Algorithmic Geometry

J-D.
Boissonnat
M. Yvinec
Algorithmic
Geometry
2LVI~tl I
A1
A / ;4
/1A, " ~
Zj
J?'
fi <I
/.
Algorithmic geometry
Algorithmic Geometry
Jean-Daniel Boissonnat
Mariette Yvinec
INRIA Sophia-Antipolis, France
Translated by Herv6 Bronnimann

INRIA Sophia-Antipolis, France
.wl CAMBRIDGE
v UNIVERSITY PRESS
PUBLISHED BY THE PRESS SYNDICATE OF THE UNIVERSITY OF CAMBRIDGE
The Pitt Building, Trumpington Street, Cambridge CB2 1RP, United Kingdom
CAMBRIDGE UNIVERSITY PRESS
The Edinburgh Building, Cambridge CB2 2RU, United Kingdom
40 West 20th Street, New York, NY 10011-4211, USA
10 Stamford Road, Oakleigh, Melbourne 3166, Australia
First published in French by Edisciences, 1995 as Geom6trie Algorithmique

( 1995 Edisciences International, Paris
English edition 0 Cambridge University Press 1998
This book is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without
the written permission of Cambridge University Press
First published in English 1998
Printed in the United Kingdom at the University Press, Cambridge
A catalogue record for this book is available from the British Library
ISBN 0 521 56322 4 hardback

ISBN 0 521 56529 4 paperback
4.i*
To Bertrand,
Martine,
Cecile,
Clement,
Alexis,
Marion,
Quentin,
Romain,
Eve,
and the others...
Table of contents
Preface xv
Translator's preface xix
Acknowledgments xxi
Part I - Algorithmic tools 1
Chapter 1 - Notions of complexity 3

1.1 The complexity of algorithms ................................. 3
1.1.1 The model of computation ............................. 3
1.1.2 Notions of complexity ................................. 4
1.1.3 Asymptotic behavior, notation .......................... 7
1.2 Optimality, lower bounds .................................. 9
1.2.1 The complexity of a problem ........................... 9
1.2.2 The example of sorting: decision trees ................... 10
1.2.3 Lower bounds by transforming one problem into another ... 11
1.3 Bibliographical notes ........................................ 12
Chapter 2 - Basic data structures 13

2.1 Terminology and features of the basic data structures ............ 14
2.1.1 Lists, heaps, and queues ............................... 14
2.1.2 Dictionaries and priority queues ........................ 15
2.2 Balanced search trees ........................................ 16
2.2.1 Graphs, trees, balanced trees ........................... 16
2.2.2 Red-black trees ....................................... 17
2.3 Dictionary on a finite universe ................................ 25
2.4 Exercises ............................................. 26
Chapter 3 - Deterministic methods used in geometry 32

3.1 The divide-and-conquer method ............................... 33
3.1.1 Overview ............................................ 33
viii Table of contents
3.1.2 An example: sorting n numbers using merge-sort ......... 35

3.2 The sweep method .......................................... 36
3.2.1 Overview ............................................ 36
3.2.2 An example: computing the intersections of line segments .. 37
3.3 Vertical decompositions ...................................... 40
3.3.1 Vertical decompositions of line segments ................. 40
3.3.2 Vertical decompositions and simplified decompositions ..... 42
3.4 Exercises ............................................. 43
Chapter 4 - Random sampling 46

4.1 Definitions ................................................. 46
4.1.1 Objects, regions, and conflicts .......................... 46
4.1.2 Random sampling ..................................... 49
4.2 Probabilistic theorems ....................................... 50
4.2.1 The sampling theorem ................................. 51
4.2.2 The moment theorem .............................. 55
4.3 Exercises ............................................. 57
Chapter 5 - Randomized algorithms 63

5.1 The randomized incremental method ........................ 64
5.2 Off-line algorithms .......................................... 65
5.2.1 The conflict graph .................................... 65
5.2.2 An example: vertical decomposition of line segments ....... 69
5.3 On-line algorithms...................................... 75
5.3.1 The influence graph ................................... 75
5.3.2 An example: vertical decomposition of line segments ....... 81
5.4 Accelerated incremental algorithms ............................ 84
5.4.1 The general method ................................... 84
5.4.2 An example: vertical decomposition of a polygon ......... 87
5.5 Exercises ............................................. 88
Chapter 6 - Dynamic randomized algorithms 95

6.1 The probabilistic model ...................................... 96
6.2 The augmented influence graph ............................... 97
6.3 Randomized analysis of dynamic algorithms .................... 101
6.4 Dynamic decomposition of a set of line segments ................ 110
6.5 Exercises ................................................... 121
6.6 Bibliographical notes .................................... 123
Table of contents ix
Table of contents ix
Part II - Convex hulls 125
Chapter 7 - Polytopes 127

7.1 Definitions ...................................... 127
7.1.1 Convex hulls, polytopes ..................... 127
7.1.2 Faces of a polytope ........................ 128
7.1.3 Polarity, dual of a polytope ............... 135
7.1.4 Simple and simplicial polytopes.............. 140
7.2 The combinatorics of polytopes .................... 141
7.2.1 Euler's relation ............................ 141
7.2.2 The Dehn-Sommerville relations ........... 144
7.2.3 The upper bound theorem ................ 145
7.2.4 Cyclic polytopes ........................... 147
7.3 Projective polytopes, unbounded polytopes .......... 148
7.3.1 Projective spaces .......................... 148
7.3.2 Oriented projective spaces ................ 153
7.3.3 Projective polytopes, unbounded polytopes.... 156
7.4 Exercises ...................................... 162
7.5 Bibliographical notes ............................. 168
Chapter 8 -Incremental convex hulls 169

8.1 Representation of polytopes ........... ................. 170
8.2 Lower bounds ........................ ................. 171
8.3 Geometric preliminaries ............... ................. 171
8.4 A deterministic algorithm ............. ................. 176
8.5 On-line convex hulls .................. ................. 180
8.6 Dynamic convex hulls ................. ................. 186
8.7 Exercises ........................... 195
8.8 Bibliographical notes ................. 197
Chapter 9 - Convex hulls in two and three dimensions 198

