
26. Helliwell, J. R . , Habash, J., Cruickshank, D. W. J., Harding, M. M.,
Greenhough, T. J., Campbell, J. W., Clifton, I. J., Elder, M., Machin, P. A.,
Papiz, M. Z., and Zurek, S. (1989). Journal of Applied Crystallography, 22,
483-97.
27. Wood, I. G., Thompson, P., and Matthewman, J. C. (1983). Acta
Crystallographica, B39, 543-7.
28. Gomez de Anderez, D., Helliwell, M., Habash, J., Dodson, E. J., Helliwell, J.
R., Bailey, P. D., and Gammon, R. E. (1989). Acta Crystallographica, B45,
482-8.
29. Moffat, K., Szebenyi, D . , and Bilderback, D . (1984). Science, 223, 1423-5.
30. Hajdu, J., Machin, P. A., Campbell, J. W., Greenhough, T. J., Clifton, I. J.,
Zurek, S., Gover, S., Johnson, L. N., and Elder, M. (1987). Nature, 329,
178-81.
31. Moffat, K. (1989). Annual Reviews of Biophysics and Biophysical Chemistry,
18, 309-32.
32. Weissenberg, K. (1924). Zeitschrift fur Physik, 23, 229-38.
33. Buerger, M. J. (1942). X-ray crystallography, Ch. 12. Wiley, New York.
34. Stout, G. H. and Jensen, L. H. (1968). X-ray structure determination, Ch. 5.
Macmillan, New York.
35. Buerger, M. J. (1964). The precession method. Wiley, New York.
36. Bernal, J. D . (1926). Proceedings of the Royal Society of London, A113,
117-64.
37. Arndt, U. W., Champness, J. N., Phizackerley, R. P., and Wonacott, A. J.
(1973). Journal of Applied Crystallography, 6, 457-63.
38. Arndt, U. W. and Wonacott, A. J. (ed.) (1977). The rotation method in
crystallography. North Holland, Amsterdam.
39. Wonacott, A. J. (1977). In The rotation method in crystallography (ed. U . W.
Arndt and A. J. Wonacott), pp. 75-117. North Holland, Amsterdam.
40. Rossmann, M. G. and Erickson, J. W. (1983). Journal of Applied
Crystallography, 16, 629-36.
41. Wooster, W. A. (1964). Acta Crystallographica, 17, 878-82.
42. Vonk, C. G. and Pijpers, A. P. (1981). Journal of Applied Crystallography, 14,
8-16.
43. Elder, M. (1985). In Methods in enzymology, Vol. 114, (ed. H. W. Wyckoff,
C. H. W. Hirs, and S. N. Timasheff), pp. 199-211. Academic, Orlando.
44. Arndt, U. W., Gilmore, D. J . , and Wonacott, A. J. (1977). In The rotation
method in crystallography (ed. U . W. Arndt and A. J. Wonacott), pp. 207-18.
North Holland, Amsterdam.
45. Wonacott, A. J. and Burnett, R. M. (1977). In The rotation method in
crystallography (ed. U . W. Arndt and A. J. Wonacott), pp. 119-38. North

Holland, Amsterdam.
46. Abrahamsson, S. (1966). Journal o f Scientzjic Instruments, 43, 931-3.
47. Xuong, Ng. H . (1969). Journal o f Physics E : Scientzjic Instruments, 2, 485-9.
48. Nockolds, C. E. and Kretsinger, R. H. (1970). Journal of Physics E: Scientific
Instruments, 3, 842-6.
49. Matthews, B. W., Klopfenstein, C. E., and Colman, P. M. (1972). Journal of
Physics E : Scientzjic Instruments, 5, 353-9.
50. Ford, G. C. (1974). Journal of Applied Crystallography, 7, 555-64.
51. Arndt, U. W. and Willis, B. T. M. (1966). Single crystal diffractometry.
52. Wyckoff, H . W. (1985). In Methods in enzymology, Vol. 114, (ed. H. W.
Orlando.
53. Luger, P. (1980). Modern x-ray analysis o n single crystals, Ch. 4. Walter de
Gruyter, Berlin.
54. Hamilton, W. C. (1974). International tables for x-ray crystallography, Vol. IV,
(ed. J. A. Ibers and W. C. Hamilton), pp. 275-81. Kynock, Birmingham.
55. Diamond, R. (1969). Acta Crystallographica, A25, 43-55.
56. Lehmann, M. S. and Larsen, F. K. (1974). Acta Crystallographica, A30,
580-4.
57. Arndt, U. W. and Faruqi, A. R. (1977). In T h e rotation method in
crystallography (ed. U . W. Arndt and A. J. Wonacott), pp. 219-26. North
Holland, Amsterdam.
58. Harrison, S. C. (1984). Nature, 309, 408.
59. Arndt, U. W. (1986). Journal of Applied Crystallography, 19, 145-63.
60. Hamlin, R., Cork, C., Howard, A., Nielsen, C., Vernon, W., Mathews,
D., and Xuong, Ng. H. (1981). Journal of Applied Crystallography, 14,
85-93.
61. Hamlin, R . (1985). In Methods in Enzymology, Vol. 114, (ed. H. W. Wyckoff,
C. H. W. Hirs, and S. N. Timasheff), pp. 416-52. Academic, Orlando.
62. Faruqi, A. R. (1977). In The rotation method in crystallography (ed. U . W .
Arndt and A. J. Wonacott), pp. 227-43. North Holland, Amsterdam.
63. Mokulskaya, T. D., Kuzev, S. V., Myshko, G. E., Khrenov, A. A.,
Mokulskii, M. A., Dobrokhotova, Z. D., Volodenkov, A. Y., Rubanov, V.
P., Ryanzina, N. A., Shitikov, B. I., Baru, S. E., Khabakhpashev, A. G., and
Sidorov, V. A. (1981). Journal of Applied Crystallography, 14, 33-7.
64. Arndt, U. W. and Gilmore, D. J. (1979). Journal of Applied Crystallography,
12, 1-9.
65. Arndt, U. W. (1982). Nuclear Instrumentation and Methods, 201, 13-20.
66. Arndt, U. W. (1985). In Methods in enzymology, Vol. 114, (ed. H. W.
Orlando.
67. Arndt, U. W. and Int'Veld, G. A . (1988). Advances in Electronics and
Electron Physics 74, 285-96. Academic, Orlando.
68. Miyhara, J., Takahashi, K., Amemiya, Y., Karniya, N., and Satow, Y. (1986).
Nuclear Instrurnentation and Methods, A26, 572-8.
69. Amemiya, Y., Kamiya, N., Satow, Y., Matsushita, T., Chikawa, J., Wakabay-
ashi, K., Tanaka, H., and Miyahara, J. (1987). In Biophysics and synchrotron
radiation (ed. A. Bianconi and A. Congiu Castellano), pp. 61-72. Springer,
Berlin.
70. Amemiya, Y., Matushita, T., Nakagawa, A., Satow, Y., Miyahara, J., and
Chikawa, J. (1988). Nuclear Instrumentation and Methods, A266, 645-53.
71. Xuong, Ng. H., Sullivan, D., Nielsen, C., and Hamlin, R. (1985). Acta
72. Durbin, R. M., Burns, R., Moulai, J., Metcalf, P., Freymann, D., Blum, M.,
Anderson, J. E., Harrison, S. C., and Wiley, D . C. (1986). Science, 232,

1127-32.
73. Howard, A. J., Gilliland, G. L., Finzel, B. C., Poulos, T. L., Ohlendorf, D.
H . , and Salemme, F. R. (1987). Journal of Applied Crystallography, 20,
383-7.
74. Xuong, Ng. H., Nielsen, C., Hamlin, R. and Anderson, D. (1985). Journal of
Applied Crystallography, 18, 342-50.
75. Blum, M., Metcalf, P., Harrison, S. C., and Wiley, D. C. (1987). Journal of
Applied Crystallography, 20, 235-42.
76. Xuong, Ng. H . , Freer, S. T., Hamlin, R., Nielsen, C., and Vernon, W. (1978).
Acta Crystallographica, A34, 289-96.
77. Howard, A. J., Nielsen, C., and Xuong, Ng. H. (1985). In Methods in
enzymology, Vol. 114, (ed. H. W. Wyckoff, C. H. W. Hirs, and S. N.
Timasheff), pp. 452-72. Academic, Orlando.
78. Kabsch, W. (1988). Journal of Applied Crystallography, 21, 916-24.
79. Giessen, B. C. and Gordon, G. E. (1968). Science, 159, 973-75.
80. Klug, H. P. and Alexander, L. E. (1974). X-ray diffraction procedures for
polycrystalline and amorphous materials. Wiley, New York.
81. Bisch, D . L. and Post, J. E. (ed.) (1989). Modern powder diffraction, Reviews
in Mineralogy, Vol. 20. Mineralogical Society of America, Washington.
82. Debye, P. and Scherrer, P. (1917). Physikalische Zeitschrift, 18, 291-301.
83. Seemann, H . (1919). Annalen der Physik, 59, 455-64.
84. Bohlin, H. (1920). Annalen der Physik, 61, 421-39.
85. Guinier, A. (1956). Theorie et technique de la radiocristallographie, pp.
185-202. Dunod, Paris.
86. Gandolfi, G. (1967). Mineralogica Petrographica Acta, 13, 67-74.
87. Parrish, W. (ed.) (1962). Advances in x-ray diffractometry and x-ray spectro-
g r a p h ~ .Centrex, Eindhoven.
88. Parrish, W. and Mack, M. (1967). Acta Crystallographica, 23, 687-92.
89. Mack, M. and Parrish, W. (1967). Acta Crystallographica, 23, 693-700.
90. Ortendahl, D., Perez-Mendez, V., Stoker, J., and Beyermann, W. (1978).
Nuclear Instrumentation and Methods, 156, 53-6.
91. Izumi, T. (1980). Nuclear Instrumentation and Methods, 177, 405-9.
92. Ballon, J., Comparat, V., and Pouxe, J. (1983). Nuclear Instrumentation and
Methods, 217, 213-16.
93. Hanawalt, J. D. and Rinn, H . W. (1986). Powder Diffraction, 1, 2-6.
94. McMurdie, H . F., Morris, M. C., Evans, E. H., Paretzkin, B., Wong-Ng, W.,
and Hubbart, C. R. (1986). Powder Diffraction, 1, 40-3.
95. Hanawalt, J. D. (1986). Powder Diffraction, 1, 7-13.
96. Frevel, L. K. (1965). Analytical Chemistry, 37, 471-82.
97. Sai-Zhu, Z., Li-Jun, C., and Xin-Xing, C. (1983). Journal of Applied
98. Alexander, L. and Klug, H. P. (1948). Analytical Chemistry, 20, 886-9.
99. Chung, F. H . (1974). Journal of Applied Crystallography, 7, 519-31.
100. Zevin, L. S., and Zevin, Sh. L. (1989). Powder Diffraction, 4, 196-200.
101. Rietveld, H. M. (1969). Journal of Applied Crystallography, 2, 65-71.
102. Albinati, A. and Willis, B. T. M. (1982). Journal of Applied Crystallography,
15, 361-74.
103. Hill, R. J. and Madsen, I. C. (1987). Powder Diffraction, 2, 146-62.
104. Hill, R. J. and Howard, C. J. (1987). Journal of Applied Crystallography, 20,
467-74.
105. Buerger, M. J. (1960). Crystal structure analysis, Ch. 7. Wiley, New York.
106. Arndt, U . W. and Willis, B. T. M. (1966). Single crystal diffractometry, Ch. 11.
107. Lipson, H . (1959). International tables for x-ray crystallography, Vol. 11, (ed. J.
S. Kasper and K. Lonsdale), pp. 237-90. Kynock, Birmingham.
108. Azaroff, L. V. (1955). Acta Crystallographica, 8, 701-4.

109. Miyake, S., Togawa, S., and Hosoya, S. (1964). Acta Crystallographica, 17,
1083-84.
110. Hope, H. (1971). Acta Crystallographica, A27, 392-3.
111. Kerr, K. A. and Ashmore, J. P. (1974). Acta Crystallographica, AM, 176-9.
112. Kahn, R., Fourme, R., Gadet, A., Janin, J., Dumas, C., and Andre, D.
(1982). Journal of Applied Crystallography, 15, 330-7.
113. Lipson, H. (1959). International tables for x-ray crystallography, Vol. 11, (ed. J.
S. Kasper and K. Lonsdale), pp. 291-315. Kynock, Birmingham.
114. Cordero-Borboa, A. E. (1985). Journal of Physics E : Scientific Instruments, 18,
393-5.
115. De Meulenaer, J. and Tompa, M. (1965). Acta Crystallographica, 19, 1014-18.
116. Bartels, K. (1977). In The rotation method in crystallography (ed. U . W. Arndt
and A. J. Wonacott), pp. 153-71. North Holland, Amsterdam.
117. North, A. C. T., Phillips, D. C., and Mathews, F. S. (1968). Acta
Crystallographica, A24, 351-9.
118. Kopfmann, G. and Huber, R. (1968). Acta Crystallographica, A24, 348-51.
119. Flack, H. D. (1974). Acta Crystallographica, A30, 569-73.
120. Katayama, C., Sakabe, N., and Sakabe, K. (1972). Acta Crystallographica,
A28, 293-5.
121. Lee, B. and Ruble, J. R. (1977). Acta Crystallographica, A33, 629-41.
122. Takusagawa, F. (1987). Journal of Applied Crystallography, 20, 243-5.
123. Walker, N. and Stuart, D. (1983). Acta Crystallographica, A39, 158-66.
124. Abrahams, S. C. (1973). Acta Crystallographica, A29, 111-16.
125. Abrahams, S. C. and Marsh, P. (1987). Acta Crystallographica, A43, 265-9.
126. Ibers, J. A. (1969). Acta Crystallographica, B25, 1667-8.
127. Blake, C. C. F. and Phillips, D. C. (1962). In Biological ejfects of ionizing
radiations at the molecular level, pp. 183-91. IAEA, Vienna.
128. Hendrickson, W. A. (1976). Journal of Molecular Biology, 106, 889-91.
129. Sygusch, J. and Allaire, M. (1988). Acta Crystallographica, A44, 443-8.
130. Rudman, R. (1976). Low temperature x-ray diffraction. Plenum, New York.
131. Hope, H. (1988). Acta Crystallographica, B44, 22-6.
132. Hope, H. (1990). Annual Review of Biophysics and Biophysical Chemistry, 19,
107-26.
133. Hamilton, W. C., Rollett, J. S., and Sparks, R. A. (1965). Acta
Crystallographica, 18, 129-30.
134. Fox, G. C. and Holmes, K. C. (1966). Acta Crystallographica, 20, 886-91.
135. Rae, A. D. (1965). Acta Crystallographica, 19, 683-4.
136. Richards, F. M. and Berger, J. E. (1962). In International tables for x-ray
crystallography, Vol. I11 (ed. C. H. MacGillavry and G. D. Rieck), pp. 17-20.
Kynock, Birmingham.
137. Blundell, T. L. and Johnson, L. N. (1976). Protein crystallography, pp. 104-6.
Academic, New York.
138. Westbrook, E. M. (1985). In Methods in Enzymology, Vol. 114, (ed. H. W.
Wyckoff, C. H . W. Hirs, and S. N. Timasheff), pp. 187-96. Academic,
Orlando.
139. Matthews, B. W. (1974). Journal of Molecular Biology, 82, 513-26.
Solution and refinement
of crystal structures
Introduction
The goal of a structural analysis is to obtain the distribution of atomic
electron density in the unit cell (in practice the atomic positions) starting
from the diffraction data. As already observed in Chapter 3 (p. 169) it is not
possible to reach this goal in a unique and automatic way, because from the
experimental data only the magnitudes, but not the phases, of the structure
factors may be obtained. Therefore, in order to compute the electron
density by means of eqn (3.45), we must somehow derive the missing
information. In this chapter we shall analyse the most important methods
commonly used to solve the phase problem.
The problem must in principle have a solution (even if not necessarily
unique), since the measured intensities are proportional to the squares of
the structure factors, which may be expressed as?
In these relationships, the number of which is equal to the number of

observed reflections, the terms f and h are known quantities, while the
atomic position vectors r are unknown. Unfortunately these unknowns
appear as argument of trigonometric functions and the solution of systems
of non-linear equations can not be obtained in any analytical way, even
though the number of relationships greatly exceeds the number of un-
knowns (there may be up to 100 reflections per atom).
It is noteworthy that, as in the case of other physical experiments (cf.
Chapter 1, p. 17), the intensities only depend on the interatomic vectors,
which are independent of the arbitrarily chosen reference system.
It is theoretically possible that the solution of equations such as (5.1) is
not unique. Indeed, more than one set of positional vectors r may
correspond to the same set of intensities; these are called homometric
t Throughout this chapter the convention of using capital letters for the I x 3 matrix of the
reciprocal lattice indices will not be followed, in order to conform with the notation generally
used in the literature on Patterson and direct methods, where generally lower case letters
indicate the general reciprocal vectors as well as the matrix of their components. With this
notation no ambiguity should arise between, for instance, the scalar product of the reciprocal
.
vectors h by the direct position vector r, indicated by h r, and the product of the indices
matrix h by the rotation matrix R of a symmetry operator, indicated by hR (the transpose sign
is usually omitted).
320 ( Davide Viterbo
sets.['] Nevertheless, in practice the constraint that the solution must obey
stereochemical rules makes it extremely unlikely that more than one
homometric set is chemically acceptable.
The possibility of solving a system of non-linear equations relies on that
of obtaining a first approximate solution, constituting the so called initial
structural model. This can then be refined until the best agreement with the
experimental data is achieved.
Before considering the different methods employed to define an initial
model, it is therefore necessary to establish the criteria which allow us to
assess its correctness. From the M positional vectors of the model the
structure factors
may be computed. A good agreement between the IFils and the observed
moduli IF,"(, obtained directly from the intensities, will indicate a correct
model. The most common parameter used to express this agreement is the R
index (also called agreement index or residual)
where K is a scale factor bringing IF;] on the same scale of IF:[, obtained as
K = C h JF,"J/C,IF;/.
In the case of equal atom structures, the R value for totally random
atomic positions has been statistically evaluated to be 0.83 for centrosym-
metric structures and 0.59 for non-centrosymmetric structures.[21Structural
models yielding values of the R index lower than these extreme values may
be considered as plausible initial guesses to start the refinement process. In
general a model with R G 0 . 5 if centrosymmetric or 0.4 if non-
centrosymmetric will be a good starting point. It may also happen that the
postulated model contains errors which can not be corrected by the
following refinement, and therefore it does not converge to the correct
solution. A quite frequent case is represented by crystals containing one or
more solvent molecules; if the presence of these molecules in the cell is
overlooked, then the initial model will be incomplete and the index R will
not decrease below 0.15-0.25, unless the positions of the solvent molecules
are taken into account. We will consider the behaviour of the R index in
some more detail in the paragraph on structure refinement.
Historically, the first crystal structures were solved by trial and error
methods, consisting in a systematic trial of all structural hypotheses
compatible with the known physical and chemical properties of the
considered crystal. These methods require a great effort, ingenuity, and skill
and can only be used with simple structures. They are seldom used today
and for this reason they will not be treated (for a comprehensive account
the reader is referred to Lipson and ~ o c h r a n [ ~ IOnly
) . the methods based on
the use of the Patterson function and the so-called direct methods will be
considered here, while those using isomorphous replacement and anoma-
lous dispersion, mainly used in solving biological macromolecular struc-
tures, will be dealt with in Chapter 8.
Solution and refinement of crystal structures 1 321
Before entering into the discussion of these methods, a short digression

will be necessary to describe some important results derivable from a
statistical analysis of the observed intensities.
Statistical analysis of structure factor amplitudes

The statistical analysis of the observed structure factor moduli gives very
useful indications on the presence or not of those symmetry elements which
do not give rise to systematic absences. It also allows us to obtain an
estimate both of the scale factor by which one has to multiply the measured
data to scale them to their absolute value, and of the temperature factor.
Finally it is a basic step in the calculation of unitary and normalized
structure factors, which will be seen to be useful quantities, especially in
connection with the use of direct methods.
The derivation of the theoretical probability distributions of structure
factor amplitudes is given in Appendix 5.A. They are derived[4351 under the
hypothesis that the atomic positions are random variables with uniform
distribution throughout the unit cell (i.e. all points in the cell have the same
probability of hosting an atom).
Their form depends on whether or not the crystal possesses an inversion
center (centric distribution or acentric distribution respectively). For the
two cases we have
Pi(lF1) =
fi exp(- IFI2/2I) (centric),
JGZ (5.4)
2 IF1 (acentric),
PI([FI) = -exp(-I FI2/2) (5.5)
E
where
and E is a factor (defined in Appendix 5.A) depending on the specific

symmetry of the weighted reciprocal lattice. 2, being a function of the
atomic scattering factors, will not have a constant value in all the reciprocal
space, but it will decrease for increasing values of sin 8/A. This is a problem
in the practical use of distributions (5.4) and (5.5). In order to overcome
this problem the unitary structure factors
and the normalized structure factors
are introduced. In the same way by which (IF[') was defined, we have
322 1 Davide Viterbo
and in the case of all equal atoms Z, = 1/N. From (5.7) we can immediately
derive
Both U and E are independent of the scattering angle 8 and correspond

to idealized point atom structures.
The centric and acentric distributions, when expressed in terms of
normalized structure factors, become
which are completely independent of the structure complexity. They are

represented in the curves of Fig. 5.1. The two curves are quite different; the
centric distribution foresees a higher percentage of reflections with extreme
intensity (weak and strong reflections), with respect to the acentric one,
showing a maximum corresponding to intermediate values of the intensities.
The comparison of the distribution of the observed amplitudes with the
theoretical distributions of Fig. 5.1 will allow us to establish the presence or
not of the inversion centre in the crystal under consideration.
From the distributions it is possible to derive the theoretical mean values
of several functions of I El or the theoretical percentage of I El > t. Some of
- these values are reported in Table 5.1 and may be compared with the
lEi corresponding experimental values to verify whether the structure is
Fig. 5.1. Probability distributions of a centrosymmetric or not.
normalized structure amplitude for
centrosymmetric (centric) and for non- So far we have implicitly assumed that the observed structure factor
centrosymmetric (acentric) structures. moduli are on an absolute scale, but in general the values of IFhI:,,,
obtained from the intensities Ihthrough eqn (3.41), are on a relative scale.
Assuming that the thermal motion is isotropic and equal for all the atoms,
we may write
(Fhl:bs= K t I O F ~ ~ ' exp (-2~s') (5.12)
where K' (reciprocal of the scale factor used in eqn (5.3)) is the scale factor,
I°Fhl is the structure amplitude in absolute scale for atoms at rest, B is the
overall isotropic temperature factor, and s = sin OIL.
Table 5.1. Theoretical values of some functions of \El obtained from the centric (5.10)
and acentric (5.11) distributions and their comparison with the corresponding ex-
perimental values for the AZOS structure
Theoretical
Centrosymmetric Non- Experimental

centrosymmetric for AZOS
Using the results of the statistical analysis of the intensities, ~ i l s o n [ ~ ]

proposed a method to derive the values of K' and B.
Let us consider a set of observed intensities falling within a restricted
range of s, such that within this range the decrease of the f s with s may be
neglected. The average value of both sides of (5.12) will be
(JFobsJ2)s ~ ) ~( - 2 ~(s2)) = K' C, exp ( - 2 ~(s2)) (5.13)

= K ' ( I o F J exp
from which
where (s2) is the mean value of sin2 O/A2 in the considered interval and
is computed using the tabulated values of the atomic scattering factors for
atoms at rest for s = m. Dividing the reciprocal lattice into several
intervals of s, (5.14) tells us that a linear relation will exist between
In ( ( I F ~ ~ J ~ ) ~and
/ & (s2)
) and that a plot of these values obtained from the
experimental data can be interpolated by the best straight line passing
through them. The intercept of the line on the vertical axis will give us In K'
and its slope the value of 2B.
Figure 5.2 is an example of such Wilson plot for a typical small organic
structure (p-carboxyphenylazoxycyanide-dimethyl su~phoxide,[~~ AZOS
hereinafter); the numerical values of the terms appearing in eqn (5.14),
obtained from 1908 observed reflections, are given in Table 5.2. The main
reason for the deviations of the experimental points from the straight line is
the breakdown of the condition of equiprobability of all atomic positions,
assumed in deriving (5.4) and (5.5); in fact, the presence of structural
Fig. 5.2. Wilson plot of AZOS together with the

formula of the compound.
Table 5.2. Numerical values of the different terms in eqn (5.14) employed to obtain the
Wilson plot of AZOS shown in Fig. 5.2
Inter- ~imitss' Reflect- (s2) (lF21), Xs ( I F I ~ ) . / ~ . In [(IFI):/~,I

val [on no.
1 0.0000-0.0655 82 0.0327 3736.8 2199.9 1.6986 0.5298

2 0.0327-0.0982 124 0.0655 1359.5 1322.9 1.0277 0.0273
3 0.0655-0.1309 156 0.0982 821.2 1020.9 0.8044 -0.2177
4 0.0982-0.1637 182 0.1309 420.2 791.3 0.5310 -0.6329
5 0.1309-0.1964 210 0.1637 292.8 646.5 0.4529 -0.7922
6 0.1637-0.2292 231 0.1964 234.4 562.7 0.4166 -0.8756
7 0.1964-0.2619 248 0.2292 146.2 507.8 0.2879 - 1.2451
8 0.2292-0.2946 269 0.2619 73.0 450.9 0.1619 -1.8209
9 0.2619-0.3274 272 0.2946 52.0 410.4 0.1267 -2.0663
10 0.2946-0.3274 134 0.3110 46.3 392.7 0.1 179 -2.1378
regularities, such as the phenyl hexagon in AZOS, is in contrast with this

assumption. The broken line in Fig. 5.2, obtained by least-square fitting, has
a slope of -9.20 and an intercept of 0.72. From these values we can derive
that B = 4.60 A2 and K' = 2.05. The normalized structure factors can then
be calculated by means of the relation
\ K' exp (-2Bs2)s Of?
The last column in Table 5.1 gives the experimental values of several
statistical indicators based on the IEl values of AZOS (space group P2,/a),
which confirm the presence of an inversion centre.
The Patterson function and its use

In Appendix 3.A (p. 182) the Patterson function was defined as the
self-convolution of the electron density p(r). According to (3.A.38) we have
P(u) = p(r) * p(-r) = j p(r)p(r + u) d r

V
(5.16)
We have shown (see (3.A.39)) that the Fourier transform of P(u) is IF(r*)I2
(in symbols I ~ ( r * )=l T[P(u)])
~ and vice versa
] r * )I exp (-2nir*
P(u) = ~ - ~ [ l ~ ( r *=) l ~IF(
v*
. U)dr*
1
=- IF,)^ exp (-2nih ' u).
V h
Since IFh/= lFphl, we may write

1
P(u) = - cos 2nh u
V h
and then P(u) = P(-u), i.e. the Patterson function is always centrosym-
metric even when p(r) is not. This is in agreement with the deductions of p.
176 concerning the centrosymmetric nature of all functions with real Fourier
transform. Since 1 ~ ~ 1 depend
' on the interatomic vectors [cf. eqn (5.1)] we
may expect that also P(u) will contain information on these quantities. This
can be verified starting from the definition (5.16). Let us, for simplicity,
suppose we have an idealized structure made up of n point atoms with an
associated weight equal to their atomic number (Fig. 5.3). The integral in
(5.16) then becomes a summation over the n points, and
In order to derive P(u), all the atoms of the original structure p(r,) are
shifted by a fixed vector u to obtain the corresponding p(r, + u), then the
+
products p(r,)p(r, u) are performed and finally all these contributions are
summed; the products will be non-zero only when p(r, + u) #O, that is
when a point in the translated structure (broken lines in Fig. 5.3) coincides
with an atom of the original structure. This condition is verified only when
the vector u coincides with an interatomic vector (in Fig. 5.3(a) u coincides
with the vector 2-4), while for a general u (Fig. 5.3(b)) all point of the
translated image fall into regions where p(r) is zero. In the first case the
value of P(u) will be proportional to the product of the weights of the two
superposed atoms, in the second P(u) = 0. In Fig. 5.3 a higher weight has
been given to atom 1 (heavier atom) and in 5.3(c) the case of a vector u
coinciding with the distance 1-4 is represented; it will be P ( u , , ~>
) P(uZ,~))
both being single Patterson peaks, as each corresponds to only one
interatomic vector. Finally in 5.3(d) the case of a vector u coinciding with
two parallel interatomic vectors of equal length is illustrated; two terms will
contribute to the summation in (5.18) and P(u) becomes twice as large as
P(u',~) and it is said to have a multiplicity of two. Let us, for instance,
suppose that atom 1 is a sulphur and the others are carbons; we will then
have: P(u2,4)= 6 x 6 = 36, P ( q 4 ) = 6 x 16 = 96 and, with reference to Fig.
5.3(d), P(u) = 2 x ( 6 6)=72.
~
From what we have seen so far it follows that the Patterson function will
have maxima corresponding to all possible interatomic vectors within the
unit cell; the height of each peak will be proportional to the product of the
atomic numbers of the atoms connected by the vector u, multiplied by the
multiplicity of the same vector.
This concept can be further clarified by considering Fig. 5.4, where in
5.4(a) a set of N = 5 points is represented, while in 5.4(b) the corresponding
distribution of interatomic vectors, and in 5.4(c) the same set of vectors,
after they have been translated to a common origin, are shown; the last
corresponds to the distribution of peaks in the Patterson function.
The Patterson function will have the same periodicity as the electron
density and therefore the size of the unit cell will be identical. On the other
hand, the number of peaks in the Patterson function is much greater than
that in p(r); given N atoms in the cell they will give rise to N' peaks in
P(u), N of which will superpose on a single peak at the origin (they
correspond to the N zero distances of each atom with itself), while the
remaining N(N - 1) are distributed over the cell. This higher density of
peaks becomes a more serious problem for real structures with non-point-
like atoms. In fact the Patterson peaks are wider than the maxima in an
electron density map. As illustrated in the one-dimensional example of Fig.
Fig. 5.3. Schemes for the construction of the

Patterson function of a point-atom structure, in
which each atom has a weight equal to its
atomic number: (a) two images separated by a
vector coinciding with the interatomic vector
between atoms 2 and 4 of equal weight; (b) two
images separated by a vector not coinciding
with any interatomic vector; (c) two images
separated by a vector coinciding with the
interatomic vector between atoms 1 and 4 of
different weight; (d) two images separated by a
vector coinciding with the two parallel
interatomic vectors of equal length between
atoms 4. 8 and 6. 7.
5.5, because of the non-zero width of the peaks in p ( x ) , the Patterson peaks
will have a width twice as large.
For these reasons the Patterson map of a structure, with even a moderate
number of atoms, may appear as an almost featureless distribution of vector
density. To overcome this problem it is convenient to employ a sharpening
procedure, consisting in computing the Patterson function with coefficients
IEhI2 or better IFhEh(.In fact the normalized structure factors correspond to
a point-atom structure with no decrease of the atomic scattering factor with
increasing sin 8/A.The (l$EhI coefficients are more convenient, because
over-sharpening is sensitive to the series truncation errors (cf. Fig. 3.16) and
may produce spurious peaks or a down-scaling of correct peaks in the map.
It is also possible to eliminate the origin peak, which may obscure some
short vectors, by subtracting from the coefficients the terms corresponding
(c) to the interaction of each atom with itself (i.e. the first term in the
Fig. 5.4. (a) Scheme of a molecule formed by right-hand side of (5.1)); the coefficient in the series (5.17) will be
five point atoms; (b) corresponding
representation of all possible interatomic
vectors; (c) Patterson function obtained by
translating all vectors in (b) to a common origin.
1 ~ sbe in an absolute scale. Figure 5.6 shows

where, of course, the 1 ~ ~ must
the comparison between a normal and a sharpened Patterson section.
From the previous considerations about the height of the Patterson peaks,
we can infer that the map will have prominent maxima when:
1. The structure contains a limited number of heavy atoms (i.e. with an
atomic number Z, significantly higher than that of the other lighter
atoms, ZI), giving rise to three types of peaks:
--
Vectors Height
heavy atom-heavy atom ZpZp very high
heavy atom-light atom Z ~ intermediate
Z ~
light atom-light atom ZIZIvery low
2. The molecular geometry is such that it will give rise to several

interatomic vectors with almost the same length and direction (e.g.
systems with condensed aromatic rings); these will correspond to Fig. 5.5. Illustration of the widening of a
Patterson maximum with respect to the electron
multiple peaks. density maxima contributing to it.
Before illustrating the methods for interpreting the Patterson function, let
us first analyse how the symmetry of the crystal is reflected into the vector
map. The following considerations apply:
1. Since all vectors are translated to a common origin, the symmetry
elements in the crystal will be translated in the Patterson function so that
they will all pass through the origin and at the same time they will lose
their translational components. By applying these rules it can be seen
that, of the 230 space groups, only 24 have a symmetry compatible with
the Patterson function. Thus, for instance, we have:
Crystal space group Patterson group
Triclinic (PI, p i ) pi
Primitive monoclinic (P2, P2,, . . . , P2,/c) P2/m
Centred monoclinic (C2, Cc, . . . , C2/c) C2/m
Primitive orthorhombic (P2,2,2,, . . . , .
Pna2,, . . . , Pbca . . .) Pmmm
2. The symmetry operators present in the crystal leave a trace in the

Patterson function, consisting of particular clusterings of vector maxima
on specific lines or planes of the map, called ~ a r k e r [lines
~ ] and sections.
These are produced by vectors, with one or two constant components,
between equivalent atoms related by symmetry elements other than the
inversion centre. Thus, for instance, the space group P2,, with equivalent
+
positions: x, y, 2; 2,y 4, Z, will have vectors between equivalent atoms
located on a Harker section with coordinates (2x, 4,22). In fact
In the space group P2, the equivalent positions x, y, z and 2, y, Z give

rise to maxima on the Harker section at (2x, 0, 22).
In a similar way it can be seen that the space groups Pm and PC give
rise to Harker lines with coordinates (O,2y, 0) and (0,2y, 4) respectively.
The Harker lines and sections corresponding to some of the most
common symmetry elements are listed in Table 5.3. The presence or lack
of Harker lines or sections may also give valuable indications on the
space group symmetry when this is not uniquely defined.
The heavy atom method

The procedures for the interpretation of the Patterson function are greatly
simplified when the structure contains a limited number of heavy atoms.
The peaks due to the vectors between these atoms will dominate the map
and it is therefore usually quite an easy task to derive the positions of the
heavy atoms. Then, if the atoms have a sufficiently high atomic number, the
information about their coordinates represents a good initial model; in fact,
the term F i computed by relation (5.2) with the contribution of the M
heavy atoms in the unit cell, generally represents the predominant contribu-
tion to the structure factor, as illustrated in Fig. 5.7.
One may then assume that the phase @iof lFtl is a good approximation
of the true phase of Fh and compute an electron density map using as
Table 5.3. Harker lines and sections relative to the most

common symmetry elements
2 axis 1) to a, b, c 0, v, w ; u, 0, w ; u, v, 0
2, axis 11 to a, b, c i,v, w ; u, 4, w ; u, v, 4
Fig. 5.7. Argand diagram i n which t w o heavy
mmirrorI.toa,b,c u,O,0;O,v,O;O,O,w
atoms (with atomic scattering factors f, and f,) a glide I to b, c 4, 4,
v, 0 ; 0, w
and six light atoms contribute t o the structure b glide I to a, c 4,
u, 0 ; 0, $, w
factor F; the resultant of the contributions of the
two heavy atoms is quite close t o F.
c glide 1. to a, b u, 0, $; 0, v, 4
coefficients the observed amplitudes (to which all atoms in the structure will
contribute) with the corresponding calculated phases $.;I The map will not
only reveal the heavy atoms but also other atoms of the structure. In the
most favourable cases the structure may be completed from the first
electron density map, but in general it is necessary to operate in more than
one cycle by the so-called method of Fourier synthesis recycling. Each
cycle requires the calculation of the structure factors from the coordinates of
the known atoms; their phases will then be used to compute a new electron
density map. If the initial model is correct, each cycle will reveal new atoms
until the structure is completed.
From the previous considerations one may get the impression that it
would be advantageous to have compounds containing atoms of high atomic
number, but one has also to consider that their contribution to the diffracted
amplitudes may became so dominant that the observed data will be almost
unaffected by the contribution of the remaining light atoms. The definition
of the final structure will then be rather inaccurate. It has been dem-
onstrated empirically that the best ratio between heavy and light atoms is
that for which
As the values of the ratio (5.20) become less than 1.0, then the
interpretation of the Patterson function and the process of completing the
structure become more and more difficult, but at the same time the accuracy
of the refined positions of the light atoms will increase. As an example let us
consider a hypothetical organic compound of formula C3,H3,04X, where X
is a halogen considered as a heavy atom; then, supposing that there are
two molecules in the cell, C 2: = 2708 and C 2; = 578, 2450, 5618 for
X = C1, Br, I respectively and the ratios (5.20) will be 0.21, 0.90, 2.07
respectively. Supposing that the data measured for the three derivatives are
equally good, the chlorine compound will be difficult to solve, but the
refined structure will be quite accurate; with bromine the solution should be
quite easy with still a reasonable accuracy of the final structure, while with
iodine it will be very easy to solve the structure but its accuracy will be
further reduced.
In order to find the positions of the heavy atoms it is necessary to locate
them with respect to the symmetry elements and to the conventional origin
of the unit cell. Let as now consider some examples, assuming for the
moment that the asymmetric unit only contains one heavy atom.
1. Space group PI. The vector between equivalent atoms related by the
inversion centre, has component u = 2x, v = 2y, w = 22; once it has been
localized on the map it will immediately give the heavy atom coordinates
with respect to the origin chosen on the inversion centre.
2. Space group P2, (twofold screw axis parallel to b). As we have seen,
the vector between equivalent heavy atoms gives rise to a peak on the
Harker section at 2x, i, 22. From its position one may easily derive the x
and z coordinates of the heavy atom; the y coordinate may be arbitrarily
assigned in order to fix the origin along the twofold screw axis.? Let us
t This is correct in the process of finding a starting model formed by a heavy atom, but
during the refinement a more robust way of fixing the origin should be used, as described in
Chapter 2, p. 107.
assume y = i; we will then have two equivalent heavy atoms at positions x,

a,
$, z and R, 2, which are related by an inversion centre (this is true for any
value of y, but the choice y =a makes it more obvious). The structure
factors computed with the contribution of the heavy atoms only will then be
real quantities and the map obtained with the corresponding phases will be
centrosymmetric, showing both the true structure and its mirror image
(enantiomorph). It may be quite difficult to unravel one image from the
other; a sufficient number of light atoms, all belonging to the same image,
should be localized in order to break down the centrosymmetric nature of
the subsequent Fourier syntheses. This may become easier when atomic
groups of known stereochemistry or one or more atoms of intermediate
atomic number (from Na to Cl) are present.
3. Space group P2Jc. This very common space group will be illustrated
by means of the structure of the lower-melting isomer of methyl-
(phenylsulphonyl)furoxan, C 9 H s ~ 2 0 4 ~ [ s 1
The heavy atom is the sulphur and the value of the ratio (5.20) is 0.37.
The general equivalent positions are:
The twofold screw axis gives rise to a Harker section of type %, 4, 4 - 22

(Fig. 5.8(a)), while the glide plane generates a Harker line at 0,4 - 2y,i
(Fig. 5.8(b)). Finally the inversion centre will produce a vector between
equivalent heavy atoms of components 2x, 2y, 22. In Fig. 5.8(a) the highest
peak is at u = -2x = 62.21100, w = 4 - 22 = 26.91100 and then: x = 0.189,
2 = 0.115. In Fig. 5.8(b) there is a large peak centred at v = 4 - 2y =
25.2/100 and therefore y = 0.124. The coordinates of the S atom found in
this way are confirmed by the presence of a fairly large peak at position
u = 2x = 0.384, v = 2y = 0.245, w = 22 = 0.230. Despite the low value of the
ratio (5.20), the electron density map, computed with the signs derived from
the contribution of the S atoms only, allowed the localization of all the
other non-hydrogen atoms. The final coordinates of the sulphur in the
refined structure are: x = 0.2880, y = 0.1249, z = 0.1143 (shifted by only
0.097 A from the original position).
4. Space group P212121. The higher-melting isomer of methyl(pheny1-
sulphonyl)furoxan[81
crystallizes in this common non-centrosymmetric space group.

Fig. 5.8. Methyl(phenylsulphonyl)furoxan LMI:

i.
(a) Harker section u, $, w; (b) Harker line 0, v,
24
28
L 1 I I
21-33 $ 1 I I I I I J '
(b)
Fig. 5.9. Methyl~phenylsulphonyl)furoxanH M I :
(a) Harker section i, v, w; (b)Harker section u, v,
f ; (c) Harker section u, w.:,
The general equivalent positions are
The three mutually perpendicular twofold screw axes give rise to three
Harker sections of the type 4, - 2y, 22; 4 - 2x, 2y, 4; 2x, 4, 4 - 22 (Fig.
5.9(a, b, c)). In the first the highest peak is positioned at v = $ - 2y =
501100, w = 22 = 6.61100, giving y = 0.0 and 2 = 0.033. In the second the
highest peak is at u = - 2x = 38.2/100, v = 2y = 0/100 (in agreement with
the first section) and then x = 0.059 and y = 0.0. In the third section the
peak at u = 2x = 11.81100 and w = 4 - 22 = 43.2/100, confirming the coord-
inates derived from the first two sections, is not the highest peak but the
fourth highest. The largest peak at 50 and 6.6 is in common with the first
section, because of the special value of y = 0; the second and third peaks are
23 27
-
I I I I I I 1
(d Fig. 5.9. (Continued)
probably due to an accidental superposition of several vectors connecting

lighter atoms. Also in this case the Fourier map, computed with the phases
due to the contribution of the S atoms only, revealed the whole molecule
(excluding hydrogens) and the coordinates of the S in the final refined
structure are x = 0.0589, y = 0.0094, z = 0.0343 (shifted by 0.0162 A from
the original position).
Let us now consider the case of a compound with more than one heavy
atom, such as the complex F~,(co),,(Noc~H~)(Nc~H,),[~~ crystallizing in
the space group p i with Z = 2. In Table 5.4 the coordinates of the highest
peaks in the Patterson map are listed in decreasing order of height. The
values of the function are on an arbitrary scale and can be scaled to their
absolute value by multiplying by the ratio C j Z?/P(O 0 0) = 82401364 =
22.6; the height of a single Fe-Fe peak is Z;, = 676 in absolute scale and
about 30 in the relative scale. The vectors between equivalent atoms linked
by an inversion centre have components 2x, 2y, 22 and correspond to single
peaks; we may therefore assume that maxima 10, 11, 12, and 13 are of this
type. If we choose peak 10 as Fe(1) - Fe(1) the coordinates of Fe(1) are
simply obtained by dividing its components by two. Once this has been
done, we are not allowed to simply divide by two the components of peaks
11, 12, 13 to obtain the coordinates of the other Fe atoms. In fact in this
way we wouldn't be sure that the other three Fe atoms are referred to the
same origin as the first. Indeed in the space group PI there are eight
different positions of the inversion centre where the origin may be chosen.
In order to resolve this ambiguity the peaks corresponding to vectors
between non-equivalent Fe atoms should be used, also taking into account
the periodic nature of the Patterson function. For instance, for peak 11, we
must consider the eight combinations:
giving for Fe(2) the coordinates referred to the eight possible origins:
Only one of these positions will account for the vectors Fe(1) - Fe(2) and
+
Fe(1) Fe(2) in the Patterson. In our case it will be the last position; in fact
u2 = 1.380, v 2 = 1.300, w2 = 1.645 will give x2 = 0.690, y2 = 0.650, z2 = 0.822.
Peak 3 and peak 5 may then be interpreted as Fe(1) - Fe(2) and
+
Fe(1) Fe(2) respectively, after these Fe(2) coordinates have been transl-
ated by -1.0 along the three axes to obtain the values listed in the bottom
Table 5.4.List of the highest Patterson peaks of the complex

Fe,(CO)11(NOC2H,)(NC2H5) and their interpretation in terms of vectors between Fe
atoms; Fe coordmates obtained from this interpretation
Peak u v w H Interpretation
Origin 0 0 0 364
1 0.080 0.260 0.1 10 144
2 0.020 0.000 0.710 138
3 0.490 0.560 0.470 120
4 0.070 0.275 0.400 72
5 0.1 10 0.245 0.820 65
6 0.485 0.455 0.810 65
7 0.420 0.180 0.420 63
8 0.405 0.290 0.360 59
9 0.480 0.570 0.760 55
10 0.400 0.195 0.700 38
11 0.380 0.300 0.645 33
12 0.430 0.285 0.075 30
13 0.450 0.165 0.135 27
part of Table 5.4. With a similar procedure we may deduce the coordinates
of the remaining two Fe atoms listed in the table.
Advanced Patterson methods

Procedures for the automatic determination of positions of heavy atoms
from the Patterson function, have been developed both for small
molecule^[^^^^] and for biological macromolecules.~1~161 The procedure
implemented within the SHELXS-86 computer program[lOlwill be briefly
described. The list of the highest peaks in the Patterson map is analysed in
terms of Harker and cross peaks. Several possible solutions are given and
the selection of the 'best' solution to be further processed (by the methods
described on pp. 365-75) is done by means of two reliability indices (or
figures of merit): an R-factor based on E values (RE) and R(Patt) which
measures the agreement between the observed and predicted (from the
atomic numbers) Patterson values.
We finally mention that there are other methods for interpreting the
Patterson function, which may be applied also in the absence of heavy
atoms. They are the so called vector and superposition methods which are
described in some detail in Appendix 5.B.
Direct methods
Introduction
With the term direct methods are indicated those methods which try to
derive the structure factor phases directly from the observed amplitudes
through mathematical relationships. In general the phase and the amplitude
of a wave are independent quantities and in order to understand how, in the
case of X-ray diffraction, it is possible to relate these two quantities, two
important properties of the electron density function should be considered:
(1) it is everywhere positive, i.e, p(r) 3 0 (positivity);
(2) it is composed of discrete atoms (atomicity).
The relation between positivity and phase values may be simply under-
stood by just imagining the computation of p(r) of a centrosymmetric
structure as a Fourier series, first with all signs correct and then with all
signs reversed: the first map will be everywhere positive or zero, while the
second wiIl be negative or zero and therefore physically unacceptable. Two
pictorial examples of how positivity restricts the possible values of the
phases are described in the Appendix 5.C, while a more formal explanation
will be given later.
Historically, the first mathematical relationships capable of giving phase
information were obtained, in the form of inequalities, by Harker and
as per['^] in 1948 and then further developed by Karle and ~ a u p t m a n [ ' ~ l
and by other authors. Because of their limited practical interest, they will
not be treated here and the reader is referred to more specialized
textbooks.[191In 1953 Hauptman and Karle['O1 established the basic concepts
and the probabilistic foundations of direct methods; the great power of
these methods in solving complex crystal structures had its highest recogni-
Fig. 5.10. Comparison between p ( x ) and p 2 ( x )

for a one-dimensional structure with equal and
well resolved atoms.
tion in the Nobel Prize for Chemistry conferred in 1985 on the mathe-
matician H. Hauptman and the physicist J. Karle.
Also in 1953 ~ayre,['~] using the atomicity condition, was able to derive a
very important relation. He considered that for a structure formed by well
resolved and almost equal atoms, the two functions p ( r ) and p2(r) are quite
similar and show maxima at the same positions. A one-dimensional example
is illustrated in Fig. 5.10.
We have seen that the Fourier transform of p ( r ) is ( 1 I V ) f i and for the
case of all equal atoms
N
Fh =fh
j=1
.
exp ( 2 n i h q ) . (5.21)
We can also define the structure factor corresponding to p2(r)

N
Gh = gh exp ( 2 n i h q ) (5.22)
j=1
where gh is the scattering factor of the 'squared' atom.

The Fourier transform of p2(r) is ( l / V ) G h and, because of the convolu-
tion theorem, it will correspond to the convolution product ( l / V ) F h *
( l / V ) F h . Since Fh is a discrete function defined only at the nodes of the
reciprocal lattice, the convolution integral (3.A.34) becomes a summation
From the ratio of (5.21) and (5.22) we also have

Fh = (fhlgh)Gh = OhGh,
and (5.23) becomes
Oh
Fh = y c FkFh-k,
k
which is Sayre's equation, valid for both centrosymmetric and non-

centrosymmetric structures. Multiplying both sides of (5.24) by F-,, we
obtain
For large values of IFhJthe left-hand side will be large, real, and positive. It
is therefore likely that the largest terms in the sum on the right will also be
real and positive. It follows that, if IFkl and IFh-kl also have large values, it
will be
@hk= 9)-h+ 9)k+ 9)h-kzO (5.26)
which for centrosymmetric structures becomes
where S(h) stands for the sign of reflection h and the symbol = stays for
'probably equal'. We note that (5.27) coincides with the indication obtained
in Appendix 5.C.
Relations (5.26) and (5.27) are expressed in a probabilistic form and
indicate the necessity of applying probability methods to estimate their
reliability. On the whole, the use of probability techniques to obtain
relationships between phases and magnitudes, has proved to be the most
important approach for the practical use of direct methods. We will
therefore describe in more detail these methods and the procedures
employed for their practical applications.
Structure invariants and semi-invariants

As already mentioned, the goal of direct methods is that of obtaining the
phases directly from the observed amplitudes. These last quantities are
independent of the chosen reference system, while in general the phases
depend on it. From the observed amplitudes we can only obtain informa-
tions on single phases or linear combinations of phases which are independ-
ent of the choice of origin (the other features of the reference system being
already defined by the choice of the space group). Since their value depends
only on the structure, they are called structure invariants (s.i.).[203221
The most general s.i. is represented by the product
when
Let us show that its value does not change when the origin is moved by a
general vector ro. The structure factor of index h, referred to the new
origin, will be
FA =
N
x fi
',
.
exp (2nih (5 - rO))
= Fh exp {-2nih rO)

=1
41 exp [i(qh - 2nh ro];
the modulus remains unchanged while the phase changes by Aq, = 2nh ro. .
The variation of the phase of the product (5.28), due to the same origin
shift, will be
which is zero because of condition (5.29).

The simplest structure invariants are:

N
1. Fooo= C Zigiving the number of electrons in the unit cell; its phase is
j=1
always zero.
2. F h R h= IFhI2, which does not contain any phase information.
3. F-,FkFh-, with phase q-h + qk+ called triplet invariant; we shall
see that these invariants play a primary role in the procedures which will
be described later.
4 . F-hFkF,Fh-k-l with phase g,-h + qk+ g,, + Vh-k-l called quartet
invariant; its importance has been recognized in recent years.
5. With a simple extension it is possible to define quintet, sextet, etc.,
invariants. Considering that the practical use of these relations is still
under discussion, we will just mention here that complex probabilistic
formulae have been derived to allow the estimation of these higher
invariants.
Structure semi-invariants ( S . S . ) [ ~ ~are
, ~ ~single
] phases or linear
combinations of phases which are invariant with respect to a shift of
origin, provided that the position of the origin is restricted to those
points in the cell which possess the same point symmetry (the so-called
'permissible origins'). A basic property of a S.S.is its capability of being
transformed into a s.i. by adding one or more pairs of symmetry-
equivalent phases. For instance, in a given space group possessing the
symmetry operator C = (R, T), the phase qHis a S.S. if it is possible to
find a reflection h such that
V = T H - Q ) ~ + Q ~ (5.32)
+
is a s.i., i.e. H - h hR = 0.
Because of (3.37), relation (5.32) may be written as
q is independent of any choice of origin, while (p, and T will depend on

it; however, the dependence is such that (5.33) holds and therefore the
value of y H will not change if we move the origin within those points
which maintain the vectors T unchanged (identical point symmetry). For
instance in P?, g,,, is a S.S. since
is a s.i. The origin is conveniently chosen on the inversion centres and

then T = (0,0,0). The permissible origins are on the eight inversion
centres present in the cell; any one of them, chosen as origin, will always
give T = (O,O, 0).
Alternatively, by applying (5.31) we can derive that the phase of g,,
.
changes by Ag, = 2n2h ro when the origin is shifted of a vector ro. It is
easy to see that Ag, = 0 (or 2nm) when the components of ro are 0 or 1 as
happens for the eight inversion centres in (cf. Fig. 5.17). The
qualitative example relative to the sign of F,,, described in Appendix
5.C, can now be related to a S.S.
As another example, let us consider, in space group P2,, the phase

q2h,0,21,
which is a s.s., because
is a s.i. for any value of k. When the origin is chosen on a twofold screw
axis, then q = Q)2h,0,21 - nk. The permissible origins are located on any of
the fqur screw axes present in the unit cell (cf. Fig. 5.18).
The two examples given above refer to single phases and are therefore
one-phase semi-invariants. We can generalize what we have seen so far
to the case of the linear combinations of more phases. Thus, the
combination C $h is a S.S. in PI if C h = 2H (i.e. if the three components
of the sum vector are all even); in P2, it is a s.s. if C h = (2H, 0, 2L). A
compact way to indicate these two conditions is: Ch = 0 mod(2,2,2)
(meaning that each of the three components of the vector Ch gives a
zero rest when divided by 2), and Ch = 0 mod (2,0,2) (with the same
meaning as before for the first and third components, while the second
must be zero). In general we may write C h = 0 mod o,, where o, is a
vector, called semi-invariant modulus, with integer components; the
vector C h is called the semj-invariant vector.
In order to identify which phases or combinations of phases are s.s. one
can refer to special tables (see, for example, ~ i a c o v a z z o [ ~in~ ]which
) the
space groups are classified in such a way that those belonging to the same
class have the same permissible origins. The same process may be carried
out automatically on a computer following an algebraic approach.[23]
Before considering probability methods, let us use the positivity property
of the electron density function to obtain an indication on the value of
triplet invariants. The development of the product F-hFkFh-k gives
F-hFkFh-k
= z
il
fjl exp [-2nih . q,] x
i2
fj, exp [-2nik q,]
x
i3
fi, exp [-2ni(h - k) .q,]
For simplicity let us assume that reflections h, k, and h - k have similar

sip B I A values and therefore the variation of the atomic scattering factors
with the indices can be overlooked. In (5.36) we can separate the terms:
2) I I 2 fifj,exp {2ni[(h - k) . (q, - ql)l>,

ilh
which is proportional to (cf. (5.1)) IFh-,I2- C, f;;

which is proportional to IFk12- f;;

i
which is proportional to IF,)^ - 2 f;;

i
+ [k . (q, - q,)]} = R.
If the electron density is positive, the atomic scattering factors, which are
the Fourier transform of the electron density around each atom, will also be
positive and the summation in (1) will then be positive. Recalling equation
(5.A.9) the terms (2), (3), and (4) are proportional to lFh12- (lF12))
IFkI2- ( 1 ~ 1 ~and
)) - ( 1 ~ )respectively
~ ) and for large values of IF1
they will be large and positive. The last term R is the only complex term,
but being a sum of positive and negative quantities, it will on average be
small; R may be considered as a 'noise' term. For large values of the
structure amplitudes we can then write
F-hFkFh-k= +
C fj + K { J F ~+~ ' IF^^^+ IFh-k12- ~ ( I F I ~ ) }R 3 0 . (5.37)
i
In the case of a centrosymmetric structure this relation implies that, when
IFh/, IFk[, and IFh-kl are large, then the product of the three signs of the
structure factors, S(h)Sfk)S(h - k) = + . This deduction is in agreement
with (5.27), derived from Sayre's equation, and with the pictorial derivation
given in Appendix 5.C.
Probability methods
For the same reasons considered on pp. 321-2, from now on we will use
the normalized structure factors. The use of normalized amplitudes is also
suggested by (5.37), which shows that the value of the triplet does not
depend on the diffraction angle. Indeed we may note that at low sin 8/A,
where the scattering power is higher, both the positive terms and the noise
term R are larger, and that they decrease by comparable amounts as the
diffraction angle increases. It is also important to point out that the use of
normalized structure factors, corresponding to a point-atom structure,
implicitly corresponds to a sharpening of the atomicity assumption for the
electron density.
In Appendix 5.D the probability formulae for triplet invariants are
derived, here we will just consider the main results. For a non-
centrosymmetric structure the distribution associated to (5.26), derived by
~ o c h r a n , ' is
~ ~given
] by (cf. (5.D.17))
where:
(1) for equal atoms
Ghk = ( 2 / @ ) IEhEkEh-kl;
Fig. 5.11. Trends of the probability distributions

(5.38) for different values of the parameter G,,,
given by (5.39).
(2) for non-equal atoms

Ghk= 2 0 ~ 0 2 ~
IEhEkEh-kl;
'~ (5.39b)
N
with on= C 27, 2, being the atomic number of the jth atom. In (5.38) L is
j=1
a normalization term depending on Ghk only. P(Qhk) is a so-called von
Mises distribution[251and Ghk is its concentration parameter; its trend is
very similar to that of a Gaussian, as it is shown in Fig. 5.11 for different
values of Ghk. All curves have a maximum at Qhk= 0, i.e, qh= qk+ qh-,
and their width around the most probable value narrows as Ghk increases.
The variance of the relation
will therefore decrease as Ghkincreases.

If more than one pair of phases qkj, qhpk1 are known (with associated
amplitudes I Ekjl and I Eh-,]I ,), with j = 1, 2, . . . , r, all defining the same
phase qh through triplet relations such as (5.40), then the total probability
distribution for qh is given by the product of the r distributions of the form
(5.38), which are assumed to be independent of one another; i.e.
where A is a normalization factor. The exponent of (5.41) may be

developed as
where
ah= [(iG, cos oj

j=1 TI1"
and
t
tan ph =
(iG, sin c+)
'=I
(C G, cos
j=1
with G, = Ghk, and mi = qk,+ qh+. Equations (5.42) to (5.44) are easily
I understood when the r relationships of type (5.40) are plotted on an Argand
1 diagram as vectors of modulus Gi and phase mi; the angle between each
vector and the real axis is an indication of a probable value of qh. In Fig.
5.12 the case with r = 5 is illustrated; it can be seen that
r r
H AH = a,, sin Ph = C G, sin m,, OH = ahcos Ph = 2 G, cos LC)~
Fig. 5.12. Vector representation, in the complex
plane, of the combination of five triplets of type and relations (5.42), (5.43), (5.44) become immediately clear.
(5.40) involving the same reflection h.
Finally eqn (5.41) becomes
which still is a von Mises distribution with a maximum for qh= Ph and a
variance[261depending on ah as shown in Fig. 5.13. For instance we may
deduce that for ah= 2, (9,) = Ph f 50", while for a,, = 10, ( q h ) = Ph It 19".
Equation (5.44) gives the most probable value of qh and is known as the
tangent formula;[271as we shall see later, this formula plays an important
role in the phase determination process.
In the case of centrosymmetric structures the probability that the sign
relationship (5.27) is true, is given byCz8](Appendix 5.D)
P' = 4 + $ tanh[(~,a;~"IEhEkEh-kl)]. (5.46)
When several relations for the same S(h) exist
the product of the different probabilities P(Eh) (given by (5.D.19)) should

be considered. With a procedure similar to that described in Appendix 5.D,
we obtain
Figure 5.14 shows the trend of (5.48); it can be seen that, when several
terms, all with the same sign, contribute to the summation, then the
absolute value of the tanh argument may become rather large and P f
approaches the extreme values 1 or 0.
In the past 10-15 years probability methods have seen new important
developments. Not only it has been possible to improve the estimate of
triplets, but also to derive reliable estimates of other phase relationships,
t
02
0 4 8 12 16
both s.i. and s.s.
At the basis of the new approaches stands the following principle: 'It is
Gora
possible to obtain a good estimate of s.i. or s.s. given "appropriate" sets of
Fig. 5.13. Trend of the standard deviation, o,of
the distribution (5.45) as a function of n,, normalized structure factor moduli, which are statistically the most effective
defined by (5.43). in determining the value of the given s. i, or s.s. '
Fig. 5.14. Trend of P+ = $ + f tanh ( x ) (cf. (5.48))

as a function of x.
The first task will then be to identify these moduli, indicated as the
phasing magnitudes, and rank them according to their effectiveness in
estimating s.i. or s.s.. Given their set {JEl),the second task will be to derive
the probability distribution
P(@ I {IEI)) (5.49)
where @ is any phase relationship we want to estimate and the vertical bar
after it stands for: 'given all magnitudes in {\El)'.
Cochran's formula[24]for triplets, derived in Appendix 5.D, is a trivial
example of such conditional distribution; in fact it may be seen as
P(@hk I IEhl, IEkl, IEh-kl).
~ c h e n k [ ~extended
~ , ~ ~ ] this principle to the case of four-phase s.i.
(quartets):
Q = Q7h + Q7k + TI + Q7-h-k-P (5.50)
Their distribution, given the four associated magnitudes, derived by
~ i m e r s k a , [indicates
~~] that Q = 0, with a variance depending on
= (2/N) IEhEkElEh+k+lI . (5.51)
Because of the 1/N factor any reasonably sized structure will have very
small B values, and for this reason quartets estimated in this way can not be
used in practice. Schenk pointed out that quartet (5.50) could be considered
-
+
as the sum of two triplets, such as TI = ~ 7 , ~ 7 , - qh+, and
- = ql
+ qh+&.Then, if also I Eh+,I is large, we will have TI 0 and G = 0
+
and therefore Q = TI T, 0 with a strengthened reliability with respect to
+
that indicated by (5.51). In a similar way we can see that the same quartet
can be written as the sum of two other pairs of triplets, and that Q also
depends on JE,+,I and IEk+,(. We can then say that the quartet not only
depends on the four basis magnitudes IEhI, \E,I, (Ell, IE,+,+,I, but also on
the three cross magnitudes (Eh+,I, IEh+lI, JEk+ll.If the last three moduli
are also large, then the indication Q = 0 is strengthened. Empirically it was
-
also found that, when the cross magnitudes have very small values, then
Q n (since cos Q = -1, these are called negative quartets).
Later ~auptrnan[~'] derived the probability distribution of Q, given the
seven basis and cross magnitudes, and confirmed Schenk's empirical
findings. He then formulated the neighbourhood principle,[331the concept
of neighbourhoods being very close to that of phasing magnitudes. ~ a t e r [ ~ ~

he also gave rules for deriving the neighbourhoods of the main s.i. and of
some S.S.in the most common space groups.
The representation theory proposed by Giacovazzo[35~361 is a generaliza-
tion and a systematization of the previous ideas, as it gives precise general
rules for identifying the phasing magnitudes, allowing, at the same time, a
completely general use of the space-group symmetry.
The first representation of a s.i. is formed by the invariant itself plus
all symmetry equivalent s.i. differing from by a constant angle, which is a
function only of the symmetry operators. The set of all basis and cross
magnitudes which appear in the first representation forms the first shell of
phasing magnitudes (first phasing shell).
This definition will be clarified through some examples. The importance
of taking the space-group symmetry into account is first illustrated by the
case of a special triplet in the space group P2,2,2,:
if the first reflection has kl even and I , odd and restricted phase 0, n , the
-
second has h2 even and phase 0, n , and the third has consequently phase
f n/2, then T = fn/2, in contrast with the indication T 0 given by (5.38)
when the three normalized amplitudes are large. The symmetry operators in
P212121 are: R, = I, T, = (0,0,0); R2 = (7, I , 1) (only the diagonal elements
are indicated, all the others being zero), T, = (1,0, 1); R, = (1, i , i ) ,
T3 = (4, 4,O); R, = (T, 1, i ) , T, = (0, 4, 4); and a triplet equivalent to T may
be written as
because 1, is odd and k, even. T and T', forming the first representation of
the triplet, have the same G but opposite phase and their combination in a
phase diagram (similar to that of Fig. 5.12) will give a null vector, which
does not contradict the expected restricted value T = f n / 2 . The general
use of the space-group symmetry in estimating triplets has been described
' ~ also showed that for non-primitive cells the l l f i
by ~ i a c o v a z z o , [ ~who
factor in (5.39a) should be replaced by l / m P , where N, is the number of
atoms in the primitive cell.
Let us now consider the example of a quartet such as (5.50). As we have
seen, Q in general depends on seven magnitudes, and when no special
symmetry conditions exist, the first representation is formed just by Q and
the seven magnitudes form the first phasing shell. On the other hand, if,
besides the identity, there is an other symmetry operator C, = (R,, T,),
which leaves some special reflections unchanged ( H = HR,) and if one of the
cross reflection (e.g. h + k) is of this special type, then the quartet
is equivalent to Q [Q' - Q = 2n(h + k)T,], but with two new cross terms
IEhR,+rland IEkRS+,I.In this case the first representation is formed by Q and
Q ' and the first phasing shell contains nine magnitudes. Since the larger the
number of phasing magnitudes the better the estimate of the s.i., a
significant advantage is obtained by considering also Q'.
A numerical example will further clarify the above ideas; let us consider
the quartet
Q = q 1 5 3 + 4)&1 + (PzI2 + Q)ji6 (C~OSS:
504, 365, 643)
in the space group P2,/c, for which R, = I, R, = ( I , 1, ?), R3 = (I, ?,I),
R4= (1, ?, 1). The first cross reflection is such that (504)R4= (504), and an
equivalent quartet may be set up
with two new cross terms.

As an example of second representation of a s.i. let us consider the case
of a triplet T. Its second representation is defined as the set of all special
quintets of type
where H is any reciprocal lattice vector for which IEHI is large. Since the
term added to T is null, then C = T, but the quintet will depend on four
basis magnitudes JE,I, IEkl, IEh+kl, (EHl and on six cross terms IEh*Hl,
IEk+nl, IEh+kfHI. If M vectors H are selected (80-100 reflections with
largest IEHI), then the second representation will be formed by M quintets
and the second phasing shell will contain 10 x M magnitudes.
Let us now consider a general n-phase s.s.
In general (the few exceptions are beyond our scope) it is possible to find a
phase qHand two symmetry operators Ci and Cj such that
is a structure invariant; h r indicates a reflection equivalent to h. Since q1

differs from @ by a constant term depending only on the symmetry
operators, the estimate of q, gives at the same time the value of @. The
collection of all s.i. q , , obtained by varying H , Ci and Cj forms the first
representation of @ and the first phasing shell will contain all the associated
basis and cross magnitudes.
Let us now consider a few examples regarding one-phase s.s., for which
(5.55) reduces to (5.32). As we have seen, in P?, the S.S. q2, may be
represented by the triplet (5.34), which alone will form its first repre-
sentation with the corresponding phasing shell including the two magnitudes
lEzhl and IEhl. The probability that EZhis positive was derived by Cochran
and ~ o o l f s o nand
[ ~ ~is~given by
which is greater than 4 when (EhI> 1 and approaches 1 as IEZhJand lEhl

increase.
Similarly in the space group P2, the first representation of the s.s. q2h,0,21
is given by the set of triplets (5.35) and the probability formula will be a
function of all the magnitudes in the first phasing shell: IE2h,0,211 and
{IEh,j,l()with j assuming all possible values of the index k within the set of
measured reflections. In P2, one-phase s.s. are special reflections with phase
restricted to 0 or ;rt and the probability formula will be of the tanh type; we
then have
P+(EZh,O,U)= 1 + tanh : ( :030;312 IE2h,0,21\ C (-l)*(IEhk,I2 - I))
k
(5.57)
where the sum is over all values of k. The term (-l)k=exp(nk)
corresponds to the constant angle nk and is obtained by applying (3.37) to
express the phase relation between qhkland qgki in (5.35). When all major
terms in the sum have k even, then P+(E2h,o,21) > $ and S(2h, 0, 21) = +,
when they have k odd, then P+(E2h,o,21) < $ and S(2h, 0,21) .= +.
Formulae (5.56) and (5.57) and those for some other space groups were
derived long before the concept of representation was introduced, and they
were known as the C formulae[201.
The second representation of a S.S. will also be illustrated by means of an
example. In space group P212121let us consider the s.s. @ = q4O6) for which
the first representation (5.55) reduces to the set of triplets
with variable k index. With an extension similar to that used to derive

(5.53), the second representation is defined as the set of quintets
with q varying over all reflections with IE,I large. The second phasing shell
will include IEHI = IE4,,I, a term IEhI = IE2,,I for each triplet (5.58) and, for
each quintet, lEql and the four cross terms IEH+,I, IEhRfql.
Once the phasing magnitudes have been identified it becomes possible to
derive the appropriate probability distribution (5.49). So far formulae have
been derived for estimating:
(1) one-phase s.s, by the second representation;[3w11
(2) two-phase S.S.by the first
(3) triplets by the second representation (PI0
(4) quartets by the first r e p r e s e n t a t i ~ n . [ ~ ~ , ~ ~ I
In most cases, the derived probability distributions for non-
centrosymmetric structures have the form of a von Mises function as (5.38)
and for centrosymmetric structures the tanh form of (5.48); the concentra-
tion parameter G or the tanh argument are now substituted by more
complex functions of the considered phasing magnitudes, which will not be
reported. Only their practical use will be illustrated in the following
sections.
Fixing the origin and the enantiomorph

In the previous sections we have seen how it is possible to derive the values
of s.i. and s.s. from the observed magnitudes. In order to pass from these
values to those of the individual phases it will be necessary to fix the
origin, and, in non-centrosymmetric space groups, it will be necessary to
choose between the two enantiomorphic forms compatible with the ob-
served amplitudes; in fact, since all the phases change their sign in going
from one form to the other, the same will apply to s.i. and S.S. For this
reason all probability formulae, such as (5.38), only define the cosine (i.e.
the absolute value) of the s.i. and S.S.In order to obtain all phases referred
to one form it will be necessary to fix the enantiomorph. The phase
determination process can then be summarized as shown in the flow-
diagram of Fig. 5 .l5.
The choice of the space group with the corresponding set of symmetry / Estimate i s . and s,s!s
I
operators already imposes some restrictions on the points where the origin
may be localized. Indeed, only the sites with the same point symmetry will
be suitable and represent the so-called permissible origins.
We will then have to choose the origin among the permissible ones and I O r i g i n (and e n a n t i o m o r p h ) I
this may be performed by fixing the value of a limited number of suitable I definition I
phases. In order to show how this is possible, let us first consider a
one-dimensional structure (Fig. 5.16(a)) formed by three atoms of different
atomic numbers ( 2 , = 3, Z 2 = 2, Z3 = 1 and C 2, = 6). The unitary structure
factor for this structure, referred to an origin at Xo, will be
Fig. 5.15. Flow diagram of the phase
determination process by direct methods.
3
(i/6) Z, sin 2nh(x, - Xo). (5.60)
j=1
When the origin is shifted so that Xo varies from zero to the identity period
a, the trigonometric terms in (5.60) will change and Ah, Bh, and therefore
Uh, will follow these changes. The modulus JUhl does not depend on the
position of the origin, and only the phase qh will depend on X,. Figure
5.16(b) illustrates this variation in the case h = 1: we can see that at Xo = 0,
q1= 109", while at Xo = 0.7, ql = 0". A change of enantiomorph is achieved
by inverting the direction of the x axis; again 1 Uhl will not change, while all
phases will change their sign ( q + 360 - q).
Since there is only one value of Xo for which q1= 0°, then the origin is
uniquely defined by fixing q, = 0". In Fig. 5.16(c) the variation of q2 with
X, is also shown: it can be seen that q2goes to zero at two points, Xo = 0.26
170 292 374 85 177 299 92 , Xo h.2 Fig. 5.16. (a) One-dimensional structure made of
three non-equal atoms; (b) values of the phase
2i6 3b8t40 131 223 3 k 5 it 4658' I U-s 0 . 4 4 .
ra. for different origin positions; (c)
I I 8
(c) corresponding values of the phase p,.

and Xo = 0.76, at a distance a12 from each other. Fixing q, = O" does not
define the origin uniquely, but only restricts its possible position to two
points; in general, by fixing qh= 0°, the origin is restricted to h possible
positions. It should finally be observed that at X o = 0.7 (position for which
q1= 0") q2= 315", and a change of enantiomorph will give q2= 45". It can
then be seen that the enantiomorph can be chosen by restricting the value of
q, (or in general of another phase q h ) within the interval 0-n (or, as in our
case, n-2n).
Let us now generalize to three dimensions the procedure just described to
fix the origin in one dimension. We will first consider the space group PI, in
which any point of the unit cell is a permissible origin and therefore no
semi-invariants which are not at the same time invariants exist. In analogy
with the one-dimensional case, we can uniquely fix the origin along the x
axis by fixing the phase of the (100) reflection and thus restricting the
possible origins to lie on planes parallel to (100). Similarly the origin may be
fixed in the other two directions by fixing the phases q,,, and q,,. The
three reflections (loo), (OlO), and (001) define a primitive cell in the
reciprocal lattice, but in a triclinic lattice there are infinite ways in which a
primitive cell may be chosen. Indeed, any three non-coplanar vectors
will define a primitive cell if

. .
HI Hz A H3= V* = a* b* A c* = volume of the reciprocal cell.
Using (5.61) and developing the mixed product we can easily obtain
?Z
The three vectors will then form a primitive cell if
A=
(1 1: :I
h2 k2 12 = f l .
(5.63)
By fixing q,, = 0" we will restrict the possible origins to lie on planes
parallel to the crystallographic planes of indices HI. If we also fix q,,, = 0°,
the possible origins will be at the same time restricted to lie on planes
parallel to the planes Hz, i.e. they will lie on the intersection lines between
+
the two sets of planes. Finally, by fixing q,, = O" (with H, # mH, nH2, m,
n positive or negative integers) the origin will be further restricted to be at
the intersection points of the above lines with the planes parallel to the H,
crystallographic planes. The number of such points within the unit cell is
given by the value of A and only when the primitivity condition (5.63) is
Fig. 5.17. Position of the eight distinct inversion obeyed, the three phases will fix the origin in a unique way.
centres in a P I unit cell, corresponding t o the
positions of the permissible origins, numbered In the space group PI the permissible origins lie on the eight distinct
as i n Table 5.5. inversion centres in the unit cell (Fig. 5.17). When the origin is shifted, for
Table 5.5. Sign variations for the reflections divided into parity groups, when the origin
is placed at the different inversion centers in the P1 cell of Fig. 5.17
Origin Parity
instance, from (O,O, 0) to (i,0, O), the phase cpH of the reflection H = (hkl)
will change by -nh; this is equivalent to saying that the sign of the
reflection will change or not depending on whether h is odd (u for ungerade
in German) or even (g for gerade). Any change of origin among the eight
permissible ones will have an effect on the sign of a reflection H which will
depend on the parity of the three components h, k, or 1 (cf. Table 5.5).
Reflections of type ggg are structure semi-invariants and may not be used to
distinguish among the possible origins. If we consider a reflection of
different parity, e.g. ggu, by imposing that its sign must be +
, we restrict
the possible origins to lie on four points (in Table 5.5 these are the origins 1,
3, 5, 7). In order to further reduce the ambiguity we will have to fix the sign
of a reflection of different parity (not ggg), e.g. ugg; with reference to Table
5.5, by fixing the sign to be + , the possible origins are restricted to points 1
and 3. In order to fix uniquely the origin, we will have to fix the sign of a
third different reflection. Its parity should not only be different from that of
the two already chosen, but it must also be different from ugu (for which
both origins 1 and 3 have a + sign); in fact the combination ggu ugg + +
ugu = ggg is a s.s. We will, for instance, choose uug.
The above rules given for PI are also valid for all primitive centrosym-
metric space groups with symmetry not higher than orthorhombic.
Similar procedures may be devised for the other space groups. If along a
given direction the origin is restricted on points separated by 4, then it can
be fixed by fixing the phase of a reflection with an appropriate parity. When
the origin can be shifted in a continuous way along an axis, it is possible to
fix it by using a reflection the phase of which only takes unique values within
the corresponding unit period.
Thus, in space group P2, (Fig. 5.18), with the b axis parallel to the
twofold screw axis, the permissible origins are all the points on the four 21
axes at (0, y, 0), ($,y, 0), (0, y, i), (4, y, 4). The choice of the twofold screw
axis corresponds to that of one of the four inversion centers on the
projection along y ; the projection reflections h01 have restricted phases 0,
n. The phases q,,,, are S.S. and indeed the (gOg) crystallographic planes pass
through all permissible origins. O n the other hand, a phase quo,will have,
for instance, a zero value if the origin is chosen at (0, y, 0) or (0, y, i), and a Fig. 5.18. Projection along the y axis of a P2, cell
n value if the origin is on the other two screw axes. By fixing quo,= 0" the with the twofold screw axis parallel t o b.
origins are restricted on the first two 2, axes. A second phase of type qgoU
(or q~,,,) will assume a zero value at (0, y, 0) and a n value at (0, y, i), and
by fixing qgoU = 0" the origins are restricted to lie on the first screw axis. In
order to fix uniquely the origin along the y axis, we will have to fix a phase
of type q,,,, because the (hll) planes intersect the y axis only once within
the period b.
Let us finally consider the very common space group P2,2,2,, in which the
permissible origins are located at the eight points midway between the three
orthogonal, non-intersecting 21 axes. The origin may be fixed in a simple
way by using reflections belonging to the three principal zones (Okl), (hol),
and (hkO). These reflections have phases restricted to two values: 0, n or
k n / 2 depending on whether the index following the zero, in a cyclic way, is
even or odd (cf. Chapter 3 , pp. 157-8). It is therefore possible to apply
the rules derived for the space group PI: the origin may be fixed by fixing
the phases of three zone reflections belonging to three linearly independent
parity groups (not ggg).
In the examples considered so far, we have always used three phases to
fix the origin, and this is true for all primitive space groups up to the
orthorhombic system. In the centred space groups some of the permissible
origins are related by translational symmetry and are indistinguishable. For
this reason the number of phases needed to fix the origin is reduced, as, at
the same time, is reduced the number of allowed parity groups. Thus, for
instance, in a C-centred lattice all ugg, ugu, gug, and guu reflections are
systematically absent and the origin is fixed by fixing the phases or the signs
of two reflections belonging to the other four parity groups.
Going from one enantiomorph form to the other will change the sign of
all individual phases and, as a consequence, the sign of all linear
combinations of phases. Therefore, the most general way of fixing the
enantiomorph will be that of restricting a suitable s.i. or S.S. within the
interval 0-n.
Let us for instance, consider the space group P21. Suppose that the two
phases q,, and qZl3have been assigned zero value to fix the origin; they
form a triplet invariant with qjil
By restricting 0 S vji1 G n also 0 G Q, S n and, if its value is sufficiently far

from the extreme values zero or n,this choice will fix the enantiomorph.
Unfortunately the most probable value of Q, is zero and in this space group
it may be difficult to properly fix the enantiomorph by restricting the phase
of one reflection. This problem does not arise in the space group P2,2,2, as
will be shown by the following example. Suppose that the origin has been
fixed using reflections of type: uOg (with phase 0, n ) , guO (0, n ) and Ouu
( f n/2). Since all linear combinations of phases for which the sum of the
indices is ggg, are S.S. in this space group, it is possible to set up S.S. with
value restricted to f n/2; the choice of one of the two values allows an
unambiguous choice of the enantiomorph. For instance
Vuo, + Vu,o = fn / 2
is a two-phase s.s.; if quo,is fixed to be zero, then the sign of the s.s. will
depend on the value chosen for q,,,. By choosing q,,, = n / 2 we fix the
enantiomorph.
For a complete and general theoretical treatment of the procedures for
fixing the origin and the enantiomorph in all space groups, the reader is
addressed to more specialized textbook^.['^,^^]
A very effective way of defining the enantiomorph, and at the same time
the absolute configuration (see later), is represented by the experimental
measurement of some triplet phases with value close to fn/2, by multiple
diffraction experiments (cf. Chapter 3, p. 191). In general the availability of
a certain number of experimental triplet-phase values, which unfortunately
are not easy to obtain, would be of great help in solving a structure by direct
methods.
Phase determination procedures

We will first describe some of the steps which are common to all direct
methods practical procedures.
Normalization
By the method described on p. 323-4 and using equation (5.15), the values of
the normalized structure factors are first calculated. Most computer
programs will supply a list of reflections sorted in decreasing order of JEl
and perform a statistical analysis of the normalized amplitudes as shown
earlier. In some cases a more detailed statistical analysis is carried out in
order to reveal the presence of pseudo-translational symmetry (cf. Appen-
dix 5.E).[4s521
Some of the most recent programs allow one to introduce the available a
priori information, such as the existence of pseudo-translational symmetry
or the coordinates of a previously located fragment;[53,541 examples of the
use of this information will be given later.
Setting up phase relationships

We will focus our attention on triplets, the procedures for setting up other
s.i. or S.S. being similar.
Since the most important relations are those estimated with the highest
reliability, we will only set up triplets relating reflections with large JEl.The
minimum IEI value is chosen in such a way that the number of reflections
employed is approximately equal to ten times the number of atoms in the
asymmetric unit and usually lies between 1.2 and 1.6.
For each reflection h all pairs of reflections k and R-k with large IEl must
be found; the set of all relationships obtained in this way is often called the
C P listing. The triplet search must take into account the reciprocal lattice
symmetry and the relations between the phases of equivalent reflections.
If all the indices are reconducted within the same independent portion of
the reciprocal lattice, each reflection h will have to be combined with all
vectors fkRi, obtained by varying R, over all rotation matrices of the space
group not related by an inversion centre; if also the JEl value of reflection
h f kR, is large then the triplet is retained. In general this third reflection
will not be in the considered independent portion of reciprocal lattice, and it
can be expressed as fHR,. The phase relationship will then become
Most computer programs store triplet relations in the form
where n h , nk, n, represent the sequential numbers of the corresponding

reflections in the sorted list, with the sign indicated by (5.64), A 9 =
+
2n(kTi HT,) and G is either defined by (5.39) or it is the parameter
obtained by means of the second representation. For centrosymmetric
structures (5.64) reduces to
and in (5.65) the signs are of no use, while G represents the tanh argument
in (5.46) or the corresponding value from the second representation.
Definition of an optimum starting set of phases

From (5.64) or (5.66), given the two phases or signs in the right-hand side
we can derive an estimate of the left-hand side phase, the accuracy of which
will increase with increasing G. We have seen that the origin may be fixed
by assigning an arbitrary value to up to three phases. Even if these are
chosen among the reflections involved in large numbers of triplets, in
general they will not be sufficient to start up the phase determination
process. It is necessary to start with a larger number of phases, and the
problem is not overcome even when a limited number of reliably estimated
one-phase S.S. (usually fewer than five) is available. Two methods are used
to define the values of the extra unknown phases necessary to start up the
process. With the symbolic addition[ss2261 method these phases are assigned
a symbolic value and all other phases are determined as combinations of
these symbols; it is more suitable for centrosymmetric structures as we shall
see later through an example. With the m ~ l t i s o l u t i o n [method,
~ ~ ~ ~ ~the
~
unknown starting phases are directly assigned permuted numerical values.
This method is easier to automate and can more readily incorporate some of
the most recent developments in direct methods; for this reason it is more
widely used and will be described in more detail.
Given the phases in the starting set, all other phases are determined one
after the other in a chain process. The choice of the reflections in the
starting set is therefore very important and the best starting set will be that
allowing one to eliminate, as much as possible, the weak links in the chain,
i.e. the poor determination of some phases by few low-reliability relation-
ships. An optimal starting set of phases can be defined by means of the
convergence procedure.[s71We have seen that a,, as defined by (5.43) is a
measure of the reliability with which a phase cph may be determined
combining several phase relations. If the vector representing ahin Fig. 5.12
is rotated by an angle Ph = ( G h ) in such a way that it falls on the real axis,
Fig. 5.19 is obtained; it can be seen that
where Qz, = ( q h )- qkl- q h p k l ZQhk1. Since at the beginning the phases are
not known, a,,can not be computed, but it can be estimated a priori by
Fig. 5.19. Figure 5.12 rotated by an angle Ph =

(v,) in order t o bring the vector ah on t o the
I real axis.
substituting each cos Qj in (5.67) with its expected value, defined as

1 2" Il(Gj) -
(COS6 , ) = cos 6, exp (G, cos 9)d6, =--- - Dl(Gj) (5.68)
Io(G,)
where Io(x) and Il(x) are modified Bessel functions of zero and first order.
These functions and their ratio are readily available[581and Fig. 5.20
illustrates the trend of (5.68) as a function of G ; it may be seen that the
ratio tends asymptotically to 1.0 as G increases.
The estimated value (cu,) is then given by
and it can be computed for each reflection, before any phase information
has been obtained, using the G values only.
The convergence process is a step process in which, at each step, the
reflection with minimum (a,,) is (temporarily) eliminated, provided it is not
an accepted one-phase S.S. or another reflection already included in the
starting set. When a reflection is eliminated, at the same time, all phase
relationships contributing to it are eliminated and the (a,,) values of the
other reflections involved in these relations are updated. Since, at each step,
the reflection which is less related to the remaining reflections is eliminated,
the process must converge towards the group of reflections which are most
strongly interrelated and are therefore the most effective in starting the
phase determination process.
Fig. 5.20. Trend of the ratio, I , ( G ) / I , ( G ) ,of the

modified Bessel functions of first and zero order,
as a function of G.
Before eliminating a reflection, the remaining ones are analysed to check

if it is still possible to fix the origin; if not, the reflection is not removed but
kept in the starting set as an origin-fixing reflection. If, in the final steps, a
reflection is eliminated with an (a,,),which, at the elimination moment, is
reduced to zero (or to a very low value), than this reflection is not (or is
only poorly) connected with the remaining few reflections; on the other
hand it is an essential link between these last reflections and those
previously eliminated. By putting this reflection into the starting set, we
assign a value to its phase and in this way we insert the missing link (or
replace the weak link) in the chain process. The convergence process is
repeated several times, each time adding a new reflection into the starting
set, until the desired number of starting-set reflections has been reached.
The flow diagram of Fig. 5.21 illustrates the convergence procedure. The
procedure may be further improved by assigning to each contributor to the
summation in (5.69) a weight depending on (ct.,) and (a,-,) .[591
Calculate <a)for all reflections -

A
Find reflect~on,not already i n t h e

startina set , w i t h smallest <d>
t
Yes
+
[/canorigin\
be defined without
\this reflect ion?/ N7
data, update all c ds f o r origin d e f i n i t i o n .
ref Iections
A / I
reflections t o
Yes
r
reflection w i t h small~st
c d > at t i m e of elimination
goes into starting set, t-
Fig. 5.21. Flow diagram of the convergence
procedure.
Figures of merit
As we shall see, the phase determination process usually leads to more than
one solution. Given several sets of phases it would be rather time
consuming to compute and interpret all the corresponding electron density
maps to see which yield the correct structure. It is instead easier to compute
some appropriate functions, called figures of merit (fom), which allow an a
priori estimate of the goodness of each phase set. Several functions have
been proposed[60]and we will analyse those most commonly used.
MABS (absolute fom) represents a measure of the internal consistency of
the employed triplet relationships in estimating the phases. It is defined
as
for a correct structure A should be close to the theoretically estimated A,

and MABS = 1.0. In practice it has been found that often for the correct
set of phases A > A , and MABS values between 0.9 and 1.3 indicate a
promising phase set.
R, fom: this is a measure of how much triplets deviate from their
expected statistical behavior and is defined as
~ c =e lo0(z
h - (ah)~)be; (5.71)
it should be minimum for the correct set of phases.
tpo fom: this is defined as
where the outer summations are over a certain number (50-150) of

reflections with lowest lEll value. The triplet-generating routine will have to
set up also the relationships ( q Otriplets) linking each of these 1 reflections
with pairs of reflections k and 1 - k with large (El, the phases of which have
been determined. The inner summation at the numerator corresponds to
Sayre's equation (5.24) (written here in terms of normalized structure
factors) relative to each reflection 1. Since [Ell= 0, the large terms in the
summation must tend to cancel each other out and the correct set of phases
will correspond to a minimum vo
value. This function differs from the
previous foms because it is not a self-consistency figure among the
determined reflections only, but it establishes the coherence of the obtained
phases with some weak reflections not used for their determination.
4.SSlFOM, SS2FOM, NTREST, NQUEST: these four functions have
the same functional form:
with j = 1 for one-phase s.s., j = 2 for two-phase s.s., j = 3 for negative

triplets, j = 4 for negative quartets respectively, and r running over all
estimated relationships of each type; as shown in (5.68), D,(lGjrl) is the
expected value of cosQjr. For instance in the case of negative quartets
(5.73) becomes
where the summation is extended to all quartets estimated to be negative

(G, < 0.0) by means of their first representation and Q, are the values of the
individual quartets computed with the determined phases. The correct set of
phases should give rise to the largest number of negative cos Q, values, thus
maximizing NQUEST. Similarly for the other three functions, a good
agreement between the estimated (through the Gj values) and the calculated
cos Qj values, maximizes their values. The four foms based on phase
relationships, are finally combined to give
TI + T2+ T3+ T4
CPHASE =
Bl+BZ+B3+B4'
A better understanding of the statistical meaning of the MABS, R, and
qofoms was recently achieved by a more rigorous probabilistic approach,[60]
which also allowed the derivation of some new effective functions. At the
basis of this approach is the use of the probability distributions P(ah) of ah
defined in (5.43) and P(a;) of a; defined in (5.72), derived by Cascarano et
a1.[611On p. 322 we saw how the expectation values in Table 5.1 could be
obtained from the distributions (5.10) and (5.11). By a similar method, from
P(ah) and P(a;) it is possible to derive various theoretical moments, which
may then be compared with the corresponding 'experimental' values
computed using the different sets of phases; MABS and R, are related to
moments of P(ah), while qo is related to a moment of P(a;), but the
expected values of other appropriate functions of a,, and a; may also be
derived and used as foms.
Once the different foms have been calculated for each phase set it is
useful to derive a combined figure of merit, CFOM; its capability of
discriminating the correct set of phases will in general be higher than that of
the individual functions. Different ways of combining the individual foms
have been used; a more objective procedure is described by Cascarano et
a1.[601in which each function is introduced with a properly estimated weight
and the resulting combined fom is normalized to fall between -1.0 and 1.0.
A correct set of phases should lead to CFOM==1.0, while CFOM << 1.0
should indicate a wrong solutioa.
The electron density map(s) corresponding to the most promising phase
set(s) (with highest CFOM) will be calculated first.
Electron density maps (E-maps)

In the calculation of the electron density maps it is more convenient to use
as coefficients of the Fourier series, E rather than F. In fact the reflections
with large ( E l , for which phases are determined, are mainly of two types
Table 5.6. AZOS: list of the 50 reflections with largest IEl value
No. h k I E No. h k I E
(this may be verified in Table 5.6): some low-angle reflections (small

indices) with very large IF( and several high-angle (large indices) ones with
relatively small (F1. The contribution of these small IF( to the electron
density would be obscured by that of the low-angle reflections, and by using
the (El this problem is overcome and the resolution of the maps is
improved. On the other hand, also for the limited number of terms in the
Fourier summations, the E-maps often show spurious peaks.
The programs for the calculation of the electron density maps usually also
perform an automatic peak search and supply a list of maxima sorted in
decreasing order of height. This list may then be analysed in terms of
distances and angles among the peaks; peaks related by a sensible chemical
geometry may be selected and labelled as a possible molecular fragment.
The peaks of each fragment, labelled by their sequential numbers, may then
be represented in projection directly on the line-printer output, where it will
be easy to reconstruct the image of the molecular fragment by connecting
the peaks related by possible chemical bonds (cf. Figs. 5.22 and 5.23).
Recently an automatic link between the E-map output and a structure
refinement program has been established;[621to this aim the map inter-
pretation procedure has been improved and the peaks are assigned an
atomic symbol. For most structures it then becomes possible to go from the
diffraction intensities to the refined molecular picture without human
intervention.
Unfortunately the phase set with the highest CFOM does not always yield
an E-map interpretable in terms of a chemically plausible structure. In these
cases it will be necessary to compute some other E-maps, with the hope that
another statistically less favoured phase set will yield the correct structure.
Besides, in the case of complex structures, the E-maps often only show a
partial image of the structure, which will then have to be completed.
We may now illustrate the most common phase determination
procedures.
Symbolic addition method
This method will be illustrated through an example. We will use the same
AZOS structure employed earlier (p. 323); the 50 reflections with largest
IEJvalue are listed in Table 5.6. When using all sign relations of type (5.47)
relating these reflections, the convergence procedure defines the following
starting set of signs:
No. h k 1 Sign Parity
Origin-fixing
3 0 1 2 + gug
reflections
8 1 1 2 + uug
16 10 4 1 + ggu
Other reflections 1 4 0 2 a
with symbolic 2 1 6 0 4 b
sign 12 8 6 4 c
In the unit cell there are four molecules of row formula C10HllN304Sand
o~o;~" = 0.141. The probability with which a sign is determined is given by
(5.46) or (5.48) and it can be seen that it is never less than 0.95, the
minimum tanh argument being 0.141 x ( 2 . 2 1 ) = ~ 1.522.
The sign expansion procedure may be followed in detail in Table 5.7. At
step (17) a new symbol d must be introduced in order to define new signs; at
step (20) we have two different, but not contradictory, symbolic indications
for the sign of the same reflection and their comparison suggests that
symbols b and c correspond to the same sign. Similarly at step (22) we have
the indication that a and b represent opposite signs. Not all the signs of the
first 50 reflections are determined; 10 signs are not defined with only four
symbols. At this point it is not convenient to introduce new symbols, but
rather to use other reflections with smaller [ E l . The probability of the
individual sign relations will decrease, but, with 40 signs already defined, we
will have several multiple indications. For AZOS 200 reflections with
(El Z- 1.61 were used. As the sign determination is carried out, several other
indications about the values of the symbols are obtained: the indications
b = c and a = - 6 are confirmed several times and new indications that
a = - and b = + (in agreement with a = - b ) are obtained. No indication is
obtained about symbol d. Out of the 24 = 16 possible sign combinations,
only two are consistent with the previous indications: - +++ and
- ++ - . Also some contradictory indications are obtained; for instance,
+
S(165) has seven contributions: - , ab, ab, ab, - , - , . The first six
indications confirm the relation a = -6, while the seventh, with very low
probability, is clearly contradictory with respect to the others and it will be
neglected. If the determination of one or more signs turns out to have
several contradictory indications of similar probability, it should be sup-
posed that a wrong choice has been made at some previous step and the
procedure should be reconsidered from the beginning. In our case this did
not happen and it has been possible to define the signs of all reflections
Table 5.7. Step by step illustration of the symbolic addition procedure for the 50
of Table 5.6
Step Relation Sign Comments and indications
abc
a b c
symbol a cancels out
a b c
abc
abc
introduced because fewer
than half the signs have been
determined
a d Two different indications

a b c d +
implying that bc = or
a b c d b = c; abcd is chosen
- b c d
- b c d Confirm b = c
- d
Indication that ab = -
or a = - b
Table 5.7. (Continued)
Step Relation Sign Comments and indications
-a bcd
-a b c d
-a bcd
-a bc
-a bc
-a bc
+
+
+
+
+
+
a
a
a
- bc
- Again b = c confirmed
-a bcd
-a bcd
- b c d
- b c d
a b
a b
using four symbols. When the most probable values are substituted for the
symbols, two sets of signs are obtained for AZOS and the corresponding
electron density maps may be computed. The values of the MABS, R,, and
I/JO foms indicate that the set with d = - is more reliable and, indeed, the
corresponding E-map shows the two fragments interpreted in Fig. 5.22 in

terms of a p-carboxyphenylazoxycyano group (5.22(a)) and a dimethylsul-
phoxide group (5.22(b)). The structure obtained in this way will then be
refined with the procedures described in the next paragraph.
The symbolic addition method, both for centrosymmetric and for
non-centrosymmetric structures, has also been automated into a number of
computer programs, the most general of which is SIMP EL;[^^@] but, SO far,
the most striking results[651have been obtained by I. and J. Karle using a
careful manual procedure proposed in 1 9 6 6 . ~ ' ~ ~
Multisolution methods
The basic idea of these methods is that of assigning approximate numerical
values to the starting phases, instead of using symbols. It has been found
empirically that initial errors of 40-50" usually do not spoil structure
solutions. The starting set, defined by the convergence procedure, will
include, besides the origin and, when needed, enantiomorph fixing reflec-
tions, also a limited number of other phases necessary to initiate the phase
expansion process by means of the tangent formula (5.44). If these are
general phases with values anywhere between 0 and 2n, we may tentatively
Fig. 5.22. AZOS: representation and give them the four quadrant values: fn/4, f37~14.One of these will be
interpretation of the E-map computed with the
best sign set obtained by application of the correct within 45". All restricted phases are assigned the values defined by
symbolic addition procedure. the space-group symmetry; e.g. 0, n , or fn/2.
If ng and ns are the numbers of general and special reflections in the

starting set, the total number of combinations of numerical values to be
developed by the tangent formula, will be
This number grows very rapidly with increasing ng and ns and only by
limiting the number of starting-set reflections it is possible to maintain the
computing time within reasonable limits. This limit can be greatly reduced
by using the so called magic integers (described in the Appendix 5.F),
which allow a considerable reduction of the number of combinations with a
minimum increase in the phase error.
Most multisolution computer programs, such as MULTAN,[~~] are mainly
based on the use of triplets estimated by Cochran's formula, to which a few
one-phase s.s. estimated by the El formulae may be added. With these
programs it is possible to solve structures with up to 60-70 atoms in the
asymmetric unit. Although more complex structures have also been solved,
it may also happen that simpler structures can not be solved. In fact,
because of the rather crude probabilistic estimate of triplets, it may happen
that at some stage of the 'chain' phase-expansion process, some triplets with
an actual value quite far from zero are used; as a result the determined
phases are completely wrong. In the recent years several new developments
have been proposed in order to overcome these problems and to make
direct methods capable of solving increasingly more complex structures. In
Appendix 5.G the most promising developments of the multisolution
techniques are outlined, while in the following we will describe, through a
practical example, the use of the SIR (semi-invariant representation)
program;[671this program is based on the multisolution strategy strength-
ened by the use of all phase relationships for which a reliable estimate may
be obtained by means of the representation method.
We will follow in detail the solution of the structure of the antibiotic
21-acetoxy-11-(R)-rifamicinol (RIFOL),[~~] C39H49N013.CH30H.H20,
which crystallizes in the space group P2, with Z = 2. Its structural formula is
Normalized structure factors are computed by the normalization routine

and 362 reflections with JE1> 1.723 selected for phase determination; the
non-centrosymmetric space group is confirmed by the statistical analysis and
no pseudo-translational symmetry (cf. Appendix 5.E) is detected.
None of the 17 one-phase s.s. estimated by means of the second
representation has a sufficiently high tanh argument to be accepted and
these indications are only used to compute SSlFOM. 326 two-phase s.s. are
362 / Davide Viterbo
estimated using their first representation; all of them are used to compute
SS2FOM, while 129 with J G J>0.6 are actively used in the phase
determination procedure.
Triplet invariants relating the 362 strongest reflections are set up and
estimated by means of their second representation (PI0 formula).[4s1The
concentration parameter of the von Mises distribution is given by
where G is the Cochran parameter defined by (5.39) and R is a function of

all the magnitudes in the second phasing shell. R may also be negative and
for some triplets a negative G' value may be obtained, indicating that their
expected value is n. The reliability of the negative estimates is in general
too small for an active use of negative triplets in the phase determination
process, but a passive use to compute an effective fom proved to be very
useful. Besides, the capability of the PI0 formula of sampling out most
triplets with a value far from zero, allows their elimination from the phase
determination process, which then becomes more stable. For RIFOL 2280
positive triplets estimated with G ' a 0.6 are actively used, while 141
negatively estimated triplets are employed to compute the NTREST fom.
At the same time 1399 I#, triplets relating the 100 weakest reflections with
the considered 362 strongest reflections are generated, to be used in the
calculation of the corresponding fom. Finally also 1009 negative quartets are
estimated by their first representation and the most reliable 500 are used to
compute NQUEST.
The convergence procedure is carried out using positive triplets and
two-phase S.S. When also some one-phase s.s. are accepted, these are
treated as known phases and directly included in the starting set. Two-phase
s.s. are treated in the same way as triplets. In fact for each reflection h
forming a S.S.with k, we have
where S2 is the value of the S.S. estimated with a reliability depending on

the concentration parameter G. The contribution of (5.77) to (a,)will be
given by G . D,(G). For RIFOL the convergence procedure defines the
starting set reported in Table 5.8. It may be noticed that the origin is fixed
following the rules given earlier (p. 349) for the space group P2,. All
permuted phases are general and are expressed in terms of magic integers,
giving rise to 24 phase permutations (cf. Appendix 5.F, enantiomorph fixed
by the magic integer procedure).
Table 5.8. RIFOL: starting set of phases defined by the convergence procedure
Type Code h k Phase
Origin 5 5 0
15 2 0
4 6 1
Permuted 18 6 4
72 10 4
149 3 1
25 1 2
17 8 4
Total number of permutations 24
In the SIR program a weighted tangent formula is used, i.e.