9.1 Representation of 2- and 3-polytopes .................... 199
9.2 Divide-and-conquer convex hulls in dimension 2 ........... 201
9.3 Divide-and-conquer convex hulls in dimension 3 ........... 205
9.4 Convex hull of a polygonal line.......................... 214
9.5 Exercises ........................................... 219
9.6 Bibliographical notes .................................. 221
Chapter 10 - Linear programming 223

10.1 Definitions ............................... ..................
10.2 Randomized linear programming ............ 225
10.3 Convex hulls using a shelling................ ..................
10.4 Exercises ................................ ..................
10.5 Bibliographical notes ................... 239
x Table of contents
Part III - Triangulations 241
Chapter 11 - Complexes and triangulations 243

11.1 Definitions ........................................... 243
11.1.1 Simplices, complexes .............................. 243
11.1.2 Topological balls and spheres, singularities ............... 245
11.1.3 Triangulations ................................... 247
11.1.4 Polygons and polyhedra ............................ 248
11.2 Combinatorics of triangulations ........................... 250
11.2.1 Euler's relation for topological balls and spheres .......... 250
11.2.2 The complexity of 2-complexes ......................... 251
11.2.3 The complexity of 3-triangulations ...................... 255
11.3 Representation of complexes, duality ........................... 258
11.4 Exercises ............................................. 260
Chapter 12 - Triangulations in dimension 2 263

12.1 Triangulation of a set of points ............................ 264
12.1.1 The complexity of computing a triangulation ............. 264
12.1.2 An incremental algorithm .............................. 264
12.2 Constrained triangulations ............................... 266
12.3 Vertical decompositions and triangulations of a polygon .......... 267
12.3.1 Lower bound .................................... 267
12.3.2 Triangulating monotone polygons .................... 269
12.3.3 Vertical decomposition and triangulation of a polygon ..... 274
12.4 Exercises ............................................. 281
12.5 Bibliographical notes ................................... 287
Chapter 13 - Triangulations in dimension 3 289

13.1 Triangulation of a set of points ................................ 290
13.1.1 The size of a triangulation .......................... 290
13.1.2 The split theorem ................................ 293
13.1.3 An incremental algorithm .......................... 296
13.2 Constrained triangulations............................... 300
13.3 Vertical and simplicial decompositions ...................... 302
13.3.1 Vertical decomposition of a polyhedral region ............. 302
13.3.2 Simplicial decomposition of a polyhedron of genus 0 ....... 307
13.4 Exercises ............................................. 317
13.5 Bibliographical notes .................................... 318
Part IV - Arrangements 319
Chapter 14 - Arrangements of hyperplanes 321

14.1 Definitions ........................................... 321
Table of contents xi
Table of contents xi
14.2 Combinatorial properties ................................... .. 322

14.3 The zone theorem ......................................... .. 325
14.4 Incremental construction of an arrangement .................. .. 330
14.4.1 The case of dimension 2 ............................. .. 330
14.4.2 The case of dimensions higher than 2 .................. .. 331
14.5 Levels in hyperplane arrangements .......................... .. 333
14.5.1 Definitions ......................................... .. 333
14.5.2 Combinatorial properties of levels ................... .. 335
14.5.3 Computing the first k levels in an arrangement ......... .. 335
14.6 Exercises ............................................... .. 343
14.7 Bibliographical notes ...................................... .. 350
Chapter 15 - Arrangements of line segments in the plane 352

15.1 Faces in an arrangement .................... .. 353
15.2 Davenport-Schinzel sequences ................... .. 353
...........
15.3 The lower envelope of a set of functions .......... ........... .. 355
15.3.1 Complexity ............................ ........... .. 356
15.3.2 Computing the lower envelope ............ ........... .. 358
15.4 A cell in an arrangement of line segments....... ........... .. 358
15.4.1 Complexity ............................ ........... .. 359
15.4.2 Computing a cell .................... ........... .. 362
15.5 Exercises .................................... ........... .. 368
15.6 Bibliographical notes .......................... ........... .. 371
Chapter 16 - Arrangements of triangles 373

16.1 Faces in an arrangement ................. ... 374
16.2 Decomposing an arrangement of triangles ... 374
16.2.1 Vertical decomposition ............. ... 375
16.2.2 Convex decomposition ............. ... 377
16.3 The lower envelope of a set of triangles..... ... 379
16.3.1 Complexity ................... ... 381
16.3.3 Computing the lower envelope ...... ... 384
16.4 A cell in an arrangement of triangles....... ... 390
16.4.1 Complexity ................... ... 390
16.4.3 Computing a cell ................. ........... ... 400
16.5 Exercises .............................. ........... ... 402
16.6 Bibliographical notes .................... ........... ... 404
Part V - Voronoi diagrams 405
Chapter 17 - Euclidean metric 407

17.1 Definition ........................... ................. 407
xii Table of contents
17.2 Voronoi diagrams and polytopes ........................... 408

17.2.1 Power of a point with respect to a sphere .............. 408
17.2.2 Representation of spheres .......................... 409
17.2.3 The paraboloid P ................................ 410
17.2.4 Polarity ........................................ 411
17.2.5 Orthogonal spheres ............................... 412
17.2.6 Radical hyperplane ............................... 414
17.2.7 Voronoi diagrams ................................. 414
17.3 Delaunay complexes ......................................... 416
17.3.1 Definition and connection with Voronoi diagrams ......... 416
17.3.2 Delaunay triangulations ............................ 418
17.3.3 Characteristic properties ........................... 418
17.3.4 Optimality of Delaunay triangulations ................... 421
17.4 Higher-order Voronoi diagrams ................................ 425
17.5 Exercises ............................................. 428
Chapter 18 - Non-Euclidean metrics 433

18.1 Power diagrams ....................................... 434
18.1.1 Definition and computation ......................... 434
18.1.2 Higher-order power diagrams ........................ 436
18.2 Affine diagrams ....................................... 437
18.2.1 Affine diagrams and power diagrams .................... 437
18.2.2 Diagrams for a general quadratic distance ................ 438
18.3 Weighted diagrams ..................................... 439
18.3.1 Weighted diagrams with additive weights ................ 439
18.3.2 Weighted diagrams with multiplicative weights ........... 442
18.4 L1 and Loometrics ..................................... 445
18.5 Voronoi diagrams in hyperbolic spaces ...................... 449
18.5.1 Pencils of spheres ..................................... 449
18.5.2 Voronoi diagrams in hyperbolic spaces ................. 450
18.6 Exercises ............................................. 454
Chapter 19 - Diagrams in the plane 459

19.1 A sweep algorithm ..................................... 459
19.2 Voronoi diagram of a set of line segments .................... 464
19.2.1 Definition and basic properties ...................... 464
19.2.2 A sweep algorithm ................................ 470
19.2.3 An incremental algorithm .......................... 471
19.2.4 The case of connected segments ...................... 476
19.2.5 Application to the motion planning of a disk ............ 479
19.3 The case of points distributed in two planes .................... 481
19.3.1 The two planes are parallel ......................... 481
Table of contents xiii
19.3.2 The two planes are not parallel ...................... 485

19.4 Exercises ............................................ 488
References 492
Notation 508
Index 513
Preface
A new field
Many disciplines require a knowledge of how to efficiently deal with and build
geometric objects. Among many examples, one could quote robotics, computer
vision, computer graphics, medical imaging, virtual reality, or computer aided
design. The first geometric results with a constructive flavor date back to Euclid
and remarkable developments occurred during the nineteenth century. However,
only very recently did the design and analysis of geometric algorithms find a
systematic treatment: this is the topic of computational geometry which as a field
truly emerged in the mid 1970s. Since then, the field has undergone considerable
growth, and is now a full-fledged scientific discipline, of which this text presents
the foundations.
Contents and layout of this book
The design of efficient geometric algorithms and their analysis are largely based on
geometric structures, algorithmic data structuring techniques, and combinatorial
results.
A major contribution of computational geometry is to exemplify the central
role played by a small number of fundamental geometric structures and their
relation to many geometric problems.
Geometric data structures and their systematic analysis guided the layout of
this text. We have dedicated a part to each of the fundamental geometric struc-
tures: convex hulls, triangulations, arrangements, and Voronoi diagrams.
In order to control the complexity of an algorithm, one must know the com-
plexity of the objects that it generates. For example, it is essential to have a
sharp bound on the number of facets of a polytope as a function of the number
of its vertices: this is the celebrated upper-bound theorem proved by McMullen
in 1970. Combinatorial geometry plays an essential role in this book and the
first chapters of each part lay the mathematical grounds and prove the basic
combinatorial properties satisfied by the corresponding geometric structures.
xvi Preface
At the same time as geometric data structures of general interest were be-
ing studied, new algorithmic techniques were devised. To general algorithmic
paradigms, computational geometry added its own geometric techniques. The
first purely geometric paradigm in the history of the field, the sweep method,
was originally used by Bentley and Ottmann in an algorithm that computes the
intersection of a set of line segments in the plane. Subsequent developments of
general techniques soon encountered important theoretical difficulties which led
to quite sophisticated variants and theoretical constructions without truly affect-
ing the practice of the field. As a reaction against this tendency, a few authors
decided it was more desirable to look for simple algorithms which were efficient
on the average, rather than algorithms whose good behavior in the worst case
did not guarantee good behavior in practical instances of the problem.
The recent body of work on randomization gave the most significant answer
in this direction. An algorithm is said to be randomized if, after making ran-
dom choices during its execution, it gives the solution to a purely deterministic
problem. No probabilistic assumptions are made about the input objects, and
randomness is used only to choose the path that the algorithm will follow to the
solution. Randomized algorithms are often simple to conceive and to program,
and their average complexity (over all the random choices made during the exe-
cution) is usually very good, often even optimal. Randomization leads to general
methods for the design and analysis of algorithms, and allows efficient compu-
tation of geometric structures, both in theory and practice. For these reasons,
randomization holds a central position in this book. The first three chapters in the
first part contain all the generic material related to randomization, and instances
of randomized algorithms are presented throughout the subsequent chapters.
Goals and limits of this book
The goal of this book is twofold. In the first place, it aims at giving a coherent
exposition of the field rather than a collection of results, and at presenting only
methods that possess a certain degree of generality. The algorithms presented in
this book have been selected to work in all dimensions whenever this was possible:
the case of dimension 2 only receives special treatment when particular methods
lead to significant improvement, which happens surprisingly seldom.
In the second place, this book aims at presenting solutions which, while theoret-
ically efficient and relevant, remain relatively simple and applicable in practical
situations. Most of these algorithms have been implemented by their authors and
their practical behavior has turned out to agree with the analyses developed in
this book.
Nevertheless, this book does not claim to be a comprehensive treatment of the
whole field of computational geometry. In particular, the reader will find mention
Preface xvii
of geometric data structures for queries and multidimensional searching only in

the exercises. Also, this book is mostly concerned with Euclidean geometry, so
the reader should not expect a full treatment of problems dealing with curves
and algebraic surfaces. However, possible extensions to curves and surfaces are
systematically pointed out, and some exercises as well as the bibliographical notes
indicate how to extend the results in these directions.
From algorithms to programs

Rather than focusing on details of the implementation, this book emphasizes the
principles underlying the algorithms. There is thus an important step to be taken
from the descriptions given here to the actual implementation.
Unfortunately, however, there is currently no satisfactory treatment of problems
related to the implementation of geometric algorithms. This topic is gaining more
momentum and other books will undoubtedly fill in this gap in the near future.
The first problem raised by the implementation of geometric algorithms has
to do with the finite precision used by the computer. More often than not, and
as is done in this book, algorithms are designed and analyzed in a model of
computation where computers can deal with arbitrarily long real numbers, and
all operations give an exact result. Of course, this is not the case in practice, and
a naive implementation that uses a standard floating point representation of real
numbers can lead to fatal errors during the execution.
The other major problem has to do with so-called degenerate cases (points on
a single line, for instance). In this book, the objects are usually assumed to be in
general position, which excludes degenerate cases. This allows a more synthetic
treatment and focuses better on the underlying principles, rather than on the
particular details allowing correct performance in degenerate situations. Clarity
is thus preferred to a comprehensive treatment which would require much longer
discussions.
How to read this book
This book assumes no particular knowledge from the reader and should be ac-
cessible to any enthusiastic geometer. Its contents have been taught in several
graduate courses both in mathematics and in computer science. It is aimed both
at mathematicians interested in a constructive approach to geometry, and at
computer scientists in need of an accurate treatment of computational geometry.
Students, researchers, and engineers in more practical fields will find here a useful
methodology and practical algorithms.
There is more than one way to read this book. The authors have tried to respect
xviii Preface
the unalienable rights of the reader, as defined by Daniel Pennac. 1 In particular,