C Gtw sin ( ~ k+, ~7h-k,)+ C Gs% sin ( ~ k+*S2s)
S
tan qh= --
- Th (5.78)
C Gtwcos(qk,+qh-k,)+C G s ~ C O S ( Q ) ~ , + SBh~ ~ )
S
where the summations over t extend to all triplets linking qh to two other
known or previously determined phases, while the sums over s refer to
two-phase S.S. The weight attributed to each relation is calculated from the
+
a = ( T ~ B')"~ values of the reflections contributing to At the
beginning the origin-fixing reflections and the restricted permuted phases
are given a = 100, one-phase S.S. (if any) are assigned their G values, while
the general phases represented by magic integers are given an a value which
depends on the root mean square deviation of the representation. Starting
from the eight reflections in the starting set, the phase extension is carried
out following the order indicated by the convergence procedure. As we
have seen, the reflections eliminated at the end are those best related to the
starting set; the phase determination process should therefore be carried out
in an order inverse to that of elimination. In Table 5.9 the final part of the
inverted convergence map (divergence map) of RIFOL is reported and
with its aid we can follow the phase determination process.
Reflection 24 is the first to be determined by a single triplet relation
(written in the form of (5.65)); it will be used to define the following
reflections with its corresponding a value. The second reflection no. 99
forms a two-phase S.S. with 24 (written as a triplet but with an asterisk
instead of the third code number) and is related by a triplet to the starting
set; each summation in (5.78) will have one term. The remaining reflections
are then determined by a similar chain process.
When a sufficiently large number (60-100) of phases has been deter-
mined, one should proceed to their refinement. It is in fact possible to
redetermine the initial phases with a greater number of contributors to the
sums in (5.78); for instance the phase of reflection 24 can now be
determined using all relations involving 24 and two other reflections among
the 60-100 determined ones. The tangent refinement process is repeated
until self-consistency and then the remaining phases are determined and
refined. At this point different foms are computed.
Table 5.9. RIFOL: divergence map illustrating the phase determination path starting
from the set of reflections in Table 5.8
Code h k I (a) Contributors
24 4 1 -17 8.41 4 -15 0 8.94

99 8 1 -19 11.01 24 * 0 6.00 4 15 0 6.09
40 2 5 14 9.50 4 17 x 8.10 72 99 x 5.72
188 4 5 13 13.75 4 72 x 6.35 17 24 x 6.15
15 40 0 5.78
31 3 1 -1 7.80 -24 25 x 6.69 -5 99 0 3.71
13 3 2 -17 10.46 15 25 0 7.29 4 31 x 6.02
135 7 6 12 8.60 13 72 0 4.88 17 25 0 4.42
31 188 0 3.92
7 7 0 -19 6.72 5 15 0 7.26
44 9 5 -5 11.47 7 40 0 7.26 5 188 0 5.28
37 3 4 14 10.22 -5 17 0 8.37 -7 72 0 6.40
Table 5.10. RIFOL: list of the ten best sets of phases with their relative figures of merit
Set Figures of merit
MABS ALCOMB PSCOMB CPHASE CFOM
1 1.093 0.920 0.925 0.885 0.91 1

2 1.090 0.843 0.921 0.878 0.885
3 1.016 0.388 0.683 0.479 0.554
4 1.056 0.457 0.624 0.440 0.539
5 1.032 , 0.348 0.601 0.495 0.517
6 1.052 0.422 0.554 0.431 0.502
7 1.051 0.432 0.538 0.435 0.501
8 1.049 0.401 0.566 0.431 0.500
9 1.044 0.383 0.567 0.431 0.496
10 1.054 0.429 0.533 0.426 0.495
The complete tangent procedure is repeated for each of the 24 permuta-

tions of the phases in the starting set. This step may become quite
demanding of computer time when the number of phase permutations is
increased. Table 5.10 gives the list of the 10 best sets output by SIR in
Fig. 5.23. RIFOL: representation and decreasing order of c F ~ M . ALCOMB and PSCOMB are combinkions of
interpretation of the E-map computed with the
first and best set of phases i n Table 5.10; only
functions based on a and qo respectively. The first set with the highest
part of the molecule can be located.
CFOM yields the E-map shown in Fig. 5.23, which may be interpreted in
terms of the RIFOL structure as indicated by the connected circled maxima.
Not all atoms are found in the map (solvent molecules plus eight atoms are
missing, two to close the macrocycle and the others in the side chains), but
the structure can be easily completed by the methods described in the next
paragraph. The final refined structure is shown in Fig. 5.24.
Partial structure recycling

It may happen that in the best E-map only a small molecular fragment can
be recognized and in this case direct methods can be used to complete the
structure. The first and simplest procedure was proposed by ~ a r l e [and
~ ~is]
known as Karle recycling. With the coordinates of the fragment the / Carbon 0
corresponding structure factors are computed in modulus IF;( and phase qi,
using (5.2). The most accurate phases, defined according to the criteria Oxygen 0
Fig. 5.24. RIFOL: final structure, without solvent
E l a p 1G1 and I J % ~1.5 ( p = ErnZrn/C, 2,) molecules; the fragment outlined by the broken
line can yield the complete molecule when input
are then used as starting set in a tangent-formula extension and refinement as a partial structure.
process. p represents the fraction of scattering matter in the fragment and is
forced within the interval 0.25-0.65. If the fragment is a sufficiently good
initial model the procedure should yield an E-map showing a larger
fragment and further recycling should lead to the complete structure.
Other procedures for completing a partial structure have been proposed
and among them we recall the DIRDIF procedure,[70]which is based on
difference structure factors, i.e. structure factors corresponding to the
unknown part of the structure; with initial phases obtained from the partial
structure, the difference structure factors are used in a weighted tangent-
formula procedure. In SIR a new probabilistic meth~d['~,'~] has been
implemented. The structure factors for the known fragment are calculated
and appropriate pseudo-normalized structure factors for the known and
unknown part of the structure are obtained. These are then used in the new
probabilistic formulae for the estimation of s.i.; on the basis of the
estimated a s computed by these formulae, a convergence procedure will
define the best starting set of phases, which will include both permuted
phases and phases calculated from the partial structure. Phase expansion
and refinement is performed by means of a new weighted tangent formula,
and two specific foms, related to R, and q O , but computed with the
pseudo-normalized structure factors, are used to select the correct solution.
The procedure was also tested on the RIFOL structure; starting from the
eight-atom fragment (11 per cent of the total number of electrons)
surrounded by a broken line in Fig. 5.24 the complete structure was
recovered.
Completing and refining the structure

The structural models obtained both by Patterson and direct methods are
often incomplete (in the sense that not all atoms have been localized) and in
all cases they only represent a crude first approximation of the real
structure.
The method of Fourier recycling was outlined earlier (p. 329) in
connection with the problem of locating the light atoms when the initial
366 / Davide Viterbo
model is formed by one or more heavy atoms. The same method can be
used to complete a molecular fragment when all atoms have approximately
the same weight. If 50-60 per cent of the electron density has been located
with sufficient accuracy it is quite easy to complete the structure. If the
initial model only contains a smaller percentage of the electron density, the
method can still be applied but the Fourier coefficients should be corrected
by appropriate statistical weights, such as those proposed by ~ i m [ ' l , ~ ~ ] ,
taking into account the different contribution of the known atoms to the
different structure factors. The derivation of these weights is given in
Appendix 5.H, where some other procedures for completing a partial model
will also be described. The Fourier cycles not only allow the location of the
new atoms but also the improvement of the positions of the model atoms.
For very small fragments the direct method procedures mentioned at the
end of the last paragraph are usually easier to apply.
Difference Fourier method

Another convenient way of completing and refining a structural model is the
difference Fourier synthesis method. A Fourier series having as coefficients
the ( F a s computed using eqn (5.2)
1
pc(r) = - F i exp (-2nih r)
V h
will show maxima at the positions of the atoms of the given model, while a
series with coefficients Fz = IFzI exp (icp,,,,)
1
p,,(r) = -
V h
Fi exp (-2nih .r)
represents the true structure. In order to see how much the initial model
deviates from the real structure, the difference series
1
Ap(r) = p,(r) - pc(r) = ; ( F i - F i ) exp (-2nih . r) (5.81)
should be computed. Unfortunately the values of cp,,, are not known and
we have to assume cp,,, = cpi; this approximation, illustrated in the Argand
diagram of Fig. 5.25, will hold better the better is the initial model.
Equation (5.81) then becomes
1
Ap(r) =
" h
(IFgI - IFiI) exp (-2nih .r + icpi).
If in the model an atom is missing, then p,(r) will be zero at the
corresponding position, while p,(r) will show a maximum. The difference
synthesis will also show a peak at the same position but it will be almost
D
zero at the positions of the model atoms (if these are correct) where
PO@> = pc(r).
An important property of the difference syntheses is that they are almost
unaffected by series truncation errors. Indeed, because of the limited
number of observations, the Fourier maps computed by means of (5.79) and
(5.80) will show some ripples around each peak (see Fig. 3.16), the size of
Fig. 5.25. Mustration of the approximation which inci-eases with increasing peak height. As a consequence a light atom
C
%truez vh. close to a heavy atom may be obscured by its ripples. Since the number of
terms in the two series (5.79) and (5.80) is the same, the truncation errors
will also be approximately the same and will cancel out in the difference
(5.82).
Let us now see how the different types of errors in the model are reflected
in a difference synthesis.
1. Missing atoms. We have already seen that they appear as positive
maxima, but, because of the approximation made for the phases, their
height is usually smaller than that corresponding to the atomic number of
the missing atom.
The lack of truncation errors allows the correct localization of light atoms
even when the model contains much heavier atoms. For this reason,
difference Fourier series, computed when the model has been corrected for
most other important errors, allow the localization of hydrogen atoms (with
only one electron they contribute very little to the X-ray diffraction) even in
the presence of medium size atoms (cf. Fig. 5.28).
2. Position errors. Their effect is shown in Fig. 5.26. If p, gives the
correct position of the atom and p, its wrong position in the model, in the
Ap map the latter will be close to a negative minimum, while the correct
position will be towards the neighbouring positive maximum along the
maximum gradient line. It is possible to have a quantitative estimate of the
shifts to be applied, but in general, when the errors are not too large, it is
easier to correct the position errors by the least-squares methods (cf.
Chapter 2, pp. 90-108, and later in this section).
3. Errors in the thermal parameters. As we have seen in Chapter 3 (p.
148), because of the thermal motion the electron density function around
each atomic nucleus becomes wider. In Fig. 5.27(a) the case in which the
thermal motion has been neglected or underestimated in the model is
represented. The p, density will therefore have a smaller and wider peak Fig. 5.26. Position error of a model atom (top)
with respect to p,, and in the difference synthesis a negative depression, and corresponding difference synthesis
surrounded by a positive ring, will appear. If, on the other hand, too high a (bottom).
thermal motion has been assumed for one atom of the model, then Ap will
show a small positive maximum surrounded by a negative ring.
Finally, the case in which an isotropic thermal motion has been assumed
in the model (the p, maximum has a spherical distribution) while the real
motion is anisotropic (the p, maximum has an ellipsoidal distribution), is
illustrated in Fig. 5.27(b) together with the corresponding difference
synthesis showing two positive maxima and two negative minima; the line
joining the two positive lobes represents the direction of largest thermal
motion. The qualitative indications obtained in this .way allow one to
recognize those atoms for which it is more important to carry out the
least-squares refinement varying the six parameters of the anisotropic
thermal motion (cf. Appendix 3.B).
Least-squares method
By far the most widely used method of structure refinement is the
least-squares method. The theory and the computing procedures have been Fig. 5.27. Thermal parameter errors and their
described in Chapter 2, pp. 90-108. Several computer programs have been effect on the difference syntheses; for the atom
of the model it is assumed: (a) too small an
implemented to carry out the crystallographic least-squares refinement; isotropic motion; (b) an isotropic motion when
usually they are integrated within complete crystallographic packages such an anisotropic model should be assumed.
Fig. 5.28. AZOP: structural formula and

projection on the mean plane of the molecule of
the difference map computed when R = 0.078
(the scheme of the molecule is added to the map
for reference).
as SHELX,['~]X T A L , [ ~N ~ ]R C V A X , ~ CRYSTALS[~']
~~] and those commer-
cially available together with most single-crystal diffractometers. Here we
will illustrate the practical application to the refinement of the fairly small
structure of 2-a~oxycyanopyridine[~~] (AZOP, C6H4N40,space group PI,
Z = 2). The structure was solved using the SIR program;[67]from the best
E-map the coordinates of all the 11 non-hydrogen atoms of the molecule
(cf. Fig. 5.28) were obtained. The structure factors computed with this
model, assuming an isotropic temperature factor equal for all atoms
( U = 0.05 A'), yield a residual R = 0.34, a sufficiently low value to indicate
the correctness of the model. As we have seen in Chapter 2 (p. 94) the
refinement must be repeated in several cycles until convergence. For AZOP
the refinement was performed using 1176 observed reflections by means of
the system of programs SHELX;['O1 the first four cycles were carried out
varying the overall scale factor and the three coordinates plus the isotropic
thermal parameter of each atom; the total number of varied parameters is
+
1 4 x 11= 45 and the R factor reduces to 0.148. In the next step the
refinement is carried out varying for each atom the six components of the
anisotropic thermal parameter as described on pp. 186-8. Four anisotropic
+
refinement cycles were performed with 1 9 x 11= 100 varied parameters
and at convergence R = 0.078.
At this point we may assume that the model formed by the heavier atoms
is sufficiently well refined and we can compute a difference Fourier synthesis
in order to localize the hydrogen atoms. Figure 5.28 shows a projection of
the map onto the mean plane of the AZOP molecule; the known atomic
positions are also shown linked by their chemical bonds, and the hydrogen
atoms are clearly seen as maxima in the map. The coordinates of the four
hydrogen atoms derived from the map are then introduced in the refinement
process with appropriate isotropic temperature factors, but this usually can
not be done in a straightforward way. In fact the contribution of the
hydrogen atoms to the observed amplitudes is very small and rapidly

decreasing with increasing sin 8/12; the accuracy of their localization is
therefore quite low and often, because of the errors in the experimental
data, the refinement shifts them into stereochemically incompatible posi-
tions (e.g. bond distances which are too short or too long) with abnormal
temperature factors (e.g. values which are too high and tend to cancel out
the H atoms or too small for H linked to atom with significant thermal
motion). As we have seen in Chapter 2 (p. 106) it is possible to impose
some geometrical restraints during the least-squares refinement; in our
example we have imposed that the C-H distances should remain within
0.98f 0.02A. In addition a common isotropic thermal parameter was
assumed for all four H atoms, which was varied as a single parameter. Four
+
cycles of refinement varying 100 3 x 4 + 1= 113 parameters converged to
R = 0.055. A frequently used alternative procedure for locating the H atoms
consists in calculating their positions on the basis of the most common
stereochemical rules. It is recommended that these positions are checked
against those obtained from a difference Fourier synthesis. Their refinement
may then be carried out as described above.
Up to this step all observations were given the same unitary weight, but
following the indications of Chapter 2 (p. 101-2), in order to get meaningful
values of the estimated standard deviations of the varied parameters, the
observational equations should be properly weighted. In Chapter 2 we also
saw how it is possible to express the weights in terms of a polynomial in IFo[.
Several weighting functions have been proposed and the first scheme
suggested by ~ u g h e s [was
~~]
w = constant = 1/16Fk,, for 1 FolG 4Fmin
w = 1 6 ~ i ~ ~ / l ~ ~ 1 ~ for IFo[> 4Fmi,
where Fminis close to the minimum observed amplitude. Later
Cruick~hank[~*I
proposed the scheme
with a, = 2Fminand a, = 2/Fm,, as initial guesses.

For AZOP the intensities were measured with an automatic
diffractometer using a scintillation counter; for each IFol the corresponding
estimated standard deviation a(Fo) due to counting errors may be derived
(cf. Chapter 4, pp. 271 and 280). Since this is not the only error affecting
our observations, it is necessary to modify the weights introducing an
empirical dependence on IFoI itself. The weighting function available in the
SHELX program is
1
where q is an adjustable parameter chosen in such a way that the (w AF,)

values remain constant when the reflections are grouped in different ways
(e.g. in lFol or sin 6/12 intervals). The last four cycles with the weighting
function (5.84) afforded a final R value of 0.052 with q = 0.0007. The
weighted R factor
C wiA4
and the goodness-of-fit measure (also called standard deviation of an

observation of unit weight)
GofF =
n-m
are also computed. Equation (5.86) is nothing but the quantity defined in
(2.58) and GofF should be close to unity if the weights of the observations
have been correctly assessed, the errors in the model are negligible in
comparison with the errors in the data and there are no significant
systematic errors. For AZOP R, = 0.054 and Go= = 1.504.
In the final stages of the refinement one should check whether some
intense low-angle reflections are affected by secondary extinction (cf. p.
97) and have JF,J systematically smaller than IF,(; some least-squares
programs allow an empirical correction of this effect but it is common
practice to simply discard these few reflections from the refinement. In the
case of AZOP three such reflections were eliminated.
It is also important to check for possible correlations between parameters
during the least-squares refinement. In Chapter 2 we saw that the
least-squares procedure gives the variance-covariance matrix of the derived
parameters (eqn (2.56)). It is therefore straightforward to calculate, for
each term in the matrix, the correlation coefficient
COV (PiPj)
Pij =
~ ' ( ~ i ) ~ ~ ( ~ j >
which is close to zero if the correlation is negligible and close to 1 for large
correlation. Most programs compute the correlation matrix and output the
off-diagonal elements which are greater than 0.5. No such correlation was
found for AZOP.
How low should the final R factor be in order to have good confidence in
the quality of the refined model? The answer to this question is not
straightforward, because it depends on the type and complexity of the
structure and on the quality of the experimental data. Nevertheless, for
small or medium size structures giving good quality crystals (as it is usually
the case for most organic or organometallic compounds), with intensities
collected at room temperature on a diffractometer, we should expect R
values in the range 0.03-0.07. For very accurate low-temperature measure-
ments these values may be further reduced. The quality of the model should
then also be assessed by looking at the standard deviations of the atomic
parameters or at those of the derived geometrical quantities, such as bond
distances, which should be less than 0.006A for non-hydrogen atoms.
Usually as the complexity of the structure increases the quality af the
crystals and of the diffraction data decreases, and, at convergence the R
value may be in the range 0.08-0.15 for large molecules and up to 0.25-0.30
for macromolecules solved at low or medium resolution. In these cases the
final model may still contain some fairly large errors, which may often be
corrected by applying stereochemical or energy constraints (cf. Chapter 8,
p. 568).
The final R value will also depend on the selection of reflections used
during the refinement. It is in fact common practice to discard the very
weak reflections, the intensities, I, of which are measured by diffractometer

with large estimated standard deviations, ~ ( 1 )reflections
; with I s na(Z),
usually with 2 G n G 3 (for AZOP n = 2 was used to select the 'observed'
are omitted. Of course the higher is n the lower will be R, but
too high a cut-off value will considerably reduce the number of observations
and the least-squares refinement will yield higher standard deviations (cf.
Chapter 2, pp. 102-4). There is no theoretical justification for discarding
weak reflections, and a more rigorous procedure requires the use of all data
with proper weights["] and a prior statistical treatment of weak reflections
to account also for negative measurements.[80]On the other hand, in most
cases, only for very accurate structural analyses would the latter procedure
reveal significant changes in the final model. When the choice between a
centrosymmetric and a non-centrosymmetric space group is ambiguous or
when the structure shows pseudotranslational symmetry, the inclusion of
weak reflections becomes essential.[811
In following the convergence of our refinement we have mainly used the
R factor defined in (5.3); other agreement indices have been proposed,
some of which are statistically more appropriate to assess the significance of
small changes in the final refinement stages. Among them we recall the
weighted residual proposed by ~ a m i l t o n [ ~ ~ ~
already considered in Chapter 2 (p. 103), where it was shown that the ratio
of the residuals R;(l)lR&(2) obtained from two refinements of different
models may be used for testing the significance of the difference between
the two models. For instance this test may be used for determining the
absolute configuration of optically active molecules from the effect of
anomalous dispersion,[831although, as pointed out by ~ o g e r s , @some~I
caution is necessary.
Although the R index is the most widely used criterion to assess the
goodness of a structural model, its indications should be regarded with some
caution. As an example of the weakness of the R index, let US consider the
case of a structure containing one heavy atom, such as Pt, Ir, Os, or Cd,
and a certain number of lighter atoms (C, N, 0 ) . Since the I$ value is less
sensitive to the position of the light atoms, a model with the heavy atom
correctly placed and the light atoms quite inaccurately localized will give a
lower R value than a model with a small error in the position of the heavy
atom and an almost correct position of the light atoms. Nevertheless the
second model, with a smaller average position error and a uniform
distribution of errors over all atoms, is certainly a better one.
When the model is an incomplete representation of the molecule, R
becomes a rather weak guide, as it measures the agreement between
quantities which depend on largely different numbers of parameters (IF:]
depends on all atomic parameters, while IFiI only depends on the limited
number of parameters defining the incomplete model). The only valid check
of the indication given by the R value is the convergence of the svbsequent
Process of completing and refining the model.
As mentioned in the introduction of this chapter, the refinement of a
structural model will not converge if the model is affected by large errors.
This is a direct consequence of the non-linear nature of the problem, which,
as observed in Chapter 2 (p. 93), implies the presence of several local
minima in the function to be minimized. Two types of errors, which occur
quite often, are particularly disturbing.
1. Incorrectly positioned molecule. It may happen that the structural

model obtained by direct or Patterson methods is constituted by a correctly
oriented molecule, but its position with respect to the symmetry elements is
incorrect. In most cases the entire molecule is shifted by a vector
corresponding to a large Patterson peak due to several superposing
interatomic vectors. An example is given by the AZOP structure; the initial
attempts at solving the structure by direct methods gave correctly oriented
molecules, shifted from the correct position and their refinement would not
converge to an R value lower than 0.25. The correct solution was only
obtained after estimating the triplet invariants by their second repre-
sentation. An alternative way, which is often successful when applied to
structures belonging to the space group PI, would be to try to solve the
structure in the lower symmetry group P I . The solution will give two
molecules related by an inversion centre in a general position; this can be
located by inspection and then, after translating one of the two molecules in
such a way that it is referred to the inversion centre at the origin (i.e, the
general coordinates of the inversion centre are subtracted from all atomic
coordinates), the structure can be refined in PI. When these methods fail
one can try to find the correct position by the Patterson method described in
Appendix 5.B .
2. Wrong space group. As explained in Chapter 3 (p. 161), the deter-

mination of the space group from the diffraction data is not unique, and a
wrong choice is more frequent than one might think. An interesting analysis
of such errors is given by Marsh and ~ e r b s t e i n [ who~ ~ ~noted
~ ~ ] that a
relatively large number of structures has been reported with space group of
unnecessarily low symmetry. Two types of structures can be considered: (i)
those for which the reported space group has a lower Laue symmetry than
that of the actual structure; and (ii) those for which the reported space
group lacks a centre of symmetry, which is instead present. For the first type
the correct space group may often be found using geometrical
consideration^[^^^^^^ such as those implemented in the computer programs
NEWLAT['~] or LEPAGE.[~'] For the second type, the refinement by
least-squares should in theory yield a singular matrix (cf. Chapter 2, p. 100),
as half of the refined parameters have a linear relation (the overlooked
inversion operation) with the other half; in practice this relation may not be
exact, but in any case the solution of the normal equations will be very
unstable with large errors in the atomic parameters and large values of their
estimated standard deviations. The most common errors occur for the
following space groups: P1 instead of PI, PC instead of P2,/c, C2 or Cm
instead of C2/m, Cc instead of C2/c, Pna2, instead of Pnam, Cmc2, instead
of Cmcm (several examples are given by Marsh and ~ e r b s t e i n [ ' ~ ~ 'They
~~).
may be recognized because they usually lead to abnormal values of some
bond distances and angles or of the thermal parameters, which are adjusted
when the refinement is carried out in the correct space group.
A critical view (possibly a bit too pessimistic) of the accuracy of the

reported structure determinations has been presented by ones[^'] and the
reading of his paper is highly recommended.
Two important effects may cause a lack of convergence in the refinement
process: twinning and disorder. As we saw on p. 83-7, the diffraction
intensities of a twinned crystal must be carefully analysed in order to
separate the contributions from the different individuals forming the twin.
This will be more difficult when all reflections from the two individuals
superpose exactly (merohedral twins, cf. p. 85). Often merohedral twinning
is not recognized and then in most cases the structure fails to be solved;
even when a solution is found, it is impossible to refine the model to an
acceptable R value because of the systematic errors on the observed
amplitudes.
Real crystals deviate from the ideal picture of a periodical perfect
repetition of a unit cell. We saw on p. 148 that, because of the thermal
motion, atoms oscillate around an equilibrium position and we may say that
they are affected by a small dynamic disorder. But larger movements are
sometimes possible in crystals, giving rise to a real dynamic disorder.
Diffraction experiments will just show a time-averaged situation and if the
movement is continuous the electron density will be smeared out. On the
other hand if the movement is stepwise between two or more energy
minima, the total electron density will be divided between the positions of
the minima. An example of such dynamic disorder in the cis form of the
complex [ ( q 5 C 5 ~ 5 ) 2 ~ e 2 ( C 0 ) 2 ( ~ 2has
- C been
0 ) 2 ] recently described;[921the
two independent cyclopentadienyl rings show different thermal motions,
which may be interpreted in terms of different rates of stepwise rotation
around their fivefold axes. This hypothesis was confirmed by a recent
solid-state NMR
Another type of disorder may occur in crystals; it is referred to as static
disorder and in this case the average contents of the unit cell is obtained by
averaging over the space, i.e. over the different unit cells. An important
example of this type of disorder, which is common in minerals and alloys, is
the substitutional disorder where the same site is occupied by different
types of atoms in different cells.
The structure factors depend only on the time- and space-averaged
contents of the unit cell and the result of a diffraction analysis is this
averaged contents. No distinction can be made between dynamic and static
disorder and the case of a molecular fragment oscillating between two
positions in all unit cells is indistinguishable from that in which half of the
cells have the fragment in one position and half in the other.
If a disordered structure is refined without taking into account the
disorder, then only an average structure is obtained with abnormal thermal
motion and the refinement will not converge. In the case of substitutional
disorder the different atomic species are introduced with the same coordin-
ates but with different occupation factors (cf. p. 87), depending on the
fraction of each species. When the disorder is between, say, two positions in
the cell, then the same atomic species are located in both possible positions
with an occupation factor depending on the 'population' of each position
(the term population will be associated with the fraction of time spent in the
given position in the case of dynamic disorder and with the fraction of cells
having atoms in that position in the case of static disorder); if this is not
known one may initially assume an occupation factor of 0.5 and than refine
the factor to convergence (the occupation factor is highly correlated to the
temperature factor and some caution should be used in the least-squares
refinement). Only when disorder is treated in a proper way will the
refinement converge to an acceptably low R value.
Absolute configuration
In Chapter 3 (p. 168) it was shown that, when the anomalous dispersion
effect is present, Friedel's law is no longer satisfied. We shall see in Chapter
7 (p. 489) that for compounds containing asymmetric carbon atoms, isomers
of opposite chirality (enantiomers) are possible and that their solutions
rotate the plane of polarized light in opposite directions. By simple chemical
methods it is not possible to decide which of the two configurations
corresponds to the isomer rotating the plane of the light to the right
+
(( )-rotamer). Fisher proposed to assume as reference that ( )- +
glyceraldehyde corresponds to the configuration 5a shown in the table on p.
487 and to establish the configuration of all other chiral molecules in a
relative way with respect to this convention; thus (+)-tartaric acid
corresponds to 7a. Since this is an arbitrary assumption there was only a 50
per cent chance of it being correct. Fortunately the first determination of
the absolute configuration of NaRb-(+)-tartrate[941by the method de-
scribed below (using the anomalous scattering of the Zr K, radiation by
Rb), proved that the assumption was correct. Since then the anomalous
dispersion method has been used to determine the absolute configuration of
a large number of molecules and it will be shown in Chapter 8 (p. 545) that
the method is also applied in macromolecular crystallography.
The two optical isomers differ in hand and may be related either by an
inversion or by a reflection operation (inverse congruence). t
In a normal X-ray experiment, with no anomalous dispersion, because of
Friedel's law, the two enantiomers are indistinguishable, as they give rise to
the same diffraction pattern. When one or more anomalous scatterers are
present, then both IFhI and IF-,I should be measured and compared with the
computed values. Since the Bijvoet["] differences 1 ~ ~ 1IF-,I2 ~ - are rather
small, and anomalous dispersion is only large for heavier atoms near their
absorption edges, it is essential to apply a proper absorption correction to
the diffraction data.
In order to illustrate how the absolute configuration may be determined it
is instructive to consider the procedure originally used by Bijvoet, Peerder-
man, and van ~ o m r n e l ,which ~~~~ gives reliable results only when the
anomalous dispersion effect is sufficiently large and for this reason it is no
longer used. The ratios IF", /IF?,,\ and IF;J/JF?,1 are tabulated for a limited
number of reflections with large Bijvoet differences; a one-by-one com-
parison of the two ratios indicates that the wrong configuration has been
chosen if, when the first ratio is greater than 1, the second is less than 1 and
i We should note that the term 'absolute configuration' is not always correct in describing
crystal structures, as it does not apply to the case of non-centrosymmetric but achiral space
groups (polar groups with reflection symmetry operations, such as Pna2,) o r to that of achiral
molecules (without asymmetric centres) crystallizing in non-centrosymmetric space roups,
such as q u a r t z The term 'absolute structure' has been recently proposed and discussed$'~"
vice versa. If this is the case, then the correct enantiomer is obtained by
&anging the sign of all atomic coordinates.
When the anomalous dispersion effect is small all the reflections should be
used and efficient methods have been proposed by and
in the latter an absolute-structure or chirality parameter x is
refined in a least-squares process (cf. p. 97) in which the structure factor is
written as
+
IF(h, x)I2 = (1 - x) IFhI2 x IF-,I2 (5.89)
where x is close to zero when the model and the crystal are in the same
chirality, and approaches 1 if they are inverted one with respect to another.
Appendices
5.A Structure factor probability distributions

We will use the central limit theorem stating that, given N independent
random variables x, ( j = 1, N), each with mean value (xi) and variance (ui2,
N
their sum X = C xj has a mean value equal to the sum of the individual
i=l
mean values and a variance equal to the sum of the variances, provided that
N is sufficiently large, i.e.
(x) = x
N
j=1
(xi) and a2= x 4.
N
j=1
(5.A.1)
Whatever the distributions of the variables xj are (they may be all

different), the sum X will have a normal probability distribution
For a centrosymmetric structure the structure factor may be written as
where the sum is extended to the independent atoms only so that Fh can be
considered as the sum of N/2 independent random variables xj =
2fi cos 2nh q with mean value
(xi) = 24 (cos 2nh q ) . (5.A.4)
Assuming a random distribution of the atomic coordinates, all q values

will be equally probable and (cos 2nh q ) = 0; hence
The variance of xj is defined as

a; = (x;) - (xj)2.
Because of (5.A.5) the second term is zero and

cy7 = 4f7(cos22xh . I;.) = 2f7 (5.A.7)
given that, for a randomly variable angle 8, (cos28) = 4.
Because of the central limit theorem, the mean value of Fh will be
and its variance
Finally the distribution function of F is given by
Pi(F) = -Lexp (- F2/2X).

VzZ
F has the same probability of being positive or negative and the probability
Pi(lF() that its modulus has a given value is given by Pi(l FI) = 2Pi(F);
substitution of (5.A.10) gives eqn (5.4).
The approximate character of (5.4) is clearly indicated by the observation
N
that the maximum value of IF1 is C f, while Pi(lFI) has a Gaussian shape
and goes to zero only at k ~ . 1 = 1
For a non-centrosymmetric structure Fh = A h + iBh and, in analogy with
the derivation of (5.A.8) to (5.A.10), we have
( A ) = 0, 0; = 4X ( B ) = 0, 02, = 4X (5.A.11)
and
1 1
m
P(A) = -exp ( - A ~ / x ) ,
m
P(B) = -exp (- B2/Z). (5.A.12)
+
The joint probability that A lies between A and A dA and at the same
time B lies between B and B + d B is given by the product of the two
distributions (5.A. 12)
1
P(A, B) dA d B = P(A)P(B) dA dB = -exp ( - ( A ~+ B2)/z) dA dB.
Jcx
(5. A. 13)
Equation (5.A.13) represents the probability that the structure factor F lies
within a region of the complex plane of area dS = dA dB. If the structure
factor is expressed in polar coordinates (i.e. in modulus and phase), then
dS = IF1 dlFJdg, and
The distribution of the amplitude IF1 alone is obtained by integrating

(5.A.14) over g, and thus (5.5) is immediately derived. From (5.A.11) we
can see that also for non-centrosymmetric structures Z = ((FI2).In deriving
the value of (1~1') we did not consider the effects due to the presence of
Fig. 5.A.1. Space group Pm: in the projection on

the plane rn the t w o related molecules
"
I superpose.
other symmetry elements besides the inversion centre. Let us, for instance,
consider the effects due to the presence of a mirror plane, by considering
the space group Pm (m perpendicular to the b axis). In Fig. 5.A.1 we can
see that on the projection along the b axis, all atoms related by the m plane
are superimposed; the projection of the electron density on the ac plane will
only contain half of the peaks, each with double weight. The Fourier
coefficients FhOI relative to this projection will have a distribution cor-
responding to a structure of N / 2 atoms with scattering power 26 and then
The mean intensity of the h01 reflections will be twice that of the general
reflections hkl. In order to take into account the symmetry element effects,
(5.A.9) is generalized to
where E will be equal to 1 for the general reflections and greater than 1 for
certain classes of reflections which are influenced by the presence of
symmetry elements; thus in Pm E = 2 for the h01 reflections. The values of E
and the relative reflection classes for the different symmetry elements are
tabulated in the International Tables.[991
We note that the calculation of the ratio of the average observed intensity
for specific classes of reflections over that of the general reflections, can be
used as an indicator of the presence of some symmetry elements. For
instance we can distinguish between the space groups Pm and P2 by
computing the ratio ( IFhOl1')/(I ~ ~ ~which ~ 1will~ be) close
, to 2 for Pm and
close to 1 for P2.
5.B. Patterson vector methods

We will consider here the so-called vector methods[1001which allow a
systematic use of the information contained in the Patterson function also
when no heavy atoms are present.
The Patterson function can be considered as formed by the superposition
of N images of the crystal structure displaced one with respect to the other
by the amount required to bring to the origin each atom in the unit cell.
Each image represents the ends of all the vectors from the atom at the
origin to all the others. Figure 5.B.l(b) illustrates how the Patterson map oj
Fig. 5.4(c) can be interpreted in terms of five displaced images of the
five-atom structure of Fig. 5.4(a), repeated in Fig. 5.B.l(a). It can be easilj
seen that the same map can also be interpreted in terms of five displaced
images of the enantiomeric structure.
It is in theory rather simple[1011to extract one image of the structure from
the overlapping set, but, as we shall see, the proposed methods may be
difficult to apply in practice; these are the so-called superposition methods.
Let us first consider the vector superposition in which one Patterson mar
is superposed on an other in such a way that the origin of the seconc
coincides with a given vector of the first; the two maps are thus separated b)
an interatomic vector. The set of coincident peaks will reveal one or more
Fig. 5.8.1. (a) Four unit cells of the five-atom

structure of Fig. 5.4(a); (b)four cells of the
Patterson map in which the peaks indicated by
r e f e r to the central origin and those indicated
by 0to the other origins; the peaks referred t o
the central origin are interpreted as the
superposition of five images and the peak
numbers refer to the atoms linked by the
corresponding vector.
0
0
@\, \
\
Solution and refinement of crystal structures
images of the structure. In particular, if the superposition is on a single

peak, the structure plus its inverse are obtained for a non-centrosymmetric
structure (Fig. 5.B.2); for a centrosymmetric structure only one image is
obtained. If the superposition is on a multiple peak the number of images
obtained is multiplied by the multiplicity of the peak (Fig. 5.B.4(c)). When
more than one image is revealed, iteration of the superposition process on
1
Fig. 5.8.2. Patterson superpositions: are the

Patterson peaks as in Fig. 5.8.1; O are the
second superposition on peak 13, indicated by

the second arrow (only one image is shown).
The reader is advised t o reproduce on
379
coinciding peaks when the origin of the identical

map of Fig. 5.8.1 is superposed on peak 12 of
the present map, indicated by the first arrow
(two enantiomorphic images of the molecule are
shown); +are the coinciding peaks after a
transparent sheets three copies of Fig. 5.8.1 and

then perform the above superpositions.
different peaks will eventually yield a single image. This is illustrated in Fig.
5.B.2, where superposition of the double image on a new peak reveals one
image only.
The basic principle of the method may be understood by considering that,
if the structure contains atoms 1 , 2 , 3 , . . . , N in the unit cell, the Patterson
map P will be the superposition of the images I(l), I(2), I(3), . . . ,I(N)
obtained by placing atoms 1 , 2 , 3 , . . . , N in turn at the origin, i.e.
After translation by a vector 1-2 the new function will be
where the different images are obtained by placing atoms 1 , 2 , 3 , . . . ,N in

turn, not at the origin, but at the end of the vector 1-2. In general I1(n) will
not coincide with any image I(1), I(2), I(3), . . . , I(N) unless n = 1 or 2. In
fact, because of the choice of the translation 1-2, 11(1)= 1(2) and
-Ir(2) = -I(l) (-I(n) is the enantiomer of I(n)) as illustrated in Fig. 5.B.3;
the two images left in Fig. 5.B.2 after the first superposition are indeed
-I(l) and I(2).
Unfortunately, this simple method is only applicable in the ideal case of a
Patterson function corresponding to a point atom structure. In practice we
have a continuous Patterson function in which the peaks of finite width
overlap; for reasonably sized molecules with no heavy atoms, also in a
sharpened map, only the peaks with high multiplicity appear, while single
peaks are not resolved from the background. Thus superposition on any
recognizable peak will give multiple images and several superpositions will
be necessary to reveal a single image of the structure. A further problem
arising with vector superpositions is that the obtained image is not referred
to the true origin of the cell as it is apparent from Fig. 5.B.2; the correct
Fig. 5.8.3. (a) Image l(1) and its enantiomorph

-1(1) with their atom 1 at the origin; (b) image
l(2) and its enantiomorph -1(2) with their atom 2
at the origin; (c) image - I ' ( l ) with its atom 1 on
peak 12 (Fig. 5.B.l(b)), DIUS the imaae ~ 'with (i) 10~ationof the structure with respect to the symmetry elements can be
its atom 1 i n peak 21 (enantiomorph); (d) image obtained by the described later.
l'(2) with its atom 2 on peak 12, plus the image
with its atom on peak 21 (enantiomorph), Despite these problems, superposition methods can still be applied in
The shadowed images -1U) in (a) and 1(2) in (b) practice, especially when some structural information is available. This can
are found in the superposition on peak 12 of Fig.
5.8.2; they are identical to -11(2) in (d) and l'(1)
be of three types:
in (c) respectively.
( 1 ) the stereochemistry of one or more rigid molecular fragments is known,
but not their orientation and position;
(2) the orientation is also known, but not the position;
(3) both orientation and position are known.
Information of the first type is often available from known bond lengths and
angles or from the Cambridge Structural ~atabase,[''~] and we shall see how
it can be exploited.
In order to carry out any superposition process it is necessary to evaluate
the degree of coincidence between the peaks of the superposing maps. This
is done by means of the so-called image-seeking functions;[100]three such
functions have been proposed:
1. The product function
( r )= P ( r ) X P ( r - u ) (5.B.1)
which takes very high values when coincidences occur, but is subject to
considerable background noise and is very sensitive to small errors in the
value of the vector u used for the superposition.
2. The sum function
E (r) = P(r) + P ( r - u) (5.B.2)
which is less sensitive to the errors on u, but can often produce spurious
peaks and in general a high background noise.
3. The minimum function
M(r) = min {P(r), P(r - u)} (5.B.3)

which is also sensitive to the errors on u but has the advantage of a low
background noise.
Let us first consider how the superposition method can be conveniently
applied when the location of one or more atoms (both independent and
related by symmetry) in the cell is known. In this case Patterson maps with
their origin at the sites of the known atoms are superposed. An example of
this atomic superposition is shown in Fig. 5.B.4(d) in which two maps, with
their origin at the position of atoms 1 and 1' (related by an inversion centre)
respectively, are superposed. Images 1(1) and I(1') coincide and reveal the
structure properly placed with respect to the origin. Atomic superposition
can be used when the crystal contains a heavy atom, the position of which
has been determined by the method described on p. 328. Atomic superposi-
tion in this case is facilitated by the fact that the coinciding images are made
up of vectors between heavy and light atoms which are well resolved over
the background of vectors among light atoms. If the ratio (5.20) is very low,
the straightforward application of the Fourier recycling procedure may be
inefficient and atomic superposition may be a convenient alternative.
We can now describe one of the procedures[10~106] proposed to solve the
Patterson function when the stereochemistry of part of the molecule is
known. It is convenient to divide the process of locating a fragment of
known stereochemistry into two steps: determination of the orientation and
determination of the position with respect to the symmetry elements.
1. Determination of the orientation. Starting from the known molecular
fragment (atomic model) it is possible to build up the corresponding vector
map (vector model). This model map is superposed on the Patterson
function with its origin coinciding with that of the Patterson function and
then rotated in all possible orientations. For each orientation the match
between the ends of the m vectors of the model and the Patterson peaks is
evaluated by means of functions similar to those considered above. Their
form is now:
(a) product function
where P(uj) are the values of the Patterson function at the ends of the
vectors;
Fig. 5.8.4. (a) Centrosymmetric three-atom

structure; (b) corresponding Patterson map; (c)
vector superposition on the double peak 23 45+
gives the t w o images in full and broken lines,
while superposition o n the single peak 25 gives
the full-line image only, shifted from the correct
origin at the inversion centre; (d) Atomic
superposition: are the known positions of the
related atoms 1 and 1'; x , C 3 are single and
double peaks of the Patterson function with its
origin on atom 1; 0, O are single and double
peaks of the Patterson function with its origin on
atom 1'; 9 are the overlapping peaks showing
the structure referred t o the correct origin.
(b) sum function
(c) minimum function

M = min {P(ul), P(uJ . . . P(u,)).
This function is the most sensitive and is usually preferred, but, when the
absolute minimum is assumed, small errors in the model may produce
small values of M even for a correct orientation. To overcame this
problem it is preferable to use the mean of the n smallest values ( M ) , ,
where n is usually equal to 10-20 per cent of the total number of
vectors. [lo']
The orientation of the model can be varied following the same Eulerian
scheme (a detailed description is given by Stout and ~ e n s e n [ ' ~ ~used
] ) in
most diffractometers to define the position of the reciprocal lattice vectors in
terms of the three angles 8, q, and x (cf. pp. 72 and 274-6). The number of
orientations to be examined can be quite large and the symmetry of the
Patterson function and that of the vector model can reduce the range of the
rotational search.
The correct orientation will correspond to large values of the functions
(5.B.4), (5.B.5), and (5.B .6).
2. Determination of the position with respect to the symmetry

elements. Once a likely orientation has been defined, the location of the
fragment can be determined by systematically translating it in all possible
positions within the asymmetric part of the unit cell and then by generating,
for each position, all the vectors between the atoms of the fragment and
those of the symmetry-related fragments. Each vector set obtained in this
way is compared with the Patterson function and the fit is evaluated by
means of one of the functions used in the orientational search. Normally the
search of the position of a fragment with respect to the different symmetry
elements is carried out separately for each symmetry element, considering
that:
(a) when the symmetry element is a plane the search is performed only
along the direction orthogonal to the plane;
(b) when the symmetry element is an axis the search is a two-dimensional
one on the plane orthogonal to the axis;
(c) when the symmetry element is an inversion centre the search is three
dimensional.
Because of the errors in the model the orientation and position are
determined by the above method only in an approximate way. They can be
further improved by shifting the atoms of the model in such a way that a
better fit of the set of interatomic vectors on the maxima in the Patterson
function is obtained.
The described procedure operates in the Patterson space, but other
methods, usually operating in the reciprocal space, such as the rotation
function,[1091 have been proposed[110-1131and successfully applied, especially
in protein crystallography (cf. Chapter 8, p. 553). They are at the basis of

the so-called molecular replacement methods.[l14]For each orientation or
position of the model fragment, instead of fitting the vector model, a fit is
searched for between the structure amplitudes calculated for the model and
the observed ones. A procedure combining direct and reciprocal space
searches has been recently described by Rius and ~iravitlles.["~]
Recently, Patterson orientation search routines have been incorporated
into direct methods programs. A highly automated routine has been built
into the DIRDIF system.[701Another example is the PATSEE routine[116]
incorporated into the SHELXS-86 system.[lolIn both these cases the correct
position of the oriented fragment is found by direct methods.
5.C Two examples of deriving phase information from

positivity
Let us first consider a centrosymmetric one-dimensional structure for which
the amplitudes IFhJand IF2hlare large. Their contributions to the electron
density function
p(x) = (lla) Fh cos 2nhx
h
will be large and are illustrated in Fig. 5.C.1, where curve 1 shows the term
+ 141 cos 2nhx (Fh positive), curve 2 the term - I FhI cos 2nhx ( F , negative),
and curve 3 the term +IF,,\ cos2n2hx ( 4 , positive). Fh would then
contribute to the electron density in the region indicated by the letter A, if
positive, and in the region B, if negative; in both cases F,, will contribute to
both regions only with a positive sign. It follows that, with p(x) everywhere
Fig. 5.C.1. One-dimensional centrosymmetric
structure with IFh/ and lFZhllarge: curve 1 shows positive, if 1 Fhl and IF,, I are large, whatever the sign of I Fh1, the sign of IF,, 1
the trend of IFh[cos Pnhx, curve 2 that of is more likely positive. This argument can be generalized to three
-IFh/ cos Pnhx, and curve 3 that of
F
I, ; hatched regions A and B are
cos 2 ~ 2 h x the
dimensions.
those in which a high value of the electron As a second example we shall consider the projection of a centrosym-
density is indicated. metric structure for which the amplitudes I FhI, I FkI, and IFh-,[ are large. In
Fig. 5.C.2 the traces of the three families of planes h, k, and h - k are
k shown as full lines, while the dotted lines are located half way. If we
h-k 1 1 1 I associate to the full lines the maxima of the terms IFhlcos2nh r, -
. .
lFklcos 2nk r, IFh-,I cos 2n(h - k) r contributing to the electron density
function p(r), then the dotted lines will represent the corresponding
minima. If I Fhl, IFkl, I Fh-klare large, the corresponding terms must add up
to give a preponderant contribution to p ( r ) . This will happen when
-
cos 2nh r = +1, if Fh is positive, and cos 2nh . r = -1 if it is negative; the
same applies to the other two terms. It follows that the possible maxima of
the electron density will be restricted to the regions indicated by A, B, C,
and D in Fig. 5.C.2, where the above conditions are satisfied for all three
terms at the same time. For each region the signs S(h), S(k), S(h - k) of
the three structure factors are
Fig. 5.C.2. Three-dimensional centrosymmetric
Region S(h) S(k) S(h - k)
structure with IFhl, IFk!, and IFh-*/ large: the full A + - -
lines represent the maxima of lFhl cos 2nh. r,
IFkIcos2nk.rand IFh_klcos2n(h- k).r,the
B + + +
broken lines the corresponding minima; the C - - +
hatched regions A, B, C, D are those in which
high values of the electron density are more
D - + -
likely. In all cases the product S(h)S(k)S(h - k) is always positive.

5.D Probability formulae for triplet invariants

For a structure with N equal atoms in a P I unit cell, the normalized
structure factor of index h can be written as the sum of N independent terms
1
Eh = - x
f i , = l
1
exp (2nih q) = -
mi4x exp (2nik - 5 ) exp [2ni(h - k) . q]
Under the assumption that all points in the unit cell have equal
probability of housing an atom, the atomic position vectors q will form a set
of independent random variables. Thus also the terms g,, which are
functions of q, will be random variables. Then, because of the central limit
theorem, Eh will have a normal distribution with mean
and variance
1
4=- 2 ( ( E )- ( 4 ) ' ) .
N ,=I (5.D.3)
Supposing that the two exponential terms in the middle formula in

(5 .D . l ) are independent, we have
(5,) = (exp (2nik q)) (exp [2ni(h - k) q]). (5.D.4)
Under the assumption that Ek has a fixed value (i.e. both its modulus and
phase are known), we can write[ll']
With the same assumption for Eh-,, (5.D.4) becomes
and substituting in (5 .D .2)
For the real and imaginary components of Eh = A h + iBh we have
Similarly to (5.D. I), we can derive
2 cos 2n(h - k + k) .q = mi=l[C,(k)C,(h - k) - S,(k)S,(h - k)]

1
m,=1
Ah = -
where C,(k) = cos 2nk q and S,(k) = sin 2nk 5. Similarly
Bh =-
fl
x1
,=I
1
[S,(k)Cj(h - k) + Cj(k)Sj(h - k)] = -
fl, = I
x Pi. (5.D.10)
where the terms (Ah)2/Nand ( B h ) 2 /have ~ been neglected, as they are of

order 1/N2 [cf. eqns (5.D.8)] and become very small for any reasonable N.
The conditional probability density function P(Ah, Bhl Ek, EhPk)can be
expressed as the product of the separate distributions of Ah and Bh. The
+
probability that Ah lies in the interval between Ah and Ah dAh and at the
+
same time Bh lies in the interval between Bh and Bh dBh, is therefore
given by
1
P(Ah, Bh I Eh, Eh-,) = -exp {-[(Ah - (Ah))2+ (Bh - (Bh))2]) (5.D.13)
n
which gives the probability that Eh lies within an area of surface dS =
dAh dBh in the complex plane. When expressing Eh in polar coordinates
(modulus and phase), then dS = IEhI dlEhl d q h and
As lEhl is known, we are interested in the conditional probability

P ( q h I IEhI, Ek, Eh-,), simply indicated as P(qh), which can be obtained by
dividing (5.D.14) by its integral with respect to qh[118]
where
and
Qhk = Q)h - q k - Th-k.
Since qk and qh-, are known dQhk= d q h and the distribution of qh is
equivalent to that of Qhk. The value of the integral at the denominator of

(5.D.15) isr5'] L = 2n&(Ghk),where &(x) is the modified Bessel function of
zero order. We can finally writeLz4]
which is equal to the Cochran distribution (5.38).

In a similar way it is possible to derive the probability formulae for
centrosymmetric structures.[281By applying the central-limit theorem to
we obtain
[
1 exp - 21 (E, - ~ -)'Im
.
k ~ h - k
(5.D.19)
Indicating by P+ or P- respectively, the probability that E, has a positive

or negative sign, we have
and
1 exp x
p+= -
1+ exp (-2x) - exp x + exp (-x)
= tanh x + exp xexp+ exp
(-x)
(-x)
+
= tanh (x) 1- P +
and finally
P+ = 4 + f tanh (*
1
EhI EkEh-,) . (5. D.22)
The function (5.D.22) is illustrated in Fig. 5.14; P+> 4 if S(k)S(h - k) =

+ and P+< i (and P- > 1) if S(k)S(h - k) = - . If in (5.D.19) we consider
the two cases S(h) = S(k)S(h - k) (with probability P + ) and S(h) = -
S(k)S(h - k) (with probability P-) we obtain eqn (5.46), where P + is
always greater than i .
5.E Pseudotranslational symmetry

We will say that a structure possesses pseudosymmetry when a non-
negligible part of the atoms approximately satisfy a higher symmetry than
that of the whole structure. It can be noted that, in the special case in which
these atoms exactly satisfy a higher symmetry, they will identify a
supergroup (see Appendix l.E) of the space group to which the whole
structure belongs.
An important type of pseudosymmetry is the pseudotranslational
symmetry occurring when a non-negligible amount of the electron density,
say p,(r) is repeated by a pseudotranslation (in the sense that it does not
necessarily correspond to a crystallographic translation) u, i.e. p,(r) =
+
p,(r u). ~ u e r g e r [ ' ~suggested
~ , ~ ' ~ ~that such structures should be treated as
the sum of a substructure, p,(r) and the complement structure, p,(r) =

P(F) - P P W .
A two-dimensional example of pseudotranslational symmetry is shown in
Fig. 5.E.l(a), where the two heavy atoms 1 and 1' are related by a centring
+
translation, u = a12 b/2, while the light atoms occupy general unrelated
positions: the heavy atoms form the substructure and the light atoms the
complement structure. If the structure only contained the heavy atoms, the
unit cell would be centred and, according to p. 160, the reflections with
+
h k = 2n + 1 would be systematically absent. For the structure of Fig.
5.E.l(a), only the light atoms will contribute to these reflections
(superstructure reflections), which will therefore be systematically weaker
+
than the reflections with h k = 2n (substructure reflections) to which
both heavy and light atoms contribute.
Because of the presence of pseudotranslational symmetry, imposing
certain relations between subsets of atomic positions, the assumption that
the atomic coordinates are independent random variables is violated. In the
reciprocal space this is revealed by a non-uniform value of the averaged
normalized intensity, (I E, 12), when the reflections are grouped in different
ways. As we have seen, the reflections have to be divided into the set of
substructure reflections with high mean intensity and the set of superstruc-
ture reflections with low mean intensity.
In Fig. 5.E.l(b) the case in which the pseudotranslation is only
approximately obeyed, is shown. Finally, in Fig. 5.E.l(c) another example
of pseudotranslational symmetry is shown, in which two similar but not
equal atoms are related by a vector u = a / 3 ; the substructure reflections will
have h = 3n.
Statistical methods for ideiltifying the type of pseudotranslational sym-
metry and the related sub- and superstructure reflections have been recently
proposed. [49,51,1201
5.F Magic integers

I (c) Magic integers were first introduced by White and W o o l f s ~ n and
[ ~ ~allow
~~
Fig. 5.E.1. (a) Heavy atoms exactly related by a one to represent several phases in terms of only one variable, through
pseudotranslation vector u = a/2 + b/2; (b)
heavy atoms approximately related by a
of the type
. +
~seudotranslationvector u = a12 b12: , - , .Id
~. cpi = mix (5.F.1)
slightly different heavy atoms related by a
pseudotranslation vector u = e/3. where the mi are the magic integers. Let us consider the example given by
l ~ ~ ~phases, expressed for convenience in cycles (0 6 cp < 1)
~ o o l f s o n . [Three
instead of degrees, are considered; by using the three integers 3, 4, 5 in
(5.F.1), we will have
91= 3x mod (11, cp2 = 4x mod (I), cp, = 5x mod (1), (5.F.2)
where mod(1) indicates that the phases are always taken within the interval
0-1 cycles (i.e. 0-360"). We will now see that, whatever the value of the
three phases, it is always possible to find a value of x (0 < x < 1) for which
the three equations (5.F.2) are approximately satisfied. For instance, if
cpl = 0.3, 147, = 0.2, cp, = 0.7, the value x = 0.766 (Fig. 5.F.1) will give:
cp, = 2.298 mod (1) = 0.298 with an error of 0.002 cycles (lo),
cp2 = 3.064 mod (1) = 0.064 with an error of 0.136 cycles (49"),
cp, = 3.830mod (1) = 0.830 with an error of 0.130 cycles (47").
The reader may assign any three values to the phases and find by trial and o 766
error, with the help of Fig. 5.F.1, the value of x which best satisfies (5.F.2). p,
So far we have used the sequence of three magic integers 3, 4, 5, but
other longer or shorter sequences of different integers may be used. The
accuracy with which phases can be represented by magic integers depends
on the values of the integers and on the length of the sequence. ~ a i n [ ~ ~ ~ , ~ ~ ~ ~
described the theory of magic integers and gave some rules for selecting
those sequences which minimize the mean square deviations (Aq2) of the
represented phases. It turns out that the optimal sequences m,, m2, . . . , m,
are those for which 2m1 = m, + 1 and the differences m, - m,-l, mnPl-
mn-2, . . . , m3 - m2, m2 - m1 form a geometric progression of integer 2
numbers with common ratio r 3 1. For r = 2 the progression 1, 2, 4, 8,
16, . . . will give rise to the magic integers listed in the top part of Table
5.F.1. An integer progression with common ratio less than 2 is the so-called
Fibonacci series 1, 1, 2, 3, 5, 8, 13, . . . , where each term is defined as the Fig. 5.F.1. Representaton o f t h e three phases In
+
sum of the two preceding ones (F, = Fn-l Fn-2) and r tends to the golden eqns (5.F.2) as a function x.
number 1.618. The corresponding magic-integer sequences are listed in the
bottom part of Table 5.F.1, where the column headed (Aq), , , gives the
root mean square error of the phases represented with the different
magic-integer sequences. The optimal sequences also have the advantage
that the errors are equally distributed among the represented phases. From
Table 5.F.1 it may be seen that (Aq), , , increases as n increases and, for a
given n it is lower when the integers are larger (r = 2).
In several multisolution programs the general phases in the starting set
are expressed in the form (5.F.1) and x is explored at regular intervals in the
range 0-1 (if the enantiomorph has to be fixed, only the interval 0-0.5
should be explored). The value of the interval Ax is chosen in such a way
that the mean variation of the phases from one value of x to the other is of
the same order of magnitude of the corresponding (Aq),,, . Ax is
therefore a function of n and of the m, values and it is smaller the larger are
the integers. In order to keep the number of explored x values (each value
corresponding to a trial starting set of phases) within reasonable limits, it is
preferable to use magic integers based on Fibonacci series, at the cost of
slightly higher (Aq), , ,. A comparison between the use of quadrant
permutations and that of magic-integer permutations is shown in Table
5.F.2. This comparison may be better illustrated by considering the case
n = 2, for which it is possible to draw in the plane ( q l , q2), the lines
q , = 2x mod (1) and q2= 3x mod (1) (Fig. 5.F.2(a)). The choice of regular
Table 5.F.1. Magic integer sequences based on the geometric progression with
common ratio r = 2 (top) and on the Fibonacci series (bottom); (Aq),,,,,, is the root
mean square error of the phases represented by each sequence
n Sequence (AQJ)~.~.~.
Table 5.F.2. Comparison between quadrant permutations (fn/4, f 3n/4) and the per-
mutations obtained with magic integers based on the Fibonacci series.
Quadrant Magic integers
No. permutations No. permutations (AV)~."I.S.
intervals along x, i.e. along these lines, gives rise to a close-packed

hexagonal disposition of points in the phase space. This corresponds to a
more efficient way of sampling the phase space than that shown in Fig.
5.F.2(b), obtained by quadrant permutation.
This is also true for n > 2 and it may be concluded that the use of magic
integers allows a considerable reduction in the number of trial permuta-
tions, while the accuracy of phase representation is kept within acceptable
limits.
5.G New multisolution techniques

In the multisolution process described on p. 360 there are two main
limitations: the restricted number of starting-set phases and the 'chain'
character of the phase expansion process by means of the tangent formula.
Fig. 5.F.2. Illustration of two phases in the plane The first factor will make the initial steps of phase expansion very critical, as
(q,, q2): (a) Lines q, = 2xand rp, = 3x: the
triangles indicate the 12 regularly spaced points, the first phases will be determined by few relationships, while the second
with intervals of the order of (Aq),,,,,,, giving factor will amplify the first effect in the following steps.
rise to the same number of phase permutations;
(b)The triangles indicate the 16 points As we have seen, the use of magic integers allows an increase in the
corresponding to quadrant permutation. starting set of phases without a great decrease in accuracy. At the basis of
all recent developments of the multisolution techniques lies the empirical
observation that it is possible to solve structures, starting with a fairly large
set of phases (more than 40-50 phases), even though these might be very
inaccurate. The different methods proposed differ in the way in which the
large initial set of phases is set up.
Those using magic integers are at the basis of the computer programs
MAGIC['^^] and MAG EX.["^] A limited number of phases (normally
10-15), called primary phases, are expressed as qpi= mix. If two of these
phases enter in a triplet with sufficiently large G, it will be possible to
express also the phase of the third reflection as a combination of magic
integers. For instance, if
+ +
then q s =(ml m,)x n.The new phases obtained in this way are called
secondary phases. The starting set of phases will therefore include the
origin-fixing phases, the primary and secondary phases, and their number
can be up to 60-70.
The selection of the x values to be tried is done by means of the q-map.
This is set up using all triplets relating the phases in the starting set. With a
procedure similar to that used in deriving relation (5.64), a triplet may be

written in terms of independent reflections as
where b is a constant angle depending only on the symmetry operators of

the space group. If in (5.G.2) we substitute the primary and secondary
phases expressed in terms of magic integers, we will obtain
where M is a combination of integers and B = b +other constant terms

appearing in the definition of secondary phases. The condition that all NR
relationships of type (5.G.3) among the starting-set phases should be as
close to zero as possible (or cos CD = +1) may be expressed as
NR
@(x)=
r=l
G, cos (M,x + B,) = max. (5.G.4)
@ ( x ) is a one-dimensional Fourier series and can be easily calculated at

regular intervals of x of the order of 1/(4Mm,,). The @-mapis thus obtained
and its maxima indicate the best values of x. Some 100 x values, correspond-
ing to the highest values of @@), are selected to obtain the same number of
sets of numerical values of the phases, which are then further refined and
expanded.
An extreme radicalization of the basic idea of enlarging the starting set by
allowing larger phase errors, is the random approach to the phase
problem.[1271 The starting phases are now assigned random values between 0
and 360" and then refined with different techniques.
The first proposed random procedure is at the basis of the
program. Some 100 best interrelated phases (selected by means of the
convergence procedure) form the starting set and are assigned several sets
of random values. Each set is then refined by a least-squares procedure
assuming as observational equations all triplet relations of type (5.G.2). The
use of a least-squares procedure in the case of cyclic variables defined
modulo 2n, needs an important warning. To obtain the correct phase values
it is necessary to know the values of the right-hand terms of relations such
as (5.G.2) not reduced modulo 2n; or, in another way, we must know n in
the equation CD = 2nn (n positive, zero, or negative integer). It may be
interesting to note that this is another way in which the 'phase problem'
shows up: it is generally possible to solve an overdetermined system of some
thousands linear equations in terms of some hundreds of unknowns, but in
our case, the cyclic nature of the variables introduces the new unknowns n,
the number of which is equal to the number of observational equations. We
must have some prior estimate of the n values, and this may be obtained
from the starting random values of the phases; the values obtained in this
way will in general not be integer and will be approximated to the nearest
integer. The least-squares procedure may now be carried out and new
values of the phases obtained; these in turn will yield new estimates of the
ns and the process is repeated until convergence. The advantage of this
refinement procedure, with respect to the tangent formula, is that all phases
are refined at the same time; the cost is a longer computing time. Each
phase set, after this refinement, is expanded by the tangent formula and
tested by some figures of merit. Applications of this method have shown

that it is possible to hit the correct solution starting with a limited number
(around 100) of random phase sets; the probability of getting the correct
solution decreases as the structural complexity increases.
The success of this and other random methods implies that, if N (=loo) is
the number of starting phases, then the N-dimensional space, in which the
refinement is carried out, only shows a limited number of minima; it will be
sufficient'to start from a limited number of random points in this space to
converge to the minimum corresponding to the correct solution.
A more straightforward procedure is applied in the RANT AN['^^]
program in which the random starting sets of phases, including all
reflections with large (El values, are directly refined by a carefully weighted
tangent formula. The initial random phases are assigned a weight w = 0.25
and at each cycle only the phases determined with a sufficiently high ah
value, so that wh =: min (0. 2ah, 1) > 0.25, are accepted. Also in this case a
limited number of trials (50-200, depending on the structural complexity) is
capable of yielding the correct structure. A similar random start procedure,
but with a different strategy, has been implemented within the SHELXS-
86['01 program.
More recently two new random procedures have been incorporated into
the computer programs X M Y [ ~and ~ ~SAY, ~TAN.['^']
~ ~ ~
In the first the initial random phases are refined by parameter shift.[1331
By this method all parameters are supposed independent so that each of
+
them can be varied systematically from uy - nAuj to u9 nAu, in 2n steps of
Aui; the value which maximizes (or minimizes) a selected function of the
parameters is taken as a better value for this parameter and is used in the
subsequent calculations with the next parameters. The process is continued
until self-consistency. The advantage of the parameter-shift method is its
ability to jump away from a local extremum. In the XMY procedure the
function which is maximized is
where
Xh = C
k
IEhEkEh-kl COS (9)-h + 9)k + qh-k)
and (5.G.6)
A large value of F is in agreement with the most probable value of the

triplets.
In SAYTAN random phases are refined by means of a modified tangent
formula derived from Sayre's equation (5.24) expressed in terms of
normalized structure factors. Its derivation is too long to be reported here
and we will just mention that it also includes quartet invariants.
All the described random procedures have proved very successful in
solving rather complex crystal structures.
Finally we mention an alternative random procedure[1341which has
recently been incorporated in the SIR program[671and is indicated with the

name MESS. When phases are generated at random they are in general
inconsistent with positivity and atomicity of the electron density, and only
the refinement process will make them comply with these conditions; on the
other hand, we have seen that these two conditions are at the basis of the
probability formulae for triplets or other i.s. or s.s.. In the MESS approach,
the starting phases are derived from triplets perturbed by random shifts A@
obeying a von Mises distribution; these phases will then comply with
positivity and atomicity. Also in this case the number of trials necessary to
get the solution is comparatively small and the procedure proved successful
in solving several complex structures.
5.H Procedures for completing a partial model

Weights for Fourier syntheses
Given the positions of some atoms, what weights should be given to the
coefficients in order to obtain the 'best' Fourier map revealing the
remainder of the structure? This problem was first tackled by ~ i m [ ~ ' , ' ~ ] .
With reference to Fig. 5.H.1, let
where FN is the structure factor of the complete structure, Fp is the

contribution of the P known atoms, and FQ is the contribution of the Q
unknown atoms.
Ideally we should compute a Fourier synthesis with FQas coefficients, i.e.
where a = qN- 9,; but in the right-hand side of (5.H.2) a is unknown and
the best estimate of FQ is obtained by replacing exp (ia) by its expected
value
,
(exp (ia)) = (cos a ) + i(sin a ) = (COSa ) ;
since a can equally have positive and negative values, then (sin a ) = 0 and
(5.H. 2) becomes
indicating that the 'best' Fourier synthesis is obtained by weighting the IFNI
factors by w = (cos a ) . In order to derive (cos a ) we must first know the
probability distribution of c?..
+
For FQ= AQ iBQ the probability distribution (5.A.13) becomes
1
P(AQ, BQ) = -exp [ - (A; + B2Q)lCQ] (5.H.4)
JGCQ
where CQ = C, f: is a summation over the unknown atoms. From the
triangle in Fig. 5 . ~1 .we have
Fig. 5.H.1. Vector representation i n the complex
plane of the structure factor as a sum of t w o
A$ + B ~ Q = IFQI2= IFNI2+ - 2 lFNlIFPIcos a (5. H.5) components.
The only unknown term is a and (5.H.6) is essentially a probability

distribution of a , i.e.
P ( a ) = (1/K) exp (X cos a ) , (5.H.7)
where X = 2 lFN( IFpIICQand K is a normalizing constant given by (see eqns
(5.D.15) and (5.D.17))
K= $ exp (X cos a ) da = 2nIo(X). (5.H.8)
Finally
exp (X cos a)
P(a) =
2"Io(X) .
Recalling eqn (5.68), we may now derive
w = (cos a ) = dm=- "(X) (5.H. 10)

Io(X) '
and (5 .H.3) becomes
When this weight is used in Fourier recycling, then IFN(= (Fobs(and the
employed coefficients are
Similarly ~ o o l f s o n [derived
' ~ ~ ~ that for centrosymmetric structures
FP+ (FQ)= IFobsl tanh (X12)Sp (5.H.13)
where Sp is the sign of Fp.
Syntheses for completing a partial model

We have seen that often only part of the structure (e.g. positions of heavy
atoms) can be initially determined; this information can then be combined
with the Patterson function to define the rest of the structure. For this
purpose Ramachandran and Raman[136]proposed several types of Fourier
with different coefficients. Among them we will recall the a
syntheses[13721381
and the /3 syntheses, with coefficients
a synthesis IF~(~F~
/3 synthesis 1 ~ ~ 1 ~(F;1 ~complex
; conjugate of Fp)
where the terms have the same meaning as in (5.H.1)
Solution and refinement of crystal structures ( 395
Considering (5.H.1), the Patterson coefficients IFNI2become
and those of the a synthesis
where 4 and F, contribute to the background and the P peaks, F,

contributes to the background and the desired Q peaks, and 4 contributes
mainly to the background.
A similar development for the coefficients of the /3 synthesis yields to
F ~-I- FQ+ F$ exp 2iqp
I F ~ ( ~=/ Fp +IFQ~~/F~ (5.H.16)
where the first two terms contribute, with equal weight, to reveal the known
P atoms and the unknown Q atoms respectively, while the last two terms
contribute mainly to the background. An advantage of the /3 over the a
synthesis is that it gives the electron density map when Fp = FN.
~ a i n [ l compared
~~] the P synthesis with the synthesis computed with
coefficients (5.H.12) (also called weighted y' synthesis). From (5.H.5) the
expected value
may be derived and, considering that

IF~12 = ((FP + F~)(F; + F;) (5.H.18)
the coefficients (5.H.12) are obtained as
where the last two terms mainly contribute to the background, the first term
contributes to the known part of the structure, and the second to the desired
unknown part, but only with half weight. In order to give the P and the Q
peaks the same weight, as in the case of the /3 synthesis, the coefficients of
the y' synthesis should be altered to
+
(2 IFNl(cos a) - IFPI) exp (iq,) = Fp + FQ FT,exp (2iqp)
+ {JFQ12- (lFQ12))lFg. (5.H.20)
The new synthesis is referred to as the 2F, - F, synthesis; it has been shown
that, for non-centrosymmetric structures, it contains fewer background
peaks than the /3 synthesis and is more effective in suppressing the peaks
corresponding to wrong atoms in the initial model.
Let us conclude this paragraph with a comparison, proposed by Tollin et
between the a synthesis and the optical image synthesis by
holography (see any optics textbook such as ~ e y e r - ~ r e n d t [ ' ~ lThe
]).
principle of holography is illustrated in Fig. 5.H.2. The waves scattered by
the object Q interfere with those from a coherent point source P and the
fringe pattern is recorded on a photographic plate to form a hologram (H).
The intensity pattern on the hologram will depend on the phase relation-
ships between the interfering waves from P and Q. By illuminating the
laser
n
(a) RECORDING
'reference beam
mirror
image
/ - -
5
'
laser
Fig. 5.H.2. Scheme of the formation of a
hologram. (b) IMAGE RECONSTRUCT ION
hologram with light identical to that from P, the image of the object Q may
be reconstructed. If t,(x) and t Q ( x )are the functions describing the source P
and the object Q, the corresponding scattered amplitudes will be their
Fourier transforms TP(s) and TQ(s). The amplitude at a point s on the
hologram will be
A ( s ) = Tp(s)+ TQ(s)exp (2nias) (5.H.21)
where a is the separation between t,(x) and tQ(x). The intensity at s on H
will then be
where the s dependence has been omitted. When the hologram is

illuminated by a light from the source P, the resulting modulated function
will be
+
IT, = T, 1 T , [ ~ T, IT,(^ + T $ T ; exp (-2xias) + 1 T,~~TQ
exp ( h i a s ) .
(5.H.23)
Its Fourier transform will be the combination of the transforms of the four
different terms and will show the image of the object Q. Comparison with
(5.H.15) shows that the four terms of the a synthesis coefficients are
formally similar to those in (5.H.23). In the case of a partially known
structure we observe the interference effects between the known (source)

and the unknown (object) parts of the structure. In 1950 ~ r a had~ ~ [ ~ ~ ~ ~
already qualitatively indicated the analogy between the heavy-atom tech-
nique and holography.
References
1. Patterson, A. L. (1944). Physics Review, 65, 195.
2. Wilson, A. J. C. (1950). Acta Crystallographica, 3, 397.
3. Lipson, H. and Cochran, W. (1953). The crystalline state. vol. III: The
determination of crystal structures. G. Bell, London.
4. Wilson, A. J. C. (1949). Acta Crystallographica, 2, 318.
5. Wilson, A. J. C. (1942). Nature, 150, 151.
6. Viterbo, D., Gasco, A., Serafino, A. and Mortarini, V. (1975). Acta
Crystallographica, B31, 2151.
7. Harker, D. (1936). Journal of Chemical Physics, 4, 381.
8. Calleri, M., Chiari, G . , Chiesi Villa, A., Gaetani Manfredotti, A., Guastini,
C. and Viterbo, D. (1976). Acta Crystallographica, B32, 1032.
9. Gervasio, G., Rossetti, R. and Stanghellini, P. L. (1979). Journal of Chemical
Research (S) 334, (M)3943.
10. Sheldrick, G. M. (1985). In Crystallographic computing 3 (ed. G. M. Sheldrick,
C. Kriiger and R. Goddard), pp. 184-9. Oxford University Press; Robinson,
W. and Sheldrick, G. M. (1988). In Crystallographic Computing 4 (ed. N. W.
Isaacs and R. M. Taylor), pp. 366-77. Oxford University Press.
11. Luger, P. and Fuchs, J. (1986). Acta Crystallographica, A42, 380.
12. Lenstra, A. T. H . and Schoone, J. C. (1973). Acta Crystallographica, -429,
419.
13. PavelEik, F. (1989). Journal o f Applied Crystallography, 22, 181.
14. Argos, P. and Rossmann, M. G. (1976). Acta Crystallographica, B32, 2975.
15. Rossmann, M. G., Arnold, E. and Vriend, G. (1986). Acta Crystallographica,
A42, 325.
16. Terwilliger, T. C., Sung-Hou Kim and Eisenberg, D. (1987). Acta
17. Harker, D. and Kasper, J. S. (1948). Acta Crystallographica, 1, 70.
18. Karle, J. and Hauptman, H . (1950). Acta Crystallographica, 3, 181.
19. Giacovazzo, C. (1980). Direct methods in crystallography. Academic, London.
20. Hauptman, H . and Karle, J. (1953). The solution of the phase problem. I. The
centrosymmetric crystal, ACA Monograph, No. 3. Polycrystal Book Service,
New York.
21. Sayre, D. (1952). Acta Crystallographica, 5, 60.
22. Hauptman, H . and Karle, J. (1956). Acta Crystallographica, 9, 45.
23. Cascarano, G. and Giacovazzo, C. (1983). Zeitschrift fur Kristallographie, 165,
169.
24. Cochran, W. (19%). Acta Crystallographica, 8, 473.
25. von Mises, R. (1918). Physikalisches Zeitschrift, 19, 490.
26. Karle, J. and Karle, I. L. (1966). Acta Crystallographica, 21, 849.
27. Karle, J. and Hauptman, H . (1956). Acta Crystallographica, 9, 635.
28. Cochran, W. and Woolfson, M. M. (1955). Acta Crystallographica, 8, 1.
29. Schenk, H. (1973). Acta Crystallographica, A29, 77.
30. Schenk, H . (1973). Acta Crystallographica, A29, 480.
31. Simerska, M. (1965). Czechoslovakian Journal of Physics, 6, 1.
32. Hauptman, H. (1975). Acta Crystallographica, A31, 671.
35. Giacovazzo, C. (1977). Acta Crystallographica, A33, 933.

37. Giacovazzo, C. (1974). Acta Crystallographica, A30, 626, 631.
39. Burla, M. C., Nunzi, A., Polidori, G., Busetta, B. and Giacovazzo, C. (1980).
Acta Crystallographica, A36, 573.
40. Burla, M. C., Nunzi, A., Giacovazzo, C. and Polidori, G. (1981). Acta
41. Cascarano, G., Giacovazzo, C., Calabrese, G . , Burla, M. C., Nunzi,
A , , Polidori, G. and Viterbo, D. (1984). Zeitschrift fur Kristallographie, 167,
34.
42. Giacovazzo, C., Spagna, R., VickoviC, I. and Viterbo, D. (1979). Acta
43. Cascarano, G., Giacovazzo, C., Polidori, G . , Spagna, R., and Viterbo, D.
44. Burla, M. C., Giacovazzo, C., and Polidori, G. (1989). Acta Crystallographica,
A45, 99.
45. Cascarano, G., Giacovazzo, C., Camalli, M., Spagna, R., Burla, M. C.,
Nunzi, A., and Polidori, G. (1984). Acta Crystallographica, A40, 278.
47. Busetta, B., Giacovazzo, C., Burla, M. C., Nunzi, A., Polidori, G., and
Viterbo, D. (1980). Acta Crystallographica, A36, 68.
48. Ladd, M. F. C. and Palmer, R. A. (ed.) (1980). Theory and practice of direct
methods in crystallography. Plenum, New York.
49. Cascarano, G., Giacovazzo, C., and LuiC, M. (1985). Acta Crystallographica,
A41, 544.
A43, 14.
A 4 4 176.
A44, 183.
53. Camalli, M., Giacovazzo, C., and Spagna, R. (1985). Acta Crystallographica,
A41, 605.
54. Burla, M. C., Cascarano, G., Fares, E., Giacovazzo, C., Polidori, G., and
Spagna, R. (1989). Acta Crystallographica, A45, 781.
56. Germain, G. and Woolfson, M. M. (1968). Acta Crystallographica, B24, 91.
57. Germain, G., Main, P., and Woolfson, M. M. (1970). Acta Crystallographica,
B26, 274.
58. Abramowitz, M. and Stegun, I. A. (1964). Handbook of mathematical
functions. National Bureau of Standards, Washington, D.C.
59. Burla, M. C., Cascarano, G., Giacovazzo, C., Nunzi, A , , and Polidori, G.
60. Cascarano, G., Giacovazzo, C., and Viterbo, D. (1987). Acta Crystal-
lographica, A43, 22.
61. Cascarano, G., Giacovazzo, C., Burla, M. C., Nunzi, A., and Polidori, G.
62. Cascarano, G. (1990). Doctorate thesis. University of Bari; Cascarano, G.,
Giacovazzo, C., Camalli, M., Spagna, R., and Watkin J . D. (1991). Acta
Crystallographica, A47, 373. (And references therein.) Altomare, A., Cas-
carano, G., Giacovazzo, C., Guagliardi, A., Burla, M. C., Polidori, G., and
Camalli, M. (1994). Journal of Applied Crystallography, 27, (in press).
63. Schenk, H. and Kiers, C. T. (1985). In Crystallographic computing 3 (ed. G.
M. Sheldrick, C. Kriiger, and R. Goddard), pp. 200-5. Oxford University
Press.
64. Peschar, R. and Schenk, H. (1987). Acta Crystallographica, A43, 751.

65. Karle, I. L., Karle, J., Mastropaolo, D., Camerman, A., and Camerman, N.
(1983). Acta Crystallographica, B39, 625.
66. Main, P., Fiske, S. J., Hull, S. E., Lessinger, L., Germain, G., Declercq, J.
P., and Woolfson, M. M. (1980). MULTAN80: A system of computer
programs for the automatic solution of crystal structures from x-ray diffraction
data. Universities of York and Louvain.
67. Burla, M. C., Camalli, M., Cascarano, G . , Giacovazzo, C., Polidori, G.,
Spagna, R., and Viterbo, D. (1989). Journal of Applied Crystallography, 22,
389.
68. Cerrini, S., Lamba, D., Burla, M. C., Polidori, G., and Nunzi, A. (1988). Acta
Crystallographica, C44, 489.
69. Karle, J. (1968). Acta Crystallographica, B24, 182.
70. Beurskens, P. T., Bosman, W. P., Doesburg, H. M., Gould, R. O., Van den
Hark, Th. E. M., Prick, P. A. J., Noordik, J. H., Beurskens, G., Par-
thasarathi, V., Bruins Slott, H. J., and Haltiwanger, R. C. (1985). Program
system DZRDZF, Technical Report, Crystallography Laboratory, University of
Nijmegen.
71. Sim, G. A. (1959). Acta Crystallographica, 12, 813.
72. Sim, G. A. (1960). Acta Crystallographica, 13, 511.
73. Stewart, J. M. and Hall, S. R. X T A L user's manual, Technical Report
TR1364.2. Computer Science Center, University of Maryland; Skelton, B.,
Hall, S. R., and Stewart, J. M. (1988). In Crystallographic computing 4, (ed.
N . W. Isaacs and M. R. Taylor), pp. 325-53. Oxford University Press.
74. Gabe, E. J., Le Page, Y., Charland, J. P., Lee, F. L., and White, P. S. (1989).
Journal of Applied Crystallography, 22, 384.
75. Betteridge, P. V., Prout, C. K., and Watkin, D. J. (1984). CRYSTALS.
University of Oxford; Watkin, D. J. (1988). In Crystallographic computing 4,
(ed. N . W. Isaacs and M. R. Taylor), pp. 354-65. Oxford University Press.
76. Viterbo, D., Calleri, M., Calvino, R., and Fruttero, R. (1984). Acta
Crystallographica, C40, 1728.
77. Hughes, E. W. (1941). Journal of the American Chemical Society, 63, 1737.
78. Cruickshank, D. W. J. (1970). In Crystallographic computing (ed. F. R.
79. Hirshfeld, F. L. and Rabinovich, D. (1973). Acta Crystallographica, A29, 510.
80. French, S. and Wilson, K. (1978). Acta Crystallographica, A34, 517.
81. Schwarzenbach, D. (1989). Acta Crystallographica, A45, 63.
82. Hamilton, W. C. (1964). Statistics in physical science, pp. 157-162. Ronald,
New York; (1965). Acta Crystallographica, 18, 502.
83. Bijvoet, J. M. (1949). Proceedings of the Koninklijke Nederlandse Akademie
van Wetenschap, B52, 313.
84. Rogers, D. (1981). Acta Crystallographica, A37, 734.
85. Marsh, R. E. and Herbstein, F. H. (1983). Acta Crystallographica, B39, 280.
86. Marsh, R. E. and Herbstein, F. H. (1988). Acta Crystallographica, B44, 77.
87. Santoro, A., Mighell, A. D., and Rodgers, J. R. (1980). Acta Crystal-
88. Le Page, Y. (1982). Journal of Applied Crystallography, 15, 255.
89. Mugnoli, A. (1985). Journal of Applied Crystallography, 18, 183.
90. Spek, A. L. (1988). Journal of Applied Crystallography, 21, 578.
91. Jones, P. G. (1984). Chemical Society Reviews, 13, 157.
92. Braga, D., Gradella, C. and Grepioni, F. (1989). Journal of the Chemical
Society, Dalton Transactions, 1721.
93. Aime, S., Botta, M., Gobetto, R., and Orlandi, A. (1990). Magnetic Reson-
ance in Chemistry, 28, 552.
94. Bijvoet, J. M., Peerdeman, A. F., and van Bommel, A. J. (1951). Nature, 168,
271.
95. Jones, P. G. (1984). Acta Crystallographica, A40, 660.

96. Glazer, A. M. and Stadnicka, K. (1989). Acta Crystallographica, A45, 234.
97. Flack, H. D. (1983). Acta Crystallographica, A39, 876.
98. Bernardinelli, G. and Flack, H . D. (1985). Acta Crystallographica, A41, 500;
99. (1959). International tables for x-ray crystallography, Vol. 11, pp. 355-7.
Kynoch, Birmingham.
100. Buerger, M. J. (1959). Vector space and its application in crystal structure
investigation. Wiley, New York.
101. Wrinch, D . M. (1939). Philosophical Magazine, 27, 98.
102. Allen, F. H., Bellard, S. A., Brice, M. D., Cartwright, B. A , , Doubleday, A . ,
Higgs, H . , Hummelink, T . , Hummelink-Peters, B. G., Kennard, O., Mother-
well, W. D. s . , Rodgers, J. R . , and Watson, D. G . (1979). Acta
103. Nordman, C. E. and Nakatsu, K. (1963). Journal of the American Chemical
Society, 85, 353.
104. Nordman, C. E. and Kumra, S. K. (1965). Journal of the American Chemical
Society, 87, 2059.
105. Nordman, C. E. (1985). In Crystallographic computing 3 (ed. G. M. Sheldrick,
C. Kriiger, and R. Goddard), pp. 232-9. Oxford University Press.
106. Nordman, C. E. and Hsu, L. Y. R. (1982). In Computational crystallography
(ed. D. Sayre), pp. 141-9. Oxford University Press.
107. High, D. H . and Krout, J. (1966). Acta Crystallographica, 21, 88.
108. Stout, G. H. and Jensen, L. H . (1989). X-ray structure determination, pp.
312-5. Wiley, New York.
109. Rossmann, M. G. and Blow, D. M. (1962). Acta Crystallographica, 15, 24.
110. Tollin, P. (1966). Acta Crystallographica, 21, 613.
111. Tollin, P. (1976). In Crystallographic computing techniques (ed. F. R. Ahmed),
pp. 212-21. Munksgaard, Copenhagen.
112. Wilson, C. C. and Tollin, P. (1986). Journal of Applied Crystallography, 19,
411.
113. Wilson, C. C. (1989). Acta Crystallographica, A45, 833.
114. Rossmann, M. G. (ed.) (1972). The molecular replacement method. Gordon
and Breach, New York.
115. Rius, J. and Miravitlles, C . (1987). Journal of Applied Crystallography, 20,
261.
116. Egert, E. and Sheldrick, G. M. (1985). Acta Crystallographica, A41, 262.
117. Woolfson, M. M. (1954). Acta Crystallographica, 7, 61.
118. Hamilton, W. C. (1964). Statistics in physical science, pp. 17-18. Ronald, New
York.
119. Buerger, M. J. (1956), Proceedings of the National Academy of Sciences of the
U S A , 42, 776.
120. Fan Hai-Fu, Yao Jia-Xing, and Qian Jin-Zi, (1988). Acta Crystallographica,
A44, 688.
121. White, P. S. and Woolfson, M. M. (1975). Acta Crystallographica, A31, 53.
122. Woolfson, M. M. (1976). In Crystallographic computing techniques (ed. F . R.
123. Main, P. (1977). Acta Crystallographica, A33, 750.
125. Declercq, J. P., Germain, G . , and Woolfson, M. M. (1975). Acta Crystal-
126. Hull, S. E., Viterbo, D . , Woolfson, M. M., and Zhang Shao-Hui, (1981). Acta
127. Baggio, R., Woolfson, M. M., Declercq, J. P., and Germain, G. (1978). Acta
128. Declercq, J. P., Germain, G . , and Woolfson, M. M. (1979). Acta Crystal-

lographic~,A35, 622.
129. Yao Jia-Xing, (1981). Acta Crystallographica, A37, 642.
130. Debaerdemaeker, T. and Woolfson, M. M. (1983). Acta Crystallographica,
A39, 193.
131. Debaerdemaeker, T. and Woolfson, M. M. (1989). Acta Crystallographica,
A45, 349.
132. Debaerdemaeker, T., Tate, C., and Woolfson, M. M. (1985). Acta Crystal-
133. Bhuiya, A. K. and Stanley, E. (1963). Acta Crystallographica, 16, 981.
134. Burla, M. C., Giacovazzo, C., and Polidori, G. (1987). Acta Crystallographica,
A43, 797.
135. Woolfson, M. M. (1956). Acta Crystallographica, 9, 804.
136. Ramachandran, G. N. and Raman, S. (1959). Acta Crystallographica, 12, 957.
137. Srinivasan, R. (1961). Acta Crystallographica, 14, 607.
138. Ramachandran, G. N. and Srinivasan, R. (1970). Fourier methods in crystal-
lography. Wiley-Interscience, London.
140. Tollin, P., Main, P., Rossmann, M. G., Stroke, G. W. and Restrick, R. C.
(1966). Nature, 209, 603.
141. Meyer-Arendt, J. R. (1972). Introduction to classical and modern optics, pp.
413-27, Prentice-Hall, Englewood Cliffs, N.J.
142. Bragg, W. L. (1950). Nature, 166, 399.
Ionic crystals
FERNANDO SCORDARI
Introduction
This chapter deals with a class of natural and synthetic compounds that, at
least to a first approximation, can be treated as being composed of
oppositely charged spheres. We start by illustrating some important
chemical and physical concepts that are later on applied to the organization
of ionic crystals. For instance, we shall see how it is possible to rationalize
some atomic structures in terms of the energetics of crystal formation. To
this end, the ionic model, attractive for its simplicity, will be used even if
departure from the predictions are common particularly when bonds have
quite prevalent covalent character. The close-packing model of spheres is
examined closely and used as a helpful tool to describe many structures.
Finally we see how the concepts and principles illustrated previously are
very useful for the systemization of several ionic structures with different
degrees of complexity.
The structure of the atom

Atoms with a single electron
We know from quantum that the motion of the electro~(e-)
in a hydrogen atom can be described by means of the wave function
Y(x, y, z), or orbital, and that IY(x, y, z)I2 dv expresses the probabjlity of
finding the electron in a given element of volume, dv.
The electron can assume different states of energy, each distinguished by
four quantum numbers n, 1, m, s:
(1) n, the principal quantum number, is related to the energy of thg
electron, and can have any positive integral value from 1 to w;
(2) 1, the azimuthal quantum number, expresses the angular momentum
module (= v m h/2n) (where h is Plank's constant) of the path ~f
the electron and determines the form of its orbital: orbitals are
indicated by the letters s, p, d, f, g . . . . and correspond respectively to
1 = 0 , 1, 2, 3, 4 . . . n - 1 ;
(3) m, the magnetic quantum number, (-1 s m S +1) represents the
angular momentum component in the direction of the magnetic field
and characterizes the orientation of the orbital;
(4) s, (s = fi), the spin quantum number, determines the way in which
the electron spins.
404 ( Fernando Scordari
Using the spherical coordinates r, 6 , q (see Fig. 6.1) Y(x, y, z ) can be

resolved into the products of three functions with separate variables:
Y(r, 6, q ) = R n , l ( r ) e 1 , ~ ( 6 ) @ r n ( ~ ) . (6.1)
While R;,,(r) gives the probability of finding the electron at a certain
distance from the nucleus, the form and the orientation of its orbital are
determined by 8 and @.
The functions R, 8 and @ take different algebraic forms according to the
quantum numbers with which they are associated.
\
The probability of finding the electron in a spherical shell of volume
4nr2 dr, at a distance of r from the nucleus, is 4nr2R;,,(r) dr. The function
X Dnal(r)= 4 ? ~ r ~ R ; , ~ is
( r )known as the radial probability function. It is
Fig. 6.1. Relationship between the orthogonal represented in Fig. 6.2 for orbitals Is, 2s, 2p, 3s, 3p, and 3d.
coordinates x y, zand the spherical coordinates
r, 8,v. The presence of secondary maxima, occurring before the main maximum,
produces some important effects. For example when the electron is located
in the 3s orbital it is more penetrating than that when it is in the 3p, (as
shown in Fig. 6.2) whereby we can deduce that the ionization energy
required by the 3s electron is greater than that required by the 3p electron.
Let us examine now the angular part of Y(r, 6 , q ) . The spatial shapes of
the angular probability functions are illustrated in Fig. 6.3. These configura-
tions allow us to estimate the magnitude of the v vector that is proportional
to the probability (summed over all distances) that the electron will be
found in that direction.
Atoms with more than one electron

The basic configuration of the electrons of an atom can easily be deduced by
means of the aufbau principle, or gradual building up of atoms. This
exploits the Pauli principle and a scale of the relative energy of the orbitals.
The Pauli principle can be synthesized as follows: in an atom no two
electrons exist which have all four quantum numbers, n, 1, m, and s, the
same. Therefore two electrons belonging to the same orbital, as defined by
the first three quantum numbers, must spin in opposite ways.
Fig. 6.2. Radial probability functions for orbitals

1s: 2% 2p. 3% 3p, and 3d for the hydrogen atom.
&&% r (4)
Ionic crystals 1 405
dXby2 d ~2 d x ~ ~ X Z dyz Fig. 6.3. Angular probability functions for: (a) s;

(b) P, P,, P,; and (c) dx2-,,2, dr2, dVr dxrl dyz
----I (d orbitals.
The energy of the orbitals for various neutral atoms increases in the
following order:
This experimental scale, though not always reliable (see Fig. 6.4),
Fig. 6.4. Relative energy of various orbitals

versus the atomic number Z.
406 1 Fernando Scordari
nevertheless constitutes a good guide for the building up of atomic

structures and especially for the location of valence electrons. For example
the outermost electrons of Rb and Sr occupy the 5s orbital rather than the
4d, exactly as predicted by the energy scale. The relative energy of the
orbitals versus Z is shown in Fig. 6.4.
Following the aufbau principle the orbitals are filled gradually as shown
below:
Chemical Orbital n I m s Electrons

element
On the same principle, the entire periodic table can be obtained in the
same way, as illustrated in Table 6.1.
Elements whose valence electrons are in the s orbitals belong to groups
IA or IIA: alkaline metals have only one valence electron in their s orbital,
while alkaline-earth metals have two. The elements in groups from IB to
VIIIB, known as transition metals, have their external electrons in orbitals
(n - l)d which immediately precede the last orbital ns.
Those elements whose valence electrons are in the p orbitals belong to
groups I11 to VIIIA. Such elements are known as post-transition metals and
non-metals, and are separated roughly in the Table 6.1 by a broken line.
Helium (in the hatched square) can be placed either in group IIA, taking
into account its electron configuration, or in group VIIIA, taking into
account its chemical behaviour similar to that of the other noble gases.
Finally there are the lanthanides and actinides which are characterized by a
gradual filling of the f orbitals preceding by two places (n - 2) the last
orbital s(n).
Interactions between ions

Notes on chemical bonds
A knowledge of the structure of an atom allows reasonable predictions to be
made about interactions between adjacent atoms. At this point it will be
useful to recall certain general concepts of chemistry.
The influence which the nuclear charge has on the external electrons
varies according to which orbitals they occupy and to what degree they are
shielded by the electrons closer to the nucleus (see p. 419). Such variations
in electrons attraction can be evaluated by means of: ionization energy
(IE); electronic affinity (EA); electronegativity (EN).
I E is the energy needed to remove an electron from the orbital of an
atom to an infinite distance. The closer the electron is to the nucleus the
greater the energy required. First, second energy of ionization, and so on,
indicate the increasing quantities of energy necessary to free electrons 1, 2,
etc. In accordance with recent thermodynamic conventions endothermic
Table 1. T h e p e r i o d i c t a b l e of e l e m e n t s a n d c o n f i g u r a t i o n s of i n t e r n a l e l e c t r o n s ( f i r s t c o l u m n ) a n d v a l e n c e e l e c t r o n s ( f i r s t r o w ) . T h e o x i d a t i o n s t a t e s in c o m p o u n d s a r e
g i v e n in t h e u p p e r p a r t of t h e cases, with t h e m o s t i m p o r t a n t u n d e r l i n e d
nner
!lectronc
10
18
16
i4
16
"Light metals.
t
bTran~itionmetals.
"The stair step running from boron to astatine, divides non-metals from post-transition metals.
Nobel gases.
Table 6.2. Partial series of the electronegativity of elements i n accordance with

Sander~on[~'
Element EN Element EN Element EN Element EN
H 2.31 V2 1.24 Rh 1.47 DY 0.94

-
+
He Cr2+ 1.35 Pb 1.57 Ho 0.96

Li 0.86 Mn2+ 1.44 Ag 1.72 Er 0.96
Be 1.61 Fe 1.47 Cd 1.73 Trn 0.96
B 1.88 Co2+ 1.70 In 1.88 Yb 0.96
C 2.47 Ni2+ 1.75 Sn2+ 1.58 Lu 0.96
N 2.93 Cu 1.14 Sn4+ 2.02 Hf 0.98
0 3.46 Zn2+ 1.86 Sb 2.19 Ta 1.04
F 3.92 Ga3+ 2.10 Te 2.34 W 1.13
Ne 4.38 Ge4+ 2.31 I 2.50 Re 1.19
Na 0.85 As3+ 2.53 Xe 2.63 0s 1.26
Mg 1.42 Se 2.76 Cs 0.69 Ir 1.33
Al 1.54 Br 2.96 Ba 0.93 Pt 1.36
Si 1.74 Kr 3.17 La 0.92 Au 1.72
P 2.16 Rb 0.70 Ce 0.92 Hg 1.92
S 2.66 Sr 0.96 Pr 0.92 TI+ 1.36
CI 3.28 Y 0.98 Nd 0.93 TI^+ 1.96
Ar 3.92 zr2+ 1.OO Prn 0.94 Pb2+ 1.61
K 0.74 Nb 1.12 Sm 0.94 Pb4+ 2.01
Ca 1.06 Mo 1.24 Eu 0.94
Sc 1.09 Tc 1.33 Gd 0.94
~i'+ 1.13 Ru 1.40 Tb 0.94
energy (absorbed by the atom) is marked positive, while exothermic energy

(given off by the atom) is marked negative.
E A is the energy given off when an electron is added to the valence shell
of an atom. The acquisition of a first electron by non-metallic atoms
involves the giving off of energy. However, the acquisition of a further
electron is an endothermic process, as the addition of the first electron has
brought the atom to a sort of 'saturation' level.
EN expresses the property of an atom to attract electrons to itself when it
combines. It is difficult to evaluate this property, since, unlike IE and EA, it
depends also on the nature and number of the atoms with which it bonds.
Since the vast majority of elements can form large numbers of combina-
tions, the E N cannot be an invariable value for each element. The various
methods u ~ e d [ ~to, ~define
] E N have led to various electronegativity series.
The semi-empirical series adopted here is that of ~ a n d e r s o n , [illustrated
~] in
Table 6.2.
Two atoms or ions are influenced by two forces: attraction (Fa) and
repulsion (F,). The algebraic sum of their relative energies, E, and E,, is
indicated by E. The minimum of this function, E,, in correspondence to
R = R e , represents the energy gain of a couple of ions in 'contact', as
opposed to the energy they would possess at an infinite distance. R, is the
bond length in correspondence with which the forces of attraction and
repulsion are balanced.
The bonds can be divided orientationally into two categories:
(1) strong bonds with energy E, > 90 kcal mol-l;

(2) weak bonds with energy E, < 35 kcal mol-'.
Type (1) bonds are due to direct interaction between the outermost
lonic crystals 1 409
orbital electrons. If the bond electrons tend to be concentrated in the region

between the atoms, then the bond is termed covalent (e.g. an H, molecule).
If, instead, the outermost electrons tend to be transferred from one atom to
the other, giving origin to high concentrations and impoverishments in
comparison with the original electronic densities, then the bond is referred
to as an ionic bond (e.g. an NaCl crystal). Finally, if the bond electrons
belong to the whole crystal, then the bond is known as metallic (e.g.
crystalline Na). Most compounds are characterized by bonds of an
intermediate nature which are referred to as heteropolar covalent bonds.
The degree of ionicity of a bond (ID) can be evaluated if the difference in
electronegativity between two atoms involves (X and Y) is known,
(AENxy = ENx - ENy). If the bond between them is simple, then, in
accordance with ~ a u l i n ~ ID
, [ ~can
] be estimated by means of the following
equation:
ID = 1 - exp [-O.~~(AEN,,)~]. (6.2)
Using Table 6.2 for NaC1, KC1, and LiF the following values for ID can
be obtained: 0.77, 0.80, 0.90.
Type (2) bonds are due to weaker, essentially electrostatic, forces like van
der Waals forces or hydrogen bond.['] Van der Waals forces are present in
all compounds: together with hydrogen bonds they play a primary role in
molecular crystal structure (see Chapter 7). Hydrogen bonds are fundamen-
tal for the cohesion of many compounds like proteins, molecular crystals,
hydrated salts, etc.
lonic crystals
The periodic table (Table 6.1) and Table 6.2 show that the E N of elements
varies with a certain regularity. In fact it increases much more from left to
right (along the periods) than from bottom to top (along the groups). The
further apart two elements are in a given period, the greater is the
difference in electronegativity between them, and the more markedly ionic
are the bonds of the compounds they form together. The various types of
bonds which form between the atoms play an important role in determining
geometrically the structural pattern and the physico-chemical properties
which characterize a crystalline substance.
In molecular crystals, more or less complex, finite, atomic aggregates can
be found, i.e. molecules, within which considerably stronger covalent bonds
are formed than the van der Waals bonds which are formed between the
molecules (see Chapter 7). In non-molecular crystals the protagonists are
atoms or ions (simple or complex) which are held together by more or less
localized bonds, and in limited cases of maximum polarization, by Coulomb
interactions.
This chapter will deal with non-molecular crystals, in particular ionic
crystals. The following questions will be considered:
1. What conditions must be satisfied in order that a combination of ions
may be considered stable?
2. What are the configurations that such combinations give rise to?
The energy associated with a given ions' disposition will be analysed
briefly in order to answer the first question, while the atomic building of
some typical structures will be analysed in order to answer the second.
Gibb's free energy, G = E - TS + PV, indicates the stability of a crystal-
line structure. The term P V (pressure x volume) can be omitted if the
pressure is not high, thus free energy takes the form:
where F is Helmholtz's free energy, E is the internal energy of the crystal,

T is the absolute temperature, and S is the entropy. E, known also as
cohesion energy, is the sum of two terms:
E = UL + Ev. (6.4)
UL is the static term, also known as lattice energy, (see p. 411), and
expresses the potential energy of the immobile ions in the equilibrium
positions. Ev is the dynamic term which expresses the vibration energy of
the mobile ions oscillating around the equilibrium positions. If the tempera-
ture is not very high and the structure is well ordered, i.e. if for Ev and
TS << UL, F = UL results, then the stability of a structure can be calculated
by means of UL, which, as we shall see, can be estimated experimentally
using the Born-Haber cycle.
UL can be calculated not only for ionic crystals but also for partially ionic
crystals (crystals in which there are prevalently covalent bonds as well as
prevalently ionic bonds), provided that the groups of atoms associated
mainly by covalent bonds are considered as single ions, having a charge
equal to that of the whole group.
U, is the sum of various values: electrostatic attraction energy E,, due to
the prevailing long radius forces; repulsion energy, E,, due to the prevailing
short radius forces (such forces depend both on the repulsion of the charges
(e--e-, p+-p+) and on the repulsion between fully occupied orbitals at any
attempt to superimpose them (Pauli's principle)); energy due to crystal field
stabilization, E,; energy due to covalent bonds; energy due to van der Waals
forces, etc.
E , is definitely the dominant term, being about 80 per cent of the entire
potential energy, while E , is about 10-15 per cent. E , is not negligible when
transition metals are present in the structure. The lower the AEN between
the ions combined, the more substantial is the percentage of energy due to
covalent bonds. Finally, the percentage of energy due to van der Waals
forces depends on momentary interactions between induced dipoles. This
type of force is always present, since the barycentre of the positive nucleus
does not correspond to the barycentre of the negative charges, due to
thermal agitation. Thus momentary dipoles are formed which interact
electrostatically with other induced dipoles. This section will deal only with
the first three contributions: E,, E,, E,. Chapter 7 should be consulted for
an analysis of the other terms of particular importance for molecular
crystals.
Lattice energy: the contributions of attractive and

repulsive terms
With reference to Table 6.1, compounds with prevalently ionic bonds are to
be found amongst combinations of alkaline metals, alkaline-earth metals
(groups I, IIA), part of group IIIA, transition metals with low oxidation
numbers, and non-metals (groups VI and VIIA). Lattice energy, or rather
the U, energy of a crystal, can be defined as the energy necessary at
absolute zero to break down a mol of crystal into its ionic components,
carrying them to an infinite distance. The formation of an ionic bond can be
considered as having two distinct phases in origin:
(1) a positive ion or cation forms in the following way:
(2) a negative ion or anion forms in the following way:
If ionization takes place by means of the taking off of more than one
electron, then the total IE is the sum of the partial I E S . [ ~ , ~ ]
For example, let us consider two opposite charges supposedly concen-
trated in a single point, +zle, -z2e (e being the charge or an electron). At a
distance of R from each other, they are mutually attracted by a force
from which the potential electrostatic energy of the charges can be derived:
In accordance with Born, the repulsion energy is given by:
where B is a constant of proportionality; n, Born's exponent, is a value

linked to the type of ions which form the crystal and variesM from 5 to 12.
Alternatively the repulsion energy can be expressed as follows:
E, = b exp (-RIP) (6.8)
where b and p are empirical parameters, which can be determined, like n,
by means of compressibility measures.
The total potential energy E for a couple of ions is given by:
The minimum of the function E, is obtained as follows:
hence:
Re is the equilibrium distance between two ions. Substituting (6.11) in (6.9)

gives:
.... @ @
%1
@ @ @ @ @ @ @ ...
Fig. 6.5. 'Touching' cations and anions
alternating i n a row. -
Let us calculate the potential energy in the case of a row of N couples of

ions of opposite charges, positioned as illustrated in Fig. 6.5.
For the two anions on either side of the reference cation (heavily marked
circle) and for the two cations immediately outside them, the potential
energy will be respectively:
E, = -2z1z2e2/R and E , = +2zlz2e2/(2~).
Taking into account all the ions around the marked cation the total E , is
given by:
Using the same procedure we will also obtain:

E , = 2B/Rn + 2B/(2R)" + 2B/(3R)" + . . . -2B/Rn. (6.14)
Thence in the case of a row of ions (6.9) takes the form:
Following the same procedure used in (6.10) we found B = 1.39

z , ~ ~ e ~ ~ " - Replacing
~ / 2 n . B in (6.15) and observing that the electrostatic
interactions between each ion and all the others are N , just equal to the
couples of opposite charged ions, (6.15) will be:
Equation (6.16) differs from (6.12) not only in the number of the
interactions N, but also in the constant 1.39. This constant derives from the
positions of the ions considered in Fig. 6.5: more generally it depends on
the type of three-dimensional lattice which the ions form. ~ a d e l u n g ' s [ ~ ~ ~ ]
constant expresses the geometric characteristics of the lattice, and will be
indicated here as AM. If No (Avogadro's number) is the number of couples
of ions contained in a mol, (6.16) becomes:
AMvalues for some important lattice types are given in Table 6.3.
Since a crystal is the content, in terms of atoms, of a cell which is
repeated many times in space, its potential energy can be calculated as
Table 6.3. Madelung constants, AM, for some

compounds
Structure Coordination A
,
number
CsCl
Halite (NaCI)
Zinc blende (ZnS)
Wurtzite (ZnS)
Fluorite (CaF,)
Cuprite (Cu,O)
Rutile (TiO,)
Anatase (TiO,)
follows:
(if i = j, I # 0). (6.18)
E L is the potential energy of an elementary cell, the coefficient ;derives

from the fact that every couple of ions is included twice in the calculation of
(6.18), N is the number of ions contained in the cell, zl and z2 are the
number of their charges expressed in terms of electrons, x , = x, - x, are the
interatomic vectors between the i and j ions of the cell, 1 is a generic lattice
vector, b, is the repulsion coefficient between the i and j ions, and p is the
hardness parameter due to its repulsive forces. The electrostatic potential,
represented by the first term of (6.18), diminishes gradually over long
distances. This term is in practice evaluated by means of particular series,
such as an optimized Ewald's series,[71which accelerate their convergence.
The Born-Haber cycle allows the enthalpy or heat of formation AHf of
an ionic crystal to be measured, starting from the elements. A scheme of the
cycle is shown in Fig. 6.6.
Born and Haber applied the Hess' law to ionic crystals. Interpreting the
principle of conservation of energy in thermodynamic terms, the Hess' law
states that AH, depends on the point of arrival and of departure, but not on
the number or nature of the intermediate stages of the reaction, that is to
say:
AHf = AHM AHx + + AHIE+ AHEA+ UL AHM+AHx
where AHMand AHx are respectively the sublimation energy of a metallic

-
M(g)+X(g)
crystal (M(,,) and the dissociation energy of the molecule X2 in its gaseous
state (X2(,J. AHIEand AHEAare respectively the ionization energies of the qEl FHEI
metal in its gaseous state and the electronic affinity. For example, for NaCl
an enthalpy equal to = -98.2 kcal mol-' has been determined, using
the Born-Haber cycle. Knowing that AHM= 24, AHx = 29, AHIE= 118.5,
-
M?g)+x(,)
AHEA= -83.3 kcal mol-' then UL= - 186.4 kcal mol-'. If in (6.17) we
insert the values A , = 1.74745, Re = 2.814 X lop8cm, n = 9, and introduce Fig. 6.6. Sketch of the Born-Haber cycle.
the conversion factor ergs -, kcal, we obtain:
= - 183.3 kcal mol-I,

which is very close to the value obtained from experimental data.
The agreement between experimental data and the results of the
calculations is particularly good in this case, although the method tends
generally to underestimate the value of UL: in particular the estimate is at
least 10 kcal mol-' below the experimental value for halogen compounds.
Lattice energy: CFSE contribution

The Born theory often gives UL values for transition metals, and in
particular those having incomplete d orbitals, which are too low compared
with observed values. In order to explain this discrepancy, Bethe and Vleck
formulated a theory known as the crystal field theory (cFT).['-'~]
This theory manages to interpret qualitatively, with conceptual simplicity,
a series of experimental data. The limitations of the CFT are in its
presuppositions, since it presupposes that between metal ions and ligands
interactions of an electrostatic nature only develop, while the ligands are
considered as point charges. To understand the type of action that the
ligands have on the d orbitals of the metal ions, their spatial geometry must
be borne in mind (see Fig. 6.3).
The five d orbitals can be divided into two distinct subgroups on the basis
of their orientation. The first encompasses the three orbitals, d,, d,,, and
d,,, whose lobes are orientated along the diagonals of their Cartesian axes.
The second group encompasses the two orbitals, dX2-,2 and d,2, whose lobes
are orientated along their Cartesian axes.
Let us analyse two cases:
(1) a transition metal in an octahedral coordination;
(2) a transition metal in a tetrahedral coordination.
For case (1) we will consider a cation which has all five d orbitals empty.
If the cation is isolated, the d orbitals all have the same energy (degenerate
orbitals). Let us now suppose that the cation is surrounded by point-like
negative charges q-, distributed symmetrically at equal distances from the d
orbitals (spherical field conditions). They will have the effect of increasing
the energy of the d orbitals, which will remain degenerate (Fig. 6.7(b)).
If, however, we imagine that six q- charges surround the cation,
positioned at the vertices of an octahedron and orientated in relation to the
orbitals as shown in Fig. 6.8(a), then the two subgroups of the d orbitals,
eg(dX2-,2, dZ2)and t2,(dxy, dxz, dyz), will have different energy values.
The e, orbitals (shaded) point towards the ligands and are nearer to them
than are the t2, orbitals (white), which are in an intermediate position. As a
result, the e, orbitals have a higher energy value than the t,,, and for this
reason the electrons prefer the latter. The difference in energy between the
e, and t,, orbitals, known as crystal field splitting, is indicated by A,,
(Fig. 6.7(c)). It follows the barycentre principle: the algebraic sum of the
energies of the d orbital electrons, when all the d orbitals are filled, is equal
At
I
,to0
Esf;!
@,',!'
'\OO000000 00,' -
metal ion in
s p h e r i c a l field',
I
y
I metal i o n in
, I
1 t e t r a h e d r a l field
1 metal ion i n
o c t a h e d r a l field
00 0 0 0 0 0 0 0 0 ~ ' ~ E ~ Fig. 6.7. Energy levels relative to the d orbitals

metal ion free in spherical, tetrahedral, and octahedral fields.
to that obtained when the cation is in spherical field conditions. Let us

indicate Ei the absolute energy of a d orbital of the cation when isolated and
E,, the absolute energy of a d orbital of the cation in spherical field
condition. When the cation is in an octahedral field the energies are
indicated respectively by X and Y for a t2, and an e, orbital.
In order that the d configuration in an octahedral field should have the
same energy as that in a spherical field, it is necessary that:
6(Es, - Y) + 4(Esf+ X) = lOE,, and X + Y = A,. (6.20)
The solution which satisfies (6.20) is: X = 0.6Ao, LY= 0.4Ao. It might seem,
judging by Fig. 6.7, that the free cation is energetically more stable than the
complex, since E,, > Ei. In reality the bond energy (disassociation energy of
a pair of touching ions separated into two atoms) is more than enough to
compensate for the increase in the energy of the d orbitals. Moreover, an
electron which occupies a t2, orbital has an energy value which is 0.4Ao less
than E,,. Thus, every electron which occupies one of the t,, orbitals
stabilizes the metal by 0.4Ao, while every one that fills an e, orbital reduces
the metal's stability by 0.6Ao.
The value obtained from the algebraic sum C i(fxi)Ao of all the electrons
of d orbitals (x = -0.4Ao if the electron belongs to a t2, orbital, or +0.6A0 if
it belongs to an e, orbital) is known as the crystal field stabilization energy
(CFSE) .
A, ranges between 20 and 90 kcal mol-' and depends:
on the cation charge (it increases as the charge does);
on the type of d orbital involved (it increases substantially from the 3d
orbitals to the 5d orbitals);
on the symmetry of the metal ion-ligands;
(b)
on the distance (R) between the metal ion and ligands, or to be more Fig. 6.8. Orientation of the orbitals e, and ,t of
precise, on 1/~;['1 a transition metal in (a) an octahedral crystal
field, and (b) a tetrahedral crystal field. For
on the nature of the ligands (some of which in order of increasing clarity's sake two of the three orbitals, ,,t and t,
stabilization are: I- < Br- < s-' < C1- < NO; < F- < OH- < H 2 0 < in the planes x y and xz respectively, are not
shown. They are oriented at an angle of 45" to
NO;, spectrochemical series). the x axis.
Let us now analyse how the electrons are positioned in the d orbitals. The
first three electrons occupy the three t,, orbitals.
The fourth electron has two possibilities: it may occupy an e, orbital, in
accordance with the Hund's first rule, or maximum multiplicity principle
(an atom, in its fundamental state, follows the maximum multiplicity
principle: in a partially filled orbital the number of electrons having the
same spin orbital is maximum), or it may associate with another electron
already in a t2, orbital, which spins in the opposite direction, thus further
stabilizing the complex.
Two factors influence the fourth electron and decide which tendency will
prevail: the A, and the pairing energy, P. The latter can be defined as the
energy necessary:
(1) to overcome the repulsion which exists between two electrons which
occupy the same orbital;
(2) to compensate the exchange energy which is lost when Hund's first rule
is violated.
If the A, > P, then the fourth electron will go to fill a t2, orbital, forming a
low-spin complex (two electrons which spin in opposite direction plus two
electrons which spin in the same way give a compound which is not very
paramagnetic). If the A, < P, then the fourth electron will go to fill an e,
orbital, forming a high-spin complex (four electrons which spin in the same
way give a very paramagnetic compound).
Table 6.4 shows the CFSE relative to high- and low-spin configurations,
for octahedral sites.
We will now examine the case of a transition metal in a tetrahedral
configuration. Figure 6.8(b) shows four ligands at the vertices of a
tetrahedron. The three d orbitals, which we shall call tZ, are nearer to the
ligand than are the other two e orbitals. This implies that, unlike an
octahedral coordination, in a tetrahedral coordination the three t2 orbitals
have more energy than the two e orbitals (Fig. 6.7(a)).
The crystal field separation value is indicated by A, and is inferior to the
Ao(A,lA, = 419) for two reasons:
(1) because there is no direct interaction between the ligands and the d
Table 6.4. CFSE for transition metals of the first series i n octahedral configurations
Number of Ion High-spin configuration Low-spin configuration

d electrons
t,, e, CFSE (A,) t,, e, CFSE(A,)
0 ca2+sc3+Ti4+ 0 0.0 0 0.0

1 1 -0.4 1 -0.4
2 -,-i2+v3+ 2 -0.8 2 -0.8
3 v2+cr3+Mn4+ 3 -1.2 3 -1.2
4 cr2+~n3+ 3 1 -0.6 4 -1.6+ P
5 ~n'+Fe~' 3 2 0.0 5 -2.0 +2P
6 ~ e ' + ~ o ~ + N i ~ 4+ 2 -0.4+P 6 -2.4 +3P
7 co2+Ni3+ 5 2 -0.8+2P 6 1 -1.8+3P
8 Ni2+ 6 2 -1.2+3P 6 2 -1.2+3P
9 cu2+ 6 3 -0.6+4P 6 3 -0.6+4P
10 2n2+~a3+Ge4+ 6 4 0.0+ 5P 6 4 O.O+5P
orbitals of the metal ion; Table 6.5. CFSE for transition metals of the
first series in tetrahedral configurations. For
(2) because there are four ligands instead of six. the ions involved see Table 6.4
This explains why, as experience has demonstrated, tetrahedral complexes Number of High-spin configuration
d electrons
have only high-spin configurations. e t, CFSE(A,)
Table 4.5 shows the CFSE relative to high-spin configurations for
tetrahedral sites.
At this point it will be useful to list the factors, which are at times
mutually opposed, which govern cation coordination:
voluminous ligands favour tetrahedral coordination (see p. 425);
the position of the ligand in the spectrochemical series (see p. 415)
favours tetrahedral coordination if on the left, octahedral if on the
right:
a high ligand charge favours octahedral coordination, because it
increases Ao, which in turn favours the formation of low-spin com-
plexes, increasing the number of electrons in tZgorbitals;
the CFSE favours octahedral complexes, both because A. = (9/4)A,,
and because the CFSE of tetrahedral complexes is generally less than
those of octahedral complexes (cf. Tables 6.4 and 6.5).
If the CFSEs of Table 6.5 are multiplied by the scale factor 419, they can
be compared with the CFSE of Table 6.4, and the octahedral site
stabilization energy, OSSE (i.e. the difference between the CFSEs con-
cerning octahedral and tetrahedral sites), can be calculated in A. units (see
Table 6.6).
Applications of lattice energy calculations

Lattice energy calculations make it possible not only to rationalize a series
of observations arising from structural analyses, but also to resolve certain
problems without recourse to X-ray techniques. Amongst the numerous
applications of lattice energy calculations, easily computed with the aid of
Table 6.6. OSSE for transition metals of the first series. For the ions involved
see Table 6.4
Number of CFSE, CFSE, OSSE CFSE, OSSE

d electrons A, unit high spin high spin low spin low spin
various automatic programmes,[72132141

are the following:
(1) the distribution of cations in different sites;
(2) the effect of the distribution of certain cations in certain sites on the
position of others;
(3) structural predictions;
(4) solutions to structural problems;
(5) the determination of charge distribution in complex ions;
(6) the estimation of the elastic constants of stable minerals subjected to
very high pressures;
(7) the effects of temperature and pressure on structures
Let us examine some examples in more detail. U, was calculated for 16
isostructural compounds with olivine, including certain berylites, borates,
phosphates, and silicates.[1s1
The structure of olivine, space group (s.g.) Pbnm, can be described as a
distorted hexagonal closest-packed system of oxygen atoms (see p. 444).
The six coordinated cations occupy two sites which are not crystal-
lographically equivalent, M(1) and M(2) respectively, at the centre
(0.0,0.0,0.0) and on the mirror plane. In silicates with an olivine-type
structure these cations may be monovalent, bivalent, or trivalent (Li+, Na+,
M ~ ' + ,Fez+, c a 2 + , Fe3+, etc.).
Potential energy calculations indicate that if the ionic radii of the cations
are comparable, then the cations with smaller charges will prefer the M(l)
site. If, instead, the ionic radii are of quite different sizes, then the cation
with the greater volume will prefer the M(2) site. In accordance with the
calculation, where re(Li+) = 0.76 = re(Sc3+)= 0,745 A, the compound
Li+Sc3+[Si04]is found to have Li+ in M(l) and Sc3+ in M(2). Moreover,
where r e ( ~ g 2 +=) 0.72 # re(Ca2+)= 1.00 A, the compound MgCa[Si04],
monticellite, has Mg2+ in M(l) and Ca2+ in M(2).
Another example is the influence of the Al/Si distribution on the sites
occupied by Na+ in albite (AlSi,O,) at high temperatures.[161Here too the
calculations agree with the experimental data, that is to say, the concentra-
tion of A13+ in a particular tetrahedral site determines the site which the
Na+ will occupy.
The determination of the structure of an inorganic crystal by means only
of calculations of potential energy has serious practical drawbacks if suitable
crystallo-chemical information is lacking initially. It would be necessary to
do an enormous number of trials, moving the atoms in all possible
directions, if their precise positions are not known initially. ~ e c e n t l ~ , [ ' ~ ]
starting from the ideal structure of olivine, it has been noticed that the
compact regular layers of oxygen atoms used as input in the calculations are
deformed in the process of potential energy minimization, just as the actual
structure shows them to be.
Ionic radius
An important factor which influences the geometry of the structure of ionic
compounds is the so-called ionic radius. It expresses the 'dimensions' of an
ion considered as a rigid sphere. Since a finite value of r, when D,,[(r) = 0

(Fig. 6.2), does not exist, 'atomic radius', and thus also 'ionic radius', are
merely conventional terms. Nevertheless, data derived from structural
determination by means of X-ray techniques suggest that d M x (the interionic
distance between a cation M and an anion X) is the sum of the two 'radii'
characteristic of the two adjacent ions.
The problem which needs solving is how to divide d M x between the anion
and the cation. With this aim, ~ a u l i n g [considered
~] certain alkaline halides,
characterized by pairs of isoelectronic ions such as NaF, KC1, RbBr, CsI,
and calculated their effective charge Z,, = Z - S. Z expresses the formal
charge, while S indicates a screening effect which depends on the distance
of the electrons from the nucleus and on which orbital they occupy. This
effect can be explained on the basis of the radial distribution curves (Fig.
6.2), which show how s orbital electrons provide better shielding for
electrons of other orbitals than they, in turn, receive from them. For
example, 2p orbital electrons are better shielded than the 2s orbital
electrons, because the latter are nearer to the nucleus. In order to evaluate
S, certain principles suggested by Slater are followed:
1. Group the electrons as follows:
2. Use the multiplication coefficients 0.35 and 0.85 for electrons belonging
to the groups (ns, np) and ((n - l)s, (n - 1)p) respectively and 1 for
those belonging to the group ((n - 2)s, (n - 2)p) or to any preceding
group. It is preferable to use 0.30 to 0.35 in the case of groups (Is).
3. If the electrons belong to groups (nd) or (nf), use the coefficients 0.35
and 1, respectively, for electrons which belong to the same group, or for
those which belong to the next group to the left.
For example, for the compound NaF:
and, in accordance with ~ a u l i n ~ ,since

[ ~ ] the ionic radius is inversely
proportional to Z,,, and it is known that d M x = 2.32 A, we can obtain:
By resolving the system (6.21), re(Na+) = 0.95 and re(F-) = 1.36 A can be
obtained, which agree perfectly with the ionic radii quoted by ~auling.[~I
This procedure can be extended to analogous isoelectronic compounds.
Various ionic radius systems have been proposed, amongst the most
accurate, thanks to the large number of structures examined, are those
which have appeared Effective ionic radii extracted from
Shannon's original work[lglare given in Table 6.7. These systems show that:
1, Cations are generally smaller than anions. The average radius of the
outermost occupied orbital varies considerably from the least electronega-
tive atom to the cation, but very little from the most electronegative atom to
the anion, e.g. ro(Li) = 1.586, ro(Li+)= 0.186 A and r,(K) = 2.162, ro(K+)=
Table 6.7. Effective ionic radii ( r e ) versus coordination number (CN) according to Shannon[lgl
Ion CN Spin re Ion CN Spin re Ion CN Spin re

Table 6.7. (Continued)
Ion CN Spin re Ion CN Spin re Ion CN Spin re Ion CN Spin re
Na'
Na'
Nb3+
Nb4+
Nb5+
~ d ~ +
Nd3+
Ni2+
Ni3+
Ni4+
yo2+
Np2+
NP3+
NP4+
Np5+
NP6+
Np7'
o2-
OH-
os4+
os5+
os6+
os6+
os7+
os8+
p3+
p5+
pa3+
pa4+
pa5+
pb2+
0.592 A. In the anions the added electron increases the radius by a few per
cent, e.g. ro(C1) = 0.725 A, ro(C1-) = 0.742 A and ro(Br) = 0.851, r,(Br-) =
0.869 A.
2. The dimensions of the atoms decrease along the period (with n a
constant), the alkaline metals having the largest dimensions, and increases
sharply along the group from top to bottom (with n a variable). This occurs
because the dimensions of an atom depend both on n and on Zeff,which act
on the atomic radius producing contrasting effects. Thus an increase in n
tends to increase the atomic volume (the most probable radius increases
with n, see Fig. 6.2), while an increase in Z,, tends to contract orbitals. For
example, the Z,, calculated with Slater's principles for Na and C1 are
respectively 2.20 and 6.10, while for Li and Cs they are 1.3 and 2.2. In the
first case the sharp increase in Z,,, which is not countered by n (which is
constant), induces a decrease in the dimensions from Na to C1. In the
second case the regular increase in n, hardly countered at all by the effects
of Z,,, induces a notable increase in the atomic volumes from Li to Cs.
Obviously this also follows for the ionic radius.
3. The ionic radius varies with the coordination number (CN). Note that
the repulsion forces increases with the CN, which in turn increases the
effective ionic radius re of the ion of the opposite charge (the cation). The
lattice energy allows this trend to be quantified approximately. (6.15) can be
rewritten as:
E = - AMzlz2e2+-CNB
R R"
where AM is Madelung's constant (which has taken the place of 1.39) and
CN, which is defined below, is the coordination number (which has taken
the place of the number 2, that is the two anions coordinated by the cation,
see Fig. 6.5 and eqn (6.13)).
For R = Re, dEldR = 0, from which we can work out:
Let us suppose that ions of opposite charges can crystallize just as well
into a CsCl type structure as an NaCl (halite) or ZnS (wurtzite) one.
Assuming that the intermediate value n = 9, we obtain:
The result of (6.24) indicates that the inter-ionic distances of a CsCl type
lattice (R,) are 2.3 per cent greater than those of an NaCl type lattice (R2),
while (6.25) shows that those of a ZnS type lattice (R3) are smaller than it
by =4 per cent. Since the anion radii remain more or less the same, the
variations occur almost entirely in the cation radius.
4. The ionic radius depends on the electronic spin state. There is a close
relationship between ionic radius and spin state in transition metals. In fact,
as explained on p. 415, A. and A, depend on the type of ligands, and the

coordination polyhedra which the ligands form around the transition metal.
If, for example, A. is high, then the first six electrons tend to occupy the
lower energy level orbitals, while if A. is low they occupy both energy levels
(Fig. 6.7). These differences in turn affect the ionic radii of the transition
metals which have from four to seven electrons in their d orbitals. To be
more precise, electrons with high spin have a larger radius than those with
low spin (Fig. 6.9).
5. The ionic radius is influenced by polarization. Two ions of opposite
charges, if they are far enough apart, have coinciding barycentres of their
positive and negative charges. When the ions are brought together a
distortion of the electronic charge known as polarization, takes place, which
is greater in the anion. Combination requires a sort of participation on the
part of the electronic charge, and therefore a certain degree of polarization,
or covalent bonding.
In accordance with ~ajan,['O]the polarization depends on:
(a) the dimensions of the cation and its charge;
(b) the dimensions of the anion and its charge;
(c) the deviation of the electronic configuration from that of a noble gas.
Fig. 6.9. Dependence of the effective ionic radii

on the spin state in transition metals of the 1st
series: (a) = bivalent; (b) =trivalent. The full
circles refer t o low-spin configuration, the open
circles refer t o high-spin configuration.
This means, for example, that for alkaline metals the polarization power
increases according to the following scale: Li+ > Nai > K+ > Rbf > Cs+
(ionic potential, 0 = Z / r , increases from right to left).
Regarding the anions, the greater the dimensions of their outermost
electrons, the weaker are their bonds, due both to the greater distance from
the nucleus and to the increasing shield effect. It follows that the
polarizability of halogens varies according to the scale: I- > Br- > C1- > F-,
analogously with that of haicogens. Anions with a high negative charge,
such as As3- and p3-, are particularly easily polarized. Moreover, it has
been shown above how the different peripheral electrons are shielded by the
s, p, and d type electrons. Therefore, although ~ i ' +has approximately the
same ionic radius as Mg,'+ it polarizes more than Mg2+ one extra
contiguous anion: this effect is even more noticeable between Cu+ and Naf.
Maximum filling principle

The lattice energy of No pairs of ions can be expressed as the sum of the
potential energy contributions (uij) relative to all the possible pairs whose
components are distant rij:
1 'No
UL=- 2
2 i.j=1
uij(ri,) (with i # j). (6.26)
The lattice energy is minimal if the total potential energy is minimal, this
occurs when:
(1) the interatomic distances are as close as possible to the equilibrium
distances (R,);
(2) the number of ions at, or very nearly at, the equilibrium distance is as
high as possible.
These conditions are general and apply to crystals having simple structural
units, or complex units connected to similar units by central or near central
forces of attraction.
The maximum filling principle can be expressed as follows: 'Simple
(atoms or ions) or complex (groups of atoms or ions) structural units on
which central or near central attractive forces act tend to increase contact
with each other to the maximum, whilst reducing their distances to a
minimum'.
By supposing that the ions behave as if they were rigid spheres on which
central, or almost central, forces are acting, we can hope to:
(1) predict the most probable arrangement of the anions which surround
the cation (coordination polyhedra), by means of a simple geometric
rule (radius ratio rule);
(2) describe certain atomic building by means of dense mechanical models

based on rigid spheres of equal radius (closest packings).
Coordination polyhedra
It is sometimes convenient to illustrate crystal structures by means of
coordination polyhedra, in order to underline the geometrical relationships
which characterize structural frameworks. Often such polyhedra continue to
exist beyond the crystal state as definite physico-chemical realities, e.g.
[siO4I4- tetrahedra in silicate melts, or Fe(H20),.-,(SO4), octahedra where
1< n < 1.3, in aqueous solution of iron s ~ l ~ h a t e . [ ~ ~ ]
The term coordination is used to refer to the atoms or ions which
surround a central atom or ion. If it is not otherwise specified below, it may
be assumed that the central ion is the cation (M) surrounded by the anions
(XI.
The Xs at a tangent to M are known as the first nearest neighbours of M
and constitute the first coordination shell. The coordination number (CN)
is equal to the number of the first nearest neighbours, i.e. to the number of
bonds formed by an atom. The next Xs, at a tangent to the first nearest
neighbours, are known as the second nearest neighbours of M, and form the
second coordination shell, and so on. Sometimes the first coordination
shell is not easily distinguished from the second. In such cases the
uncertainty is expressed by attributing two numbers to the CN, e.g. 6 2. +
The distances dMx are all equal when M occupies certain symmetrical
positions, e.g. M in 3, with CN = 6; otherwise the dMx vary, if only a little,
from their average.
The coordination polyhedron of M can be obtained by 'joining the dots',
i.e. all the centres of X. Some of the possible coordination polyhedra are
shown in Figs. 6.10 and 6.11. They can be found in structures as isolated
polyhedra, and/or connected to polyhedra of the same or another type.
Radius ratio rule

If the ions are considered as behaving like rigid spheres, an attempt can be
made to predict by geometrical means which coordinations will be pre-
ferred. The parameter used for this purpose is p, = (r,/r,).
Figure 6.11 shows some of the most common regular geometrical
dispositions and their relative p, ratios. The p, values for a given anion
increases as rM, and thus the number of contacts also increase. From Fig.
6.11 it can be deduced, for example, that within the range 0.155 S p, <
0.225 a triangular coordination is stable, since for p, < 0.225 cations and
anions are no longer adjacent in tetrahedral coordination; thus it is no
longer possible. However, the results obtained by geometrical methods
must be integrated with further energetic considerations. Fig. 6.10. Some coordination polyhedra. (1)
tetragonal pyramid, (2) trigonal bipyramid, (3)
Figure 6.12 shows the trend of U, versus p, for CsCl, NaC1, and ZnS type trigonal prism, (4) one-cap trigonal prism, (5)
structures having the following ratios: seven-vertex polyhedron, (6) pentagonal
bipyramid, (7) trigonal dodecahedron, (8) two-
cap trigonal prism, (9) three-cap trigonal prism,
(10) icosahedron, (11) truncated tetrahedron,
(12) rhombododecahedron, (13)
pentagonododecahedron, (14) truncated
This formalism not only indicates the coordination number around each octahedron. The coordination number of each
ion, but also allows their stoichiometry to be deduced, which for the three polyhedron is indicated below it.
Anions Polyhedron Relationships between rM and rx
Dumb-bell
B
A
H C
Triangle
Tetrahedron
B
Square
-. D ~ A
/ C B
Octahedron
Square antiprism A B = 2 r x , A O = r M + r X , m 1= (AD - F G ) / 2

or AH' = r X g ,AH = r x m , 00' = = / 2 , O'C = r X G
Thomson's cube (00')' + (m)' +
= ( A O ) 2 , (rM/rX)' 2rM/rx- ( G + 2 ) / 2 = 0
Cube
Hexagonal cuboc.
Cuboctahedron
Fig. 6.11. Other very common coordination

polyhedra. At the side of each polyhedron there
is its coordination number, name, and the
procedure followed to calculate and p,.
compounds in question is M:X = 1:1. From Fig: 6.12 it results that:

(1) the UL trends are characterized by three discontinuity points in
correspondence to the values p, = 0.732, 0.414, 0.225;
(2) the UL trends favour a CsCl type structure when p, > 0.71, and an NaCl
type when 0.71 > p, > 0.32;
(3) the UL(NaCl) curve is closer to that of the UL(CsC1) than to that of the
UL(ZnS).
Fig. 6.12. U, pattern versus p,. From the point

of view of the energy, the minimum p, values
for CsCI, NaCI, and ZnS type structures are
lower than those predicted by geometrical
means. The difference is most accentuated i n
the range between an NaCl and a ZnS type
structure (0.32 instead of 0.41).
The reasons why these discontinuities exist become clearer on observing

Fig. 6.13. In (b) an octahedron of M(X), composition is shown, while in (a)
and in (c) the same polyhedron for a larger cation (M' > M) and for a
smaller one ( M < M) are illustrated respectively. In accordance with (6.17),
the value of UL(NaC1) decreases gradually from configuration (a) to
configuration (b), since Re < R:. However, UL does not vary in the range
0.32 < p, < 0.414, that is to say, between configurations (b) and (c), since
the anions already adjacent in (b) cannot come any closer together. Thus Re
has the same value in both configurations (b) and (c). Configuration (d) is
preferred to (c) if, in accordance with the UL(NaC1) trend of Fig. 6.12, there
is another cation M"' < M so that p, < 0.32. In the same way, an NaCl type
structure is likely to be preferred to a CsCl type when p, < 0.71, i.e., for p, PO @ .-,
values a little below the discontinuity point (p, = 0.732).
L-
--3 (4 (b)
Applications of the concept of ionic radius
To assume that an atom is rigid, and also spherical, is obviously a gross
simplification of the physical reality. Accurate studies[221to determine p(r)
have shown, for example, that in MgO the electronic cloud of the oxygen
only partially wraps the cation: therefore the terms sphere and ionic radius
lose any meaning. Despite this, however, the concept of ionic radius often Fig. 6.13. (a), (b), and (c) illustrate the changing
produces satisfactory results. Alkaline halides and, to a lesser extent, relationships between equal anions (white
alkaline-earth metal oxides are the compounds to which a mechanical circles) in an octahedron, as the cations vary i n
size (broken circles). Configuration (c) becomes
atomic model is most easily applied. Such compounds with an MX unstable and could evolve into configuration (d)
stoichiometry should have structures with a CN = 8, if the ionic radii r,(M) if pe < 0.32.
LI + Na" K+ RbiCs' and r,(X) are more or less the same. Otherwise, such compounds are
governed by the cation coordination number (since the cation is generally
smaller than the anion) and the stoichiometry. The dimensions of the cation
depend on the CN (see p. 422). Therefore ambiguity remains as to the ionic
radius dimensions when the structure (and thus the CN) is not known. Let
us compare the theoretical data with the experimental data for the halides.
'Average' ionic radii for a CN = 6 are used, for example, CsBr (r, = 1.67,
rx = 1.96 A) forms a compound Cs': Br- = 8:8, in accordance with both the
radii ratio: (p, = 0.85) and the stoichiometry (M:X = 1:l). Also for NaCl
both the p, = 0.53 and the stoichiometry indicate a compound Na+:Cl- =
6:6, in accordance with the results of structural analysis.
The rule of the radius ratio predicts M:X = 4:4 structures in the higher
field T (p, < 0.41), M:X = 6:6 in the intermediate field 0 (0.41 G p, < 0.73),
and M:X = 8:8 in the lower field C (p, > 0.73). Experimental data, on the
other hand, shows that almost all the structures are of M:X = 6:6 type, two
are of M:X = 8:8 type (CsI, CsBr), and one, CsC1, can crystallize in either
type of coordination.
In Fig. 6.14 a broken line is shown for p, = 0.32, which corresponds to
the point of intersection of the NaCl and ZnS curves in Fig. 6.12. The first
Fig. 6.14. Three fields are shown (T, 0, C), anomaly involves the T field. The incongruence between the observed
delimited by t w o unbroken lines (p, = structure M:X = 6:6 and the predicted structure M:X = 4:4 by the radius
0.73,0.41). In accordance with geometrical
criteria, they represent stability areas for the ratio rule is only apparent. Considering Fig. 6.12, NaCl type configurations
alkaline halides. The broken line (p, = 0.32) are stable for 0.32 < p, < 0.41, i.e. the values observed. The other anomaly
shows the reduced T field according t o the
energy indications of Fig. 6.12. The diamonds involving the lower field, C, is constituted by M:X = 6:6 structures and can
and squares show the type of coordination be explained by considering not only the interactions between ions of
polyhedron on which the real halide structures opposite charge, but also those between ions of similar charge in the CN
are based: octahedra and cubes respectively.
function, and the geometric relationships between the coordination poly-
hedra (see p. 435).
However, if re(IVLi+)were used for LiI, p, = 0.27 would be found to be
the right value. The configuration Li+:I- = 4:4 would gain potential energy
and would therefore seem likely to be preferred, which is contrary to the
experimental data.
Another example can be illustrated by means of Fig. 6.15, as deduced by
~ h i l l i p s . [ ~The
~ I 'fields of stability' for the most important structures of
A2B04 composition are shown. The diagram was obtained empirically on
the basis of about 130 structures using the effective radii of the cations in
octahedra1 (r,) and tetrahedral (r,) coordinations. Such a diagram makes it
possible to work out the field in which a certain A,BO, structure prevails,
even if some borderline cases may be contrary to the general trend.
Improvements in structural predictions can be obtained by considering not
only geometrical factors but also chemical factors, such as polarization.
Graphs for compositions of M,X, type have been obtained by plotting
the average quantum number n (which indicates roughly the dimensions of
the ions) on the y axis and the electronegativity difference, AEN (which
indicates the degree of polarization of the bond),[241on the x axis. The
Fig. 6.15. Taking into account ionic radii r,(A) prediction is improved if p,AEN is used instead of AEN.
and r,(B) the stability fields for A2B04 structures
are shown. These fields conform t o the Finally using effective mean ionic radii, several have found
following structure types: (a) 0-K2S04, (b) quantitative relations between crystallographic and compositional para-
Na2S04(thenardite), (c) (Mg, Fe), SiO, (olivine),
(d) Be2Si04 (phenakite), (e) A12Mg0, (spinel), (f) meters in garnets. So according to as so,[^^] it is possible to calculate the
Sr2Pb04, (g) K2NiF4, (h) FeJaO,, (i) A12Ba04. oxygen fractional coordinates, the metal-oxygen distances, and the cell
edge of oxide garnets and hydrogarnets as follows:
where r(X), r(Y), and r(Z) are the mean ionic radii respectively of cations
8, 6, and 4 coordinated, and (OH) are the number of OH groups present in
the formula unit (f.u.) of garnets:
Closest packings
Most metals, some compounds of the type MX, MX, etc., and others with a
more complex stoichiometry, take on structures which can be described by
means of closest packings of rigid spheres of equal radius. Let us denote by
packing coefficient (ci) the ratio:
ci = C (VaIVc) (6.27)
where C Va is the volume of the atoms or anions contained in the
elementary cell of volume Vc. For discussion of the concept of ci see
Appendix 6.A. According to the maximum filling principle, ci + max.
Hexagonal closest packing (HCP) and cubic closest packing (CCP) satisfy
this requirement. When analysing their characteristics, it should be borne in
mind that in the common structures, ions, atoms, or groups of either,
replace rigid spheres forming the so-called closest-packing structures. The
basis of the HCP and CCP dispositions is a layer of spheres (layer A)
packed as closely as possible similar to that illustrated in Fig. 6.16.
This layer has the following characteristics: every sphere is in contact with
another six spheres; between every three spheres there is a space, known as
a hole, every sphere is surrounded by six holes, one third of each of which
belongs to it, so a total of two holes can be associated with each sphere.
Such holes, which are all alike from a physical point of view, are indicated
by b (white) and c (black). Spheres belonging to the next layer may
correspond to the b or the c holes of layer A, and are thus said to belong to
Fig. 6.16. A compact layer of spheres

characterized by triangular cavities indicated by
white spaces (b) and black spaces (c). At the top
of the sketch on the left is shown the symmetry
of the layer A (6m), while on the right is shown
the symmetry of a pair of layers AB (3m).
layer B or layer C. With reference to a hexagonal mesh of sides a, and a2,

in general a sphere may correspond to any of three possible holes: a (for
layer A) belonging to site (Oa, + Oa2+ nh) b to site ((2/3)a, + (1/3)a2+
nh), and c to site ((1/3)al + (2/3)a2 + nh), where n is a whole positive
number including zero and h is a vector perpendicular to the compact layer
of a module h (which is equal to the distance between two successive
compact layers). HCP is characterized by an ABAB . . . type sequence,
while CCP is characterized by an ABCABC . . . type sequence.
The two characteristic sequences which represent HCP and CCP and their
orientation with respect to the compact layers, are illustrated in Fig. 6.17.
The HCP cell is projected in the direction [OOl] 11 h, while the CCP cell is
projected in the direction [ I l l ] 11 h. An HCP cell contans two spheres (that
of a simple hexagonal cell contains one); a CCP cell contains four, as does a
face-centred cubic (FCC) lattice (for this reason it is possible to use FCC or
CCP with the same meaning). It is possible to identify four types of
polyhedra in both packings: cuboctahedra, octahedra, tetrahedra, and
triangles. The first, illustrated in Fig. 6.11, can be traced by joining the
centres of the 12 spheres around a central sphere of equal radius. The other
three can be traced by joining the centres of the spheres which correspond
to the three possible holes of the closest packings: octahedral, tetrahedral,
and triangular.
Fig. 6.17. An ABAB.. . sequence of layers

determines a compact hexagonal cell (top),
while an ABCABC . . . sequence determines a
compact cubic cell (bottom).
Many cations have a radius of just the right size, compared with that of
the oxygen atoms, that they can occupy tetrahedral or octahedral holes. Let
us examine, therefore, these two types of holes more closely in order to find
out what is their quantitative ratio to the spheres. In Fig. 6.18 a layer A
(black spheres) and a layer B (white spheres) are shown.
Supposing that the sequence is an ABAB . . . type, a sphere marked S
(belonging to layer A and not visible in the figure) is surrounded by a total
of 12 spheres (six of A and six of two B layers, above and below A). It
contributes to the formation of:
(1) six octahedral holes ( 0 ) three of which are shown in the figure; the -.-.
other three can be obtained by a symmetry plane m which passes
Fig. 6.18. Octahedral, 0, (top right) and
through A; tetrahedral. T. (bottom right) holes with relative
coordination polyhedra &a& out from t w o
(2) eight tetrahedral holes, three of which, shown in the figure, are in fo~~o~in~compact~a~e~~.
correspondence with the three spheres marked with a thick T and one
of which is in correspondence with the sphere marked S and three of
the spheres marked with the same T; the other four can be obtained by
a symmetry plane m which passes through A.
Since six spheres are needed to form an octahedral hole and every sphere
helps to form six holes, obviously there is a ratio of one hole to each sphere.
Four spheres are needed to form each tetrahedral hole, while each sphere
helps to form eight tetrahedral holes, so there is a ratio of two tetrahedral
holes to each sphere.
A layer of octahedral holes (OL) is enclosed by two layers of spheres, as
are two layers of tetrahedral holes (TL), through they are orientated in
different directions (see Fig. 6.32). This situation is common both to HCP
and CCP, consequently a double layer of spheres has the following
characteristics in both HCP and CCP:
1. Octahedra which share edges.

2. Tetrahedra which, differently orientated (TL, and TL,), share edges.
3. Octahedra and tetrahedra which share faces.
The differences between HCP and CCP are in the geometric elements
shared by the polyhedra in the direction perpendicular to the layers; these
are the differences which concern us most here. HCP has all the octahedra
and half of the tetrahedra sharing faces with polyhedra of the same type, the
other half of the tetrahedra sharing vertices, and finally the octahedra and
tetrahedra sharing vertices and edges. Instead, CCP has (Fig. 6.19) all the
octahedra sharing faces with half of the tetrahedra, the other half of the
tetrahedra sharing vertices, and finally the octahedra which share edges, as
do half of the tetrahedra.
It follows that, for example, while in HCP the octahedra share faces in
the direction perpendicular to the OL and edges parallel to the OL, in CCP
the octahedra share edges only. Such differences are very important for
certain ~toichiometries[~~] as we shall see below, since they affect the Fig. 6.19. An FCC cell. The tetrahedral (black
stability of the structure and thus which type of packing is preferred. triangles) and octahedral (white diamonds)
There are two coefficients which characterize closest packings. The first is holes of a CCP are shown. Note that the number
q (6.27), which has a value of 0.74 both for HCP and CCP. It is easily ~~~~~[~~j:~:","~",:~~~~~$~~
found, remembering that an elementary HCP cell contains two spheres of that of the tetrahedral holes.
Fig. 6.20. The centres of the three spheres

marked A and the centre of the sphere marked B
together form a trigonalpyamid. It canbe
deduced from this t h a ~ BH': = D V 3 / 2 , HH'=
l / 3 ( B H 1 )= DV3/6, h = ~ 9 6 1 3h, / D = q 6 / 3 =
0.8165.
while a CCP one contains four (Fig. 6.17). For example, the side of an
elementary CCP cell will be a,, = ~ j 6 where, D is the diameter of a sphere.
The volume of a sphere is V, = 4.189(D/2)3 from which c, = 4v,/a3 = 0.74.
The other parameter is the ratio hlD. Since h = [ ( ~ V ' 3 / 2 ) ~ -
" ~ Fig. 6.20) it follows that h l D = 0.8165. For example, in
( ~ f i / 6 ) ~ ](see
HCP, co = 2h, therefore colD = 1.63.
Combinations of HCP and CCP can create an indefinite number of
closest-packing structures. The repetition period, nh, will be in the range
2h S nh ~ p , where
, p , is the maximum dimension of the crystal perpen-
dicular to the closest-packed layers which the sequence can reach when the
structure is completely disordered.
As n increases, the number of possible sequence combinations grows very
rapidly. For example, there is only one possible packing for n =4:
ABACA; for n = 6 there are two: ABCACBA and ABABACA; but for
n = 20 the number of possible packing rises to 4625. One of the following
ways can be chosen to derive the possible packings: the first is to count all
the possible sequences once having generated them;[301the second is to
obtain them by means of combinatorial analysis.[311Figure 6.16 shows the
symmetry of a closest-packed layer (6mm) and of a pair of such layers (3m).
The minimum symmetry content common to all closest packings is P3ml.
Other symmetries are represented by all those space groups which can have
as a subgroup P3ml (see also Fig. 1.E.1), i.e. ~ 3 m 1 ,~ 6 m 2 ,Pb3mc,
P6,/mmc, R3m, ~ g m ~, m 3 m .For HCP and CCP the space groups are
respectively: P63/mmc and ~ m 3 m .
When real structures are considered the symmetry may be reduced, due
not only to the arrangement of the cations which occupy the various holes,
but also to the distortions which the cations produce in the lattice.
Therefore, some structures belonging to systems different from those just
examined can be described as closest packings, distorted to a greater or
lesser degree.
There are other close packings characterized by spheres with fewer next
neighbours. These are the PTP (primitive tetragonal packed), the BCT
(body-centred tetragonal), and the BCC (body-centred cubic). The first has
c, = 0.72 and CN = 11, the second (not found in the known structures) has
c, = 0.70 and CN = 10, the third (taken on by several structures) has
+
c, = 0.68 and 8 6 spheres of the same type coordinated around a central
Fig. 6.21. A BCC cell. The central sphere is sphere, eight at a distance of D and six at a distance of 1.150 (Fig. 6.21).
surrounded by eight spheres at a distance of D
and six more distant spheres at 1.150 ( 2 0 = 1V3, HCP and CCP, on the other hand, have 12 spheres all at the same distance
where / i s the side of the cell). from the central sphere.
Pauling's rules
Ionic bonds play an important role in the building up of structures for the
vast majority of minerals. In fact, even the groups (Si04)4-, (SO4)'-,
(C0,)'-, which have prevalently covalent bonds internally, can be seen as
large negative ions bonded to the cations by essentially ionic interactions.
The general principle of the minimization of potential energy can be applied
to such compounds. The Pauling rules are derived from this principle.
Pauling's first rule

This rule synthesises the geometric relationships which tend to be estab-
lished between cations and anions in a stable structure, introducing the
concept of the ionic radius.
A coordination polyhedron is formed around each cation. The distance
between the cation and the anion is determined by the sum of the radii,
while the coordination number of the cation is determined by the radius
ratio.
The maximum number of anions, compatible with the equilibrium
conditions imposed by the minimum potential energy, surround the cation.
Too many anions would increase the cation-anion distance, thus making the
structure less stable than if the distances are very close to Re. The optimum
+
distance is the sum of the ionic radii: rM r,. However, the cation-anion
distances do not depend only on the ionic radii of the cation and anion
involved, but also on the average radius of each of the other cations.[321In
structures of a certain complexity, it can happen that not all the cations have
optimum CN and dMx distances. In such cases some cations, above all the
largest ones with the lowest valence (formal charge), are the first to take on
higher-energy coordinations, followed by smaller higher-valence cations
which tend to form more rigid coordinations.
It follows that the lowest coordination expected for Na+ (which taking
into account the oxygen has p,=0.71) is 6, but in sodalite,
Na8C12[Si6A16024], it has CN = 4 (302- + ICl-), while Si4+,half of which is
substituted by AP+, has its usual CN, i.e. 4.
Pauling's second rule

Known as the valence sum rule, this rule analyses the electrostatic balance:
The valence of an anion in a stable ionic structure tends to compensate
the strengths of the electrostatic bonds which reach it from the cations
situated at the centres of the polyhedra of which the anion is a vertice,
and vice versa.
In practice, this rule means that the electrostatic balance is satisfied by the
cations belonging to the primary coordination.
If -V is the valence of an anion, vi and (CN,), are respectively the
valence and the coordination number of the ith cation, then we can define
the electrostatic bond strength as si = V~/(CN,)~. The rule can be sum-
marized as:
v=& (6.28)
i
where i is extended to all the cations coordinated by the anion in question.

For high-symmetry, simple ionic structures it is possible to give a single

value to the bond strengths and to find the valence of a given ion by
multiplying the bond strength i times. For example, for NaCl (space group
~ m 3 m ) the
, bond strength which reaches a given C1- from a given Na+ is
116. Since there are 6 Na+ surrounding each C1-, all equidistant from it
follows:
The same procedure can be applied as a rough guide to rutile, Ti02

(space group P42mnm) whose structure is based on slightly distorted Ti06
~ + coordinates six 02-and around each oxygen
octahedra. Each ~ i cation
there are three Ti4+, This means that each oxygen receives 416 of the
valence of titanium. Since very 0'- is surrounded by three ~ i ~then:
+ ,
which is exactly the opposite of the oxygen valence.

In structures with irregular polyhedra, like brookite (a polymorph of the
compound TiO,) which is characterized by strongly distorted Ti06 oc-
tahedra, this simple model needs to be modified slightly. In order to
calculate the strength of each electrostatic bond, or more correctly the bond
valence (s), Brown and ~ h a n n o n [proposed
~~] the following equation:
where so, Ro, and N are empirical constants. For so = 1, Ro = R,, the
parameter R1 represents the expected bond length when the valence is
assumed to be one. In such as a case (6.29) becomes:
By inserting the correct values for R, and N for each cation[341and the
bond distance R, which can be deduced from the structure, s can be
determined. For example Na+, Mg2+, A13+, Si4+,P5+, and s6+have values
of R1 = 1.622 A and N = 4.290. The experimental bond valence can be
obtained by applying (6.30). The agreement between the experimental and
the theoretical valence is very good, generally the difference does not
exceed 5 per cent. If the agreement is not satisfactory for an ionic structure
it can be ascribed to the following:
(1) the structure has not been correctly determined;
(2) the structure has not been interpreted properly, or certain bonds have
been overlooked, or the wrong atoms have been ascribed to certain
sites (e.g. Si4+ in place of A13+ or vice versa).
(3) The cations have stereochemically active lone-pair electrons (e.g. sb3+).
Amongst the various applications of this method[351it is often used by
crystallographers to distinguish 0'- from (OH)- and H 2 0 in some complex
mineral structures.[361Calculated bond valences can be used through (6.28)
to calculate experimental atomic valence and so the oxidation state of an
Pauling's third rule

This summarizes, from a qualitative point of view, the influence that the fact
that the coordination polyhedra share geometric elements has on the
potential energy and thus on the stability of the structure:
The sharing of sides and, in particular, of faces between coordination
polyhedra reduces the stability of a structure. This effect is accentuated
where there are cations with high valence and low coordination number.
In order for the potential energy to be minimal it is necessary that:
(1) the distances between the cations should be as great as possible,
compatible with the existence of the structure itself,
(2) the cations should be shielded as much as possible from each other.
Figure 6.22 illustrates the distances between cations for certain regular
polyhedra. As can be seen, the fulfilment of conditions (1) and (2) are less
likely, for any given pair of polyhedra, when they share an edge than when
they share only a vertice, and least likely when a face is shared. The least
probable of all the configurations are two tetrahedra which share an edge,
and above all two tetrahedra which share a face. This explains why in
silicates the tetrahedra [SiO4I4- are found either isolated or sharing vertices;
they hardly ever share edges and never share faces. For tetrahedra
containing cations with a higher charge, like [PO4I3-, even the likelihood
Fig. 6.22. Geometrical relationships between

some regular polyhedra. Considering the
oxygens as touching rigid spheres with a radius
of re = 1.40 A, the following inter-cation
distances can be calculated: = 3.43, 4, =
1.98, 0(,=1.14, do,=3.96, %,=2.80, d,,=2.29,
4,=4.85, dc, =3.96, d,, = 2.80A. The sketch at
the bottom shows the different shielding effects
produced by three or four anions together on
the underlying cation.
that they will share vertices is reduced, so much so that in the case of group
having yet higher cation charges, like [S0,I2-, that it is almost nil.
Figure 6.22 shows that do, = dCf.However, in octahedral coordinations
the shielding of the two cations is more accentuated. This explains why
numerous halites tend to keep the NaCl structure rather than that of the
CsCl, favoured by p,. Structures like CsCl are only possible if the anions
are large enough to keep the cations far enough apart, as can be observed in
the case of CsC1, CsBr, CsI (in the NaCl structure type, the distance
between two cations for CsF is 4.24 A and 4.92 A for CsCl). It should be
remembered here that, in accordance with the indications of Fig. 6.22, ionic
compounds with an MX stoichiometry (see p. 437) crystallize not with an
HCP lattice, but with a CCP lattice (e.g. NaC1, MgO, CaO, etc.). This is
because the energy levels involved in the formation of the two packings seen
along h is different: in the case of CCP the octahedra MX6 share edges,
while in the case of HCP they share faces.
On the basis of what has been said, the question of how to realize the
relative stability of certain ~ i 0 polymorphs
~ [ ~ can ~ ~be understood. This
compound has three polymorphic modifications: rutile, brookite, and
anatase. The structural differences concern mainly the way in which the
TiO, octahedra are linked to one another. In rutile, brookite, and anatase,
each octahedron shares respectively two, three, and four edges with a
similar number of other octahedra. The structural stability of brookite is less
than that of rutile, while anatase is the least stable, as predicted by the third
Pauling rule.
Pauling's fourth rule

The fourth rule is an obvious corollary of the third, and so no further
comment is necessary:
In a crystal containing various cations, those with high valence and a
low coordination number tend not to allow the coordination polyhedra
which they form to share any features.
Pauling's fifth rule

The fifth rule, known also as the parsimony principle, states that:
The number of essentially different constituents in a crystal tends to be
small.
This rule summarizes the common observation that most of the non-fluid
substances that surround us are crystalline and not amorphous. This means
that the number of structural units, having considerable differences for
instance with regard to volume and chemical properties, tends to be low for
any given substances.
Source applications of these rules together with symmetry and other
crystallochemica1parameters are given in Appendix 6.B.
Ideal a n d defect structures

The preceding paragraphs have dealt mainly with the principles which
underlie the structural building of ionic compounds.
Some simple, or at least not very complex, atomic configurations,

characterized at most by few dozen atoms per elementary cell, will now be
examined. Particular attention will be given to ideal structures, i.e. those
structures in which it is possible to single out an elementary cell 100 per cent
representative of all the cells that make up the crystal. In such cases all the
crystallographic sites which are symmetrically equivalent (including the
translation) must be occupied by the same chemical species and the
resulting compound is defined as an ideal chemical formula.
In defect structures the relative positions of the atoms are not regular
throughout the crystal. Thus the representative cell is a statistical cell which
expresses the probability that a certain site will be occupied by a given
atom. In such cases the compound is better defined by a crystallochemical
formula, which, apart from taking into account the stoichiometry, considers
also the percentage of presences (occupancy) of a certain chemical element
in a given crystallographic site. For example, the compound MgA1204
(spinel) parent of the spinels, is a normal spinel (see p. 442) and adopts the
following ideal structure: Mg2+ occupies only the tetrahedral sites, while
A13+ occupies the octahedral ones. The compound Fe304 (magnetite) is
instead an inverse spinel (see p. 442) in which Fe2+ occupies only
octahedral sites, while Fe3+ occupies half tetrahedral sites and half
octahedral ones.
Therefore the structure of magnetite can be represented by a particular
cell of statistical significance having its tetrahedral sites occupied by Fe3+
and its octahedral sites occupied by 0.5 Fe2+ and 0.5 Fe3+. This structural
peculiarity is evident from the crystallochemical formula Fe3+ (Fe::,
Fe;:),O4 which substitutes the less explanatory stoichiometric formula
Fe304. Moreover, the composition of a non-stoichiometric magnetite
Fe,,O,, can be rationalized from its crystallochemical formula in the
following way: ~ e ~ + ( ~ eFe:&, , 00.0,)204.In fact the reason why it is not
stoichiometric becomes clear immediately: it is due to the substitution of
Fe2+ by Fe3+,with the necessary abandoning of 5 per cent of the octahedral
sites and the formation of gaps (represented by the square).
From the point of view of symmetry, the space group ~ d 3 madopted both
by the spinel and by the magnetite, represents in the first case true
symmetry and in the second statistical symmetry.
IG the next section some classes of simple structures, defined on the basis
of their stoichiometry, will be analysed briefly. The silicates, included
amongst the ionic compounds by tradition, will be considered briefly from
the point of view of their classification.
MX structures
As predicted by the third Pauling rule, the packing definitely preferred by
MX compounds is CCP, while BCC is rare, and HCP is never adopted. It
can be seen from the stoichiometry of these compounds that all the
octahedra are occupied and that therefore an HCP-based structure is
decidedly less probable than a CCP-based structure. In the first case, in fact,
the octahedra share edges and faces, while in the second the octahedra only
share edges. The non-existence of hexaognal packing in the structures of
MX compounds can be best explained with help of Fig. 6.23.
Fig. 6.23. In (a) an unstable layer of cations

(white spheres) and anions (black spheres) of A compact layer like that shown in Fig. 6.23(a) is not stable due to the
equal radius R / 2 , is illustrated. By moving
alternate rows in the direction of the arrows a strength with which like charges repel each other. Such effects are
distance of R / 2 , the stable layer illustrated in (b) considerably attenuated if alternate rows are translated a distance of R/2 in
is obtained. In (c) the structure of NaCI, or halite
is shown.
the direction of the arrows (Fig. 6.23(a)) to give a regular one positive-one
negative ion pattern (Fig. 6.23(b)). In order that a succession of layers of
the same type (along the perpendicular to the layer examined) should be
stable it is necessary that a second layer be translated by R. The successive
translation (equal to R) of a third layer brings us back to the initial
configuration. Such a succession of layers gives rise to a cubic structure like
that of NaCl, space group Fm3m, or in other words, a CCP of C1- ions with
all the octahedral sites filled by Na+.
This tendency is strengthened if the cations are bivalent, which is why
MX compounds almost always adopt CCP type packings. Apart from the
halides previously mentioned (see p. 427), other examples are: periclase
(MgO), wustite (FeO), bunsenite (NiO), manganosite (MnO), and lime
(CaO). Some alkaline halides also crystallize naturally: villiaumite (NaF),
halite (NaCl), sylvine (KCl), and carobbite (KF).
MX2 and M,X structures

Such compounds present a wide range of structures, of which only a few are
illustated here.
Fluorite (CaF,), cubic, space group Fm3m, is shown in Fig. 6.24. Each
cation Ca2+ is surrounded by eight F- positioned at the vertices of a cube,
while each anion F- is surrounded by four Ca2+ positioned at the vertices of
a tdtrahedron (Ca2+:F- = 8:4). The structure adopts a quasi-closest packing
of F- similar to BCC, with half of the sites at the centre of the cube empty
and half occupied by Ca2+ alternately, in substitution of F- (p, > 0.73) at
112, 112, 112. Due to the alternate filling of these sites the cubes share edges
instead of faces, and consequently the cations are further apart and better
shielded (see Fig. 6.22). Various compounds such as SrCl,, BaCl,, and
others with more complex formulae in which some cations are substituted
by other cations e.g. 2Ca2+ -t K+ + ~ a ~ 3Ca2+ + , + 2K+ + U4+ (for +
Ca2+ OF- read 'are substituted by'), etc. have a CaF, type structure.
Fig. 6.24. A representation of the structure of The ratio M:X = 6:3 is favoured when p, < 0.73 allowing the formation of
fluorite, CaF,. In the figure the cubic
coordination polyhedra of the ca2+ (black
structures similar to that of rutile (TiO,). The tetragonal structure of rutile,
circles) are outlined, together with the space group P4,/mnm, can be described as HCP. Figure 6.25 illustrates a
geometrical relationships between them. The layer of octahedra (OL) seen (a) perpendicular to and (b) parallel to its
elementary cell (dotted lines) consists of four
such polyhedra occupied by calcium and four extension, while Fig. 6.26(a) shows a packing of OL layers of the type
similar empty polyhedra. shown in Fig. 6.25(b). The black octahedral are occupied, the white are
empty. Figure 6.26(b) shows a structure of rutile obtained by the distortion

of the ideal HCP as seen in Fig. 6.26(a). This is in agreemwnt with p, and
with the stoichiometry of rutile which predicts that half of the octahedral
holes will be occupied. The occupation of the holes occurs in such a way
that the octahedra share edges and vertices, never faces, as is illustrated in
Fig. 6.26(c).
This feature means that HCP is preferred to CCP, in which the octahedra
share edges only. Minerals with a structure like that of rutile include:
sellaite (MgF,), pyrolusite (MnO,), cassiterite (SnO,), stishovite (SiO,) a
high-pressure compound, with Si4+hexacoordinated. A TiO, type structure
is certainly one of the most common among strongly ionic binary com-
pounds such as fluorides and oxides. It is also common amongst more
complex compounds in which the Ti4+ is substituted by cations with a
similar radius: e.g. if 3 ~ i - +
~,+~ e , + 2 ~ a ~
then
+ tapiolite (FeTa206) is Fig. 6.25. An octahedral layer (OL) obtained by
formed, or if instead 3Ti4++ ~ g , +2Sb5+ + bystromite (MgSb206) is combining t w o compact layers of spheres of
equal radius. In (a) it is seen perpendicular t o its
formed. extension, in (b) parallel t o its extension.
Another structure typical of compounds with a ratio of M:X = 6:3 is that
of brucite (Mg(OH),): trigonal with space group ~ 3 m l It . has a layered
structure which can be described by means of HCP. There are two possible
ways for half of the octahedral HCP holes to be occupied without the
octahedra sharing faces: the first has already been illustrated for rutile; the
second involves all the octahedra of alternate octahedral layers being
occupied (e.g. OLA occupied entirely and OLB left completely free in Fig.
6.26(a)). Brucite is an example of the second type. It is more advantageous
from the point of view of energy to fill every other octahedron of each row,
as happens in rutile, rather than to occupy all the octahedra of OLA and
none of OLB. This is because in the first case the octahedra are linked by
(six) vertices and (two) edges, but in the second by (six) edges only. The
layered structure of rutile is due to the permanent OH- dipoles which link Fig. 6.26. In (a) (on the right) a sequence of ideal
together the fully occupied octahedral layer;@]they are known as 'brucitic .
compact layers ABAB.. with the corresponding
sequence of octahedral layers OLA-OLB-OLA-
layers'. Minerals with a brucite type structure include: portlandite .
OLB.. (on the left), seen parallel t o their
(Ca(OH),), pyrochroite (Mn(OH),), nibrucite (Mg4NiO(OH),). Another extension. The rows of black octahedra
group of minerals with brucitic layers, space group ~ 3 mwhich
, conform to (perpendicular to the sheet) are occupied, while
the white ones are empty. In (b) the distorted
CCP is: chlormagnesite (MgCl,), lawrencite (FeCl,), scacchite (MnCI,). closest packing is shown, while in (c) the rows of
If the cation-anion positions of fluorite are inverted, an anti-fluorite occupied octahedra are outlined, and their
extension along the axis c, i n the structure of
rutile TiO,.
-A
OLB
-B
OLA
-A
OLB
-B
OLA
-A
structure is formed with a ratio of M:X = 4:8, as seen in the compounds

Li20, Na20, K 2 0 , Rb20. Such a structure can be described as a CCP
having all its tetrahedral holes occupied by monovalent cations. Note that in
this case an HCP would be impossible, since the tretahedra would have to
share faces, thus bringing the alkaline cations into close contact.
MX3 and M2X3 structures

Bayerite and gibbsite, two polymorphic modifications of the compound
Al(OH),, have MX, stoichiometry: they are both monoclinic with space
groups C2/m and P2,ln respectively. Their structure (M:X = 6:2) is based
on a layer similar to that of brucite. As can be seen from their
stoichiometry, only one third of their octahedral holes are occupied, while
the presence of OH- groups indicates that their structure is layered like that
of brucite. It is useful to double the stoichiometric formula: A12(OH),, and
then to divide it into (OH); and Al2(OH);+. This expedient makes it easy
to illustrate that there is a layer of unoccupied octahedra, which overlies
another layer of octahedra two-thirds occupied (Fig. 6.27(a)). An HCP type
packing, with partially occupied, similarly orientated octahedral layers,
alternating with empty octahedral layers results[391(bayerite, Fig. 6.27(b)).
Leaving the relative positions of the cations out of consideration, in
gibbsite, two occupied octahedral layers, like that shown in Fig. 6.27(a), are
related by a plane of symmetry m which passes through the empty layer, so
the same orientation of the layers recurs every other full layer (Fig.
6.27(c)). In actual fact the arrangement of A13+ cations does not allow for
the plane m. Moreover, the substitution of 3 ~ g 4 ~ 2A13+
+ + 0,reduces
the symmetry from trigonal (brucite) to monoclinic (bayerite and gibbsite).
Both corundum (a-A1203) and haematite (a-Fe203)have an M:X = 6:4
ratio, are rhombohedra1 with space group R ~ C Their . structures are
characterized by a bayerite-type layer (Fig. 6.27(a)). It can be deduced from
their stoichiometry that the octahedral holes are two thirds occupied. The
structure of corundum can be obtained translating the bayerite layers by 113
(-01 + a,).
Perpendicularly to the bayerite-type layers the octahedral holes are
ordered so that one empty octahedron is followed by two occupied
Fig. 6.27. (a) The layer characteristic of bayerite,
AI(OH),; 1 indicates the mesh of bayerite, 2 that octahedra which share a face. This explains not only the type of lattice, R,
of corundum, a-Al,O, whose structure is based but also why the structure is stable. In fact the translation has the same
on the same type of layer; (b) shows the
structure of bayerite seen in the direction (010)
effect as a rotation of the bayerite type layer through 120" would have with
as a sequence of alternately empty and occupied the consequent turning of the empty octahedral holes around a threefold
OL; (c) shows the sequence of OL typical of
gibbsite, ANOH),.
OLB
-
1
I
OLA C
- I
OLB I
screw axis. As far as stability is concerned, it is indeed the presence of the

two empty octahedra, above and below the two occupied octahedra, that
allows the two cations which populate them to move as far apart as possible.
This improves the shielding, since the three oxygens shared are brought
closer together, and thus reduces the electrostatic repulsion. The octahedra
occupied by ~ l and ~ Fe3+
+ are, in fact, rather distorted, having three longer
distances and three shorter ones (Al-0 = 1.86(x3) and 1.97 A(x3);
Fe-0 = 1.95(x3) and 2.21 A(x3)).
A,B,X, structures
A very large number of compounds of great geological, crystallochemical,
and applicative interest belong to this class. Some structures are illustrated
below in which A and B represent monovalent to pentavalent metals and X
is generally oxygen.
llmenite (FeTiO,) and perovskite (CaTiO,) have ABX, stoichiometry.
Ilmenite is isostructural with corundum, in which the cations are substituted
+
according to the following scheme: 2A13+ (Fez+,~ g ' + ) ~ i ~ Such + .
cations are ordered so that the octahedra of one layer are occupied by Fez+
and those of the next layer are occupied by Ti4+. It follows that two
contiguous octahedra belonging to two adjacent layers are occupied by ~ e ' +
and by ~ i ~so+that, the average valence (+3) of the two cations remains the
same as that of the two corresponding cations in corundum. The specializa-
tion of the octahedral sites, half occupied only by Fez+ and half only by
Ti4+,determines the disappearance of the glide c. Thus the space group R ~ C
(corundum) becomes R3. A number of structures with various valences
(shown in brackets) have a similar structure to ilmenite, including
LiNbO,(+ 1, +5); MgTiO,, FeTi03(+2, $4); Mn(Fe, Sb)03, a-AlZO3, a-
Fez03 (the last two can be considered as particular cases of the structure of
ilmenite) and Ti203, V203(+3, +3). In these compounds A and B have
similar ionic radii ( ( r e ) = 0.65 A), therefore they occupy two octahedral
holes of a CCP type packing.
If, however, the cation is too large to fit into the octahedral holes,
another type of structure with the same stoichiometry is formed (perovskite
type). Thus, as Na+, K+, Ca2+,Sr2+,Ba2+,pb2+,etc., cations have an ionic
radius equal to that of the anions when have CN = 12, they can substitute an
X (0'-, F-) anion. If one quarter of the X anions are substituted by such
cations, then a particular type of CCP is obtained, a representative cell of
which is illustrated in Fig. 6.28.
Fig. 6.28. Contents of an elementary cell of

perovskite, CaTiO,. Note the coordination 12 of
the ca2+ (hatched circles). The octahedral holes
not occupied by caZ+ are marked X.
Ti4+ occupies one quarter of the octahedral holes available, i.e. only of
those which have no direct contact with the larger cations. Compounds with
a similar structure to that of perovskite include: NaNb03(+l, + S ) ;
BaZr03(+2, +4); LaA103, LaFe03(+3, +3).
The term spinels is used to indicate a considerable group of compounds
with a similar structure, having the general formula AB2X4.This term is the
extension to the group of the name of spinel (MgA120,), a mineral
characterized by a slightly distorted CCP type structure. In normal spinels
the cations marked A occupy tetrahedral holes (t), while those marked B
occupy octahedral holes (0). The range of their radii depends not only on
the tolerance of the structure, but also on the type of anion: X = 0'-, s*-,
se2-, Te2-, F-, CN-.
According to the stoichiometry of spinels one eighth of the tetrahedral
and half of the octahedral holes of an FCC cell should be occupied. In
actual fact, such an arrangement would result in an unstable structure
because of the need for occupied tetrahedral and octahedral holes to share
faces. However, the structure becomes stable if two FCC subcells share, so
to speak, their work: essentially one gives space to cations in coordination
of 4(T), while the other only gives space to those in coordinations of 6(0)
(see Fig. 6.29).
In this case, in fact, the occupied tetrahedra and octahedra share vertices.
This leads to the succession of cells TOT . . . along a l , a2, and a3, which in
turn leads to a cell consisting of eight subcells with a total atomic content of
A,B1,X3,. In this section only spinels with AB204will be considered. Type
B cations can have a charge of +1, +2, or +3, while type A cations can
Fig. 8.29. (a) A projection, parallel to a,, of the have a corresponding charge of +6, +4, or $2 V.U.(valence unit). Thus
elementary cell of spinel (MgAI,O,). The number from the point of view of valence, there are three possible types of spinel:
next to the circles indicate the heights ( ~ 1 1 8 )
respectively of M ~ (odd)~ + AI,+ (even, above the (+6, +I), (+4, +2), (+2, +3).
circles) and 0'- (even pairs, below the circles). In inverse or anti-spinels, the type A cations occupy octahedral sites,
In (b) the contents of two subcells T and 0,
which alternate regularly along, a,, a,, and a,
whereas the type B cations are equally divided between octahedral and
giving rise to the spinel cell (illustrated in (c)),
are outlined.
Ionic crystals ( 443
tetrahedral sites. Consequently, the formula that describes normal spinels is

(A),(B2),04 while that of inverse spinels is (B),(AB),O,.
The two formulae represent extreme cases, since spinels of various
degrees of inversion commonly exist. The following general formula for
spinels can be given (A,-,B,),(A,B2-,),04, where i (inversion parameter)
represents the fraction of A cations which are hexacoordinated.
The choice of configuration, inverse or normal, and the degree of
inversion depend on a delicate network of factors[61which can be sum-
marized as follows: (1) temperature; (2) Madelung's constant and u
parameter; (3) order-disorder phenomena; (4) ionic radius; (5) charge; (6)
CFSE; (7) polarization.
If we establish the origin of the cell of spinels (space group ~ d 3 m at
) 43m,
the equivalent 32 positions of the anions in an ideal packing, can be
generated, starting with the atomic coordinates u, u, u (u = 318). Therefore
the distortion of the actual packing can be measured by means of the u
parameter. In order to illustrate better the structure of spinels, Fig. 6.29 was
obtained by translating the origin of the cell by 118, 118, 118. Only (+2, +3)
spinels will be analysed here.
Factor (1). Recently, by means of neutron and X-ray diffraction, the
range (750-850 "C according to some authors,[401600-700 "C according to
others[411)within which transition of the order-disorder type occurs in the
spinel MgA1204 has been specified. Figure 6.30(a) shows the effect of
temperatureL411on the u parameter. u falls rapidly in the area within the
broken lines, which corresponds to the transition normal + inverse spinel.
However, it would seem that this transition is not complete, since it stops at
1200 "C, when i = 0.30.
Factors (2) and (3) can be examined together. It has already been shown
the Madelung's constant, AM, depends on the lattice geometry. The
distortion of the lattice, which can be measured by means of u, influences
the value of AM;moreover, since AM and U, are in direct proportion (see
eqn (6.17)), the configuration with the highest AM value is favoured. The
trend of AMplotted against u is illustrated in Fig. 6.30(b) for normal spinels
(ns) , ordered inverse spinels (ois) , and disordered inverse spinels (dis) .
From the figure it can be seen that for u > 0.379, AM(ns)> A,(dis) and that
for u >0.381, AM(ns)>AM(ois). This is the main reason why almost all
($2, $3) spinels with u > 0.381 are normal.
Factors (4) and (5) counteract one another. Factor (4) indicates that the
smallest cations ought to occupy the tetrahedral sites, while the largest
ought to occupy the octahedral sites. Since B ~ type + cations are very often
smaller than type cations, this means that (+2, +3) spinels should be
prevalently inverse. Experience, however, shows that exactly the opposite is
true. Factor (5) takes into account the Pauling7s second rule, according to
which the charge of the cation is more or less neutralized by the anions of .375 -380 .385
the first coordination shell. Consequently, the cations with the highest U
charges (B3+) would tend to occupy the sites with the highest coordination (b)
number (in this case octahedral sites), thus favouring normal (+2, +3) Fig. 6.30. Shows the trend of the distortion
spinels.
Factor (6) considers the effects of the crystal field on transition metals.
-
parameter, u,versus the ternDerature ("C) for
MgAI,O,. Note the rapid fall bf u (at 650 "C)
which indicates the rapid inversion of the spinel.
The latter, due to the weak crystal field of the oxygens, almost all adopt (b) Illustrates the trend of the Madelung
constant A, versus u. The abbreviations ns, ois,
high-spin configurations in spinels. It should be remembered that cations and dis indicate normal spinel, ordered inverse
with a do or d5 electronic configuration (spherosymmetrical) include: Mn2+, spinel, and disordered inverse spinel.
~ e ~Zn2+,+ , Ga3+. It is possible to predict accurately which type of spinel

will occur, if the AM and the OSSE indications agree. If, on the other hand,
they do not agree, one or other type of spinel may prevail, depending on
the balance of all the factors involved. For example, (+2, +3) type spinels
such as MgV204, MgCr204, MgMn204, are normal, since they have
u = 0.385 (and thus AM favours ns); moreover, Mg2+ is spherosymmetrical
and the OSSE definitely suggests that octahedral sites will be preferred by
v 3 + , Cr3+, ~ n ~ + .
Now let us consider the spinels NiMn204, NiFe204. They have respec-
tively u = 0.383, 0.381, therefore there is a slight AM bias towards ns. On
the other hand, the OSSE favours Ni2+ rather than Mn3+ while for Fe3+ it is
nil. Hence spinels with highest degree of inversion are the result.
Finally the minerals magnesioferrite (MgFe204), magnetite (FeFe204),
and trevorite (NiFe204) have respectively i = 0.90, 1.0, 1.0. In the first
compound the OSSE of Fe3+ is zero and AM= 0.382, so there is no clear
indication. However, Fe3+ tends to prefer tetrahedral sites, owing not only
to the polarization (factor (7)), but also to its relatively small ionic radius
(re(Fe3+)= 0.645, rc(Mg2+)= 0.72 A). In the other two compounds AMdoes
not give clear indications, while the OSSE of Fe2+ and Ni2+ decidedly
favour these two cations for octahedral sites, therefore the degree of
inversion is maximum. t
Factor (7) can play a decisive role when the other factors counteract each
other. Therefore, if AM does not clearly favour either of the two
configurations, polarization can play an important part either strengthening,
or even overturning the A, indications.
The effects of polarization are particularly noticeable where ~ e ~Ga3+, + ,
and 1n3+ are concerned. For example, in jakobsite (MnFe204) the AM
favours a normal configuration (u = 0.385)) the OSSE = 0 both for
and ~ e ~The + .effect of polarization is to partially invert the AMindications,
giving rise to a spinel with i = 0.20. For other interesting aspects concerning
the question of spinels see, for example, Greenwood,[6]~ u r n s , [ ' ~ ]v a n s , [ ~ ~ ]
and ~ t t o n e l l o . [ ~ ~ ]
Olivines are minerals with a general formula M2[Si04] where M repre-
sents various cations, which are essentially bivalent and hexacoordinated.
Belonging to the olivine group are: forsterite Mg2[Si04],fayalite Fe2[Si04],
tephroite Mn2[Si04], monticellite CaMg[SiO,], kirschsteinite CaFe[SiO,],
glaucochroite CaMn[SiO,], knebelite MnFe[Si04]. In particular, olivine is
a solid solution of forsterite and fayalite (Mg,, Fe,-,) [SiO,], while in the
Ca-olivines pairs of cations such as Ca2+ and Mg2+, Ca2+ and Fe2+, Ca2+
and Mn2+ occupy quite distinct sites. Referring to the two minerals
forsterite and fayalite, the composition given above can also be given as
Fo,Fal-,, or simply Fox. Olivine is found in large quantities in the earth's
lower crust, and it is generally supposed to be particularly abundant in the
earth's upper mantle.
The ideal structure of a-olivine can be described as HCP with (according
t Incidentally, the inversion and the complete disorder between Fez' and Fe3+ in octahedral
sites make magnetite an excellent conductor of electricity. This is due to the ease with which
~ e ' +and ~e~~ exchange electrons. Below -153 "C the Fez+ cations become ordered, reducing
noticeably the capacity of magnetite to conduct electricity.
to its stoichiometry) half of its octahedral holes occupied by Fe2+ or Mg2+

and one eighth of its tetrahedral sites occupied by Si4+.The closest packed
layers are parallel to (100) and the octahedral holes are occupied in such a
way that chains of occupied and empty octahedra form zig-zags along the c
axis (Fig. 6.31).
Two layers similar to that shown in Fig. 6.31, at a distance of a12 are
translated by b/2, so that occupied octahedra overlie empty octahedra. In
this way, in fact, as the Pauling's third rule predicts, it is not possible for the
occupied octahedra to share faces. As pressure increases forsterite["I
becomes unstable and transforms itself giving rise first to a spinel type
,,, s,,l, a ,aver of distoned octahedra tyflcal
of the structure of olivine, (Mg, Fe),[SiO I. The
structure (/3 and y), then to an ilmenite type structure and finally, when f d by ~ g "Or Fez+
P > 200 kbar, to two structures: a perovskite type structure, and periclase as
follows: 2Mg2Si04-, 2MgSi03 + 2Mg0.
O n the classification of silicates

First of all, can the structures of silicates be described by means of closest
packings? The answer to this is not always affirmative: let us examine why
this is so. If AEN(Si-0) = 1.72, then eqn (6.2) predicts that between the
silicon and the oxygen interactions will be established, with a percentage of
covalent bonds more or less equal to that of the ionic bonds. As far as the
coordination of the Si4+ is concerned, the covalent bond necessitates the
formation of four hybrid sp3 orbitals, or in other words, the angular
probability of finding an electron is greatest in the direction which connects
the centre of a regular tetrahedron to its vertices (these directions form
angles of 109'28' between each other). On the other hand, the rule of radius
ratio predicts a CN = 4 for the ionic bond. Experimentally Si4+ is nearly
always found to be tetracoordinated, but occasionally hexacoordinated Si4+
has also been observed.
The cations which most frequently occur in the structures of silicates are:
A13+, Fe3+, ~ e ~Mg2+,+ , Mn2+, Ca2+, Nai, K+. Some of these, positioned
around the theoretical limit of two possible coordinations, can form various
polyhedra with the anions: e.g. for ~ 1 p, ~ =' 0.38, therefore CN = 4, 6.
Other cations are of a size to prefer coordinations of 4, 6, 8, or even 12.
It has already been mentioned that the structure of olivine can be
described as HCP, even if a rather distorted one (c, = 0.60). The distortion
is due to the fact that octahedra and tetrahedra share edges. In fact the
distortion depends both on the difference in length between the 0-0
distances of the tetrahedron [SiO4I4- (-2.6A) compared to that of the
octahedron(z2.8 A), and on the electrostatic repulsion.
Other structures with characteristics less similar to those of closest
packings are also possible in silicates. Figure 6.32 illustrates an octahedral
layer (OL) and two diversely orientated tetrahedral layers (TL, and TL,),
which are fitted one into another to form a double layer of spheres (anions),
typical of closest packing. This complex layer will be referred to as OTL.
There are several ways of filling the holes of a close-packed structure. Let
us consider two border-line events. Firstly, both the tetrahedral and
octahedral holes of the layer OTL shown in Fig. 6.32 could be filled, and of
each successive layer. Secondly, the octahedra of one layer (OTL,) and the
Fig. 6.32. A layer of octahedra (OL) and two

diversely orientated layers of tetrahedra (TL,
and TL,), whose polyhedra form the holes
characteristic of a double compact layer of equal
spheres.
tetrahedra of the overlying layer (OTL,) could be filled alternately, in

accordance with the stoichiometric constraints.
The configurations relative to the first hypothesis imply structures of
higher potential energy generally, due either to the more frequent sharing
of geometric elements, or to other factors. For example, if there is not
sufficient oxygen to form isolated [si0,I4- groups the tetrahedra would tend
to condense. The ideal closest packing would require an ~ i - 0 - ~angle
i of
109"28', but, owing to the considerable percentage of ionic bond between
silicon and oxygen, the two Si4+ atoms tend to move away from each other
(thus improving the shielding) forming angles which on average are of 140".
If, however, one of the two Si4+ is substituted by A13+, then ~ 1 - 0 - s i
angles well below 140" (-118") can be formed.
Intermediate figures (e.g. fully filled octahedral and partly filled tetra-
hedral holes) lead to structures with potential energy half way between
those of the two configurations hypothesized above. Bearing in mind what
has been said, the structure of phlogopite, KMg,[AlSi3O,,](OH), will now
be analysed. Supposing that it conformed to closest packing (c, = 0.56) its
stoichiometry would indicate that one quarter of the octahedral holes are
occupied by Mg2+ and one sixth of the tetrahedral holes are occupied by
A13+ or Si4+.In actual fact, of four successive OTL, one is completely filled
by Mg2+ (octahedral holes), two immediately above and below each have
one third of the tetrahedral holes filled, while the last layer contains K+
cations (CN = 12) equal in number to the A13+ which substitute Si4+. To
summarize, in phlogopite a layer of tetrahedra TL, is followed by a layer of
octahedra OL, which is followed by another tetrahedral layer TL,, and
finally there is a layer of K+ ions which connects two tetrahedral layers TL1
and TL,.
Figure 6.33(a) represents one of the two tetrahedral layers (TL,). Every
Fig. 6.33. A layer of tetrahedra is represented in

(a) (TL,). The black tetrahedra are occupied by
si4+,while the white ones are empty. This
determines, as can be seen in (b), the
reorganization of the layer in rings with an OH-
group at the centre, and also, as shown in (c),
the relationship between the ring of tetrahedra
and a group of octahedra of an OL typical of
phlogopite, KM,,[AISi,O,,I(OH),.
tetrahedron shares three vertices with similar tetrahedra, as shown in the

figure, so that every vertex is shared by three tetrahedra. If three of the
contiguous tetrahedra (e.g. 5, 6, 7 of Fig. 6.33(a)) were occupied by cations
with average valence >3, the Pauling's second rule would be infringed. In
+
fact, supposing that they are occupied by either 3Si4+, or 2si4+ AP+, or
+
even Si4+ 2A13+,the shared oxygen still received respectively 3, 2.75, 2.50
v.u., far higher than the necessary value of 2.
Therefore, with respect to closest packing, there are two important
differences. The first is that the group of seven tetrahedra, shown in Fig.
6.33(a), gives rise to a ring of six tetrahedra (Fig. 6.33(b)), leaving an OH-
group in the centre instead of the oxygen which acted as the vertex of the
empty tetrahedron (7 of Fig. 6.33(a)).
This pseudo-hexagonal ring fits well into the overlying octahedral layer
(Fig. 6.33(c)), and allows a ~ i - 0 - ~ angle
i of about 140". The layer of
oxygens which constitutes the base of the ring-forming tetrahedra has
c, =0.76, while the layer of oxygens to which the vertices of the tetrahedra
and the OH- group contribute has c, = 0.88. The latter of these two values
is nearer to the theoretical value, c, = 0.91.
The other difference is represented by the anions around K f . This cation,
unlike the Ca2+ of perovskite, is not surrounded by six anions in the plane
(OOl), which is characterized by K+ cations; instead the twelve coordination
is achieved by six oxygens belonging to the ring of tetrahedra of the
overlying layer (TL,) and six belonging to that of the underlying layer
(TL).
The examples examined above refer to silicates with hexacoordinated
cations which are relatively small, such as A13+, M ~ ' + ,Fez+, Fe3+.
For such cations the edge of the octahedron (-2.8 A) is similar in length
to that of the tetrahedron (-2.6 A), as the structure of olivine demons-
trates, in which they are shared. If larger cations are present, such as Na+
and Ca2+, they adopt coordinations with CN 3 6. If CN > 6, then the holes
made available by closest packing do not fit the requirements of the cation.
If CN = 6, octahedra with edges >3.8 A are formed, edges considerably
longer than the edges of the tetrahedra. In this case a pair of tetrahedra
inscribing a trigonal prism of side 4.1 A is often necessary. Such a prism is
incongruous with the polyhedra of a closest packed OTL.
As a result, in order to describe in rational terms the structures of
silicates, it is necessary to abandon the mechanical model used so far. It is
helpful to make use of particular coordination polyhedra, the type of links
which are established between them, and other parameters.
Liebau's crystallochemical classification

The simplicity of a classification system depends basically on the number of
parameters on which it is based. For the silicates this number was originally
low (Bragg 1930) and has gradually increased as our knowledge of atomic
structures has grown ( ~ o l t a i , [~~i ~e b] a u [ * , ~ ~ ~ ) .
The crystallochemical classification proposed by takes into
account the close relationships which exist between structure, chemical
composition, thermodynamic stability, and reactivity in silicates. Given the
strong prevalence of structures with Si4+ in tetrahedral coordinations over
Si4+ in octahedral coordinations, attention will be given here only to the

former, while Liebau considers both. Since the silicon can be substituted by
other cations such as A13+ and to a lesser extent by p5+,Ge4+,Fe3+, Be2+,
the tetrahedron [ ~ i 0 ~will
] - be
~ represented more generally as [TO4]below.
The first problem to be solved is that of establishing which tetrahedra can
reasonably be considered as part of the silicate anion. Liebau proposes the
following criterion: 'If, under certain thermodynamic conditions, an element
replaces some of the silicon atoms in a given silicate crystal structure so that
the crystallographic positions are statistically occupied by silicon and the
other cation, then this cation is recognized as part of the silicate ion7.
For example, two phases of potassium feldspar, KAlSi308, known as
sanidine hT (high temperature) and microcline 1T (low temperature) are
similar since they have A13+ and Si4+ in tetrahedral coordination, but differ
in that the two cations show different degrees of disorder. This disorder is
maximum in the first phase, K[(A1Si,)08] while it is minimum in the second,
KAl[Si308], since here the tetrahedral sites tend to be completely occupied
by either A13+ or si4+. These two phases are the end members of a series
characterized by different degrees of ~ l ~ + / disorder
~ i ~ +and they are an
example of the phenomenon known in thermodynamics as second-order
phase transition. As the temperature rises the feldspar passes from one
phase to the other as the degree of Al/Si disorder varies, without the first
phase being destroyed (in this case it would be a first-order phase
transition). For this reason, in accordance with Liebau, the formula for fully
ordered microcline 1T is not KAl[Si308], but is similar to that of sanidine
hT, i.e. K[A1Si,08]. It follows that, assuming [A1Si308]is the building unit,
both phases belong to so-called framework silicates.
Noting that Si-0 bonds are generally stronger than M-0 bonds (where
M is a cation other than silicon) the crystallochemical classification of
silicates takes into account the various possible modes of condensation of
the tetrahedra [TO,].
First of all, the number of polyhedra which are linked to a given
polyhedron by vertices, edges, or faces is defined as s (connectedness).
When these polyhedra are tetrahedra [TO,] vertices are shared, and since
each tetrahedron has four vertices, s = 0, 1, 2, 3, 4. If the tetrahedron is
isolated it is called singular (QO),if it shares one vertice it is called primary
(Q1), two secondary (Q2), three tertiary (Q3), and four quaternary (Q4);
concisely QS = QO,Q1, Q2, Q3, Q4.
The parameters on which the crystallochemical classification of silicates is
based are:
1. Nan= number of anion types in a silicate. Nan= 1, uniform-anion
silicates; Nan> 1, mixed-anion silicates.
2. CN = coordination number. In naturally occurring compounds only
tetrahedra [TO,] and exceptionally octahedra [TO6], are present.
3. L = linkedness, type of link. L = 0, 1, 2, 3: for L = 0 the tetrahedron is
isolated, for L = 1, 2, 3 the tetrahedron shares respectively one vertex,
one edge, one face with a similar tetrahedron. In Fig. 6.22 (first row) the
relationships between two tetrahedra for L = 1, 2, 3 are shown.
4. B = branchedness, type of ramification. The tetrahedra [TO,] can be
condensed linearly (see Fig. 6.34), giving rise to limited groups, single
.b Fig. 6.34. Some unbranched fundamental

anions (u6f.a.); (a) a pair of tetrahedra, (b) an
unbranched group of three tetrahedra, (c) a
single unbranched chain, (d) a single
(4 (b) (c) (d) unbranched ring.
chains and rings. They are characterized by tetrahedra with s G 2 and

constitute the so-called unbranched fundamental anions (uBf.a.1.
Complex anions of the same type (uBc.a.) can result from the
condensation of uBf.a. Branched fundamental anions (brf.a.) can also
be obtained by condensing tetrahedra. They include tetrahedra with
l S s 6 4 (Fig. 6.35). By condensing brf.a., brc.a. can be obtained.
Branched anions can be divided into open-branched (oB) or loop-
branched (1B) ones. In the first case, one or more tetrahedra constitute a
sort of 'branch' which is joined to the non-branched anions by means of
a single geometric element (vertex, edge, or face). In the second case,
more than one element is shared, e.g. two vertices (see Fig. 6.35). The
condensation of more than one fundamental anion generates the
following types of anions (see Fig. 6.36):
uB f.a. + o B f.a. = hB c.a. (h =hybrid branched c.a.)
oB f.a. + IB f.a. = olB c.a. (mixed-branched c.a.