one may read parts, jump ahead, or search here and there for the needed piece
of information. Chapters 1 and 2 are mere reviews that the informed reader
may skip. Chapters 4, 5, and 6 introduce the formal framework for randomized
algorithms. They hold a central position in this book. Nevertheless, the reader
who is mostly interested in the applications may skip them in a first reading and
just get familiar with the results. Or this reader may read chapter 5 to get a
further glimpse of the algorithms without reading the full text. The subsequent
parts are dedicated to, respectively, convex hulls, triangulations, arrangements,
and Voronoi diagrams. They are essentially independent. As a sampler for an
introductory course, the teacher may choose to present chapters 8 or 9, 12, then
the section in chapter 14 about dimension 2, then chapters 15 and 17, and chapter
19 if time permits.
Exercises are included to complement the text, rather than to solidify the
understanding. They offer extensions as well as applications of the results. The
solutions to the most difficult ones are sketched in the hints.
Bibliographical notes at the end of each chapter give pointers to the literature
available on the topics of the chapters as well as on the exercises. These references
are by no means comprehensive and only introductory references are given for
related topics.
1In his essay Comme un Roman, the French writer Daniel Pennac describes the unalienable
rights of the reader as:
1. The right not to read.
2. The right to jump ahead.
3. The right not to finish a book.
4. The right to read again.
5. The right to read anything.
6. The right to Bovarysm (textually transmissible disease).
7. The right to read here and there.
8. The right to thesaurize.
9. The right to read aloud.
10. The right not to say anything.
Translator's Preface
The original text was written in French. The translator's task was constrained by
the fact that most of the French words used in the original text were originally
coined by their authors in English publications, or have a commonly accepted
translation into English. The problem was thus one of reverse engineering! For-
tunately, there are now many textbooks in computational geometry which helped
to resolve conflicts in terminology. Whenever possible, the translation conformed
to the standard terminology or, for the more specialized vocabulary, to the ter-
minology set up in the original papers.
For graphs, however, the use of the word edge overlapped with that of 1-faces
for common geometric structures. Similarly, the word vertex is also used for
polytopes in a different meaning than for graphs. The situation is somewhat
complicated by the fact that sometimes graphs are introduced whose nodes are
edges of a polygon. We have followed the French text in systematically using
the words node and arc for the set underlying a graph and the symbolic links
between the elements of this set. The terminology related to graphs is recalled
in subsection 2.2.1.
We have departed from the French text for the word saillant (meaning salient)
to follow the usage with convex vertices/ edges, as opposed to reflex. Although a
vertex or an edge is always convex in the original meaning of convexity, here it
means (as most people would understand it) that the internal angle around the
vertex or around the edge is smaller than 7r. Luckily, this definition is never used
for higher-dimensional faces, and therefore should not create confusion.
Vertical decompositions as they are introduced in this book have also been
called by various names, such as trapezoidal maps, vertical partitions, and verti-
cal visibility maps. As with other authors, we have preferred the phrase vertical
decomposition or even decomposition for short, in order to emphasize the rela-
tion with other geometric decomposition schemes, for example decompositions of
arrangements, polygons, or polyhedra into simplices (also called triangulations).
We should properly speak of the decomposition of the plane induced by a set of
segments. The reader will forgive us for using the phrase decomposition of (a set
of) line segments.
xx Translator'spreface
A translation is an excellent opportunity to enhance the text with added refer-

ences, a broader index, more exercises, more explanatory figures, and sometimes
more concise proofs. Examples of these are found everywhere in this book, espe-
cially in the exercises about data structures (exercise 2.6) or in the proof of the
upper-bound theorem (see also exercises 7.8 and 7.10). The translator wishes to
thank the authors for their guidance, their willingness to answer his questions,
and for bringing him back to orthodoxy when his mood was getting whimsical.
Herve Br6nnimann.
Acknowledgments
This book benefited from the joint work of researchers of the PRISME project at
INRIA and is inspired by much common work with Panagiotis Alevizos, Andre
CGrezo, Olivier Devillers, Katrin Dobrindt, Franco Preparata, Micha Sharir, Boaz
Tagansky, and Monique Teillaud. To proofread the manuscript, Jean Berstel and
Franck Nielsen provided their help with an unfailing friendship. The translation
has been carried out by Herv6 Br6nnimann who not only translated but also
corrected the original manuscript in many places. Many thanks to all of them! A
book about geometry could not exist without drawings, and a book about com-
putational geometry could not exist without computer generated drawings. The
Jrdraw software provided the ruler and compass, and together with its designer
Jean-Pierre Merlet it played an essential role in the conception of this book.
Part I
Algorithmic tools
The first part of this book introduces the most popular tools in computational
geometry. These tools will be put to use throughout the rest of the book.
The first chapter gives a framework for the analysis of algorithms. The concept
of complexity of an algorithm is reviewed. The underlying model of computation
is made clear and unambiguous.
The second chapter reviews the fundamentals of data structures: lists, heaps,
queues, dictionaries, and priority queues. These structures are mostly imple-
mented as balanced trees. To serve as an example, red-black trees are fully
described and their performances are evaluated.
The third chapter illustrates the main algorithmic techniques used to solve
geometric problems: the incremental method, the divide-and-conquer method,
the sweep method, and the decomposition method which subdivides a complex
object into elementary geometric objects.
Finally, chapters 4, 5, and 6 introduce the randomization methods which have
recently made a distinguished appearance on the stage of computational geom-
etry. Only the incremental randomized method is introduced and used in this
book, as opposed to the randomized divide-and-conquer method.
Chapter 1
Notions of complexity
Computational geometry aims at designing the most efficient algorithms to solve

geometric problems. For this, one must clearly agree on the criteria to estimate
or measure the efficiency of an algorithm or to compare two different algorithms.
This chapter recalls a few basic notions related to the analysis of algorithms.
These notions are fundamental to understanding the subsequent analyses given
throughout this book. The first section recalls the definition of algorithmic com-
plexity and the underlying model of computation used in the rest of this book.
The second part introduces the notion of a lower bound for the complexity of an
algorithm, and optimality.
1.1 The complexity of algorithms

1.1.1 The model of computation
From a practical standpoint, the performances of an algorithm can be evaluated
by how much time and memory is required by a program that encodes this algo-
rithm to run on a given machine. The running time and space both depend on the
particular machine or on the programming language used, or even on the skills
of the programmer. It is therefore impossible to consider them relevant measures
of efficiency that could serve to compare different algorithms or implementations
of the same algorithm. In order to compare, one is forced to define a standard
model of a computer on which to evaluate the algorithms, called the model of
computation. Thus, to define a model of computation is essentially to define the
units of time and space. The unit of space specifies what types of variables a
memory cell can hold; these are called the elementary variables (or elementary
types). The model specifies what elementary operations can be realized in one
time unit. The running time complexity is therefore the number of elementary
operations that have to be performed in order to realize the operations as de-
4 Chapter 1. Notions of complexity
scribed in the algorithm. Likewise, the spatial complexity describes how many
memory units are needed in order to store all the data required for the execution
of the corresponding program.
The model of computation underlying all the algorithms given in this book is
the so-called real RAM model. In this model, each memory unit can hold the
representation of a real number, and accessing a memory location takes constant
time, that is, time independent of the particular location to be accessed. The
machine can work on real numbers of arbitrary precision for the same cost. The
elementary operations are:
1. the comparison of two real numbers,
2. the four arithmetic operations,
3. all the usual mathematical functions, such as logarithm, exponential, trigo-
nometric functions, etc.
4. the integer part computation.
The assumption that all numbers can be represented exactly allows us to ignore
all the problems related to numerical accuracy, as they occur in the real world. In
particular, the otherwise very relevant problems of robustness of these algorithms
in relation to rounding and numerical inaccuracies are not mentioned in this book.
1.1.2 Notions of complexity
Worst-case and average-case complexity
Each instance of a problem (be it geometric or not) is specified by a set of data

called the input to the problem. The size of the input is the number of memory
units needed to represent this input. When all the input data are elementary,
that is, can be represented in a bounded number of memory cells, this input size
is simply proportional to the number of input data.
When an algorithm is run on a given set of data, one expects the number of
elementary operations executed to depend primarily on the size of the input.
However, the running time also depends on the input itself. The worst-case
complexity of an algorithm, or complexity in the worst case, is a function f(n)
that gives an upper bound on the number of elementary operations run by the
algorithm when the input size is n.
This worst-case complexity is a pessimistic estimator of the running time of an
algorithm. For many algorithms, the upper bound on the number of operations is
reached, or even approached, only for very peculiar inputs which occur marginally
if at all. Sometimes the worst case can easily be avoided by an appropriate pre-
processing. Therefore a better choice to evaluate the efficiency of an algorithm is
1.1. The complexity of algorithms 5
often the average-case complexity, or complexity on the average. This is a func-

tion g(n) that gives the average number of operations (in the statistical sense) if a
probability measure is given on all inputs of size n. The average-case complexity
is generally harder to estimate than its worst-case counterpart. Moreover, it also
depends on the probability measure over the space of all inputs of size n. It is
only useful when this probability measure accurately models the real distribution
of the input to the algorithm.
In the same way, we can define the worst-case and the average-case space com-
plexities of an algorithm.
In this book, the complexities will only be given for the worst case, and the
word complexity will be used as a shorthand for the worst-case complexity (in
time or space). Occasionally, we say that an algorithm runs in time f(n) when
its worst-case time complexity is f (n). Likewise, we say that an algorithm requires
space (or storage) g(n) when its worst-case space complexity is g(n).
Output-sensitive complexity
An algorithm that solves a given problem usually builds, for a given input, a
result called the output, which embodies the solution to the problem. The size
of the output equals the number of memory units needed to store this result.
Obviously, the size of the output depends on the size of the input, but also on
the input itself.
For a given problem, the worst-case output size, or output size in the worst
case, is the function s(n) that upper bounds the output size for all inputs of
size n. The algorithm under consideration needs to at least write the output,
therefore the size of the output in the worst case is an elementary lower bound
on the running time complexity in the worst case.
For a given problem and a given input size, however, the output size can some-
times change a lot depending on the actual input given to the algorithm. For
instance, consider the problem of computing all the intersecting pairs of a set
of line segments in the plane. For a set of n segments, the input consists of 4n
real numbers, two for each endpoint. There might be as few as no intersections,
and as many as n(n)2 In this case it is interesting to have at hand adaptive
algorithms whose time complexity is a function of the output size. The number
of elementary operations executed by such an algorithm depends on the size of
the output for the instance of the problem, and not on the size of the output in
the worst case. For instance, in the problem of reporting all pairs of intersecting
line segments, the number of elementary operations carried out by the algorithm
should be a function of the number of intersecting pairs, which is not true of the
naive algorithm that tests all the pairs for intersection.
An adaptive algorithm can be analyzed in terms of both variables n and s,
the respective sizes of the input and the output. The worst-case complexity of
an adaptive algorithm is the function f (n, s) that upper bounds the number of
elementary operations needed for solving all the instances of the problem with
input size n and output size s. Likewise, the average-case complexity of such
an algorithm is the function g(n, s) that upper bounds the number of operations
carried out by the algorithm, averaged over all the instances of the problem with
input size n and output size s.
The complexity of randomized algorithms
In this book, the reader will find many randomized algorithms, that is, algorithms
whose execution is to some extent random. Such an algorithm will make ran-
dom choices during its execution, and these choices will influence its subsequent
behavior. In all cases, the algorithm will output the correct answer to the given
problem, but the number of elementary operations needed for this will greatly
depend on the random choices. The efficiency of a randomized algorithm is then
evaluated as an average over all possible random choices. The analysis is then
called a randomized analysis. However, such an analysis by no means involves
any statistical hypothesis on the data itself. Rather, the complexity is averaged
over all possible executions of the algorithm in the worst case for the input.
Preprocessing, queries, amortized analysis
It happens frequently that we have to answer many different questions of the same
kind about a given set of data. For example, given a set of lines in the plane, the
questions might ask for some kind of localization. Each query consists of a point
in the plane, and the question asks for the enclosing cell in the subdivision of the
plane induced by the lines. In cases such as this, it often pays off to compute
a data structure during a preprocessing phase, which in turn will be queried
repeatedly for all the different requests. The analysis therefore concerns both
the complexity of the preprocessing phase and that of answering the requests. In
some cases, the data structure is semi-dynamic, which means that it is possible
to add more data on-line; it may also be fully dynamic, meaning that deletions
as well as insertions are allowed. Each type of operation (insertion, deletion,
query) has its own associated cost. Sometimes, the cost of a single operation
is hard to evaluate, but one may estimate the compounded cost of a number of
these operations. The complexity of such a sequence divided by the number of
operations gives the amortized complexity of one operation. Such an analysis is
then called an amortized analysis.
1.1. The complexity of algorithms 7
1.1.3 Asymptotic behavior, notation