+
combination of open loop-branched anions)
In conclusion the branchedness (B) of silicates can be uB or br

(unbranched or branched); while branching can be of various type: oB,
IB, olB, hB.
5. M = multiplicity, i.e. the finite number of tetrahedra that are linearly
linked in a 'short chain' of tetrahedra, or the number of rings, single
Fig. 6.35. Some branched fundamental anions

(6rf.a.). The branches can consist of one or more
tetrahedra (dotted). The figure illustrates: (a) oB
triple tetahedron; (b) and (c) oBsingle rings; (d)
oBsingle chain; (e) IBsingle chain; (f) olBsingle
chain.
Fig. 6.36. Fundamental anions (f.a.)and

complex anions (c.a.1: ( a ) uBf.a., (b) 0Bf.a.. ( c )
1Bf.a.. ( d ) uBc.a., ( e ) 0Bc.a.. (f) IBc.a., (g) hB
c.a., (h) 0Bc.a.. ( i ) IBc.a., (j) hBc.a., (k) 01Bc.a..
(I) h6c.a. (i)
chains, or single layers linked to a complex ring, chain, or layer

respectively; rarely is M 5. The condensation of groups of connected
tetrahedra (i.e. multiple tetrehedra), rings, single chains, or single layers
does not alter the dimensions of the fundamental silicate anions which
are involved in the condensation.
6. D =dimensionality. The silicate anions can have the following exten-
sions: finite (D = 0), infinite in one (D = I), two (D = 2), or three
(D = 3) directions.
When D = 0, the symbols t and r are used respectively to indicate
groups of tetrahedra (terminated anions), and cyclic anions (rings).
7. P or Pr = periodicity of the chain (P) or ring (P'). Silicate anions with
D = 1, 2, 3 are based on fundamental chains of tetrahedra. The
parameter P expresses the number of tetrahedra which determines the
repetition period of an unbranched chain (if a branched chain is
considered the tetrahedra which constitute the branches should be left
out). Figure 6.37 illustrates some fundamental chains of different
periodicity. If the anions are cyclic the symbol Pr, which expresses the
periodicity of a ring (i.e. the number of tetrahedra within a single ring) is
used, excluding, therefore any branches. In order to avoid confusion,
when choosing which fundamental chain is characteristic to a silicate with
D = 1, 2, 3, it is necessary to take into account certain rules:
(a) Fundamental chains with lowest periodicity running parallel to the
Fig. 6.37. (a), (c), and (el illustrate unbranched

chains, while (b), (d),(f), (g), and (h) illustrate
branched chains. The following pairs have the
same periodicity: for (a) and (b) P = 2, for (c) and
(d) P = 3, for (e) and (f) P = 4, for (g) and (h)
P=5.
direction of the shortest identity period within the silicate anion are
chosen.
If rule (a) is satisfied by more than one chain, the fundamental
chains are chosen such that their number is lowest.
Once rules (a) and (b) have been satisfied, the third rule indicates
the following order of preference (for fundamental chains): un-
branched > loop-branched > open-branched > mixed-branched >
hybrid (for > read 'is preferred to').
order in which the various parameters have been presented repre-
sents the classification hierarchy: superclass (N,,), classes (CN), sub-
classes ( L ) , branches ( B ) , orders ( M ) , groups (D), sub-groups (r or t),
families (P, Pr).
If there are a number of silicates with the same type of silicate anion, i.e.
silicates belonging to the same family further subdivisions can be made. The
following criteria are particularly important:
(1) the Si:O atomic ratio of the silicate anions;
(2) the degree of stretching of a chain.
As far as criterion (1) is concerned, it should be noted that two or more
rings, or two or more single chains having P > 1, can join together via all or
only part of their tetrahedra. As a result, complex anions with various Si:O
ratios are generated, depending on the portion of the tetrahedra involved.
As far as criterion (2) is concerned, Liebau proposes that the degree of
stretching should be measured by means of the stretching factor (A), which
can be obtained as follows:
where I, is the identity period of the chain, 1, is the length of the edge of the
tetrahedron, both in A, and P is the periodicity of the chain (the number of
tetrahedra needed to identify the period).
Since amongst all the silicates so far discovered shattuckite,
C U ~ [ S ~ , ~ ~ ] (has
~ Hthe
) , ,most stretched chain, it is taken here as a point of
reference. From it the value 1, = 2.7 A can be obtained (equal to half of the
repetition period of the chain, which is two tetrahedra [Si04]-4). Thus for
shattuckite f, = 5.40 A12 X 2.70 A = 1.00 while for enstatite, Mg2[Si206],
(I, = 5.21 A, P = 2) f, = 0.956 and for alamosite, Pb,,[Si,20,6], (I, =
19.63 A, P = 12) f, = 0.606.
The crystallochemical classification presents the same periodicity charac-
teristics as the periodic system of the elements, so that, for example, by
gradual condensation of [TO4] tetrahedra in a linear way, uB anions can be
generated that are linked only by means of vertices in the following way:
(1) for D = 0 the number of tetrahedra that can be condensed linearly
increases as M increases, so that for M -, w, D i. 1; in which case a
single uB chain will result;
(2) for D = 1 the number of chains increases with M, and for M w,
D + 2, thus forming a single uB layer;
(3) when single layers are condensed and for M -, w, D + 3, a three-
dimensional building of tetrahedra, i.e. uB framework is obtained.
By varying M and D the whole of Table 6.8 can be obtained, which lists
Table 6.8. Chemical and mineralogical (in round brackets) nomenclature of silicates
M=l M=2 M=3 ...
D= 0 Oligosilicates Monosilicates Disilicates Trisilicates . ..
(-1 (Nesosilicates) (Sorosilicates)
D= 0 Cyclosilicates Monocyclosilicates Dicyclosilicates Tricyclosilicates ...

(-1 (Cyclosilicates)
( D=1 Polysilicates ( Monopolysilicates ( Dipolysilicates ( Tripolysilicates I ... ' 1

I (-1 I (Inosilicates)
I
D =2 Phyllosilicates Monophyllosilicates Diphyllosilicates Triphyllosilicates ...
(Phyllosilicates)
D=3 Tectosilicates Tectosilicates

(Tectosilicates) I
the chemical and mineralogical (in round brackets) nomenclature of

silicates.
Structural formulae
A structural formula should contain the largest possible amount of
information and therefore should include the parameters used in crystal-
lochemical classification. Some of the parameters CN and P are best used
only when there is some perplexity: the first is written in square brackets as
a right-handed superscript to the cation, while the second is written without
brackets as left-handed supercrript to the cation. Other parameters, Nanand
L, can be deduced directly from the structural formula. Therefore a
structural formula containing the essential parameters is as follows:
M, indicates cations which do not belong to the silicate anion; the

meaning of the other symbols has already been given. For D = 0 the
interpretation of the symbol is not unequivocal and can therefore be
substituted by r or t, depending on whether the silicate anion is cyclic or
not. For D = 1, 2, 3 the symbols L, 2, 2 indicate clearly that the silicate
anions are respectively chains, layers, and three-dimensional framework of
tetrahedra. Sometimes it is useful to complement the formula by the
following suffixes: (IT), (mT), (hT), (lP), (mP), (hP). They indicate that the
phase described by the formula is of low (l), medium (m), or high (h)
temperature (T) and/or pressure (P).
Some examples:
A @ A ~ [{OB,
~ ] 3t) [Si50,,](OH, F),,04Cl zunyite
The bracketed right-hand superscripts to A13+, indicate the two coordina-
tions of the A13+ in the structure of zunyite.
Since there are no indications for Si4+, it is implicit that CN = 4. The
symbols in curly brackets indicate that the silicate anion consists of a group
of open-branched tetrahedra {oB), three of which determine the multi-
plicity M = 3 (see Fig. 6.3.5(a)). The content of the square brackets indicates
that the group consists of five tetrahedra and that therefore the 'branches' of
the silicate anion are the remaining two tetrahedra [SiO4I4-.
Ca2A12Sn{oB,l r ) [ 4 S i 6 0 1 , ] ( O ~ ) 2 . 2 ~ 2eakerite
0
The structural formula informs us that the silicate anion is open-branched
{oB) and that it is a single ring {lr). The content of the square brackets
indicates that the unbranched part consists of four tetrahedra [SiO4I4- and
that the 'branches' consist of two tetrahedra [SiO4I4- (see Fig. 6.35(b)).
For other structural formulae and for illustrations of the silicate anions
described therein, see Figs. 6.38-6.42.
Relationship between classification parameters and properties

of the cations
The properties of non-tetrahedral cations, in particular, electronegativity
and valence, have a strong influence on the structure of silicate anions.[451
Here, the direct influence of the non-tetrahedral cations on the parameters
which govern crystallochemical classification will be examined.
Fig. 6.38. Ideal configurations of certain silicate

anions and structural formulae: (a) olivine
(Mg, Fe), {US, lt}[SiO,]; (b) ilvaite
CaFe,(Fe, Mn){uB, 2t}[Si,O,IO(OH); (c) benitoite
BaTi{uB, lr)lSi3091; (d) taramellite 1. CN: the coordination number of silicon depends in particular on the
Ba,(Fe3+, Ti),B,{uB, 1r}[Si4O,,I,O,CIX; (e) EN of the anions (X) of the first coordination shell. If the anions are
tourmaline XY Z B {US, 1r} [Si,0,9~0,(~,+0H, F),
where x = ~ a ;af2+; Y = ~ i +Mg , , Fe ,
strongly electronegative, as is fluorine (see Table 6.2), the tendency CN > 4
~ n ' + ,~ e ~ A13+;
', Z = A13+, M ~ ' + ;(f) steacyite is increased (SiF,), if weakly electronegative tetrahedral groups are formed.
K,-,(Na. Ca),-,Th,-,{uB,2r}[ S,O
i ,I the In the case of oxygen, the EN is lower than that of fluorine, but is still
fundamental anion is dotted.
high. Therefore the silicon of both natural and artificial compounds is found
both in tetrahedral or octahedral (a few cases) coordinations. The Si-0
type of bond is influenced by the cations (M) belonging to the second
coordination shell of the silicon: the highly electronegative M cations give a
Fig. 6.39. Single chains with different periodicity more ionic character to the Si-0 bond. This, obviously, affects the CN
(P), structural formulae and stretching factors
(f,) for the following minerals: (a) a hypothetical (CN > 4 is slightly favoured), the effective charge of the silicon (positive),
mineral, M(uB,I L}[s~o,I, f, = 1; (b)enstatite, and, when CN = 4 the overall charge of the tetrahedra (negative).
Mg,{uB, 1 ,}[S1,0,1, f, = 0.965; (c) wollastonite,
ca,{u~, 1 L)[s.,I 0 I,f, = 0.904; (d) krauskopfiie, The effects of repulsion between the silicate anions can be evaluated by
H,Ba,{uB, ,}[Si4O,,I~4H,O, fs=0.783; (e)
rhodonite, (Mn, Ca),{uB, 1 ~}[s,~,o,,], f, = 0.906;
means of the distortions which can be seen above all in the S i - o - ~ i bond
(f) stokesite, Ca,Sn,{uB, 1 ,)[S1,0,,1~4H,0. angles, but also to a lesser extent in other parameters.
f, = 0.718; (g) pyroxmangite,
(Fe,Ca, Mn),{uB, 1 ~}[S~,O,,I, f,= 0.923; (h) 2. L: the stability of a silicate anion decreases as L increases, an inverse
alarnosite, Pb,,{uB, 1 ~}[s~,,o,,], f, = 0.606. The
periodicity (P) is as the silicon stoichiometric
tendency would violate Pauling's third rule. This explains why no silicate
coefficient.
Fig. 6.40. Double chains in which the

fundamental anion is hatched, and the structural
formulae of: (a) sillimanite,
Alie'{u~,2 ~}['(AIs~o,)I(~T);(b) tremolite,
Mg,Ca,{uB, 2 ~ } 1 2 ~ i 4 ~ l , 1 2 ( (c) ~ ~xonotlite,
)2;
Ca,{uB, 2 l}13si 0 (d) narsarsukite,
Na,Ti,{uB, 2 ~ } ? ~ / ~ h , ~ ] 0(d)
, ; can also be
obtained by condensating the unbranched rings
{US, lr}[Si,O,,l (dotted in figure) and the latter
formula has to-be changed into Na,Ti,{uB, mr)
I~S~,O,~IO,.
anions with CN = 4 are found with L > 1 (at present only one case has been
found with L = 2). Highly electsonegative cations favour a higher degree of
linkedness because the [SiOJ effective charge tends to be reduced.
3. B: branched silicate anions, in particular branched ring anions, are less
stable than unbranched varieties, due Do the shwter average Si-Si distances.
If, however, cations with high EN are present, then the stability of such
silicate anions is increased, since they attenuate the effective negative
charge thus reducing the repulsion between the tetrahedra. The stabilizing
effect of such cations decreases as D increases, because in such cases the
effective charge of the tetrahedra [TO,] ako decreases.
4. M: the number of structures with increasing M falls drastically both in
the oligosilicates and in the cyclosilicates, and in polysilicates and phyllosili-
cates. This is due to the decided increase in the potential energy of
structures as M increases. To examine this phenomenon the differences in
energy between linear groups of tetrahedra will be analysed. It should be
borne in mind that the greater their distance from the tetrahedron which
terminates the group, the less tmo contiguous tetrahedra will vary from the
point of view of energy. For example two Q1 tetrahedra are similar from the
point of view of energy because they each start and terminate a group. Two
Q1 tetrahedra and one Q2, belonging to a group of three tetrahedra are very
different from the point of view of energy, because the distance that
separates Q2 from the external tetrahedron Q1 is only that of one
tetrahedron. Finally, in a single unbranched chain every tetrahedron has
practically the same energy as that next to it, since both are more or less at
the same distance from the end tetrahedron. In contrast a linear group of Fig. 6.41. Single and double layers (the
fundamental chain is hatched) and the structural
tetrahedra could, from an energetic point of view, be regarded as being formulae of: (a) muscovite,
constituted by different structural units. KAI,{uB, 1 ~ ) [ 2 ( ~ ~ ~ i , ) ~ l , and
l ( ~ talc,
~),
To this the principle of parsimony is applied (see p. 436, Pauling's fifth ,
MgAuB, 1 _.l2si 0 NOH),; (b) apophyllite,
KCa,{uB, 1 , } ~ ~ i , 0 , , l & ~ ~ 0 ~ ) ~ 8(c)
~~0;
rule), according to which the smaller the number of structural units a hexacelsian, Ba{uB, 2 ,}I (AISi)O,],(hT).
structure is composed of, the more stable it is. This explains the uniform
number reduction of the groups as M increases.
What has been said above regarding linear groups can be extended to
chains or layers, by considering the whole chain or the whole layer as a
structural unit, instead of a tetrahedron.
5 . D: the general rule is that where there is a fixed Si-0 ratio, the silicate
anions tend to join together, in accordance with D+max. Though there
are exceptions to this rule, it can be justified by the fact that condensation
of the [siO4I4- tetrahedron results in a better local electrostatic valence
balance, in accordance with Pauling7ssecond rule.
6: t, r: multiple tetrahedron silicate anions are more stable than cyclic
ones. In fact, since the average Si4+-Si4+ distances are greater for the
former, the force of repulsion between Si4+ and Si4+is less.
7: P: the periodicity of a chain can be evaluated, in general, by means of
the stretching factor f,. The higher this factor, the more the chain is
stretched, and the lower is P. There are, however, important exceptions
where P increases with f, (pyroxenoids and pyroxenes).
(c)
Fig. 6.41. (Continued) 8. f,: this factor depends, in particular, on EN and on the average valence
( v ) of the cations. Strongly electropositive cations tend to increase the
negative charge of the [TO4] groups; therefore the repulsive forces which
act between these groups are greater and as a result f, + 1. Vice versa,
strongly electronegative cations, by reducing the charge of the [TO4]
groups, reduce f, and so P tends to increase.
An increase in the ( v ) of the cations has a similar effect. In fact, when
( v ) is high, a greater number of oxygens of the chain must be involved in
cation-oxygen bonds. As a consequence f, is reduced and P increases.
The ionic radius of the cation has more effect on the distortion of the
polyhedra than on the periodicity P. The greater the difference between the
silicon radius and that of the other cations, the greater this effect will be.
9. Nan: bearing in mind still the principle of parsimony, it can be stated
that the number of the structural units, distinguished either chemically or
geometrically, must be as small as possible. This explains why the vast
majority of silicates have Nan= 1; only a few structures have Nan= 2 (e.g.
okenite, Ca,,{uB, 2 ~)[3Si6016]{~B, 1 ~)[3Si6015]2.18H20) and there are no
known crystal structures at present with Nan> 2; however in glasses, melts,
etc. Nan> 2 is common. Silicates with mixed anions are favoured by cations
with high EN.
Fig. 6.42. Three-dimensional buildings of [TO,]
tetrahedra, and the structural formulae of: (a) 10. s: for a given Si:O ratio of the silicate anion, high EN cations favour
trid mite, {US, ~}[2~i,041; (b) cristobalite, {uB,
J'
:}[ Si,04] (note the thickened fundamental
a high s value.
chain and how in cristobalite the layers
perpendicular to [ l l l ] are all orientated the
same way, while in tridymite the same layers
perpendicular to [00011are turned through 180" Appendices
compared to one another); (c) orthoclase,
K{IB ~ } [ 3 ( ~ ~ ~ i 3 ) ~The
, l ( structure
m ~ ) . of
orthoclase is represented schematically. The 6.A Application of the concept of the packing coefficient
letters U and D indicate the position of the [TO,]
tetrahedra with one vertex pointing either up (U) (ci)
or down (D); (d) the fundamental chain IS, on
which the structure of orthoclase is based (the In general, calculated packing coefficients almost never match the expected
'branch' consists of the dotted tetrahedra). value (0.74). It is sometimes found that structures with lower c: ( ~ 0 . 6 0 )
conform to closest packing models whereas others with ci ( ~ 0 . 7 0 )depart

from them. The main reason is that in fact ions are not hard spheres but
charged, elastic, and polarizable entities. In addition the atoms are
sometimes too bulky to be included in the octahedral holes of a closest
packing scheme; in such cases the structure would not have the greatest
economy of space, i.e. it would not be the closest-packed structure. Thus
the ideal conditions demanded by the mechanical atomic model are in
practice never achieved.
In spite of this, the packing coefficient provides useful information,
mostly for structures with a high degree of ionic bonding. Assuming, for
oxygen, re = 1.40 A, packing coefficients are calculated for some compounds
belonging to several classes of minerals with different ionic bond-character.
The abbreviation c.p. shows those compounds which adopt a close-packed
arrangement of anions, while 6 represents density.
Oxides
Periclase [MgO]
Rutile [TiOz]
P-Tridymite [SiO,]
Coesite [Si02]
Stishovite [SiO,]
Brucite [Mg(OH)2]
Gibbsite [A1(OH)3]
Ilmenite [FeTi03]
Spinel [MgA120,]
Carbonates
Aragonite [CaCO,] ci= 0.61 (6 = 2.93)
Calcite [CaC03] ci= 0.56, c.p. (6 = 2.71)
Alstonite [BaCa(CO,),] ci= 0.51
Ewaldite [Ba3Ca2(C03)5] ci= 0.55
Shortite [Na2Caz(C03)3] ci= 0.53
Tychite [NazMgz(C03)4(S04)] ci= 0.55
Gaylussite [Na2Ca(H20)5(C03)z] ci= 0.51 (a
Fig. 6.42. (Continued)
Borates
Kotoite [Mg3(B03),] ci= 0.67, C.P.
Tincalconite [NazB405(OH)4.3H20] ci= 0.55
Kernite [NazB406(OH)2~3H20] ci= 0.53
Borax [Na2B405(OH)4.8Hz0] ci= 0.53
Sinhalite [MgAl(BO,)] ci= 0.76, c.p.
Aksaite [Mg(B304(0H)2)2.3H20] ci= 0.44
Gowerite [Ca(B304(OH)z)z~3H20] ci= 0.42
Sulphates
Baryte [BaSO,] ci= 0.53
Chlorothionite [K2Cu(S04)C12] ci= 0.50
Fibroferrite [Fe(OH)S04.5H20] ci= 0.53
Parabutlerite [Fe(OH)S04-2H20] ci= 0.60
Hohmannite [Fe(Hz0)4[(S04)20]~4HzO]
ci= 0.58
Coquimbite [Fez(S04)]-9H20 ci= 0.55
Phosphates
Triphylite [ ~ i ( ~ e ' Mn2+)P04]
+,
Heterosite [(Mn, Fe)P04]
Hydroxylapatite [Ca,(OH)(PO,),]
Brushite [CaHP04.2H20]
Moraesite [Be2(OH)P04.4H20]
Struvite [NH4MgP04.6H20]
Silicates
Forsterite [Mg2Si04]
Larsenite [PbZnSiO,]
Monticellite [CaMgSiO,]
Humite [Mg7(0H,F)&304),]
Zircon [ZrSiO,]
Grossular [Ca3Al2Si3OI2]
Enstatite [Mg2Si206]
Anthophyllite [Mg7Six022(OH)2]
Pyrophyllite [A12Si40,,(OH)2]
Orthoclase [KAISi,Ox]
Some remarks concerning the relationships between the anion arrange-
ment and the packing coefficient are helpful.
The compound CaCO, crystallizes in two structures (polymorphs): calcite
(c, = 0.56) and aragonite (c, = 0.61). Calcite can be described by means of a
close-packed oxygen arrangement with & triangular and 4 octahedral sites
filled respectively by C4+ and Ca2+. This packing is not the closest possible
for CaC0, because Ca2+is surrounded by six oxygens, i.e. in this structure,
calcium shows the lowest CN among those adopted (see Table 6.7). If
pressure increases, a more compact structure (aragonite in which Ca2+ has
CN = 9) originates; this structure no longer conforms td the closest packing
model.
The compound SiO, gives rise to several phases. Some of them are
considered here, they are: (1) P-tridymite, stable between 870 and 1470 "C;
(2) coesite, roughly stable from 30 to 100 kbar; and (3) stishovite, a very
high-pressure phase, stable above 100 kbar. P-tridymite structure (c, = 0.50,
6 = 2.22) is an infinite three-dimensional framework of [SiO4I4- tetrahedra.
It can be sliced into sheets of tetrahedra like muscovite (Fig. 6.41(a)) but
arranged in such way that the tetrahedra vertices alternate up and down.
The structure of coesite (c, = 0.67, 6 = 2.91) is quite dense and somewhat
more complex compared with that of P-tridymite. The main difference
between them concerns the second and following Si4+coordinations that, of
course, are greater for coesite. This explains the different packing
coefficients, though none of them (see p. 429 and following) conforms to the
closest packing. Stishovite has an exceptionally high packing coefficient
(c, = 0.99) and a density (6 = 4.29) much greater than that of /3-tridymite or
coesite. The structure is rutile type, TiO,, (Fig. 6.26, (c)), so it conforms to
the closest packing model. The very high pressure under which stishovite
crystallizes forces Si4+ to renounce its habitual four for an unusual six
coordination.
According to the mechanical closest packing model it is impossible to
have c, = 1, unless we suppose that the available space of the structure is
almost completely filled by oxygens having an approximate volume of a ball

with re = 1.40 A but geometrical form by no means spherical.
The compound grossular, Ca3Al2Si3OI2,belongs to an important group of
minerals known as garnets. In spite of the high packing coefficient
(ci = 0.66), grossular does not have a close packed structure (see Appendix
6.B). Here we note that Ca2+ cannot replace 0'- as occurs in perovskite.
Let us consider the structure of perovskite (Fig. 6.28) and its chemical
formula in the form Ca4Ti4 hIz4Ol2 (El represents the empty tetrahedral
site).
Figures 6.19 and 6.28 show that if grossular were to adopt a perovskite-
type structure in half of the cells shown in Fig. 6.28, si4+ in tetrahedral
holes would be in contact with Ca2+. After all, according to the examples
quoted, the packing coefficient is trustworthy only in a few cases, i.e. for
those structures having suitable cations and a high degree of ionic bonding
(as happens in several oxides). Its value can give little information on the
real atomic packing, which can sometimes be enhanced only by further
crystallochemical parameters (see Appendix 6.B).
6.B Structural inferences from crystallochemical

parameters
The effective ionic radius re, the packing coefficient (ci) and the valence
electrostatic rule (V = 2,) can sometimes be profitably used to foresee some
important structural features particularly when simple crystal structures are
considered. First of all one should extract all the possible information from
the chemical formula. On p. 437 we have seen that for the compounds of
MX type, the coordination numbers of M and X can vary provided that the
ratio 1:l is preserved; for instance in NaC1, M:X= 6:6 and in CsBr
M:X = 8:8. Again, in the MX2 compounds the coordination number of M
and X can change, the ratio M:X = 2:l being equal. So, salts like S O 2 , Ti02
and CaF2 have the ratio M:X = 2:l and respectively the following coordina-
tion numbers: 4:2, 6:3 and 8:4.
From these remarks the following rule can be derived: in a given
structure, if no bond between identical atoms forms, M and X coordination
numbers of compounds M,Xp, are in the ratio p:m.
This rule can also be applied to more complex structures like A,B,Xp.
First let us define the average cation coordination number ( m , ) :
where CNA and CNB are respectively the coordination numbers of cations A
and B.
The relationship between m, and the average anion coordination
number ( m , ) is expressed by
--
CN,/CN, = p l(m + n). 64.2)
Replacing A . l in A.2, the latter can be written as
Some examples are given, in order of increasing structural complication,

to understand the relationships between chemical formula, crystallochemical
parameters, and symmetry, showing how they can sometimes have predic-
tive value.
At first, we will assume tentatively that the ionic structures having
ci> 0.60 can be described in terms of a close-packed anion arrangement
with cations filling the holes, even if in some cases this hypothesis does not
hold (as some of the following examples will prove).
1. Periclase, MgO, Fm3m, and Z = 4, has a packing coefficient ci= 0.62.
According to the accredited Mg2+ CN, 4, 5, 6, 8 (see Table 6.7) this cation
can fill the tetrahedral (half) or the octahedral site of a structure based on a
close-packing model. Moreover the space group Fm3m and the number of
formula units Z = 4 informs us that Mg2+ and 02-lie at 4,i, 4 and 0 , 0 , 0
or vice versa, from which m, = CN, follows. Using (A.3) in the more
simple form mCN, =pCN, we reach an obvious conclusion, i.e. Mg2+ and
02-have the same CN, (4 or 6) and of course for both coordinations the
electrostatic-valence principle is satisfied. To solve the CN dilemma, we
observe that periclase is isotype with lime, CaO, and that accredited CN, for
Ca2+ are 6, 7, 8, 9, 10, 12 (Table 6.7). Only 6 and 8 belong both to Mg2+
and Ca2+ CN, sets, consequently only the octahedral sites of a closed-
packed anion arrangement can be filled.
2. Rutile, Ti02, P4,/mnm, and Z = 2, has a packing coefficient c, = 0.74.
The accredited CN, of Ti4+ are 4, 5, 6, 8 (Table 6.7). Therefore, according
to the close packing and valence requirements, Ti4+ can fill tetrahedral or
octahedral sites. The octahedral site stabilization energy (OSSE = 0) cannot
remove the uncertainty regarding the site, even if the high electrical charge
informs us that Ti4+ prefers the octahedral site. El. can be observed that,
according to the space group, Ti4+can be placed at 0, 0, 4 or 0, 0, 0 and that
02-can lie on 2/m or 4 or mm (from 4e to 4g according to International
Tables of X-ray Crystallography Vol. A) in this case m, = CN,. No
profitable information can be drawn from isotype structures, but we know
that Nb4+ and ~ a can ~ replace
+ Ti4+ in very high percentage ( ~ 0 . 4 0 )The
.
common CN for these two cations, i.e. 6, indicates that very probably, Ti4+
prefers the octahedral site.
From (A.3) the calculated oxygen coordination number is m, = 3, i.e.
three Ti4+ surround one 02-whereas, as shown above, Ti4+ seems linked to
six oxygens.
3. Ilmenite, FeTiO,, ~ 3 and, Z = 2, has a packing coefficient c, = 0.66.
According to the space group, two ~ e and ~ two
+ Ti4+ can lie on a three-fold
axis or one of them on a three-fold axis and the other on the two
independent 3 (la, lb). As for the oxygen atoms, they can be situated on
the two independent i (3e, 3d) or in the general position 1 (6f); in both
cases (i or 1) CN, must be an integer. The accredited CN, for Fe2+ and ~ i ~ +
are respectively 4, 6, 8 and 4, 5, 6, 8 (Table 6.7), so either Fe2+ or ~ i may
~ +
fill tetrahedral and/or octahedral sites. From (A.3) three different m, can
be calculated for ilmenite:
- -
1.4 + 1-4= 3CNx;
- CN,
- = 813
1.4 + 1-6= 3CNx;
- -
-
CN, = 1013
1.6 + 1.6 = 3CNx;CN, 4.
Only the third result agrees with the statement that m, must be an
integer and precisely it tells us that four cations are linked to each oxygen
without specifying their nature. Applying the electrostatic-valence rule to
+ +
the three possible combinations 1Fe2+ 3Ti4+, 2Fe2+ 2 ~ i ~ 3Fe2+ +, +
lTi4+,we found respectively 2.33, 2, 1.67 V.U.(valence units). So 2Fe2+ and
2Ti4+ are the cations coordinated by one oxygen atom, in agreement with
the stoichiometry.
4. Perovskite, CaTiO,, ~ m j m ,and Z = 1, has a packing coefficient
c, = 0.62. The space group and Z inform us that Ca2+ and Ti4+ are in m3m
(la, lb) and that oxygen lies on 4/mmm (3c or 3d), i.e. the number of
cations arotlnd it is an integer and, what is more, even. The accredited CN,
for Ca2+ are 6, 7, 8, 9, 10, 12 and for ~ i they
~ ' are 4, 5, 6, 8. So according
to c, and CN, Ca2+ and Ti4+ should fill, like ilmenite, octahedral sites. This
eventuality would require that 5 of the octahedral si;es be filled by Cd2+and
Ti4+ and that consequently several edges (three are in the sheet) be shared
between Ca2+ and Ti4+ octahedra. Now we observe that for CN = 6,
different from ilmenite (re = 0.61 (Fe2' low spin) and re = 0.605 A (Ti4+),
the Ca2+ and Ti4+ effective ionic radii are quite different: re = 1.00 (Ca2+)
and re = 0.605 A (Ti4+).This strong difference is reflected on the octahedral
edges that, if shared, would involve a strong strain in the structure and
consequently the increase of its potential energy. Moreover for CN = 12,
the Ca2+ effective ionic radius (re = 1.34 A) matches well with the 02-one,
a
so we have a special close packing form in which of oxygens are replaced
by Ca2+. For Ti4+ in tetrahedral or octahedral sites and Ca2' 12-
coordinated, (A.3) gives respectively m, = or m, = 6, supporting the
hypothesis that Ti4+ is six-coordinated like the oxygen. As far as the nature
of the cations around the oxygen is concerned, the valence sum principle
indicates that around the oxygen atom there are two Ti4+ and four Ca2+.
5. Zircon, ZrSi04, I4,/amd, and Z = 4, has a packing coefficient c, =
0.70. In accordance with the space group and the number of formula unit Z,
zirconium and silicium are located at 42m (4a and 4b). In theory, oxygen
can be placed at 2/m (8c and 8d) or at 2mm (8e x 2), but it can easily be
shown that starting from the unit cell parameters, a = 6.60 and c = 5.98 A,
the calculated interatomic distances zr4+-0;- and Si4+-02- do not match
well with those obtained from the effective ionic radii (Table 6.7). For
instance, supposing the cation (M4+), i.e. zr4+ or Si4+, at 0, 0, 0 (4,) and
02-at 0, $, (8c), then the M4+-02- calculated distance is 1.81 A # 1.66
(Si4+-0'-) and 2.12 (zr4+-02-). On the other hand the situation does not
improve if 02-is located at 0, 0, z because the calculated Zr4+-0 and
Si4+-0 distance sums are: 2.99 A (5.9812 A) # 3.78 A (1.66 + 2.12 A).
These kinds of argument support the hypothesis that 0'- is situated on 2 or
m (16f or 16g or 16h), so m, = CN,.
The packing coefficient would suggest that zircon adopts a close-packing
structure. However, if Si4+ cations are located at the habitual tetrahedral
holes, m, assumes integer values (A.3) when zr4' fill tetrahedral sites or
when it is 8- or 12-coordinated. Let us now consider, besides ZrSiO,, the
zircon isotype structures: coffinite, USi04; thorite, ThSi04; Hafnon,
HfSiO,. The accredited CN, (Table 6.7) of the bigger cations are: 4, 5, 6, 7,
8, 9 (zr4+);6, 7, 8, 9, 12 (u4+);6, 8, 9, 10, 11, 12 (Th4+);4, 6, 7, 8 (HP+).
They show that some CN, (6 and 8) are present in all the four cations. This
peculiarity together with the results of (A.3) suggest that a packing,
different from the close packing, in which Zr4+ is 8-coordinated, charac-

terizes the crystal structure of zircon. In addition we note that the
electrostatic valence balance is not compatible with a structure having Zr4+
in octahedral sites. In fact, since the oxygen atom is shared by Si4+ and
Zr4+, the bond strengths coming from Si4+ (+1 v.u.) and from zr4+ (+4/6
or +8/6v.u.) do not balance 02-valence. By means of (A.3), we can
conclude that around 02-there are three cations, the nature of which can
be specified by means of the electrostatic-valence rule: two zr4+ and one
Si4+.
6. Spinel, MgA1,04, ~ d g m Z , = 8 has a packing coefficient ci= 0.70. In
the unit cell there are 8Mg2+, 16A13+,and 320'-. Magnesium lie on 43m
(8a or 8b), aluminium on 3m (16c or 16d), and consequently oxygen is on
3m (32e), so m, = CN, is an integer. If the civalue and the accredited CN,
for Mg2+ (4, 5, 6, 8) and for A13+ (4, 5, 6) are taken into account, then the
spinel structure would seem to conform to a close-packing one with
tetrahedral and/or octahedral sites partiallyfilled. On the other hand, (A.3)
leads to three possible configurations for CN, integer: (1) Mg2+ and ~ l in ~ +
tetrahedral; (2) Mg2+ in tetrahedral and A13+ in octahedral sites; and (3)
half ~ l in~tetrahedral
+ and half in octahedral together with Mg2+. The
calculated m, of oxygen is three for the first and four for the other two
configurations. As concerns other factors influencing crystal structures, the
effective ionic radius seems to favour A13+ in tetrahedral and Mg2+ in
octahedral sites, whereas the formal charge indicates that cations follow the
opposite behaviour pattern. The quoted ambiguity allows us to understand
why direct and inverse spinel structure exists, as is better explained in a
more general way at p. 443.
Finally, for cations located in tetrahedral sites, both the chemical formula
and the valence balance require two ~1~~ and one Mg2+ around 0'-
whereas when Mg2+ and ~ l are~ respectively
+ in tetrahedral and octahedral
holes the chemical formula indicates that oxygen is surrounded by one Mg2+
and three ~ f + .
7. Pyrope, Mg3Al2Si3O1,, 1a3d, Z = 8, has packing coefficient 0.73. The
number of formula units Z and the space group suggest that A13+ lies on 32
or 3 (16b or 16a) and that Mg2+ and Si4+are on 222 and 4 or vice versa. As
regards the oxygen atoms these can be located on 3 (32e), 2 (48f and 48g),
or 1 (96h) axes. The shortest inter-ionic distances between cations, lying on
222 (24c) and 4 (24d), and oxygen lying on 3 (32e) or on 2 axes (48f and 48g)
have the following values: 3.09 and 2.865 A, far from those inferred from
Table 6.7, i.e. Mg2+-0 = 2.29 A or Si4+-0 = 1.66 A. Besides symmetry,
valence balance and chemical formula requirements would seem to exclude
the location of an oxygen atom on the quoted special positions, so the most
reliable point symmetry for oxygen atom seems 1. The CN, used for Mg2+in
the latter analysis were limited to 6 and 8 because pyrope is an isotype with
grossular, Ca3A12Si3012 and the common CN, between Mg2+ and Ca2+ are 6
and 8.
The modified form of equation (A.3) applicable to pyrope is
It can be easy verified that only when Ca2+ has CN = 8, m, become a

whole number, i.e. m, = 4.
Finally by means of the chemical formula and the application of the

electrostatic-valence rule it is possible to characterize the oxygen coordina-
tion, i.e. two Mg2+, one A13+, and one Si4+.
Generally speaking it is difficult to imagine making a correct inference on
the atomic organization of a complex structure without having some
preliminary information or without any assumption concerning it. For
instance, coquimbite has (Fe3+,A1),(S0,),.9H20, ~ 3 1 c , 2 = 4, V =
1760.5 A3 and ci= 0.55. If the usual CN, 6 and 4 for this kind of structures
are respectively assumed for Fe3+ and S6+ and if it is supposed that Hf
forms no bifurcated hydrogen bonds in coquimbite, then the CN, of oxygen
+ +
can b e inferred. In fact using (A.4) we have 2.6 3-4 2.18 = 21CN,, from
which m, = 2.86 can b e estimated. The structure informs us that: O(1) and
O(2) are shared by one ~ e (or ~ A13+)
+ and one S6+; O(3) and O(4) are
shared by one S6+ and two H + ; 0,(1) and 0,(2) are shared by one ~ e (or~ +
A13+) and two H + ; 0,(3) is shared by four HC. This configuration leads to
the quoted m, = 2.86.
References
1. Cartemel, E. and Fowles, G. W. A. (1966). Valency and molecular structure.
Butterworth, London.
2. Huheey, J. E. (1983). Inorganic chemistry. Principles of structure and reactivity,
(3rd edn). Harper, New York.
3. Vainshtein, B. K., Fridkin, V. M., and Indenbom, V. L. (1982). Modern
crystallography II. Springer. Berlin.
4. Pauling, L. (1959). The nature of the chemical bond. Cornell University Press,
Ithaca.
5. Hamilton, W. C. and Ibers, J. A. (1968). Hydrogen bonding in solids.
Benjamin, New York.
6. Greenwood, N. N. (1970). Ionic-crystals, lattice defects and non-stoichioimetry.
Butterworths, London.
7. Catti, M. (1978). Acta Crystallographica, A34 974.
8. Cotton, F. A. and Wilkinson, G. (1980). Advanced inorganic chemistry, (4th
edn). Wiley, New York.
9. Orgel, L. E. (1960). A n introduction to transition-metal chemistry ligand field
theory. John Wiley, New York.
10. Basolo, F. and Johnson, R. (1964). Coordination chemistry. Benjamin, New
York.
11. Ballhausen, C. J. (1962). Introduction to ligand field theory. McGraw-Hill, New
York.
12. Burns, R. G. (1970). Mineralogical applications of crystal field theory.
13. Busing, W. R. (1981). W M I N , computer program to model molecules and
crystals in terms of potential energy functions, U.S. National Technical Informa-
tion Service, ORNL-5747.
14. Catlow, C. R. A. and Cormack, A. N. (1984). Acta Crystallographica, B40, 195.
15. Alberti, A. and Vezzalini, G. (1978). Zeitschrift fur Kristallographie, 147, 167.
16. Brown, G. E. and Fenn, P. M. (1979). Physics and Chemistry of Minerals, 4, 83.
17. Parker, S. C., Catlow, C. R., and Cormack, A. N. (1984). Acta
18. Shannon, R. D. and Prewitt, C. T. (1969). Acta Crystallographica, B25, 925.
19. Shannon, R. D. (1976). Acta Crystallographica, A32, 751.
20. Fajans, K. (1923). Naturwissenschaften, 11, 165.
21. Magini, M. (1979). Journal of Chemical Physics, 70, 317.

22. Redinger, J. and Schwartz, K. (1981). Zeitschrift fur Physik, B40, 269.
23. Phillips, J. C. (ed.) (1974). Treatise o n solid state chemistry, Vol. 1. Plenum,
New York.
24. Pearson, W. B. (ed.) (1972). Crystal chemistry and physics of metals and alloys.
Wiley, New York.
25. Basso, R. (1985). Neues Jahrbuch fur Mineralogie. 3, 108.
26. Hawthorne, F. C. (1981). Journal of Solid State Chemistry, 37, 157.
27. Lamgley, R. H. and Sturgeon, G. D. (1979). Journal of Solid State Chemistry,
34, 79.
28. Novak, G. A. and Gibbs, G. V. (1971). American Mineralogist, 56, 791.
29. Moore, P. B. and Smith, J. V. (1970). Physics of the Earth and Planetary
Interiors, 3, 166.
30. International tables for x-ray crystallography, Vol. 11, (1959). Birmingham:
Kynock Press.
31. Iglesias, J. E. (1981). Zeitschrift fur Kristallographie, 155, 121.
32. Basso, R. (1985). Neues Jahrbuch fur Mineralogie, 3, 108.
33. Brown, I. D. and Shannon, R. D. (1973). Acta Crystallographica, A29, 266.
34. Brown, I. D. and Wu, K. K. (1976). Acta Crystallographica, B32, 1957.
35. Brown, I. D. (1978). Chemical Society Reviews, Vol. 7, No. 3. The Chemical
Society, London.
36. Scordari, F. and Stasi, F. (1990). Neues Jahrbuch fur Mineralogie, 161, 241.
37. Siisse, P. and Tilmann, B. (1987). Zeitschrift fur Kristallographie, 179, 323.
38. Evans, R. C. (1976). A n introduction to crystal chemistry (2nd edn). Cambridge
University Press.
39. Ottonello, G. (1986). Physics and Chemistry of Minerals, W, 79.
40. Baumgartner, O . , Preisinger, A , , Heger, G., and Guth, H. (1981). Acta
Crystallographica, A37, C187.
41. Yamanaka, T. and Takeuchi, Y. (1983). Zeitschrift fur Kristallographie, 165, 65.
42. Liu, L. G. (1976). Physics of the Earth and Planetary Interiors, 11, 289.
43. Zoltai, T. (1960). American Mineralogist, 45, 960.
44. Ribbe, P. H. (1982). Orthosilicates, (2nd edn), Reviews in Mineralogy No. 5.
Mineralogical Society of America, Washington.
45. Liebau, F. (1985). Structural chemistry of silicates. Springer, Berlin.
Molecules and molecular
crystals
GASTONE GILL1
Chemistry and X-ray crystallography

Crystal and molecular structure
X-ray diffraction from crystals was discovered at the beginning of this
century and it soon became evident that it was suited to the investigation of
the structure of matter at atomic resolution level. Diffraction applications of
X-rays and, later on, of other radiations of comparable wave length, such as
electrons or thermal neutrons, developed rapidly allowing the scientific
world to obtain, for the first time, a detailed knowledge of the intimate
atomic distribution in elements, organic and inorganic substances, natural
and synthetic fibres, and even liquids. It became clear, however, over the
years, that the level of image resolution achievable was strongly dependent
on the aggregation state of matter. Currently we know that a resolution of a
few thousandths of an angstrom is actually obtainable only by diffraction
experiments carried out on relatively perfect single crystals, both organic
and inorganic, while macromolecular crystals of proteins (haemoglobin,
myoglobin, and enzymes) cannot give an accuracy better than, say, 0.5 A
and fibres or liquids produce poorly resolved images, which can be
considered barely giving much more than a hint of the true molecular
structure.
The experimental resolution achievable is to be compared with the
natural resolution needed by classical chemistry. For example, carbon-
carbon distances are typically 1.54 for single bond, 1.39 for aromatic
compounds, and 1.33 and 1.20 A for double and triple bonds, respectively;
thus a total range of 0.034 A covers the fantastic change of properties and
reactivities between alkynes and alkanes and the much smaller range of
0.006 A separates aromatic from alkene chemistry. Taking into account that
two bond distances cannot be reasonably compared at less than four e.s.d.
(estimated standard deviations), we must conclude that no valuable
chemical information is contained in C-C distances known with an e.s.d.
greater than 0.01 A and, therefore, in crystal structures where the e.s.d.s on
atomic positional coordinates are greater than about 0.007 A. This value
sets a clear limit on diffraction experiments. Only elements, minerals, and
small molecules easily develop into long-range ordered three-dimensional
arrays of perfect or ideally imperfect crystals, and only diffraction experi-
ments on such crystals can provide independent experimental evidence of
direct relevance to chemistry. Conversely, synthetic or natural macro-
466 1 Gastone Gilli
molecules, which crystallize as highly defective hydrated monocrystals in the

most favourable case of globular proteins, but often as fibrous materials
having simple one-dimensional order, cannot give much more than the
overall shape of the diffracting molecule and their study is to be considered
more a branch of high-resolution microscopy than a part of the structural
chemistry in a strict sense.
The expression structural crystallography is often used for identifying
the branch of sciences producing direct knowledge of the structure of matter
by diffraction methods. Within structural crystallography there are divisions
which have become traditional: crystal chemistry of metals and alloys,
minerals, molecular and macromolecular crystals are usually taught in
different courses, and their scientific achievements published in different
journals. This is a kind of functional partitioning, different parts of
crystallography being dealt with according to their usefulness to other
disciplines.
There are, however, more scientifically founded reasons for a subdivision,
which originate from the differences that the knowledge of the crystal
structure may have for different classes of chemical substances. This
knowledge is of primary value when the atomic binding forces within the
crystal can be classified as true chemical bonds, as is the case for covalent,
metallic, or ionic crystals, where the nature of the bond can be understood
only on the basis of the knowledge of the crystal itself. For instance Na,, G,
and Si, molecules are well known in the gas phase but their properties
cannot explain why metallic sodium, diamond or graphite, and silicon may
be electronic conductors or insulators or semiconductors. In the same way
the reason for the stability of ionic substances is found in the ordered
arrangement of alternating anions and cations within crystals where any
simple relationship between chemical formula and crystal structure is
completely lost. A good example, taken from wells,[ll may be that of the
series of AX iodides: HI, AuI, CuI, NaI, and CsI; only the first term gives
biatomic molecules at room temperature, while the others are solids where
A is surrounded, in order, by 2, 4, 6, and 8 atoms of X.
The other extreme is represented by the class of molecular crystals where
chemical substances, typically organic and organometallic compounds, exist
as discrete molecular entities packed together by weak dispersion or dipolar
forces. Interatomic distances within the crystal can be generally grouped in
two clearly distinct distributions, a first of bond distances within the
molecule and a second one of the much longer contact or van der Waals
distances among atoms of different molecules. Non-bonded atoms on
different molecules tend to behave much like non-interacting hard spheres
that cannot approach at a distance shorter than the sum of their en-
cumbrance radii, called van der Waals radii, rvdW,and reported in Table
7.1. If the crystal is fused or vapourized by heating or dissolved by a solvent
Table 7.1. Atomic van der Waals radii
(A). H in aromatic rings 1.00 A; C perpen- the crystalline order is destroyed, but the geometry and pattern of
dicularly to the benzene ring 1.77 A. Data interatomic connections within the molecules, apart from possible partial
according to ~ o n d i [ ~ l rotations around single bonds, remain unchanged.
H 1.20
In this case the crystal structure (intended as crystal packing) is less
C1.70 N1.55 01.52 F1.47 important, the chemical information being mainly concentrated in the
Si 2.10 P 1.80 S 1.80 CI 1.75 molecular structure which, in principle, could be determined in any other
As 1.85 Se 1.go Br 1.85
1 1.98
state of aggregation of matter. In fact, methods for accurate determination
of structure in liquids or in solutions are not actually available and gas phase
Molecules and molecular crystals 1 467
methods (microwave spectroscopy and gas electron diffraction) allow us to

study only small and preferably symmetrical molecules, so that crystal
structure determination often remains the only practicable way in which to
obtain detailed information on the molecular structure of chemical com-
pounds. Taking into account the importance of the knowledge of the
molecular structure of both synthetic and natural substances, it is not
surprising that X-ray structure determination of molecular crystals has so far
been used as a simple tool for determining molecular configurations and
conformations and that crystal packing has mostly been seen either as an
irrelevant factor or as an unwanted cause of perturbation of the free-
molecule geometry. Only recently has the idea that crystal packing could be
important in itself in the study of molecular interactions been widely, if not
generally, accepted.
The growth of structural information

X-ray crystallography of molecular crystals has been so successful in
chemistry that the last three decades have registered an almost exponential
growth of the number of structures published per year. Structural data had
to be collected in computerized databases, and the Cambridge Structural
Database (csD),[~]as at April 1990, contained bibliographic, chemical, and
numerical results for some 84 464 organic and organometallic compounds.
Figure 7.1 show a histogram of the new entries in the CSD in the period
1965-86.13] The number of complete structural analyses published up to
1960 was only 579 (45 from 1935-45, 212 from 1945-55, and 322 from
1955-60) and this number is now exceeded tenfold by the structures
determined per year.
The factors which have determined such a rapid evolution in the field are
scientific (development of direct methods theory), technological (improve-
ment in automatic single-crystal diffractometers), and organizational (dis-
tribution of complete systems of crystallographic computer programs). All
these factors must also be considered with the parallel evolution of fast
electronic computers, without which X-ray crystallography would have
remained that slow and patient skilled work well known to the scientific
world until the end of the 1950s.
The effects produced in chemistry by this extraordinary wealth of
structural information are enormous, and for a vivid analysis of this
Fig. 7.1. Number of new crystal structures

registered each ear in the Cambridge Structural
Database (CSD)81 from 1965 t o 1986. Structures
containing d- or f-block metals are indicated by
1965 1970 1975 1980 1985 shading. (Reproduced by permission from
Year Orpen eta/.[ ''I)
468 1 Gastone Gilli
phenomenon the reader is referred to the second part of the Dunitz's book
X-ray analysis and the structure of organic molecules.[41The attempts aimed
to interpret, rationalize, and understand this continuously growing mass of
data give rise to problems far beyond the limits of a single chapter, being
dealt with by the scientific discipline of structural chemistry, initiated more
~ ] his famous book The nature of the
than 50 years ago by Linus ~ a u l i n g ' in
chemical bond and the structure of molecules and crystals. A n introduction to
modern structural chemistry.
The present chapter is intended to summarize the structural chemistry of
molecular crystals. As regards its content, it has been necessary to be
severely selective of topics. The discussion is strictly confined to the nature
of molecular crystals and to the stereochemical aspects of molecules which
are directly derived from the results of structural investigation. Final
applications, such as structure-property studies in chemistry and molecular
biology or pharmacology have been completely excluded. The topics have
been divided in two parts, according to whether they refer to single
molecules or molecular crystals as a whole and can be so summarized:
(1) molecular crystals: molecular interactions, their nature and effect on the
crystal packing; elements of crystal thermodynamics and polymorphism;
(2) single molecules: classical stereochemistry; molecular geometry and
chemical bond; molecular mechanics; interpretation of molecular struc-
tures and the structure correlation method.
The nature o f molecular crystals

Generalities
Molecular crystals are ordered packings of discrete physical entities, i.e.
molecules. Most of them consist of neutral molecules, and it is those which
are mainly discussed here, though some may sometimes contain ions (e.g.
organic carboxylates, alkylammonium halides) as well. The energy of
formation of the crystal from the isolated (gas phase) molecules is called
lattice energy, U; typical values of -U for molecular crystals are 0.5-
4 kcal mol-' for noble gases, 10-20 kcal mol-' for most neutral organic
compounds and up to 40-50 kcal mol-' for neutral molecules of relevant
complexity. By comparison with some typical - U values for ionic (e.g. 188
in NaC1, 629 in CaF,, 3804 in A1203 and 3137 kcal mol-' in Si0,) or
covalent crystals (some 171 and 283 kcal mol-' for diamond and Sic,
respectively) it can be seen that the forces determining the packing in
molecular crystals are much weaker than those of ionic or covalent chemical
bonds. suchweaker intermolecular forces can be approximately classified as
follows.
Repulsion or exchange forces

Most stable molecules have closed electron shells, that is all their molecular
orbitals are doubly occupied and cannot accept other electrons without
violating the Pauli principle. Accordingly, they repel other molecules trying
to come too close. Repulsive forces exert their effects at very short range,
decrease with the ninth to twelfth power of the interatomic distance and
grow quite rapidly only when two non-bonded atoms try to get closer than
C-H
_---
1 .
C----,.---
\-.'
Fig. 7.2. Curves of non-bonded energy as a
function of internuclear distance, E,,(r)
for the C-C, C-H, and H-H interactions
versus r,
calculated by using the atom-atom potentials

3 4 k given by ~ i g l i o [ '(full
~ ] curves) and ~ l l i n ~ e r [ ' ~ '
r (A) (broken curves).
their proper contact distance, given by the sum of their van der Waals radii
(Table 7.1).
Dispersion or London f o r c e ~ [ ~ l l ]
These are weak short-range attractive forces which decrease with the sixth
power of the intermolecular distance and are caused by the mutual
attraction of the small transient dipoles that molecules can induce in each
other. They give the greatest contribution to the lattice energy in crystals of
neutral molecules even if the energy of any single interaction is very small.
Dispersion energy between two molecules is usually expressed as the sum of
the crossed interactions between all pairs of atoms on the two molecules and
the sum of repulsion and dispersion energies is named non-bonded energy,
En,. When plotted versus the interatomic distance, r, the non-bonded
energy displays a typical minimum at a distance nearly equal to the sum of
van der Waals radii of the two atoms (Fig. 7.2). Such a minimum is quite
shallow, being of the order of a few tenths of a kcal mol-l.
Dipolar forces
Molecules having permanent dipole moments experience electrostatic at-
traction when properly oriented (orientation forces). According to
~ i t a i ~ o r o d s k i ,such
[ ~ I attraction forces must cancel out in crystals having
only translational symmetry operations (space group PI); for other space
groups it has been estimated that the dipole-dipole interactions may
contribute one tenth of the lattice energy for molecules having dipole
moment of 3-4 D, while their contribution becomes negligible when the
dipole moment is less than 1D.
Monopolar forces
Monopoles are associated with ions, a situation which is not very common
in molecular crystals and is not discussed here in detail. Two main points,
however, deserve particular attention: ionic interactions are known to cause
a quite relevant increase in lattice energies as can be shown, for example, by
the comparison of these energies for acetic acid and sodium acetate,
respectively 17.4 and 182 kcal mol-l; monopole-monopole interactions are
long-range forces whose energy decrease only with the first power of the
470 1 Gastone Gilli
interionic distance (Coulomb law) and in this respect differ from all other
short- or very-short-range intermolecular forces.
Hydrogen bonding[l"lsl
With the exception of monopolar forces, H-bonds are the highest-energy
interactions in molecular crystals. They greatly affect the way in which
molecules are packed, in the sense that the observed packing is almost
inevitably that allowing the maximum number of such bonds to be made.
Moreover, H-bonding is, by itself, the most relevant non-bonded interaction
in nature, being the main factor determining the structure of water, the
folding of proteins, and the pairing of bases in DNA. For this reason most
crystal packing studies are essentially attempts to understand the laws
governing the intermolecular H-bonding in an easily reproducible ex-
perimental environment, that given by the molecular crystal.
H-bonding occurs when a hydrogen atom is bonded to two (or sometimes
more) other atoms. This situation may be depicted schematically as
D-H--A, where D is the H-bonding donor and A the acceptor. In
principle all atoms more electronegative then hydrogen (C, N, 0 , F, S, C1,
Se, Br, I) can play the role of A and D , though stronger hydrogen bonds
are necessarily associated with the most electronegative ones (N, 0 , F, Cl).
Several theoretical studies have been devoted to clarifying the nature of
the bond in H-bonding complexes and in particular the relative contribu-
tions of different terms to its total energy. Probably the most popular and
quoted partitioning scheme is that developed by Umeyama and
~ o r o k u m a [ for
l ~ ~the treatment of (H20), and (HF), dimers. It makes use
of the energy decomposition analysis developed within the ab initio
SCF-MO theory[203211 where the total H-bonding energy is partitioned in
four terms: the repulsion or exchange energy and the electrostatic,
polarization, and charge transfer attraction energies. The authors were able
to conclude that the main attractive term is electrostatic and that the
contribution of charge transfer is small, so that H-bonding can be
qualitatively defined as an electrostatic more than charge transfer or simply
electrostatic interaction.
H-bonds can be classified (Fig. 7.3) according to their topology as
intramolecular, intermolecular, or bifurcated and, according to their
energy, as going from weak to very strong.
1. Weak H-bonding can be observed for any couple of donor and
acceptor atoms whenever the two groups cannot achieve the correct
approach for some sterical reason. The main factor is usually the D-H--A
angle which, for maximizing the electrostatic interaction between the D-H
dipole and the negatively charged acceptor, must be in the range of some
160-180". A good example comes from the intramolecular H-bonds closing
five-, six-, or seven-membered rings; the H-bond closing a five-membered
ring is always weak (and so weak that the hydrogen of its D-H group forms,
whenever possible, a second bifurcated hydrogen bond with another
acceptor (Fig. 7.3(g1)). A second reason why a H-bond is to be classified as
weak comes from the small intrinsic electronegativities of the H-bonded
partners, the most classical case being that of the C-H--A interactions.[221
2. Medium H-bonding is typical of water, alcohols, amines, amides, and
carboxylic acids. Its geometry is rather well defined: the 0-H--0 group
---$ \
R-C
//O---H-0
\OAH
'C-R
- - -04
Fig. 7.3. Different types of H-bonds.

Intermolecular H-bonds forming three-
dimensional nets in water (a), dimers in
carboxylic acids (b), chains in hydrofluoric acid
(c), and bicyclic dimers between guanine and
cytosine in DNA Id), Intramolecular H-bonds
closing six-membered rings (e) are stronger
than those closing five-membered rings (f).Two
examples of intermolecular bifurcated H-bonds
(g, g'). Charge assisted H-bonds (CAHB) (h) and
resonance assisted H-bonds (RAHB) (i) are
stronger than the usual H-bonds between
neutral groups. An example of a very strong
H-bond, the difluoride ion (j).
tends to be linear; the D-H distance is not significantly lengthened with

respect to that observed in the absence of H-bond; the D---A contact
distance is practically identical to the sum of the van der Waals radii of A
and D, that is the van der Waals radius of the interleaving H atom is almost
zero. A wealth of thermodynamical data['*] shows that its formation
enthalpy is in the range 2-8 kcalmol-'. In more detail, it is some
4.8 kcal moly' for the water dimer and 4.7, 5.4, 5.8, 8.6 kcal mol-I,
respectively, for the H-bonds formed by carbonyl, amide, ether, and amine
acceptors with the phenol as donor.[151
3. Medium-strong H-bonding can be thought to derive from medium
H-bonding strengthened by the presence of positive and/or negative
charges. A possible range of energies is 6-20 kcal mol-'. It seems
reasonable to distinguish two different cases: (a) charge assisted and (b)
resonance assisted H-bonding.
(a) Charge assisted H-bonding (CAHB) arises from net ionic charges on
the donor and/or the acceptor groups. The effect of charge in
strengthening the H-bond is documented by the data collected[231
on the
H-bonds formed between the intrachain or terminal groups of oligo-
peptides and aminoacids: the average N---0 contact distance is
2.840, 2.908, 2.912, and 2.929 A, respectively, for the H-bonds
-COOe---H3N+-, ) C=O---H,N+-, -COO----H-N ( and
) C=O---H-N (.
472 1 Gastone Gilli
(b) Resonance assisted H-bonding (RAHB) is similar to CAHB but with

the difference that the strengthening charges arise from polar resonance
form^;[^^,^'] the only system extensively studied is that of P-diketone
enols where the resonance O=C-C=C-OH(---)-0-C=C-C==OH+
strengthens the intramolecular 0-H---0 bond in such a way that the
0---0 distance can become 2.417 A and the calculated H-bond energy
12.7-19.7 kcal mol-' (values to be compared with those for the water
dimer: d,-, = 2.74 A , -AH = 4.8 kcal mol-'). In these compounds an
appreciable lengthening of the 0 - H bond with consequent shift of the
proton towards the central interoxygen position is also observed.
4. Very strong H-bonding appears to be inevitably associated with ions.

It is characterized by relevant lengthening of the D-H distance, nearly
central position of the proton, D---A distances definitely shorter than
the sum of van der Waals radii, and bond energies intermediate between
true H-bonding and covalent bonding. Complexes like [F--H--F]- or
[H,O--H--H,O]+ have H-bond enthalpies of 150-250 and 130-
150 kcal mol-I and must be considered true chemical species where the
charge transfer (covalent bonding) is relevant and we may speak of
H-bonding only in a formal sense.
Charge transfer[2G291
Intermolecular charge transfer or donor-acceptor interactions occur be-
tween electron donors (Lewis' bases) and acceptors (Lewis' acids). They
establish an at least partially covalent bond between highly polarizable
groups, which is often described as the formation of a molecular orbital by
electron donation from the highest occupied molecular orbital (HOMO) of
the donor to the lowest unoccupied molecular orbital (LUMO) of the
acceptor. Classical examples are the molecular complexes NH, BF3 = +
+
H3N-BF,, I, I- = [I-I-I]-, or the molecular crystal of iodine, where the
I, molecule is both an acceptor along the interatomic axis and a donor
perpendicular to it. Such a type of interaction occurs in crystals only in the
presence of large and easily polarizable (soft) atoms. Another type of
donor-acceptor interaction which has been actively studied in the recent
past is that present in metallic or semiconducting organic crystals, a class of
mixed crystals which contain planar donor and acceptor molecules packed in
separated (segregated) infinite stacks and have given rise to great interest
for their potential applications in electronics. For a recent review on their
structural aspects see references.[301
Intermolecular forces and crystal packing

We may wonder whether a generic knowledge of intermolecular forces
can give us some general insight into the way in which molecular crystals are
packed. It is not difficult to show that the relative molecular positions are
mainly determined by short-range forces. Calling r the interatomic distance,
the repulsion energy drops nearly with r-12, the attractive dispersion energy
tends to zero with r-6, and the attractive electrostatic energy decreases with
r-'. These facts have important implications, because the electrostatic
energy changes so slowly within the lattice that it is practically unable to
displace the molecules from their sharp minima, already determined by the
short-range balance of repulsion and van der Waals interactions of the outer
atoms of the molecule, so that these latter remain the true controlling factor
of the packing arrangement. The only exception to this rule is represented
by the hydrogen bond, which is electrostatic in nature but requires a quite
specific geometry in the donor-acceptor D-H--A interaction and, more-
over, involves energies much greater than dispersion interactions do.
The relevant role played by short-range van der Waals forces in the
crystal packing helps us to understand the apparently paradoxical fact that
the large majority of molecular crystals belong to a few space groups having
second-order symmetry elements. Since van der Waals forces are both
adirectional and additive, the energy of any single atom is lower the higher
the number of atoms of other molecules surrounding it at contact distances
is or, in other words, the most stable crystal is that in which molecules pack
themselves with the highest coordination number. A simple geometrical
analysis of the problem has been carried out by ~ i t a i ~ o r o d s kwho
~ [ ~has
'
shown that rows of molecules staggered by a glide lattice operation can
produce a very efficient close packing (coordination of twelve) by repeating
a molecule of arbitrary form in the space groups PI, P2,, P2,/c, Pca2,)
Pna2, and P2,2,2, or a centrosymmetric ~iloleculein the space groups PI,
P2,/c, C2/c or Pbca. These space groups are actually those most frequently
observed for molecular crystals.
A more detailed analysis of intermolecular forces
Dispersion energy
The first theoretical treatment of dipersion forces is due to ond don[^] (so
that they are often called London forces), who used the perturbation theory
to obtain the following simplified equation for the dispersion energy E(r) of
two molecules whose centres are separated by the distance r
E(r) = -(3/2)a;a;1,1,r-~/(1, + I,) (7. la)
where a; and a; are the molecular polarizability volumes and I, and I2their
first ionization potentials. For identical molecules it becomes
E(r) = -(3/4)a'21r-6 (7. lb)
showing that the attraction energy increases for molecules having high
polarizability and ionization potential values and is of the general form
- ~ r - ~where
, c is a constant mainly determined by the value of a'.E(r) is
the only attractive energy among neutral molecules without permanent
dipole moment and determines all phase transition temperatures of the
substance. Polarizabilities are known to increase with the molecular volume
and with the number of n bonds (particularly extended systems of
conjugated bonds) and are essentially a measure of how much electron
clouds can be displaced from their equilibrium positions around the nuclei;
higher polarizabilities imply stronger intermolecular attractions and this is
the reason why, at room temperature, larger molecules give crystals, smaller
ones give liquids, and only very small ones are gaseous. It may be said that
dispersion forces (or more basically polarizabilities) are the true reason why
molecular crystals can exist.
474 1 Gastone Gilli
Units for dipole moments and polarizabilities

Magnitudes of dipole moments are generally quoted in debyes (D), where
1D = 3.336 x I O - ~ O C ~A. dipole consists of two charges +q and -q
separated by a distance R is calculated as p = qR and is directed from the
negative to the positive charge. Taking as unit of charge the electron
(le = 1.602 x 10-19C) and of length the angstrom (1 A = 10-'Om) or the
bohr (1 bohr = 0.529177 A = a" = radius of the first hydrogen orbit) the
following equivalences hold: l e A = 4.800 D and l e Bohr = 2.542 D.
A moderate electric field E induces in a molecule a dipole moment p
according to p = a E , where a is called molecular polarizability; p and E
being vectors, a is a second rank tensor but usually only its scalar value
a = (Ciai)1'2/3 is given as polarizability volume a' = a/4ne0, whose units
are simply A3 = lopz4cm3. Typical values of a' are 0.82, 2.63, 1.48, and
10.5 .A3 for Hz, C12, H 2 0 , and CCl,, respectively.
Repulsion (or exchange) energy and total non-bonded energy

The repulsion forces two molecules experience when their outer atoms
come into contact are of quantum nature and derive from the Pauli
principle; they register the impossibility for closed-shell molecules to accept
other electrons in their doubly occupied molecular orbitals. Unfortunately
no theoretical treatment for the calculations of repulsion energies is known,
so that they must be evaluated empirically and this is done according to two
different approximations. In the first, the repulsion energy is expressed as
a exp (-br), where a and b are constants to be determined; the total
non-bonded energy assumes the so-called 6:exp or Buckingham's form["]
Enb(r)= a exp (-br) - ~ r - ~ . (7.2)
In the second approximation the repulsion energy is of the type ar-"
where a is a constant to be determined and n an integer in the range 8-15
(usually 12); in such a case the non-bonded energy is in the so-called 6:n or
Lennard-Jones' form[lol
Atom-atom potentials
When the interacting bodies are atoms (e.g. noble gases) the distance r can
be taken as the internuclear distance. In the case of molecules some
complications arise and the atom-atom approximation is most commonly
used: the total interaction energy between two molecules is expressed as the
sum of those among the constituent atoms. The two-atom potentials in the
generalized form
Enb(r)= A exp (-Br)r-D - Cr-6 (7.4)
are often named atom-atom potentials. Values for the A, B, C, and D
parameters are not derived from theory but determined by experiment, such
as gas deviations from ideality, compressibility of liquids and solids, and
neutron scattering by liquids. Otherwise, the parameters minimizing the
difference between observed and calculated values of some physical quantity
(typically lattice energies and cell parameters for crystals; bond distances,
bond angles, and torsion angles for molecules) are used.
Equation (7.4) requires four constants for any pair of atoms, which can be
a problem when the number of different atoms increases. For this reason
other forms of (7.4) are used, such as
which corresponds to (7.4) with B = 0, D = 12, A = &*r*12,and C = 2 ~ * r * ~

or
E,,(r) = ~ * [ kexp
, (-k,r/r*) - k , ( ~ * l r ) ~ ] (7.5b)
which is identical to (7.4) by putting D = 0, A = kle*, B = k,lr*, and
C = ~ * k , r *and
~ where k,, k,, and k, are simple rescaling constants. For
any atom i only two values are tabulated: its van der Waals radius rl and an Enb
energy parameter E,, which are used to calculate the parameters of any i-j
interaction as r* = rl + r, and E" = (E,E,)'/~.The quantities r* and E* assume
the meaning of position and depth of the potential well of the curve En,
versus r (Fig. 7.4). This formula greatly simplifies data tabulation because it
requires only two values per atom instead of three or four for each pair of E* ---
atoms.
Several tabulations of atom-atom potentials are available and an exhaus- Fig. 7.4. Meaning of the parameters eI and rl in
tive literature is reported in the recent book by Pertsin and ~itai~orodski.['] eqns (7.5a, b).
Generally speaking, potentials are not transferable from one application to
another, being more suited to fit phenomena they are derived from. Tables
7.2 and 7.3 report typical potentials used for crystal packing studies[12]and
for molecular mechanics and conformational analysis calculations.[131As an
example, Fig. 7.2 compares the curves calculated for the C-C, C-H, and
H-H interactions by means of the two different potentials; although the
general shape and the position of the minima are similar, the agreement is
rather poor. This is typical of semi-empirical quantities since they do not
derive from a unique theoretical fundamental but are just a set of
parameters which only need to be internally consistent in such a way as to
reproduce a given physical quantity; what really matters is that they work,
which they usually do if used within their own range of applicability.
Electrostatic energies
The study of interactions among permanent molecular multipoles is
simplified by the fact that the energy involved rapidly decreases with the
order of the multipole itself and it is generally admitted that the role played
by quadrupoles is negligible. Even considering only the first two terms of
multipolar expansion the number of terms remains relevant, that is
monopole with monopole, dipole, and induced dipole together with dipole
with dipole and induced dipole. However, a great simplification can be
reached in some cases because interactions with induced dipoles can be
usually neglected in lattice energy calculations and crystals of neutral
molecules (practically the only case so far studied) do not need monopoles
to be taken into account.
A point of general interest concerns the value of the electric permittivity
E (normally expressed as the product of the dielectric constant E, and of
the vacuum permittivity E,, so that E = E,EJ which is included in all of the
following equations. In interactions decreasing rapidly (e.g, with rP6 in the
dispersive or dipole-dipole interactions) E, is assumed to be unitary as in a
vacuum because the space around the atom considered can be taken as
empty within the short distance (usually 15 A) to which the calculations are
476 1 Gastone Gilli
Table 7.2. Atom-atom potential parameters ac-

cording to ~ i ~ l i o to
[ ' ~be' used in crystal pack-
ing calculations. Units of kcal mol-'. Q repre-
sents CI or S and Me the methyl group. Poten-
tials are of the form given in eqn (7.4)
Pair A B x ~ o -C~ D
Table 7.3. Atom-atom potential parameters

according to ~ l l i n g e r [ ' ~for
' molecular
mechanics calculations in kcal mol-'.
Potentials are of the form given in eqn
(7.5b) with k, = 8.28 x lo5, k2= 13.59, and
k3 = 2.25
Atom r* E*
extended. When monopoles are involved the energy decreases only with the
first or second power of distance; calculations must be extended to a wide
range and the value of E, to be used becomes a complex and barely known
function of the distance in consequence of the shielding effects produced
around the ion by the interleaving atoms.
1. Monopole-monopole interactions occur in crystals containing ions.
The interaction energy of ions i and j having formal charges qi and qj located
at a distance rij is
where E = E,EO is the electric permittivity. The constant (1/4neO)depends on

the units used and is 332.145 kcal mol-' A e-2 or 1389.70 kJ mol-' A e-'; it
has a value of one when atomic units are used, that is energies are in
hartrees (1 hartree = 627.7 kcal mol-I) and distances in bohr (1 bohr =
0.529177 A).
2. Monopole-dipole interaction energy is calculated according to
where r is the distance between the charge q and the central point of the
dipole p, and @ the angle between r and the direction of the dipole.
3. The case of dipole-dipole interactions is of great interest as most
neutral molecules have small dipole moments associated with chemical
bonds between atoms of different electronegativities, whose vector sum
produces, unless it does not vanish because of symmetry, the overall
molecular dipole moment. The interaction energy is calculated according to
two main models.
In the dipolar model the bond dipole moments are tabulated for all bonds
of interest or directly calculated from the known values of atom
electronegativities. The interaction energy between two dipoles pi and pj
whose centres are separated by the vector q is
and the total dipole-dipole energy, ED,, in the crystal is one half of the
sum of all the intermolecular terms A slightly different way could be
that of summing up all the small bond dipoles into the overall molecular
dipole moment and of computing the total energy over all molecules.
In the monopolar model the total energy is calculated as the sum of all
atomic partial charges qi according to
where atoms i and j are on different molecules. Partial charges can be

obtained from standard semi-empirical or ab initio quantum mechanical
calculations, approximate semiempirical m ~ d e l s [ ~ or,
~ ~ more
~ ' ] recently, as
deformation densities[331experimentally determined by X-ray diffraction.
4. Hydrogen bonding has already been treated and only potential fields
for the calculation of its energy remain to be discussed. Since the H-bond is
essentially ionic in nature, the total interaction energy of the fragment
478 1 Gastone Gilli
D-H--A (D = donor, A = acceptor) can be divided in two terms: the usual

non-bonded energy, Enb, and an electrostatic (or dipolar) term that takes
into account the partial charges on the donor and acceptor groups. Starting
from the empirical observation that the D . . . A contact distance is
practically equal to the sum of the van der Waals radii of D and A as though
the H atom does not exist, it seems clear that any potential assuming a van
der Waals radius of zero for the hydrogen (see eqn (7.5)) and localizing
suitable partial charges (or dipoles) on D-H and A will describe reasonably
well the H-bonding energy, the preference for a linear D-H--A arrange-
ment included, as this optimizes both the van der Waals repulsion and
electrostatic attraction between A and D. This is essentially the strategy
used by the potential energy fields more recently proposed for describing
the bond,[^*^^] though other approaches are also
Thermodynamics of molecular crystals

It is of relevant interest to obtain even an approximate evaluation of the
different terms contributing to the total free energy of a molecular crystal.
The discussion essentially follows the lines of the treatment given by
~ i t a i ~ o r o d s k iand
[ ~ ] is restricted to the simple case of a crystal constituted
by rigid molecules at atmospheric pressure. These conditions allow some
simplifications. As regards the effect of pressure, the Gibbs free energy
function G = H - TS differs from the Helmholtz function F = U - TS for
the term pV = G - F = H - U, which is the work done by the system
against the external pressure. As the coefficient of thermal expansion is very
small in crystals, the expansion work is irrelevant at atmospheric pressure
and we can put G = F. Moreover, if the molecule is conformationally rigid,
intermolecular forces are unable to modify the internal molecular geometry
and the following partitioning of the free energy is acceptable
where F,, and Fmo, are the inter- and intramolecular contributions to the
total crystal free energy. In this approximation Fmo, could be separately
calculated at any temperature from the frequencies of the internal vibra-
tional modes of the rigid molecule by the usual methods of statistical
thermodynamics; the same methods allow us to calculate another molecular
quantity which will be shown to be necessary, the contribution of internal
vibrational modes to the molar heat capacity, Cmo,.
Within our approximations the molar free energy of the crystal can be
written as
U is the lattice energy or total potential energy of the intermolecular

interactions. Even if the interatomic forces are temperature independent, U
is temperature dependent as a consequence of the slight increase of the cell
parameters with temperature, so that it must be written U(T) = Uo+
AU(T), where Uo is the lattice energy at absolute zero (a term large and
negative) and AU a small positive term increasing with the temperature and
equal to zero for T = 0, representing the expansion work against the
internal cohesion forces. It causes a difference between the molar heat
capacities at constant pressure and volume and from this difference can be
actually calculated as A U(T) = J": (C, - C,) dT.
KOis the zero-point energy, that is the crystal vibrational energy at zero
kelvin; it is known that it may have some relevance only for molecules of
extremely small moment of inertia (N2, 0 2 , CO) and strongly directional
bonds (H20). In common crystals it is very small (<0.2 kcal mol-') and can
be neglected. It cannot be dissociated from the lattice energy and the global
+
term U KOcan easily be obtained from the experimental molar subEima-
tion enthalpy at the temperature T according to
where 2RT is a small correction term (some 1.2 kcalmol-' at room

temperature) due to the differences of molar heat capacities between solid
(6R) and gas (3R) and to the expansion work of a mole of gas (pV = R T for
an ideal gas). Of particular interest is the molar sublimation enthalpy
extrapolated to absolute zero, which has the meaning of lattice energy at
this temperature, that is
E,,, is the vibrational part of the internal energy. At not too low
temperatures an oscillator gives a contribution of kT to the internal energy
(equipartition principle). The Avogadro number N of molecules having six
degrees of freedom (three rotational and three translational) accumulate, at
the temperature T in gaseous phase, a kinetic energy of 6NkT/2 = 3RT; in
the solid state translations and rotations are hindered and become oscilla-
tions and librations around the equilibrium positions which, having both
kinetic and potential components, can accumulate an energy E,,, = 6NkT =
6RT and the crystal should have a molar heat capacity C, = 6R. In practice
C, tends to zero for T tending to zero because an ever decreasing number
of vibrational levels is accessible and E,,, = $:CV d T becomes increasingly
smaller than 6R.
Svibis the vibrational entropy. Clearly lim Svib= SOfor any crystal of a
T-0
pure substance without static disorder, where So is the residual entropy at
absolute zero. So, a quantity strictly related to KO, is very small and it
is assumed to be zero when the entropies are measured according to
the third law of thermodynamics. This can be calculated as Svib= Scr=
SF (Cp - CmoJ d(ln T).
The thermodynamic equations needed to evaluate the different terms in
(7.11) and an actual calculation concerning naphthalene at two different
temperatures are summarized in Table 7.4. The experimental data necessary
are not easily found, being the molar heat capacity at constant pressure, C,,
the isobaric thermal expansivity, a = (l/V)(6V/6T),, the isothermal com-
pressibility, b = -(l/V)(6V/6p),, and the volume of the unit cell, V) from
zero to the temperature of interest. Two other quantities are needed, Uo,
obtained from the molar sublimation enthalpy according to (7.12), and the
contribution of internal vibration modes to the molar heat capacity, Cmo,,as
a function of T, which has already been shown to be obtainable without
great difficulty from spectroscopic data in the case that there is no mixing of
internal modes (intramolecular vibrations) and external modes (vibrations
and librations of the molecules as a rigid body). Figure 7.5 reports the plot
480 1 Gastone Gilli
Table 7.4. Quantitative evaluation of the free energy of the crystal of

naphthalene at two different temperatures according to ~itaigorodski['' (the
zero-point energy KO has been neglected)
Isobaric thermal expansivity Isothermal compressibility

a = ( 1 1 V)(6V/6T), B = -(1 IV)(6 V/6p),
Calculated values (cal mol-')
Quantity Way of calculus T=140K T=300K
uo lim AH""^)
T-0
-16700 -16700
Fig. 7.5. Plot of the molar vibrational free

energy, F,,,, and of its components as a function
of the temperature for the crystal of
naphthalene. Data from ~ i t a i ~ o r o d s k i . ' ~ ]
of the calculated values of molar vibrational free energy for the crystal of
naphthalene over a much wider range of temperatures.
From these data it is possible to obtain some rules regarding crystals in
general and molecular crystals in particular. The lattice energy constitutes
by far the greatest part of crystal free energy, F,,; at low temperature and
even at higher temperatures it is the prevailing part. The general effect of
the vibrational part is that of causing a continuous decrease of F,, because
ITSvibl increases more rapidly than E,,, with temperature. In other words
the crystal is increasingly stabilized by higher temperatures; this, of course,
is not accidental but the expression of a fundamental physical law as
differentiation of the free energy expression dG = V dp - S d T implies that
(6G/6T), = - S and that, S being always positive, the free energy decreases
with increasing T. A last point can be that the work done against the
internal cohesion forces of the crystal, AU, is small up to room temperature
and can be neglected in a first approximation.
Free and lattice energy of a crystal from atom-atom

potentials
Knowing all the individual intermolecular energy terms, the lattice energy
of a crystal built up of rigid molecules can be calculated according to
where the first term is the non-bonded energy (eqn (7.4) or (7.5)), the
second the electrostatic energy (7.9) and the last one the H-bonding energy.
The index i runs over all atoms of a reference molecule and the index j over
those of all the surrounding molecules. The summation can be truncated at
some 15 A in computing En, because dispersive forces decay rapidly with

distance. This is not true for electrostatic forces and special mathematical
techniques have to be used to increase series convergence.
In the crystal a rigid molecule has six degrees of freedom, three
translational and three rotational (or librational); if x, y, and z are the
coordinates of its centre of gravity and 8, QI and 3 the Eulerian angles
describing its orientation (Fig. 7.6), U is a function of these six variables and
of the cell parameters, ai (i = 1, 3), that is
The number of parameters for the unit cell goes from one for the cubic
system to six for the triclinic system. Moreover, the molecule may have
fewer degrees of freedom as it is in a special position; for example it has
only three rotational degrees of freedom if located on a symmetry centre.
So, the lattice energy of naphthalene (P21/a, Z = 2) depends on only seven Fig. 7.6. Eulerian angles 0, q, and yt defining the
orientation of the (x', y', 2') orthogonal base of
degrees of freedom (a, b, c, P, 8, QI,3 ) . the rigid molecule with respect to the
If n is the total number of parameters, it is useful to think of U as a orthogonal base of the crystal, (a, b, c ) .The
+
hypersurface in an (n 1)-dimensional space where the geometrical para- broken line OA is the intersection of planes
( x ' y ' ) and (ab).
meters are the abscissae and the energy the ordinate: the different minima
on this surface correspond to all possible crystal structures in the space
group chosen. The relevant number of calculations done on different
crystals indicate that the experimental structure usually corresponds to the
deepest minimum or, at least, to one of the deepest minima of the potential
surface and that the calculated and experimental values of U compare
within a few kcal mol-l. This seems to indicate that the structure can be
predicted from the simple evaluation of lattice energy, independently of the
vibrational part of the free energy.
The reasons for this fact will be discussed in the next section. What is
important here is that it allows us to obtain the best potential energy
parameters (globally called a force field) to be used in (7.14) by a
least-squares procedure on a minimum number of known crystal structures
where the quantities to be reproduced are the lattice energy, the unit cell
parameters, and the positional parameters of the molecule. This method has
been applied to several classes of chemical compounds and is reported to
give far better results than the use of potentials derived from other
molecular properties. Unfortunately it has been impossible to find a force
field able to give a very accurate reproduction of the crystal properties that
is valid for all chemical compounds. For instance, the calculations can be
very good for a class (e.g. hydrocarbons) but any attempt to extend the
force field to molecules containing heteroatoms causes a worsening of the
final results. Moreover, very few cases of intermolecular H-bonding have
been studied so far[823s351and no general evaluation of the term EHBin
(7.14) is therefore possible.
Another point concerns molecular flexibility. In theory, lattice energy
calculations can be extended to the case in which the molecule has internal
degrees of freedom caused by rotation around single bonds. If t, are the j
torsion angles of interest, (7.15) can be rewritten as
U = U a , , 8, QI,V , x , Y , x, z,) (7.16)
where 8, QI,v, x, y, z concern a reference fragment of the molecule and z,
are the torsion angles defining the orientation of the other fragments with
482 1 Gastone Gilli
respect to the first one. To a first approximation the conformational energy

can be calculated as the non-bonded interaction energy among the atoms
belonging to the different molecular fragments by means of the same
potentials used for intermolecular interactions. To obtain more accurate
results, however, it will be necessary to use a specific intramolecular force
field such as that described on p. 509.
The calculation of the crystal free energy turns out to be much more
complicated problem because it requires evaluating, besides the lattice
energy, the term Kib= Evib- TSvib,that is all the possible vibrational modes
of the molecules around their equilibrium positions within the crystal. This
can be done by the methods of lattice dynamics within the Born-von
Karman but the topic is too long and difficult to be
treated here. Several general treatments of the method and a recent review
on the results of the calculations carried out on molecular crystals['] are
available and the interested reader is referred to them.
Polymorphism and the prediction of crystal structures

Polymorphism is an important aspect of the thermodynamics of molecular
crystals, even if little attention has been devoted to this specific problem so
far. However, it frequently happens that different polymorphic forms are
obtained by crystallization from different solvents (in particular when the
molecule has different possible conformers) and the number of polymorphs
would certainly increase steadily if a larger number of solvents and a wider
range of tempertures and pressures were investigated.
The basic reason for the phenomenon lies in the fact that the hypersur-
face of the lattice energy U has many minima in the twelve-dimensional
space of (7.15) (six cell parameters and six positional values defining the
molecule position) and their number is even larger in the (12Sn)-
dimensional space of (7.16), where n is the number of possible torsions. The
deepest minimum corresponds to the structure with the closest packing,
which has the minimum enthalpy because it can establish the most profitable
pattern of intermolecular interactions. It represents the most stable struc-
ture in the absence of the vibrational contributions, i.e. at absolute zero.
When the crystal is heated, however, the total free energy changes from Uo
+ +
to F,, = Uo AU Evib- TSvib,which is smaller than Uo (it has already been
shown that Fvih = Evib- TS,, is always negative and AU is negligible).
Comparing two hypothetical polymorphic forms of a same substance having
energies of (Uo), and (U0)2 at absolute zero (Fig. 7.7), the higher-energy
form is less efficiently packed and will have a greater vibrational entropy
and therefore a faster decrease of the free energy when the temperature
increases. From this point of view, the hypothetical molecular arrangement
having the average structure of the liquid at zero kelvin can be considered
as the polymorph of highest enthalpy and having the steepest possible
decrease of free energy with T. With reference to Fig. 7.7, the polymorphic
Fig. 7.7. Free energies of two hypothetical
transition 1+ 2 is possible whenever the corresponding energy curves cross
polymorphic phases of the same substance as a at temperature T, lower than the melting point. At the transformation point
function of temperature. The crossing of the two (Uo + AU + Evib- TSVib),is equal to (Uo + AU + Evib- TSvib)Zand a very
curves before the melting point defines the
transition from form 1 to form 2 at temperature small increase of temperature will cause a sudden phase transition by
T. +
absorbing the quantity of heat AH, = A(Uo AU + Evib)2-l which is easily
measured experimentally and must be numerically equal to A(TSvib)2-l=

TASv,b.
We may wonder what is the energy range of the minima on the
hypersurface of U, which could give rise, at higher temperatures, to phase
transitions or, in other words, whether polymorphism could be predicted
from simple lattice energy calculations. This seems unlikely, at first sight,
since it has been already shown that the vibrational part of free energy may
amount to a large percentage (22 per cent for naphthalene at room
temperature; Table 7.4) of the total free energy. Let us reconsider,
however, the expression for the transition enthalpy in the form AH,=
+ +
AU, A(AU) AEvi,; AH, is known from experience to rarely exceed
I kcal mol-' in molecular crystals, A(AU) is a small variation in an already
small term and can be neglected and AE,,, = T(ACV),-, (see p. 479) is a
small term which is necessarily positive because the heat capacity of a less
packed crystal must be slightly larger. Therefore, the total lattice energy
variation energy variation, AU, = 1kcal mol-' - AE,,,, observable in a
polymorphic transition should be less than 1kcal mol-', which seems to
imply that the number of possible polymorphs can actually be determined
from lattice energy calculations because no polymorphism will be observed
at ordinary pressures when the second deepest minimum differs from the
absolute one by more, say, than 1-2 kcal mol-l.
Effect of crystal forces on molecular geometry

A subject often discussed in structural literature is whether the crystal field
of van der Waals crystals can modify molecular geometries. Following
~ i t a i ~ o r o d s k ~two
, [ ~ cases
~ ~ ~ I can be distinguished: deviations from the
minimum-energy (or equilibrium) geometry and different choices among
possible geometries.
As regards the first case and calling AE the hypothetical molecular energy
variation caused by a displacement from the equilibrium geometry and AU
the concomitant change of lattice energy, the deformation will be possible
only when IAE I < I A UI . For dispersive interactions, /AUI has been
estimated[441from the changes of lattice energy consequent upon small
displacements of C and H atoms in the structure of benzene: it amounts to
less than 0.01 kcal mol-I for an atom shift of 0.03 A. Values for lAEl can be
easily calculated by the methods of molecular mechanics (see p. 506). A
shift of 0.03 A in the stretching or compression of a C-C or C-H single
bond requires an energy of some 0.30 kcal mol-' which is much larger than
IAUI: therefore, crystal forces cannot affect bond distances. Moreover, a
shift of 0.03 A corresponds to a bond angle change of 1.2-1.6" in common
organic compounds. The energy necessary to change by this amount a
C-C-C or C-C-H angle in benzene is about 0.03 kcal mol-' which is still
three times larger than IAUI; accordingly, it seems unlikely that relevant
(say, greater than 1-2") changes of bond angles can be produced by the
crystal field.
More complex is the case of rotations around single bonds, where the
same shift of 0.03 A can be produced by rotation around a C-C-C-C
torsion angle by expending only 0.0012 kcal mol-' as far as the simple
torsional potential is concerned. Here the controlling factor is represented
by the intramolecular contacts of the atoms, or groups of atoms, which are
484 1 Gastone Gilli
moved by the rotation around the single bond. Let us assume that, when the
molecule is in its conformational minimum, all contact distances of interest
are not far from the minimum of their atom-atom potentials (Fig. 7.2): the
global conformational energy minimum will inevitably be shallow and the
crystal field might even produce relevant changes of the torsion angle. A
classical example is diphenyl, which is not planar by itself but becomes
planar under the slight compression of the crystal field. Conversely, crystal
forces will weakly affect the value of the torsion angle when the walls of the
potential well are steep, a situation occurring when the molecule is in
tension (or rigid) because, in the conformational minimum, the atomic
contact distances are in part shorter and in part longer than the optimal
contact distances. From this point of view it can be stated that overcrowded
molecules are the least affected by the weak crystal forces.
It remains for us to consider the case of molecules having different
possible conformations which crystallize in a conformation which is not that
of minimum energy for the free molecule. This is quite possible and, in fact,
many conformationaly flexible molecules are found to crystallize in different
space groups with different conformations (conformational polymorphism).
It has already been remarked that the enthalpy of polymorphic transitions
seldom exceeds 1 kcal mol-' in molecular crystals and this makes it possible
that a molecule, whose second most stable conformation differs from the
first one by not more than 1-2 kcal mol-l, may gain from a more efficient
crystal packing that energy (or even more) which has been lost in
consequence of the choice of an unpreferred conformation.
Elements o f classical stereochemistry

Structure: constitution, configuration, and conformation
X-ray crystallography produces a quite detailed description of the spatial
arrangement of atoms in molecules, a description indicated by the generic
term of structure. However, a rigorous definition of what is structure is not
a trivial problem and modern s t e r e o c h e r n i ~ t r y [ ~tries
~ - ~to
~ ~give an answer
by means of the more basic concepts of constitution, configuration, and
conformation.
Molecular constitution indicates the way in which atoms are intercon-
nected, making due distinction between single and multiple bonds. It is
therefore a topological concept which can be reconnected to that of
graph.[551
The configurations of a molecule of given constitution indicate all
possible spatial arrangements of its atoms, ignoring those derived from
rotation around single bonds. In carbon chemistry different configurations
are mainly caused by the presence of double bonds or asymmetric carbon
atoms. A change of configuration requires, as a rule, the breaking and
remaking of a chemical bond and is a high-energy process.
The word conformation refers to the different spatial arrangements of a
molecule of given configuration produced by rotation (torsion) around
single bonds. The concept of conformation and the relative nomenclature
are more easily illustrated by an example. Figure 7.8 reports the potential
energy curve of 1,2-dichloroethane as a function of the Cl-C-C-Cl torsion
angle as calculated by the methods of molecular mechanics (p. 506-11). The
486 1 Gastone Gilli
The term structure includes the constitutional, configurational, and

conformational aspects of the molecule. The same three aspects do not
always need to be considered. For CH,, CHCl,, and CH2C12the structure is
defined by the simple constitution, for CHClBrI (asymmetric carbon) and
ClHC==CHCl (cis or trans configurations) by constitution and configuration,
and for H,C-CH,, H2BrC-CBrH2 and cyclohexane by constitution and
conformation; finally, the structure of HRC=CH-CH=CHR1 needs the
definition of constitution, configuration, and conformation, having four
possible cis-trans configurations around the two double bonds besides the
conformations produced by rotation around the central single bond.
It is important to remark that, while the crystal structure is always
univocally defined by atomic coordinates, cell parameters, and symmetry
operations, molecular structure is a concept which can be given a precise
meaning only within the frame of the valence bond theory (p. 502-6) since
all classical stereochemistry is centred on the idea of simple and multiple
localized bond.
Isomerism
Molecules having the same formula but different structures are called
isomers and are said to display isomerism. There are three different types of
isomerism: constitutional, configurational, and conformational, and the
two last are grouped under the common term of stereoisomerism.
Constitutional (or bond or topological) isomerism