The choice of an algorithm for solving a given problem is guided by the associated
complexity, in time or in space, in the worst case or on the average. This choice
is not crucial if the input set remains small. The complexity analysis really
matters when the input size becomes big enough. As a consequence, we are
mostly interested in the growth of the complexity as a function of the input size
n, that is the asymptotic behavior of this function when the variable n approaches
infinity. To analyze an algorithm is thus to determine or at least to upper bound
the dominating term in the time or space complexity. Most of the time, the
order of magnitude will suffice, and we will neglect the numerical constants. We
will then speak of the order of magnitude of the asymptotic behavior of the
complexity.
The usual functions 1,log n, n, n log n, n2, n3 , ...., 2 , whose orders of magnitude
form an increasing sequence, give a natural scale for comparing or evaluating the
complexity of the algorithms.' Of course, this scale can be refined to arbitrary
precision by factoring in other slow-growing functions. Such functions encoun-
tered later in this book are the iterated logarithm log(') n, the very slow-growing
log* n function, or the inverse Ackermann a(n) function. These functions are
defined as follows. The i-th iterated logarithm log(') n of a number n is the num-
ber log log ... log n obtained by composing the logarithmic function with itself i
times and evaluating it on n. The function log* n of n stands for the number of
successive iterations of the logarithm function needed to yield a number smaller
than or equal to 1, starting from a value n. In other words,
log* n = i{•> log( ')n > 1

I log(') n < 1
The value of the log* function remains smaller than 5 for all numbers from 1 up
to 265,535
The Ackermann function is obtained by expanding the following recurrence:
Al (n) = 2n
Ak(n) = An (I),
where Akn) is the function obtained by composing the function Ak with itself
n times. Henceforth, we will write A(n) for An(n). The Ackermann function is
increasing, and its rate of growth is very fast. Here are the first values of this func-
tion: A(1) = 2, A(2) = 4, A(3) = 16, A(4) is a tower of 65,536 powers of 2. The
functional inverse of this function, defined by a(n) = min{p > 1 : A(p) > n},
'The notation log stands for the logarithm in base 2, which in this book will be assumed as
the base for all logarithm functions unless otherwise stated.
is thus an extremely slow-growing function. In fact, ae(n) is at most 4 for all

practical purposes.
In order to compare the growth of different functions, the following notation
is extremely useful. Let f and g be two positive real-valued functions of the
integer-valued variable n.
* By f(n) = O(g(n)), to be read "f(n) is a big-oh of g(n)," we express the

fact that there is an integer no and a real-valued constant c such that
Vn > no, f(n) < cg(n).
* By f(n) = Q(g(n)), to be read "f(n) is a big-omega of g(n)," we express

the fact that there is an integer no and a real-valued constant c such that
Vn > no, f(n) > cg(n).
* By f(n) = E3(g(n)), to be read "f(n) is a big-theta of g(n)," we express

the fact that there is an integer no and two real-valued constants c1 and c2
such that
Vn > no, cig(n) < f(n) < c29(n).
In particular, a function f(n) is 0(1) if and only if it is bounded above by a

constant. A function f(n) is said to be linear if f(n) = E3(n) and quadratic if
f(n) = E)(n 2 ).
Let A be an algorithm and f (n) its complexity, for instance in the running time
and in the worst case. We consider the complexity of the algorithm to be known
if we can specify another function g(n) such that f(n) = E3(g(n)). It will only be
bounded from above if f(n) = O(g(n)) and from below if f(n) = Q(g(n)).
A last piece of notation is sometimes useful. We will write f(n) = o(g(n)), to
be read "f(n) is a little-oh of g(n)," if and only if
lim f( )=0.
n-~oog(n)
Note that this necessarily implies that f(n) = O(g(n)).

An algorithm is a priori all the more interesting if its complexity is of small
order of magnitude. Indeed, all resources being equal, such an algorithm will
work for a greater input size. Nevertheless, one must remain aware of the short-
comings of such a limited view. The asymptotic analysis predicts the existence
of certain constants but does not give any information about the values of these
constants. Consider, for example, two algorithms A and B solving the same prob-
lem. Suppose further that the complexity f(n) of algorithm A is dominated by
1.2. Optimality, lower bounds 9
the complexity g(n) of algorithm B, that is f(n) = O(g(n)). The latter asymp-
totic statement implies that, from a certain input size on, algorithm A will beat
its competitor B in terms of running time. Nothing is said, however, about the
threshold beyond which this is the case (the value of this threshold depends on
the constants no and c concealed by the big-oh notation). One must therefore
refrain from choosing, for a particular practical situation, the algorithm whose
asymptotic analysis yields a complexity with the smallest order of magnitude.
The elegance and simplicity of an algorithm are both likely to lower the order of
magnitude of the concealed constants, and should be taken into consideration if
appropriate. For these reasons, this book usually presents several algorithms for
solving the same problem.
1.2 Optimality, lower bounds

1.2.1 The complexity of a problem
Given a model of computation, the complexity of a problem is the minimum
number of elementary operations needed by any algorithm, known or unknown,
to solve that problem. To put it differently, it is a lower bound on the complexity
of all possible algorithms solving that problem, its running time being evaluated
in the worst case.
An algorithm A that solves a problem P is called optimal if its running time
complexity has the same order of magnitude as the complexity of the problem,
when the size of the input approaches infinity. By definition, any algorithm A
that solves a problem P whose complexity is g(n) has a complexity f (n) such that
f(n) = Q(g(n)). It is therefore optimal if moreover f(n) satisfies f(n) = O(g(n)).
It is of the highest importance to determine the complexity of a problem for
which one seeks a solution, since this complexity bounds the complexity of any
algorithm that solves the problem. The size n of the input and the size s(n) of
the output are natural lower bounds on the complexity g(n) of a problem P:
g(n) = Q (max(n, s(n))).
By definition, the complexity f(n) of any algorithm A that solves problem P is
an upper bound on the complexity of P:
g(n) = O(f(n)).
In particular, if there is an algorithm A with complexity f(n) = O(max(n, s(n)))
that solves problem P, then this algorithm is optimal and the complexity of the
problem is g(n) = e(max(n, s(n))). In other cases, the complexity of a problem is
much more difficult to establish, and there is no general method for this purpose.
The next two subsections give two short examples of methods that might be used
to determine the complexity of a problem.
1.2.2 The example of sorting: decision trees
Sorting n numbers according to the natural (increasing) order is one of the rare
problems whose complexity can be found by direct reasoning. Given a finite
sequence of n numbers, X = (xl,...,xx), all in some totally ordered set (for
instance, N or R), to sort them is to determine a permutation a of {1,..., n}
such that the sequence
MY = (Yl v Yn), Yj = X,:(j)

is totally ordered, that is,
Y1 < Y2 < .< n
If no particular property of the elements of X is known, the only operation

at hand to sort X is the comparison of two elements. One speaks of sorting
by comparison only. The following theorem shows that the complexity of the
sorting problem is Q(nlogn), in the comparison model of computation. In the
same model, there are algorithms that solve this problem in the corresponding
O(nlogn) time: for instance, such an algorithm is given in chapter 3. In this
model, the complexity of the problem is thus e(nlogn).
Theorem 1.2.1 Sorting n numbers using only comparisons requires at least

Q(n log n) comparisons.
Proof. The proof of this theorem is based on the idea of a decision tree. One
can always assume that the sequence under consideration does not contain the
same element twice; thus all the numbers in the sequence are distinct. For lack of
other information on the input data, the algorithm can only perform comparisons,
then branch accordingly, depending on the result of this comparison. Branching
is a binary process since there can be only two results to the comparison. The
execution of such an algorithm can be represented by a binary tree, the decision
tree. Each leaf represents a possible output from the algorithm; in our case,
an output is one of the n! possible permutations of the set {1,...,n}. Each
internal node represents some state of the algorithm, at which the algorithm will
perform a comparison. Depending on the result of this comparison and on its
current state, the algorithm will then branch to its right or left descendant in the
tree, and subsequently perform the comparison stored at that node, or output
the corresponding permutation if it reaches a leaf. All computations begin at the
root of the tree, and each execution therefore corresponds to a path from the root
to a leaf of the tree. The number of comparisons performed by the algorithm in
the worst case is thus the height of the decision tree. A possible decision tree
sorting three elements a, b, c is shown in figure 1.1. For our sorting problem, the
1.2. Optimality, lower bounds 11
Figure 1.1. A decision tree sorting three elements a, b, c.
ri (n)
Problem A Problem B
73 (n)
Figure 1.2. Transforming one problem into another.
decision tree has at least n! leaves, and its height h is thus at least log(n!) which,
according to Stirling's approximation formula, 2 is Q(n log n). 0
1.2.3 Lower bounds by transforming one problem into another

The reduction method is undoubtedly the method used the most frequently to
determine the complexity of a problem. It consists of transforming the instance
of the problem into an instance of another problem, the complexity of which is
well known, or conversely of transforming the instance of another problem into an
instance of the problem under consideration. To make the method more explicit,
let A and B be two problems. We say A is transformed into B in time i-(n) if:
1. the input to problem A can be converted into an input suitable for problem
B, using r7i(n) elementary operations,
2. it is possible to convert the solution to problem B on the latter input
into a solution to problem A on the former input, using T3(n) elementary
operations, and
3. ri(n) + r 3 (n) = 7(n).
2
Stirling's approximation formula states that n! = 7 (n)fn (1 + 1 + o( )) where e
stands for the base of natural logarithms.
Theorem 1.2.2 If a problem A, whose complexity is f(n), can be transformed

in time T(n) into a problem B whose complexity is g(n), then
f(n) = O(g(n) + T(n)),
g(n) = Q(f(n) -T(n)).
Proof. If the complexity of B is g(n), then there is an algorithm that solves

problem B in g(n) elementary operations, and the transformation allows problem
A to be solved using g(n) + T(n) elementary operations. Conversely, if f(n) is
the complexity of problem A, there is no algorithm that solves B using less than
f(n) - T(n) operations. El
Hence, the complexity of B gives an upper bound on the complexity of A, and
that of A gives a lower bound for the complexity of B. We will show below that
numerous geometric problems contain a sorting problem, for instance computing
the convex hull or the Voronoi diagram of n points in the plane. The lower bound
Q(n log n) holds for these problems in a suitable model of computation.
1.3 Bibliographical notes