Constitutional isomers represent the different modes of connecting the same
set of atoms. Accordingly, 1-butene, 2-butene, and iso-butene (la, l b , Ic),
the three o-, m- and p-disubstituted benzene derivatives (2a, 2b, 2c), and
the two nitro- or nitrito-complexes (3a, 3b) are constitutional isomers.
Usually isomers of this sort cannot interconvert and are to be considered as
distinct chemical compounds. In case an interconversion equilibrium can be
established the expressions tautomers, tautomerism, and tautomerie
equilibrium are employed; tautomerism is normally associated with the
migration of a hydrogen atom and a typical example is that of the
acetylacetone keto-en01 tautomerism (4a, 4b). It should be remarked that
less recently constitutional isomers were termed structural isomers; it is
clear that such an expression has to be avoided.
Configurational isomerism
Molecules having the same constitution but different configurations are
called configurational isomers. Two main cases are to be distinguished:
when the two stereoisomers can be related by a symmetry operation of
reflection they are enantiomers and when this is not possible they are called
diastereoisomers or, sometimes, diastereomers. The distinction has a
precise physical meaning. Both enantiomers and diastereoisomers are
identical as far as their chemical bonds are concerned. Only the former,
however, have indistinguishable non-bonded interactions; they have the
same physico-chemical properties in all respects except for their optical
properties and reactivity towards other enantiomeric species. In particular,
enantiomers rotate the plane of polarized light by the same angle but in
Motecules and molecular crystals 1 487
COOH I COOH COOH a a l a

I
0, /a
\bj
1
H O b i 4 H H*:+OH Hbi4Oi-l >=c=c\\\ brc=c=c

I I I \b
- I
H C - ~ ~ O HHO + C + H HWC4OH
-
b
COOHI
I =COOH E 0 0 ~ 7 a , b , c
8 a.b
488 1 Gastone Gilli
opposite directions, a property by which they can be identified and which

explains their older name of optical isomers.
Molecules having an internal symmetry plane cannot have enantiomers
because they are their own enantiomer. In such molecules the possible
diastereoisomers are often and traditionally called geometrical isomers.
Common cases are planar molecules (ethylene derivatives, planar rings, and
square-planar complexes). However, geometrical isomerism is not another
type of isomerism but just a particular name given to diastereoisomerism in
a specific situation.
The condition for a molecule to show enantiomerism (i.e. to exist as a
couple of enantiomers) is that it cannot be superimposed on to its mirror
image in any of its possible conformations. Such a molecule is said to be
dissymmetrical or chiral. Since rotations are allowed operations in trying to
superimpose the molecule with its mirror image, the most general condition
for chirality is the lack of molecular rotary-reflection axes, a condition which
excludes the presence of both internal symmetry centres (S2 = 1) and planes
(Sl = m) .
The most frequent cause of chirality is the presence of an asymmetric
carbon atom, that is a tetrahedral carbon bonded to four different atoms or
atomic groups as, for instance, in glyceraldehyde (5a, 5b). The number of
possible isomers increases with the number of asymmetric atoms. The
cr,P-dibromocinnamic acid, having two asymmetric carbons, produces four
different stereoisomers (6a-6d); the pairs (6a, 6b) and (6c, 6d) are enan-
tiomers while all other pairs are diastereoisomers. Tartaric acid (7a-7c)
illustrates the case where one of the two enantiomeric pairs vanishes
because the parent diastereoisomer has an internal symmetry plane (7c: a
meso form). This last example elucidates another interesting aspect of
enantiomerism. The meso form depicted in (7c) cannot exist in practice
because it is a high-energy eclipsed conformation. However, any pair of
right- and left-handed torsions of the same entity will generate a pair of
enantiomers which are isoenergetic, independently of the shape of the
potential energy curve; being isoenergetic, they will be present in the pure
liquid or in solution with exactly the same populations causing the
compound to be optically inactive.
Chirality can be associated with a variety of other bond situations which
can produce tetrahedral arrangements, as in allenes (Sa, Sb), spiro-
compounds, quaternary ammonium salts, or bis(benzoy1acetonato)-
beryllium (9a: one enantiomer shown). It can be also associated with non-
aromatic cyclic systems: for example, all rings with an odd number of
terms and two different non-geminal substituents have two asymmetric
carbons producing two diastereoisomeric pairs of enantiomers, as shown
by the cyclopropane derivative (10a-10d). An example of chirality in
coordination chemistry is given by the chelate octahedral complexes of the
type [ C ~ ( e n ) ~ ](~l l+a , l l b ) or [cis-CoC12(en),]+, where en stands for
ethylenediamine.
Of particular interest are pyramidal atoms, i.e. those having tetrahedral
hybridization but a hybrid orbital occupied by a non-bonding pair. It was
suggested many years ago that molecules similar to the amines NHR,R2 and
NR,R,R, should exist in two enantiomeric forms which, however, have
never been observed. The reason is that the two enantiomers (12a) and
(12b) exchange rapidly in solution by pyramidal inversion through a planar
transition state having an energy higher only by a few kcal mol-l. Further
investigations have shown than other potentially chiral compounds of
trivalent phosphorus and arsenic have higher interconversion barriers and
that their enantiomers may be actually resolvable, at least at low
temperature.
The absolute configuration of an enantiomer (i.e. its actual atomic
disposition in space) is a problem which cannot be tackled by chemical
methods. All attempts to obtain it from the sign of optical activity in
solution, the property which is more strictly related to enantiomerism, have
failed. The first absolute configuration of an enantiomer was only accom-
plished in 1951 by Bijvoet and c o - ~ o r k e r s [on
~ ~(+)-tartrate
] of sodium and
rubidium by anomalous scattering methods, as discussed in Chapter 5. Since
then absolute configuration determination has become routine.
The nomenclature for identifying enantiomers has changed over the years
and reflects the discovery of new techniques for their study. The oldest one
simply reports the sign of the optical activity (+ or -); later on Fisher, in
his fundamental studies on carbohydrates, discovered the stereochemical
series and enantiomers were named after the series they belonged to (D or
L). More recently Cahn, Ingold, and i re lo^[^^] introduced the R-S
nomenclature which describes exactly the spatial arrangement of groups or
atoms as obtainable from X-ray diffraction studies.
In this context it is worthwhile to remark that chemical nomenclature has
been developed for solutions and not for the crystal state, so that some
misunderstandings can sometimes occur. As far as enantiomers are con-
cerned, it is known that the usual chemical synthesis can only produce a 1:l
mixture of the two enantiomers which is called racemic mixture; it can be
separated (resolved) by reacting it with other chiral molecules by which the
two enantiomers are transformed into diastereoisomers; only the more
recent asymmetric synthesis can produce single R or S enantiomers. When
a pure enantiomer crystallizes, it can only adopt a polar space group (i.e.
without S, rotary-reflection axes), otherwise the symmetry element would
generate the other enantiomer; this is the case, for instance, for all natural
aminoacids or proteins. The crystallization of the racemic mixture may be a
more complex problem. Usually the mixture is thermodynamically more
stable and crystallizes as a homogeneous solid containing equimolecular
amounts of both enantiomeric molecules which is termed racemate. The
crystal obtained is usually centrosymmetric with the two enantiomers
related by a symmetry centre, though sometimes it happens to be polar with
a double (or of higher even multiplicity) asymmetric unit allocated to both
enantiomers. In the rare case that the racemate is less stable than its
components, the racemic mixture crystallizes in a polar space group with
spontaneous resolution of the enantiomers, that is it produces two types of
enantiomeric crystals, each one containing just one enantiomer (which is,
incidentally, the way Louis Pasteur discovered enantiomers for the first
time). This fact is the origin of a not uncommon error in structural
determination, that of determining the absolute molecular configuration
without taking into account the fact that the bottle from which the crystal
was taken contains one half of crystals of the opposite configuration. The
subject of chiral crystals can become quite difficult to understand in its
generality because there are also space groups which are chiral by
themselves and not because they have to allocate chiral objects (e.g. P31
490 1 Gastone Gilli
and P3J. Moreover, what has been said for configurational enantiomerism
has to be extended to conformational enantiomerism because conformations
become fixed within the crystals. For more details the reader is referred to a
more specialized treatment.[571
Going back to the other isomers, it has already been remarked that
geometrical isomerism is the traditional way of indicating diastereoisomer-
ism in molecules having an internal S,. Classical examples are cis-trans
isomers in ethylene derivatives (13a, 13b), syn-anti isomers in oximes
(14a, 14b) or azocompounds (15a, 15b), cis-trans isomers in square planar
(16a, 16b) or octahedral (17a, 17b) complexes, and mer-fac (meridional-
facial) octahedral complexes (18a, 18b). Geometrical isomerism is also
observed in saturated cyclic compounds where the substituents can be over
or under the ring plane. For instance, four-membered rings substituted as in
(19) can only show geometrical isomerism in consequence of their internal
mirror plane; accordingly (19a) and (19b) are cis-trans isomers. Complete
nomenclature rules for geometrical isomers are a ~ a i l a b l e . [ ~ ~ > ~ ~ ]
Conformational isomerism
Conformational isomers (rotational isomers, rotamers, conformers) are
the molecular states corresponding to minima of the potential energy curve
expressed as a function of the torsion angle around a single bond. With
reference to Fig. 7.8, states b, d, and f are conformers while a, c, and e are
transition states; all are possible conformations. As is the case for all
stereoisomers, conformers can have mutual relationships of enantiomerism
or diastereoisomerism; so conformers b and f are enantiomers while b and c
(or c and f) are diastereoisomers.
The energy barrier protecting conformers is usually rather small (5-
15 kcal mol-l) and much smaller than that protecting configurational iso-
mers. For reference, the thermal activation barrier for cis-trans
isomerization around a double C-C bond (13a, 13b) is some 40 kcal mol-l.
However, it is not the height of the barrier that can distinguish between
conformational and configurational isomers but the fact that conformers
differ in a rotation around what is a single bond (even if with partial
double-bond character) in the ground state of the molecule. This rule allows
us to classify as conformers the enantiomers produced by hindered rotation
in o, or-disubstituted diphenyls (20a, 20b), although the interconversion
barrier is so high that they can be resolved at room temperature. Such
resolvable conformers are sometimes reported as atropoisomers.
Ring conformations
It seems reasonable to assume that a ring containing single bonds, e.g.
cyclohexane, could have different conformations describable by a potential
energy curve of the type shown in Fig. 7.8, having minima and maxima
corresponding to ring conformers and transition states. The difficulties arise
when trying to define the independent variables that the potential energy is
a function of, and this for the evident reason that the torsion angles fixing
the conformation are not mutually independent but related by the condition
of ring closure. The problem is not easy to solve and only some basic
concepts will be discussed here; the mathematical complications will be
discussed in the next section.
Let us focus our attention on the origins of the energy barrier which
separates the different conformers. As it will be discussed in more detail
later (p. 507), the deformation of a molecule can be conceived of in terms of
displacements of its bond distances, bond angles, and torsion angles from
their optimal equilibrium values. The deformation energy needed rapidly
decreases in the order bond stretching or compression, bond angle bending,
and torsion around single bonds, while the torsion around double bonds
requires energies in between stretching and bending. Since energies of
single bond torsions are very small, transitions between conformations
which can occur without angle bending will have an almost null activation
barrier. In this case the ring is flexible because many different combinations
of the two isoenergetic (degenerate) conformations become possible. The
pathway interconverting isoenergetic conformations is called, for reasons to
be discussed later, a pseudorotation path. Classical examples of degenerate
conformations are the envelope (E) and twisted (T) conformations of
cyclopenthane (Fig. 7.9) and the boat (B) and twist-boat or twisted (T)
conformations of cyclohexane (Fig. 7.10). The opposite occurs when it is
impossible for the molecule to change conformation without bond angle
deformation (or, in general, bond stretching or rotation around double
bonds). In this case the molecule is rigid and its actual conformation is
protected by a barrier which can be of the order of magnitude of Fig. 7.9. The two degenerate conformers of
10 kcal mol-' for saturated rings and somewhat greater if there are double cyclopenthane (E =envelope, T = twisted) and
bonds stiffening the ring. Figure 7.10 shows the somewhat idealized shape of their symmetries. The ring conformation is
flexible because the t w o conformers can
the potential energy curve of cyclohexane as a function of a generalized interconvert with zero activation energy along a
coordinate of conformational interconversion as could be obtained by the pseudorotation path. Any observed
conformation can be described as a linear
molecular mechanics methods. The rigid conformation of cyclohexane is the combination of E and T.
chair (C), which is the most stable and is transformed into the T form
through a transition state called half-chair (H) which is some 11kcal mol-'
higher in energy. T and B are almost isoenergetic and can be interconverted
practically without any activation barrier (pseudorotation). C, B, and T are
the three low-energy conformations of cyclohexane and therefore its three
conformers according to the definitions given in the previous section.
The number of possible independent conformers of a ring can be
determined by the use of simple considerations. A planar ring of N atoms
has 2N degrees of freedom, out of which two are of translation, one of
rotation, and 2N - 3 of in-plane vibration. Allowing the atoms to vibrate
out of plane, that is the ring to become puckered, the number of vibrational
Fig. 7.10. Idealized potential energy profile of

cyclohexane as a function of a generalized
coordinate of conformational interconversion as
could be obtained by the methods of molecular
mechanics. C = chair, B = boat, T = twist-boat or
twisted, H = half-chair. B and T a r e degenerate
and ~ r o d u c flexible
e ring conformations; C is
CONFORMATIONAL COORDINATE the rigid low-energy conformer.
492 1 Gastone Gilli
degrees of freedom becomes 3N - 6 with an increment of (3N - 6) - (2N - 3)

= N - 3, which is just the number of out-of-plane independent vibrations or
independent puckering conformers. So three- four- five- six- and seven-
membered rings should have, respectively, zero, one, two, three, and four
puckering conformers. This kind of reasoning allows us to know the total
number of possible conformers but not their nature, a much more difficult
task to be discussed in the next section. Another qualitative consideration
can be derived from the examples already given. For cyclopentane there are
5 - 3 = 2 conformers (E and T), which are isoenergetic or degenerate and
flexible because they can interconvert along a pseudorotation path. For
cyclohexane the conformers are 6 - 3 = 3, out of which two (B and T) are
degenerate and flexible while the other one (C) is single and gives the
lowest-energy rigid conformation. By generalization: a N-membered ring
has N - 3 possible conformers which are arranged two by two in flexible
mutually interconverting degenerate pairs; if N - 3 is odd, the last single
conformer is rigid and has the lowest possible energy. Accordingly,
cycloheptane will have four possible conformers arranged in two flexible
pairs while cyclooctane will have one more rigid low-energy conformation,
which is the ground state of the ring.
Ring conformation and group theory

The problem of determining the N - 3 independent low-energy conforma-
tions (conformers) of a N-membered ring is reducible to that of finding the
normal out-of-plane vibrational modes of the ring, which can be easily
solved by applying group theory[581 to equilateral polygons. The method will
be illustrated for five- and six-membered rings and its understanding
requires some previous knowledge of group theory. From a formal point of
view the problem is the same as that of finding the molecular orbitals of
planar rings within the Hiickel theory.
Let us consider the regular hexagon of Fig. 7.11, where the out-of-plane
displacements are represented by the small vertical arrows. The hexagon
belongs to the point group D,,, whose character table is reported in Table
7.5. The reducible representation for out-of-plane vibrations is obtained by
Fig. 7.11. A system of coordinates for describing

the out-of-plane vibrations of a six-membered
ring.
Table 7.5. Character table for the point group D,
Dgh E 2C6 2C3 Cz 3C; 3C; i 2S3 2S6 ah 3ad 3av
A l , l l l l l l 1 1 1 1 1 1 xZ + y2, z2
A
,, 1 1 1 1 -1 -1 1 1 1 1 - 1 -1 R,
B,, 1 -1 1 -1 1 -1 1 -1 1 -1 1 -1
B
,, 1 -1 1 -1 -1 1 1 -1 1 -1 -1 1
E,l 2 1 -1 -2 0 0 2 1 -1 -2 0 0 (R,, R,,) (x:, yz),
E
,, 2 -1 -1 2 0 0 2 -1 -1 2 0 0 (X -Yr x ~ )
A,, 1 1 1 1 1 1 -1 -1 -1 -1 -1 -1
A, 1 1 1 1 -1 -1 -1 -1 -1 -1 1 1 z
El, 1 - 1 1 -1 1 -1 -1 1 -1 1 - 1 1
B," 1 -1 1 -1 -1 1 -1 1 -1 1 1 -1
El, 2 1 -1 -2 0 0 -2 -1 1 2 0 0 (x,y)
E", 2 -1 -1 2 0 0 -2 1 1 -2 0 0
applying all the group symmetry operations to the six small vertical vectors
of Fig. 7.11; the character for any class of symmetry operations (or column
of the character table) is given by the number of small vectors not moved by
the symmetry operation. It is easily seen that the characters of such
representation, r 6 , are
which can be resolved by standard methods"rn a combination of the four

irreducible representations (i.r.)
according to the equation r6= A,, + El, + B,, + EZu.

The normal modes of out-of-plane vibration (or modes of puckering) are
now to be found and, for the sake of clarity, let us make reference to the
final results shown graphically in Fig. 7.12. The modes transforming as AZu
and El, are immediately found from Table 7.5 to correspond to rigid body
motions (respectively translation along z and rotation around x and y).
A base for B,, (i.e. a representation transforming under the group
operations as B,,) is easily found because it has to be symmetrical with
respect to C, and antisymmetrical with respect to C6 and therefore
corresponds to alternate positive and negative out-of-plane displacements
(conformation C in Fig. 7.10); calling zi one of such displacements, the
vibration is described by the function
where 6-'I2 is a simple normalization factor granting that C iz: = 1. A base

t The number of times nithat the i.r. Ti appears in the reducible representation T' is given by
where h is the order of the group (in the present case 24) and x,(R) and x ' ( R ) are the
characters of the symmetry operations Tiand T', respectively.
494 1 Gastone Gilli
Fig. 7.12. Normal modes of out-of-plane

vibration (ring puckering) of a six-membered
ring and their symmetries. Any mode belongs to
an irreducible representation of the point group
D
,, and is therefore orthogonal to the others.
for E2, is not intuitive and must be found by ordinary group theory
methods. A first base can be obtained by associating the out-of-plane
displacement to atom 1 of the ring and operating on it by all symmetry
operations multiplied by their characters. The required base is the sum of all
the terms obtained, that is
which is a base for (or is transformed as) the boat conformation B and can
be written in normalized form as
As E2, is two-dimensional, the degenerate partner of 4, remains to be

found; it can be done by the same technique operating on z2 instead of zl
and obtaining the normalized function
which, however, is not orthogonal to @, because S = $ @,$J dv = -112. To

make the two functions orthogonal the Schmidt orthogonalization proce-
dure is used,? obtaining the orthogonal function required in normalized
form
GT = 4-112(-~2+ 23 - z5+ z6) (7.19)
which is the base for describing the puckering of the twisted conformation
T. GB and GT belong to the same i.r. and are degenerate, at least for
infinitesimal zi displacements, and all their linear combinations are allowed
t If the two already normalized functions 4, and 4, are not orthogonal, i.e. [ 4,4, du = S #
0, the new function 4' = 4, - SG, is orthogonal to +,, as can be shown by simple substitution.
solutions; this means that any linear combination of T and B conformers is

an allowed conformation of the ring or, in other words, that the two
conformers T and B can interconvert with zero activation energy along a
pseudorotation path.
It is now useful to find an analytical expression for describing the
displacements z, of a ring whose puckering amplitude is (C, zf)'I2 = q. It is
convenient to define the z, values with reference to the mean ring plane,
which is fixed by the condition of non-translation (C, z, = 0) and by those of
non-rotation around the x and y axes; these last conditions can be obtained
by equating to zero the sum of the vertical displacements weighted by the
functions representing the two components of the rotational i.r. El, of Fig.
7.12, that is
2 zisin [2x(i - 1)/6] = 0

i
where i can assume the integer values from one to six. Moreover, it can be
verified by simple computation that the following functions give the same
displacements as 4, in (7.17) and 4, and 4, in (7.18) and (7.19), and
therefore are transformed by the operations of the point group D6,, as the
i.r. BZgand E2, of Fig. 7.12
E;,: @,(i) = zi= 3-'I2q2 cos [4n(i - 1)/6]

El,: @,(i) = zi= 3-112q2sin [4n(i - 1)/6] (7.23)
where 6-'I2 and 3-'12 are normalization factors and q3 and q2 the mean
puckering amplitudes of the two vibration modes.
The function describes the out-of-plane displacements of the C
conformer while any linear combination k,& +
k2GTdescribes those of the
degenerate B and T conformers. For the properties of trigonometric
functions, such combinations can be written as
where 42 is a phase angle having values of 0, 60, 120, 180, 240, or 300" for
the pure B conformations and 30, 90, 150, 210, 270, or 330" for the pure T
conformations.
Equations (7.21) and (7.24) describe analytically the conformations of a
six-membered ring in terms of the three parameters q3, 9,) and G2 and the
N - 3 possible conformations remain characterized as the three C, B, and T
conformers. The characterizing parameters (9,) q2, G2) are more generally
identified as the N - 3 generalized puckering coordinates, that is the N - 3
variables necessary and sufficient to define any out-of-plane deformation of
the ring. q3 and q2 are called puckering amplitudes and G2 the phase angle
and any possible conformation of the six-membered ring corresponds to a
point in the three-dimensional space spanned by the orthogonal base
( q 2 , $ 2 ) q3).
496 1 Gastone Gilli
For example 'C4t corresponds to 9, = q, q2 = 0, and @, undefined, while

4C1 differs only for having 9, = -9. Moreover, 1 , 4 ~B2,s, , 3 , 6 ~B4,',
, 2,5B,
and B3,6 have all 9, = 0, q2 = q, and @, = 0, 60, . . . , 300°, respectively,
while 4T2, 6 ~ 23, ~ 1 'T4,, ' T ~ , and 'T, still have 9, = 0 and 9, = q but
$J2 = 30, 90, . . . , 330") respectively. The actual value of q = (C, z?)li2 is
determined by the equilibrium value of the bond angle, a , defined by the
ring atoms. It is self-evident that q will be zero for an equilateral hexagonal
ring having a = 120" and it can be shown that 9, is equal to 6 6 ' " ~ for an
equilateral C ring of side R and tetrahedral internal ring angles (e.g.
cyclohexane) .
It has already been remarked that any possible ring conformation is
defined by the triplet of values (q,, $J2,9,) in a Cartesian system of axes;
otherwise these coordinates can be expressed as polar coordinates
(Q, 0, $1, where Q is the total puckering amplitude defined as
and = @2,while 0 is an angle in the range 0 - ?G such that
q2 = Q sin 0 and 9, = Q cos 0. (7.26)

In this way any conformation is represented by a point on a sphere of
radius Q, 'C4 being at the north pole ( 0 = 0", $J undefined) and its
enantiomer 4C1 at the south pole, while B and T interconvert along a
pseudorotation path which is on the equator (6 = 90") in correspondence to
all possible values of @.
The previous discussion was mainly concerned with the low-energy
300
--
-5 -
--
-2 -
conformers C, B, and T. The ring, however, also has high-energy transition
states and there is general agreementUs9]that these are the envelope (E),
half-chair (H), and screw-boat (S) conformations of C,, C2, and C2
symmetry, respectively, as shown in Fig. 7.13. Cremer and ~ o p l e [have
shown that the E and H forms are located at tan 0 = fv(3/2) and fv 2 ,
~~]
' ] the S form occurs for tan 8 = f(1 v2).

respectively, and ~ o e y e n s [ ~that
This last author has reported a complete map of the six conformations,
+
usually called canonical, in polar coordinates, which is reported in Fig. 7.14
)
l
'
( (")
with some modifications. The figure has the further meaning that all
Fig. 7.13. The three activated states of puckering maximum circles indicated are possible interconversion among
of a six-membered ring and their symmetries
(E = envelope, H = half-chair, S = screw-boat).
conformations. The T-B pseudorotation circle is the equator, while along
the meridians the C conformation is converted to B through E or to T
through H and S.
The problem of the conformations of a five-membered ring can be solved
by the same methods. The ring has only two puckering coordinates because
N - 3 = 2; it belongs to the point group D,,, whose character table can be
obtained from any group theory text. The reducible representation for the
out-of-plane vibrations is
t According to the nomenclature usually employed, the conformations chair, boat, and
twisted (or twist-boat) are called C, B, and T and the atoms which are over or under the mean
plane of the ring are indicated by upper left or lower right indices. So, 2 3 5 is~ a boat ring having
the atoms number two and five directed upwards and B,,, its enantiomer while *T, is a twisted
ring with the second atom pointing up and the fifth one pointing down.
Fig. 7.14. Polar projection of the northern part of

the sphere reporting, for an arbitrary value of Q ,
the positions Band $C of the canonical
conformations of a six-membered ring. The
southern hemisphere contains the enantiomeric
conformations in centrosymmetric positions.
+ +
which can be decomposed as T, = A'! E'; Ei, where A; and E;' cor-
respond to translation and rigid body rotations, respectively. The puckering
modes belong to the two-dimensional i.r. E; and a base for them, found by
the usual methods, is shown in Fig. 7.15. The two conformers found,
envelope (E) and twisted (T),are degenerate and all their linear combina-
tions are possible conformations on a pseudorotation pathway. By analogy
with (7.20-7.24) the non-translation and non-rotation conditions are
Fig. 7.15. Normal modes of out-of-plane

zicos [2n(i- 1)/5]= 0, vibration (ring puckering) of a five-membered
i
ring and their symmetries. E =envelope, T =
twisted.
zisin [2n(i- 1)/5]= 0,
i
and the functions describing the puckering displacements are
@=(i)= zi= (2/5)lIzq2sin [4n(i - 1 ) / 5 ] , (7.29)

where i can assume all integer values from one to five and (2/5)'12 is a
normalization factor providing that C iz? = q:, q2 being the puckering
amplitude. Since E' is two-dimensional, any linear combination k1GE+
k2GT belongs to Ei and can be written as
498 ( Gastone Gilli
4,C;3E
where 4, is a phase angle having values of 0, 36, 72, 108, 144, 180, 216, 252,
1~ 'T,
288, and 324" for the ten E conformations and 18, 54,90, 126, 162, 198, 234,
270, 306, and 342" for the other ten T conformations, intermediate values of
@, corresponding to mixtures of the two along the pseudorotation path. The
30
complete pathway is shown in Fig. 7.16.
$2 "v
E A
*,
4 ~ FE
3
YJT4
Computation of puckering coordinates
The theory given in the previous section concerns equilateral and isogonal
rings. However, the equations obtained can also be used for irregular rings
d : 4 without loss of generality, at least as regards the Cremer and Pople
treatment,[601which makes use of the displacements zifrom the ring mean
plane. Other methods using the values of the endocyclic torsion angles (e.g.
T~
EI that proposed by Altona et al.[621for five-membered rings) may not have the
Fig. 7.16. Pseudorotation wheel describing the Same degree of generality.
conformations which can be assumed by a
five-membered ring during the changes of the
Several Computer programs[63264] are available for the computation of the
phase angle $,. puckering parameters according to Cremer and ~ople.[~'] The procedure
usually includes: transformation from crystal to orthogonal coordinates;
passage to the system of coordinates defined by (7.20) or (7.27), having
their origin at the centre of gravity of the ring, the z axis perpendicular to
the mean ring plane and the y axis passing through the first ring atom;
determination of the parameters q and $J (or Q, 8, and $) satisfying eqns
(7.21) and (7.24) or (7.30).
An example of application to the fructose and glucose rings of the
molecule of sucrose is reported in Table 7.6. The five-membered fructose
ring has q2 = 0.353 A and G, = 265.1". By comparing the calculated value of
Gz with those of Fig. 7.16, it is found that the ring conformation is very near
4 ~ with
3 a small component of E3. The six-membered ring has a Q value of
0.556& i.e. it is more puckered than the five-membered ring. The
calculated values of 8 = 5.2" and $ = 183.7" indicate, when compared with
Table 7.6. Atomic coordinates and puckering parameters for fructose and glucose rings
of the molecule of sucrose (coordinates from Brown and Levy, Acta Crystallographica,
839, 790 (1973))
Atom N Orthogonal coordinates Puckering coordinates
Puckering parameters: q, = 0.353 A, +, = 265.1"
Puckering parameters: q, = 0.050 A, 4, = 183.7"

q, = 0.554 A
0 = 0.556 A, 8 = 5.2", @I = 183.7"
those of Fig. 7.14, that the ring conformation is an almost perfect 'C, with a
very small distortion towards 5H4 and E,.
Molecular geometry and the chemical bond

An overview of bond theories
One of the central topics of structural chemistry concerns the interpretation
of the observed molecular geometries in terms of the existing theories of
chemical bonding. Theories derived from the fundamentals of quantum
chemistry include ab initio[65,661and ~ e m i e m p i r i c a l [ ~ ~methods.
-~~] In
principle both classes of methods (in particular the first ones) are able to
reproduce all the details of the experimental geometries with the great
additional advantage of giving them a theoretical justification in terms of
molecular orbital energies, shape, and symmetry. The main difficulties arise
from the great complexity of the calculations involved even for molecules of
relatively small dimensions. The situation is not so different from that
discussed in connection with the problem of molecular structure determina-
tion by microwave spectroscopy or gas electron diffraction. Great accuracy
can be achieved at a reasonable cost of both human effort and computer
time only for molecules of very small dimensions in comparison with those
X-ray crystallography is used to deal with. Another important aspect
concerns the objective difficulties many crystallographers may have in
understanding the fundamentals of quantum chemistry needed for a not
completely acritical use even of the most known and generally available
compter programs (e.g. GAUSSIAN[~~] for the ab initio programs or
MIND013, MNDO[~"and AM^[^'] for the semiempirical ones).
It is then not surprising that, in spite of the fact that quantum-mechanical
calculation programs have been implemented in most molecular graphics
systems commercially available, the most popular bond theories used in
interpreting molecular geometries remain the qualitative or
semiquantitative theories, that is those always based on quantum chem-
istry but with the different aim to produce concepts of easier application for
practical scientists. It is these theories which will be briefly reviewed here.
1. Valence bond (VB) In VB theory the molecular

geometry is interpreted in terms of the concepts of hybridization and
separability of localized a bonds and localized or delocalized n bonds. The
x bond delocalization is treated by means of the concept of resonance
between canonical forms, through, more recently, the Huckel molecular
orbital method limited to the n electrons (Pi H M O ) [ ~has
~ ] been frequently
employed. The idea of localized bonds is the logical foundation of
molecular mechanics (see p. 506-ll), which gives an interpretation of the
molecular geometry in terms of balls (atoms) connected by springs (the
chemical bonds), and, as previously discussed (see p. 486), is the basis of all
definitions of classical stereochemistry.
2. Valence shell electron pair repulsion (VSEPR) theory.[763771 This is
probably the most used qualitative tool for the interpretation of the shapes
of molecules containing main-group elements, while its applicability to
compounds of transition elements is rather limited. The VSEPR model has
506) 1 Gastone Gilli
been considered destitute of theoretical background for a long time and

only recently has it been interpreted in terms of basic quantum chemistry.[781
3. Transition metal-ligand bond theories. The VB and crystal field
(CFT)[~~-~'] theories were the first ones to try an interpretation of structure
and bonding in coordination compounds. The introduction of molecular
orbital methods in coordination and organometallic chemistry has given
origin to the ligand field theory ( L F T ) , [ ~ ~which
, ~ ~ ] is of great value for the
understanding of the energetics and geometry of these compounds. A
specific semiquantitative tool developed in the frame of LFT for calculations
concerning electronic spectra, magnetic properties, and equilibrium
geometries has been the so called angular overlap model ( A O M ) , [ ~ ~ ~ ~ ]
whose application has been successively extended to the study of the
molecular geometry of compounds of main-group
4. Qualitative molecular orbital theories (QMOT). The first attempts to
interpret geometries of simple compounds of the AB2 type in terms of
molecular orbitals were made by ~ a l s h . [ His ~ ~ ]method was successively
extended to a series of small molecular systems, such as AB3, AB4, A2B4,
A2B6, functional groups, e t ~ . [ ~ ~h ~ e, ~development
~ , ~ ~ ~ of~ QMOTs
~ ] in
interpreting molecular shapes has been paralleled by the great impact these
theories have had in the field of organic chemical kinetics (frontier orbitals
Moreover, QMOTs are having increasing applications in the
interpretation of the geometry of organometalic compounds by the methods
developed by ~ o f f m a n n , [ ~which
~ - ~ ~are
] based on the partitioning of the
molecule into fragments and which are of semiquantitative nature as
molecular orbitals are evaluated by E H T (extended Huckel theory),[951an
extension of the classical Hiickel method taking into account also o
electrons.
5 . Perturbative methods. These are based on the first-order (FOJTE)
and second-order (SOJTE) Jahn-Teller effects. F O J T E [ ~ ~ is] of great
importance in understanding the geometrical distortions which lower the
molecular energy by removing the orbital degeneration[971and has found its
most typical applications in the study of the coordination polyhedra of d9
c ~ m ~ l e x e s .The
[ ~ ~SOJTE
, ~ ~ ~ has similar applications but in closed-shell
molecules. It has been mainly developed by earso on['^^] extending a first
paper by arte ell;['^'^ it has been extensively used for predicting geometries
of compounds of main-group elements.[862871
6. Other qualitative theories. A rather simple model which has provided
remarkable agreement between experimental and predicted structures is the
hard sphere model proposed by ~ a r t e l l [ l ~and
' ] developed by ~ l i d e w e l l . [ ' ~ ~ ]
It makes the hypothesis that molecular geometry is mainly determined by
the repulsions among first non-bonded neighbours, each having an in-
dividual non-bonded radius. Other theories having a more limited range of
application have been proposed, such as the three centres-two electrons
bond in boron chemistry, the trans influence in square planar complexes or
the theories based on the acid-base concept, which have proved to be of
great utility in interpreting charge transfer effects in molecular crystals and
have been reviewed by several
In view of the concise character of the present exposition, it has been
decided to limit the discussion to the VSEPR and VB theories, which are
certainly the most popular models used in structural crystallography.

Moreover, only compounds of main-group elements will be taken into
account.
The VSEPR
The VSEPR theory assumes that the relative ligand arrangement around a
central atom is determined by the mutual repulsions of bonding or
non-bonding (lone) pairs of electrons. Bonding and non-bonding pairs are
identified from the Lewis-type electronic structure of the molecule. The
method can be summed up in the following few rules: Rule 1. Electron pairs
repel each other and adopt the geometrical disposition which minimizes
mutual repulsions; Rule 2. Lone pairs take more room than bonding pairs;
Rule 3: The space occupied by a bonding pair decreases with the increasing
electronegativity of the substituent; Rule 4. The two bonding pairs of a
double bond (and, more so, the three bonding pairs of a triple bond) occupy
more space than a single bond does.
Calling A the central atom, X the bonding pair (or the generic ligand
connected by it) and E the non-bonding pair, Rule 1 can be used to predict
the geometry of a large number of compounds (Fig. 7.17). BeCl,, BF,, and
CF, are AX,, AX,, and AX, molecules and will have linear, trigonal, and
tetrahedral geometries, respectively. Molecules of type AX3E (NH,, pH3,
NF,, PF,, or AsC1,) are pyramidal and those of type AX,E, (H20, H2S,
H,Se, or OF,) are bent (angular). Molecules of the type AX, may be either
trigonal bipyramidal, or square pyramidal. According to VSEPR calcula-
tions, the first geometry is slightly more stable and, in fact, AX, molecules AX, AX3E AX2E2
(PF,, AsF,, and PCl,) are usually trigonal bipyramidal. Finally, molecules
belonging to the types AX, (e.g. SF6) and AX,E (e.g. IF5) display the
expected octahedral and square pyramidal geometries.
For AX4E (TeCI,, SF,), AX3E2 (ClF,, BrF,), and AX2E3 (XeF,, IC1;) it
is necessary to establish whether the non-bonding pairs are in the axial or AX, AX,E AX3E, AX2E31
equatorial positions of the trigonal bipyramid. According to rule 2 the
non-bonding pair takes more room and should prefer the equatorial position
where it makes angles of 120, 120, and 90" with the vicinal electron pairs
(the three angles would be 90" for the axial position). Similar considerations
predict that AX,E, molecules (e.g. [ICl;]) are square planar with the AX, AX,E AX,E2
non-bonding pairs in trans positions. L
An interesting feature of VSEPR is its ability to interpret fine structural Fig. 7.17. Prediction of the shape of simple
details. For example, Rule 2 predicts that increasing substitution of X by E ~ ~ ~ ~ ~ e R :Ectron ~ ~ ~
will cause a narrowing of the remaining X-A-X angles because of the pairs (lone pairs), TP =total number of electron
greater space taken by the lone pairs. The effect is generally observed, as pairs(bondingandnon-bonding); HYB=typeof
VB hybridization.
exemplified by the series CH, (AX,), NH3 (AX,E), and H 2 0 (AX2E2)
where the X-A-X angle decreases in the order 109.5, 107.3, and 104.5".
Rule 2 can be considered a particular case of Rule 3 since the
non-existent substituent E can be thought of as the least electronegative of
all the possible X ligands. This last rule is thus the most suited for
understanding the angular changes caused by substitution and predicts that
the X-A-X angle decreases while the electronegativity of X increases.
Although a few exceptions are known, the effect is correctly observed in a
variety of cases. For example, the X-A-X angle is 103.8 in OF, and 104.9
in H,O, 102.1 in NF, and 106.6" in NH,. In the halides of the main-group
502 1 Gastone Gilli
elements of the type AX2E2or AX3E the bond angles are known to increase
in the order F < Cl < Br = I. Another well documented[1061series of
electronegativity-dependent angular variations concerns the monosubsti-
tuted benzenes. The endocyclic C-C-C angles in ipso (that is that carrying
the substituent) changes with the substituent itself, being 117.2 for N(Me),,
118.1 for CH,, 120.00 for H (unsubstituted benzene), 121.4 for C1, 121.8 for
-CN, and 123.4" for F. It is evident that the angle is wider the greater is the
electronegativity of the substituent group.
Rule 4 explains why the bond angles trans to a multiple bond are smaller
than usual, as can be illustrated by some examples in the X2A=Y system.
The X-A-X angle should be 120" in the absence of perturbing effects but
becomes respectively 108 and 111" in F2C=0 and C12C=0, where both the
double-bond repulsion and the high electronegativity of the substituents
contribute to the squeezing of the angle. The effect of the double bond is
better seen when the X substituent is less electronegative, for example in
H2C=0 and (NHJ2C=O, where the X-C-X angle is 108" in both
compounds.
Valence bond (VB) theory[5,73,741

The VB theory is too well known to require a detailed exposition and only
some aspects concerning the interpretation of molecular geometries will be
discussed here.
The mutual orientation of the bonds departing from a central atom is
interpreted in terms of hybrid orbitals (HO), which are linear combinations
of the atomic orbitals (AO) of such an atom; typical sp, sp2, sp3, sp3d, and
sp3d2 HOS assume relative orientations which are respectively linear,
trigonal, tetrahedral, trigonal, bipyramidal, and octahedral (Fig. 7.17) and
the overlapping of the HOs with the correct AOs of the bonded atoms
generates the frame of the a bonds within the molecule.
After the formation of all the a bonds between the central atom and its
surrounding ligands, the generic set s'pmdn of hybrid orbitals can allocate
+
only (1 m + n ) bonding or non-bonding (lone) pairs, which require twice
as many valence electrons. The hybridization state of the central atom
remains determined by the total number of valance pairs available, while its
connectivity is also affected by the number of non-bonding pairs which
occupy an H O without adding any bonded atom. Moreover, electron pairs
implied in n bonding are irrelevant for determining the hybridization state.
So, CH,, NH,, and H 2 0 have all eight valence electrons (sp3 hybridization)
but only CH, is tetrahedral while NH3 is pyramidal (one non-bonding pair)
and H 2 0 is bent (two non-bonding pairs). Application of these rules to the
examples of the previous section shows that VB gives exactly the same
results obtainable by the use of VSEPR theory (see Fig. 7.17).
In VB theory the double bonds do not affect the shape of the molecule,
which is strictly determined by the net of its a bonds. Double and triple
bonds are conceived of as a bonds to which one or two further n bonds are
added and the only effect of such an addition is to shorten the bond
distances. The idea of single or multiple localized bonds between two given
atoms is clearly an oversimplification of thexomplex many-body interactions
between nuclei and electrons within the molecule. However, it has proved
to be accurate enough to allow tabulations of characteristic bond
Table 7.7. A selection o f characteristic b o n d distances i n organic

molecules ( i n A); R = alkyl, A r = aryl, ar = aromatic
C-C sp3-sp3 1.54 C-P c(sp3)-P

sp3-sp2 1.51 C-H C(sp3)-H
sp3-sp 1.46 C(ar)-H
C-C aromatic 1.39 C- F c(sp3)-F
r-C )*c( 1.33 C(sp2)-F
)C=C=C( 1.31 C-CI C(sp3)-CI
)C==C=C=C( 1.28 c(spZ)-CI
1.20 C(sp)-CI
C- N sp3-sp3 1.47 C-Br c(sp3)-~r
r-N 1.27 C(sp2)-Br
r=N 1.16 C(sp)-Br
C-0 R-0-R 1.43 C- l C(sp3)-l
RCO-OR 1.34 C(sp2)-l
RCO-OH 1.31 N-H
RCOO-R 1.44 0- H
C(ar)-OR 1.36 S-S RS-SR
C-0 RCOO- 1.25 0-0 RO-OR
r-0 )c=O 1.20 N-N ) N-N (
=C=O 1.16 N=N -N=N-
C-S C(sp3)-S- 1.82 N-0 ) N-O-
C(sp2)-S- 1.76 )N+O
r-S 1 .7l -NO2
C-Si c(sp3)-~i- 1.87 N=O -N=O
distance^[^^'-^^^] which are reasonbly reproducible from molecule to mole-

cule. Table 7.7 reports a selection of characteristic bond distances observed
in organic molecules.
Bonds of intermediate multiplicity are interpreted within the VB theory

in terms of resonance, as shown for the simple cases of benzene (7.1) and
carbonate ion (7.11). The structure of the molecule is represented as a
resonance hybrid between different limit forms, called canonical. The
hybrid justifies the fractional bond order, for instance 1.5 in benzene (7.1)
and 1.33 in the C-0 bond of the carbonate ion (7.11). Such a fractional
number has been called bond number, n, by pauling,['I who has proposed
different formulae for its evaluation. The most used is
Ad = d(n) - d(1) = - c log,, n (7.31)
where d(1) and d(n) are the distances of a single and n-ple bond,
respectively, and c is a constant to be determined. Some applications of this
equation will be discussed in the last part of this chapter.
VB is particularly suited for rationalizing small changes of bond distances
and angles in terms of substituent electronegativities. The discussion follows
the treatment given by ~ e n t [ " land
~ starts from the elementary considera-
tion that sp, sp2, and sp3 hybrids have decreasing s character (50, 33, and 25
per cent, respectively). This leads to important consequences:
1. Since the s A 0 is at a lower energy than the p orbital, the
504 1 Gastone Gilli
electronegativity of an atomic valence increases with the s character of the

valence itself, that is an atom binding by one of its sp HO is more
electronegative than when it binds by an sp2 HO or by an sp3 HO. If the
HO in question allocates a non-bonding pair (NB), the basicity of the atom
will be affected according to the rule that the greater is the s character of the
HO donating the lone pair, the worse will be the atom as a donor.
Accordingly ketones, where the lone pair is on an sp2 HO, are weaker bases
than ethers, where it is on an sp3 HO. Likewise nitriles R-C=N are weaker
bases than pyridine C,H,N, that, in turn, is weaker than ammonia NH3, in
agreement with the fact that nitrogen hybridization changes from sp to sp2
to sp3.
2. The best superposition of two ns AOs occurs at a smaller internuclear
distance than that between np AOs and then bond distances become shorter
with the increasing s character of the HOs involved. In fact the C-H bond is
1.09 in methane, 1.O7 in ethylene, and 1.O6 A in acetylene, the single bonds
C(sp3)-C(sp3), C(sp3)-C(sp2), and C(sp3)-C(sp) have decreasing bond
lengths of 1.54, 1.51, and 1.46 A and, finally, double-bond distances
decrease along the series ) C=C( , ) C=C==C( , and ) C=C=C==C( in the
order 1.33, 1.31, and 1.28 A. It may be remarked that this contraction is not
a constant for all the elements but seems to increase with the electronegati-
vity of the atom, as shown by the series carbon-halogen in Table 7.7.
3. The bond angles steadily increase with the increasing s character of the
HOs, the angle between sp3, sp2, and sp HOs being respectively 109.47,
120, and 180". Smaller changes caused by electronegativity differences are
rationalized by the so-called Bent's rule:11111The p character of an atom
tends to concentrate in its HOs pointing towards more electronegative
substituents, the effect being stronger the less electronegative the atom itself is.
All the substituents can be arranged, starting from the lone pair, in order of
increasing group electronegativity and the narrowest bond angles will be
those implying the most electronegative ligands. The effect is well docu-
mented. The X-0-X angle is 111, 109, 104.9, and 103.8" in (CH3)20,
CH30H, H 2 0 , and F 2 0 , but only 100 and 92" in CH3SH and H2S in view of
the smaller electronegativity of sulphur. In the AX3 series the X-A-X angle
is 109, 106.6, and 102.1" in N(CH3)3, NH3, and NF,, 100 and 93.3" in
P(CH3)3 and pH3, 96 and 91.5" in As(CH,)~and ASH,. Bent's rule is the
VB equivalent of Rules 2 and 3 in the VSEPR theory and it can be shown to
include even Rule 4. In the Y=AX2 fragment the n bond contributes to
shorten the X=Y bond length, causing a parallel increase of s character in
the a bond and, in turn, an increase of the p character of the HOs pointing
to the X substituents so that the X-A-X bond angle is narrowed.
Hybridization. The machinery

A generic HO is a linear combination of AOs, 4,
and the problem consists of finding the set of aij values matching the
required molecular fragment geometry. A few rules allow us to set up a
system of linear equations:

1. The coefficients must be such that any HO is transformed into one of its
equivalent orbitals by all the symmetry operations of the fragment.
2. The HOi must be orthonormal, that is
3. Since the atomic orbitals @ j are orthonormal the sum over all HOs of the
squares of the coofficients of any @ must be unitary, that is
Ci a; = 1 for any j.
As an example, let us build up the three trigonal HOs in the plane (xy).
The axes are chosen in such a way that HO, points along x and H 0 2 and
HO, are in the (-x, y) and (-x, -y) quadrants, respectively. Let us write
the HOs as
The s orbital has spherical symmetry and must be equally shared by the
three HOs; since C ia: = 1 for condition (3), it must be that a, = a, = a, =
11Y3. c, is zero because p, is orthogonal to HO, in view of the axes chosen
+
and therefore a: b:= 1 for (2), so that b1 = 21v6. Since H 0 2 and H 0 3 are
+
symmetrical with respect to the a, containing x and c: c: = 1, it results
that c2 = -c3 = 1/v2. Likewise b2 = b, = -1Iv6 because of condition (2).
The final HOs are
HO, = 1/Y3 s + 21v6 p,,

H 0 2 = 11Y3 s - 1 l v 6 p, + 1 / v 2 p,, (7.36)
H 0 3 = 1 / v 3 s - l / v 6 - 1/Y2 p,.
It is easy to check that condition (1) is fulfilled for all the symmetry
operations of the point group D,, that the trigonal AX, fragment belongs
to. The hybridization index n of each HO is defined as the ratio between
the sum of the squares of the p A 0 coefficients and the square of the s
coefficient. In this case n1 = n2 = n, = 2 and the hybridization is termed sp2.
Alternatively, the s and p characters (S and P, respectively) of the HO can
be used, as defined by the conditions n = PIS = P l ( 1 - P ) = (1 - S)/S and
+ +
S P = 1 and by the inverse relationships S = ll(n + I), P = nl(n 1). In
the present case S = 0.33, and P = 0.666.
By the same methods it can be calculated that the two spHOs
(n, = n2 = 1) oriented along x have equations HO1 = (s + p,)/v2 and
H 0 2 = (s - p,)/v2 and that four equivalent sp3 HOs (n, = n2 = n, = n4 = 3),
oriented in such a way that each HO makes the same angles with the
506 1 Gastone Gilli
Cartesian axes, have equations
Both HO coefficients ail and hybridization indices n have been worked out
starting from symmetry considerations. If the HO geometry is not exactly
linear, trigonal, or tetrahedral the n values are no longer integer but
become fractional numbers and not necessarily equal. Equations for the
calculus of the ail and n values have been reported[11231131 for some particular
geometries, that is the quasi-tetrahedral systems AX2Y2 of symmetry C2,
and AX2YZ of symmetry C, and the planar quasi-trigonal AX2Y system of
symmetry C,,.
T /*p;,
For the AX2Y2system of C2, symmetry (left side of Fig. 7.18) it has been
shown that n, = n2 = -sec a , n3 = n, = -sec /3 and that the angles a and /3
+
are related by the equation cot2 (a/2) cot2 (/3/2) = 1. By putting a =
X
109.47" we obtain nl = n2 = n3 = n4 = 3, /3 = 109.47", S = 0.25, and P = 0.75
/ 4 Ir
, , a , HO, I a I' for any HO, in agreement with the theory. In the water molecule
0 - -- - , HO, - - - -- "02
( a = 104.9") the two HOs pointing to the hydrogens have n =3.89,
1
Y I / Y
I
1'
I
S = 0.20, and P = 0.80 and the angle between the lone pairs is calculated to
Fig. ,.,*.
Definition of the coordinate system
and labelling of the relevant angles between
be /3 = 114.8", in agreement with Bent's rule because the p character is
found to concentrate on the HOs pointing to the more electronegative
hybrid orbitals in the AX2Y2 (left) and AX2Y substituents, that is the hydrogens. The opposite case occurs in (CH3)20,
(right) systems discussed in the text.
where the methyl is an electron donating group; the observed angle a is
111°, allowing us to calculate that n = 2.79, S = 0.26, and P = 0.74 for the
HOs directed towards the CH, groups and /3 = 108".
In the planar system AX2Y (Fig. 7.18, right side) the important
relationships become nl = n2 = -sec cu and n, = tan2 (ct.12) - 1. Only if
cu = 120" it will be that nl = n2 = n, = 2, while n1 = n2 > n3 for a < 120" and
nl = n2 < n, for a > 120". Since a is smaller or greater than 120" when H 0 3
is directed towards substituents less or more electronegative, Bent's rule is
obeyed.
These few examples can help to illustrate some interesting and not always
well understood aspects of VB theory that can be summarized as follows:
(1) The hybridization indices are, in general, fractional numbers which
become integers only for a few very special geometries;
(2) the hybridization indices and the p and s characters of the HOs are
straightforwardly calculable from the bond angles and vice versa;
(3) Bent's rule is not a simple qualitative rule but can be quantified, at least
in a number of simple cases.
Molecular mechanics
Molecular mechanics is a method for calculating the equilibrium geometries
and other properties of ground state molecules on the basis of a purely
classical mechanical model. The molecule is considered to be a set of atoms
connected by elastic springs, and to any internal molecular coordinate (bond
distances and angles, torsion angles, distances between non-bonded atoms,

etc.) is assigned a natural equilibrium value. To any displacement Ax from
the natural value is associated a restoring force proportional to the
displacement itself, F = -k AX, and an energy proportional to its square,
E = k(Ax)'/2. Molecular mechanics is essentially an attempt to assign
parametric values to all natural equilibrium values and force constants k,
whose complete set is called a force field and whose knowledge will allow us
to calculate the potential energy hypersurface for the movement of the
atoms within the molecule. The energy calculated assumes the meaning of
the deformation or steric energy of a particular geometry of the molecule
with respect to its equilibrium geometry.
The values of the parameters defining the force field are obtained
empirically by comparison of observed and calculated molecular geometries.
Other properties can be optimized as well, most frequently the formation
enthalpies and the vibrational frequencies. In particular a force field able to
calculate both geometries and frequencies is called a consistent field. The
central point remains that of being able to calculate the properties of the
maximum number of different molecules using the minimum number of
constants transferable from one molecule to another.
The present discussion is necessarily concise and more details can be
obtained from r e ~ i e w s [ ~ ~and
~ ' ' ~books['23,1241
] available on the subject.
Several molecular mechanics computer programs are distributed by QCPE
(Quantum Chemistry Program ~ x c h a n ~ e ) . [ ~ ' ~ ]
The total energy, E, is considered to be the sum of different contributions
which are, in the order, the stretching, bending, stretching-bending,

out-of-plane bending, torsion, non-bonded, and electrostatic energies and
have the following meaning:
1. Bond stretching. The total energy of stretching or compression of all
chemical bonds of the molecule is of the form
N
Es = (112) C ks,i(Al,)'
i=l
(7.38)
where the sum is extended to all the N bonds within the molecule, ks,i are
their stretching force constants and Ali = li - the differences between the
actual and equilibrium bond lengths. This model tends to overestimate the
energy needed to produce very large lengthenings because, in such a case,
the bond becomes increasingly yielding in consequence of the decreased
overlapping of orbitals. It is common practice to add a small cubic term of
the type k , , , ; ( ~ l ~ )where
~, kScjiis a small negative constant taking into
account this last effect.
2. Bond angle bending. Its total energy is expressed as
where the sum is extended to all the M bond angles, AOi = Oi - Oo,; are the
differences between the actual and equilibrium bond angles, and kb,i the
bending force constants. Bending energies are smaller than stretching
energies by an order of magnitude because angles are much more yielding
508 1 Gastone Gilli
than distances. Also in this case a cubic term can be added, which is of the
type kb,,i(h8i)3and where kbc,iis a small negative constant.
3. Stretching-bending terms. An improved force field is obtained if
proper allowance is made for the fact that the narrowing of a bond angle is
paralleled by the legthening of the two encompassing bonds and vice versa.
If 8i,ABcis the angle delimited by the bonds li,ABand li,BCconnecting the
three atoms A, B, and C, the stretching-bending energy is defined as
where ksb,iare the corresponding force constants and Ali and AOi have the
usual meaning. The term Esbis small in comparison with the previous terms
and is neglected in some force fields. A field containing it is called a valence
force field; this differs from the Urey-Bradley force field which presents a
different treatment of the A . . . C geminal interactions and is not discussed
here.
A C
B
H
/
O - A~- - - -c- P>
B
o
4. Out-of-plane bending. In the deformations concerning trigonal sp2
atoms it is useful to distinguish between in-plane and out-of-plane deforma-
tions (Fig. 7.19). The in-plane deformations concern the angles A-P-B,
Fig. 7.19. in-plane and out-of-plane A-P-0, and B-P-0 and are dealt with by eqn (7.39), while the
deformations of a trigonal sp2 atom. out-of-plane deformations are the displacements from zero (A8) of the
angles C-A-P, C-B-P, and C-0-P. The associated energy is
for any single sp2 atom of the molecule.

5 . Torsions. All force fields include energy terms accounting for the
rotations around bonds. The corresponding force constant is relatively large
for rotations around double bonds but very small for rotations around single
bonds. A common expression for the torsional energy is
E , = C [Vl(l + cos @)I2+ V2(1- cos 2@)/2+ V3(1+ cos 3@)/2] (7.42)
where @ is the torsion angle (-180" s @ s 180°), V,, V2, and V3 are the
force constants, and the sum is extended to all sequences 1-2-3-4 of
bonded atoms in the molecule. The second term has the value zero for
@ = 0 and f180" and maxima for @ = f90; it is used to describe the rotation
around double bonds and in this case V1 = V3 = 0. The third term is equal to
zero for @ = f60°, f180" and has maxima for @ = 0, f120"; it describes the
rotation around single bonds connecting sp3 atoms. The first term is zero for
@ = 180°, has a maximum for @ = 0°, is used as a small corrective term in
special cases, and is often neglected.
6. Non-bonded interactions. The corresponding energy, Enb, can be
expressed by the atom-atom potentials discussed on p. 474. The interac-
tions among all the atoms of the molecule are taken into account, with the
exclusion of those between first neighbours (1-2 or bonded atoms) and
second neighbours (1-3 or vicinal atoms). Third-neighbour (or 1-4)
interactions are considered in both the Enband E , terms. Special treatment
has sometimes been reserved for the lone pairs on etheric oxygens or aminic
nitrogens by considering them as pseudo-atoms localized at about 1A from
the atom and having their own atom-atom potentials.
7. Electrostatic interactions. It has already been shown on p. 477 that
electrostatic interactions can be evaluated as interactions among partial
charges localized on the atoms or among small dipoles associated with each
chemical bond. Both methods are used in molecular mechanics.
A relevant number of different force fields proposed by different
~ ]currently available and Table 7.8 reports, as an example, the
a ~ t h o r s ' ' ~are
Table 7.8. Force field parameters for alkanes and non-conjugated alkenes according to
M M ~ / M M P ~ . [ ' ~Force * ' ~ ~ ~ i n kcal mol-' A-2 or i n kcal mol-' deg-'; bond
, ' ~ ~constants
moments (b.m.) in D. In the symbol (e,),, n indicates the number of additional hydrogen
atoms bonded to the central atom. I= C(sp3), 2 = C(sp2), 3 =H. Final energies i n
kcal mol-'
'Stretching' (k,, = -k,)

Atoms k,/2 lo b.m.
1-1-1 0.0083 109.47 110.51 110.20

1-1-3 0.0053 108.50 108.51 107.90
3-1-3 0.0042 112.80 111.10 109.47
1-1-2 0.0083 109.47 110.51 110.20
2-1-2 0.0083 109.47 110.51 110.20
2-1-3 0.0053 108.50 108.51 107.90
1-2-1 0.0083 116.60
1-2-2 0.0083 121.70 120.50
2-2-2 0.0131 120.00
1-2-3 0.0053 119.50
2-2-3 0.0053 120.00 118.80
3-2-3 0.0042 122.40
'Stretching-bending' 'Out-of-plane bending'

Atoms k,, Atoms ko0&
'Torsions' 'Non-bonded interactions'

Atoms 4 & 5 Atom ri Ei
510 1 Gastone Gilli
parameters used by the program M M ~ / M M P ~for [ ~ ~hydrocarbons.

] The
parameters of any force field are optimized to reproduce the experimental
geometries of some reference molecules and used for predicting those of
other chemically similar molecules. A force field developed for the alkanes
will reproduce their observed geometries with a precision comparable with
the experimental e.s.d. For other classes of molecules, such as alkenes,
alcohols, ethers, thioles, esters, carboxylic acids, etc., the agreement
between observed and calculated quantities is definitely worse in view of
complications arising from the treatment of n-delocalized systems, partial
double bonds, lone pairs, intramolecular hydrogen bonding, or highly polar
groups. The discrepancies increase again when attempting to develop a
force field able to deal with several functional groups at the same time and
its seems unlikely that a universal field of sufficient accuracy will ever be
obtained.
Let us now consider a few practical aspects of molecular mechanics
calculations. Conformations undergo by far the greatest changes in a
molecule, bond distances and angles being inevitably constrained to the
proximity of their equilibrium values. It is therefore useful to conceive of
the total potential energy as a hypersurface in a space where it is the
ordinate, while the torsion angles are the abscissae. On this hypersurface
there are as many minima as there are possible conformers (i.e. conforma-
tions of relative minimum energy; see pp. 484 and 490-2) of the molecule.
Such minima are separated by regions of higher energy, where it will be
possible to identify passages of relative minimum mapping the pathways of
interconversion from one conformer to another. A typical molecular
mechanics calculation starts from a given geometry, either obtained from
the crystal structure or built up from standard values of distances and
angles, and optimizes all the parameters in such a way as to reach, in a
minimum number of program cycles, the molecular geometry corresponding
to the nearest energy minimum. This is not necessarily the absolute
minimum; that can be identified with certainty only by repeating the
calculations starting from as many initial geometries as are necessary to
systematically map the energy hypersurface. When the main minima have
been found, the surrounding region can be scanned to evaluate the shape
and the height of the energy barriers which separate one minimum from
another. The energy profiles for 1,2-dichloroethane of Fig. 7.8 and for
cyclohexane of Fig. 7.10 are typical results of molecular mechanics
calculations.
A great number of organic molecules have been studied by the methods
of molecular mechanics, whose success and popularity are due to the fact
that it is a theoretical method which is very fast and extremely easy to
understand. It does not require experimentation because the initial model
can be obtained from tabulations of characteristic distances and angles. The
computing time is very short with respect to the time needed by quantum-
mechanical calculations. A benchmark carried out on propane[651has shown
that the full geometry optimization requires times in the ratios
1:12:l3:662:5665 when accomplished, in order, by MM2, MIND0/3,
MNDO, and GAUSSIAN82 with 3-21G and 6.31G* basis sets. Since
molecular mechanics is so fast that it can optimize the geometry of
molecules having the complexity of cholesterol in a few minutes of
computer time, it has the additional advantage that large parts of a potential
hypersurface can be calculated at a reasonable cost, giving information not

only on the geometry of the different conformers but also a general view of
the molecular motions and interconversion pathways.
This is not to say that molecular mechanics can give the best solution to
all problems of conformational analysis. It presents several pitfalls, some of
which have already been discussed, and all of which are essentially related
to the lack of transferability of the force field parameters among different
chemical classes. However, it cannot be denied that its use gives to the
crystallographer the possibility of understanding aspects of the molecular
geometry (particularly its dynamics) which are not accessible through crystal
structure determination.
Molecular hermeneutics: the interpretation of

molecular structures
Correlative methods in structural analysis
The mass of structural data obtained by X-ray crystallography of molecular
crystals has been the object of many attempts at systematization and
rationalization. Altogether they define a wide borderline area of interest
which does not really belong to any traditional chemical discipline and is
developing by itself along often not well defined tracks and with purposes
which can appear somewhat confused, though all are essentially aimed at
understanding the relationships between the experimentally determined
molecular structure and the properties the molecule will display in gases,
liquids, solutions, and in chemical and biochemical reaction environments.
Though we may suppose it is a branch of structural chemistry, such an area,
for the moment, does not even have a precise name, so that we are left free
to use the name of our choice. Structure-property relationships has often
been used though interpretation of molecular structures is certainly much
more correct. To pay homage to the scholarly tradition of translating
scientific sentences into ancient Greek, I would suggest the name molecular
hermeneutics which has the meaning of the science, or the art, of molecular
interpretation.
Irrespective of the name we choose, all attempts at systematic inter-
pretation of molecular structures have one point in common: they make use
of correlative methods, being based on the comparison of extended sets of
structural parameters (distances, angles, geometries of specific molecular
fragments, inter- or intramolecular contact distances, etc.) which are
correlated among themselves or with other physical properties. The simplest
of these attempts are intended to produce tables of standard bond distances
and angles, but most of them are devoted to clarify the systematic aspects of
a variety of chemical situations such as the characteristics of the inter- and
intramolecular hydrogen bonding in crystals and molecules, the typical
geometries of organic functional groups, the trans influence in coordination
compounds, the role of n back-bonding in organometallic compounds, the
structural requirements which organic metals or semiconductors must fulfill,
or, as extensively discussed in connection with the VSEPR and VB theories,
the effects that electronegativity and hybridization changes produce on the
geometries of simple molecules.
512 1 Gastone Gilli
As it is impossible to give even a summary of these systematic studies, it

has been chosen to discuss in some detail a single method of systematic
investigation of molecular geometries, the so-called structure correlation
m e t h ~ d [ ~ , " " ~which
~ ~ ] , can be considered to be an attempt to obtain
information on the dynamic behaviour of molecules from the inevitably
rather static crystal data. The basic idea of the method is as follows.
Usually, internuclear distances found in molecular crystals fall into the two
distinct fields of bonding and non-bonding distances, which are separated by
a forbidden gap going from the longest bond to the shortest contact
distance. For instance, this gap is extended from 1.48 (single-bond distance)
to 3.25 A (sum of van der Waals radii) in the case of the C-N distances. In
real crystals, however, several interatomic C-N distances have been found
to fall in this interval, a fact that seems interesting because any situation
intermediate between bonding and non-bonding is typical of chemical
reactions.
As early as 1931 Eyring and ~ o l a n y i [ l suggested
~~] that any chemical
reaction could be described as a low-energy path on a potential energy
hypersurface (or surface) of the reaction itself, which is the function giving
(within the Born-Oppenheimer approximation) the total potential energy
of the system as a function of the relative positions of all the atomic nuclei.
The reaction pathway is that trajectory (reaction coordinate) which
connects the points of relative minimum on the surface and leads from the
reagents to the products through the higher energy transition or activated
state. For the scope of this account it is sufficient to make reference to the
classical examples reported in all books on physical chemistry, such as the
+ + + +
reactions H Hz Hz H or Hz D H HD, or to the elementary
accounts on the subject which are a ~ a i l a b l e . [ ~ ~ ~ - ~ ~ ~ ]
If it were possible to take snapshots of the molecules taking part in the
reaction at different points of the reaction pathway, the relative positions of
the atomic nuclei along the course of the reaction would become known.
The basic idea of the structure correlation method consists in assuming that
the instantaneous displacements of the nuclei during a reaction can be
represented by the images of the reaction centre obtainable from a great
number of crystal structures where the centre itself happens to be deformed
by local inter- and intramolecular perturbations; in this hypothesis, the
different images need only to be ordered in a rational sequence to give the
progressive deformations of the reaction centre along the reaction
coordinate.
This picture of a chemical reaction is directly and even more naturally
applicable to a particular class of chemical transformations, the
conformational interconversions, for which it has already been shown that
the relative conformers' populations are determined by the depth of the
minima on the potential energy hypersurface (the enthalpic term) and by
their shape (the entropic term), while activation barriers for interconver-
sion of conformers are straightforwardly interpretable as heights of the
saddle points on the lowest-energy routes (conformational interconversion
pathways) connecting different conformers on the hypersurface itself.
Some three-centre-four-electron linear systemsr126~128~1341

It is well known that the solubility of iodine in water is increased by the
presence of iodide ions, a fact explained by the formation of polyiodide I;
Fig. 7.20. Correlation diagrams of the

interatomic distances (in A) for three linear
three-centre-four-electron systems: (a) tri-
iodide anions, (b)thiathiophthenes, and (c)
linear 0-H ... 0 hydrogen bonds. (Reproduced
by permission from ~ u r ~ i . ' ' ~ ~ ' )
ions by a charge transfer reaction where the Lewis base I- donates to the
empty n* orbital of I,. Several crystal structures containing the I; ion are
known: the anion is nearly linear with an I,-I, I, angle of 175-180". The
bond distance in the I, molecule is 2.67 A and the sum of the van der Waals
radii is 4.30 A. Actual structures show many interleaving distances, the
shortening of the I, . I, distance being always associated with a lengthen-
ing of the I,-I, one, and vice versa. In general the anion is strongly
asymmetric in presence of small cations and tends to become symmetrical
with larger cations. A plot of the I,-I, versus I, - I, distances is shown in
Fig. 7.20(a). The experimental points are clearly arranged on an equilateral
hyperboloid whose analytic form can be derived from Pauling's formula
(7.31)
Ad = d(n) - d(1) = -c log,, n
where n is the bond number, d(n) and d(1) the bond distances for the bond
numbers n and 1, respectively, and c a constant easily determined as
c = Ad'llog 2, being Ad' the bond length increment for the symmetrical
anion having n = 112 (currently c = 0.85 for a Ad' of 0.26 A). Assuming that
the single bond is shared among the three atoms, it can be written
and the resulting function for c = 0.85 is drawn as a continuous curve in Fig.
7.20(a); only one branch of the hyperboloid is shown since the second one,
obtainable by interchanging I, with 13, is identical because of ion symmetry.
The points on the curve may be supposed to represent the geometrical
changes occurring along the reaction coordinate of the process
The shape of the upper part of the curve indicates that the approach of the
I; ion to the iodine molecule is much more rapid than the lengthening of
514 1 Gastone Gilli
the I,-I, distance up to the point of formation of the activated complex,

whose dissociation will quickly occur along the lower branch of the curve.
From a slightly different point of view, all the curve can be considered to
map the deformations along the antisymmetric stretching vibration of the
activated complex up to its complete dissociation.
Similar correlations have been reported for the groups S-S . . . S (Fig.
7.20(b)) and 0 - H - 0 (Fig. 7.20(c)), respectively found in thiathio-
phthenes (7.111) and in different structures containing nearly linear hydro-
gen bonds. The rule of bond number conservation expressed by (7.43) is
found to hold also in these two cases with parameters d(1) =2.02,
A d 1 = 0.31, and c = 1.03 A for (7.111) and 0.96, 0.26, and 0.85 A for the
hydrogen bond. The curves calculated in this assumption are indicated by
the continuous lines and are seen to approximate well the experimental
data.
Nucleophilic addition to organometallic compounds

Cadmium complexes have been studied by Biirgi in 1973 in the earliest
application of the structure correlation method.[1351These compounds are
normally tetrahedral but can assume quasi-trigonal bipyramidal five coor-
dination by adding a further ligand at a distance greater than that of the
other four ligands (Fig. 7.21). Eleven crystal structures were found having
three sulphur atoms as equatorial ligands and a fifth ligand Y at different
Fig. 7.21. Coordinate system used for cadmium
complexes. Ax and Ay are the differences approaching distances on the normal to the triangular face of the tetra-
between the actual Cd-X and Cd-Y distances hedron and lying on the same line containing the Cd atom and the trans
and the corresponding single bond standard
values. Azis the vertical displacement of the Cd
ligand X. X and Y happen to be iodine, sulphur, or oxygen atoms. The
atom from the S,S,S, plane. system of coordinates used is described in the caption of Fig. 7.21.
By plotting the experimental values of Ax versus Ay (Fig. 7.22) it is found
that the points are arranged on a curve which can be interpreted in terms of
bond number conservation (eqn (7.43)) with Ad' = 0.32 and c = 1.05 A.
Another correlation is obtained by plotting Az as a function of Ay (upper
part of Fig. 7.23) and Ax (lower part). Each structure is represented by two
points having coordinates (Ax, Az) and (Ay, Az); Ax = 0 corresponds to
the fragment CdS3X of tetrahedral geometry, which is represented at the
same time by the point in the left lower corner and by another point at
Ay = a outside the upper right corner. The leaving of X caused by the
approaching of Y moves the two representative points towards the centre,
where they collide at the point Ax = Ay = 0.32 and Az = 0 corresponding to
the trigonal bipyramidal structure.
Figures 7.22 and 7.23 may be interpreted as the geometrical changes
occurring along the reaction of nucleophilic substitution S,2
Fig. 7.22. Correlation diagram for the bond

.
Y+CdS3XS[Y...CdS3..*X]SYCdS3+X
length increments Axand Ay in cadmium
complexes. (Reproduced by permission from
The reaction starts with CdS3X tetrahedral and Y at infinite distance. The
B"rgi,[1351 ) approach of Y causes an increasing lengthening of the trans Cd-X bond and
a flattening of the CdS3 pyramid until trigonal bipyramidal transition state

formation. Afterwards X starts leaving and finally the new tetrahedral
species YCdS3 is formed.
The continuous curve of Fig. 7.23 is calculated by the equations
Ax = -c log [(0.84 + Az)/1.68],
Ay = -c log [(O.84 - Az)/1.68],
making use of the already known value of c = 1.05 A. The two equations
are derived from the usual bond number conservation rule n, n, = 1, +
where n, is the bond number of Cd-X and n, that of Cd-Y, but expressed
as a function of the angle 8, the average X-Cd-S angle. The two relations
n, = (1 - 3 cos 8)/2 and n, = (1 + 3 cos 8)/2 satisfy the rule because for
8 = 90" (trigonal bipyramidal geometry) nx = n, = 0.5 and for 8 = 109.47"
(tetrahedral geometry) nx = 1 and n y = 0. Pauling's formula (7.31) becomes
Ax = -clogn,= -clog [(I -3cos 8)/2],
(7.45)
+
Ay = -c log n y = -c log [(I 3 cos 8)/2],
and, since the equatorial Cd-S bond has a nearly constant value of 2.52 A,
we can put Az = -2.52 cos 8, which gives eqns (7.44) when substituted in
(7.45).
Fig. 7.23. Correlation diagram for the bond
Another case studied is that of the organometallic compounds of length increments A x a n d A y against the vertical
s ~ ( I v ) . [ ' ~Similar
~] methods give a convincing mapping of what could be the displacement Az. (Reproduced by permission
from ~ i i r g i . " ~ ~ ' )
reaction pathway for the S,2 nucleophilic reaction with configuration
inversion
It is interesting to remark, however, that the same paper reports the

mapping of the reaction pathway of the S,3 process
SnR2X2+ 2Y * [SnR2X2Y2]G SnR2Y2+ 2X,
a trimolecular reaction which cannot realistically occur according to the
principles of chemical kinetics. This seems to suggest that structure
correlations do not simply map possible reaction pathways but, more
generally, the lower regions of the potential energy hypersurface for the
relative movements of the nuclei irrespective of the fact that they
correspond or not to real chemical reactions or conformational
rearrangements.
Nucleophilic addition to the carbonyl group

Nucleophilic addition of aminic nitrogen to carbonyl is one of the first and
more studied examples of structure correlation^,['^^^'^^] though the analo-
gous addition of oxygen has been also reported.['39] Molecules where the
nitrogen is forced near the carbonyl carbon at a distance shorter than the
sum of the van der Waals radii are mainly rings of the type (7.IV) and (7.V)
or 1,8-disubstituted naphthalenes (7.VI). In the original paper 14 structures Fig. 7.24. Coordinate system used for describing
having the required characteristic were reported and the coordinate system the approaching of the aminic nitrogen t o the
for their analysis is shown in Fig. 7.24. Final results are summarized in Fig. .
carbonyl. dl and 4 are the N . . C and C==O
distances, respectively: the distance of the
7.25, which describes the course of the geometrical deformations of the carbon atom from the (RR'O) plane is a measure
carbonyl group caused by the approaching of the nucleophile. As this of carbonyl pyrarnidalization and is termed A.
516 ( Gastone Gilli
approaches the carbon (i.e. d l decreases), the plane containing the carbon
and the two R and R' substituents bends away causing an increasing
pyramidalization of the carbon which rehybridizes from sp2 to sp3, while the
C=O distance is slightly increased. It seems of interest that the nucleophile
does not approach the carbonyl plane perpendicularly but makes an almost
constant N . C==O angle of 110°, which has been interpreted in terms of
elementary banana bond considerations but also shown to match the results
of ab initio quantum-mechanical calculations.[41The constraint arising from
Fig. 7.25. The coordinate system is that of Fig. the fixed 110" angle is more clearly seen in (7.VI). The two -NR2 and
7.24 projected on the (NCO) plane. The open
circles indicate the positions of the nitrogen -COR groups would both be expected to be splayed outwards in order to
atoms (top), of the bisector of RCR' (bottom left) reduce their van der Waals repulsions but the C-N bond is actually found to
and of the carbonyl oxygen (bottom right) i n the
14 (A-L) structures considered. (Adapted from
be splayed inwards by the need to maintain the correct approach angle to
Bijrgi and ~ u n i t z . ' ' ~ ~ ' ) the carbonyl C=O bond.
When looking at the dependence (Fig. 7.24) between A (carbonyl group
pyramidaiity) and d l (N . C distance) it is found that they are correlated
according to the regression line
a function which gives the maximum pyramidal character of the carbon

atom, A,, = 0.437 A, for d l = 1.479 A (taken as standard distance for the
C-N single bond). By assuming that the C-N bond number can be
expressed as n = (A/A,,,)~ the above equation changes into
d l = -0.85 log n + 1.479 (7.47)
which is nothing more than the usual Pauling equation (7.31).
A case of conformational rearrangementr140,1411

The fragment c ( s p 2 ) - ~ ( s p ~(7.VII)
) is contained in a variety of organic
molecules, such as anilines (7.VIIIA) and naphthylamines (7.VIIIB),
amides (7.VIII 0 ) and thioamides (7.VIII S) (X = 0 , S), amidines
(7.VIII N) (X = NR), and enamines (7.VIII D) (X = CR,). It displays a
definite tendency to be planar, which can be accounted for by the
contribution of the polar form (7.VIIa) to its ground state. The cis-trans
isomerization barrier has been measured by a number of dynamic NMR
experiments to be in the range 5-22 kcalmol-' (on average, 20.7, 18.1,
12.8, 9.0, and 5.1 kcal mol-' for thioamides, amides, amidines, enamines,
J-x)
Molecules and molecular crystals
and anilines respectively, in agreement with the fact that the contribution of
the polar form decreases with the decrease of electronegativity of X). In
spite of the general tendency to planarity, there are many crystal structures
c
Z
+
Xc
+
1
XN
517
where the group is deformed by rotation around the C-N bond or by

nitrogen pyramidalization, or by both, mainly in consequence of in- - -
tramolecular steric hindrance. The structure correlation method was used
with the aim of mapping the coordinate of cis-trans isomerization of the Fig. 7.26. lnternalcoordinatesystem describing
the out-of-plane deformation of the c ( s p 2 ) - ~ ( s p ~ )
group. fragment. (Reproduced by permission from Gilli
A coordinate system for the out-of-plane deformation can be obtained by et al.[14")
the usual methods of group theory; the resulting orthogonal coordinates are
the angle of twisting around the C-N bond, t ( t = 0 and 180" for the planar
cis and trans geometries, t = f90" when the two halves of the group are
perpendicular) and the two out-of-plane bendings of nitrogen, x,, and
carbon, x,, ranging from zero (sp2 atom) to 60' (sp3 atom). Calling $,, 4,)
$31 and $, (Fig. 7.26) the four torsion angles 1-2-3-4, 5-2-3-6, 5-2-3-4,
and 1-2-3-6, it is found that[142,143]
A total of 90 fragments belonging to 68 crystal structures were found from

CSD.[~]The xc values are always very small (on average 1.7"), indicating
that the carbon sp2 has greater resistance to out-of-plane bending and that
this motion can be neglected. Conversely X, shows a wide variation
assuming values up to f55". Figure 7.27 shows the correlation diagrams of
Fig. 7.27. Correlation diagram of X , versus r for

anilines (A), naphthylamines (B), enamines (D),
and amidines (N). Data for amides and
thioamides (not shown in the figure) would
cluster in the lower left corner with r s 15" and
,yN S 27'. (Reproduced by permission from Gilli
et
518 1 Gastone Gilli
xN against z for all the chemical classes except amides and thioamides.
These latter cluster in the lower left corner with t c 1.Y and X, S 27",
showing the greater resistance to out-of-plane deformation of these com-
pounds in agreement with the larger electronegativities of the oxygen and
sulphur atoms.
Compounds of the other classes undergo more severe distortions which
appear to be of two different types. The first, producing simple nitrogen
pyramidalization, occurs for t nearly equal to zero and for increasing values
of xN and can be called a butterfly deformation. The second, which could
be called combined, associates nitrogen pyramidalization (increasing xN)
with a rotation around the C-N bond (increasing z). The diagonal straight
line of Fig. 7.27 ideally separates mostly butterfly (on the left) from mostly
combined (on the right) distortions.
The combined motion admits a simple interpretation. In the planar
fragment the nitrogen is sp2 hybridized and its p, A 0 is implies in a n bond
with the p, A 0 on the carbon. The rotation around the C-N bond causes a
decoupling of the n system while the nitrogen rehybridizes engaging its p,
A 0 into an sp3 HO carrying the lone pair. The opposite mechanism ia also
possible: the planar nitrogen undergoes an out-of-plane vibration (butterfly
motion) which causes the rehybridization of the atom from sp2 to sp3. The
sp3 HO on nitrogen is essentially decoupled from the p, A 0 on carbon, the
double bond fades, and the rotation around C-N becomes possible. This
second mechanism seems to be preferred because Fig. 7.28 shows that XN (a
measure of the nitrogen rehybridization) is strongly correlated with a
lengthening of the C-N distance which almost encompasses the full
variation range of dC-N (from 1.27 to 1.44 A going from double to single
C(sp2)-N(sp3) bond), while no similar correlation can be established with z,
the rotation angle around the C-N bond.[1411
From the point of view of the structure correlation method the butterfly
motion corresponds to a simple out-of-plane vibration and does not produce
Fig. 7.28. Correlation diagram between the &-,

bond distances and the x, values for all
fragments studied (A = anilines, B =
naphthylarnines, D = enarnines, N = arnidines,
0 = amides, and S = thioamides). (Reproduced
by permission from Gilli et
Molecules and molecular crystals ( 519
Fig. 7.29. The relative movements of the

fragment atoms along the reaction pathway
identified by the broken curve of Fig. 7.27,
showing the steps leading from the planar
.ground state Toto the two enantiomeric
transit~onstates T, and TZ. Further rotation
around the C-N bond would transform the cis
isomer on the lee into the trans isomer on the
right (not shown). (Reproduced by permission
from Gilli et at.[14")
any chemical change. Conversely, the combined motion can be considered

to be representative of the geometrical changes experienced by the fragment
along the pathway of a cis-trans isomerization reaction. The broken curve
of Fig. 7.27 maps the most probable pathway leading from the planar
ground state To, having t = X, = 0°, to the transiton state located at
( t , x,) = (90, 4 0 " ) ; since the rotation can occur in two opposite directions,
two enantiomeric TI and T2 transition states will be equally possible and
Fig. 7.29 shows the geometrical changes the fragment is undergoing along
the two enantiomeric pathways indicated by the broken curve. It is seen that
the atom movements start with an out-of-plane deformation of the nitrogen
and are followed by the rotation around the C-N bond.
It seems of great interest to show whether the proposed reaction pathway
does actually map a relative minimum path on the potential energy
hypersurface. What is needed is an energy model for the out-of-plane
motion of the c(sp2)-N(SP~)fragment. A rather simple molecular mechani-
cal model has been proposed11411by modification of a first model given by
Winkler and ~ u n i t z , [ lthat
~ ~ ]is
V ( T ' ,xN)= (CTIB + IB)(l- cos t r ) / 2
4- QP(1-I- COS tr)&/2 d- IB(1- COS Z')(COS3xN- 1)/4, (7.49)
where the total potential energy expressed as a function of t ' = 22 and X, is

assumed to depend on three predetermined parameters, CTIB = cis-trans
isomerization barrier, ZB =inversion barrier of the pyramidal sp3 nitrogen,
and QP = force constant for the out-of-plane bending of the sp2 nitrogen.
The energy is zero at the origin in correspondence to the planar geometry
and increases with X& for t' = 0" along the out-of-plane vibration of the
planar nitrogen; for x,=O and 2' = 180" the function has the value
+
CTIB IB, differing from CTIB for the energy needed to make planar the
sp3 nitrogen, while it assumes the exact value of CTIB only for the
hypothetical transition state geometry ( t ' = 180" and X, = 60"). The two
+
expressions (1 cos t ' ) and (1 - cos t ' ) are introduced in order to progres-
sivelv cancel the second term of (7.49) re~lacineit bv the third term, while
520 1 Gastone Gilli
z' changes from zero to 180' and, accordingly, the nitrogen hybridization
from sp2 to sp3.
The values of the constants have been chosen as average values from
those of the different classes of compounds. The CTIB value is in the range
5-22 kcal mol-', QP is evaluated from vibrational spectroscopy and mole-
cular mechanics to be some 5-10 kcal mol-' radP2, and IB can be assimi-
lated to the inversion barrier of ammonia, usually reported as =6
kcal mol-l. Values chosen were CTIB = 14, IB = 8 kcal mol-', and QP =
8 kcal mol-I rad-2. Equation (7.49), however, evaluates the position of the
xN
transition state at t ' = 180' and = 60", which does not correspond to that
found from Fig. 7.27 as far as xN is concerned. The reason is that some
allowance must also be made for the 1-4 non-bonded interactions, that is
where En,,can be obtained by the methods of molecular mechanics. Here it

has been calculated assuming R, = R, = R, = methyl and X equal to a
nonexistent atom intermediate between oxygen and methyl group. The final
V' function calculated in such a way is shown in Fig. 7.30 and represents the
potential energy hypersurface for an imaginary molecule which is the
average of all different compounds of all chemical classes investigated. In
spite of the approximations made, it has the correct shape and the
agreement with experiment becomes more convincing when the experimen-
tal points are plotted on it. They appear to be nicely located along the
energy valley leading from the planar ground state to the transition state,
which is now more correctly located at t' = 180" and X, = 50".
Fig. 7.30. Total potential energy (kcal mol-')

surface as calculated from eqn (7.50) as a
function o f t = t1/2and x,. Experimental points
include all compounds except amides and
thioamides. (Reproduced by permission from
Gilli et
Resonance assisted hydrogen bonding (RAHB)

6-diketones are known to undergo easy enolization and in two recent
structure correlation methods have been applied to the
P-diketone fragment in its en01 form (7.IXa,b) in order to understand what
happens to the fragment geometry when perturbed by intramolecular,
(7.X), or intermolecular, (7.XI), hydrogen bonding. The reason why this
particular system was studied originates from the empirical observation that the
n-conjugated system present in the 6-diketone en01 HOCR=CR-CR=O
fragment undergoes greater delocalization when it forms either intra-
molecular or infinite chain intermolecular hydrogen bonding. The nature
and the entity of this effect is shown in Table 7.9. Here the standard
distances are those tabulated['091 for pure single and double bonds, the
unperturbed ones are an average of nine fragments that cannot make
H-bonds (having OR instead of OH). The unperturbed geometry is a 87:13
mixture of the resonance forms (7.IXa) and (7.IXb) according to Pauling's
formula (7.31). However, when the fragments form H-bonds all distances
undergo changes consistent with an increased contribution of the polar form
(7.IXb), which becomes respectively 29 per cent for the intermolecular and
48 per cent for the intramolecular cases reported in the table; at the same
time the contract 0--0 distance, which is typically 2.74 A in the 0 - H . 0
bond in ice, becomes much shorter, being respectively 2.575 and 2.485 A in
Table 7.9. Selected bond distances (A) for the P-diketone enol fragment. %(7.IXb) = per
cent contribution of the polar form (7.IXb) according to Pauling's formula (7.31); the
standard 0--0 contact distance in parentheses is that observed in ice
Perturbed by:
Intermolecular
H-bond (7.XIIa 1.316 1.372 1.431 1.238 2.575 29
Intramolecular
H-bond (7.X)b 1.281 1.398 1.410 1.279 2.485 48
aNeutron data from Jones, R. D . G. (1976). Acta Crystallographica, 8,32,2133.

bSernrningsen, D . (1977). Acta Chem. Scand. Sect. 8, 31, 114.
522 ( Gastone Gilli
the two cases, while the 0 - H bond distance (0.95 A in the absence of
H-bonding and according to Table 7.7) is lengthened up to 1.24 A in the last
compound of Table 7.9.
In a first 25 X-ray or neutron structures of p-diketones and
P-ketoesters were studied, 22 of type (7.X) and 3 of type (7.XI), and the
discussion reported here concerns this first analysis. Later on[2591491 the
investigation was extended to a much larger set of fragments (81 and 37 for
(7X) and (7.XI), respectively) but without obtaining significant differences
in the final results. In summary, all the following phenomena are observed
to occur together:
(1) increased delocalization of the n-conjugated system;
(2) strengthening of the 0 - H - . 0 bond as indicated by the shortening of
do--, and the lengthening of d,, distances;
(3) shift of the proton position towards the middle point of the 0 . . . 0
contact.
The three effects may occur with different intensities but maintain the
same intercorrelation for both intramolecularly (7.X) or intermolecularly
(7.XI) bonded fragments (the intramolecular H-bond is usually stronger and
causes greater n delocalization) and even for different classes of compounds
(e.g. P-diketone enols form stronger H-bonds and are more delocalized
than the enols of P-ketoesters or /3-ketoamides).
While the H-bond strengthening is easily measured by both d o - - , and
dSH changes, a specific geometric parameter is needed in order to describe
the n system delocalization. The simplest way consists in using symmetry
coordinates for the in-plane antisymmetric vibration of the group (7.X) or
(7.XI), that is q , = d4 - d l and q , = d , - d,. Clearly q , = q , = 0 for the
totally delocalized structure, while the greatest values will occur for the
standard bond distances of Table 7.9. Figure 7.31 reports the scatter plot of
s
4'I
N
-a
0
'y
?4
k .
U
It
N .,A' INTRA HE
-.05- /
*
o INTER HB
a&/
' 0"
,
' AVERAGE UNPERTURBn

Fig. 7.31. Scatter plot of q, versus q, (in A) for
-.lo- STANDARD
the 25 fragments studied. The plot is
centrosymmetric as any point can be plotted /
/
twice in view of the equivalence of the enol-
ketonic (EK) and keto-enolic (KE) isomers. The
full squares and stars refer t o the standard and
unperturbed geometries of Table 7.9,
.
respective1 (Reproduced by permission from
Gilli et al!2'1)
q2 versus ql for the 25 compounds studied. The plot is centrosymmetric as

any point can be plotted twice in view of the equivalence of the enol-ketonic
(EK) and keto-enolic (KE) isomers, and its centre identifies the fully
delocalized structure. The full squares and stars refer to the standard and
unperturbed geometries of Table 7.9, respectively. The broken line con-
necting EK and KE is calculated from the condition of bond number
conservation (p. 513). Since q1 and q2 are linearly dependent, a single
coordinate Q = q1+ q2 can be used, which has values of 0.0, -0.320, and
+0.320 A for the fully n-delocalized and the fully n-localized EK and KE
structures, respectively. Alternatively the fragment geometry can be de-
scribed by a coupling parameter, A, according to which the state of any
fragment is a mixture A(EK) + (1- A)(KE); it assumes the values of 1.0 for
EK, 0.0 for KE, and 0.5 for the fully delocalized form.
The scatterplot of do--o against Q = ql + q2 for the 25 fragments
considered is reported in Fig. 7.32. The plot displays m symmetry with
respect to the delocalized structure having Q = 0 or A = 0.5 owing to the
equivalence of the EK and KE structures. Very short 0 . . - 0 distances are
seen to be associated with small values of lQl, that is with relevant
delocalization of the n-conjugated system. The plot of the variations of the
dSH and of the contact do,-, distances against ds-o is shown in Fig.
7.33. Although data are necessarily more disperse owing to the known
difficulties in locating protons by X-ray diffraction (only three were neutron
structures), the plot clearly indicates that the shortening of the 0 . . . 0
distance below 2.5 A causes the two bond (0-H) and contact (OH . . . 0 )
distances to become progressively equal, though a perfectly centred
positioning of the proton is never achieved.
The observed correlations are in agreement with the model which, in its
qualitative aspects, is illustrated in Fig. 7.34. Let us assume firstly that the
double C=C is substituted by a single C-C bond; the 0 - H . . 0 bond
established will be the balance between two different factors: the energy of
Fig. 7.32. Scatter plot of do--, versus Q. The

2.4- plot has rn symmetry because of the
t I I b equivalence of EK and KE structures. Full and
-.32 -.20 -.lo 0 .lo .20 -32 open circles indicate intra- and intermolecular
H-bonds, respectively. (Reproduced by
permission from Gilli et
524 ( Gastone Gilli
Fig. 7.33. Scatter plot of &, and do,--,

distances against the contact do--, distance.
the H-bond itself (EHB)and the van der Waals energy originated from the
attractive and repulsive terms of the 1-4 interactions (Evdw). Re-
establishing now the double C=C bond, the resonance (7.IXa) tt (7.IXb)
will cause a neat shift of electrons from left to right (Fig. 7.34), which will
+
stop when the minimum of ERES E B P is attained, where the first term is
the resonance energy gain and the second the energy needed to dissociate
the opposite partial charges on the terminal oxygens. The charges have the
correct sign for strengthening the H-bond with consequent shortening of
d o - - , and lengthening of d., The movement of the proton to the right is
equivalent to a negative charge going to the left and the global effect is the
annihilation of the partial charges initially set up by resonance, so allowing
an increased contribution of the polar forrn (7.IXb) and a further
strengthening of the H-bond, an imaginary process going on until the
Fig. 7.34. A graphical representation of the minimum of the function
resonance assisted hydrogen bonding (RAHB)
model. (Reproduced by permission from Gilli e t E = E H B+ E v m+ ERES+ EBP (7.51)
is attained. The H-bond formed has quite specific features and, in view of
the strict interplay of n delocalization and H-bond strengthening which
causes it, has been called[241resonance assisted h y d r ~ g e n bonding
(RAHB). It is essentially a charge assisted H-bond where, however, the
charges do not arise from the presence of ions but from the resonance in a
heteroconjugated system.
The RAHB model is supported by a wealth of spectroscopic data and
theoretical calculations for which the reader is addressed to the original
paper. What is more important here is to discuss whether the correlation
among geometrical parameters can be interpreted in terms of the potential
energy hypersurface of the fragment. The approximate energy partitioning
used in (7.51) can be evaluated quantitatively since semi-empirical equa-
tions are available for all the four terms. E H Bis the total energy, including
both attraction and repulsion terms, of the 0 - H . . . 0 bond as a function of
dH
, = r and d o _ - , = R. It can be written as E d r ., R ) and has been
\
calculated by the equation proposed by Lippincott and ~ c h r o e d e r [ ~ for

~,~*]
all the desired values of r at the R value for which the energy was a relative
minimum. In such a way R turns out to be a function of r and is known for
any value of r. ERE, can be calculated by the method proposed by
Krigowski et a1.[1441which is known to give the resonance energy of a
n-conjugated system from its bond distances with fairly good accuracy. EBp
is the energy required to create the opposite fractional charges fq on the
two terminal oxygens and can be easily evaluated by means of the
coefficients of the atomic ionization energy versus electron affinity curves
tabulated by Hinze and ~ a f f 2 for [ ~the~ main
~ ~ elements.
~ ~ ~ ~Finally, EVdw
can be calculated by the methods and equations already discussed in the
section on molecular mechanics.
The final potential energy map calculated according to (7.51) for
acetylacetone (R, = R, = CH,, R, = H) is shown in Fig. 7.35 as a function of
two coordinates: the bond number of the 0 - H bond, n(0-H) (which is 1.0
in the absence of H-bonding when d,, assumes the value of 0.97 hl, and
0.5 when the hydrogen is equally shared by the two oxygens and
,d
, = do,-, = 1.25 hl), and the coupling parameter A (which assumes the
values of A = 0, 0.5, and 1 for n-localized EK, fully n-delocalized, and
n-localized KE fragment structures, respectively) or its related quantity Q.
Figure 7.36 illustrates the physical meaning that can be given to the four
corners and the central point of the plot of Fig. 7.35 while Fig. 7.37 shows a
three-dimensional representation of the potential energy map in the same
coordinate system.
The potential energy map is centrosymmetric in consequence of the
fragment symmetry and displays a diagonal valley of lower energy which can
Fig. 7.35. Total potential energy map (kJ mol-')

calculated according t o eqn (7.51) for
acetylacetone ( R , = R, = CH,, R, = H) as a
function of the coupling parameter A and the
n(0-H) bond number. The physical meaning of
the four corners and of the cental point of the
plot are illustrated i n Fig. 7.36. The star indicates
the gas electron diffraction structure of
acety~acetone['~"and the full points the X-ray or
neutron structures analysed. (Heproduced by
PROTON TRANSFER permission from Gilli et
526 1 Gastone Gilll
*PROTON TRANSFER-
I J
0 FORMAL 0.5 CHARGE 1
Fig. 7.36. Geometries assumed by the fragment

in the five positions of the energy map of Fig.
7.35corresponding to the four corners and to
the central point. The diagonal line is a zero-
charge line for the fragment because it is the
locusof points for which the partial charges
generated by resonance are exactly
compensated by the shift of the proton.
be considered as the path of the reaction of proton transfer transforming the

EK form into its KE isomer. Figure 7.38 shows the geometrical changes
which the fragment undergoes along the reaction pathway and Fig. 7.39
reports the energy profiles along the same path. The curve A' has been
calculated without taking into account the two terms related to resonance
(ERE,and EBp);it is essentially representative of H-bonding in water and,
in fact, it reproduces quite well the water distance values (r = 1.01 A,
R = 2.74 A) and of the proton transfer barrier (some 40 kJ mol-l). The
curve A is calculated for acetylacetone by the use of all terms of (7.51)
(Figs. 7.35 and 7.37); the introduction of the resonance term lowers the
total energy by 42.5 kJmol-l, increasing the H-bond energy (already
20 kJ mol-' in water) to the relevant value of 62.5 k3 mol-l; at the same
time the proton transfer barrier is reduced to 34.8 kJ mol-' and the 0 - H
distance is lengthened until r = 1.08 A (associated R = 2.50 A).
The structure of acetylacetone has been recently determined by gas
electron diffra~tion"~''and the corresponding geometry, marked by a star in
Fig. 7.35, is seen to be very close to the calculated energy minimum. In the
same figure are reported as full points the geometries of all the 25 fragments
investigated. It seems interesting that all of them are correctly located inside
the diagonal energy valley though, of course, not in the minimum, which
has been specifically calculated for acetylacetone. As far as their exact
position is concerned two main factors can be discerned. The first is
electronic, and we have already remarked that P-diketone enols form much
stronger H-bonds than the enols of P-ketoesters or P-ketoamides. The
Fig. 7.37. Three-dimensional representation of second is steric in nature, the strongest H-bonds being associated with bulky
the potential energy map of Fig. 7.35 in the R, and R, substituents. For example, strictly similar calculations carried out
same coordinate system. The path for the
reaction of proton transfer transforming the EK On hexamethylacetylacetone by the use of (7.51) show that the energy
into the KE isomer is clearly shown as an energy minimum is shifted to r = 1.11 and R = 2.48 A, while the H-bond energy
valley whose saddle point corresponds to the
fully n-delocalized form having the proton in the increases to 82.5 kJ mold' and the proton transfer barrier decreases to
middle of the two oxygens. 27.5 kJ m0l-I.
Resonance assisted hydrogen bonding (RAHB) is therefore a mechanism H

of synergistic interplay of resonance and H-bond formation. It can be
generalized by calling it a mechanism of interplay between H-bond and
resonance in heterodienes or, more generally, heteroconjugated systems:
Fig. 7.38. Geometrical deformations

experienced by the fragment along the pathway
of the proton transfer transforming the EK into
the KE isomer.
7. Xll 7.Xlll
From this point of view, the P-diketone enols so far discussed correspond to
the heterodienic scheme (7.52) (X = Y = O), but the RAHB mechanism is
supposed to operate independently of the number of interleaving carbon
atoms provided the conjugation between the lone pair on X and the C=Y
double bond is maintained. In the X = Y = 0 series the scheme (7.55)
represents carboxylic acids which, in fact, are well known to give strongly
H-bonded dimers (Fig. 7.3, scheme b), though a more convincing proof of
RAHB could come from the observation of strongly delocalized and
H-bonded heterotrienes (7.53) and heterotetraenes (7.54). Structures cor-
responding to these fragments have been recently analysed[251and found to
display the expected behaviour. Compounds (7.XII) behave as totally
Fig. 7.39. Energy profile along the reaction

coordinate of proton transfer transforming EK
into KE as mapped along the mainly diagonal
pathway of lowest relative energy of the map of
Fig. 7.35 (curve A). Curve A' is the same curve
but calculated omitting the ERE,and E,, terms
i n eqn (7.51); it can be considered t o represent
the proton transfer in water. The horizontal lines
mark the zero-point energies which are slightly
different i n water and acetylacetone.
528 1 Gastone Gilli
7. X I V 7. x v 7.XVI 7.XVll
"k
8--
N-H--0
/
7. X I X I
1 1
7.XXlll
n-delocalized heterotetraenes with bond 5 not participating to the conjuga-

tion; the opposite bonds n and n ' (n = l, 2, 3, and 4) are nearly equal and
the 0 - H . . 0 is extremely short (on average, do--, = 2.47 and
do--, = 1.18 A). In a strictly similar way compounds (7.XIII) show total
delocalization of the heterotrienic system and display one of the shortest
d o - - , distances ever observed, i.e. 2.425 A.
Up to now we have proved that RAHB is active in the enolone fragment
(7.XIV) (or in other diketone enols with longer conjugated systems)
forming intramolecular or, sometimes, intermolecular H-bonds. All these
compounds have the oxygen as the only heteroatom. However, other
heteroconjugated systems having different heteroatoms are putatively im-
plied in this phenomenon via intramolecular or intermolecular (including
dimerization) H-bond formation, such as enaminones (7.XV), enaminoim-
ines (7 .XVI) , enolimines (7 .XVII) , amide-amidine complexes (7 .XVIII)
and amide dimers (7.XIX). Preliminary data confirming the strong n
delocalization of H-bonded enaminones (7.XV) and enolimines (7.XVII)
have already been reported,[1481and the formation of amide dimers (7.XIX)
has been discussed by several a ~ t h o r s . [ ~ ~ , ~ ~ ]
Possible biochemical and biological implications originate from the fact

that the amide-amidine complex (7.XVIII) is present in both thymine-
adenine (7.XX) and cytosine-guanine (7.XXI) PI-bonded couples deter-
mining the double-helix structure of DNA; in addition, cytosine-guanine
coupling (7.XXI) implies another much wider cycle of H-bonds and
mconjugated double bonds. Strictly similar considerations can be made in
connection with the H-bonds determining the secondary structure of
proteins: both a-helices (7.XXII) and 0-pleated sheets (7.XXIII) contain
long chains of n-heteroconjugated systems connected by H-bonds and differ
only because each a-helix contains three of such chains which are nearly
parallel and isoriented, while in 0-pleated sheets the arrangement of the
chains is antiparallel.
References
1. Wells, A. F. (1975). Structural inorganic chemistry (4th edn). Clarendon,
Oxford.
2. Bondi, A. (1964). Journal of Physical Chemistry, 68, 441.
3. Allen, F. H . , Bellard, S., Brice, M. D . , Cartwright, B. A., Doubleday, A.,
Higgs, H., Hummelink, T., Hummelink-Peters, B. G., Kennard, O., Mother-
well, W. D. s . , Rodgers, J. R., and Watson, D. G. (1979). Acta
4. Dunitz, J. D. (1979). X-ray analysis and the structure of organic molecules.
Cornell University Press, Ithaca, N.Y.
5. Pauling, L. (1960). The nature of the chemical bond and the structure of
molecules and crystals. A n introduction to modern structural chemistry (3rd
edn). Cornell University Press, Ithaca, N.Y.
6. Kitaigorodski, A. I. (1973). Molecular crystals and molecules. Academic, New
York.
7. Kaplan, 1. G. (1986). Theory of molecular interactions. Elsevier, Amsterdam.
8. Pertsin, A. J., and Kitaigorodski, A. I. (1987). The atom-atom potential
method. Springer, Berlin.
9. London, F. (1930). Zeitschrift fur Physikalische Chemie, B11, 222, 236; (1930).
Zeitschrift fur Physik, 63, 245.
10. Lennard-Jones, J. E. (1931). Proceedings of the Physical Society, 43, 461.
11. Buckingham, R. A. (1938). Proceedings of the Royal Society of London, A168,
264.
12. Giglio, E. (1969). Nature, 222, 339.
13. Allinger, N. L. (1975). MMlIMMP1, QCPE No. 318. Indiana University.
14. Hamilton, W. C., and Ibers, J. A. (1968). Hydrogen bonding in solids.
Benjamin, New York.
15. Pimentel, G . C., and McClellan, A. L. (1960). The hydrogen bond. Freeman,
San Francisco; Pimentel, G. C., and McClellan, A. L. (1971). Annual Reviews
of Physical Chemistry, 22, 347.
16. Kollman, P. A , , and Allen, L. C. (1972). Chemical Reviews, 72,283; Kollman,
P. A. and Joesten, M. D. (1982). Journal of Chemical Education, 59, 362.
17. Schuster, P., Zundel, G., and Sandorfy, C. (ed.) (1976). The hydrogen bond,
Vols. I, 11, and 111. North-Holland, Amsterdam.
18. Emsley, J. (1980). Chemical Society Review, 9, 91.
19. Umeyama, H., and Morokuma, K. (1977). Journal of the American Chemical
Society, 99, 1316.
20. Morokuma, K. (1971). Journal of Chemical Physics, 55, 1236.
530 1 Gastone Gilli
21. Kitaura, K., and Morokuma, K. (1976). International Journal of Quantum

Chemistry, 10, 325.
22. Taylor, R., and Kennard, 0. (1982). Journal of the American Chemical
Society, 104, 5063.
23. Gorbitz, C. H . (1989). Acta Crystallographica, B45, 390.
24. Gilli, G., Bellucci, F., Ferretti, V., and Bertolasi, V. (1989). Journal of the
American Chemical Society, 111, 203.
25. Gilli, G., and Bertolasi, V. (1990). In The chemistry of enols (ed. Z .
Rappoport). Wiley, New York. p. 713.
26. Bent, H. A. (1968). Chemical Reviews, 68, 587.
27. Gutmann, V. (1978). The donor-acceptor approach to molecular interactions.
Plenum, New York.
28. Kitaigorodski, A. I. (1984). Mixed crystals. Springer, Berlin.
29. Huheey, J. E. (1983). Inorganic chemistry (3rd edn). Harper and Row, New
York.
30. Ibers, J. A., Pace, L. J., Martinsen, J., and Hoffman, B. M. (1982). Structure
and Bonding, 50, 1.
31. Del Re, G. (1958). Journal of the Chemical Society, 4031.
32. Gasteiger, J., and Marsili, M. (1980). Tetrahedron, 36, 3219.
33. Berkovitch-Yellin, Z., and Leiserowitz, L. (1982). Journal of the American
Chemical Society, 104, 4052.
34. Lifson, S., Hagler, A. T., and Dauber, P. (1979). Journal of the American
35. Dauber, P., and Hagler, A. T. (1980). Accounts of Chemical Research, 13,
105.
36. Spackman, M. A. (1986). Journal of Chemical Physics, 85, 6579, 6587.
37. Lippincott, E. R., and Schroeder, R. (1955). Journal of Chemical Physics, 23,
1099.
38. Schroeder, R., and Lippincott, E . R. (1957). Journal of Physical Chemistry,
61, 921.
39. Born, M., and von Karman, T. (1912). Physikalische Zeitschrift, 13, 297.
40. Born, M., and Huang, K. (1954). Dynamical theory of crystal lattices.
Clarendon, Oxford.
41. Willis, B. T. M., and Pryor, A. W. (1975). Thermal vibrations in crystal-
lography. Cambridge University Press.
42. Califano, S., Schettino, V., and Neto, N. (1981). Lattice dynamics of molecular
crystals, Lecture Notes in Chemistry, Vol. 26. Springer, Berlin.
43. Pawley, G. S. (1970). In Crystallographic computing (ed. F. R. Ahmed).
Munksgaard, Copenhagen. p. 243.
44. Kitaigorodski, A . I. (1970). In Advances in structure research by diffraction
methods, Vol. 3, (ed. R. Brill and R. Mason). Pergamon, Oxford. p. 173.
45. Cahn, R. S., and Ingold, C. K., (1951). Journal of the Chemical Society, 612.
46. Cahn, R. S., Ingold, C. K., and Prelog, V. (1956). Experientia, 12, 81.
47. Klyne, W., and Prelog, V. (1960). Experientia, 16, 521.
48. Mislow, K. (1965). Introduction to stereochemistry. Benjamin, New York.
49. Hallas, G. (1965). Organic stereochemistry. McGraw-Hill, New York.
50. Cahn, R. S., Ingold, C. K., and Prelog, V. (1966). Angewandte Chemie
International Edition in English, 5, 385.
51. Eliel, E. L., Allinger, N. L., Angyal, S. J., and Morrison, G. A. (1966).
Conformational analysis. Wiley, New York.
52. Natta, G., and Farina, M. (1968). Stereochimica. Mondadori, Milan.
53. IUPAC 1968 Tentative rules, section E , fundamental stereochemistry,
Information Bulletin No. 35; (1970). Journal of Organic Chemistry, 35, 2849.
54. Stoddard, J. F. (1971). Stereochemistry of carbohydrates. Wiley, New York.
55. Balaban, A. T. (ed.) (1976). Chemical applications of graph theory. Academic,
New York.
Molecules and rnolecular crystals 1 531
Bijvoet, J. M., Peerdeman, A. F., and van Bommel, A. J. (1951). Nature, 168,
271.
Rogers, D. (1975). In Anomalous scattering (ed. S. Ramaseshan and S. C.
Abrahams). Munksgaard, Copenhagen. p. 231.
Cotton, F. A. (1971). Chemical applications of group theory (2nd edn). Wiley,
New York.
Cano, F. H., Foces-Foces, C., and Garcla-Blanco, S. (1977). Tetrahedron, 33,
797.
Cremer, D. and Pople, J. A. (1975). Journal of the American Chemical Society,
97, 1354.
Boeyens, J. C. A. (1978). Journal of Crystal and Molecular Structure, 8, 317.
Altona, C., Geise, H. J., and Romers, C. (1968). Tetrahedron, 24, 13.
Parkanyi, L. (1980). RING. Hungarian Academy of Sciences, Budapest.
Nardelli, M. (1982). PARST. University of Parma, Italy.
Clark, T. (1985). A handbook of computational chemistry. Wiley, New York.
Hehre, W. J., Radom, L., Schleyer, P. v. R., and Pople, J. A., (1985). Ab
initio molecular orbital theory. Wiley, New York.
Pople, J. A., and Beveridge, D. L. (1970). Approximate molecular orbital
theory. McGraw-Hill, New York.
Dewar, M. J. S. (1969). The molecular orbital theory of organic chemistry.
McGraw-Hill, New York.
Murrell, J. N., and Harget, A. J. (1972). Semiempirical self-consistent-jield
molecular orbital theory of molecules. Wiley, London.
Frisch, M. J., Head-Gordon, M., Schlegel, H. B., Raghavachari, K., Binkley,
J. S., Gonzalez, C., et al. (1988). GAUSSIAN 88. Gaussian Inc., ~ittsburgh,
PA.
Bingham, R. C., Dewar, M. J. S., and Lo, D. H. (1975). Journal of the
American Chemical Society, 97, 1285, 1294, 1302, 1307, 1311; Dewar, M. J. S . ,
and Thiel, W. (1977). Journal of the American Chemical Society, 99, 4899,
4907.
Dewar, M. J. S., Zoebisch, E. G . , Healy, E. F., and Stewart, J. J. P. (1985).
Journal of the American Chemical Society, 107, 3902.
Coulson, C. A., (1961). Valence. Oxford University Press, London.
McWeeny, R. (1979). Coulson's valence. Oxford University Press, London.
75. Streitwieser, A. Jr. (1961). Molecular orbital theory for organic chemists.
Wiley, New York.
76. Gillespie, R. J. (1972). Molecular geometry. Van Nostrand-Reinhold, London.
77. Gillespie, R. J. (1963). Journal of Chemical Education, 40,295; (1970). 47,18.
78. Bader, R. F. W., Gillespie, R. J., and MacDougall, P. J. (1988). Journal of the
79. Orgel, L. E., (1966). An introduction to transition-metal chemistry (2nd edn).
Wiley, New York.
80. Bethe, H. (1929). Annalen der Physik, 3, 135.
81. Van Vleck, J. H. (1932). Physical Review, 41,208; (1935). Journal of Chemical
Physics, 3, 803, 807.
82. Ballhausen, C. J., (1962). Introduction to ligand field theory. McGraw-Hill,
New York.
83. Dunn, T. M., McClure, D. S., and Pearson, R. G. (1965). Some aspects of
crystalfield theory. Harper and Row, New York.
84. Schaffer, C. E., and J@rgensen,C. K. (1965). Molecular Physics, 9, 401.
85. Schaffer, C. E. (1973). Structure and Bonding, 14, 69.
86. Burdett, J. K. (1978). Chemical Society Review, 7, 507.
87. Burdett, J. K., (1980). Molecular shapes. Theoretical models of inorganic
stereochemistry, p. 23. Wiley, New York.
88. Walsh, A. D. (1953). Journal of the Chemical Society, 2260, 2266, 2288, 2296,
2301, 2306.
532 1 Gastone Gilli
89. Gimark, B. M. (1979). Molecular structure and bonding. Academic, New

York.
90. Lowe, J. P. (1978). Quantum chemistry. Academic, New York.
91. Woodward, R. B., and Hoffmann, R. (1970). The conservation of orbital
symmetry. Academic, New York.
92. Fujimoto, H., and Fukui, K. (1974). In Chemical reactivity and reaction paths
(ed. G. K. Klopman). Wiley, New York.
93. Elian, M., and Hoffmann, R. (1975). Inorganic Chemistry, 14, 1058.
94. Hoffmann, R. (1981). Science, 211, 995.
95. Hoffmann, R. (1963). Journal of Chemical Physics, 39, 1397.
96. Jahn, H. A., and Teller, E . (1937). Proceedings of the Royal Society of
London, A161, 220.
97. Jotham, R. W., and Kettle, S. F. A. (1971). Inorganic Chimica Acta, 5, 183.
98. Gaio, J., Bersuker, I. B., Garay, J., Kabesova, M., Kohout, J., Langfel-
derova, H . , Melnik, M., Serator, M., and Valach, F. (1976). Coordination
Chemistry Reviews, 19, 253.
99. Hathaway, B., Duggan, M., Murphy, A., Mullane, J., Power, C., Walsh, A.,
and Walsh, B. (1981). Coordination Chemistry Reviews, 36, 267.
100. Pearson, R. G. (1969). Journal of the American Chemical Society, 91, 1252,
4947; (1970) Journal of Chemical Physics, 52, 2167; 53, 2986.
101. Bartell, L. S. (1960). Journal of Chemical Physics, 32, 827; (1968). Journal of
Chemical Education, 45, 754.
102. Glidewell, C. (1975). Inorganic Chimica Acta, 12, 219; (1976). Inorganic
Chimica Acta, 20, 113.
103. Cotton, F. A., and Wilkinson, G. (1980). Advanced inorganic chemistry (4th
edn). Wiley, New York.
104. Pimentel, G. C., and Spratley, R. D. (1969). Chemical bonding clarified
through quantum mechanics, Holden-Day, San Francisco.
105. Purcell, K. F., and Kotz, J. C. (1977). Inorganic chemistry. Saunders,
Philadelphia.
106. Domenicano, A., Mazzeo, P., and Vaciago, A. (1976). Tetrahedron Letters,
13, 1029.
107. Sutton, L. E . (1958). Tables of interatomic distances and configuration in
molecules and ions, Special Publication, No. 11. The Chemical Society,
London.
108. Sutton, L. E. (1965). Tables of interatomic distances and configuration in
molecules and ions, Special Publication, No. 18. The Chemical Society,
London.
109. Allen, F. H . , Kennard, O . , Watson, D. G., Brammer, L., Orpen, A. G . , and
Taylor, R. (1987). Journal of the Chemical Society Perkin Transactions 11, S1.
110. Orpen, A. G., Brammer, L., Allen, F. H . , Kennard, O . , Watson, D. G., and
Taylor, R. (1989). Journal of the Chemical Society Dalton Transaction, S1.
111. Bent, H . A. (1960). Journal of Chemical Education, 37, 616; (1961). Chemical
Review, 61, 275.
112. Dewar, J. S., Kollmar, H . , and Li, W. K. (1975). Journal of Chemical
Education, 52, 305.
113. Gilli, G. and Bertolasi, V. (1983). Journal of Chemical Education, 60, 638.
114. Allinger, N. L. (1976). Advances in Physical Organic Chemistry, 13, 1.
115. Altona, C., and Faber, D. H . (1974). Topics in Current Chemistry, 45, 1.
116. Engler, E. M., Andose, J. D . , and Schleyer, P. v. R. (1973). Journal of the
117. Ermer, 0. (1976). Structure and Bonding, 27, 161.
118. Hursthouse, M. B., Moss, G. P., and Sales, K. D. (1978). Annual Reports of
The Chemical Society. Section B: Organic Chemistry, 75, 23.
119. Kitaigorodski, A. I., (1978). Chemical Society Review, 7, 133.

120. White, D. N. J. (1978). In Molecular structure by diffraction methods, Vol. 6,
(ed. L. E. Sutton and M. R. Truter). The Chemical Society, London. p. 38.
121. Osawa, E., and Musso, H . (1982). Topics in Stereochemistry, 13, 117.
122. Boyd, D. B., and Lipkowitz, K. B. (1982). Journal of Chemical Education, 59,
269.
123. Ermer, 0 . (1981). Aspekte won Kraftfeldrechnungen Volfgang Bauer, Munich.
124. Burkert, U. and Allinger, N. L. (1982). Molecular mechanics. American
Chemical Society, Washington, D.C.
125. Allinger, N. L., et al., M M l I M M P l , QCPE 318; Allinger, N. L., and Yu, Y.
H. MM2, QCPE 325; Warshel, A. and Levitt, M. QCFFIPI, QCPE 247;
Huler, E., Sharon, R., and Warshel, A. MCA, QCPE 325; Andose, J. D.,
Engler, E. M., Collins, H. B., Hummel, J. P., Mislow, K., and Schleyer, P. v.
R. BIGSTRN, QCPE 348; Iverson, J. D., and Mislow, K. BIGSTRN2, QCPE
410; Browman, M. J., Carruthers, L. M., Kashuba, K. L., Momany, F. A.,
Pottle, M. S., Posen, S. P., et al. UNICEPP, QCPE 361.
126. Biirgi, H. B. (1975). Angewandte Chemie, International Edition in English, 14,
460.
127. Murray-Rust, P. (1978). In Molecular structure by diffraction methods, Vol.
6, (ed. L. E. Sutton and M. R. Truter). The Chemical Society, London.
p. 154.
128. Biirgi, H. B., and Dunitz, J. D. (1983). Accounts of Chemical Research, 16,
153.
129. Biirgi, H. B., (1985). In Static and dynamic implications of precise structural
information, Reports of the 11th International School of Crystallography,
Erice, Italy. p. 245.
130. Eyring, H., and Polanyi, M. (1931). Zeitschrift fur Physikalische Chemie, B12,
279.
131. Atkins, P. W. (1986). Physical chemistry (3rd edn). Oxford University Press.
132. Sathyamurthy, N. and Joseph, T. (1984). Journal of Chemical Education, 11,
968.
133. Miiller, K. (1980). Angewandte Chemie, International Edition in English, 19, 1.
134. Biirgi, H. B. (1975). Angewandte Chemie, International Edition in English, 87,
461.
135. Biirgi, H. B. (1973). Inorganic Chemistry, 12, 2321.
136. Britton, D . , and Dunitz, J. D. (1981). Journal of the American Chemical
Society, 103, 2971.
137. Biirgi, H. B., Dunitz, J. D., and Schefter, E . (1973). Journal of the American
138. Biirgi, H. B., Dunitz, J. D . , Lehn, J. M., and Wipff, G. (1974). Tetrahedron,
30, 1563.
139. Biirgi, H. B., Dunitz, J. D., and Shefter, E . (1974). Acta Crystallographica,
B30, 1517.
140. Gilli, G . , and Bertolasi, V. (1979). Journal of the American Chemical Soceity,
101,7704.
141. Gilli, G . , Bertolasi, V., Bellucci, F., and Ferretti, V. (1986). Journal of the
American Chemical Society, 108,2420.
142. Winkler, F. K., and Dunitz, J. D. (1971). Journal of Molecular Biology, 59,
169.
143. Biirgi, H. B., and Shefter, E., (1975). Tetrahedron, 31, 2976.
144. Krygowski, T. M., Anulewicz, R . , and Kruszewski, J. (1983). Acta
145. Hinze, J. and Jaffk, H . H. (1962). Journal of the American Chemical Society,
84, 540; (1963). Journal of Physical Chemistry, 67, 1501.
534 1 Gastone Gilli
146. Hinze, J., Whitehead, M. A., and Jaffk, H. H., (1963). Journal of the
American Chemical Soceity, 85, 148.
147. Iijima, K., Ohnogi, A., and Shibata, S. (1987). Journal of Molecular Structure,
156, 111.
148. Ferretti, V., Bellucci, F., Bertolasi, V., and Gilli, G. (1985). Proceedings of the
IX European Crystallographic Meeting, Torino, p. 337.
149. Bertolasi, V., Gilli, P., Ferretti, V., and Gilli, G. (1991) Journal of the
Protein crystallography
Introduction
Protein crystallography, the subject of this chapter, is a specialized branch
of crystallography that investigates, by using diffraction techniques on single
crystals, the three-dimensional structure of biological macromolecules.[l-sl
For a long time, good quality single crystals were obtainable only for
globular proteins. Nowadays, other biological macromolecules, like t-RNA,
polysaccharides, and polynucleotides, give crystals suitable for X-ray
analysis. Therefore, the techniques for structure solution and refinement
described in this chapter apply to all cases where crystals with a high portion
of unordered solvent can be grown.
Despite the fact that crystals of proteins have been known since the
beginning of the century, only around 1960 were the first structures,
myoglobin and haemoglobin, e l ~ c i d a t e d . ~Their
~ ] resolution was made
possible mainly by the development of the isomorphous replacement
technique. Since then, a lot of theoretical and technical advances have been
made: among them, the use of anomalous dispersion, molecular replace-
ment, and the development of oscillation techniques in data collection of
crystals with large cell dimensions. At the end of the 1970s the growth of
computing power made possible the refinement in reciprocal space. In the
meantime, the introduction of powerful and affordable graphic stations
brought the development of interactive graphic software, allowing faster
interpretation of electron density maps and model building. Only in very
recent years has the availability of synchrotron radiation and two-
dimensional detectors made data collection a routine procedure.
An idea of the 'state of the art' of protein crystallography is given by the
Brookhaven Protein Data Bank (thereafter called PDB),[~]which collects
coordinates of the macromolecular structures solved, mainly by X-rays.
Looking at the January 1989 release, one finds, perhaps with some surprise,
that only 413 sets of coordinates are available (plus 86 bibliographic entries,
that is structures solved but not yet deposited).? They include 10 polysac-
charides, 26 DNA fragments, 6 t-RNA, and 18 model structures. The
remaining 353 are protein structures, including 14 virus coat proteins: taking
into account the fact that most of them are related proteins or variants of
the same molecule (there are for example 34 lysozyme variants or mutants,
12 haemoglobins, and 8 myoglobins), this number must be considered quite
small and demonstrates that solving the structure of a new macromolecule
? I t must be remembered that the Protein Data Bank cannot be considered fully
representative of the macromolecular structures so far solved, since not all of them are quickly
deposited.
536 1 Giuseppe Zanotti
still remains a long and demanding job. But possibly the reason for the
relatively low number of new structures published every year is the
crystallization process, a field in which technical (and theoretical) advances
have proceeded very slowly and which remains the more uncertain step in
protein structure determination.
Protein crystals
It is the solvent content that makes the difference between a classical
molecular crystal and a protein crystal: in the former, all the atoms can be
described in terms of a regular lattice, whilst in the latter a crystalline array
coexists with a high portion of material in the liquid state. The mother
liquid, whose content can range approximately from 30 to 75 per cent or
more, has a strong influence on the behaviour of this kind of crystal, making
them very peculiar and creating some advantages along with some obvious
disadvantages. Among the latter, the major one is that protein crystals are
much less ordered than classical crystals, not only for the large amount of
unordered material present in the crystal itself, but also because surface
groups of the macromolecule in contact with the solvent can show a great
mobility. As a consequence, diffraction data cannot be measured to the
resolution normally attainable with 'small' molecules. On the other hand,
among the advantages, the environment of the macromolecule in the crystal
is not too different from that of the solution from which the crystal was
obtained (the influence of the solvent on the conformation of the protein
cannot be underestimated) and we can profit from the solvent in the
preparation of heavy-atom derivatives, as will be described in the next
paragraph.
A last important point must be remembered about protein crystals: since
inversion symmetry elements are not allowed, the number of possible space
groups is reduced from the original number of 230 to 65 (see Chapter 1,
p. 24).
Principles of protein crystallization

The process of crystallization of a macromolecule is very complex and
poorly understood. A theoretical treatment of it, despite recent progress,[s1
is at present impossible. Nevertheless, a lot of experience has been
accumulated on the crystallization of water-soluble p r ~ t e i n s , [ ~ , and
~ , ~ a~ J ~ ]
crystallographer starting with enough pure material has a reasonable chance
of succeeding in obtaining X-ray quality crystals. The same does not yet
apply to membrane proteins, the crystals of which have been very seldom
obtained.[12,13]
Growth of a protein crystal starts from a supersaturated solution of the
macromolecule, and evolves towards a thermodynamically stable state in
which the protein is partitioned between a solid phase and the solution. The
time necessary before the equilibrium is reached has a great influence on the
final result, that can go from an amorphous or microcrystalline precipitate
to large single crystals. The supersaturation conditions can be obtained by
the addition of precipitating agents and/or by modifying some of the
internal parameters of the solution, like p H and temperature. Since a
Protein crystallography 1 537
Table 8.1. Precipitating agents commonly used for growing

protein crystals suitable for X-ray analysis. The list is not
intended to be exhaustive
Class Examples
Inorganic salts (NH4),S04, Na,SO,, NaCI, KCI,

NH4CI, MgSO,, CaCI,, NH,NO,, LiCl
Organic salts Citrate, acetate, formate
cetyltrimethyl ammonium salts
Organic solvents Ethanol, isopropanol, acetone,
dioxane, 2-methyl-2,4-pentanediol(MPD)
protein is a labile molecule, all the extreme conditions of precipitation, pH,

and temperature should be avoided. And indeed, the understanding of the
biological and physiological properties of the macromolecule via the
determination of the three-dimensional structure is much easier if crystals
are grown in conditions not too far from the physiological environment in
which the molecule operates.
The precipitants frequently used can be divided into three categories:
salts, organic solvents, and polyethylene glycols (PEG). A list of the more
common among them is reported in Table 8.1.
1. pH. The pH value has a strong influence on the solubility, which has a
minimum at a pH close to the isoelectric point of the macromolecule.
2. Salt concentration. Ionic strength can have opposite effects on protein
solubility, i.e. solubility decreases exponentially with increasing ionic
strength, a phenomenon known as salting-out, but it also has a minimum at
very low ionic strength, the salting-in effect. In practice, precipitation can
be achieved by increasing the salt concentration or by dialysing the solution
of protein against water.
3. Organic solvents. Precipitation properties of organic solvents can be
ascribed to the double effect of subtracting water molecules from the
solution and to decreasing the dielectric constant of the medium. They can
sometimes have effects on the protein conformation too, and for this reason
they should be used with caution.
4. PEG. PEG is a precipitating agent with peculiar properties: it is a
polymer, available in molecular weights ranging from about 200 to
20 000 Da;t its effect on solubility, along with some properties in common
with salts and organic solvents, is due to volume exclusion property: the
solvent is restructured and phase separation is consequently promoted. PEG
is of wide applicability, and has been demonstrated to work with proteins
that were crystallized with other precipitating agents.[lO]
Several others parameters and factors can influence the crystallization
process: among them, the protein concentration, the temperature, the
presence of cations that sometimes stabilize the conformation of the
protein, and the purity of the sample. The presence of contaminants can be
extremely important in preventing the formation of crystals suitable for
X-ray analysis.
t The molecular weight is expressed in Dalton, I Da = 1.6604 X g.
The solvent content of protein crystals

The amount of the solvent contained in a protein crystal has been discussed
] analysed data from 116 distinct crystal forms. He
by ~ a t t h e w s , [ ' ~who
defined the quantity VM, given by the ratio of the volume of the unit cell
divided by the total molecular weight of the protein in the cell. Therefore,
,V represents the crystal volume per unit of protein molecular mass: it is
practically independent from the volume of the asymmetric unit and mostly
depends on the solvent content. In the crystals examined, , V has a value
ranging from 1.6A3 ~ a - 'to about 3.5A3 Da-l, but more extreme values
have been found. A special trend can be observed if the molecular weight of
the proteins is taken into account: small proteins tend to have small ,V and
big proteins a large one, which is understandable, since a low ,V means a
low solvent content. Calling V,,,, the crystal volume occupied by the
protein, V; its fraction with respect to the total crystal volume V and M,,,,
the mass of protein in the cell:
since the first term in parentheses represents the specific volume of the
protein and the second the reciprocal of VM, and remembering that
molecular weight is expressed in Da:
If we assume a value of about 1.35g cmP3 for the protein density, as a first
approximation:
V; - 1.23/VM (8.3)
v:,," =1- v;. (8.4)
The most important information contained in VM is not the solvent
content, but the molecular mass of macromolecule in the crystal cell. Very
often the number of molecules per asymmetric unit can be determined
unambiguously, when the molecular weight of the protein is known, at least
approximately. For monomeric proteins, or for those without the possibility
of internal symmetry, the number of molecules in the cell is relevant only
for the subsequent steps in structure determination. For multimeric pro-
teins, composed of identical subunits, this number may be of relevant
biological consequence: in fact, if some of the internal symmetry elements
are coincident with the crystallographic ones, the relative arrangement of
such subunits in the molecule is immediately available.[151
The experimental determination of the density of the protein crystal,
along with that of the dried content of the crystal itself, can allow a quite
accurate determination of the molecular mass of the protein contained in
the asymmetric unit, and consequently the molecular weight of the protein.
These techniques have been reviewed by ~ a t t h e w s , [ ' ~but
] the reader must
be warned that the experimental measurement of the density of such kinds
of crystals is difficult and the results very often uncertain.
Preparation of isomorphous heavy-atom derivatives

Isomorphous heavy-atom derivatives are of paramount importance for the
solution of the phase problem in protein Their prepara-
Table 8.2. Representatives of heavy-atom compounds used

i n protein crystallography for preparation of derivative^[^,^'
(Ac = acetate)
Classes Examples
Platinum compounds K,PtCI, K,Pt(CN),, K,Pt(NO,),,

Pt(NH,),CI, or -Br,, PtCI,,
PtCI,, Pt(ethylenediamine)CI,
Mercury compounds HgCI,, HgAc,, mersalyl acid,
p-chloromercuribenzoic acid,
methylmercury acetate,
ethylmercury chloride
Uranyl salts UO,(NO3),
Rare earths SmAc,, EuCI,
Others KAuCI,, KAu(CN),,
PbAc,, PbCI,, (CH,),PbAc
ASNO,
K,lrCI,
tion is made possible by their quite high solvent content: the presence of
channels of disordered or only partially ordered liquid allows the diffusion
of relatively small compounds into the crystal. Reactions among the diffused
compounds and eventual accessible reactive sites of the protein can take
place. In some special cases, protein molecules are derivatized in solution
and subsequently crystallized, but the former procedure is simpler and it is
advisable to try it first. In practice, the crystal is soaked in a solution of its
mother liquid in which the heavy-atom compound has been dissolved. A list
of the more commonly used heavy-atom derivatives is given in Table 8.2.
Concentration of heavy atoms and time of soaking are the most relevant
variables: they can range from 0.001 M to 0.1 M and from a few hours to
several days (these numbers must be considered only a rough indication,
since extreme cases have been reported). These conditions strongly depend
on the protein under study, on the precipitating agents, on the pH used, and
on the temperature.
1. pH. At high pH (greater than 9) the acidic groups of the protein are
mostly negatively charged, and potentially they can react with cations. At
the same time, many heavy atoms form insoluble hydroxides. At low pH,
on the other hand, potentially reactive groups will be protonated and
prevented from reacting.
2. Precipitating agents. High salt concentration is not the ideal medium
for heavy-atom reaction with proteins, not only because the solubilization of
the compound can be prevented by a very high salt concentration, but also
because ions in the solution will compete with the protein. Polyethylene
glycol solutions on the contrary provide favourable conditions for heavy-
atom reactions, since PEG does not react with most of the compounds
commonly used.
3. Temperature. There is an obvious kinetic effect of temperature on
heavy-atom reactions, which are slower at 4 "C than at room temperature.
Very often preparation of heavy-atom derivatives is a trial and error
technique, unless some special features of the macromolecule are known a
priori. When the three-dimensional structure has been solved, a rationale

for the behaviour of the specific system can be inferred: unfortunately, at
that time it becomes useless.
How isomorphous are isomorphous derivatives?

The term 'isomorphous derivative' ideally indicates a crystal where some
solvent molecules have been replaced by some group of atoms with more
electrons, without any alteration of the structure of the protein or of the
crystal lattice itself. In practice, this never happens: the introduction of a
bulky compound which interacts with some of the atoms on the surface of
the protein will give rise to local movements and displacements of atomic
groups, at least in the close vicinity of the binding site.
To check if some reaction has taken place after the soaking of a crystal in
the heavy-atom solution, diffraction data have to be collected. To avoid
wasting time, a visual inspection of a single precession photograph,
compared with the same zone of the native protein, is often enough to judge
about the quality of the derivative. A more quantitative test is given by the
comparison of structure factor data: an agreement index, R (see pp. 309 and
312), between native and derivative data can be calculated and the entity of
the substitution can be roughly evaluated from it; experience indicates that
a value from 0.15 to 0.25 can be considered a reasonable agreement factor.
A smaller value is an index of little substitution, too high a one must be
regarded suspiciously, as a clue to non-isomorphism.
Lack of isomorphism can be confirmed by other parameters: a change in
cell length of less than about 1.5 per cent can be tolerated to a resolution of
about 3 A , but it could be deleterious at 2 A . Furthermore, the mean
difference among structure factors amplitudes must be practically constant
over the entire range of 8 used: quite large differences at low resolution and
very low at high resolution indicate disordered heavy-atom binding, whilst
an increase of the differences with d spacing is an index of non-
isomorphism.['l
The solution of the phase problem

The following section is entirely concerned with primary phasing, that is the
determination of approximate phase angles of protein structure factors.
First the two main methods used to solve the phase problem in protein
crystallography, isomorphous replacement and anomalous scattering, are
described, without taking into account the experimental errors. Their
treatment is introduced later, along with the techniques of phase refine-
ment. Methods to improve the electron density maps and rotation and
translation functions, that in some cases can help to solve the phase
problem, have been left for the end of the chapter. A brief mention of the
application of direct methods to macromolecular crystallography is given in
the last subsection.
The isomorphous replacement method

The method of isomorphous replacement is a corner-stone in protein
crystallography: not only has it been used to solve the phase problem for the
first protein structures, but it remains, with few exceptions, the only
procedure available to solve the structure of completely new proteins.['']
Let Fpbe the structure factor of the native protein, Fp its magnitude and
rpP its phase. FpH, FpH, and rppH are the corresponding quantities of the .%
2
heavy-atom derivative. If we assume a perfect, 'ideal' isomorphism, the d,
relationships between Fp and FpHis illustrated in Fig. 8.1 and can be -2
written:
where FH is a vector representing the contribution of the heavy atom I Real axis
'alone'. Fig. 8.1. Representation in the Argand plane of

Our final goal is to derive the value of pp, but the only quantities in the ~ ~ i ~ ' ~ ~ ~ ~ ~ ~
figure that can be measured are Fp and FpH.On the contrary, FHcannot be atom derivative (F,,), assuming a perfect
measured, but since it represents the contribution of few atoms, it can be i~omorphismofthetwocrystals.
calculated if their atomic coordinates and thermal parameters are known.
Let us first describe a way to determine the positions of heavy atoms and,
from them, to calculate FHand derive an estimate for rpp.
The determination of heavy-atom positions

For a small molecule, the heavy-atom positions can be determined via a
Patterson map calculated using measured amplitudes (see p. 328). The same
map is not useful for a macromolecular crystal: even for a small protein, the
ratio among the number of electrons of the heavy atoms and those of the
macromolecule is so low, that a Patterson map calculated with the
coefficients of the protein derivative will look absolutely meaningless. But if
the structure factor amplitudes of two isomorphous crystals are known,
various kinds of difference-Patterson syntheses can be calculated.
The 'true' difference-~atterson[l']synthesis would be that calculated using
as coefficients IF;, - F;\. This map represents the difference between the
Patterson of the derivative minus the Patterson of the native protein. In
fact, remembering (8.5):
The right-hand side of eqn (8.6) shows that the Patterson map calculated
with these coefficients will contain peaks corresponding to heavy-atom
heavy-atom vectors, FL, and to heavy-atom protein-atom vectors, the
mixed terms.
A more common choice among protein crystallographers is to calculate a
Patterson synthesis using coefficients (FpH- F,)'. This was originally called
modulus difference squared synthesis, but is commonly known as
isomorphous difference-~atterson.['~] The resultant map has several inter-
esting features, but the main reason for its success is that it is more
representative of the heavy-atom situation. In particular, in the centrosym-
metric case (that is for centric? projections in a non-centrosymmetric space
group), since FpHand Fp are collinear, these coefficients represent a true
estimate of the value of FH. In fact:
t The word centric is used here in a general sense, indicating that phases are restricted to two
values, not necessarily 0 or x.
Since FH is usually small if compared with the modulus of the structure

factors of protein and derivative, in general the relationship
FH= IFPH- FPI (8.8)
will be satisfied. In few cases, when FpHand Fp are both very small, it may
+
happen that FH=FPH FP This is called a cross-over, and it can be
removed from the calculation. For a centric projection a Patterson synthesis
calculated with coefficients (8.8) will correspond to the Patterson map of the
heavy atoms alone, that is it will contain peaks corresponding only to the
vectors relating heavy-atom positions. Methods to derive the atomic
coordinates from a Patterson map are described on pp. 328 to 335. In
favourable cases, i.e. when more than one centric projection is available,
the three coordinates of the heavy atoms can be obtained.
For general reflections, IFpH- FPI will not be a correct estimate for FH.
From Fig. 8.1:
F&= Fc + FEH - 2FPHFPCOS ( q p - qPH). (8.9)
If the phase angles of protein and derivative structure factors are very
similar, that is if the two vectors FpHand Fp are nearly collinear, the cosine
term is close to 1 and:[17]
F&= F: + F : ~- 2FPHFP= (FPH- F P ) ~ . (8.10)
The probability distribution of the values of q, - qpHhas been analysed by
~ i m [ ~(see
~ , also
~ ~ 1p. 393): the larger the value of the product FpHFp,the
more likely the difference Iqp- qPHIis to be small, an vice versa. Equation
(8.10) does not represent a good approximation for small structure factors,
but the latter will give a small contribution to the Fourier transform. In
practice, a Patterson map calculated with coefficients (8.8) will tend to the
Patterson-difference of the heavy atoms alone in the non-centrosymmetric
case too (for a general discussion of the features of difference-Patterson
maps, see Ramachandran and ~rinivasan['~]).
The single isomorphous replacement (SIR) method

Once the position of heavy atoms is more or less accurately known, the
calculated value of I;;I can be used to estimate the phase angle of the
structure factor of the native protein. From Fig. 8.1, using simple
trigonometric considerations,? we can write:
Fc, = F$ + F ~ H + 2FpFHCOS 6, (8.11)
cos 6 = (F;, - Fc - F & ) / ~ F ~ F ~ ; (8.12)
8 corresponds to qp- q H , and since qHis known, q,, the phase angle of
Fp, can be derived:
Fig. 8.2. Graphic construction of S.I.R. method,

showing the two possible values for the phase Equation (8.13) contains a cosine term, consequently two solutions exist for
angle of the native structure factor F,. The
structure factor of the protein is a vector lying qp. This ambiguity is illustrated in Fig. 8.2, where two circles of radius Fp
on a circle of radius Fp and centre 0. If a circle of and FpHare drawn with centre respectively in 0 and at the end of vector
radius F,,, the modulus of the structure factor of
the heavy-atom derivative, is drawn with centre
-FH: the intersections of the two circles represent the two possible solutions
in the end of the vector -F,, the intersection of
the two circles represents the two possible
solutions of (8.13). t An alternative way of deriving eqn (8.11) is given in Appendix 8.A
for cpp. It is important to notice that this problem is immediately solved for
restricted phase reflections, for which eqn (8.12) becomes: A
F, FP
-
*
Since the value of FHis known from calculation and the values of FpHand Fp
from observation, the relative sign of Fp with respect to FH can be
determined unambiguously from the comparison of the magnitude of the
three vectors (Fig. 8.3). For general reflections this is unfortunately not
true, and the ambiguity between the two possible values for the phase angle
must be solved with other techniques.
In principle, a three-dimensional electron density map could be calcu- - --- -
lated, using the two possible phase values: this kind of synthesis is called an FPH FP
SIR map and it will contain information about the true structure plus (c)
noise.['lI In fact, if we call cp, the correct phase value of cpp and cppw the Fig. 8.3. The sign of reflections of centric zones
wrong one, looking at Fig. 8.4 we can write: can be fully determined using only one heav
atom derivative (after Blundell and Johnson ). 16
When F, is small compared with F, and F,,,
WSIR= FPexp (i~,) + FPexp (~vPw). (8.16) case (a) and (b)apply: if F,< F,,, the sign of F,
is equal t o that of F,; if F,> F,, the sign of F, is
Since cp, = 1/2(cppT+ cppw), then: reversed. The so-called 'cross over', represented
in (c), may occur when F , is larae and F. is
(8.17) small: despite the condiiibn F,; F,,, thk sign of
F, is reversed with respect t o F,.
From (8.17) it is evident that FsIRwill approach the correct value of Fp if

cpH .= cp,,. But since we have no reason to believe the numerical values of
cpH and vPTare correlated, the first term of (8.17) will give the correct
electron density, while the second one will only contribute to noise. In
practice the level of noise in this kind of map is so high that its
interpretation is extremely difficult.
A
The classical solution of the problem of phase ambiguity: the

MIR technique
The multiple isomorphous replacement technique, MIR, has been the first
\- Real.
method used to solve this phase ambiguity, and it remains the more
common and is widely adopted. It is based on the preparation of a second
heavy-atom derivative having heavy atoms bound to the protein in
position(s) different from those of the first one. Let us assume we have Fig. 8.4. The same diagram shown i n Fig. 8.2,
measured coefficients from two derivatives, called FpHland FPH2,and in except that now the 'true' and the 'wrong' value
for the native structure factors are explicitly
some way we have calculated FHland FH2(that means we know the indicated as F,, and Fpw respectively. It can be
positions of the heavy atoms of the two derivatives, relative to the same seen that the further the t w o vectors are from
each other, the more the S.I.R. phase
origin). The complete solution of the phase problem can be illustrated using (corresponding t o rp,) is different from rp,.
a diagram due to ~ a r k e r , [illustrated
~~] in Fig. 8.5: a circle of radius Fp
centred on the origin 0 of an Argand plane is drawn, and from 0 the two
vectors -FHl and -FH2; they are in turn the centres of two circles of radii
FpHland FPHZ respectively. Since the two derivatives are assumed perfectly
isomorphous, the three circles must intersect in one point (B in Fig. 8.5):
the vector OB will coincide with the structure factor of the protein. The
graphical construction just described is equivalent to solving a pair of
simultaneous equations like (8.13):
In principle, two different heavy-atom derivatives supply a unique, unam-

biguous value for the phase angle of the native protein. In practice, errors
-Real from various sources prevent system (8.18) from having an exact solution.
axis
Anomalous scattering: a complementary (or alternative)

approach to the solution of the phase problem
According to Chapter 3 (p. 167), the structure factor may be rewritten as
Fig. 8.5. Harker construction illustrating the
M.I.R. method. A second circle of radius F,,,
the sum of two parts:
and centred at the end of vector -F,, has been
added to the construction of Fig. 8.2. The three
circles have only one common intersection,
representing the solution of the two equations where the real part of anomalous scattering (the so-called anomalous
(8.18). dispersion) is included in F1(hkl). If F"(hk1) is not zero, the practical result
is the breakdown of Friedel's law: F(hk1) is no longer equal to ~ ( i i i and)
the Friedel pairs are more correctly called Bijvoet pairs. Anomalous
scattering is negligible for light atoms like carbon, nitrogen, or oxygen (that
is, nearly all the atoms of a protein) and can be of practical use for a native
protein molecule only in special cases, when heavy ions are bound to it.
Even if for a very small protein the sulphur's anomalous scattering has been
used to solve a structure ab i n i t i ~ , [in
~ ~general
] the anomalous scattering
phenomenon becomes relevant for heavy-atom derivatives. Equation (8.5)
must be in fact modified, as shown in Fig. 3.15 and Fig. 8.6:
I Real axis
Fig. 8.6. Vector diagram showing F& and F&

+
(superscripts and - indicate_t_he two terms of
the Bijvoet pairs, i.e, hkland hkl; the
components are drawn reflected across the real where superscripts + and - stands for (hkl) and (iRi), respectively. If the
axis). The two vectors have different heavy-atom derivative contains only one kind of anomalous scatterer,
magnitudes, due to the anomalous contribution
of the heavy atoms. F,!, is the real and FA the vectors F h and Ffl, are mutually perpendicular, and if we call QI, the phase
imaginary component of F,. angle of vector F h , QI, f n / 2 will be that of Ffl,. This fact can be considered
practically always true for protein derivatives, and has the important
consequence that anomalous effect cannot be observed if the vectors Fpand
FHare collinear, for example for all the centric reflections (see Fig. 8.7).
An accurate measurement of the Bijvoet pairs allows the calculation of an
anomalous-difference Patterson synthesis, with coefficients:
Notice that we do not need native coefficients for the calculation, so (8.21)
is also useful when an anomalous scatterer is present in the native protein
crystal. A map calculated with coefficients (8.21), supposing the anomalous
contribution Ffl, is small compared with Fp, which is very often true in our
I Real axis case, will contain maxima corresponding to vectors relating the positions of
Fig. 8.7. For restricted-phase reflections, if only anomalous scatterers. This synthesis was for example used by ~ o s m a n n [ ~ ~ ]
one type of anomalous scatterer is present, F:,
and F& have the same magnitude, independent to determine the Hg position in two derivatives of horse haemoglobin.
of the size of the imaginary component. Other kind of anomalous Patterson maps with coefficients different from
(8.21) can be produced, like the a-anomalous or the P-anomalous

synthesis.[Is1
The most important application of anomalous scattering in protein
crystallography remains the possibility of solving the phase ambiguity
discussed in the previous ~ a r a g r a p h . [ ~Let
~ - ~us~ assume
] we have only one
heavy-atom derivative, and we have measured F;,, FFH, and FP.T
Moreover, the position of the heavy atom(s) are known. Looking at Fig.
8.6, an expression for the phase angle of the derivative, whose complete
-
Real
axis
derivation is given in Appendix 8.A, can be written:
F;, - FFH= 2FA sin (qpH- qH) (8.22)
qPH= qHf sin-' [(F:, - F&)/2FA]. (8.23)
Equation (8.23) gives again two possible solutions for qp,, but since it
contains a sine term, its information is complementary to that of (8.13): the Fig. 8.8. A graphical representation of the
phase angle can now be fully determined. Information from both sources, solution of the phase problem using one heavy-
atom derivative and its anomalous absorption
isomorphous replacement and anomalous scattering, is used in Fig. 8.8, component. Two circles of radius F+,, and F,
P
where the role of the second derivative of Fig. 8.5 is played by anomalous are drawn at the end of vectors -F, and -F,
respectively. OA is the native protein structure
scattering. factor, whose phase angle is uniquely
determined by the intersection of the three
circles.
The use of anomalous scattering in the determination of
the absolute configuration of the macromolecule
Two isomorphous derivatives are, in principle, enough to determine the
correct value of the phase angle qp of the structure factor, but the M.I.R.
method alone does not allow us to select the correct enantiomorph for the
protein. Once a right-handed system has been chosen in indexing the
reciprocal lattice, the initial arrangement of the heavy atoms for the first
derivative is arbitrary. In other words, the phase of FHcan be q,, as in Fig.
8.2, or -qH. Once a hand has been arbitrarily selected for a heavy-atom
derivative, all the others must simply be consistent with it, but they can all
be relative to the wrong enantiomorph. If this is the case, the electron
density map will represent the mirror image of the correct structure, that is
all the amino acids will have D configuration. When a medium (or high)
resolution electron density map can be produced, some overall features may
indicate that the wrong enantiomorph has been chosen (for example,
a-helices will appear left-handed instead of right-handed-the amino acid
configuration and a-helices are described on p. 574). Then the problem can
be immediately corrected by recalculating the phases with the heavy atoms
arranged in the other way.
However, anomalous scattering differences can be used to identify from
the beginning the correct enantiomorph, using an extension of the method
proposed for small molecules by B i j v ~ e t . [We
~ ~ ]have seen in the previous
subsection that, at least in principle, structure factors for one single
heavy-atom derivative along with its anomalous differences can solve the
phase ambiguity of the protein. This is no longer valid if two possibilities are
taken into account for the value of q,, since now there are two possible
t Readers must be warned that measurement of Bijvoet pairs requires very accurate data
collection, since anomalous differences are very often of the same order of magnitude as the
experimental errors on intensities. Synchrotron radiation is extremely helpful in this respect:
thd appropriate wavelengths that optimize the anomalous effect can be selected (see p. 239).
value of q, associated with every qH. Let us assume the solution for the
phase angle of the derivative structure factor, from (8.A.7) and (8.A.15), is
qpH= qH+ 6. We are left with another ambiguity, since the original choice
of q H was arbitrary: the two possibilities are in fact qpH= qH-t 6 or
qpH= - qH- 6, and they determine the selection of the enantiomorph
(Fig. 8.9).
At least two heavy-atom derivatives and anomalous differences for one of
Real axis them are necessary to select the correct hand of the molecule: using phases
from one derivative along with anomalous contribution and the heavy atoms
in the two possible arrangements, two difference-Fourier of a second one
are calculated. Peaks in the difference map calculated with atoms in the
correct hand will be reinforced, those with the wrong one lowered.[']
Fig. 8.9. Graphical illustration of the ambiguity

in the choice of enantiomorph. This figure is The treatment of errors
equivalent to Fig. 8.2, except that now two
possible arrangements of the heavy atoms for In the previous paragraphs, we assumed that all the measurements were
the same derivative are shown. Vectors OA and error-free and that a perfect isomorphism existed among crystals of the
OB are the two possible structure factor values
for the native protein if v, is selected as the
native protein and derivatives. This is obviously not realistic, since Fp and
phase of the heavy-atom derivative; vectors OA' FPHare experimentally measured quantities and FH is calculated from
and OB' those corresponding to -q,, that is to approximate heavy-atom positions. Experimental errors can be considered
the enantiomorph.
to arise from three different sources:
(1) errors in intensities, mainly attributable to inaccuracy in measurements,
scaling etc. ;
(2) errors in position, occupancy, and temperature factor of heavy-atoms;
(3) lack of isomorphism, due to displacement of protein atoms and of
solvent molecules close to the heavy-atom binding site, etc.
As a consequence, the circles of Fig. 8.5 are not expected to intersect at the
same point and the system of equations (8.18) will not have an exact
solution (Fig. 8.10). Following Terwilliger and ~ i s e n b e r g , [let
~ ~us
] denote
the different contributions to the total error deriving from points (I), (2),
and (3) as a, q, and p, respectively. It must be noticed that, whereas a, and
apH,the standard deviations for Fp and FpH,are routinely evaluated (see pp.
271 and 280) an a priori estimate of q and p is impossible.
The systematic treatment of errors in the isomorphous replacement was
pioneered by Blow and Crick.[301They made two basic assumptions: all the
errors reside in the calculated structure factor of the derivative, FpH(c,Ic);
errors follow a Gaussian distribution. Consequently, a phase probability
distribution for derivative j is given by:
Real
axis The difference FPH(obs) - FPH(calc)
in (8.24) is called lack of closure and is
usually denoted by E,; its physical meaning is illustrated in Fig. 8.11, which
is a realistic picture of what was idealized in Fig. 8.1: now the triangle
formed by the three vectors F,, FPH, and FHdoes not close. The lack of
closure depends from the phase of the protein, since:
Fig. 8.10. Harker construction analogous to Fig.
8.5. where, due to lack of isomorphism and
errors in the data, the three circles of radius F,, ni in (8.24) is a normalizing factor and Ei is a measure of the total error. The
F,,,, and F,,, do not intersect at the same point. choice of .E, is crucial and not obvious. In practice, the value commonly
.-V) Fig. 8.11. (a) Representation of the lack of

i3 closure E, following the Blow and Crick
assumption that all the errors reside i n the F,,
9 amplitude. As a consequence of it, F,,(,,,,, is
-E equal t o Fp~
,,, + E . (b) components of the error
described in (a) are now explicitly shown: ,u is
the lack of isomorphism and q the difference
I Real axis
-
Real axis
between the true and the estimated value of F
The third source of error, the inaccuracy i n the
measurement of Fp and F,, prevents the
.,
(a) (b) closure of the polygon.
used is:[311
The right-hand side of (8.26) represents the mean square value of the lack
of closure residual. Since it depends from the resolution, it is evaluated in
ranges of sin 8/A.
An a priori derivation of (8.24) along with that of E; for centric and
acentric reflections has been obtained by Terwilliger and ~ i s e n b e r ~asl ~a~ ]
function of all the different sources of errors:
Equations (8.27a) and (8.27b) are not of direct use, since we lack a value
for q and p, but from them a theoretical justification of (8.26) can be
obtained.[291
If more than one derivative is present, (8.24) can be modified to give the
total probability distribution of the phase of a reflection:
The sum is extended to all the j derivatives used in calculating the phases. A
diagram showing P,(q,) for the hypothetical derivatives of Fig. 8.10 is
reported in Figs. 8.12(a) and (b), and the total probability in Fig. 8.12(c).
The lack of closure for anomalous scattering measurements assumes a
different form:[25,26,291
but the probability for the phase of the structure factor due to the
anomalous scattering is analogous to (8.24):
E,,, is usually smaller than Eiso and can be estimated for example from the
agreement among centric reflections.[251If isomorphous and anomalous
information are available for the derivative j, they can be combined to give
a total probability:
P,(?'P
') = (q(iso,(q~>)(P,(an01(9~)). (8.31)
On the basis of a different formulation of the error model. Hendrickson and
~atmann[~'] proposed a different representation of the phase probability

distribution:
~ ( v P=) exp (K + A cos qp+ B sin qp+ C cos 2 q p + D sin 2qp). (8.32)
Equation (8.32) has the advantage of being completely general and that new
information can easily be included. Moreover, the old probability can be
expressed using coefficients A, B, C, and D through a least-squares
fitting.[331Since the overall probability distribution is given by the product of
individual probabilities:
if every one is cast in the form (8.32), Ptot(qp)can be obtained by the sum
of the individual coefficients:
+ C. Ci cos 2 q p + C D~sin 2 q p
i i
From (8.33) or (8.34) we can calculate the probability distribution of the

phase angle for every structure factor and use the most probable phase in
the calculation of an electron density map. This is called the most probable
Fourier, and in principle it would be a good approximation. Unfortunately,
P(qp) tends to be bimodal, and the use of the most probable phase can
introduce large errors (see Fig. 8.12(c)). According to Blow and Crick,[301
the Fourier which is expected to have the minimum mean square difference
from the 'true' Fourier is that obtained using the centroid of the probability
distribution of Fp, that is:
2n
F ~ ,= NFP
~ ~ ~~ )( 9exp
~ (MP)
) dqP (8.35)
where N is a normalization factor:

Fig. 8.12. (a) Lack of closure, cj, for the t w o
derivatives shown in Fig. 8.10. c, assumes the
value 0 at the intersections of the circles of
radius Fp and FpHj. Since the three circles do not
intersect at the same point, this happens for
values of @,different for each derivative. (b) Equation (8.35) is equivalent to:
q(qp), calculated on an arbitrary scale using
(8.24), for the t w o derivatives: q(p,) has a
maximum every time cj = 0.(c) Combined
probability for the t w o derivatives. p, where m is a weighting function called the figure of merit:
represents the most probable phase, p, the Best
phase, that is the centroid of the probability (see
text for more details). (8.38)
qP(best)
can be calculated easily by developing (8.38) and observing that
P(qp) can be sampled and the integral substituted by a sum:
The sum is estimated for the i values of qp at which the probability is

evaluated, usually at intervals of 10 or 20 degrees. The figure of merit is a
value ranging from 0 to 1 and is a measure of the reliability of the phase qp:
in principle, m = 1 means an error of O" in the phase angle, m = 0.5 means
an error of 60".
The refinement of heavy-atom parameters

Once an approximate position of one or more heavy atoms has been found,
the problem immediately arising is the refinement of their coordinates,
thermal parameters, and occupancy. Refinement is made difficult by the
lack of a correct value of FH(obs):we can only have estimates of it.
Historically, the first method of refinement of heavy-atom parameters was
proposed by ~ossmann,["] who suggested minimization, using classical
least-squares methods:
FH(calc) can be calculated in the usual way from the known heavy-atom
parameters, but a correct estimate of FH(,bs) is available only for centric
reflections (see p. 541), so (8.41) should in principle be employed to refine
that class of reflections and the refined parameters used to calculate all the
phases. Nevertheless, since (8.8) can be considered an approximate estimate
of FH(obs),expression (8.41) has sometimes been used to refine acentric
data.$
Possibly the most used and most popular refinement scheme among
protein crystallographers is the so-called phase refinement, introduced for
the first time in the refinement of myoglobin.[341 The quantity minimized is:
which represents the least-squares minimization of the lack of closure,

where:
) IFp(obs) exp ( ~ v P+
F ~ ~ ( c a l c= ) FH(calc)l. (8.43)
In general w is chosen as the reciprocal of the total error E, calculated from
(8.26) for the derivative being refined.§
t In the rest of the chapter, unless otherwise specified, the sums are extended to the
appropriate set of reflections.
$If anomalous scattering data are available, a more accurate approximation of FH (see
Blundell and ~ohnson['])is represented by
+
The - and values are called lower (F,,) and upper (F,,) estimates, respectively. It has
been shown that the lower estimate would represent the correct value of FH for the large
majority of the reflections. It must be considered, in any case, that when anomalous differences
are low or the measurements are affected by large errors, coefficients (8.8) can be more safely
used.
5 A modification of the phase refinement scheme, called MVFC (minimum variance Fourier
coefficients) has been proposed by ~ ~ ~ u s c Theh . [ expression
~~] minimized is again (8.42), but
the sine and cosine of the phase are considered as variables. The values of the trigonometric
functions estimated in the refinement are used directly for the electron density calculation,
without the need to postulate a probability distribution for the phase error.
Whatever refinement scheme is selected, different procedures can be

followed: refining only centric reflections is quite safe, if reflections with
restricted phase represent a good portion of the total. The number of
variables is also important in this respect: in general, three positional
parameters, the occupancy, and an isotropic temperature factor are varied
for every atom, but since the last two parameters are highly correlated, they
must be refined separately. If general reflections are used, in order to avoid
bias it is advisable to refine the parameters of one derivative omitting it
from the phase calculation, that is refine one derivative and calculate the
phases with all the others.[361This can obviously be done if several different
derivatives are available.
One of the difficulties in heavy-atom refinement is the check of the
correctness of the parameters used: this is particularly true at the beginning
of the refinement, where it is sometimes difficult even to decide if the
solution found is correct or not. Many quantities can be used to monitor the
progress of refinement: none of them taken alone gives an absolute
indication, but together they can be considered a reliable check. Different
kinds of R factors have been defined:
The Cullis R factor, Rc (8.442, is calculated only for centric reflections. A

value between 0.40 and 0.60 is generally considered reasonable, the only
drawback being the low number of reflections used in the calculation. For
general reflections, the Kraut R factor (8.45) is used instead. Since its
numerator is the quantity being minimized in phase refinement, RK can be
considered a good parameter to monitor the course of refinement. Unfortu-
nately, a high degree of substitution of the derivative implies a statistically
large RK, and a very low substitution a small one, so the Kraut R factor
cannot be considered an indication of the correctness of the variables being
refined. i
An idea of the relative importance of a single derivative in the phase
calculation is given by the R(modulus), defined by (8.46), which is the lack
of closure divided by the modulus of the heavy-atom structure factor. If its
value is greater than one, the circles of Fig. 8.8 will never intersect and
(8.18) will not have any solution at all. R(modulus), if calculated in shells of
resolution, gives an index of the utility of that derivative at different
resolutions. Often the reciprocal of R(modulus), which is called the phasing
power, is quoted. Its value must be bigger than one.
The last important quantity used in refinement is the figure of merit,
defined in (8.38). The figure of merit gives a direct estimate of the errors in
t In the first refinement scheme, R, can also be used: R,,, = C (IF,, -

F,(,,,,,I)/C F, the disadvantage of it being that it is possible to obtain a completely
incorrect structure with a value of R,,, ranging from 0.40 to 0.60.
phases, but it is strongly dependent on the value of the total error E : an

overestimate of E will give an over-optimistic figure of merit.
Picking up minor heavy-atom sites: the difference-

Fourier synthesis
When the positions of heavy-atoms have been identified and some refine-
ment has been carried out, the next step is to check the correctness of the
heavy atoms identified and the localization of other minor sites that may be
present. This can be accomplished in principle by using difference-Fourier
techniques (see p. 366): the map that better represents the differences
between observed and calculated positions of heavy atoms should be that
with coefficient^:[^']
dF= F ~ ( o b s ) - FH(calc). (8.47)
I Real axis
Since FH(obs)
is not directly measurable, (8.47) can be approximated only if
Fig. 8.13. Argand diagram of the quantities used
anomalous differences are available. in a Fourier-difference calculation. AFis the
Often a map with coefficients: magnitude of the coefficient used instead of F, if
coefficients (8.48) are used. GFis used i n the
AF = m(FpH- Fp) exp (iq,) (8.48) Fp, - Fp,,, synthesis. (Adapted from
Henderson and offa an.'^'')
is calculated. Unfortunately, FpH-Fp does not correspond to FH, and
moreover qp is not the phase of FH.Furthermore, since Fourier maps are
strongly dominated by phases, a difference-Fourier map calculated with
coefficients (8.48) will tend to reproduce the peaks of the heavy atoms used
in calculating the phases. A difference-Fourier map of a derivative
calculated with phases obtained from that derivative is normally useless, and
the technique of cross-difference Fourier, that is the Fourier difference of a
derivative calculated using phases for the protein obtained from all the
others, is highly recommended. Unfortunately this is not always possible,
since it requires at least three different derivatives.
A synthesis, proven useful in picking up minor sites, is the derivative
difference-map, calculated using coefficients:
- F~H(calc))exp ( i ~ P ~ ( ~ ~ l ~ ) ) . (8.49)
dF = rn(F~~(obs)
If the position of all the major heavy atom peaks has been determined
correctly, coefficients (8.49) allow the localization of low-occupancy sites.
Figure 8.13 shows the coefficients used in (8.48) and (8.49) compared with
the 'true' ones.[381
A third approach to the solution of the phase

ambiguity: real-space filtering
We have already seen in Chapter 4 that the positivity of the electron density
can be used to obtain phase information. The same principle may be used to
improve electron density maps of protein^.[^^-^^]
This method is called real-space filtering and can produce correct phases
when good SIR data are available. Phase information can be extracted by
an electron density map by a simple back-Fourier transform: if a suitable
modification has been introduced in the map, the phases obtained by the
back-transformation will be more correct than those used in calculating the
map and so the new phases can be used to modify the original ones. The
procedure described here, suggested by ~ a n g , [is~based
] on the same basic
principle.
We have seen on p. 543 that an electron density map calculated using SIR
phases will contain the correct density of the molecule, covered by an
enormous amount of noise. Information about the correct phases can be
obtained via the following automated procedure.
1. After calculating an electron density SIR (or single anomalous scatter-
ing, SAS) map, errors are removed in direct space by evaluating for
every point an average value p, from the formula:
where S is a constant estimated from previously solved protein

structures.
2. The value of p, estimated from (8.50) is added to any point in the map.
Any negative density remaining is considered an error and set to zero.
The density in solvent region is substituted with the value p, (p,,,,). +
3. Structure factors are calculated by Fourier inversion. Now new phases
are available, and they can be combined with the old ones, using one of
the well known phase combination procedures.[19~20]
4. Combined phases will produce new maps, that will undergo the
procedure again, until convergence is reached.
An important feature of the Wang procedure is that an automatic method
has been devised to decide what is solvent and what is protein, based on the
practical assumption that solvent regions, like protein regions, are con-
tiguous to some extent. The value of electron density at point j is substituted
by a new value which depends from the points surrounding it, within a
sphere of radius R,? according to the criteria:
where
w i= 1 - rijlR if pi > 0 and R > rij (8.52)
Since the resulting map has large homogeneous connected regions, the
molecular boundary can be revealed by a threshold density level appropri-
ately selected. Real-space filtering methods can also be used to extend the
resolution of the electron density map.
Rotation and translation functions and the molecular

replacement method
Sometimes it may happen that we know the three-dimensional structure of a
molecule and we are interested in solving the same structure in a different
t The optimum value of radius R depends mainly on the resolution: a value between 8 and
10 8, is often used to average a 3 8, resolution isomorphous replacement map.
Fig. 8.14. A simplified two-dimensional

\ illustration of (8.53).In (a) and (b)the same
\ 'molecule' is represented in different positions
\.' with res~ectto the same reference svstem. Bv a
rotation'of an angle @anda translation of t, the
(c) object in (a) can be superimposed to that in (b).
space group. At other times we have reasons to believe that the conforma-
tion of a protein is quite similar to that of another that has been previously
solved, which is often the case for the same protein from different species.
In all the cases mentioned above, six variables, three rotational and three
translational, will approximately describe the transformation from one set of
coordinates to the other. In fact, if we call X the set of vectorst representing
the atoms of the original molecule and X' the transformed ones, the
transformation is simply described by:
where [C] is a matrix that rotates the coordinates X into the new orientation
and t is a translation vector. Equation (8.53) is illustrated for a two-
dimensional situation in Fig. 8.14, where a 'molecule' formed by three point
atoms can be superimposed to an identical molecule in a different
orientation by the translation of a vector t, after the rotation of an angle cu.
As mentioned in Appendix 5.B (p. 383), the technique of positioning a
molecule or a fragment of known structure in a crystal cell is called
molecular replacement. In principle it is possible to simultaneously search
for the six variables which minimize the difference between Fobs and F,,,,,
but in practice this is a very hard task, even for the fastest computer.$ The
solution of the problem was pioneered by Rossmann and v low,[^^] who
explored the possibility of finding the orientation of similar subunits in a
crystal cell without any knowledge about the translation t, making use of the
Patterson function. After the correct orientation has been found, a search
for the translation vector can be carried out (a collection of papers on
molecular replacement is found in the book by ~ o s s m a n n ) . [ Let
~ I us first
describe the methodology and the problems connected with the rotation
function.
The first step in molecular replacement: the rotation function

The idea of the rotation function can be easily understood by a simple
two-dimensional example. In Fig. 8.15(a) a 'molecule' of three idealized
t In the following discussion all the rotations will be performed in a Cartesian reference
system. It is assumed that, if required, an appropriate orthogonalization is applied before a
rotation is performed.
$This statement is becoming untrue, due to growing availability of computing power.
Subbiah and ~ a r r i s o n have
[ ~ ~ shown,
~ in the test case of the human histocompatibility antigen,
that an exhaustive three-dimensional search at low resolution can be performed. The correct
solution can also be obtained, starting from a random position, using the simulated annealing
approach (see p. 569).
Fig. 8.15. An isolated, simplified 'molecule' of

three atoms. Its self-convolution is shown in (b):
since atoms are considered as points, i t is
everywhere 0,except when t w o points
superimpose exactly.
point atoms is represented isolated, in a orthogonal reference system. Let us

imagine a two-dimensional lattice of similar 'molecules' in a different
orientation. In the lattice of Fig. 8.16(a) the unit cell is made by two oLsuch
molecules, related by a twofold axis, denoted by 1 and 2. Maxima of its
idealized Patterson function, shown in Fig. 8.16(b), can be divided in two
categories: those arising from intramolecular vectors, or self-vectors, and
those from intermolecular or cross-vectors. Maxima belonging to the first
class are indicated in the figure by circled points and are confined to a short
distance from the origin. It is easy to see that by a simple rotation of 112"
anti-clockwise the isolated molecule can be superimposed to molecule 1 of
Fig. 8.16(a), after an appropriate translation, or, by a rotation of 292", to
molecule 2. The self-convolution function of the isolated molecule (Fig.
8.15(b)) can also be superimposed to the Patterson of the crystal if we
perform the same rotation. Let us define a function R(C):
where C is a matrix that rotates the coordinates of the model molecule with
respect to the reference system of the crystal, P,,,,,(u) is the Patterson
Fig. 8.16. (a) A portion of a two-dimensional

lattice of a molecule identical t o that of Fig. 8.15 l------
w
(the unit cell is dashed). (b) Its corresponding
.IIIIJ
Patterson map. Circled points indicate self-
vectors. Squared points are cross-vectors close
to the origin: some of the points of Fig. 8.15(b)
accidentally superimpose t o them during
rotation, giving rise t o false maxima in the
rotation function. (4
function of the crystal and Pmol(Cu)is the self-convolution function of the

isolated molecule, rotated by C. The function R(C) will have a maximum
when the peaks of the two functions superimpose, at least partially. The
calculation of function R(C) for all the possible values of the rotational
variables will allow us to determine the orientation of the known molecule
in the reference system of our crystal.
The right-hand side of (8.54) can be Fourier t r a n ~ f o r m e d [ ~and
~'~~]
reduced, neglecting a constant, to:
F(h) are the Fourier coefficients of the crystal and Fmol(p)the coefficients of
the Fourier transform of the isolated molecule, rotated by C (h, h' are used
here to indicate different terms of (hkl) values, p represents a point in
reciprocal space of a continuous transform). Gh,h,is an interference function
whose magnitude depends on h, h', and the volume used in the integration
of (8.54).
The function R(C) can be evaluated in real space using (8.54) or in
reciprocal space, using (8.55). In both cases the computing time is strongly
dependent on the sampling chosen, which in turn is related to resolution. In
real space PC,,,, and Pmo,must be sampled finely enough for the resolution
selected (generally this means a value around 112 or 113 the d spacing). The
volume of integration is a sphere whose radius depends on the size of the
isolated molecule, and this value determines the steps of the angular
variables used in evaluating R(C). In reciprocal space problems are quite
similar, since Fmol(p) is a continuous function, defined over all the
reciprocal space. The isolated molecule can be put in an artificial cell,
generally a cube whose edges can be about two to thiee times the size of the
molecule, and the continuous function evaluated with a sampling appropri-
ate to the resolution used. In practice, since (8.55) is dominated by large
Fourier coefficients, it is possible to limit the numbers of F ( p ) used.
A faster but more complex approach in evaluating the rotation function,
the so-called fast-rotation function, has been devised by ~ r o w t h e r . If~ ~we~ ]
express the Patterson function in terms of spherical polar coordinates,
(r, 0, q ) , for a rotation C, corresponding to the three angles cq,a 2 , ag, the
rotation function can be written:
1
R(C) = PwSt(r,8, q)RPmol(r,8, q ) r 2sin 0 dr d o d q (8.56)
where RPmolis Pmolafter a rotation C. Equation (8.56) can be expanded

using Bessel functions, more appropriate to a rotation group than a Fourier
series, well suited for translation operations. The use of Bessel functions
requires a lot of difficult mathematics, outside the scope of this book, but
the final result is that R(C) can be evaluated as a summation of two terms,
one of them independent of the rotation itself. The computation time is
consequently greatly reduced with respect to the use of (8.54) or (8.55).
The rotation matrix C and the choice of variables

Rotation is usually performed with respect to an orthogonal system, making
use of different rotational variables. Quite common are the Eulerian
rotation angles el, 02, and 83, illustrated in Fig. 2.3(a): O 1 is the rotation
angle about the z axis and is positive when the rotation is clockwise looking
from the origin; O2 is a rotation about the new x axis and 83 a rotation about
the new z axis. The matrix C describing such a rotation is given in (2.32b).t
An appropriate rotation for the three angles will cover all the space (see p.
72), but if the Patterson map presents some rotational symmetry, the
rotation function will also have symmetry and a partial rotation will be
sufficient.
The symmetry of the rotation function is a combination of the symmetries
of the two Patterson functions, PC,,,,and P,,,.[~~] The Eulerian angles make
easy to describe the symmetry of the rotation function.[491Any triplets of
angles 0 , ) 82, and 8, can be considered as a point of a three-dimensional
system, whose unit cell has dimensions 2n in all directions: a rotation a is in
+
fact equivalent to a 2n. The resulting rotation space groups are some of
those described in the International tables for x-ray crystallography. $
A disadvantage in using 8 angles is that when O2 is small, 8, and O3
represent a rotation about nearly the same axis, and maxima will resemble
strips rather than maxima. The distortion effect can be avoided if a
combination of Eulerian angles is used instead:['']
A different possibility is the use of spherical polar angles, (p, I/J, and x
(Fig. 2.3(b)). Angles (p and I/J define a spin axis, and a rotation of around x
this axis is performed. Polar angles are very useful when a particular
direction has to be exploited or when a defined rotation has to take place, as
is sometimes the case for self-rotation (see p. 558).
Translation functions
Once the orientation of a known molecule in an unknown cell has been
found, the next step is the determination of its absolute position. Only when
one molecule is present in space group PI is this problem non-existent, since
in this case the origin of the crystal cell can be chosen arbitrarily with
respect to all three axes. In all the other cases, when the reference
molecule, exactly oriented, is translated in the unknown cell, symmetry-
related molecules move accordingly and all the intermolecular vectors
change: only when all the molecules in the crystal cell are in the correct
position, the calculated Patterson cross-vectors superimpose to those of the
observed Patterson (intramolecular vectors are insensitive to translation).
Figure 8.17 illustrates the method. Molecule 1 is positioned in the crystal
cell of the unknown structure in the correct orientation: s1 is the vector
defining its position with respect to the origin. Since we do not know yet the
correct position of the molecule in the cell, s1 is arbitrarily chosen. Molecule
2 is generated by the twofold axis, and its position is defined by vector s2.
The correct solution is shown in Fig. 8.17(a), where the correct origin of
molecule 1 is indicated by sy. As vector sl varies, all the intermolecular
vectors among symmetry-related atoms will change: they will coincide with
t In Chapter 2 the rotation matrix C is called RE, for Eulerian angles and R,, for spherical
polar angles.
$The reader must be warned that the Eulerian rotation matrix is not Hermitian, that is
reversing the order of the Patterson functions does not produce the same rotation-equivalent
positions.
Fig. 8.17. (a) 'Molecules' 1 and 2 correctly

oriented and positioned with respect t o a
symmetry element. s : is the vector from the
origin to an arbitrary point of molecule 1, and sz
the corresponding one for molecule 2, which is
generated from 1 by twofold rotation, to = sz -
sy is the translation vector from molecule 1 t o 2.
(b) Molecule 1 has been translated (but not
rotated, so that orientation is unchanged).
Molecule 2 has moved accordingly, and now all
vectors defining the molecular positions in the
cell are changed. Only when t = to the t w o
models, and consequently cross-vectors of
Patterson maps, superimpose. Determination of
sy from to is straightforward.
those of Fig. 8.17(a) only when s, = s:, that is when the local origin is
correctly defined with respect to the symmetry element. The determination
of the translation vector is performed by comparing two Patterson maps,
just as before in the rotation case, except that now we are interested in
maximizing the superposition of a different class of peaks. The problem
described above is in practice quite difficult, since the function representing
the superposition is generally very noisy and with many small maxima.
Several translation functions have been proposed, and some of them are
briefly summarized in Appendix 8.B. To illustrate the principles of
translational search, only the T function of Crowther and low['^] will be
described here, following the treatment of atm man.['^] In the case illustrated
in Fig. 8.17, the set of cross-vectors of the calculated Patterson from
molecule 1 to molecule 2 can be written as:
where p , and p, represent the electron density of the two molecules. If

molecule 1 is now translated, a new vector s, will define its origin. At the
same time molecule 2 will move into the cell, and a new function P12 can be
calculated for every value of s,. Since we are looking at intermolecular
vectors, it is more useful to define the translation as a function of vector
t = s2 - s l , which defines a local origin with respect to a symmetry element.
If Pohs(u)is the value of the observed Patterson at point u, the T translation
function is defined as:
Function Twill have a maximum when the two Pattersons superimpose. In

reciprocal space, (8.59) can be written:[511
T(t) = lohs(h)Fl(h)F~(hA)
exp (-2niht) (8.60)
h
where F,(h)
.\ is the Fourier transform of the model molecule 1 and Fl(hA)
/
the calculated structure factor of molecule 1 after application of symmetry

operation A. Looking at Fig. 8.17, since t = s, - s,:
T(t) = I,,,(h)F,(h)FT(hA) exp [-2nih(s2 - sl)]. (8.61)

h
The T function will have a peak at position s2- s1 = t o e s02- s1 0

(or
s l - s2= -to). The determination of sy from function T is equivalent to
solving a Patterson map with only two atoms in the crystal cell. In general
the T function does not need to be evaluated for the entire cell: if for
example two molecules are related by a screw axis along 2, the maximum to
will have coordinates (2x, 2y, 1/2), that is it will be confined only to section
2 = 112.
Other types of translation functions have been developed, and some of
the more commonly used in protein crystallography are summarized in
Appendix 8.B. A general review on translation functions is reported by
Beurskens et al. [531
Self-rotation and self-translation functions: improving the

electron density maps
Sometimes more than one molecule is present in the crystallographic
asymmetric unit. If we assume that they are identical, or at least very
similar, we can take advantage of the independent information present in
the structure factors. In fact, if we are able to identify the non-
crystallographic symmetry elements relating the independent molecules, the
electron density map can be averaged and substantially improved. The
presence of three molecules in the asymmetric unit has allowed the solution
of the structure of the haemaglutinin of the influenza virus using only SIR
phases
The self-rotation function is very similar to the general rotation function
defined in (8.54):
where P,(Cu) is the Pl(u) Patterson function rotated by matrix C, and U(u)
is a function which is 1 inside a sphere and 0 elsewhere. The function U is
necessary since both Patterson maps extend to all space, but we are
interested only in the superposition of self-vectors, confined to a region
around the origin of the cell. The sphere defined by U generally has a
diameter slightly larger than the maximum supposed molecular dimension.
The choice of polar rotation angles is quite common for self-rotation and
deserves a brief comment. Quite often the non-crystallographic symmetry is
represented by a rotation axis, in a direction different from the crystal-
lographic ones. In that event, the use of polar angles reduces the search for
the position of the axis from a three-dimensional problem to a two-
dimensional one: a twofold axis, for example, will correspond to a rotation
of 180" around the polar axis X, and a two-dimensional map (calculated for
q from 0" to 360°, Il,from 0" to 360", and x = 180") will show the presence of
the axis. A clear example of that is presented by Evans et a1.[551
The definition of the translational component of the non-crystallographic
symmetry represents the last and possibly the more difficult step. Let us for
example assume that the direction of a twofold non-crystallographic axis is
known. It has been shown[561that only the component of the translation

vector t in the direction of the axis can be determined precisely. The other
component of vector t, that is that perpendicular to the axis, is intrinsically
an imprecise parameter, unless the molecular structure is perfectly known,
which is not the case. In general, the self-translation function (analogous to
the T function previously described) is used to detect the existence of a
translational component of a rotational symmetry.
The steps and the possible different pathways described in the previous
paragraphs for the solution of a crystal structure of a macromolecule are
summarized in the scheme reported in Fig. 8.18.
I Crystallization I
I Space group determination I
Isomorpous derivative
preparation and
data collection replacement
Difference-
Patterson map *
1 1 Phase refinement 1- - 1 Difference-

Fourier maps
Yd, srdetc.
measurements derivative
Real space
Fig. 8.18. Simplified scheme summarizing some

of the possible steps in the determination of the
structure of a macromolecule. Some of them
I Electron density I : can be used alternatively or combined,
depending on the specific problem, that is size
map calculation and Refinement of the protein, previous knowledge of the
model building (see Fig. 8.23) structure, number of molecules in the
asymmetric unit, and so on.
560 ( Giuseppe Zanotti
Direct methods and the maximum-entropy principle in

macromolecular crystallography
Direct methods, extensively described in Chapter 5 , are not yet applicable
in the present form to the ab initio structural determination of a macro-
molecule. The main reason for that resides in the very large number of
atoms to locate: additional information is needed in order to increase the
reliability of the phase indication. Isomorphous replacement and
anomalous-dispersion data can be classified in this category, and indeed
they have been often used to locate the positions of heavy atoms. For
example coefficients (8.8) can be used with conventional direct methods
programs and they can be very helpful in the case of a very large number of
sites.LS7] Furthermore, classical direct methods have allowed, using
coefficients ( F L - F,), the localization of the positions of anomalous
scatterers in m e t a l l ~ ~ r o t e i n s . [ ~ ~ ~
Quite promising, but not yet fully exploited, is the use of direct methods
for the determination of the phase of the protein structure factor taking
advantage of isomorphous replacement and anomalous-dispersion data (for
a review see ~ i a c o v a z z o ) [ ~Let~ ] . us call @,(h) and @,(k) the phases of
reflections h and k, respectively, for the native data set. @pH(h)and @pH(k)
are the phases of the corresponding reflections of the derivative. In addition
to the classical triplet invariants introduced on p. 338:
we can make use of triplets containing mixed terms, that is relationships

between phases of the two crystals:
The eight different combinations of the six phases give rise to eight
conditional probability distribution for Qj, whose formulation is not
discussed here. An important point, however, must be stressed: the
probability distributions do not depend only on the total number of atoms
present in the crystal cell (see Appendix 5.C), but also on the difference
between native and derivative data. The reliability factor for a triplet phase,
G, defined in (5.37) and (5.C.16), in the presence of isomorphous data must
be rewritten:[60]
where Np and NH are the number of protein atoms and heavy atoms in the
unit cell, respectively, and Ah represents the normalized difference for
reflection h. The second term of (8.63) may substantially increase the value
of the reliability factor G. A relevant application of (8.62) is that one single
derivative allows in principle the determination of the phase of the protein
structure factor.
A similar approach can also be used in presence of anomalous dispersion
data, where in addition to the classical triplet invariants:
six other equations containing mixed terms, analogous to (8.62b), can be

introduced. The only limitation in the previous treatment is represented by
the very high degree of accuracy necessary in the measurement of the
intensities, particularly in the case of anomalous scattering data. Neverthe-
less, anomalous dispersion techniques are becoming more relevant, due to
the possibility of performing multi-wavelength experiments with syn-
chrotron radiation.
An apparently different approach to the solution of the phase problem is
based on the maximum-entropy principle, the application of which to
crystallography was proposed by Collins[61]and that has been shown to be
virtually equivalent to direct methods by ~ r i c o ~ n e .If' ~p(r)
~ ] is a density
distribution, its entropy is defined as:
H =- \ p (r) ln p (r) dr.

J V
(8.65)
The concept of entropy is borrowed from information theory, where H

measures the expected value of the total information. The 'better' distribu-
tion is that which maximizes H and which is at the same time consistent with
other observations: in our case, obvious conditions are that the electron
density over the entire cell is equal to F(000) and the Fourier transform of
p(r) is in agreement with the diffraction data. Equation (8.65) can
consequently be modified to include constraints. If some sort of previous
density distribution is available, the relative entropy can be defined:[631
where q(r) is a distribution inferred by a priori information: in the absence

of other information, that is if q is constant, (8.66) reduces to (8.65). The
maximum-entropy method has been used to expand a set of initial phases,
in conjunction with other t e ~ h n i ~ u e s . Besides,
[ ~ ~ , ~ ~the
I structure of 15 base
pair of DNA (a structure of about 600 atoms) has been recently solved ab
initi0:[~~1
starting phases were determined by conventional direct methods
and expanded using the maximum-entropy algorithm. Although this ap-
proach has been demonstrated to fail in other simpler test cases, promising
applications of direct methods to macromolecular crystallography are
expected in the near future.
The interpretation of electron density maps and

the refinement of the model
The interpretation of electron density maps

Once a phase angle estimate for the protein structure factors is available,
the calculation of an electron density map is straightforward, using (see p.
170):
p(r) = Fp(h)exp (-iq,) exp [-2ni(h r)] (8.67)
h
where the Fourier coefficients are usually weighted by their figure of merit.
The initial interpretation of the map is in general not easy, unless very good
phases at high resolution are available, which is seldom the case. Therefore,
the strategy generally adopted is to calculate first an electron density map at
low resolution, say 5-6 A: these maps allow us to identify the contours of
the molecules in the crystal cell, and to distinguish between solvent regions
and protein. Eventually, some elements of secondary structure can be
identified: a-helices will appear as cylindrical rods of diameter of about
4-6A. 0-sheets are more difficult to distinguish, and in any case single
0-strands are not visible.
When the position of the molecule has been located in the unit cell, a
map at medium resolution, say 3.5-2.5 A resolution, is calculated and an
attempt to trace the polypeptide chain is made. Chain trace at this
resolution is made easier, and sometimes possible, if the amino acid
sequence is known. Mistakes are quite common in the interpretation at
medium resolution: the connections among secondary structure elements
are often difficult to recognize and amino acids can be positioned along the
chain shifted from their correct position by one or more residues.
Higher-resolution phases, say 2 A or more, allow us to correct for this
kind of mistake and to locate more accurately the amino acid side chains.
Unfortunately, MIR phases very seldom extend to that resolution, and
high-resolution maps can be obtained using calculated or combined phases,
as will be discussed later.
Interactive computer graphics and model building

Low-resolution maps are traditionally drawn on a small scale on transparent
sheets and are known among protein crystallographers as 'minimaps': they
are very useful in giving a global view of the electron density of the unit cell.
Such maps can also be used at medium resolution, since they allow a
preliminary, approximate tracing of the polypeptide chain.
The complete building of the molecular model in the old days was
performed using an apparatus, generally home-made, called an optical
comparator or 'Richard's box', from the name of its inventor.[661Nowadays,
the interpretation of the electron density can be entirely performed on a
graphic display: the map is shown on a video and the operator is allowed to
fit a piece of chain into the density. For this purpose, the modelling software
most popular among protein crystallographers is FRO DO.[^^,^^] The prin-
ciple of the program is that objects displayed on the screen are divided in
Fig. 8.19. 'Chicken wire' representation of the

electron density of a side chain of a histidine
residue. Maps are calculated at a resolution of
2.5 A, with coefficients 2F0,, - F,,,,
and
calculated phases. Drawing produced by an
Evans and Sutherland PS330 graphic station,
using program FRO DO.'^"
two categories: those that can be manipulated, called 'foreground objects',

and those that cannot be modified interactively, defined 'background
objects'. Background and foreground models can be seen superimposed on
each other. Program FRODO contains a dictionary of stereochemical
information on natural amino acids and groups often occurring in protein
molecules or nucleic acids. From this dictionary, building of a part of a
polypeptide chain or a portion of a macromolecule in the preferred
conformation is straightforward. The electron density, which does not need
to be modified, is displayed as a background in a 'chicken wire' repre-
sentation or similar, and the atomic model, or a part of it, as a foreground
object. Figure 8.19 illustrates a simple model as it appears on a calligraphic
screen. Atoms can be easily manipulated, by a simple rotation of a dial or
movement of a light pen: they can be moved, alone or in groups, rotation
around dihedral angles performed, or bonds stretched. The fitting of the
built model into the electron density can be fast, if the starting phases are
good enough; otherwise, a lot of time can be spent in trials. A convenient
method to build the initial model of a protein has been devised by Jones and
~ h i r u ~ based
, [ ~ ~on
] the idea that short elements of secondary structure can
be taken from a data base consisting of coordinates of a restricted set of well
refined protein structures: once some a-carbons (say 10-15) are roughly
positioned into the density, the piece of model from the data base that
better fits these atoms is searched for and used.
The only disadvantage in the use of a graphic system during electron
density map interpretation is that a global view of the molecular model is
practically impossible, since a drawing of all the atoms of a protein gives
results which are quite confusing. Nevertheless, people who built protpjy
models using a 'Richard's box' are strongly in favour of interactive graphic
systems.
The refinement of the structure

The refinement of protein structures, with very few exceptions,[701cannot be
carried out using the classical least-squares methods. This is not due to the
size of the problem, since nowadays computers are powerful enough to
handle systems of equations containing thousands of variables, but to the
limited number of X-ray data. It has been shown in fact on p. 98 and 104
that, for an accurate definition of the parameters, the system must be
largely overdetermined, that is the ratio of observations to variables (atomic
Table 8.3. Number of theoretical independent reflections at

different resolutions for a protein crystal with one molecule of 182
amino acids in the asymmetric unit. The solvent content is 40 per
cent. The number of parameters is 4408 (1469 atoms times 3) if an
overall B factor is considered, or 5876 if an individual isotropic B
thermal parameter is assigned to each atom
Resolution Independent Ratio obs./var. Ratio obs./var.

range (A) reflections (x, y, z ) (x, Y, 2, 6)
coordinates, thermal factors, and sometimes occupancy) must be of the

order of 10 or so, and this is indeed the case for small molecules, where
diffraction data can be collected to a spacing of 0.7 8, or even more. Protein
crystals are intrinsically less ordered: diffraction data are often measured to
a resolution of 3.0-2.5 8,, sometimes 2.0 8,. A resolution of 1.5 8, can be
considered quite good, and only in exceptional cases have 1.0 A data been
collected.[231A typical situation is illustrated in Table 8.3, where the number
of independent reflections for a medium-size protein (182 amino acids) with
a solvent content of 40 per cent are calculated at different ranges of
resolution. For a ratio of observations to parameters of 10 it would be
necessary to collect all possible diffraction data to a spacing of 1.28,, a
resolution never attainable for such a type of crystal. For the above-
mentioned reasons, the first attempts to refine a protein structure were
performed in real space.[711The method seeks to minimize the difference
between the observed electron density, p,,,, computed by (8.67), and a
calculated model density, peal,, obtained by assuming a Gaussian distribu-
tion of the electron densities centred at the atomic positions of the current
model:
This technique suffers from all the drawbacks of a real-space refinement

procedure, the most relevant being that if poor phases are available, p,,,
will be quite incorrect and the convergence of the method very slow, or it
will not converge at all. There are many reasons to favour the reciprocal
space refinement methods and different solutions have been proposed to
overcome the problem of the underdetermination of the system: in fact,
improvement of the ratio of observations to parameters can be achieved by
decreasing the number of variables or by artificially increasing the number
of observations. The former is called constrained least squares and the
latter restrained least squares.
Constrained versus restrained least squares

Constrained or rigid-body refinement? is a well-known and widely used
technique in crystallography (see p. 105): when the geometry of a group of
t Despite the distinction between them described on p. 105, rigid-body and constrained
refinement are taken as synonyms i n this chapter.
atoms is accurately known and there are reasons to believe that it will not
be significantly modified by the environment, the entire group can be
treated as a rigid entity. In the classical case of a phenyl ring, the eighteen
positional variables can be reduced to only three translational and three
rotational.
Bond length and valence angles in amino acids are very well known from
the structures of hundreds of small peptides. In a protein, they can be held
fixed to their theoretical values and only torsion angles around single bonds
allowed to vary. This approach was used by ~ i a m o n d [ ~in ' ] real-space
refinement, but it can be used in reciprocal space as well. Taking into
account the fact that the peptide bond can be considered planar, only two
torsion angles, called g, and I) (see Appendix 8.D), need to be varied for
the backbone chain of every amino acid: for a protein of n residues, the
parameters are reduced to about 2n for backbone plus the torsion angles of
side chains. An illustration of a possible choice of constrained parameters is
reported in Fig. 8.20 for a simple dipeptide. This solves the problem of
underdetermination, but the model becomes in some way too rigid and the
radius of convergence, that is the maximum displacement allowed for an
atom in a wrong position to be corrected, becomes quite small.
Constrained least squares can be applied to very different extent: the
definition of rigid body can be applied to only some group of atoms or to
the entire molecule. If, for example, an approximate solution of the
structure has been found using the molecular replacement technique, a first
refinement can be performed by considering the entire protein (or a
subunit) as a rigid group and the best position in the new crystal cell can be
searched for using only three translational and three rotational variables. In
that event, there is the supplementary advantage that, since the number of
variables is very limited, only low-resolution data need to be included in
refinement, greatly increasing the radius of convergence of the method.
Increasing the number of observations is another possible solution of the
underdetermination problem in macromolecular refinement (see p. 107).
Information from other sources can, in fact, be introduced and treated in a
way similar to that used for observations coming from X-ray diffraction. The
use of geometrical restraints has been proposed by Konnert and
end ricks on,[^^'^^] following a procedure devised by ~ a s e r [ for ~ ~ small
]
Fig. 8.20. Schematic drawing of the dipeptide

phenylalanine-alanine, used to illustrate the
constrained least-squares technique. Arrows
indicate free rotation about the bond. All the
bond lengths and valence angles are held fixed,
and the peptide group and the phenyl ring
planar. The total number of variables amount t o
eleven: three rotation and three translations (not
indicated in figure) plus five internal torsion
angles. (Some of the hydrogen atoms are
indicated in figure only for clarity, but they are
usually not taken into account i n the
refinement.)
molecules. In addition to the classical quantity minimized in crystallographic

least squares:
where the summation is extended to all the i reflections, other observational

functions can be added. Since distances and valence angles of amino acids
are well kitown and they are not expected to deviate significantly from the
ideal value, instead of considering them as fixed, we can also minimize:
d,(ideal,
is the ideal value for the specific distance we are considering, d,(,,,,) is
that calculated from our present model and wi is usually chosen as the
reciprocal of the standard deviation of the distribution expected for the
distances of type j. Notice that since di(calc)is a function of the atomic
coordinates, (8.70) does not increases the number of variables. The total
number of equations like (8.70) is equal to the distances that are restrained:
bond lengths, the distances between one atom and the next-nearest-
neighbour (which is equivalent to restraining valence angles), and the
first-to-fourth atom distances, where the dihedral angle described by the
four atoms is in some way fixed (this is for example the case of the planar
peptide bonds). An example of the number of distances that can be
restrained for a simple dipeptide is illustrated in Fig. 8.21. Other possible
restraints in the Hendrickson and Konnert formulation are:
S3 represents the sum of the deviations of the atoms i from the plane k,
which is defined by its unit normal m k and by the origin to plane distance
Fig. 8.21. The same dipeptide as in Fig. 8.20

illustrates the restrained least-squares
technique. The coordinates of any atom are
allowed t o vary, but the stereochemistry is
preserved by applying restraints on bond
distances (full lines), bond angles (broken lines),
torsion angles (dotted lines), and planarity. Non-
bonded contacts are not shown i n the figure.
(Adapted from Sussman, ref. [3],Vol. 115, p.
274).
d k ; ri is the vector that defines a point i whose distance from plane k has to
be minimized.[751S, restrains the volume of chiral atoms, defined for an
a-carbon by the product of the interatomic vectors of the three atoms
bound to it:
Since the sign of Vc, depends upon the handedness, S4 restrains chiral
centres to their correct configuration (Fig. 8.22). S, is applied to all
non-bonded atoms (except those taken into account in S2) and avoids too
close contacts. Other kinds of restraints can be considered, i.e. on isotropic
thermal parameters, occupancy, and non-crystallographic symmetry. It may
sometimes happen, particularly during the first stages of refinement, that
some part of the structure is poorly determined and the model 'blows up'.
Fig. 8.22. The chiral volume for an cu-carbon
In that event, a restrain on the excessive shifts can be applied: atom. The central atom is chosen as the origin
of the coordinate system, and vectors (r, - ,),r
(rc8 - ),r and (r, -),,r (8.74) are denoted s,
sc8, and s, respectively. The cross product
s, X sC8 is a vector perpendicular to the plane
CC,C8. If it is on the same side of vectors,, as
where rk and ro are the atomic vectors of the target and the initial model in the figure, that is if angle 0 is less than 90".
respectively. Using eqns (8.70)-(8.74), the number of observational func- the dot product between the t w o vectors is
positive. If s, and sq are reversed, that is if the
tions is now greatly increased from the original number, represented by eqn wrong configuration IS chosen for the cu-carbon,
(8.69). Equation (8.75) has effect only on the diagonal terms of the normal the vectors, x sc8 points i n the opposite
matrix. The number of restrained parameters for the example described in direction and the value of V, becomes
negative.
Table 8.3 is shown in Table 8.4.
It must be remembered that in protein crystallography 'experimental'
phases are very often available. They can be included in least-squares as an Table 8.4. Number of restraints, following
additional information that imposes another Hendrickson and ~ o n n e r t , ~ ' for
~ ' a protein
molecule of 1469 atoms (excludina H) in the
(8.76)
'97 = Ci wp(qi(obs)- 2
Q)i(calc)) . asymmetric unit. The example is Ye~ativeto
the case of Table 8.3
qobsis the estimate of phase angle from isomorphous and anomalous data Number of distances:
and qolCis the phase calculated from the model. Weights for (8.76) must bond distances
take into account the cyclic nature of phase angles. angle distances
planar 1-4 distances
Phase information is also used by Lunin and ~ r z h u m t s e v . [They
~ ~ ] suggest
Planes
that only differences among crystallographic quantities be minimized, that is
structure factor amplitudes and phases. Since phase probability distribution Chiral centres
may be represented by (8.34), they assume an analogous probabilistic Torsion angles
distribution for the module of the structure factor F for reflection i of the Possible contacts:
form: contacts due t o single torsion
contacts due to multiple torsion
P(F;) = exp [ - ( F : -F ~ ( , , , , ) ~ / ~ u ~ ] , (8.77) possible H-bonds
and if structure factors moduli and phases are assumed to be mutually

independent, the joint probability distribution will be given by the product
of (8.34) and (8.77). The most probable model will consequently be that
which minimizes:
S=
i
{ ( 1 / 2 a 2 ) [-~ F:(ob,,]2
f +
- [ A cos qi B sin q i
+ C cos 2 q i + D sin 2 q i ] ) . (8.78)
Using (8.78) the multimodality of the phase distribution is taken into
account.
A different approach to using restraints has been proposed by Jack and

~ e v i t t : ' instead
~ ~ ] of restraining stereochemistry, they minimize:
where D represents the difference among observed and calculated structure

factor amplitudes given by (8.69) and E is a potential-energy function:[801
+ C k,{l + cos (me, + 6)) + C (Ar-'' + BrP6). (8.80)
The four terms on the right-hand side describe bond, valence angle,
dihedral torsion angle, and non-bonded interactions, respectively. Kb is the
bond stretching constant and K, the bond angle bending force constant; k,
is the torsional barrier and m and 6 the periodicity and the phase of the
barrier. A and B the repulsive and the long-range non-bonded parameters.
The summation extends to the j bonds, the I valence angles, the 8 torsion
angles, and the n non-bonded interactions between all pairs of atoms
separated by at least three bonds. Despite the apparently very different
approach, the energy minimization and the geometrically restrained least-
squares are not too different in practice, since the final effect of (8.80) is to
impose restraints on the model.
Whatever method is used, special care is needed about the weights
applied to the different functions. We are in fact dealing with non-,
homogeneous quantities, like structure factor amplitudes and interatomic
distances, so the weights of the relative observational functions must be
chosen in such a way that everything is put on the same scale: an
overestimate of geometric restraints will in fact produce a stereochemically
perfect model associated with a very high crystallographic R factor; on the
contrary, an underestimate of the same weight will result in a good R factor
with unreasonable bond lengths and angles.
Restrained and constrained least squares

The two methods described above, restrained and constrained least
squares, can be combined:[s11the molecule(s) is(are) considered as made up
of rigid groups, and restraints are applied to distances among such groups.
The quantity minimized, S, is the sum of three terms:
where DF is (8.69), DD restrains the stereochemistry, analogously to eqns

(8.70)-(8.73), and DT restrains the structure from moving away from the
starting set of coordinates, (8.75). All of the terms of (8.81) are functions of
the atomic coordinates, generally referred to an orthogonal reference
system. If a subset of these atoms is considered as a rigid group, S can be
expressed, for that particular group of atoms, as a function of six rigid-body
parameters, three rotational and three translational, and an arbitrary
number of torsion angles, that is:
where ti and R, are the translation vector and the rotation matrix of the
entire group i, Yi,, . . . , Vim are the m torsion angles and B,,, . . . , Bin the
n temperature factors of group i.
Since the definition of rigid group is left to the user, the entire molecule
or a portion of it can be constrained, or eventually some subunits. The
restrained-constrained approach was originally devised for nucleic-acid
refinement, but it has been successfully used in refinement of protein
structures too.[822831A computationally quite efficient method of combining
sterical restraints and rigid-body refinement has also been recently
proposed. [s41
Crystallographic refinement by molecular dynamics

The development of vectorial and parallel computers offers nowadays the
possibility of performing molecular dynamics calculations on complex
systems, including proteins in the crystal state. The application of molecular
dynamics calculations to macromolecules is a quite widespread technique
(for a review, see Karplus and ~ c ~ a r n m o n [ ~but
~ ]its
) , introduction in the
crystallographic refinement of protein structures had been proposed only
re~ently.[~~,~~]
Molecular dynamics of free atoms consists in solving the classical Newton
equation of motion:
To take into account the effect of the medium and the approximations used
to calculate the total energy, dynamical effects can be better represented by
a set of Langevin equations:
where bi is a frictional coefficient, used to prevent atoms from moving away

too much from their original positions, kB is Boltzmann's constant, To the
temperature, and i ( t ) a random force with Gaussian distribution and
properties:
The simulation starts from an initial set of coordinates. To each atom is

assigned a velocity, usually at random from a Maxwellian distribution
corresponding to the temperature selected, and eqn (8.83) or (8.84) is
integrated at a given temperature for a given time.? New velocities are then
assigned, eventually at a new temperature, and the calculation continued.
The simulation is normally performed for a short period of time, usually of
the order of few picoseconds.
In the crystallographic refinement of macromolecules by molecular
dynamics, the X-ray information is used to restrain the energy of the
system. The total potential energy is, in fact, considered as the sum of two
E,,, represents an empirical potential energy, analogous to that defined by

(8.80), E,, is a sort of 'experimental' potential energy, and is considered as
t Numerical integration can be performed using for example the Verlet algorithm.
the sum of three terms:
Exrayin (8.87) describes the difference between observed and calculated

structure factor amplitudes:
w, is a factor which puts Exrayon the same scale as the empirical potential
energy term and NA is given by C w,(F,,,)~, to ensure that wA is
independent of the resolution range used. The terms E p and ENBcan be
included to take into account experimental information about MIR phases
and crystal packing respectively.
Molecular dynamics simulations can be performed at ambient
temperature,[871or at higher temperature, as in the version called simulated
annealing.[s69881 The latter consists in starting the simulation at room
temperature, say 300 K, and heating up the system (for example at
2000-4000 K) and subsequently cooling down at the initial value. The
advantage of going to high temperatures, unreasonable from the biological
point of view, is that model can come out of local minima, and the ratio of
convergence of the method is increased with respect to classical least
squares. t
The result of molecular dynamics calculations is a family of conforma-
tions, but the constraints imposed by X-ray data restrict these conforma-
tions to all those with the lower crystallographic R factor.
The strategy of the refinement of protein structures

The initial model of a protein structure is, very often, not good enough to
allow for a fully automated refinement. Indeed, if some serious errors are
present in the model, for example the polypeptide chain is positioned more
or less correctly but the amino acids are shifted one residue or more along
the chain, an automatic procedure will hardly recover from that error.
Besides, the radius of convergence of constrained or restrained least-squares
methods can be evaluated to be around a half the resolution of the data
used, that is not more than 1-1.5 A. At the beginning of the refinement, to
speed up convergence, medium-resolution data (3.0-2.5A) can be
employed. Since the number of observations at that resolution is quite low,
an overall temperature factor for all the atoms is used. Afterwards, the
resolution is gradually extended, solvent molecules included, and isotropic
individual B factors applied.
The same seems not to be true using the simulated annealing technique,
which allows a more rapid and automatic convergence: the heating makes it
easier to get out of the false minima without manual intervention. Some
more experience nevertheless must be accumulated. A fully automated
refinement was possible in the test case of the enzyme aspartate aminotran-
sferase, refined with data at 2.8 A resolution starting from MIR
t T h e term temperature must be regarded cautiously here: it does not indicate a physical
temperature, but rather a parameter controlling the refinement. The simulated annealing is in
fact virtually equivalent to the Metropolis algorithm (Metropolis, N., Rosenbluth, M.,
Rosenbluth, A , , Teller, A. and Teller, E. (1953). Journal of Chemical Physics, 21, 1087-92).
coordinates.[89] A careful comparison between the model of myo-

haemerytrin refined, starting from the same model, in one case with several
cycles of restrained least-squares and manual rebuilding and in the other
case with the 'simulated annealing' technique without manual intervention
has been reported.[901The two structures compare quite well, but molecular
dynamics procedure could not bring the refinement to completion in a fully
automated way: manual intervention is still necessary to correct for gross
errors (say more than 3-5 A in the main chain) and to include solvent
molecules. Nevertheless, simulated annealing can save a lot of human
effort, at the expense of quite a long computational time.
For the above mentioned reasons, some cycles of automatic minimization
are usually followed by recalculation of the electron density maps and
manual adjustment or rebuilding of the model.
In recalculating electron density maps, a major problem is the choice of
the phases and the coefficients to be used. MIR phases suffer from all the
errors described on p. 546, and the isomorphism of heavy-atom derivatives
does not extend generally beyond 3.A or so: high-resolution electron density
maps are seldom achieved with MIR phases. On the other hand, calculated
phases tend to reproduce the model used in calculating them, and an
electron density map obtained wih calculated phases may be strongly
biased. For these reasons, phases coming from independent sources, e.g.
the phases from isomorphous derivatives data and those calculated from the
model, can be combined to produce an improved electron density
The probability distribution of calculated phases, Pcalc(q),can be evaluated
by using a procedure due to ~ i m (see [ ~Appendix
~ ~ 5.H, p. 393) and can be
used, along with the 'experimental' probability Ptot(q), to obtain a
combined probability distribution:
Pcomb(c~>
= Ptot(~)Pca~c(q). (8.89)
The new figure of merit, mcomb,obtained by (8.38) can be used to calculate
a best combined electron density map. If only calculated phases are
available, the Sim formula can be used to weight the Fourier coefficients. A
scheme of the possible refinement procedure is illustrated in Fig. 8.23.
The coefficients and phases more commonly used for Fourier syntheses
are listed below, but other combinations of them are possible:
mF0bs exP ( i q ~ ~ ~ ) (8.90)
W C O Fobs
~ ~ exP ( i q c ~ ~ b ) (8.91)
W S IFobs
~ - &a~cI exp (iqca~c), (8.92)
- & a d exP (i~comb),
~cornb(~~obs (8.93)
~ ~ l r n ( ~-
~&o ba ds exp (i~calc). (8.94)
Equation (8.90) gives the coefficients of a classical observed Fourier
synthesis. In principle, they could be used during all the stages of the
refinement, but MIR phases can be improved by phase combination.
Furthermore, very often they do not extend to high resolution, and
calculated phases must be used instead, when a reasonable atomic model is
available. To reduce bias, a Fourier map with combined phases can be
calculated. Coefficients (8.92) correspond to Fourier-difference maps with
calculated phases. If they are calculated from a partial model, they are
Starting model
(see Fig. 8.18)
Calculation of new
Some cycles of least- electron density maps
Fig. 8.23. Block diagram summarizing the squares refinement with new phases
phases of structural refinement. The starting
model is obtained by one of the procedures
schematized in Fig. 8.18. The numbers of
iterating cycles necessary t o reach convergence --. Structure factor
can be quite variable, depending o n the quality
of the initial model. If M.I.R. phases are not
available (i.e. the structure has been solved b y I calculation
molecular replacement techniques), only maps
with calculated phases can be used. Some of the
coefficients used in electron density map I End of refinement I
calculation are described on p. 571.
known as omit maps and can be useful in positioning portions of the

molecule that did not appear clearly in the M.I.R. maps. Coefficients (8.93)
and (8.94) correspond to a combination of a Fourier electron density map
and a difference-Fourier: if Fobsand F,,,, are very similar, the magnitude of
the coefficients approaches to (8.91); otherwise, terms with Fobsgreater than
F,,,, will have a higher weight. The practical result is an enhancement of the
regions of density where severe errors are present in the model. The weight
used for map calculation can be a modification of the Sim scheme or the
weight obtained by the combination procedure.
It must be finally remembered that the crystallographic R factor of a
refined protein structure is not comparable with that of a small-molecule
crystal. Besides, owing to the low ratio between observations and variables,
the R factor alone does not represent an index of the reliability of the
model: molecular stereochemistry must be correct also. For example,
deviations from the ideality of bond lengths, valence angles, and restricted
torsion angles of about 0.01 A, 2", and l o , respectively, are considered
reasonable. The crystallographic R factor of a crystal structure refined with
these restraints at 2 A resolution can be reduced to a value between 0.15
and 0.20.
Protein structure
The three-dimensional structure of a globular protein is the resultant of a
very large number of interactions. This makes it at the same time stable and
flexible: a modification in a specific site can, in fact, be assimilated by small
local adjustments, without altering the overall conformation. In other cases,
modifications taking place in a region of the macromolecule can be
transmitted and influence the conformation in zones far apart from where
the movement originated, a phenomenon called cooperative or allosteric
effect.? It must also be remembered that the medium may have a relevant
influence on the conformation of the macromolecule, that in vivo operates
in a non-homogeneous environment, like for example the cell cytoplasm or
the cellular membrane. Flexibility can consequently be useful, or even
necessary, to the protein to fulfil its biological functions. On the contrary,
the picture of a molecular model resulting from the X-ray structure
determination is, at least apparently, a static one, the only 'dynamic'
information residing in the atomic temperature factors. Nevertheless, this
must not be considered in contrast with reality, since the X-ray structure
represents an average, over time and space, of single structures around a
conformational minimum, which is representative of the conformation of
the protein molecule in a solution of the same medium. On the other hand,
a description of a dynamic situation is quite difficult, particularly for a
complex molecule: the sort of static image that will turn out from the
following discusion must be considered a necessary simplification as well as
the starting point for the comprehension, at the molecular level, of the
behaviour of biological systems.
General aspects
Proteins are polymers of the 20 natural aminoacids (some symbols and
conventions relative to amino acids are reported in Appendix 8.D). They
may contain other groups, often relevant for the conformation of the
molecule, like haems, prosthetic groups, carbohydrates, and so on, or they
can coordinate cations (see p. 582). Nevertheless, the building blocks of
proteins are the amino acids, and the protein fold* is the first and perhaps
more general aspect of its structure.
A polypetide chain of a given sequence, in the appropriate conditions,
can refold spontaneously into its final three-dimensional structure:[931all the
necessary structural information must be contained in the amino acid
sequences. It should therefore be possible, at least in principle, to predict
from it the structure of a protein. Unfortunately, despite several efforts, this
remains a hope for the future. Moreover, some simple, schematic rules
relative to the forces that contribute to stabilize a structure can be
summarized: they are, however, only indications and exceptions are quite
comm~n.[~'~~]
1. Groups potentially able to form hydrogen bonds will be positioned
accordingly: or at the surface of the molecule, where they can be
hydrogen bonded with the solvent, or, when in the interior, a hydrogen-
bond donor will be close to an acceptor. Not necessarily all the possible
interactions of this kind will be realized, but most of them usually are.
2. Polar residues in a water-soluble protein will preferably be located on
the surface of the macromolecule and the hydrophobic ones in the
interior (the contrary apply for lipid-interacting proteins).
3. Charged groups will be close to the surface, where they can be
t Cooperativity and allostery are not exactly synonymous, but in this context no distinction is
made about them.
I The term 'fold' is not used here to indicate a process, but simply the way in which the main
chain is wrapped on itself.
neutralized by solvent ions. When forced to be inside, they will be

stabilized by a proper counter-ion.
As a corollary of the previous rules, particularly rule (2), the three-
dimensional structure of the proteins will be as compact as possible. For
example, the packing density of ribonuclease S, calculated approximating
the atoms to spheres of van der Waals radius, is about 0.75,[951approxi-
mately the same value found for close-packed spheres.
Levels of organization of proteins: secondary structure

The concepts illustrated before, despite generalities, refer to single interac-
tions and in practice they are not well suited to describe overall features or
general aspects of the structure. A different, and perhaps more useful,
approach to understanding the structure of a protein is the identification,
under the apparent complexity, of regular schemes and motifs. Looking at
the structure under this aspect, a hierarchical organization becomes
apparent and several levels of organization, of different complexity, can be
distinguished. The first local order is imposed by the hydrogen bonds
formed between the N-H and the carbonyl oxygen of the polypeptide
chain: the regular structure generated by these quite strong interactions is
called secondary structure? and results in a limited number of possible
conformations, briefly described below.
1. Helices. The more common among the repetitive motifs found in
polypeptide chains is the a-helix. The theoretical a-helix, predicted by
Pauling et al. in 1951, is characterized by 3.6 residues per turn and a pitch of
5.4 A. A hydrogen bond is formed between the carbonyl oxygen of residue
+
n and the N-H of residue n 4. The number of atoms covalently bound in
the ring closed by the hydrogen bond is 13 (Fig. 8.24). Along with the
number of residues per turn, this number is sufficient to characterize the
secondary structure: the a-helix is consequently called a 3.613-helix.Other
less common helical structures are the 310-helix and the n- or 416-helix,
characterized by an exact repetition after three and four residues respec-
+
tively and a hydrogen bond between residues n and n 3 or n 5. A special +
case is represented by the a,,-helix, which is a sort of mixture between the
a- and the 3,0-helix.
Every type of helix can also be defined by its g, and q torsion angles:
their indicative values for the most common repetitive structures are
reported in Table 8.5. Of course, real helices present backbone torsion
angles that deviate from the ideal ones and, in addition, bending and local
irregularities are observed.[961Nevertheless, they are the most regular
structures found in proteins. Since natural amino acids, except glycine, are
chirals, right-handed and left-handed helices are energetically not equiv-
alent: helices found in proteins are in practice always right handed. Amino
acids assuming the torsion angles characteristics of a left-handed helix, when
found, are confined to very short pieces of chain (one or two residues),
often at the end of a right-handed helix connecting a P-strand.
t For historical reasons, the amino acid sequence is called primary structure.
Fig. 8.24. (a) Stereo view of t w o a-helices of
m y ~ ~ l o b i n(data
[ ~ ~from
] PDB), connected by a
short piece of chain (to simplify the
representation, for every side chain only the
@-carbonis drawn). Helix 7 goes from residue
0C 100 to 118 and h e l i x 8 f r o m 125 t o 148. Residues
from 116 t o 119 participate also i n a type I
@-turn.From I 1 9 t o 124 the structure is less
0N regular. The t w o helices are not exactly aligned,
but their axes are slightly bent, a classical
situation of the packing of helices, (b) Scheme of
00 the hydrogen bond pattern for the most
common types of helices: in the a-helix, 13
atoms are included in the ideal ring from the
O H carbonil oxygen t o the N-H. Only 10 atoms in
the 3,,-helix.
Table 8.5. Approximate torsion angles for some

common regular structures (from IUPAC-IUB
commission on biochemical nomenclature,
Journal o f Molecular Biology, 52, 1-17 (1970)).
(References about the sources of the reported
values can be found in the reference; w is as-
sumed to be 180")
Right-handed a-helix
Left-handed a-helix
3,,-helix
Parallel-chain plated sheet
Anti-parallel-chain pleated
sheet
Polyglycine
Polyproline II
One of the reasons that helices are quite common is the possibility to
accommodate in a very short length a large number of residues: all side
chains in fact point away from the helical axis, giving rise to quite an
efficient packing of bulky groups, one on the top of the other.C9'] The only
residue that cannot be accommodated in a whelix is proline, due to the
steric hindrance of its side chain and to the fact that its nitrogen is lacking
the hydrogen necessary to form the hydrogen bond. Proline can be fitted
properly only in the first turn of a helix, but will interrupt it, or introduce a
bend, if present in any other place. A special kind of secondary structure,
called polyproline helix, has sometimes been observed.
Another important peculiarity of helical structures is represented by their
electrostatic properties: since all the dipoles of the peptides are pointing in
the same direction, the result is a net total dipole for the helix[981that may
contribute, for example, in stabilizing the binding of charged species.
2. ,&structures. A 0-strand is a portion of polypeptide chain in a nearly
extended, pleated conformation. It is less regular than helices and its torsion
angle values reported in Table 8.5 must be considered only indicative, since
large variations are observed in practice. A P-strand is normally not stable
in itself, but it is stabilized by the contact with another P-strand, with which
it forms hydrogen bonds. According to the relative direction of the two
chains? 0-strands can be parallel or antiparallel (Fig. 8.25): they differ in
the way the hydrogen bonds are formed, making the latter slightly more
stable[1011and more commonly observed in proteins. Since P-chains are
quite flexible, this kind of structure can present irregularities and defects: a
jump of one residue in a strand is for example called a P-bulge.
Several parallel or antiparallel P-strands form a so called P-sheet. Since in
a 0-strand amino acid side chains point alternatively in opposite directions,
a P-sheet has the possibility of having a polar surface and an apolar one.
Finally, strands in a sheet are not exactly aligned, but are twisted with
respect to each other.[lo2]
Fig. 8.25. (a) Schematic drawing of the

hydrogen bond pattern for different types of
6-sheets. The first t w o O-strands run antioarallel.
ihe third is parallel t o the second one (diiections
of strands are indicated by arrows). Hydrogen
bonds are differently oriented i n the t w o cases,
the antiparallel being more favoured and more
commonly found in proteins. A plus sign on the
cu-carbon indicates the side chain coming up
from the sheet, and a minus a side chain going
away from the reader. (b) Stereo view of three
~ ~ ' from PDB,
P-strands of r i b o n ~ c l e a s e [ ' (data
only Cg is drawn for every side chain). The three
strands 71-75.105-1 10, and 120-124 run
antiparallel each other. Notice that residues 121
and 122 have the side chains pointing i n the
same direction, an irregularity called a P-bulge.
t The chain is considered to run from the N to the C terminus.

Fig. 8.25. (Continued).
3. Others. Short tight loops connecting two strands are called p-turns, or
hairpins, or reverse turns. Different kinds of classifications have been
attempted: one of them, based on the hydrogen bond patterns, is that of
Milner-White and ~ o e t . [ ' The
~ ~ ] more classical p-turns are called type I, 11,
and I11 (and type I t , 11', and 111' those with g, and q~ torsion angles
reversed). A schematic pattern of two turns is illustrated in Fig. 8.26.
In real structures, short pieces of polypetide chains not falling in one of
the previous categories are normally found. In general their g, and 1C, torsion
angles correspond to those of one of the structures discussed above, but
nevertheless that portion of the molecule cannot be classified as one of the
previous categories.
A last important feature strongly influencing the polypeptide chain
conformation is the presence of a disulphide bridge: the sulphur atoms of
two cysteines form a covalent bond, connecting two pieces of polypeptide
chain or even two different chains. This results in a strong stabilization of
the three-dimensional structure, at the expense of some local irregularities.
Polypeptide chain description

In describing a polypeptide chain, the only relevant variables are in practice
the two dihedral torsion angles around the N-C, and C,-C bonds, called cp Fig. 8.26. Schematic representation of the
and v : the torsion angle w is very close to 180°, owing to its partial hydrogen bonds pattern for a y-turn (upper) and
a fi-turn (lower). Symbols for atoms are as in
double-bond character. Therefore, only 2n - 1 numbers are enough to Fig. 8.25. (Adapted from Milner-White and
describe the fold of a protein composed by n amino acids. ~oet."~~')
One convenient way of representing the main chain dihedral angles is the
so-called Ramachandran plot,[1041where every amino acid is represented by
a point of coordinates g, and q . In such a plot, regions forbidden for sterical
reasons can be individuated and they represent a large portion of the total
area: the backbone torsion angles of all the amino acids, with the exception
Fig. 8.27. Ramachandran plot for the model of

bovine P-trypsin, orthorhombic form, refined to
a crystallographic Rfactor of 0.18 up to a
resolution of 1.8 A['051(data from PDB). Small
squares represent glycine residues, crosses all
the others. The origin of the diagram is in the
centre. Contour lines correspond roughly to fully
allowed and outer limit regions.
of glycine which has much more conformational freedom owing to the

absence of substituents at the a-carbon, should not fall into those areas.
Ramachandran plots can be used during the refinement to check for the
quality of the structure: amino acids falling into forbidden regions must be
corrected, and values at the border or in unfavourable zones considered
suspiciously. An example of such a plot is given in Fig. 8.27. Since every
type of secondary structure described in the previous subsection is charac-
terized by its own pair of theoretical g, and q values, it corresponds to a
point on the plot. From the diagram it is consequently possible to
individuate immediately the amount of a-helix, P-sheet, and so on, present
in the structure. Any information about the connectivity among residues is,
on the contrary, lost.
A possibly more informative, but less used, method of mapping the
folding of the main chain is the diagonal plot.[106]It is a matrix with the
amino acid numbers along both axes: if the distance between two residues is
less than a predefined amount, the corresponding matrix element is
darkened. In this kind of graphic, the major structural elements are easily
visualized: an a-helix will produce a thickening along the diagonal and two
antiparallel P-strands a narrow line perpendicular to it; domains can be
recognized too, and structural homologies among proteins individuated.
Furthermore, the information of a diagonal plot is in a form that can easily
be stored in a computer. An example of a diagonal plot is reported in Fig.
8.28.
Higher levels of organization: tertiary and quaternary

structure, domains, and subunits
We have previously seen that, despite the apparently very large degree of
freedom of a polypeptide chain, proteins can be considered as made up by a
very limited number of secondary structure elements, that is a-helices,
/3-strands and turns, plus some pieces of chain in conformations more

difficult to classify. The three-dimensional relationships among secondary
structure elements is called tertiary structure or, preferably, super-
secondary structure. A careful examination of the structures solved to the
present has allowed us to identify some macroscopic concepts that simplify
the description of the way in which the secondary elements are held
together. Of course these rules are not necessarily general, since they are
based on quite a limited set of structures.
Fig. 8.28. (a) The bovine pancreatic trypsin

Inhibitor, a small protein (58 amino acids)
refined t o an Rfactor of 0.16 at 1.5 8,
resolution['07' (data from PDB). The backbone
only, that is the atoms C ,, C, N, and 0 of the
main chain, are shown. (b) @-carbonchain trace
of the same structure. It is composed by one
turn of helix, t w o long bent P-strands and a
final a-helix from residues 47 t o 56. All these
elements are disposed in a quite complex way,
difficult t o describe. (c) Diagonal plot for the
same protein. The plot was produced using the
program DGPLOT (Swanson and Bernstein,
distributed with PDB). Amino acid numbers are
reported on both axes, and numbers inside the
diagram correspond to the distance i n A,
rounded to the next integer, between the t w o
cu-carbons. Distances above 9 A are represented
by letters. In the diagram the a-helix close t o the
C terminus of the molecule is represented by the
thickening along the diagonal (A in figure) and
the strands from 29 t o 43 and from 6 t o 25, close
to each other, by a line perpendicular t o the
diagonal (B in figure). Other spatial relationships
are immediately evident from the diagram:
region 43-48 is close i n space t o 20-33 (marked
C in figure).
ARG
PRO
ASP
PHE
CYS
LEU
GLU
PRO
PRO
TYR
THR
GLY
PRO
CY S
LY S
ALA
ARG
ILE
ILE
ARG
TYR
PHE
TYR
ASN
ALA
LY S
ALA
GLY
LEU
CY S
GLN
THR
PHE
VAL
TYR
GLY
GLY
CYS
ARG
ALA
LY S
ARG
ASN
ASN
PHE
LYS
SER
ALA
GLU
ASP
CY S
MET
ARG
THR
CY S
GLY
GLY
ALA
TER
Fig. 8.28. (Continued).

One example is given by packing of a-helices: it has been shown that

helices tend to pack not parallel to each other, but with the axes forming
specific angles: according to the way they are disposed, they can be grouped
in only three classes, each of them characterized by a specific angle.[loS1
Connections between strands in a sheet too follow quite strict rules,
thoroughly described by ~ i c h a r d s o n . [ ' ~
The
~ ] topology of the sheet can be
illustrated by simplified diagrams, like those reported in Fig. 8.29. The
physical meaning of such a diagram is discussed on p. 585.
Every polypeptide chain does not necessarily fold into one single unit: a
portion of it can be structurally independent and is called a domain.
Although the concept is not unequivocal, its existence is widely recognized.
The most simple definition of domain is that given by ~ i c h a r d s o n : [ a~ ~ ~ ]
domain is a contiguous portion of polypeptide chain folded into a compact,
local, independent unit. Domains are sometimes easy to recognize, as in the
case of immunoglobulins, or obvious, as for calmodulin (Fig. 8.30).
Sometimes they are not simple to individuate, or are even questionable.
Whether a domain must be thought of as a potentially independent, stable
unit or only as a compact globular region contiguous in space, is matter of
discussion. The domain concept is in any case justified by the classification (i)
given on p. 585.
q It-
Proteins may eventually be made up by more than one polypeptide chain:
often several of them, equivalent or different, interact to form a protein
molecule. The assembly of such subunits is called quaternary structure and
Fig. 8.29. Topology of the proteins classified as
is in general essential for function: dissociation of a multimeric enzyme may 'Greek key p - b a r r e l ~ " ' ~(see
~ ' p. 585 for more
-
result for example in a change or loss of activity. Nevertheless, subunits can details about the classification). Barrels are
be considered as independent structural eldments: they can often be
a ~~~~~~n~;9~s~~~d
separated and crystallized and their overall conformation is in general not among them. Dashes indicate that the first
too different from that of the subunit in the macromolecular complex from ~ ~ ~ ~ ~ ~ ~ ~ ~ f ~
which they originate. A relative reorientation of subunits has been pyruvate kinase domain 2; (c) trypsin
associated with allosteric properties. A classical example of it is mammalian and 2; (d) immunoglobulin C; ( 4 Cu, Zn
haemoglobin, a tetramer formed by two distinct chains, denoted a and P. superoxide dimutase; (f) prealbumin; (g)
plastocyanin; (h) plasma retinol-binding protein;
Two a$ dimers are related by an exact twofold axis, which in some crystal (i) immunoglobulin V; (I) fattyacid-binding
form corresponds to a crystallographic one: from the oxygenated to ~ ~ ~ ~ ~ ~
deoxygenated state the molecule undergoes a substantial change in the B - I ~ C ~ O ~ I Oand ~ Ubi~in-bindins
I~~ protein.
quaternary (and also tertiary) structure, paired with a variation of affinity
for oxygen.['l11
Fig. 8.30. C, tracing of calmodulin, a small

protein that binds calcium in four different
binding sites["0' (coordinates from PDB). Two
'globular' domains, composed of about 60-70
amino acids each, are completely separated in
space, connected only by a long a-helix.
When a macromolecule is made up by equivalent subunits, a relevant

aspect is represented by its internal symmetry. The simplest case is
represented by a rotation axis: very often, in a dimeric protein the two
subunits are related by a twofold axis. Multimeric proteins may present
threefold and also five- or sixfold axes. A combination of rotation axes can
give rise to higher point symmetries, like 222 or 32. Possibly the most
complex internal symmetry is represented by the coat of spherical viruses,
composed by hundreds of copies of the same protein arranged in a
icosahedral symmetry.[1121It must be remembered that some or all of the
internal symmetry elements may or may not correspond to those of the
crystal (see p. 538).
Groups other than amino acids

The product of biosynthesis is a molecule containing only the 20 natural
amino acids. To fulfil their function, proteins may need the presence of
other groups: therefore, they can undergo post-translational modifications
or interact with other molecules. These modifications may have a strong
influence on their structure.
A post-translational modification is a chemical reaction generally involv-
ing the side chain of some amino acid.? The functional groups bound are
called prosthetic groups and may have the function to assist enzymes in
chemical reactions, or to add to natural amino acids properties they lack,
for example photochemical ones. A quite common modification is phos-
phorylation, largely used in nature to control enzymatic activity.[1131
Glycoproteins are the result of binding of polysaccharides. They may be
bound through the oxygen of hydroxylated residues, like serine or thre-
onine, or through the nitrogen of asparagine. Carbohydrates are quite
hydrophilic and their content can reach a high percentage of the total mass
of the protein molecule: glycoproteins consequently present peculiar pro-
perties and their function is matter of discussion.["41 Very few of them have
been crystallized, and in general the polysaccharide moiety has been found
disordered in the crystal.
Cations bound to a protein can be divided in two categories: those which
simply neutralize charged groups at the surface of the molecule and depend
on the ionization conditions of these groups (and ultimately on pH), and
those strongly coordinated, required for the stability of the conformation.
The latter can influence protein activity by stabilizing one particular
conformational state. The more common bound cation is ca2+, which is
very abundant in cytoplasm. Among proteins that bind it, the most peculiar
is calmodulin (Fig. 8.30), which has four calcium binding sites: the level of
calcium bound to the protein modulates its function. Other bivalent cations
commonly interacting are Zn2+, Cu2+, and ~ e ~ but + ,many other metals
present in nature can be found, in different oxidation states. As an example,
the iron-sulphur binding site in ferredoxin is shown in Fig. 8.31.
A last important case is represented by the small, generally organic,
compounds non-covalently bound to proteins. The macromolecule in this
case may have the function of carrier or of storage site.
t Other possibilities are: a modification of the N or the C terminal part of the molecule, or a
proteolytic cleavage of a peptide bond.
Fig. 8.31. The t w o sulphur-iron clusters in

Peptococcus aerogenes ferredoxin are shown,
along with the polypeptide chain trace of the
(coordinates from PDB). Fe and S
atoms are disposed alternated at the vertices of
a distorted cube: each Fe atom is surrounded by
four S atoms, the fourth being that of a cysteine
residue.
Thermal parameters and disordered structures

Special care must be used in evaluating temperature factors obtained from
the refinement of a protein crystal structure. Experimental B factors
include, besides thermal vibrations, the static disorder, which in such kind
of crystal is particularly relevant. Owing to the high content of solvent,
superficial groups of the macromolecule have the effect of partially ordering
the solvent, but at the same time, as a consequence of this contact, they
become very mobile. It has been shown[115]that it is possible, for very well
refined structures, to distinguish the contribution to B factors of real
thermal vibration from the static disorder. Moreover, during the refinement
procedure thermal parameters are usually restrained, that is their variation
is in some way smoothed down. In any case, the comparison of temperature
factors for the same structure, ribonuclease A, independently refined by two
different groups using different data[l16]shows a quantitative agreement in
the trends, that is regions with an high B factor roughly correspond in both
structures. As a general rule, in well refined structures main chain atoms
present lower thermal motion than side chains, or in any case less disorder.
In looking at crystallographic results, it must be kept in mind that a very
high B factor for some residues could be either due to an intrinsic disorder
of that part of the molecule, or an indication of a misinterpretation of the
electron density. In some cases, small parts of the structure, often at the
beginning or at the end of the polypeptide chain, or some loops protruding
towards the solvent regions, are very mobile and cannot be seen at all in the
map. In exceptional situations, two conformations could be differentiated
for some residues.[231
The disorder can also have some functional role, as is sometimes the case
for allosteric proteins, where two conformational states are present, one of
them characterized by a portion of a disordered chain that becomes ordered
in the other state.[117,1181In any case, as has been pointed out by
~ i c h a r d s o n , [ ' this
~ ~ ] is a case in which the crystallization process introduces
a bias in the results, since less ordered or disordered proteins are likely to
be more difficult or even impossible to crystallize.
Solvent structure
A high portion of the solvent contained into a crystal cell can be considered
not to be relevant for the macromolecule; it is simply there to fill the
channels produced by molecular contacts in the crystal. And indeed, this

unordered solvent cannot be seen in an X-ray diffraction experiment. Water
molecules closely bound to the protein, on the contrary, can be considered
as part of the structure of the macromolecule itself: a protein cannot be
completely dehydrated without a complete crash of the architecture of its
three-dimensional structure. Tightly bound solvent molecules in the crystal
are identified during the process of refinement, and can be distinguished in
three groups:
1. Water molecules making hydrogen bonds with hydrophilic side chains on
the surface of the protein, where often they take part in a tetrahedral or
trigonal network of hydrogen bonds. Ordered waters are substantially on
the first shell of coordination around the protein, or eventually in the
second shell, bound to the water molecules of the first one.
2. Water molecules that serve as a bridge among parts of the main chain or
other structural elements that are too far apart to form hydrogen bonds:
for example, if two strands of a P-sheet diverge slightly, a water
molecule can make an H-bond in the middle, filling the gap. This kind of
solvent molecule is essential in stabilizing the protein structure.
3. Solvent molecules located in the internal cavities of the protein, where
they sometimes do not form very stable interactions but simply fill a
vacuum.
It should be noted that the arrangement of the solvent structure around a
protein determined by X-ray analysis is strongly influenced not only by the
crystal packing, but also by the pH and the solvent used in the crystal-
lization, and it cannot be considered fully representative of the situation in
vivo.
The influence of crystal packing

The question possibly most often asked since the beginning of protein
crystallography can be summarized as follows: how is the structure in the
crystal representative of the 'real' in vivo structure? Proteins are quite
stable but largely flexible molecules: comparison of the same protein
obtained in different crystal forms, and consequently subjected to com-
pletely different packing forces, present in general the same fold, with some
differences, usually small, in the regions of contact among molecules in the
crystal. Figure 8.32 shows the a-carbon chain trace of TRP aporepressor
Fig. 8.32. a-carbon chain trace of TRP-

aporepressori'20' (data from PDB). Thick and
light lines stands for the orthorhombic and the
trigonal crystal forms respectively. The t w o
models superimpose quite well, except for the
first five residues and for the region around
67-80.
from two crystal forms, trigonal and o r t h o r h ~ m b i c . [ ' The

~ ~ ~ fact that
molecules crystallized in different conditions of pH and precipitants keep
the same overall conformation is indirect evidence of the stability of protein
conformation, and of the validity of the structure obtained using the
crystallographic technique. On the other hand, the local variations easily
observed suggest that great care has to be taken in drawing specific
conclusions on functional aspects from details of the structure.
Protein classification
The secondary structural elements of a protein could be combined, at least
in principle, in a nearly infinite number of ways to produce the three-
dimensional structure. An examination of the structures solved till now has
on the contrary shown that these elements are put together in quite a
limited number of ways, giving rise to a small set of possible patterns. The
previous statement must nevertheless be regarded cautiously: it could be
partially ascribed to the limited size of the data base available (as explained
in the introduction, in the Brookhaven Protein Data Bank only 353 sets of
protein coordinates are reported, most of them relative to parent mole-
cules); in addition, a bias could be introduced by the fact that only some
classes of molecules have so far been crystallized. In any case, regularities
and similarities in tertiary structures can be observed: a distribution of
proteins or domains into classes and subclasses has been attempted.
According to ~ i c h a r d s o n , [ 'protein
~ ~ ~ molecules solved up to the present
belong to one of the following classes:
up-and-down helix bundle

(1) antiparallel a structures Greek key helix bundle
miscellaneous
up-and-down P barrels
Greek key /3 barrels
(2) antiparallel /3 structures
open-face P sandwiches
miscellaneous
singly wound parallel P-barrels
(3) parallel alp structures doubly wound parallel P-sheets
miscellaneous
(4) small irregular proteins.
Set (4) is a very small one, and nearly all the domains of solved structures
belong to one of the first three sets. A first classification in three main
groups is quite obvious: domains may be composed mainly by helices, by
P-strands differently arranged, or by a combination of both elements. What
is more puzzling is that, inside each main group, strict rules are observed:
for example, all the a-helices belonging to class (1) are arranged in an
antiparallel way, that is nearby helices are pointing in opposite directions.
Two structures belonging to the first two subclasses, the up-and-down and
Greek key antiparallel a, are shown in Figs 8.33(a) and (b), respectively.
Fig. 8.33. Sketches of protein models produced

following the directives of Richardson, usin
program RIBBON.['^" (a) myohernerythrin' "
8
and (b) m y ~ ~ l o b i n ,representatives
[~~' of the
antiparallel a-structures; (c) p r e a ~ b u m i n , [ ' ~a~ '
Greek key P-barrel, a subclass of antiparallel P
structures; (d) triose phosphate isomerase
TIM),"^^' one of the more elegant examples of
a parallel CYIPprotein. All coordinates were
taken from PDB.
The same holds for class (2), the all-p domains: the strands in this group are
antiparallel to each other (only if more than seven strands are present, can a
parallel chain be observed). Figure 8.33(c) shows a picture of prealbumin,
an example of a protein of the subclass called greek key $ barrel, and Fig.
8.29 illustrates all the topologies found in that class, which does not mean
all the possible, but simply those found in the structures solved till now.
In proteins containing both structural elements, classified a l p , a-helices

are arranged parallel to each other and so are P-strands. There is not a
rational explanation for that, since they could be antiparallel as well. An
example of this last class, the enzyme triose phosphate isomerase, classified
in the subclass called singly wound parallel p-barrel, is shown in Fig.
8.33(d).
It is difficult to judge the real biological meaning of the previous
classification, since oversimplification could have a part in it. Nevertheless,
it remains surprising that molecules which are so complex give rise to such a
limited number of three-dimensional patterns.
Appendices
Appendix 8.A Some formulae for isomorphous
replacement and anomalous dispersion
In this appendix, formulae (8.13) and (8.23), relative to isomorphous
replacement and anomalous scattering, will be derived in a general way.
Equation (8.13) can also be obtained from Fig. 8.1, using simple trigono-
metric considerations. It will be derived to illustrate the formalism used in
the rest of the appendix. Let us call:
FPH= FPHexp G ~ P H )
FH= FHexp (iq,).
From (8.5):
= (FP+ FH)(Fc + Fh).
FPHF:H (8.A.1)
By substituting in (8.A.1):
FPHexP ( ~ ~ P H ) FexpP H(-~vPH)
= {FPexp (MP) + FHexp ( ~ ~ H ) ) {exp
F P (-iqP) + FHexp (-iqH)), (8.A.2)
F$, = F$ + F&+ FHFpexp [i(qp - qH)]+ FpFHexp [-i(qp - q,)], (8.A.3)
F;, = F$ + F&+ 2FPFHCOS (qp - qH). (8.A.4)
Equation (8.A.4) is equivalent to (8.11). An analogous expression can be
derived for qp,. Since, from (8.5):
FP= FPH
- FH
we can derive as before:
F$ = F$H+ F&- 2FpHFHcos (qpH- qH), (8.A.6)

hence:
= qH+
Q l p ~ [(F$H + F&- F$)/2FpHFH].
COS-I (8.A.7)
Equations (8.A.4) and (8.A.6) are, in fact, the same equation and both give
two possible solutions for the phase of the protein (or of the derivative).
For anomalous dispersion measurements, let us define:
Subtracting the two:

F;;-F-~PH - 4 F P ~ F cos
& (qPH- qH- n/2) = 4FpHFksin (qpH- qH).
(8.A.10)
We can also write:
and, since the anomalous contribution to the structure factor of a protein is

small if compared with the non-anomalous part:
F;H + FFH -'. 2FPH.
Substitution of (8.A. 12) in (8.A. 11) gives:
~ p ' ; - FG -'. 2FPH(F;~- FFH), (8.A.13)
and comparison of (8.A. 13) and (8.A. 10):
FH
: - F& = 2F; sin (qpH- 9,) (8.A.14)
Finally:
qPH= plH + sinP1[(F& - F,)/2F;]. (8.A.15)
The combination of (8.A.15) and (8.A.7) allows a unique determination
of YPH.
Appendix 8.B Translation functions

Several types of translation functions have been defined and most of them
are quite similar and sometimes virtually equivalent (for a general review,
see ~eurskens['~]). In the following, the more common among those used in
protein crystallography are described.
1. The Q function ( ~ o l l i n [ ' ~ ~The
] ) . Q function was devised by Tollin to
determine the position I;. of a known group of atoms in a crystallographic
cell, with respect to an arbitrarily chosen origin. Let us consider only two
molecules, related by the symmetry operator A. If one of them (that we will
call reference molecule) is correctly oriented in the cell of the unknown
crystal, the problem of positioning it with resp,ect to translation reduces to
finding the correct origin with respect to the symmetry operator A. In the
case of a superposition of m maps translated by vectors 5, the sum function
defined in (5 .B.5) ( ~ u e r ~ e r ; [ "see
~ I Appendix 5 .B) becomes:
S(u) = 2 Z(h) Ci cos [2nh(u - q)].
h
(8.B.1)
Equation (8.B.1) will present a maximum in position 5 = A(X + t) - t,

that is when the translation t will be the vector defining the position of the
absolute origin with respect to the symmetry operator under consideration.
Substituting in (8.B. 1):
Q(t) = C I(h) C cos 2nh[Xj + t - A(X,, + t)]. (8.B.2)
h i,i'
From the maximum of (8.B.2) it is straightforward to obtain X,.
The Q function will show several false peaks, whenever two atoms have
coordinates XI and X2 such that the relationships:
is satisfied, and these peaks will have the same magnitude as those defining
the correct origin. However, the position of false peaks can be predicted, or
they can be removed by modifying the Z(h) value.
2. The T function (Crowther and low['^]). The function for two
molecules related by symmetry operator A is given by (8.60). It has been
that the T function is virtually equivalent to the Q function,
except that the origin is looked at in a different way.
Modified T functions have also been defined. If more than one symmetry
operation is considered, (A,, . . . , A,), the T function becomes:
Tl(t) = exp (-2nih . t) (8.B.4)

C (bbs(h) - C F,(~A~))F,(~)F:(~A,)
h i
h(t) = 2 z O b s ( h ) (C
h i
~ F,(~A,)F:(~A,)) exp (-2nih . t)
I
(8.B.5)
3. The TH function (Harada et U L . [ ~ ~TH ~ ] )represents

. an improvement of
the T function, since it takes into account the interpenetration of molecules
in the crystal cell. TH is defined as the ratio of two functions:
Vector t here does not have the same meaning as in Fig. 8.17: function TH
gives in fact the absolute value of the translation that has to be applied to
the model in order to position it correctly in the crystal cell. Equation
(8.B.7) is similar to the Crowther T function, but O(t) is a new term, taking
into account the interpenetration among molecules. TH will present a
maximum when cross-vectors of Patterson maps superimpose and molecules
do not overlap each other.
Appendix 8.C Macromolecular least-squares refinement

and the conjugate-gradient algorithm
The normal matrix used in restrained least-squares deserves a special
comment. If the molecule under refinement is composed of N atoms, and
we are refining atomic coordinates and individual isotropic thermal para-
meters, the normal matrix will be a square matrix of dimension 4N. Even
employing rigid-body constraints whenever possible, the use of classical
full-matrix methods may be still prohibitive. To speed up calculations,
different algorithms have been devised. The fast calculation of gradients,[1291
and the use of the fast Fourier transform (FFT, see p. 131) in calculating
structure factors.[1301
For matrix inversion, any algorithm able to handle system of equations
with thousands of variables can be used, like the ~ a u s s - ~ i e d e l . Since
[ ~ ~ ] in
most of the restrained least-squares programs a 'sparse matrix' is used, the
algorithm of conjugate gradients appears well suited for solving a system of
linear equations of this kind. The only drawback of it is that it does not
directly provide the diagonal elements of the inverted matrix, from which
the standard deviations of the parameters can be obtained. However,
approximate values for them can be calculated. A brief description of the
conjugate gradients algorithm is given here, taken from ~ o l l e t [ ' ~and ~]
~ o n n e r.[721
t
Let us have a system of n simultaneous equations in n unknowns:
In the case of crystallographic least squares, the vector of unknowns X

represents the shifts to be applied to the parameters. Classical inversion of
matrix A for a very large number of parameters, as is the case of
macromolecules, can be very inefficient or even impossible. Gradient
methods are based on the iterative process that consists in obtaining,
starting from an approximate solution Xk, a new one from:
where rk is the vector of residuals, defined by:
and akis an appropriate constant.

The conjugate gradients method, discovered by Hasteness and stiefel,[1311
is an iterative process that gives the solution for an n-dimensional problem
in approximately n steps (but generally much fewer than that). The method
starts with an arbitrary value for vector X: let us call it X, (usually zero).
The vector of residuals, ro, is given by:
r o = p o =b -AXo.
Let us also define:
~t;.= (rT~i)~/(pTApi)
Pi = -(rT+~Api)~/(pTApi).
Using (8.C.4) and (8.C.5), a new value of X , called X I , will be given by:
X 1 = Xo + mopo. (8.C.7)
Now a new residual and a new value for pi can be calculated:
From p l we can now evaluate a new vector X and a new residual vector:
By analogy with (8.C.9) a new value of p can be calculated, and so on. The
method can proceed until convergence, that is when the vector of residuals
r;: is less than a predefined amount. The advantages of the method are that
every estimate of X improves the previous one and that the elements of the
matrix A are not moved from their positions: the non-zero elements of A
can be stored contiguously, taking advantage of a scarcely populated
matrix.
Appendix 8.D Conventions and symbols for amino acids

and peptides
Fig. 8.D.1. Schematic representation of a
peptide unit, with symbols according the
recomendations of the IUPAC-IUB Commission
on Biochemical Nomenclature, Journal o f
Molecular Biology, 52, 1-17 (1970). Each
repetitive unit, delimited in the figure by broken
lines, has general formula -NH-CHR-CO-,
where R is one of the 20 side chains illustrated
in Fig. 8.D.3. Amino acids are connected by a
peptide bond CO-NH. In all the protein
structures known, the peptide torsion angle, w ,
assumes a value of about 180°, that is the
carbonyl oxygen and the H of nitrogen assume a
trans planar conformation. The only exception is
represented by proline, that in few cases has
been found to be cis. Amino acids i n a
polypeptide chain are numbered from the N t o
the carboxyl terminal.
Fig. 8.D.2. A mnemonic rule for remembering

the L- configuration of an amino acid: looking
down from the H atom toward the C ,, the
substituents (CO, R, and N ) must form the word
CORN if read in clockwise sense.
1
592. Giuseppe Zanotti
Fig. 8.D.3. The side chains of the 20 natural

amino acids.
Leu. L
-
Ala, A -CB
CG1
Val, V -CB' - C B - CG - SD - CE Met, M
\CGZ
,CD1-CEl,
Phe, P -CB-CG cz - C B - OG Ser, S
OG1
Thr, T -cs'
\ CG2
I
CD1 - CE1 , OD1
Tyr,Y -CB-CG: c z - OH -CB -CG Asn, N
CD2-CE2 ' ' ND2
OEl ND1 -CE1

G1n.Q -CB-CG-CD' -CB-CG' I His, H
ASP,D - C B - CG :
OD1
OD2
-CB-CG-CD
OEl
' OE2
G ~ uE,
Fig. 8.0.4. Symbols used i n Protein Data Bank
for the atoms of the 20 side chains of Fig. 8.D.3.
NH1 The same nomenclature is also used, with some
small modifications, in most of the refinement
Arg, R CB-CG-CD-NE-CZ: -CB - C G - CD - CE - N Z L~s, programs. The three-letter and the one-letter
NH2 code for the amino acids are used.
References and further reading

Further reading
1. Blundell, T. L. and Johnson, L. N. (1976). Protein crystallography. Academic,
New York.
2. McPherson, A. (1982). Preparation and analysis of protein crystals. Wiley, New
York.
3. Wyckoff, H. W., Hirs, C. H . W., and Timasheff, S. N. (ed.) (1985).
Diffraction methods for biological macromolecules. In Methods in
enzymology, Vols 114-15. Academic, Orlando.
4. Rossmann, M. G. (ed.) (1972). The molecular replacement method: a collection
of papers on the use of non-crystallographic symmetry. Gordon and Breach,
New York.
5. Cantor, C. and Schimmel, P. R. (1980). Biophysical chemistry. Freeman, San
Francisco.
References
6. Kendrew, J. C., Dickerson, R. E., Strandberg, B. E., Hart, R. G., Phillips, D.
C., and Shore, V. C. (1960). Nature, 185, 422-7; Cullis, A. F., Muirhead, H.,
Perutz, M., Rossmann, M. G., and North, A. C. T. (1961). Proceedings of the
Royal Society of London, A265, 161-87; Perutz, M. (1985). In Methods in
Enzymology, Vol. 114 (ed. H. W. Wyckoff, C. H. W. Hirs, and S. Timasheff),
pp. 3-46. Academic, Orlando.
7. Bernstein, F. C., Koetzle, T. F., Williams, G. J. B., Meyer, E. F. Jr., Brice,
M. D., Rodgers, J. R., Kennard, O., Shimanouchi, T., and Tasumi, M.
(1977). Journal of Molecular Biology, 112, 535-42.
8. Pusey, M. L., Snyder, R. S., and Naumann, R. (1986). Journal of Biological
Chemistry, 261, 6524-9.
9. Zeppezauer, M. (1971). Methods in Enzymology, 22, 253-69.
10. McPherson, A. (1976). Journal of Biological Chemistry, 251, 6300-3.
11. McPherson, A. (1976). Methods in Biochemical Analysis, 23, 249-345.
12. Garavito, R. M., Jenkins, J., Jansonius, J. N., Karlsson, R., and Rosenbusch,
-
J. P. (1983). Journal of Molecular Biology, 164, 313-27.
13. Michel, H. (1983). Trends in Biological Science, 56-9.
15. Matthews, B. W., and Bernhard, S. A. (1973). Annual Reviews in Biophysics
and Bioengineering, 2, 257-317.
17. Green, D. W., Ingram, V. M., and Perutz, M. F. (1954). Proceedings of the
Royal Society of London, A225, 287-307; Rossmann, M. G. (1960). Acta
18. Ramachandran, G. N. and Srinivasan, R. (1970). Fourier methods in crystal-
lography (ed. M. J. Buerger) pp. 96-235. Whiley, New York.
19. Sim, G. A. (1959). Acta Crystallographica, 12, 813-15.
20. Sim, G. A. (1960). Acta Crystallographica, 13, 511-12.
21. Blow, D. M., and Rossmann, M. G. (1961). Acta Crystallographica, 14,
1195-202.
22. Harker, D. (1956). Acta Crystallographica, 9, 1-9.
23. Hendrickson, W. A. and Teeter, M. M. (1981). Nature, 290, 107-13.
24. Rossmann, M. G. (1961). Acta Crystallographica, 14, 383-8.
25. North, A. C. T. (1965). Acta Crystallographica, 18, 212-16.
26. Matthews, B. W. (1966). Acta Crystallographica, 20, 82-6.
27. Parthasarathy, S. (1970). Acta Crystallographica, A27, 45-7.
28. Bijvoet, M. (1954). Nature, 173, 888-91.
29. Terwilliger, T. C. and Eisenberg, D. (1987). Acta Crystallographica, A43,

6-13.
30. Blow, D. M. and Crick, F. H. C. (1959). Acta Crystallographica, 12, 794-802.
31. TenEyck, L. F. and Arnone, A. (1976). Journal of Molecular Biology, 100,
3-11.
32. Hendrickson, W. A. and Latman, E. E. (1970). Acta Crystallographica, B26,
136-43.
33. Hendrickson, W. A. (1971). Acta Crystallographica, B27, 1472-5.
34. Dickerson, R. E . , Kendrew, J. C., and Strandberg, B. E. (1961). Acta
35. Sygusch, J. (1977). Acta Crystallographica, A33, 512-28.
36. Blow, D. M. and Matthews, B. W. (1973). Acta Crystallographica, A29,
56-62.
37. Petsko, G. A. (1976). Acta Crystallographica, A32, 473-6.
38. Henderson, R. and Moffat, J. K. (1971). Acta Crystallographica, B27,
1414-20.
39. Davies, A. R. and Rollett, J. S. (1976). Acta Crystallographica, A32, 17-23.
40. Schevitz, R. W., Podjarny, A. D., Zwick, M., Hughes, J. J., and Sigler, P. B.
(1981). Acta Crystallographica, A37, 669-77.
41. Bath, T. N. and Blow, D. M. (1982). Acta Crystallographica, A38, 21-9.
42. Cannillo, E . , Oberti, R., and Ungaretti, L. (1983). Acta Crystallographica,
A39, 68-74.
43. Bryan, R. K. and Banner, D. W. (1987). Acta Crystallographica, A43,556-64.
44. Subbiah, S. and Harrison, S. C. (1989). Acta Crystallographica, A45, 337-42.
45. Rossmann, M. G. and Blow, D . M. (1962). Acta Crystallographica, 15, 24-31.
46. Tollin, P. and Rossmann, M. G. (1966). Acta Crystallographica, 21, 872-6.
47. Crowther, R. A. (1972). In The molecular replacement method: a collection of
papers on the use of non-crystallographic symmetry (ed. M. G. Rossmann), pp.
173-5. Gordon and Breach, New York.
48. Tollin, P., Main, P., and Rossmann, M. G. (1966). Acta Crystallographica, 20,
404-7.
4 9 . R a 0 , S. N., Jyh-Hwang, J., and Hartsuck, J. A. (1980). Acta
50. Latman, E. E. (1972). Acta Crystallographica, B28, 1065.
51. Crowther, R. A. and Blow, D . M. (1967). Acta Crystallographica, 23, 544-8.
52. Latman, E . E. (1985). In Methods in enzymology, Vol. 115, (ed. H. W.
Wyckoff, C. H . W. Hirs, and S. N. Timasheff), pp. 55-7. Academic, Orlando.
53. Beurskens, P. T., Gould, R. O., Bruins Slot, H . J., and Bosman, W. P.
(1987). Zeitschrift fur Kristallographie, 179, 127-59.
54. Wilson, I. A , , Skehel, J. J., and Wiley, D. C. (1981). Nature, 289, 366-73.
55. Evans, P. R., Farrants, G. W., and Lawrence, M. C. (1986). Journal of
Molecular Biology, 191, 713-20.
56. Rossmann, M. G . , Blow, D. M., Harding, M. M., and Coller, E. (1964). Acta
57. Wilson, K. (1978). Acta Crystallographica, B34, 1599-608.
58. Mukherjee, A. K., Helliwell, J. R., and Main, P. (1989). Acta
Crystallographica , A45, 715-18.
59. Giacovazzo, C. (1990). In International tables for x-ray crystallography, vol. B.
Reidel, Dordrecht.
60.Giacovazz0, C., Cascarano, G., and Chao-De, Z. (1988). Acta
61. Jaynes, E. T. (1957). Physical Review, 106, 620-30; Collins, D. M. (1982).
Nature, 298, 49-51.
62. Bricogne, G. (1984). Acta Crystallographica, A40, 410-55; (1988). Acta
63. Harrison, R. W. (1989). Acta Crystallographica, A45, 4-10.

64. Podjarny, A. D., Moras, D., Navaza, J., and Alzari, P. M. (1988). Acta
65. Prince, E., Sjolin, L., and Alenljung, R. (1988). Acta Crystallographica, A44,
216-22.
66. Richards, F. M. (1968). Journal of Molecular Biology, 37, 225-30.
67. Jones, T. A. (1978). Journal of Applied Crystallography, 11, 268.
68. Pflugrath, J. W., Saper, M. A., and Quiocho, F. A. (1984). In Methods and
applications in crystallographic computing (ed. S. Hall and J. Ashida), p. 404.
Oxford University Press.
69. Jones, A. T. and Thirup, S. (1986). E M B O Journal, 3, 819-22.
70. Watenpaugh, K. D., Sieker, L. C., Herriot, J. R., and Jensen, L. H. (1973).
Acta Crystallographica, B29, 943-956.
71. Diamond, R. (1971). Acta Crystallographica, A27, 436-52.
72. Konnnert, J. H. (1976). Acta Crystallographica, A32, 614-17.
73. Hendrickson, W. A. and Konnert, J. H. (1980). In Computing in crystal-
lography (ed. R. Diamond, R. Ramaseshan, and K. Venkatesan) pp.
13.01-13.23. The Indian Academy of Sciences, Bangalore.
74. Waser, J. (1963). Acta Crystallographica, 16, 1091-4.
75. Schomaker, V., Waser, J., Marsh, R. E., and Bergman, G. (1959). Acta
76. Rees, D. C. and Lewis, M. (1983). Acta Crystallographica, A39, 94-7.
77. Haneef, I., Moss, D. S., Stanford, M. J., and Borkakoti, N. (1985), Acta
78. Lunin, V. Yu. and Urzhumtsev, A. G. (1985). Acta Crystallographica, A41,
327-33.
79. Jack, A. and Levitt, M. (1978). Acta Crystallographica, A 3 4 931-5.
80. Levitt, M. (1974). Journal of Molecular Biology, 82, 393-420.
81. Sussman, J. L., Holbrook, S. R . , Church, G. M., and Kim, S. H. (1977). Acta
82. Sussman, J. L. and Podjarny, A. D. (1983). Acta Crystallographica, B39, 495.
83. Shoham, M., Yonath, A., Sussman, J. L., Moult, J., Traub, W., and Kalb, A.
J. (1979). Journal of Molecular Biology, 131, 137-55.
84. Tronrud, D. E . , Ten Eyck, L. F., and Matthews, B. W. (1987). Acta
85. Karplus, M. and McCammon, J. A. (1983). Annual Reviews in Biochemistry,
53, 263-300.
86. Briinger, A. T., Kuriyan, J., and Karplus, M. (1987). Science, 235, 458-60.
87. Fujinaga, M., Gros, P., and van Gunsteren, W. F. (1989). Journal of Applied
88. Brunger, A. T. (1988). In Crystallographic computing (ed. N. W. Isaacs and
M. R. Taylor), pp. 127-40. Oxford University Press.
89. Brunger, A. T. (1988). Journal of Molecular Biology, 203, 803-16.
90. Kuriyan, J., Briinger, A. T., Karplus, M., and Hendrickson, W. A. (1989).
Acta Crystallographica, A45, 396-409.
91. Rice, D. W. (1981). Acta Crystallographica, A37, 491-500.
92. Stuart, D. and Artymiuk, P. (1984). Acta Crystallographica, A40, 713-16.
93. Anfinsen, C. B., Haber, E., Sela, M., and White, F. H. Jr (1961). Proceedings
of the National Academy of Sciences of the U S A , 47, 1309.
94. Schulz, G. E. and Schirmer, R. H. (1979). Principles of protein structure.
Springer, Berlin.
95. Richards, F. M. (1974). Journal of Molecular Biology, 82, 1-14.
96. Blundell, T., Barlow, D., Borkakoti, N., and Thornton, J. (1983). Nature,
306, 281-3.
97. Piela, L., Nemethy, G. E., and Scheraga, H. A. (1987). Biopolymers, 26,
1273-86.
98. Hol, W. G. J., van Duijnen, P. T., and Berendsen, H. J. C. (1978). Nature,
273, 443-6.
99. Takano, T. (1977). Journal of Molecular Biology, 110, 569-84.
100. Borkakoti, N., Moss, D. S., and Palmer, R. A. (1982). Acta Crystallographica,
B38, 2210-17.
101. Richardson, J. S. (1977). Nature, 268, 495-500.
102. Chothia, C. (1973). Journal of Molecular Biology, 75, 295-302.
103. Milner-White, E. J. and Poet, R. (1987). Trends in Biological Science, 189-92.
104. Ramachandran, G. N., Ramakrishnan, C., and Sasisekharan, V. (1963).
Journal of Molecular Biology 7, 95-9.
105. Bode, W. and Schwager, P. (1975). Journal of Molecular Biology, 98,693-717.
106. Phillips, D. C. (1970). British biochemistry past and present (ed. T. W.
Goodwin), pp. 11-28. Academic, London.
107. Wlodawer, A., Deisenhofer, J., and Huber, R. (1987). Journal of Molecular
Biology, 193, 145-56.
108. Chotia, C., Levitt, M., and Richardson, D. C. (1971). Proceedings of the
National Academy of Sciences of the USA, 74, 4130-4.
109. Richardson, J. S. (1981). Advances in Protein Chemistry, 34, 167-339.
110. Babu, Y. S., Sack, J. S., Greenhough, T. J., Bugg, C. E., Means, A. R., and
Cook, W. J. (1985). Nature, 315, 37-40.
111. Dickerson, R. E. and Geiss, I. (1983). Hemoglobin: structure, function,
evolution and pathology. Benjamin Cummings, Menlo Park.
112. Jurnak, F. A. and McPherson, A. (ed.) (1984). Biological macromolecules and
assemblies. Wiley, New York.
113. Krebs, E. G. and Beavo, J. A. (1979). Annual Reviews in Biochemistry, 48,
923-59.
114. Paulson, J. C. (1989). Trends in Biological Science, 14, 272-6.
115. Petsko, G. A. and Ringe, D. (1984). Annual Reviews in Biophysics and
Bioengineering, 13, 331-71.
116. Wlodawer, A., Borkakoti, N., Moss, D. S., and Howlin, B. (1986). Acta
117. Perutz, M. F. (1970). Nature, 228, 726-39.
118. Barford, D. J. and Johnson, L. N. (1989). Nature, 340, 609-16.
119. Adman, E. T., Sieker, L. C., and Jensen, L. H. (1976). Journal of Biological
Chemistry, 251, 3801-6.
120. Zhang, R.-g., Joachimiak, A., Lawson, C. L., Schevitz, R. W., Otwinowski,
Z., and Sigler, P. B. (1987). Nature, 327, 591-7.
121. Priestle, J. P. (1988). Journal of Applied Crystallography, 21, 572-6.
122. Sheriff, S., Hendrickson, W. A., and Smith, J. L. (1987). Journal of Molecular
Biology, 197, 273-96.
123. Blacke, C. C. F., Geisow, M. J., Oatley, S. J., Rerat, B., and Rerat, C.
(1978). Journal of Molecular Biology, 121, 339-56.
124. Banner, D. W., Bloomer, A. C., Petsko, G. A., Phillips, D. C., and Wilson, I.
A. (1976). Biochemistry and Biophysics Research Communications, 72, 146.
125. Tollin, P. (1966). Acta Crystallographica, 21, 613-14.
126. Buerger, M. J. (1967). Vector space. Wiley, New York.
127. Tollin, P. (1969). Acta Crystallographica, A25, 376-7.
128. Harada, Y., Lifchitz, A , , Berthou, J., and Jolles, P. (1981). Acta
129. Agarwal, R. C. (1978). Acta Crystallographica, AM, 791-809.
130. Ten Eyck, L. F. (1973). Acta Crystallographica, A29, 183-91.
131. Hestenes, M. R. and Stiefel, E. (1952). Journal of Research of the National
Bureau of Standards, 49, 409-36.
132. Rollet, J. S. (ed.) (1965). Computing methods in crystallography. Pergamon,
Paris.
Phvsical ~ r o ~ e r t i of
es
crystals
MICHELE CATTl
Introduction
The discipline named crystal physics has been associated traditionally with
two topics:
1. The study of phenomenological and macroscopic properties of crystals,
concerning mainly the effects of crystal anisotropy and symmetry on the
physical properties of matter.
2. The microscopic investigation of crystal defects, i.e. deviations from the
ideal periodicity of the crystal structure, and of their influence on the
macroscopic physical properties.
Both subjects are closely related to crystallographic results: in particular,
the first one relies upon consideration of symmetry, while the second one is
founded on the structural analysis of crystals.
Crystal physics is not only a discipline of great fundamental interest, but it
also has important technological applications. Crystals are used in industry
because of their useful physical (optical, electrical, magnetic, etc.) pro-
perties, which are studied by crystal physics. It suffices to mention
piezoelectric transducers, magnetic oxides for tapes for recorders and
computers, crystals for non-linear optics of laser technology, and many
other examples. Analogously, defects have large effects on the crystal
properties, such as mechanical strength, electrical and thermal conductivity,
etc., so as to play an outstanding role in several branches of technology. It is
also worthwhile to point out the close relations between crystal physics and
solid state physics. The latter discipline, in a restricted meaning, emphasizes
such topics as the quantum study of electronic and vibrational energy levels,
the interaction of radiation with matter, and related items, both in ideally
periodic and in defective solids. All these subjects try to explain on a
microscopic basis the phenomenological properties which are dealt with by
crystal physics; on the other hand, their treatment relies heavily upon the
geometrical features of solids studied by crystallography. Thus crystal
physics can be considered to be a bridging discipline between crystal-
lography and solid state physics, partially overlapping in some areas with
both of them.
600 1 Michele Catti
Crystal anisotropy and tensors

The behaviour of crystals is characterized typically by anisotropy: a physical
agent associated with a given space orientation usually causes effects with
different orientations with respect to the crystal lattice. A central role is thus
played by the geometrical aspect of the problem, particularly in relation to
the choice and transformations of reference frames, and taking into account
the need to comply with symmetry constraints of the crystal.
As shown in Chapter 2, most crystallographic issues are treated in a
lattice reference frame, where the basis vectors are three non-coplanar
translations associated with the periodic symmetry of the crystal structure.
However, orthonormal (Cartesian) reference frames were also introduced
(p. 68), and the transformation properties between lattice A and Cartesian
E bases were analysed, according to the relation E = MA involving the
orthonormalization matrix M. Both types of reference frames can be used in
the study of physical properties of crystals; yet treatments relying upon
lattice bases need a particular formalism (covariant-contravariant subscripts
and superscripts) which is often awkward to use. So orthonormal frames are
generally preferred, and the relations of transformations turn out to be
much simpler. This method will be followed in the present book.
By working with Cartesian frames, a conventional orientation with
respect to the lattice frame must be defined. In the general triclinic case, the
most widely accepted convention involves: e3= clc, e2= b * / b q ,el = e2 x e3.
This implies that, for the monoclinic system, e2= b l b ; for the trigonal and
hexagonal cases, el = ala. In the latter systems, the crystallographic basis
vectors a and b are chosen parallel to the twofold axes (when they are
present) which are normal to the three- or sixfold unique direction. For
orthorhombic, tetragonal, and cubic systems, the Cartesian basis vectors are
parallel to the corresponding lattice basis vectors. The orthonormalization
matrices can be found by the methods of p. 68. It should be stressed that
other types of conventions are possible (for instance: e3= c/c, el = a*la*,
e2= e3 x e l , leading to e2= blb in the trigonal and hexagonal systems).
Since numerical values of the physical properties are strictly dependent on
the choice of reference frame, it is necessary to know whether a different
convention from the standard one is associated with tabulated physical data.
Tensorial quantities
Let us consider a functional relationship between two vectorial physical
quantities, X and Y. If they have a common direction, the functional
relation between their moduli is dealt with by writing a Taylor expansion
about X = 0:
In a crystal, where anisotropy is present, the Y vector generally has a

different direction from X, so that each of the three Cartesian components
Y , of Y has to be considered as a function of all three Cartesian components
Physical properties of crystals 1 601
In the Taylor expansion (9.1) a single coefficient (Yo, y', y", etc.) is required
for each term. On the other hand, according to (9.2) the constant part of
the Y(X) dependence is expressed by 3 quantities Yo,,, the linear part by 9
coefficients y,,, the quadratic part by 27 coefficients y,,,, and so on. In
general, for a term of nth order in the Taylor expansion a number of 3"+'
coefficients is necessary to define the Y(X) dependence.
We now want to examine how the different coefficients appearing in the
expansion (9.2) are transformed, when the orthonormal reference basis E is
changed into another one E'. Let T be the transformation matrix (cf. p. 65)
relating the two Cartesian bases: E' = TE. The metric matrices of both bases
are equal to the identity matrix I, so that G' = G = I; by substituting into
G' = TGT, equivalent to (2.21), the result fi = I is obtained. This condition
is equivalent to
and characterizes T as an orthogonal matrix, representing an isometric

rotation in a Cartesian basis (cf. p. 36). Taking into account (2.20) and
(9.3), the X, Y, and Y , vector representations whose components appear in
(9.2) are transformed as X ' = TX, Y' = TY and Y , = TY, when the E basis is
changed into E'. The explicit relation for the Yoxicomponents is:
As for the nine coefficients yih expressing the linear part of the Y(X)
dependence, they can be considered to be components of a 3 x 3 square
matrix y, so that the first-order term in (9.2) is written in matrix form as:
Y = yx. (9.5)
By substituting Y = T-'Y' and X = T-'x' into ( 9 . 9 , and taking into account

(9.3), one obtains: Y' = T ~ T X ' . Thus the transformation law for the y
matrix is found to be: y' =TyT. The explicit expression involving com-
ponents is:
An entity represented by nine components y,, with respect to a given

Cartesian basis, which obey the (9.6) law of transformation, is defined to be
a tensor of second rank. Analogously, a vector whose three components
follow the transformation rule (9.4) is a tensor of first rank, while a scalar is
a zero-rank tensor. It should be emphasized that a tensor as such, being a
generalization of the vector concept, is independent of the reference basis;
its components, instead, are transformed when the basis changes.
The transformation properties of coefficients y,,, in the expression (9.2)
602 1 Michele Catti
can be found by similar methods, and turn out to be:
An entity represented by 27 quantities yillkas components with respect to a

given Cartesian basis, which obey the transformation law (9.7), is a
third-rank tensor. In a general way, a tensor of rank n is defined as a set of
3" coefficients with n subscripts, associated with a given Cartesian basis,
which transform according to the formula:
where T is the matrix relating the new basis to the old one. It should be
noticed that the general rule (9.8) is equivalent to rule (9.4) multiplied by n
times: thus we can also say that a tensor of rank n transforms in the same
way as a product of n coordinates (or vector components). It is important to
point out that any set of 3" coefficients with n subscripts does not necessarily
obey the (9.8) transformation rule, and so need not represent a tensor of
rank n. Take for instance the 32 components of the matrix T, or those of the
orthonormalization matrix M relating the Cartesian basis E to the lattice
basis A: in no way can an expression of type (9.6) be applied to them, so
that they are not components of second-rank tensors. Tensors of rank
higher than two could be represented by matrices with more than two
dimensions, but this is usually avoided for simplicity and the matrix
formalism is limited to tensors of first and second rank. Further, in tensor
calculus the Einstein convention is commonly adopted, according to which
summation symbols are omitted and understood; however, for the sake of
clarity this convention is not followed in this text. Because of the particular
importance of second-rank tensors in representing the physical properties of
crystals, their features are analysed in detail in Appendix 9.A.
Tensors of rank n have been introduced as coefficients of terms of order
n - 1 in the Taylor expansion (9.2) expressing a functional dependence
between two vectors. However, tensors can represent a physical property
relating not only vectors, but also other tensors. Let us consider a linear
dependence only, for the sake of simplicity. Then the coefficients expressing
such a dependence between vector components and second-rank tensor
components are clearly themselves components of a third-rank tensor:
Analogously, a linear dependence between two second-rank tensors is

represented by the components of a fourth-rank tensor:
The following general rule can be formulated: the coefficients of a linear

dependence of the components of an nth-rank tensor on the products of the
components of n,, . . . , nm-rank tensors are themselves the components of a
+ +
tensor of rank n nl . . . + n,.
From the physical point of view, tensors can represent either an intrinsic
property of the crystalline medium, or an external field applied to the

crystal with an arbitrary orientation. In the first case they are called matter
tensors, in the second one field tensors. Several examples of both kinds of
tensors will be given in the following sections.
Symmetry of tensorial properties

The symmetry constraints which must be satisfied by the physical properties
of crystals are expressed in a very general form by Neumann's principle (see
p. 14), which states that a crystal point group must be either the same or a
subgroup of the symmetry group inherent to the physical property
considered.
A useful way of applying Neumann's principle is to constrain the tensor
representing the required physical property to be invariant with respect to
any symmetry operation of the crystal point group. Of course it suffices to
check the invariance with respect to the group generators only. For
instance, in the case of a physical quantity expressed by a second-rank
tensor y, the invariance relationship is
this equation has to be satisfied for all symmetry matrices R corresponding

to point-group generators. By solving the equation for all independent
symmetry operations of a given crystal point group, the corresponding
symmetry constraints on the y, components can be derived. In particular,
all ylh components would turn out to be zero for a point group more
symmetrical than the tensor itself, thus violating Neumann's principle. This
algebraic procedure has been already applied in Appendix 3.B (p. 185) for
deriving restrictions on the thermal factor, and will be used in subsequent
sections to obtain the symmetry conditions for a number of tensorial
physical properties in all 32 crystal point groups.
Neumann's principle can also be interpreted from a geometrical point of
view, by considering the symmetry of the geometrical representation of the
tensorial property and comparing it straightforwardly with the crystal point
symmetry. This is certainly feasible for first- and second-rank tensors, while
for higher-rank tensors the method would become quite complicated. A
first-rank tensorial property is represented by a vector, which may be polar
(segment with arrow) or axial (segment with direction of rotation)
(Fig. 9.1). For instance: velocity, force, electric field intensity are repre-
sented by polar vectors, while angular velocity, moment of force, magnetic
field intensity are represented by axial vectors.
Polar and axial vectors differ by their groups of symmetry, as they belong
to the mm and m/m limit groups, respectively. Limit groups of symmetry
(also called Curie groups) are point groups including the infinite symmetry
axes, i.e. rotations of any angles. The mm group of polar vectors (for
instance, a moment of electric dipole) is not centrosymmetric, whereas the
m/m group of axial vectors (e.g. a moment of magnetic dipole) is. Thus
Neumann's principle states that a spontaneous electric polarization, typical
of pyroelectric and ferroelectric crystals, can be observed only for the
so-called polar point groups, which are subgroups of the limit group mm: 1,
2, 3, 4, 6, m, mm2, 3m, 4mm, 6m. The polar vector must be parallel to the
symmetry axis (polar axis), when this is present; in point group m, the Fig. 9.1. Polar and axial vectors.
604 1 Michele Catti
vector may have any direction lying in the mirror plane, while in group 1 its
orientation is unrestricted. On the other hand, a spontaneous moment of
magnetic dipole (ferromagnetism) is consistent only with crystal symmetry
represented by subgroups of the limit group w/m: 1, 2, 3, 4, 6, m, 2/m, 6,
4/m, 6/m, 1, 3, 4. The axial vector may have any orientation in point groups
1 and T, it must be normal to the mirror plane in group m, and its direction
has to be parallel to the symmetry axis in all other cases.
A symmetric second-rank tensorial property is represented geometrically
by a second-order surface (quadric), which may be an ellipsoid or a
hyperboloid of one or two sheets (cf. Appendix 9.A). In general, such a
surface has mmm symmetry, but in special cases higher symmetries can be
displayed. When two tensor eigenvalues are equal but different from the
third one, then the quadric becomes a revolution ellipsoid or hyperboloid
with symmetry a/mm. Because of Neumann's principle, this must occur for
all crystal point groups belonging to the tetragonal, trigonal, and hexagonal
systems, which are subgroups of the limit group w/mm, but not of the group
mmm. Besides, the principal direction of the tensor corresponding to the
unique eigenvalue must be parallel to the symmetry axis 4, 3, or 6. When all
three eigenvalues are equal, the quadric is a sphere with symmetry m/m;
such a situation is required for all cubic point groups (subgroups of w/m),
as both mmm and w/mm symmetries are lower than those consistent with
the cubic system. For the orthorhombic, monoclinic, and triclinic systems,
the second-rank tensor can be represented by a general mmm quadric with
all three eigenvalues different, but with a constrained orientation in the
orthorhombic and monoclinic cases. In the former instance, the principal
directions are to be parallel to the crystallographic axes, while in the latter
case one of the principal directions must be parallel to the unique
monoclinic axis.
The forms shown by first- and symmetric second-rank tensors as imposed
by symmetry are given in Table 9.1. Of course if the symmetric second-rank
Table 9.1. Forms displayed by first- and symmetric second-rank tensors

referred to Cartesian axes
(a) Polar vector p (b) Axial vector a
1 (PI PZ ~ 3 ) 1, i (a, a, a),

m (P, 0 p3) 2, m. 21-m (0 a, 0)
2 (0 P, 0) 3, 3, 4, 4, 6,
3, 4, 6, mm2, 6, 4/m, 6/m (0 0 a),
3m, 4mm, 6mm (0 0 p3)
"
.
(c) Symmetric second-rank tensor y
Triclinic Monoclinic
("I
Orthorhombic
Cubic
iYl1
i2 .)
("l ") Y33
Yl 1
Tetragonal,
hexagonal " .) Y33
Y33
tensor is referred to a lattice basis A rather than to the Cartesian basis E,

unlike the convention followed until now, then the symmetry constraints on
its components may be different from those of Table 9.1. This is true in
particular for the hexagonal system, where non-zero off-diagonal com-
ponents may appear (see also Tables 3.B .1 and 3.B.2).
Overview of physical properties

A number of important tensorial properties of crystals will now be
examined, using the results of tensor theory previously discussed. The
whole subject would include all topics of classical phenomenology applied to
the behaviour of matter: mechanical, thermal, electrical, magnetic, optical
properties. However, only some of them are considered here, because of
limited space. First, the dielectric behaviour of crystals will be illustrated, as
it is particularly suitable for a simple application of the tensorial concepts
just developed. This topic is closely linked to that of optical properties,
because of the relation between refraction index n and relative permittivity
E , (n = V E ,in isotropic bodies); yet only a short overview will be given of
that subject. Then mechanical properties (stress-strain relations, crystal
elasticity), which require use of the highest (fourth) rank tensors presented
in this book, will be considered in some detail. Thermal behaviour is
included in that section through examination of the thermal expansion
phenomenon. Eventually the piezoelectric properties, bridging electrical
and mechanical behaviour of crystals, and providing an example of
third-rank tensors, are analysed.
Electrical properties of crystals

Dielectric media are insulators, they ideally do not allow the transport of
free charges, but only the presence of static polarization charges. Polariza-
tion of atoms is due to electrostatic induction operated by an applied
electric field E . The electronic component of polarization is a separation of
the centre of negative from the centre of positive atomic charges, giving rise
to a moment of atomic electric dipole. If atoms are ionized, then they are
slightly displaced by the electric field from their equilibrium positions, and
contribute to polarization by an additional ionic component. In Fig. 9.2 an
unpolarized pair of ions is shown, where the positions of nuclei A and B
coincide with the centres of negative (electronic) charge distributions X and
Y, respectively. Under the action of an electric field E, A' separates from
X' and B' separates from Y', so as to give rise to an electric dipole within
each ion (electronic polarization). Besides, the relative position A'B'
changes as well with respect to AB, modifying the natural electric dipole of
the ionic pair (ionic polarization).
An example of the order of magnitude of the effect can be given for NaCl
crystals. The C1- ion displays an electronic polarizability (=ratio between
induced electric dipole moment and electric field) of 3.3 x C m2 V-',
while the Naf contribution is an order of magnitude smaller. This means
Fig. 9.2. Above: unpolarized pair of ions. Below:
that a potential difference of 1000 V applied to a NaCl crystal plate 1mm electronic and ionic polarization induced by the
thick would induce a dipole moment of about 3 X C m on each C1- electric field E.
606 ( Michele Catti
ion, corresponding to a separation of centres of positive (nucleus-core

electrons) and negative (valence electrons) charges of -3 x 10-6A. The
contribution of ionic polarizability ( = 6 x C m2V-' for the pair Na+
C1-) would cause the Na+-C1- separation to change by about 4 x lo-' A.
The overall effect is measured by the P vector, the intensity of
polarization, which is equal to the induced moment of electric dipole per
unit volume. For an isotropic body, the first-order approximation gives a
relationship of proportionality:
E~ is the vacuum permittivity (or dielectric constant), and x is a dimension-

less proportionality constant, called dielectric susceptibility, which actually
expresses the magnitude of the polarization phenomenon. The material
response is fully incorporated in Maxwell equations, which for a static
problem take the form: div D = p, curl E = 0, where p is the volume
density of free electric charges. The electric displacement (or electric
induction) vector D is related to the intensity of polarization P according to:
+
D = eOE P. By substituting (9.11) for P , the dependence of D on E for an
isotropic body is found:
where the relative permittivity E , is defined as E ,= 1+ X , and the product

E , E=~E is the permittivity (or dielectric constant) of the isotropic dielectric
medium. When the electric field depends sinusoidally on time (e.g. in an
electromagnetic wave), then E and the related quantities are functions of the
corresponding frequency.
Equation (9.12) relating D to E holds for an isotropic body and within a
linear approximation is valid when the electric field E is not very large. A
generalized equation extended to anisotropic media and to any field
intensity takes the form of a Taylor expansion of type (cf. eqn (9.2)):
Of course a similar expansion could also be written to express the P(E)

dependence, replacing equation (9.11). The coefficients E~~ appearing in
- (9.13) are components of the second-rank permittivity tensor E; on the basis
of thermodynamic arguments, this can be proved to be symmetrical
(chi = E ~ with
~ ) positive eigenvalues. Related tensorial quantities are the
relative permittivity E,= and the dielectric susceptibility x = E, - I.
According to (9.13), now D and E are no longer parallel vectors; within a
linear approximation corresponding to (9.12), they are related by the
L
tensorial equation D = EE. However, the Maxwell equations given above
and the relation D = E ~ +EP still hold also for anisotropic crystals. A
Fig. 9.3. between electric electric
displacement, and electric polarization vectors i n
picture of the three vectors E , D, P in a crystalline slab between the plates
a plane condenser with crystalline dielectric. of a condenser is given in Fig. 9.3.
Pyroelectricity
The presence of the zeroth-order term Dopiin the expansion (9.13) implies
that, in an anisotropic crystal, a spontaneous polarization given by the
vector Po = Do is possible when E = 0. This adds to the normal induced
polarization depending on E and expressed by higher-order terms. When a

crystal shows a spontaneous electric dipole moment Po, it is said to be
pyroelectric; thus pyroelectricity is a property associated with the first-rank
tensor Po,, or Do,,. As Po and Do are polar vectors, all the symmetry
considerations of p. 603 relative to that case hold: therefore pyroelectric
crystals can only belong to point groups which are subgroups of w m (cf.
Table 9.1). Of course an isotropic medium has symmetry higher than ~0 m,
so by Neumann's principle it can not display pyroelectricity.
It should be noticed that, from a practical point of view, the spontaneous
polarization of a pyroelectric crystal has a very short life. Indeed, the
combined effects of a very small crystal conductivity due to defects and of
ionized particles in the air contribute to 'discharge' the condenser cancelling
the electric dipole moment of the crystal. However, a quick temperature
change causes a variation of Powhich makes the spontaneous polarization
detectable as an electric signal. Thus pyroelectricity is properly revealed as a
change of Po with temperature, and the three pyroelectric coefficients are
the derivatives dPo,,/dT.
A particular type of pyroelectricity is that of ferroelectric crystals; for
these, the vector Poof spontaneous polarization can be inverted by applying
an electric field in the opposite direction.
An example of a pyroelectric and ferroelectric crystal of technological
importance (used for infrared detectors) is that of LiTa03 (symmetry group
3m). Its spontaneous electric dipole moment is aligned with the threefold
axis, and the corresponding pyroelectric coefficient has value of 1.9 X
Cm-2 K-'. Thus a temperature change of 1K, brought about for
instance by infrared radiation, causes a dipole moment per unit volume of
1.9 x C m-2 to arise. Taking into account that the principal value of
relative permittivity along the 3 axis is E,,, = 46, we can compute the electric
field which would be necessary to produce the same polarization: E =
- 1) = 4.77 x lo5 V m-'. This is a large field, showing that the
magnitude of the pyroelectric effect for LiTaO, is noteworthy.
Dielectric impermeability and optical properties

When the first-order term in the expansion (9.13) is sufficient to account for
optical behaviour of the crystal, we speak of linear optical properties. The
dielectric impermeability tensor is defined as b = E-', so that E = bD; the
relative impermeability is B = E;' = &,b, and its positive principal values are
Bi = The representation surface of tensor B (cf. Appendix 9.A) is an
ellipsoid of equations
according to whether the reference is a general Cartesian basis or the basis

of eigenvectors. This ellipsoid is called the optical indicatrix, and has the
important property that its semi-axes have lengths equal to B;'I2 = E:!; = Iti,
i.e. to the principal refraction indices of the crystal. Thus the optical
indicatrix is simply the representation quadric of the tensor of relative
dielectric impermeability.
The peculiar optical properties of crystals can be derived by use of this
ellipsoid. Let us consider a propagation direction OP of an electromagnetic
608 1 Michele Catti
wave, drawn through the centre of the optical indicatrix (Fig. 9.4). The
central section of the ellipsoid normal to OP is an ellipse with semi-axes OM
and ON. It can be shown that not just one but two wave fronts propagate
through the crystal normally to OP with different velocities v, = cln, and
v 2 = cln, (c is the velocity of light in a vacuum); the corresponding
refractive indices nl and n2 are equal to the lengths of semi-axes OM and N
ON. The two waves are plane polarized with directions of vibration of the D
vectors parallel to OM and ON, respectively. The whole subject of optical
crystallography, which will not be dealt with here in detail, is based on the
study of properties of the optical indicatrix and other related surfaces. We
would just point out that the orientation and symmetry behaviour of the
optical indicatrix in the different crystal point groups follows all results
derived from Neumann's principle for symmetrical second-rank tensors on
p. 604 and in Table 9.1.
Thus in the cubic system all three principal refraction indices are equal,
the optical indicatrix is a sphere, and the crystal is optically isotropic in all
directions. In the tetragonal, trigonal, and hexagonal systems two principal Fig.9.4. The Optical
indices are equal but different from the third one, whose corresponding
principal axis (called optic axis) is parallel to the high-symmetry direction.
The indicatrix is an ellipsoid of revolution about the optic axis. The two
wave normals propagating along that direction have equal refractive indices,
and therefore coincide; no double refraction is observed, and the optic axis
is a direction of optical isotropy. Crystals belonging to these symmetry
systems are called uniaxial. For the orthorhombic, monoclinic, and triclinic
systems, the indicatrix is a general ellipsoid with two circular central
sections; the corresponding normal directions, lying on the plane of the
longest and shortest semi-axes, are the optic axes and the crystals are said to
be biaxial. The two optic axes are directions of optical isotropy, just as in
the uniaxial case.
In some crystalline materials the optical behaviour depends significantly
on terms of order higher than one (typically, second-order terms) in the
D(E) expansion (9.13). This subject is called non-linear optics, and is
associated with important technological applications. Optical non-linearity is
also known as the electro-optical effect, and the electro-optical coefficients
eZhk are evidently the components of a third-rank tensor. The (9.13)
expansion truncated to second order can be rewritten as:
so that the eihk coefficients may be interpreted as derivatives of permittivity

components dependent on the field E with respect to the field
components themselves; from this comes the name of electro-optical effect.
Elastic properties of crystals

The mechanical behaviour of solid bodies is characterized by a change of
geometrical shape (deformation) as a response to applied forces, apart from
rigid translations and rotations. When the mechanical stress is small enough,
then the corresponding strain is reversible and follows Hooke's law, being
simply proportional to the stress itself: this is the regime of elastic
deformation, well known for isotropic bodies from texts of elementary

mechanics. In the case of crystals, the force field acting on the body and the
ensuing deformation field have to be specified in three dimensions with
respect to the crystal Cartesian reference basis. Hooke's law is generalized
by linear relationships between all stress and strain components, leading in a
natural way to the tensor formalism for anisotropic elastic behaviour.
The study of the mechanical properties of crystals is a valuable subject for
at least two reasons. First, because of technological applications: whatever
use is made of a crystal in applied science or in industry, its response to
mechanical agents is of primary importance and has to be known in detail.
Second, there are important fundamental implications of the elastic
behaviour of crystals, which concern the nature of interatomic forces giving
rise to cohesion and geometrical features of the crystal structure.
Crystal strain
In a general way, the state of strain of the crystal is defined by the vector
field u = x' - x, which gives for every point the change between equilibrium
x and strained x ' position vectors. As usual, the dependence u(x) of the
vector field can be expanded in a power series of type (9.2) (we assume that
the displacement of the point at the origin is vanishing):
Elastic phenomena involve small deformations, for which terms of order

higher than one may be neglected. In this case the strain is said to be
homogeneous, and the ui components are linear transformations of the
position vector components xh of a general crystal point:
The coefficients eih are dimensionless components of a second-rank tensor e

which is generally non-symmetrical.
To understand the geometrical meaning of eih quantities, a two-
dimensional case is illustrated in Fig. 9.5. A square is drawn with sides OA
and OB parallel to the Cartesian xl and x2 directions; then points A and B
have coordinates [xl(A), 01 and [O, x2(B)], respectively. After deformation,
A and B change into A' and B'; according to (9.16), the displacement
vectors u(A) = AA' and u(B) = BB' have the following components:
ul(A) = e11~1(A), u2(A) = e2ixi(A),

ul(B) = e12x2(B), u2(B) = e22~2(B).
The meaning of ell and ez2 components follows immediately: ell =
ul(A)/xl(A), e2,= u2(B)/x2(B). Further, the sides OA' and OB' are
rotated by the angles cpl and 9, with respect to the original directions OA
and OB, and evidently:
n
w
u.I
tan ql = u2(A) e21
tan cp2 = 4 3 ) - e 12 Fig. 9.5. Homogeneous deformation of a square
x2(B) + uz(B) 1 + e22' into a parallelogram.
610 1 Michele Catti
Fig. 9.6. Decomposition of the strain of Fig. 9.5

X1
into a symmetrical strain plus a rigid rotation. __C
Taking into account that, for small strains, ell << 1, e,, << 1, tan q1= q,,
tan q2= q,, we obtain: q, = ezl, q2= el,.
If ell = e,, = 0 and el, = -eZ1, then the strain reduces to an anti-clockwise
rigid rotation by the angle q = ezl; in this case e is antisymmetrical
(eij = -eji). Generally, the strain tensor e can always be written as the sum
+
of a symmetrical E = i ( e e) plus an antisymmetrical w = i ( e - e) com-
ponent: cij + wij = i(eij + eji) + i(eij - eji) = eij. As o is antisymmetrical, it
represents a rigid rotation; therefore E = e - o corresponds to the physically
relevant part of the strain. A geometrical picture of the decomposition
+
e = E o for the planar deformation of Fig. 9.5 is shown in Fig. 9.6.
The Lagrangian strain tensor E is called infinitesimal, because it is suitable
to represent small deformations only; the use of the same symbol as that for
dielectric permittivity may sometimes be confusing. Another tensor which is
more convenient to express larger deformations is the finite Lagrangian
strain tensor:
q = i(e + e + ee). (9.17)
For small strains, the difference q - E = 1/2ee is vanishing, so that use of q
or E tensors is quite equivalent. Let us look now for a relation between the
previous macroscopic representation of strain and the lattice microscopic
nature of the crystal. M is the orthonormalization matrix of the undeformed
lattice basis (E = M A ) , and M' that of the deformed basis (E = MIA'); as
fractional coordinates of points are not changed by a homogeneous
deformation, then M ' x ' = &. From (9.16), x ' - x = ex; by substituting
X' = MI-'&, one obtains:
Substitution into E = $(e + e) and into (9.17) yields the required expressions
for the E and q tensors:
Using the property of orthonormalization matrices ( M - ~ M - ~= G, cf. p. 69),

one has that M'-'M'-l = G ' and M G M = I, so that:
By comparing (9.20) with (9.19), an important difference appears between

the two strain tensors E and q: the former depends on M' and then on the
relative orientation of the deformed and undeformed lattices, while the
latter does not. The q tensor, unlike E, depends only on the metrics of
Physical properties of crystals ( 611
the deformed lattice, and is simply the transformation in a Cartesian basis of

the change of metric tensor G' - G induced by the crystal deformation.
The q tensor is symmetrical, and can be diagonalized by determining its
principal strains ql, q2, q3 and principal directions; these are the only
directions in the crystal which are not changed by the deformation. The
relative change of volume due to strain is given by the expression
AVIV = q1 + q 2 + q 3 = t r q .
The crystal deformation is usually caused by an applied stress arbitrarily
chosen, so that the tensor is a field tensor: it does not depend on intrinsic
properties of the crystalline medium, but rather on an external field, being
thus unconstrained by crystal symmetry. The symmetry, as well as the
crystal system, may be altered by deformation. The same holds, for
instance, for the vector of induced polarization P , which is a field tensor of
first-rank as it depends on the applied electric field E. On the other hand,
another important kind of crystal strain is that caused by temperature
changes AT; the effect is represented by the tensor of thermal expansion,
defined as:
Its eigenvalues cui are always positive, so that the representation quadric of
equation
is an ellipsoid, whose radius vector has a length equal to the inverse square
root of the coefficient of thermal expansion along its direction. The
corresponding volume thermal expansion is
1 dV
--=C
VdT i=l
cuii=tra.
The tensor of thermal expansion a is independent of any external vector

field, but is related to the inner properties of a given crystal. It is thus a
matter tensor, just as tensors of dielectric permittivity or impermeability; its
symmetry properties and orientation must be consistent with the point
group of the crystal, according to Neumann's principle and the discussion of
p. 604. An example of first-rank matter tensor is the vector of spontaneous
polarization Poof pyroelectric and ferroelectric crystals.
Inner deformation
The strain tensor q was related by (9.20) to a change of metric tensor
G' - G, i.e. of unit-cell geometry. This corresponds to a purely homoge-
neous deformation of the crystal structure, leaving the atomic fractional
coordinates constant (lattice strain). If, on the other hand, such coordinates
vary, then in addition to the lattice deformation, an inner strain arises,
which is just defined by the coordinate changes Axi for all atoms in the
asymmetric unit. The inner strain generally occurs as a relaxation of atomic
positions to minimize the energy of the deformed lattice, so it is a function
of the lattice strain. The overall deformation of the atomic arrangement is
612 1 Michele Catti
the sum of these two effects. Changes of interatomic distances due to the
total strain can be decomposed into the separate effects, according to
where d , and d(, are the distances between atoms i and j before and after
deformation, respectively. The first term is due to lattice strain only, while
the second one (in square brackets) is mainly ascribed to inner strain
contribution.
Let us consider an application of these concepts to the strain induced by
thermal expansion in mica muscovite. Two crystal-structure refinements at
25 and 700 "C allowed us to determine the changes of unit-cell constants and
of atomic fractional coordinates for the corresponding temperature range
(Table 9.2). By calculating the metric tensor change G' - G and applying
equation (9.20), the strain tensor q can be obtained; dividing it by the
temperature difference A T yields the tensor of thermal expansion a,which
represents the lattice component of thermal strain: a,, = 1.12, a22= 1.18,
a33= 1.89 x lo-' K-l. The inner component of thermal strain is given by
the changes of atomic fractional coordinates divided by AT. It appears that
the largest contribution to inner deformation is related to basal oxygen
atoms 0(1), 0(2), 0(3), while the effect is very small in all other cases. This
corresponds to a substantial decrease of the ditrigonal distortion of the (001)
layer of (Si, Al)O, tetrahedra sharing corners, which approaches a more
symmetrical hexagonal configuration (cf. Fig. 6.41). The contributions of
lattice and internal deformations to changes of interatomic distances can be
analysed in the case of K - 0 bond lengths; in Table 9.3 the two components
Table 9.2. Lattice constants and atomic fractional coordinates of muscovite

KAI,~Si,AIO,,I(OH), (space group C2/c), at 25°C (above) and 700°C (below)
(Catt~et a/. 1989)
Physical properties of crystals ( 613
Table 9.3. K-0 distances (in A) in muscovite at 25°C ( q j )and 700°C

(dij),and differences between squared distances and corresponding
lattice and inner components (in A2)
of quantities (d,;)2- d$ are reported. A positive inner strain contribution

indicates that the bond expands more than expected from unit-cell dilatation
alone, whereas the opposite is implied by a negative value. Thus, in the K
coordination sphere a very large effect of inner thermal strain is observed,
with a positive sign for the six short bonds and a negative (even the absolute
bond-distance changes are negative) for the six long contacts.
Stress tensor
The other field tensor which is necessary to define the elastic properties of
crystals is the stress tensor. The mechanical forces applied externally to the
crystal are 'contact forces' or 'surface forces' acting on the external surface
of every volume element, and not 'body forces' acting on every point of the
body (like the force of gravity). Therefore the force field is represented by
the vector p (force per unit area) as a function of the unit vector n normal to
the surface element dS. For a homogeneous stress, p depends not on the
absolute position of dS but only on its orientation, so that p =p(n);
furthermore, the dependence is simply linear, being expressed by a tensorial
relationship:
The ti, coefficients are components of the second-rank tensor of stress z. A

general tijcomponent has the physical meaning of oriented pressure along
the ith direction acting on to the dS surface normal to the jth Cartesian
direction. The simple case of surface forces acting onto the three faces of a
cube is sketched in Fig. 9.7. The unit vector n normal to the (100) face has
components [1 0 01, so that using (9.22) the components of the force per unit
area acting on to (1 0 0) turn out to be p1 = z l l , p 2 = t21, p3 = t g l similarly
;
for the (0 1 0 ) and (0 0 1) faces. The diagonal values tiiare called normal
components, and the off-diagonal ones ti,(i # j ) are the shear components
of stress. Analogously to what is true for the strain tensor e, it can be easily
proved that if no rigid rotation of the volume element is produced by the
applied stress, then the z tensor is symmetrical: tji= tij. Therefore,
/
according to the analysis in Appendix 9.A, the stress tensor has real /
/
eigenvalues and can be diagonalized; along its principal directions, the
applied pressure is normal to the surface element. Important particular
cases of stress are that of isotropic (hydrostatic) pressure, occurring when all
4
9.7. Stress field on (1 0), (0 1 01, and
three eigenvalues are equal: tij= -p ajj (the minus sign corresponds to the (o o 1) faces of a crystal cube.
614 ( Michele Catti
convention of considering negative a compressive stress, and positive a

tensile one); and that of uniaxial pressure, when two eigenvalues are zero
(or, in a more general case, different from zero but equal).
Elasticity tensor
A solid is said to be elastic when the dependence of strain on the applied
stress is linear (Hooke's law). If the solid is anisotropic, as is the case for
crystals, then Hooke's law is expressed in tensorial form. The coefficients of
the linear dependence must be characterized by four subscripts, as they
relate the qij and rhk components of two second-rank tensors (cf. eqn
(9.10)):
3
Til =Ch,k~ljhk~hkt (9.23)
1
and, inversely:
1
The quantity cijhkhas the physical meaning of stress component zij which
must be applied to the crystal so as to achieve a deformation state
characterized by a qhk component of unit value. Similarly, syhk is the qij
strain component resulting from application of a unit stress rhk to the
crystal. As the q components are dimensionless, and those of z have the
dimension of pressure, it follows that the quantities cijhk and sijhk have
the dimensions of pressure and of pressure-', respectively.
The coefficients cijhkand sijhkobey the transformation rule (9.8), and then
are components of the fourth-rank tensors c and s , respectively; c is called
tensor of elastic constants or stiffness coefficients, while s is the tensor of
elastic moduli or compliance coefficients. The two tensors are related by the
generalized inversion relationship:
Unlike q and z, which are field tensors, c and s are matter tensors, that is
they characterize an intrinsic property of the crystalline medium and are
independent of the applied force field. On the basis of the symmetry
relations qii = qji and rij= rji for the strain and stress tensors, and of the
condition that the total moment of force applied to the crystal be zero
(otherwise the whole crystal would rotate rigidly), a number of symmetry
relations can be derived for the subscripts of cijhkand sijhkcomponents.
These are:
and similarly for sijhk.Thus of the 81 components of each tensor only 21 are
actually independent.
The mechanical work per unit volume of an infinitesimal elastic deforma-
tion of the crystal is given by
by substituting the expression (9.23) for zij, one obtains:
and integrating with respect to dqii the work per unit volume necessary to
produce the finite strain q is derived:
The linear compressibility PI of the crystal is the relative change of length

along a given direction 1 when the crystal is subjected to a unit isotropic
pressure. By the general properties of second-rank tensors (cf. Appendix
9.A and eqn (9.A.8)), the linear strain along the unit vector 1 is
by substitution of (9.24) for qij, and remembering that z~~= -pShk for
isotropic pressure, one obtains:
Analogously, for the volume compressibility P = -(dVldp)lV; the defor-

mation corresponding to a volume change is AVIV = ELl qii, and sub-
stituting (9.24) for qii gives:
The expression for the reciprocal of volume compressibility, the elastic bulk
modulus K, is obtained accordingly:
K = 110. (9.30)
A simplified convention, based on a condensation of subscripts, is often
used to express components of stress, strain, and elasticity tensors (Voigt's
notation). The symmetrical pair of indices ij (i, j = 1, 2, 3) is substituted by a
single subscript p ( p = 1, 2, 3, 4, 5, 6)) according to the rule: 11 -t 1,
22 -, 2, 33 + 3, 23 + 4, 13 + 5, 12 + 6. Then rp = tii and cpq= cijhk,
with the above correspondence law for subscripts implied. The tensor z is
now represented by a 6 x 1 linear matrix, instead of a 3 X 3 symmetrical
square matrix:
Similarly, in Voigt's notation the elastic stiffness tensor c is represented by

a 6 x 6 symmetrical square matrix (cpq = c,). Things are slightly more
616 1 Michele Catti
complicated for the q and s tensors. In fact, in order to have relations (9.23)
and (9.24) transformed into the corresponding ones
the qp and spqcomponents have to be defined in the following way: qp = qii

(P = 1, 2, 3), =2qij (P = 4 , 5, 6); spq=siihh (P, q = 1, 2, 3), Spq = B i i h k
( p = 1, 2, 3; q = 4, 5, 6), spq= 4sijhk(p, q = 4, 5, 6). For example, q1 = qll,
q4 = 2qZ3,S13 = S1133, S26 = 2s2212,s 4 =~4 ~ If the
~ coefficients
~ ~ 2 and
~ 4.were
omitted in the definition of quantities qp and s,,, they would appear
explicitly in the linear forms (9.31) and (9.32). Furthermore, with the
chosen convention the relation of matrix inversion
holds for the two 6 x 6 square matrices representing the elasticity tensors.
Thus the relations (9.31) and (9.32) can be rewritten in matrix form as:
t= cq, (9.34)
By use of Voigt's notation the expressions (9.27), (9.29), and (9.30) for
the energy of elastic deformation, volume compressibility, and elasticity
bulk modulus become, respectively:
As matter tensors, the stiffness and compliance tensors c and s have to

comply with the requirements of the crystal point symmetry, according to
Neumann's principle. On the basis of p. 603, the simplest method to derive
the symmetry constraints on cpq and spq components is as follows. The
symmetry operations R which are generators of the crystal point group are
considered. For each of them, the Cartesian basis E is transformed into
E' = RE, and correspondingly the transformation of cpq components into c;,
is obtained; by symmetry, the conditions cbq = cpq must be satisfied leading
to the wanted relations. This procedure will be illustrated by some simple
examples.
A twofold symmetry axis parallel to the e2 Cartesian vector transforms
the indices of Cartesian coordinates as follows: 1 + -1, 2 4 2, 3 4 -3.
As the components of a second-rank tensor transform as products of two
coordinates, the corresponding transformation rule for the pair ij of indices
is: 11 + 11, 22 + 22, 33 + 33, 23 4 -23, 13 4 13, 12 -, -12; in Voigt's
notation: 1 -, 1, 2 + 2, 3 4 3, 4 4 -4, 5 5, 6 + -6. Therefore, a
rotation by 180" of the crystal about e2 causes the elastic constants with only
one subscript equal to 4 or 6 to change sign: but, because of symmetry
invariance, such components can only have zero values. Thus in the
monoclinic system, where a twofold axis parallel to e2 is present, the elastic
constants c14, C24) c ~ C ~~ S ,,C16, C261 C36, C56 are always equal to zero. It is
easy to show that no further constraints are imposed on the cp, components
by the other symmetry operations in monoclinic point groups. In fact, by
the same procedure as above the c tensor can be proved to be invariant to
action of the inversion centre. Thus elasticity is a centrosymmetrical
property, and in order to derive the symmetry constraints on the cpq
components only the generators of the point group not containing the
inversion centre need be taken into account. For instance, a mirror plane
normal to e2 is the product of a twofold axis parallel to e2 and of the
inversion centre, so that it is completely equivalent to the twofold axis as far
as the elastic behaviour is concerned.
In the orthorhombic system, two twofold axes parallel to e2 and e3 can be
considered as generators for the 222 and mmm point groups (excluding the
inversion centre). Thus in this case the symmetry constraints on the elastic
constants are the sum of those already found for the monoclinic system, plus
those due to the twofold axis parallel to e3. Such a rotation transforms the
indices of vectors ei according to: 1 + -1, 2 + -2, 3 + 3, and the Voigt
condensed subscripts according to: 1 + 1, 2 + 2, 3 -, 3, 4 + -4, 5 +
-5, 6 + 6. Following the same reasoning as before, this implies that all cpq
components with only one index equal to either 4 or 5 must have zero value.
By combining that result with the conditions found previously for invariance
to a rotation parallel to e2, one obtains that the elastic constants which may
differ from zero are only ell, cz2, c ~ c12, ~ C13,
, C23, c,,, c ~ C6+~ This
, can be
easily shown to be true for point group mm2 as well, and then holds for the
whole orthorhombic system.
The symmetry restrictions on the components of the elasticity tensor c
(Voigt's notation) are summarized in Table 9.4 for all crystal point groups.
Examples and applications

A list of experimental values of elastic constants is given for some inorganic
and organic crystals in Table 9.5. As an example of numerical application,
let us consider the case of anhydrite, CaSO,. For orthorhombic symmetry
the orthonormalization matrix has the diagonal form
Then by means of (9.20), and using the Voigt notation, the following
expressions are obtained for the strain components related to changes of
lattice constants:
ql = $ [ ( ~ ' / a-) ~11, q2 = $[(brlb)2- 11, q3 = $[(C'/C)~
- 11,
77, = $(bllb)(c'lc) cos a ' , qs = 4(a1/a)(c1/c)cos p',
q6 = $(alla)(b'lb) cos y'.
We want to apply, for instance, a uniaxial compression of 1GPa
(= lo9N mT2) to a crystal of CaSO, along the x crystallographic direction,
and to determine the corresponding deformation. The stress tensor takes
618 1 Michele Catti
Table 9.4. Elastic constants and crystal symmetry
Table 9.5, Independent values of elastic stiffnesses c,, (GPa) and compliances s,
(TP~C')of some crystals (Landolt-Bornstein Tables, 1983)
MgO CaCO, CaSO, (C&.)&O CaS0,.2H20 C,,H,

periclase calcite anhydrite benzophenone gypsum naphahalene
cubic trigonal orthorhom. orthorhom. monoclinic monoclinic
p q c s c s C S C S C S C S
the form z = [-1 0 0 0 0 0] GPa, and, by using (9.32) or (9.35), the strain
components are obtained: q = [-0.011 0.00076 0.00128 0 0 01. Taking
into account the above relations between q components and lattice
constants, the unit-cell edges of anhydrite undergo changes of -1.10 per
cent (a), +0.08 per cent (b), +0.13 per cent (c). The mechanical work per
unit volume required to perform this deformation can be calculated by
(9.36), yielding a value of 5.5 MJ m-3. Let us consider now an isotropic
compression of 1GPa of the same crystal, corresponding to the stress
z = [-1 -1 -1 0 0 0] GPa. Again, by (9.35) we obtain the resulting
deformation: q = [-0.00896 -0.00344 -0.00675 0 0 01, corresponding to
relative decreases of the a, b, c cell edges by -0.90 per cent, -0.34 per
cent, and -0.68 per cent, respectively. The energy per unit volume amounts
to 9 . 6 ~ ~ m -and ~ , the relative volume decrease is -1.9 per cent
qq = -0.0192). This value could also have been computed by deriving
the volume compressibility /3 = 0.01915 ( G ~ a ) - lby (9.37), which gives for a
pressure of 1GPa the same volume contraction as that calculated
previously.
Piezoelectricity
Some physical properties of crystals are expressed by a relation between a
vector and a second-rank tensor, rather than between two vectors or
between two second-rank tensors, as have been considered until now. If the
dependence between the components is linear, the relative coefficients
represent a third-rank tensor. This type of third-rank tensor is different
from that examined previously (cf. the electro-optical effect), which
expresses a second-order dependence between the components of two
vectors.
In piezoelectric crystals, by applying a mechanical stress z an electric
dipole moment (per unit volume) P arises, whose components P;: are related
linearly to the stress components:
The dihkcoefficients are called piezoelectric constants (or moduli), and, as

they obey the transformation rule (9.7), represent the third-rank tensor of
piezoelectricity d . On the basis of simple thermodynamic arguments, and
taking into account the symmetry of the z tensor, the piezoelectric constants
dihk can be proved to be invariant with respect to an exchange of h and k
subscripts: dihk= dikh.Thus only 18 out of 27 components of tensor d are
independent.
From a microscopic point of view, piezoelectricity is a polarization of the
crystalline medium due to displacements of ions from their equilibrium
positions, by action of an external stress field. In this case the P vector is
associated with an ionic polarization produced by the second-rank tensor z,
while in pyroelectric or ferroelectric crystals P is due to a quite spontaneous
ionic polarization, and in normal dielectrics to a mainly electronic polariza-
tion induced by the electric field vector E. Besides the direct piezoelectric
effect (9.39), also the inverse effect is observed: this arises when an electric
field E applied to the crystal produces a deformation represented by the
620 1 Michele Catti
second-rank strain tensor q . Thermodynamics demonstrates that the inverse

piezoelectric effect is a necessary consequence of the direct effect, and that
the coefficients relating linearly the q and E components are the same
piezoelectric moduli dihkwhich appear in (9.39):
As in the cases of second- (z, q) and fourth- (c, s) rank tensors, also for
the third-rank tensor d the number of subscripts can be reduced using the
contracted notation of Voigt. Of course only the second and third subscripts
of dihk components, i.e. the pair of indices referring to the c or q
components, are affected by contraction; the first subscript, relative to
components of vector P or E , is not involved. Similarly to what is done for
the strain tensor q , coefficients equal to 2 have to be introduced in the shear
components in order to have the relations
6 3
P, = C diptp
p=l
and rl, = C dipEi
i=l
satisfied: dil = dill, di2= di2,, di3= di33, di4= 2diZ3, di5= 2di13, di6= 2diI2.
According to Voigt's notation, the piezoelectric tensor d is represented
simply by a 3 x 6 rectangular matrix with two-subscript components dip
( i = l , . . . , 3 ; p = 1 , . . . , 6).
Symmetry properties of the piezoelectric tensor

As the piezoelectric tensor is a matter tensor, it must be invariant with
respect to symmetry operations of the crystal point group. Restrictions on
the values of d, components according to the crystal symmetry can be
derived by a method similar to that used in the case of elastic constants.
First, it can be shown that centrosymmetric crystals are never piezoelectric.
The action of the inversion centre operation leaves the centrosymmetric
crystals and the stress tensor c unchanged: on the other hand, the vector P
is reversed, changing its sign. Of course the same inversion operation can
produce either a vector P or a vector -P only if P = 0. The same situation
was found for any property relating a polar vector to an even-rank tensor,
which can thus be observed only in non-centrosymmetric crystals.
Let us now consider a twofold axis parallel to the e2 vector of the
Cartesian basis: this corresponds to the point group 2 of the monoclinic
system, in the standard orientation. The symmetry operation causes the
indices of vectors el, and also of Cartesian coordinates, to transform
according to: 1 + -1, 2 + 2, 3 + -3. In Voigt's notation, the contracted
subscript changes as: 1 -+ 1, 2 + 2, 3 + 3, 4 + -4, 5 + 5, 6 + -6. As
the components of the third-rank tensor d transform as products of three
coordinates with corresponding indices, using Voigt's convention we obtain
that the quantities d, change sign for a rotation of 180" about e2 in the case
of dll, dl,, dl37 d15) d24) d26, d31, d32) d33, d3* But, since by symmetry the
same components cannot change, they must necessarily be equal to zero.
The form taken by tensor d in point group m is different from that shown in
point group 2, otherwise than the tensor c of elasticity which had a unique
form for all monoclinic point groups. A mirror plane perpendicular to e,
transforms the i index of d , as 1 += 1, 2 + -2, 3 + 3, and the contracted

p index as 1 +- 1, 2 +- 2, 3 -+ 3, 4 + -4, 5 + 5, 6 +- -6. Therefore,
repeating the previous reasoning shows that the components d l 4 ) dl6, d Z 1 ,
dZ2,d23j dZs, d34, d36 must be zero.
When the generators of the point group are more than one, the
constraints on the dip components imposed by all the corresponding
symmetry operations must be satisfied at the same time. In the orthorhom-
bic system, the non-centrosymmetric point groups are 222 and mm2. The
former group has two twofold axes, parallel to e2 and to e3 for instance, as
generators; by the previous methods the axis parallel to e3 can be shown to
require that the components d l , , d I 2 ,d l 3 , dl6, d21,dZ2,d Z 3 ,dZ6)d34,d35be
zero. Thus by combining these restrictions with those imposed by the
symmetry axis parallel to e2, we find that in point group 222 only the
piezoelectric components d l 4 , d Z s , and d3, can differ from zero. Analog-
ously, in point group mm2 the symmetry constraints of the twofold axis
parallel to e3 can be combined with those of the mirror plane normal to e2,
Table 9.6. Piezoelectric tensor and crystal symmetry

622 ( Michele Catti
yielding the result that the only components of tensor d which have
non-zero values are dI5) dZ4) d31, d3Z) d33. By similar reasoning the
conditions imposed by symmetry on the piezoelectric moduli can be derived
for all other non-centrosymmetric point groups. In particular, it should be
remarked that in the cubic point group 432 all dip values turn out to be zero:
so this crystal symmetry, though lacking the inversion centre, is forbidden
for piezoelectric crystals. The forms taken by the d tensor in all permitted
point groups are shown in Table 9.6.
Crystal defects
In previous chapters, some properties of the crystalline state were presented
and discussed on the basis of a simplified model, the ideal crystal, relying
upon the fundamental assumption of an atomic structure with perfect
translational periodicity. A large number of very important crystal pro-
perties can be accounted for in such a way: for example, the diffraction of
X-rays, electrons, and neutrons, the main aspects of dielectric and optical
behaviour, and most features of electronic and thermodynamic properties.
However, some crucial phenomena of crystalline solids can not be
explained at all by the perfect crystal model. First, the mechanical
behaviour in non-elastic conditions. The force per unit surface required to
break brittle crystals (break modulus), and the energy necessary to produce
a plastic deformation in ductile crystals have experimental values lower by
several orders of magnitude with respect to theoretical values calculated on
the basis of the perfect crystal model. The maximum shear stress which an
ideal crystal can withstand in elastic conditions is estimated to be about
K1100, where K is the elastic bulk modulus: on the other hand, in real
crystals stresses of K/105 are sufficient to start plastic slip processes. Second,
all transport phenomena in crystals can not be understood by a perfectly
ordered lattice model, where all atomic sites are occupied. This includes
very important processes such as diffusion, which is involved in all chemical
or phase transformations in crystalline solids, and the electrical conductivity
due to ionic motion within the crystal structure. Furthermore, several
particular features of electronic, spectroscopic, and thermal behaviour of
real crystals are not accounted for by the perfect crystal model.
Thus the deviations from ideality, or defects, have to be examined in
detail in order to build up a more general model: the real or defective
crystal. Two classes can be distinguished, point defects and extended
defects. The former ones are a violation of translational symmetry in a
single lattice site: for instance, the absence of an atom from its expected
position (vacancy), or the presence of an atom in an unexpected position
(interstitial). Extended defects, on the other hand, break the lattice
periodicity by relating two portions of the crystal, each of which is perfect in
its interior, in a 'wrong' way. The misfit may involve an extended part of a
plane, and in this case we have a planar defect (e.g. stacking faults in layer
structures); or it may involve just a region closely surrounding a line, so that
we speak of a linear defect or dislocation. Extended defects are weak points
of the crystal, from a mechanical point of view, and are then responsible for
plastic behaviour and for the lower observed strength with respect to
theoretical values. Point defects, on the other hand, provide vacant lattice
sites which make atomic motion possible in crystals, and consequently

diffusion and ionic conduction phenomena. The two kinds of defects are not
quite independent of each other: it will be shown below that extended and
point defects may interact strongly, so that the defective behaviour of
crystals is actually a single field of investigation.
Experimental methods
Methods for direct observation (e.g. microscopy techniques) can be applied
to extended but generally not to point defects, as these usually have
dimensions smaller than the limit of resolution of the best instruments.
Transmission electron microscopy (TEM) is the experimental technique best
suited to observation of extended defects; in the case of high-resolution
instruments, the resolution power can reach 2-3 A (HRTEM). An electron
beam accelerated by a potential of 100 kV is associated with a de Broglie
wave of wavelength A = h l p = 0.037 A, where h is the Planck constant and p
is the linear momentum of electrons (cf. p. 185). Such a small wavelength is
consistent with a much higher resolution than can be attained by conven-
tional optical microscopy (not better than lo4A for monochromatic green
light). A system of electromagnetic lenses allows the electron beam to be
focused on to the sample and on to the fluorescent observation screen,
analogously to what happens with light in an optical microscope. The
regions of the crystal which contain extended defects give rise to changes of
intensity of the transmitted beam (effect of 'contrast'), so as to visualize the
defects themselves.
A scheme of the optical path of a TEM is shown in Fig. 9.8. The beams
which are scattered by the sample at small angles (1 to 2') to the transmitted
beam are focused by the objective lens to form a diffraction pattern at its
back focal plane. By properly focusing the intermediate and projector lens =
system, a magnified image of the back focal plane of the objective lens is
projected on the viewing screen. The intermediate aperture can limit the
sample area from which the diffraction pattern is obtained, so as to explore
very small portions of the specimen (selected area diffraction mode). In the
imaging mode, the aperture at the back focal plane of the objective lens is
inserted in order to block all diffracted beams and let out the transmitted
beam only (bright-field technique); alternatively, all but a single diffracted I
beam are blocked by the aperture (dark-field technique). Intermediate
methods are based on selection of a number of diffracted beams, which
recombine at the image plane giving a contrast contributed only by the
corresponding diffracting directions.
Important sources of errors in operating conditions are due to the
spherical aberration of the objective lens and to an incorrect focusing of the
objective lens. The spherical aberration error causes a slight displacement
A = C,a3 of points on the image plane. cx (rad) is the maximum angle of
electron scattering which can pass through the objective lens (effective (a) (b)
aperture of the lens); C,is the coefficient of spherical aberration (==focal
Fig. 9.8. Optical path of a transmission electron
length, of order 1-3 mm). Taking into account the Rayleigh formula for the microscope: (a) selected area diffraction mode;
resolution of the instrument (R is the size of the resolved object): (b) imaging mode. (1) Specimen; (2) objective
lens; (3) back focal plane of objective lens; (4)
0.61A first intermediate image plane; (5) intermediate
R =-- lens; (6) second intermediate image plane; (7)
0 projector lens; (8) viewing screen.
624 ( Michele Catti
it appears that as a increases the resolution improves, but the aberration

error A becomes larger. Thus an optimum intermediate lens aperture a
would minimize both R and A, and this corresponds to aop,x( A ~ c ~ ) " ~ ,
314 114
AoptxA Cs .
The only drawback of the very powerful TEM method is the need for a
very thin sample (not thicker than about 2000 A), because of the low
penetration of the electron beam into matter. Clearly such thin crystals or
regions of crystals may only have a very small area, so that defects can not
be observed on a large scale. A different technique which is not affected by
this disadvantage is X-ray topography. The diffracted intensity of a very
strong reflection varies locally in crystal regions where extended defects are
present; thus a direct image of the defect structure of the crystal can be
obtained, similarly to what occurs in electron microscopy. However, in
X-ray topography the resolution is much smaller, so that the density of
defects can not be very high. On the other hand, the sample is a large single
crystal and then an extended area of the sample can be investigated. Thus
the two techniques of X-ray topography and electron microscopy appear to
be quite complementary.
The most widely used topographic technique is the transmission Lang
method (Fig. 9.9). A collimated X-ray beam is allowed to pass through a
portion of the crystal, which is oriented for Bragg reflection from a
particular set of planes. The directly transmitted beam is blocked by a
shield, and the diffracted beam is recorded on a photographic plate. The
whole crystal is either immobile (section topograph), or is scanned by
translating it slowly together with the plate (traverse topograph). Very thick
samples should be avoided, to prevent too large an absorption; otherwise, a
back-reflection instead of transmission technique can be used, so that only
the surface crystal area contributes to diffraction. Dislocations and other
extended defects are revealed by diffraction contrast, which is caused by
differences of intensity of the diffracted beam from perfect and from
distorted portions of the crystal. A more detailed discussion of applications
of the Lang method to the study of crystal defects is presented after
introducing dislocations on p. 630.
An older experimental method, which, however, can be very useful and
informative, applies particularly to the study of line defects and is based on
etching of the crystal surface by chemical or electrolytic methods. At the
site where a dislocation meets the surface an etch pit is formed, revealing a
number of details about the nature of the defect.
Indirect experimental techniques have usually to be used to investigate
point defects. For instance, simple (but accurate) measurements of the
Fig. 9.9. Lang method for X-ray topography: (a)

transmission mode; (b) back-reflection mode. S,
shield; P, photographic plate; C, crystal.
unit-cell volume and of the crystal density may give important indications
about the number and nature of point defects present in the crystal.
Another physical quantity which is closely related to such defects is the
electrical resistivity, particularly in ionic crystals where the very small
conductivity observed is entirely due to the thermally activated motion of
ions through the crystal. Ionic transport is made possible by vacancies, so
that resistivity measurements as a function of temperature are able to
characterize many features of point defects in the crystal. Other very
important techniques are based on resonance phenomena (e.g. electron spin
resonance or ESR) which may be particularly sensitive to impurities present
at the concentration level of point defects. These topics will be considered
in more detail in the following sections.
Planar defects
The most important planar defects are observed in layer structures and are
called stacking faults. Let us examine the simplest case of layer structures,
already mentioned on p. 429: the cubic and hexagonal close-packed
arrangements. A single close-packed layer is shown in Figs. 9.10 and 9.11; it
corresponds to a (111) or a (001) crystallographic plane in the cubic or
hexagonal case, respectively. Vectors are shown along the important lattice
directions lying on the plane; for example, in the cubic case (Fig. 9.10) the
arrow along [I121 means the vector a + b - 2c, and is denoted by the
conventional notation a[llZ]. The stacking sequences of layers
ABCABC. . . (cubic) and ABAB. . . (hexagonal) are projected on to the
plane (110) (cubic) or (130) (hexagonal) normal to the close-packed layers
and containing the [112] (cubic) or [210] (hexagonal) direction (Fig. 9.12).
The three positions A, B, C of a general layer are related to one another by
translation vectors equal to a[113]/6 (cubic) or a[210]/3 (hexagonal). Full
circles represent atoms belonging to different layers, linked by thick lines to
emphasize the stacking sequence.
A stacking fault occurs when a single layer takes a different position with
respect to that required by the periodic sequence. This corresponds to a
Fig. 9.10. Closed-packed (1 11) atomic layer of an

FCC structure. The cubic cell and the significant
lattice directions on the plane are outlined.
626 1 Michele Catti
Fig. 9.11. Closed-packed (001) atomic layer of an

HCP structure. The hexagonal cell and the
significant lattice directions are outlined.
rigid translation of a crystal portion by the a[112]/6 vector (FCCcase) on

the closed-packed plane of the fault (Fig. 9.13(a)): the stacking sequence
becomes ABCAICABC. . . . A physical process which can produce such a
defect is a plastic glide caused by a shear stress applied to the crystal; hence
the name of deformation stacking fault given in this case. Another process,
however, can give rise to the same type of fault: the removal of a single
layer from the stacking sequence (a layer of type B in Fig. 9.13(a)), caused
by a condensation of vacancies on the corresponding plane. This is a first
example of correlation between extended and point defects. An example of
a deformation stacking fault for the hexagonal close-packed structure is
shown in Fig. 9.13(b). In this case, by removing, say, an A-type layer the
sequence ABABIBABAB. . . would be obtained, with an unstable con-
figuration due to two equal B planes facing each other. Then the upper B
layer shifts by the vector a[210]/3 so as to take the C position, attaining the
stacking sequence ABABICBCB. . . .
It may also happen that layers following the fault plane are stacked in a
symmetry-related way to those preceding the fault: we then have the
sequence ABCABIACBA. . . (Fig. 9.14(a)), and the fault (corresponding to
a twin boundary) is called a growth stacking fault, because it occurs typically
during the process of crystal growth.
1r l l l l rllolJ
C
B
A
C
B
A
C
Fig. 9.12. Stacking sequences of perfect FCC B
(left) and HCP (right) closeIpacked structures
shown on the (110) and (120) planes,
respectively. A B C A B C A B C [1121
A B C A B C A B C A B C A B C A B C Fig. 9.13. Deformation stacking faults in FCC (a)

(a) (b) and HCP (b) close-packed structures.
Other kinds of stacking faults in closed-packed structures are generated

by a condensation of interstitial point defects, rather than vacancies as
considered before. This corresponds to inserting an additional layer in the
stack. An example for the FCC lattice is shown in Fig. 9.14(b), with the
stacking sequence ABCAICIBCAB. In this case the fault is referred to as
an extrinsic fault, characterized by two breaks in the stacking sequence and
by an extra plane not belonging to either of the lattice patterns on both
sides of the fault. On the other hand, the deformation stacking fault (Figs.
9.13(a) and (b)) is also called an intrinsic fault, with just a single break in
the continuing pattern of the stacking sequence.
All the examples of stacking faults discussed above refer to the simplest
cases of layer structures, the FCC and HCP lattices. However, similar
defects are also observed in more complicated structures built up by parallel
layers related by a finite number of possible translations or rotations (cf.
micas, clays, etc.). When the periodic order of the stacking sequence is
violated by just a single layer, then a planar defect occurs. The basic
processes are always the following: removal of a layer (condensation of
vacancies), insertion of a layer (condensation of interstitials), plastic
deformation by glide of atomic planes by a fraction of lattice vector. In Figs.
9.15 and 9.16 the images of extrinsic stacking faults in muscovite mica
KA12(Si,A1010)(OH)2 are shown, obtained by high-resolution transmission
electron microscopy. The structure of micas (cf. p. 455) displays 1 0 A thick
(001) composite layers stacked in a variety of ways. A single layer is built up
Fig. 9.14. Growth stacking fault (twin boundary)

(a) and extrinsic stacking fault (b) in an FCC
close-packed structure.
628 ( Michele Catti
Fig. 9.15. An image of an extrinsic stacking fault

in mica muscovite, by high-resolution electron
microscopy. (Courtesy of M. Amouric.)
by two (Si, Al)04 tetrahedral sheets sandwiching cations (typically A13+ or

~ g ~ in+octahedral
) coordination. Adjacent layers are separated by sheets
of alkali cations. The unit cell of ordinary monoclinic muscovite contains
two layers, so that the d(ool,spacing amounts to 20 A. However, varieties of
muscovite are known with one (d,,,, = 10 A) and three (d(,,, = 30 A) layers
per unit cell. Figure 9.15 shows a single three-layer slab in a matrix of
two-layer structure, giving rise to a fault in the regular stacking sequence. In
Fig. 9.16 all slabs are one-layer (10 A), but the layer corresponding to F is
rotated by 120" around c* with respect to the normal orientation, so that a
fault again occurs.
Line defects: dislocations

A slip process, induced by shear stress, can give rise not only to stacking
faults in layer structures, but also to line defects in any type of crystal. We
want to consider a particular crystallographic plane, usually characterized by
high atomic density, as a slip plane. Now let a portion of the crystal glide
over the other one by a lattice vector, so that the two portions superpose
perfectly after slipping. This differs from the process generating a planar
Fig. 9.16. HRTEM image of stacking fault in 10 A

muscovite. (Courtesy of M. Amouric.)
Fig. 9.17. Perfect crystal (left) and crystal with a
,
'
,, slip plane but with perfect lattice continuity
(right).
Fig. 9.18. Edge dislocation (left) and screw

dislocation (right). The Burgers vector b a n d the
dislocation line At3 of length l a r e emphasized.
fault, where glide occurs by a fraction of lattice vector breaking the lattice
continuity over the whole slip plane. In the present instance if the slip plane
cuts the whole crystal, no defect is created (Fig. 9.17). On the other hand, if
only a part of the slip plane, limited by a boundary line, is involved in the
glide, then that line separating the slipped from the unslipped portion of the
crystal is a defective region: a dislocation.
Thus a dislocation is a linear defect defined completely by the cor-
responding curve (which is a planar line of length 1 and then determines the
slip plane as well), and by the lattice vector coplanar with it which measures
the magnitude and direction of slip (Burgers vector b). A perfect superposi-
tion of lattice points of the two slipped portions of the crystal is observed far
from the dislocation line, while close to it the lattice periodicity fails. It is
important to examine the orientation of the Burgers vector with respect to
the dislocation line. Two limit cases are observed for a straight line: the
edge dislocation, with b normal to the line, and the screw dislocation, with b ,
parallel to the line (Fig. 9.18). If the dislocation line is a general curve
',
(Fig. 9.19), then the dislocation character (edge or screw) changes from C
point to point, according to whether the unit vector 1 tangent to the line in a b
given point is normal or parallel to the Burgers vector; for an intermediate Fig. 9.19. Genera, dislocation with mixed edge-
angle, the dislocation is said to have a mixed character. screw character.
The Burgers circuit

Let us now try to give a more precise definition of the Burgers vector,
which, as was shown before, is unique and constant for a given dislocation,
630 ( Michele Catti
Fig. 9.20. Burgers circuits for an edge

dislocation (left), and for the corresponding
perfect crystal (right).
Fig. 9.21. Burgers circuits for a screw dislocation

(left), and for the corresponding perfect crystal
(right).
independent of the position along the dislocation line. A Burgers circuit is

defined as a closed path that starts from a lattice point and comes back to
the same point by encircling the dislocation line without crossing it. Now if
the defect is removed, the previous Burgers circuit is modified so that it is
no longer a closed line: the part exceeding or missing with respect to a loop
is a lattice translation called the Burgers vector b of the dislocation. In Figs.
9.20 and 9.21 the Burgers circuits relative to an edge and a screw
dislocation, respectively, are shown.
It was stressed on p. 626 that planar defects can arise not only by a slip
process, but also by a condensation of point defects (vacancies or
interstitials). This is true for dislocations as well. A condensation of
vacancies corresponds to removing a lattice half-plane, while in the case of
interstitials a new half-plane is inserted. Both processes lead to the
formation of an edge dislocation.
X-ray topography of dislocations

An example of dislocations detected by X-ray topography in quartz is shown
in Fig. 9.22. The nature of contrast caused by line defects in X-ray
topographs is generally quite complex, as can be seen in the magnified
images of two dislocations in a silicon crystal (Fig. 9.23). The contrast is
contributed by three different superposing effects:
1. A direct image, due to diffraction of the X-ray beam by the defective
crystal region close to the dislocation line; it is accounted for by the
simple kinematical (or geometrical) theory of diffraction.
2. A dynamic image, related to the coupling of the reflected and incident
waves into wave fields with directions comprised between those of the
primary and diffracted beams (Borrmann fan), as required by the
dynamical theory of diffraction. The dynamic image is the shadow cast
by the dislocation line in the fan of these wavefields.
Fig. 9.22. X-ray topograph of quartz

reflection), showing a great number
dislocations. (Courtesy of A. Zarka.)
3. An intermediary image, consisting of fringes due to new wave fields

which are created by interaction of the dynamical wave fields with the
defective region. When the crystal and photographic plate are translated
(traverse topograph), the direct images form projections of the disloca-
tions in the reflected direction (black lines).
In the section topograph of Fig. 9.23, the direct images are simply black
points; the dynamic images are thick white lines, and the intermediary
images a series of black fringes. Knowing both the direct and dynamic
images, it is possible in most cases to reconstruct the position of the
dislocation line within the crystal.
Moreover, X-ray topography is very useful for determining the direction
of the Burgers vector of the line defect. Generally speaking, any crystal
defect is invisible in a given Bragg reflection if its displacement field has no
component parallel to the Bragg vector rH. Let u be the vector field of
displacement from regular lattice periodicity caused by a dislocation of the
Burgers vector b and line vector 1; the invisibility condition is then
u r, = 0. It can be shown that, in a general case and under the
approximation of isotropic elasticity, there are non-zero displacement
components of the defect parallel to b, to 1 X b and to 1 X b X 1,
respectively. For a pure screw dislocation, where b is parallel to 1, we have
that 1 X b = 0 and 1 X b X 1 = 0, and thus there is a single invisibility
.
condition b rH = 0; this corresponds to complete invisibility of the disloca-
tion for the zone of all Bragg planes which contain 1. In the case of a pure
edge dislocation, b is normal to 1 and then is parallel to 1 X b X 1: therefore
we have two invisibility conditions, b . rH = 0 and 1 X b . r~ = 0,
corresponding to the single Bragg plane orthogonal to the line vector 1.
In principle, mixed dislocations always have a u component parallel to rH
and then are never invisible. However, the displacement component parallel
to b is dominant in causing diffraction contrast, at least when the absorption
is low, so that in these conditions all dislocations are hardly visible if
.
b r, = 0. Besides, it can be shown that the width and the integrated excess ~ ~ ; ~ ~ ~ ; ~ \ s,atl ~ v s i ; f
density of the dislocation image on the topograph are approximately face(ll1).(courtesyof~.Authier.)
632 1 Michele Catti
proportional to b rH and (b . rH)2, respectively. The Burgers vector

direction of a given dislocation can thus be determined unambiguously by
choosing properly the diffraction conditions in the X-ray topograph. By
using a number of Bragg reflections, the differences in diffraction contrast of
the dislocation image allow us to detect the Bragg vectors r, for which
.
b rH = 0, and then the direction of b. From measurements of the image
width it is also possible to derive or to estimate the magnitude, b, of the
Burgers vector, obtaining a complete characterization of the line defect.
I
Energy of a dislocation
Dislocations form, move, and interact with one another in a crystal
according to the energy of elastic distortion of the lattice associated with
them. It is thus important to analyse the elastic strain around line defects
using the basic ideas of crystal elasticity developed on p. 614, and to find out
which factors control the dislocation energy. The simplest case is that of a
pure screw dislocation, which can be represented for convenience as a shear
deformation of a cylindrical ring of isotropic material (Fig. 9.24). A radial
slit LMNO is cut parallel to the z axis, and the free surfaces are displaced
rigidly with respect to each other by the distance b in the z direction (cf.
Fig. 9.18(b)). Owing to the symmetry of the problem, cylindrical r, q, z
rather than Cartesian coordinates are best suited to represent the strain and
stress field in the ring. The only non-zero strain component turns out to be
the shear term e,, = b/2nr. The corresponding stress is z,, = G E ~=,
Gb/2nr, where G is the shear modulus (the equivalent of a shear
component of the elasticity tensor).
It is easy to calculate the strain energy per unit volume W of the
Fig. 9.24. Elastic strain of a cylindrical ring
deformed ring as half the product of stress x strain:
simulating the lattice distortion of a screw
dislocation. w = ~ t , , ~ ,=,$ G ( b / 2 n ~ ) ~ . (9.41)
As the radius r of the cylindrical ring varies between the inner r, and the
outer rl values, it is necessary to integrate the above expression in order to
obtain the total deformation energy per unit length of the ring. For a section
of thickness dr and unit length, the volume is 2nr dr; thus the overall energy
is given by:
1 '-1
n r= - Gb2log - .
G (b / 2 n ~ ) ~ 2 dr
4n ro
This formula represents the elastic energy per unit length of a screw
dislocation, neglecting the contribution of the dislocation core (r < ro) where
strains are very large and the elasticity theory fails. However, estimates
suggest that the core energy is only a small fraction of the elastic energy,
owing to the much smaller crystal volume involved in the core distortion.
The ro length is of the order of 1 0 - ~ m ,rl is of the order of the crystal
dimensions, and a typical value of the shear modulus G is 4 x 10lONm-2;
for a Burgers vector of, say, 2.5 x 10-lo m, and a crystal of 0.01 m the
energy of a single screw dislocation amounts to 3.2 x lop9J m-'.
The energy per unit length of an edge dislocation is given by expression
(9.42) divided by 1 - v, where v = (3K - 2G)l[2(3K + G)] is Poisson's ratio
( ~ 0 . 3 ) This
. is related to the fact that not only shear but also compressive
stresses are involved in edge dislocations; thus the bulk elastic modulus K in
addition to G appears in the corresponding energy expression.
An important aspect of eqn (9.42) is that the dislocation energy is
proportional to the square of Burgers vector magnitude b. As a conse-
quence, for a dislocation with a large value of b it is energetically more
favourable to separate into several dislocations characterized by b values as
small as possible (i.e. equivalent to a lattice spacing). Partial dislocations
(cf. p. 634 below) arise from splitting of a normal dislocation so as to attain
Burgers vectors shorter than a lattice spacing, in association with a stacking
fault.
Motion and interaction of dislocations

An important feature of dislocations is that they can move through the
crystal, as a consequence of applied stress, of thermal effects, and of change
of point defect distribution. The simplest kind of motion is the conservative
glide, which is equivalent to a shift of the dislocation line in a direction lying
within the slip plane. This corresponds to displacing the strained part of the
crystal by a glide on that plane.
In a pure edge or screw dislocation, the straight dislocation line (of length
1) moves in a direction perpendicular to itself on the glide plane, covering a
distance x. Thus the final result of the process would be that the line
emerges on the crystal surface with disappearance of the defect (Fig. 9.17),
and with an overall slip of a crystal portion by the length b of the Burgers
vector. Therefore, the glide process can be considered to be driven either by
a force F' applied to the moving portion of the crystal through the distance
b, or by a force F applied to the moving dislocation through the distance x.
The former one is evidently equal to tlx, the product of stress (=force per
unit surface) by the area; the latter can be derived by equating the work
performed by the two forces:
W = tlxb = Fx,
whence:
F = zbl.
So the force per unit length experienced by a dislocation when an external
stress z is applied to the crystal is simply the product of stress by the length
of Burgers vector.
If two screw dislocations are close enough to each other, the stress
z, = GbI2nr associated with one of them will act on the other as if it were
an external stress. Thus r in this case is equal to the distance d between the
two dislocations, and substituting the expression for t into (9.43) a formula
is derived for the force per unit length between the two line defects:
Using the G and b values assumed in the numerical example on p. 632, the
interaction force per unit length between two dislocations at a distance of
100 A turns out to be 0.04 N m-'. It should be noticed that the force (9.44)
decays very slowly with distance, according to the l l d dependence, so that
the interaction between dislocations is quite long range.
634 1 Michele Catti
A kind of non-conservative motion of dislocations, which must be

thermally activated, is the climb. This is typical of edge dislocations, when
the line moves in a direction normal to the slip plane. Such a process is due
to a condensation of vacancies or interstitials along the dislocation line. In
the first case the extra lattice half-plane is eroded, with climbing of the line
on the same side with respect to the slip plane. In the second case the
half-plane is extended beyond the slip plane, and the dislocation line climbs
on the other side of that plane.
Evidently, a dislocation line cannot end within the crystal: it either forms
a closed loop, or joins another dislocation, or ends on the crystal surface.
Thus in general a network of dislocations (Frank net) is present in the
crystal, with its extremities ending on the surface. When three or more
dislocations meet in a point (a node of the network), then the sum of their
Burgers vectors can be easily proved to be zero; this result is formally
analogous to Kirchhoff's law of electrical networks. A global property
characterizing the network of linear defects in the crystal is the dislocation
density N, defined either as number of dislocations intersecting a unit area,
or as total length of dislocations per unit volume; in both cases the quantity
N is measured in units of cm-'.
Partial dislocations
A peculiar type of linear defect is observed in crystals showing layer
structures, and is closely connected with stacking faults (p. 625). Let us
consider the close-packed cubic structure: atomic layers (Fig. 9.10) are
represented by (111) lattice planes (i.e. the system of symmetry equivalent
planes ( I l l ) , ( l l ? ) , (1?1), ( i l l ) ) , and clearly most processes of plastic glide
related to formation of dislocations occur on these layers. The shortest
lattice vector on the (111) plane, a[li0]/2, is the most probable Burgers
vector for dislocations in this case. Similarly, in a general layer structure the
layer plane is preferred as slip plane for formation of dislocations. However,
we have learnt (p. 626) that a slip on these layers by a suitable vector,
different from a lattice vector, may give rise to a stacking fault. In the FCC
structure such a vector is typically a[112]/6, translating for instance a B
layer into a C position, while the whole stack above the slip plane follows it
(cf. Fig. 9.13(a)). Now if the glide process occurs not on the whole plane
area, but just on a part of it, then at the border between slipped and
unslipped portions a linear defect arises. This is called partial dislocation,
or Shockley partial: it is quite equal to a normal dislocation, except for its
Burgers vector not being a lattice vector. Thus a stacking fault not
extending through the whole crystal is bounded by a partial dislocation. If
the fault occupies a ribbon on the slip plane, then its boundary is formed by
two partial dislocations, one for each side. In this case the Burgers vectors
of the two partials have as sum a lattice vector, which is the Burgers vector
of the unit dislocation sum of the two partials themselves. Let us consider
an ordinary dislocation with Burgers vector a[li0]/2 on the slip plane (111)
of an FCC structure (cf. Fig. 9.10); it can be decomposed into two partials
separated by a stacking fault according to the following relation between the
corresponding Burgers vectors:
It should be stressed that the vector on the left-hand side of the equality is a
latttice vector, while those on the right-hand side are not; this is consistent
with the characters of unit and partial dislocations, respectively. The driving
force of the above dislocation reaction is the minimization of the dislocation
energy (9.42), brought about by a decrease of the Burgers vector b.
Small-angle grain boundaries

-
In a densely pressed or sintered polycrystalline sample, the different

crystallites (also called grains) usually have different orientations which may
be quite random in some cases. Their interfaces are known as grain
boundaries, and can be considered to be extended surface defects of the
solid specimen. However, even a single crystal is normally divided into a
mosaic of crystallites, which are only slightly misoriented with respect to
one another. So in this case we speak of small-angle grain boundaries for
the surfaces separating the crystallites. Two parts of a crystal with slightly
different orientations, separated by a small-angle grain boundary, are
sketched in Fig. 9.25. The wedge-shaped gap in between tends to be filled
by portions of lattice planes, giving rise to an array of edge dislocations
(points P, Q, R) on the surface of the grain boundary. Let 3 be the small
angle between the surface lattice planes of two adjacent crystallites. The
points P and R are separated by a distance blsin (312) = 2b13, so that the
spacing between adjacent dislocations P and Q is simply b l q . For a very
small ?) value of 0.1" (= 0.0017 rad) and a Burgers vector of 2.5 x 10-lo m,
the distance between edge dislocations at the grain boundary turns out to be
1430 A.
Point defects
Besides vacancies and interstitials (intrinsic defects), which have already
been mentioned on p. 622, a third kind of point defects should be
M
Fig.9.25. small-anglegrain boundary,showing
considered. These are impurities, and correspond to substitution of an atom the array of edge dislocations at the interface.
at a regular lattice site by another atom of a different chemical species
(extrinsic defect). All such types of point defects are usually observed in
ionic, covalent, and metal crystals. If the crystal is not monoatomic, it is
necessary to distinguish between stoichiometric and non-stoichiometric
compounds. In the first case the defect concentrations of different atomic
species are related by the constraint of constant ratios between numbers of
atoms, while this does not occur for non-stoichiometric crystals. Further-
more, in ionic solids the numbers of point defects concerning anions and
cations are always constrained by the need for electroneutrality. We shall
examine stoichiometric ionic crystals in more detail.
A single vacancy, either cationic or anionic, would violate the crystal
electroneutrality; thus it must be associated with another vacancy of
opposite sign, or to an interstitial of the same sign. In the former case we
have a pair of vacancies of different signs called Schottky defect: in the
latter one, a vacancy + interstitial pair (Frenkel defect) is observed. Also an
ionic impurity with different charge with respect to the substituted ion
introduces an electric unbalance and has then to be compensated by a
vacancy or an interstitial of the same sign. For instance, a M2+ impurity in a
636 1 Michele Catti
sublattice of A' cations would be accompanied by a vacancy of A+, while a

A' impurity in a matrix of M2' would require an additional A' interstitial.
Let us consider the class of alkali and silver halides AX with the NaCl
type structure. In alkali halides the predominant defects are Schottky
defects: pairs of A+ and X- vacancies, corresponding to empty sites in the
two FCC sublattices forming the structure (Fig. 9.26(a)). The missing ions
have migrated to the external surface of the crystal, or to internal surfaces
(e.g. grain boundaries in polycrystals) or to extended defects (dislocations,
stacking faults). Every vacancy is characterized by a relative electric charge
(called effective charge), which is evidently equal in absolute value and
opposite in sign to the charge of the missing ion. Thus a vacancy of Na'
bears an effective charge of -1, and a vacancy of C1- a charge of + l ; the
symbols V i and V; are used, respectively, so that a Schottky defect is
+
denoted by the pair V, V;. The space distribution of positive and
negative vacancies should not be random, in principle, but is conditioned by
the mutual electrostatic interaction. Vacancies of opposite signs attract each
other, favouring the formation of pairs in adjacent positions, or of clusters
of vacancies with peculiar configurations. However, this tendency increases
with the concentration of defects, so that for very high dilution of vacancies
in the crystal their interaction can be often neglected to a first
fk\ approximation.
In silver halides AgX, Frenkel defects involving cations are dominant. An
Fig. 9.26. A Schottky defect (pair of cationic and
anionic vacancies) (a) and a Frenkel defect Ag+ ion leaves its regular lattice position, corresponding to an octahedral
(vacancy + interstitial) (b) in the NaCl type hole in the FCC sublattice of X- anions, and occupies an interstitial position
structure.
in a tetrahedral hole (Fig. 9.26(b)). In a sense, the presence of Frenkel
defects in this case suggests an incipient transformation of the NaCl type
structure into that of ZnS type, where coordination environments of both
cations and anions are tetrahedral instead of octahedral. Similarly to
vacancies, interstitial defects are characterized by an effective charge,
relative to the regular structure: this is simply equal to the charge of the
interstitial ion itself, and the defect is denoted by the atomic symbol with
the i subscript (i.e. Ag?). A Frenkel defect in the AgX halide is symbolized
+
by the pair Vi, Ag:.
Thermal distribution of defects

An important question which should be asked is: why do intrinsic point
defects form, and which physical variables control their concentration? To
give an answer, the problem has to be tackled from a thermodynamic point
of view. The formation at constant pressure of a single vacancy in a
monoatomic crystal requires a change of enthalpy AH, necessary to remove
the atom from its lattice site and to bring it to the crystal surface. On the
other hand, a disorder arises in the crystal due to the random position of the
vacancy in the lattice. Let N be the number of atomic sites per unit volume,
and n, the number of vacancies per unit volume in the crystal. As the
number of ways of distributing the n, defects over the N sites is
w = N ! / [ ( N- n,)!n,!], the corresponding increase of configurational en-
tropy brought about by this disorder is
The Stirling approximation for the natural logarithm of the factorial

(log N! = N log N - N) was used; k is Boltzmann's constant. The change of
Gibbs free energy of the crystal due to the formation of n, vacancies
includes the enthalpic and entropic effects, according to AG = AH -
TAS = nvAHv- TAS. At a given temperature T, equilibrium is attained
when n, is such to make AG a minimum:
By solving this equation with respect to n,, we obtain the result

log [n,l(N - n,)] = -AH,lkT; remembering that n, << N (so that N - n, =
N), the formula relating the equilibrium concentration of vacancies to
temperature is derived:
n, = N exp (-AH,/kT). (9.45)
If the crystal is ionic, the number of cationic vacancies must be equal to the
number of the anionic ones. Thus it can be easily shown that the
concentration of Schottky defects as a function of temperature is the
following:
n, = N exp (-AHJ2kT). (9.46)
In the case of Frenkel defects, the corresponding distribution is:
nF = VNN, exp (-AHF12kT), (9.47)
where Ni is the number of available interstitial sites per unit volume.
It should be stressed that the relationships (9.49, (9.46) and (9.47) were
derived under some simplifying assumptions, the most important of which
are:
1. Defects do not interact with one another, and then are distributed in a
perfectly random way. This holds to a first approximation only when the
concentration of electrically charged defects is very small.
2. The contribution to entropy due to atomic vibrations and the depend-
ence of defect enthalpy on temperature can be neglected. Such an
approximation is valid for temperatures which are not very high.
The key significance of the above equations is that formation of intrinsic
point defects is brought about by an entropy increase, and that their
concentration changes with temperature according to an exponential law.
Diffusion
One of the fundamental phenomena made possible by point defects in
crystals is the mobility of atoms and ions in the solid state. The process
concerning mass transport by purely thermal activation, with absence of
electric fields, is called diffusion. It is governed by Fick's laws, the second of
which is the following:
where c is the concentration of the diffusing atomic species at time t and at

638 1 Michele Catti
position x along the diffusion direction in the crystal. By integrating the

differential equation (9.48) the theoretical concentration profile c ( x , t ) with
respect to space and time is obtained. Measurements based on radioactive
tracers or electron microprobe can give an experimental concentration
profile: by comparison with the theoretical one, the diffusion coefficient D is
derived. If the whole procedure is repeated at different temperatures, the
quantity D can be shown to depend on temperature according to the
Arrhenius law:
D = Do exp (-HaIkT), (9.49)
where Do is a constant factor and Ha is the activation enthalpy of the
process.
From an atomistic point of view, diffusion consists of thermally activated
jumps of atoms or ions from regular lattice positions into empty neighbour-
ing sites (vacancies). The quantity Ha represents the enthalpy needed by the
atom to overcome the potential barrier between starting and ending sites.
Let jumps occur in a random way, n, be the total number of vacancies per
unit volume, and a the distance between regular site and vacancy; then the
following expression can be proved to hold for the quantity Do of equation
(9.49):
v is the frequency of vibration of the atom attempting to hop, which is

proportional to the probability of jumping into a neighbouring vacancy; g is
a geometrical factor depending on the detailed crystal structure (for a
primitive cubic lattice, g = 116). Therefore, Do and the diffusion coefficient
D are proportional to the concentration of vacancies n,.
\
log D A However, the diffusion coefficient D as a function of temperature shows a
more complex experimental behaviour (Fig. 9.27). Two regions, of high and
low temperature respectively, can be distinguished: these are characterized
by quite different slopes of log D versus 1IT. Moreover, in the low-
temperature region the observed straight line may be shifted up or down by
changing the crystal sample of. the same substance. The explanation is
related to the presence of impurities of aliovalent ions, which, as mentioned
mtr~ns~c, \ on p. 635, give rise to vacancies distinct from those due to thermal disorder.
reg1on 1 lmpur~ty For instance, crystals of NaCl can contain small, different quantities of
1 \, region
" divalent impurities (e.g. ~ n ' +ions) substituting Na+ cations: then for each
1/T
Mn2+ impurity a vacancy VNa is created in order to keep the electroneutra-
Fig. 9.27. Arrhenius plot Of log
versus 1/T, showing the impurity and intrinsic
lity. The number of these vacancies is independent of temperature, changes
regions characterizing diffusion in crystals. from sample to sample, and at low temperature is much larger than that of
thermal (intrinsic) vacancies. Thus at low temperature (impurity region) the
quantity n, of eqn (9.50) represents essentially the number of vacancies due
to impurities and is a constant; the slope of the straight line log D versus
1 / T is simply Ha, according to (9.49). At high temperature, on the other
hand, intrinsic vacancies outnumber extrinsic ones (intrinsic region), and
expression (9.46) for the thermal distribution of Schottky defects must
replace n, in equation (9.50) for Do. The result obtained for the diffusion
coefficient D is:
D = g v a 2 exp
~ (- Ha + k y / 2 ) .
Therefore, in the intrinsic region the slope of the line log D versus 1 / T
equals (Ha+ AHJ2)Ik instead of Halk, and is then larger than the
corresponding one in the impurity region.
Ionic conductivity
Alternatively to the thermally activated process (diffusion), ionic transport
in crystals is driven by an applied electric field. In this case electric
conduction based on migration of ions rather than electrons occurs;
however, the atomistic mechanism always relies upon the presence of point
defects, just as for diffusion. Generally, the relevant contribution to
conductivity is given either by cations (e.g. alkali or silver ions) or by anions
(halide or oxide ions). In either case the conductivity a is proportional to
the number of charge carriers per unit volume, n, and to the mobility of the
migrating ion, p (e is the electron charge); for cations:
By considering the equilibrium condition between the field-induced drift

current and the opposing diffusion current due to the concentration
gradient, a very important relationship between mobility and diffusion
coefficient (Nernst-Einstein equation) can be derived:
This equation provides a link between the phenomena of diffusion and of

ionic conductivity. By substitution into (9.52) one obtains:
If the process occurs in the intrinsic (high-temperature) region of Fig. 9.27,

where the number of Schottky defects follows Boltzmann's thermal dis-
tribution (9.46), then eqn (9.51) holds for D and can be substituted into
(9.54), yielding:
On the other hand, eqn (9.54) holds in the low-temperature impurity

region, where the number of vacancies in the expression (9.50) for Do is
independent of temperature.
From an experimental point of view, it is much easier to perform
measurements of electrical conductivity than of diffusion coefficients. Hence
the importance of this technique for studying the energetics of point defects
appears clearly: by plotting log a T against 1 / T for an ionic crystal, one
obtains diagrams quite similar to that of Fig. 9.27, showing two straight lines
with different slopes for the intrinsic and impurity regions. On the basis of
+
eqns (9.55) and (9.54), these slopes are equal to (Ha AHJ2)lk and to
Halk, respectively. Thus it is possible to derive from these measurements
both the activation enthalpy Ha for hopping of ions into vacancies, and the
formation enthalpy of Schottky defects AH,. Results of experiments on
NaCl give values of 0.68(1) eV for Ha, and 2.30(1) eV for AH,.
640 1 Michele Catti
Appendix
9.A Properties of second-rank tensors

Second-rank tensors represent physical laws of linear type in crystalline
media. A very important class is that of symmetrical second-rank tensors,
which are characterized by the property yj, = yij for all their components in
any reference basis. Antisymmetrical tensors, on the other hand, follow the
rule yji = -yi,, implying that diagonal components are zero (yii = 0). It may
be shown easily that any tensor can be written as the sum of a symmetrical
plus an antisymmetrical tensor. Symmetrical tensors are associated to the
quadratic form:
a
which is also the equation of a general second-order surface (a quadric); this

surface is obviously invariant with respect to a change of reference basis,
and is thus a correct geometrical representation of the symmetrical tensor y.
Eigenvalues and eigenvectors

In order to study the features of the representation quadric (9.A.1), the
equation Y = AX, or
yX = AX, (9.A.2)
should be considered. The solutions of this equation are X vectors which are
transformed by action of the tensor y into parallel vectors AX; so the
corresponding directions define axes of isotropy in the crystal. Such
solutions are called eigenvectors, and the corresponding A values are the
eigenvalues of the tensor. It is important to stress that eigenvalues (scalars)
and eigenvectors (vectors) are invariant quantities which do not depend on
the reference basis; only their components do. The matrix equation (9.A.2)
is equivalent to a system of three homogeneous linear equations in the
unknowns X I , X,, X 3 which admits non-zero solutions if and only if the
determinant of coefficients is zero:
det (y - Al) = 0. (9.A.3)
By calculating the determinant explicitly, an equation of third degree in the
unknown A is obtained; it can be proved that all its three roots are real
because y is symmetrical. By substitution of each real eigenvalue A,, A2, A3
into (9.A.2) the corresponding eigenvectors XI, X2,X3 can be found. One
of the three linear equations implicit in (9.A.2) has to be rejected, because
it is linearly dependent on the other two (cf. eqn (9.A.3)); therefore only
directions and relative (not absolute) moduli of eigenvectors can be
determined. It can be demonstrated that if all three eigenvalues are
different, then the eigenvectors are unique and orthogonal to one another;
otherwise, they may not be unique but can always be chosen to be
orthogonal. It is also convenient to normalize the eigenvectors, so that the
normalized components QLhare dimensionless quantities which obey the
orthonormality condition:
3
C.
h=l
Qih Qjh = dij,
where the first subscript is the identity number of the eigenvector, and the
second one specifies the Cartesian component.
It is very useful to change the reference basis from the original one to that
formed by the three normalized eigenvectors Q,, Q2, Q3, and to find out
the representation of tensor y in the new basis. The transformation matrix
Q has rows given by the components Qij of each eigenvector:
By applying the tensorial transformation rule (9.6), we obtain:
taking into account (9.A.2) and (9.A.3). This means that the representation
of tensor y in the basis of its eigenvectors is a diagonal matrix whose
non-zero components are just the eigenvalues A,, h2, hg. The eigenvalues
take the names of principal components (yl = dl, y2= A,, y3 = S),and the
directions of the eigenvectors are called principal axes (or principal
directions) of tensor y; this is represented with respect to the basis of its
eigenvectors by the matrix:
Along the principal directions the crystal behaves isotropically, i.e. each
eigenvector Xi is transformed by tensor y into a parallel vector k; according
to = y i X i , which is the equivalent of (9.A.2).
Representation surfaces and their properties

With respect to a general basis, the action of tensor y is expressed as:
while in the basis of eigenvectors the correspondent relation is simply:

Y;: = yiXi. (9.A.6)
In the latter basis, the equation of the representation quadric (9.A.1)
becomes:
From the general features of second-order surfaces and their equations, it

turns out that if all eigenvalues are positive then (9.A.7) is the equation of
an ellipsoid; if only one is negative, the surface is a hyperboloid of one
sheet; if two eigenvalues are negative, we have a hyperboloid of two sheets,
and if all of them are negative, an imaginary ellipsoid is obtained
(Fig. 9.A.1). Of course in the latter case one usually considers as repre-
642 1 Michele Catti
Fig. 9.A.1. Representation surfaces of second-

rank tensors: ellipsoid, hyperboloid of one and
two sheets.
sentative surface the real ellipsoid of equation
Let us consider now the case of all eigenvalues positive. Then by writing
equation (9.A.7) as EL, (xi/aJ2 = 1, it turns out that the semi-axes of the
representation ellipsoid have lengths given by ai = l / v y i , i.e. they are equal
to the inverse square root of principal tensor components. The symmetry
axes of the quadric correspond to the principal directions of the tensor.
We want now to derive an expression for the component y, of tensor y
along a given direction characterized by direction cosines 4, 12, l3 (then I is
the unit vector parallel to that direction). Such a component is defined as
the projection onto I of the Y vector corresponding to a X vector parallel to
1, divided by the modulus X : y, = Y . IIX. With respect to a general
Cartesian basis, by applying the usual formula of scalar product and taking
into account (9.A.5) one obtains:
With reference to the basis of eigenvectors of y, a simpler expression is

derived:
A geometrical interpretation of the quantity y, can be obtained by

considering the equation of the representation surface of y (9.A.1), and by
substituting xi = xli. The equation becomes:
where of course x represents the length of a radius vector joining the origin
to a general point on the quadric surface. Because of (9.A.8), it follows that
x = llvy,, i.e. the radius vector of the representation quadric has a length
equal to the inverse square root of the tensor component along the same
direction.
Further reading
Amelinckx, S., Gevers, R., and Van Landuyt, J. (ed.) (1978). Diffraction and
imaging techniques in material science. North-Holland, Amsterdam.
Amouric, M. (1987). Acta Crystallographica, B43, 57-63.

Amouric, M. and Baronnet, A. (1983). Physics and Chemistry of Minerals, 9,
146-59.
Authier, A. (1967). Advances in x-ray analysis, vol. 10, pp. 9-31. Plenum Press,
New York.
Bollmann, W. (1970). Crystal defects and crystalline interfaces. Springer, Berlin.
Catti, M. (1985). Acta Crystallographica, A41, 494-500.
Catti, M. (1989). Acta Crystallographica, A45, 20-5.
Catti, M., Ferraris, G., and Ivaldi, G. (1989). European Journal of Mineralogy, 1,
625-32.
Henderson, B. (1972). Defects in crystalline solids. Arnold, London.
Hull, D. (1968). Introduction to dislocations. Pergamon, Oxford.
Nye, F. (1985). Physical properties of crystals. Clarendon, Oxford.
Rosenberg, H. M. (1975). The solid state. Clarendon, Oxford.
Sands, D. E. (1982). Vectors and tensors in crystallography. Addison-Wesley,
Reading, MA.
Sirotin, Yu. I. and Shaskolskaya, M. P. (1982). Fundamentals of crystal physics.
Mir, Moscow.
Tilley, R. J. D. (1987). Defect crystal chemistry. Blackie, Glasgow.
Vainshtein, B. K., Fridkin, V. M., and Indenbom, V. L. (1982). Modern
crystallography, Vol. 11. Springer, Berlin.
Vainshtein, B. K. (ed.) (1988). Modern crystallography, Vol. IV. Springer, Berlin.
Zarka, A . , Lin, L., and Sauvage, M. (1983). Journal of Crystal Growth, 62,409-24.
Subject index
absolute configuration 97,374-5, 489 background 266,280,282

conventions of 591 balanced filters 242
determination of 545 barycentre principle 414
absolute structure 374 basis and cross magnitudes 343-6
absorption bayerite 440
coefficient 233, 241 benitoite 454
correction 304 Bent's rule 504
edge 241 benzophenone, (C6H5)&O 618
linear 241 Bessel functions 353, 387
mass 241, 299 best plane 74
acentric distribution 321-2 beta structures 576
alamosite 652 turns 577
albite 418 Bijvoet pairs 544
allostery 572 body-centred cubic (BCC) 432
amides, amidines 516, 528 bond angle calculation 120-4
amino acids 573 distances (table) 503
conventions 591 length calculation 120-4
formulas 592, 593 number conservation rule 514
amorphous solids, scattering from 201, 212 valence 436
anatase 436 Born-exponent (n) 411
angular overlap model (AOM) 500 Born-Haber cycle 413
angular probability functions 404 Bragg's law 154,243,284, 291
anilines 516 reflection intensity 163
anisotropic thermal motion 367, 368 branched fundamental anions (br f. a.) 424
anomalous branchedness (B) 448
alpha-anomalous synthesis 545 Bravais lattices 18-22
beta-anomalous synthesis 545 law 15
-difference Patterson synthesis 544 brightness 238
scattering 165, 320,374, 544, 545, 587 brilliance 238
anti-fluorite structure 439 Brookhaven protein data bank 535,585
apophyllite 455 brookite 434, 436
area detectors 281 brucite 439
Arndt-Wonacott rotation camera 261 Buerger precession camera, see camera, precession
Arrhenius law 638 Burgers circuit 629
asymmetric unit 24 Burgers vector 629-34
atom-atom potentials 474, 476 bunches 240
atomic scattering factor bunsenite 438
for electrons 195 bystromite 439
for neutrons 198
for X-rays 147
Aufbau principle 404 C(sp2)-N(sp3) fragment 516-20
646 1 Subject index
cadmium complexes 514 convolution 181

calcium carbonate (calcite) CaCO, 618 cooperativity 572
calcium sulphate (anhydrite) CaS0, 617, 618 coordination number (CN) 425,459
calcium sulphate dihydrate (gypsum) CaSO, . H,O, 618 coordination polyhedron 425
Cambridge Structural ata abase 380,467 coquimbite 463
camera correlation coefficient 370
Arndt-Wonacott rotation 261 correlation matrix 370
cylindrical film rotation 247 correlative methods 511
Debye-Scherrer 289 corundum 440
Gandolfi 293 cosets 48
Guinier 292 covalent bond 409
precession 254 cristobalite 456
Seeman-Bohlin 292 critical
Weissenberg 247 energy 236
carobbite 438 wavelength 236
Cartesian frames 600 cross-over 542
cassiterite 439 crystal field splitting (A) 414
central limit theorem 375-6, 385, 387 crystal field stabilization energy (CFSE) 415
centric distribution 321-2 crystal field theory (CFT) 414, 500
charge assisted hydrogen bond (CAHB) 471 crystal forces and molecular geometry 483
charge transfer 472 crystal packing 472
chemical bond 499-506 crystal size 304
chirality 488 crystal systems 17
chirality parameter 375 crystallization 536
chloromagnesite 439 crystallochemical formula 437
close-packed structure cubic closest packing (CCP) 429
cubic 625 Curie groups 603
hexagonal 626 cyclohexane 491
closest packing 429
Cochran distribution 340, 361, 387
cohesion energy (E) 410 data collection techniques
coincidence site lattice 81 for polycrystalline materials 287
collimators 244 for single crystals 245
colour symmetry 59 data reduction 301
combined figure of merit (CFOM) 340,361,387, 356-8, Debye formula 207
364-5 Debye-Scherrer camera 289
complement structure 388 decay, see radiation damage 308
completing the structure 365-6 defect
compliance coefficients 614 enthalpy 636, 637
Compton scattering 144, 185 entropy 636,637
computer graphics 562 extended defects 622
configuration 484, 486 free energy 637
conformation 484,490 line defects 622, 628
conformational interconversion 512, 516-20 planar defects 622,625-7
conformers 490 point defects 622,635-7
conjugate classes 48 defect structure 437
conjugate gradient 590 densitometers 271
consistent field 507 densitometry 268
constitution 484, 486 density of protein crystals 538
convergence map 363 diagonal plot 578
convergence procedure 352-4, 358,363 diastereoisomers 486
Subject index 1 647
dielectric constant 475 effective charge (ZeE)419

dielectric impermeability 607 electron affinity (EA) 406
dielectric permittivity 606 electron density
dielectric susceptibility 606 atomicity of 335-6
difference interpretation 562
Fourier 366-7,551,571 mapping 216
Patterson 541 positivity of 335, 339-40, 384
diffraction by a crystal 151, 161 electron density function 88-90, 131, 169-71
diffractions cones 287 electron microscopy 623,627, 628
diffractometer electron scattering 195
Bragg-Brentano 295 electronegativity (EN) 406
four circle 274 electro-optical effect 608
geometry 273 electrostatic bond strength (s,) 433
Seeman-Bohlin 296 electrostatic energy 475, 477, 509
single crystal 273 E-maps 356-68,360, 365
diffusion 637-8 emittance 238
diffusion coefficient 638 enamines 516
P-diketone enols 521-6 enantiomers 486
dimensionality (D) 450 enantiomorph, choice of 545
dipolar energy 477 enantiomorph, fixing 346-8,350-1
dipolar forces 469 energy
dipole moment 474 critical 236
Dirac delta function 173 dispersive methods 238
direct methods 321,335-65, 390-1, 560 enstatite 452, 454
directions, indexing of 7, 22 entalpy (AH3 413
Dirichlet region, see Wigner-Seitz cell entropy 479
dislocation 622, 628-34 equivalent reflections 157
climb 634 errors, treatment of
density 634 in protein crystallography 546
edge dislocation 629 etching 624
energy 632 Eulerian
motion 633 cradle 274
partial dislocation 634 rotation angles 72, 556
screw dislocation 629 Ewald sphere 154,246,287
disorder 373 exchange energy 474
dynamic 373 exchange forces 468
static 373 exposure 271
substitutional 373 extended Hiickel theory 500
disordered structures in proteins 583 extinction
dispersion energy 473 primary 164
dispersion forces 469 secondary 97, 164
displacement-shift-complete lattice 82
disulphide bridges 577
divergence map 363
DNA, hydrogen bonding in 529 fast Fourier transform 131
donor-acceptor interactions 472 fayalite 444
dynamic range 272,284 ferroelectricity 607
dynamic theory of diffraction 162, 191 festoons 252
Fibonacci series 225,389
Fick's law 637
eakerite 453 figures of merit 355-6, 360-5,548, 550
648 1 Subject index
film Hamilton test 103

exposure 271 Hamilton weighted residual, 371
speed 271 hard sphere model 500
filtering 551 Harker and Kasper inequalities 335
filters 241 Harker diagrams 544
first coordination shell 425 Harker lines 328,330
first order phase transition 448 Harker sections 328-33
fluorite 438 heat of formation (AHf) 413
flux spectral 235 heavy-atom
force field 481, 507, 509 (table) derivatives 536
Z,formulae 346 method 328-35,541
forsterite 444 parameters, refinement of 549
Fourier helices 574
best 548 Helmholtz free energy (F) 410
difference- 551, 571 heteropolar covalent bond 409
maps 88-90,131,169-71,216,562-71 hexacelsian 455
most probable 548 hexagonal closest packing (HCP) 429
transform 175 hexagonal indices 22
Fourier recycling 329, 365-6 higher invariants 338
frames 286 holography 395-7
Frank net 634 homometric sets 319-20
free energy 478 Hiickel theory 500
Frenkel defect 635-7 Hund first rule 416
Friedel's law 155, 544 hybrid branched complex anions (hB c.a.) 449
frontier orbitals theory 500 hybrid orbitals 502,504, 506
hybridization index 505
hydrogen bond 409,470-2,477,513,521-9
Gandolfi camera 293
gases scattering by 201, 208
generator (symmetry) 35, 46 ideal chemical formula 437
generators (of X-ray) ideal structures 437
conventional 229 ilmenite 441
rotating-anode 231 ilvaite 454, 460
sealed tube 231 image-seeking functions 380
geometrical isomers 488, 490 imaging plate 283
Gibbs free energy 410 impurities 635, 638-9
gibbsite 440 indexing
glaucochroite 444 precession photographs 254
glide planes 11, 30 rotation photographs 267
detection of 159 Weissenberg photographs 252
glycoproteins 582 initial structural model 320, 328
goniometer k (kappa) 275 insertion devices 238
goodness-of-fit measure 98, 370 integrated diffraction intensities 163
grain boundaries 635 intensity measurements 268, 271
group theory 45, 492 intensity statistics 321-4, 375-7
Guinier camera 292 intermolecular forces 468-78
internal energy (E) 410,479
interstitial 622, 627, 635, 636
haematite 440 intrinsic region 638, 639
half spot, see partially recorded reflections 262 inverse spinel 437, 442
halite 413, 422,438 inversion axes 5
Subject index ( 649
inversion parameter (i) 443 plane 74
ionic bond 409 refinement 94
ionic conductivity 639 restrained 107, 364,564,568
ionic crystals 409 weighting functions 369
ionic mobility 639 ligand field theory 500
ionic radius 418 lime 438
ionization energy (IE) 406 limiting sphere 154
isomerism 486-90 line group 30
isometric transformations 2, 35 linkedness (L), 448
isomorphism 540 liquids, scattering from 201, 212
ideal 541 lithium tantalate, LiTaO, 607
lack of 540,546 London forces 469
isomorphous loop branched anions (IB) 449
derivative 538 Lorentz correction 301
difference-Patterson 541 low temperature measurements 310
preparation 538 lunes 261
replacement 320, 540-87
isotropic thermal parameter 148, 368-9
madelung constant (A,) 412, 413, 443
magic integers 361,363, 388-90
Jahn-Teller effect 500 magnesioferrite 444
jakobsite 444 magnesium oxide (periclase), MgO 618
magnetite 437, 444
manganosite 438
k (alpha) lines 231,233,242 maximum entropy 560
k (beta) lines 231,233,242 maximum-filling principle 424
k parameter 238 maximum-multiplicity principle 416
Karle recycling 365 membrane proteins 536
keto-en01 tautomerism 486-521 merohedral twinning 85, 373
kirschsteinite 444 metallic bond 409
knebelite 444 metric matrix 61
krauskopfite 454 mica 612,627, 628
microcline 448
microdensitometers 271
lack of closure 546,549 Miller indices 8
Lagrangian strain tensors 610 minimum function 381, 383
Lang method 624 MIR technique 543
lattice energy (UL) 410,468,478,480-2 mirrors 245
lattice frames 600 mixed-branched complex anions (olB c. a.) , 4 9
lattice function 174 model building 562
lattices 6, 18-22 modulated structures 191, 221
Laue classes 17 modulus difference squared synthesis 541
determination of 156 molecular crystals 409, 466
Laue condition 151 molecular dynamics
method 246 in refinement 117,569
lawrencite 439 molecular hermeneutics 511
layer groups 57 molecular mechanics 506-1 1
layer line screen 249,256 molecular replacement 384, 552
least-squares molecular structure 466
constrained 92, 104,564,568 monochromatization of X-rays 241
method 9O-lO9,367-74 monochromators, crystals 243
650 ( Subject index
monopolar energy 477 origin fixing 329, 334, 346-51,358, 362-3, 379-80, 383
monopolar forces 469 orthoclase 456
monticellite 444 oscillation method 259
mosaic 162 out-of-plane deformation 508, 517
Moseley's law 231
mother liquid 536
movements 2,36-41 packing coefficient (c,) 429,456
multiple diffraction 191, 351 packing, in protein crystals 548
multiplicity 288 pairing energy 416
multiplicity (M) 449 parameter-shift method 392
multisolution methods 352, 360-5, 389, 390-3 partially recorded reflections 262
multiwire proportional counter 282 partial structure recycling 365
muscovite 455, 612, 627, 628 Patterson
atomic superposition 381
function 182,320,324-35,377-84,541
naphthalene C,,H, 618 group 328
naphthylamines 516 sharpening 326
narsarsukite 455 superposition methods 335, 377-84
negative quartets 343, 356, 362 vector methods 335, 377-84
negative triplets 356, 362 vector multiplicity 325
neighbourhood principle 343 Pauli principle 404
Nernst-Einstein equation 639 Pauling's bond number 503, 513, 516
neutron scattering 198 Pauling's rules 433
Newmann's principle 14, 603, 616 first 433
nibrucite 439 second 433,443,447,456
Niggli reduced cell 77 third 435, 454
non-bonded energy 469,474,508 fourth 436
non-linear optics 608 fifth 436, 455
non-molecular crystals 409 PDB 535, 585
normal spinel 437,442,443 PEG 537,538
normalization 351, 361 periclase 438, 460
normalized structure factors 321-4,326, 340, 351, permissible origins 338, 347-50
356-7,361 periodic table of the elements 407
nucleophilic addition 514, 515 periodicity of the chain or ring (P or Pr)450
number of anion types (N,,) 448 perovskite 441,445, 461
number of molecules in unit cell 312 phase
combined 562, 571
permutation 352,360-1,362, 364
problem 169, 319,540
occupancy 437 refinement 365-75,549
octahedral site stabilization energy (OSSE) 417 phasing magnitudes 343-6
olivine 418, 428, 444, 454 phasing power 550
omega phasing shell
scan 278 first 344-6
2-theta scan 278 second 345-6
omit maps 572 phenakite 428
open-branched anions ( o B ) 449 phlogopite 446
optical density 269 photographs
optical indicatrix 15, 608 oscillation 267
optical isomers 488 precession 254
orientation matrix 276 rotation 247
Subject index ( 651
upper level 257,252 Q translation function 588

Weissenberg 248 quadratic forms 75
zero level 248, 256 qualitative molecular orbital theory (QMOT) 500
piezoelectricity 16, 619-22 quantum-chemical methods 499
pixels 269,286 quantum numbers 403
plane groups 30 quartet invariants 338, 343-5,346, 356, 362, 392
planes, indexing of 8, 22 quasicrystals 223
point groups 11, 16 quaternary structure, in proteins 578
polar angles 73,556
polarizability 474 R factor 312, 320, 368-71,369,572
polarization 423, 443, 605 Cullis 550
correction 303 Kraut 550
factor 143 modulus 550
of synchrotron radiation 240 racemates 489
polycrystalline materials 287 radial density distribution 180
polymorphism 482 radial probability function 404
polypeptide chain 573, 577 radiation damage corrections 308
portlandite 439 radius of gyration 214
position sensitive detectors 282 radius ratio rule 425
curved 297 Ramachandran plot 577
potential energy hypersurface 481, 485, 510,512, 519, random phases 391-3
525 reaction pathways 512-16,519, 526
powder diffraction 109,287 real-space filtering 551
powder diffraction file 297 reciprocal lattice 63
precession camera 254 reduction of data, see data reduction 301
precipitants 537, 538 refinement 94,104,107,365-75
primitive cell 6 constrained 105, 106,564,568
primitivity condition 348 by molecular dynamics 569
probability distributions 321-2, 340-6, 375-6, 385-7, in protein crystallography 562
547,560,571 in real space 564
product function 380-1 restrained 107, 564,568
profile fitting 112,273,280,286 rigid-body 105, 569
proportional counter strategy 570
multiwire 282 reflecting sphere, see Ewald sphere
protein reflections, partially recorded 262
classification 585 refraction indices 608
crystallization 536 relative scaling 310
crystals 536 reliability index, see R index 312
domains 578,585 Renninger effect 191
globular 535 representation
hydrogen bonding in 529 first 344-6, 362
structure 572 second 36-61, 361
subunits 578 representation theory 342-6, 361
pseudorotation 491 repulsion energy 474
pseudotranslational symmetry 351,371, 387-8 repulsion forces 468
puckering coordinates 495, 498 resonance 503
pyrochromite 439 resonance assisted H-bonding (RAHB) 472, 521-9
pyroelectricity 11, 606, 607 restraints
pyrolusite 439 geometrical 107,565
pyrope 462 on phases 567
pyroxmangite 454 with potential energy 568
652 1 Subject index
rhodonite 454 self

Rietveld refinement 109-17 -rotation 556, 558
ring conformations 490-9 -translation 558
rod groups 56 sellaite 439
Ross filters see balanced filters 242 semi-invariant (s. s .)
rotamers 490 modulus 339
rotating-anode generators 231 one-phase 339,345-6,352-3,355-6,361,363
rotation 2, 36, 69 two-phase 346, 350, 355-6, 361-3
rotation camera 247,261 vector 339
rotation functions 383,552, 553,558 series truncation error 170, 366
rotation matrix 555 shattuckite 452
rotation method sillimanite 455
with area detectors 286 silver halide 636
cylindrical film 247 Sim weights 366,393-4
geometry 261 simulated annealing 570
for macromolecular crystallography 259 single crystals
roto-inversion 5, 35 data collection methods for 245
roto-reflection 5, 35 SIR technique 542
rutile 434, 436,438, 460 slip plane 628, 629, 634
small angle scattering 213
soaking 539,540
salting-in 537 sodalite 433
salting-out 537 sodium chloride, NaCl605,636, 638
sanidine 448 solvent
Sayre equation 336,340, 355 calculation of 538
scacchite 439 in protein crystals 536, 538
scattering structure of 583
by amorphous solids 201,212 space groups 22
by gases 201,208 determination of 161
by liquids 201,212 spectral
electron 195 brightness 238
neutron 198 flux 235
X-ray 146 spectrochemical series 415
scattering factor sphere of reflection 245, 287
for atomic electrons 146 spherical polar angles 73, 556
for atoms 147 spinel(s) 428, 437,442, 462
for molecules 150 stacking fault 622,625-8
scaling 310 deformation 626
scanner, see densitometer 271 extrinsic 627
Schottky defect 635-7,639 grow 627
screening effect (S) 419 twin boundary 627
screw axes standard deviation 98, 122,271,273, 280
definition of 10 starting set of phases 352-4, 358, 360-3
detection of 159 stationary picture method 286
sealed tube generators 231 steacyte 454
second coordination shell 425 stereochemistry 484-99
second-order phase transition 448 stereographic projection 40-1
secondary structures, in proteins 574 stiffness coefficients 614
Seeman-Bohlin stishovite 439
camera 292 stokesite 454
diffractometer 296 storage ring 235
Subject index 1 653
strain 609-11 tautomerism 486

stress 613 television detectors 283
stretching-bending energy 507 temperature factor 149, 186-90
stretching factor (f,) 452 tensors 601,602,640-2
structural chemistry 468 eigenvalues 640, 641
structural crystallography 466 eigenvectors 640, 641
structure correlation method 512 representation surfaces 641, 642
structure factor 87, 152 tephroite 444
structure invariants (s.i.) 337-9 terminated anions ( t ) 450
structures tertiary structure, in proteins 578
&Bod 428 TH translation function 589
A,B,X, 441 thenardite 428
MX 437 thermal expansion 611
MX2 438 thermal motion 76, 148, 583
MX3 440 analysis of 117
M2X 438 thermodynamics of molecular crystals 478-80
M2X3440 thiathiophthenes 513
subgroups 47,49,51 Thomson scattering 142
sublattices 80 three-centre-four-electron systems 512-14
sublimation enthalpy 579 time resolved experiments 241,247
substructure 388 tin complexes 515
substructure reflections 388 torsion angles 73
sum function 381, 383 torsion energy 508
supergroups 51 tourmaline 454
superlattices 80 transformations
superstructure reflections 388 basis 65, 128
sylvine 438 isometric 35
symbolic addition methods 352, 358-60 orthonormalizing 68
symmetry translation function 552, 556, 588
axes 3, 10, 32 transmission factor 305, 306
classes 11, 16 tremolite 455
colour 57 trevorite 444
generalization of 55 trial and error methods 320
generators 46 tridimite 456
groups 45 triiodide ion 513
planes 5,32 triplet invariants 338, 340-6, 350-3, 355-6, 362, 385-7,
symmorphic space groups 25 390-3
synchrotron radiation 234 triplets
&-synthesis 394-6 setting-up of 351-2
P-synthesis 394-5 twins 83, 133
2F,-F, synthesis 395 twinning 373
zl-synthesis 395
systematic absences 159
systems 17 u parameter 443
unbranched fundamental anions ( u B f.a.) 449
undulators 238
T translation function 557, 589 unit cell 7
tangent formula 342,363-4, 392 unit cell parameters
tangent refinement 363-4 from the orientation matrix 276
tapiolite 439 from a precession photograph 259
taramellite 454 from a rotation photograph 248
654 1 Subject index
unitary structure factor 321, 347 Wilson plot 322-4

wollastonite 454
Wooster effect 270
vacancies 622, 627,635, 636 wurtzite 422
valence bond (VB) 499,502-6 wiistite 438
valence shell electron pair repulsion (VSEPR) 499, 501 Wyckoff positions 27
van der Waals
forces 409
radius 466,574
variance-convariance matrix 91, 95, 105, 107, 370 X-ray
vibration energy (E,) 410 absorption 241
villiaumite 438 characteristic spectra 230
VM 438 collimators 244
Voigt's notation 615 filters 241
Von Mises concentration parameter 341,346, 362 generators 231
distribution 341, 346, 393 mirrors 245
monochromators 243
position sensitive detectors 282
wavelength rotating anode generators 231
characteristic 230 sources 229
critical 236 tubes 231
minimum 230 X-ray topography 624,630-2
weak reflections 371 xonotlite 455
weighted tangent formula 363, 365
Weissenberg camera 247
white radiation 230 zero-point energy 479
wigglers 238 zircon 461
Wigner-Seitz cell 41 zunyite 453

Cristallografia PDF

Uploaded by

Copyright:

Available Formats

Cristallografia PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cristallografia PDF

Uploaded by

Copyright:

Available Formats

Fundamentals of

C. GIACOVAZZO, H. L. MONACO, D. VITERBO

INTERNATIONAL UNION OF CRYSTALLOGRAPHY

Crystallography, the science concerned with the study of crystals, is a very

Bari, Italy C.G.

We have the pleasant task of expressing our gratitude to Dr C. K. Prout and

Least-squares refinement of crystal Introduction

Monochromatization, collimation, and Phase determination procedures

Closest packings Ring conformations

7 Molecules and molecular crystals 465

The refinement of heavy-atom References and further reading

Lucretius, De Rerum Natura, IV, 386-7.

Carmelo Giacovazzo, Dipartimento Geomineralogico, Universith di Bari,

The crystalline state and isometric operations

basic crystal periodicity is modulated by periodic distortions incom-

(1) a translation, when all points of the object undergo an equal

(3) a rototranslation or screw movement, which may be considered as the

Axes of rotational symmetry

Fig. 1.1. Arrangements of symmetry-equivalent

Axes of rototranslation or screw axes

(2) the 2 axis is equivalent to a reflection plane perpendicular to it; the

(3) the 3 axis is equivalent to the product of a threefold rotation by an

(4) the 4 axis is also a 2 axis;

Reflection planes with translational component (glide

The rational properties of lattices

When the cell is not primitive u, v, w, and n will be rational numbers.

x'lpa + y'lqb + z'lrc = 1.

Multiplying both sides by m we obtain

The intercepts of the plane (1.5) on X, Y , Z will be l l h , l l k and 111

Symmetry restrictions due to the lattice

Let us now consider the restrictions imposed by the periodic nature of

For instance, for a screw axis of order 4 the allowed translational

We will now consider the restrictions imposed by the periodicity on the

Point groups a n d s y m m e t r y classes

Table 1.2. Single-axis crystallographic point groups

Proper axis Improper axis Proper and improper

Combination of cu (ded B (ded Y (ded

repeat in R the object in Q. P and Q are therefore directly congruent and

Fig. 1.10. Arrangement of proper symmetry

PPP PI1 I P I IIP

results so far described can be easily derived by recalling that:

independently of the morphological aspect of the samples. Thus, each

Point groups in one and two dimensions

The Laue classes

The seven crystal systems

(b) m ; 2 m m (c) m;2mm

Table 1.7. The five plane lattices

Cell Type of cell Point group Lattice parameters

Symbol Type Positions of Number

Fig. 1.16. Monoclinic lattices: (a) reduction of a Cubic lattices

It is worth noting that a rhombohedral description of a hexagonal P lattice

Fig. 1.17. Rhombohedra1 lattice. The basis of the

ck can be obtained from a H , bH, CH by choosing:

The space groups

Fig. 1.19. The 14 three-dimensional Bravais

Crystal Point Space

Monoclinic P2, P2,, C2

Tetragonal p4, p41, p4,, p4, 14, 14,

Cubic (231 P23, F23, 123, P2,3,_12,3 -