The exposition of the concepts of complexity and optimality given in this chapter is
purposely kept to a strict minimum. The reader seeking a more detailed discussion on
all these notions is referred to one of the classical textbooks on the analysis of algorithms
such as those by Aho, Hopcroft, and Ullman [6], Knuth [142], Sedgewick [200], Cormen,
Leiserson, and Rivest [72], and Froidevaux, Gaudel and Soria [108] (in French).
Chapter 2
Basic data structures
Data structures are the keystone on which all algorithmic techniques rely. The
definition of basic yet high-level data structures, with precise features and a well-
studied implementation, allows the designer of an algorithm to concentrate on
the core issues of the problem. For the programmer, it saves the tedious task of
creating and administrating each pointer.
Throughout this book, we describe data structures especially designed for rep-
resenting geometric objects and dealing with them. But computational geometers
also make extensive use of data structures that represent subsets or sequences of
objects. These structures can be used directly by the algorithms, or modified and
augmented for geometric use. The first part of this chapter recalls the terminol-
ogy and features of each basic data structure used in this book. It is useful to
know how these structures can be implemented and what their performances are.
The most delicate problem is undoubtedly the one addressed by dictionaries and
priority queues, which treat finite subsets of a totally ordered set (the universe).
To achieve better efficiency, these structures are usually encoded as balanced bi-
nary trees. For instance, the second part of this chapter describes red-black trees,
a class of balanced trees that can be used to implement dictionaries and priority
queues. Finally, when the universe is finite, dictionaries and priority queues can
be even more efficiently implemented by other more sophisticated techniques, the
characteristics of which are given without proof in the third part of this chapter.
The sole purpose of this chapter is to present, as far as data structures are
concerned, the information necessary for a thorough understanding of the forth-
coming algorithms. In particular, the authors by no means claim to present a
comprehensive account of this topic, and the interested reader is urged to refer
to the references given in the bibliographical notes.
14 Chapter 2. Basic data structures
2.1 Terminology and features of the basic data struc-

tures
2.1.1 Lists, heaps, and queues
Lists are the basic data structures used to represent a sequence of elements of a
set.
Let X = {X1 ,X 2 ,...,XXn} be such a sequence. Any structure that wishes
to represent this sequence should, at the very least, allow sequential access to
these elements. The basic operation that achieves this is the successor operation
which gives a pointer to the element Xi+l following the current element Xi. In
some situations, both directions may be needed, and the data structure should
also allow the predecessor operation which gives a pointer to the element Xi-
immediately preceding the current element Xi.
A list must also handle insertions of new elements and deletions of any of its
elements. Therefore the list should allow an insert operation, for inserting a
new element after a given position, and a delete operation, for deleting a given
element. Insertion is often a must, if only for building the data structure, and
deletion is often required as well.
Finally, we sometimes need two other operations on lists: concatenationwhich
appends one list at the end of the other, and a converse split operation which
breaks up a list into two parts at a given position.
There are implementation variants that efficiently realize these operations.
Most of the time, a singly or doubly linked list suffices (see figure 2.1). Each
element of the list is put into some memory location called a record, which in-
cludes two fields: one contains the value of that element, and the other a pointer
to the record of the next element in the order of the list. When the list is doubly
linked, a pointer to the record of the previous element in the order of the list is
also supplied.
The data structure also contains a pointer to the first element of the list.
Sometimes a pointer to the last element is also useful.
Such a data structure occupies a space that is proportional to the number of
elements in the list. If this number is n, the space needed for the structure is O(n).
It allows the following operations to be performed in constant time: successor,
predecessor (for a doubly linked list), insertion, and deletion. Such a list can
therefore be built and enumerated in O(n) time if n is the length of the sequence
it represents. Also, pointers to the first and last elements allow concatenation
and partition to be performed in constant time.
Stacks and queues are particular implementations of lists when the insertions
and deletions only occur in special positions.
In the case of a stack, all insertions and deletions happen after the last element
2.1. Terminology and features of the basic data structures 15
Em-- FE---
Figure 2.1. Singly and doubly linked lists.
in the list, which is called the top of the stack: to stack an element means to
insert it as the last element of the list, and to pop consists of deleting the element
that was stacked the most recently. Stacks are therefore particularly suited to
process the elements of a set in the LIFO order, which stands for "last in, first
out."
In the case of a queue, all insertions occur at the end of the list, whereas all
deletions take place at the beginning of the list. Queues are therefore well suited
to process elements in the FIFO order, which stands for "first in, first out." This
is the normal order for a waiting line, or queue, hence the name given to this
data structure.
Stacks or queues can always be implemented as general lists. There are more
specific methods to implement these data structures but we will not expand on
them in this book.
2.1.2 Dictionaries and priority queues

A data structure that represents a set S, subset of a universe U, must at least
allow the following operations:
* query: given an element x of U, find out whether x belongs to S,
* addition: add an element x of U into S,
* deletion: delete an element x from the set S.
When the universe is totally ordered, the data structure is generally based on
the order of the elements of S. Additions are then preferably called insertions
and one may wish to perform locations (to be defined below) rather than queries.
Elements of a totally ordered universe are called keys and the order on U is
denoted by <. The words smaller, greater, minimum, and maximum refer to the
total order on U. If S is a subset of the totally ordered universe U, then the data
structure may be required to handle, in addition to the three previous operations,
some of the following operations:
* location: given an element x of U, find the smallest element y of S such

that x < y,
* minimum: find the smallest element in S,
* maximum: find the greatest element in S,
* predecessor. find the element of S immediately preceding a given element

x of S,
* successor- find the element of S immediately following a given element x of

S.
We call a dictionaryany data structure that can perform queries, insertions and
deletions. If it supports searching for the minimum as well, we call it a priority
queue. If it supports all the operations detailed above, we call it an augmented
dictionary.
Priority queues and dictionaries can be implemented using lists or arrays. When
the universe is totally ordered, it is often more efficient to use balanced data
structures such as red-black trees, described below.
2.2 Balanced search trees

2.2.1 Graphs, trees, balanced trees
A graph is a pair (X,£), where X is a set of elements called nodes of the graph,
and £ is a set of pairs of nodes of X, these pairs being called arcs. The graph
is directed if the arcs are considered as ordered pairs. A path in the graph is an
ordered sequence of nodes such that any two consecutive nodes are joined by an
arc. The graph is connected if any two nodes can be joined by a path, and acyclic
if no non-empty path can start and end at the same vertex without passing some
other vertex of the graph at least twice. A tree is a directed, connected, and
acyclic graph. As a consequence, one of the nodes stands out as having no arc
coming into it; this node is commonly referred to as the root. Conversely, nodes
having no arc coming out of them are called the leaves. Graphs, trees and their
specific vocabulary are extensively described in the reference works cited in the
bibliographical notes. In these references, nodes are sometimes also called vertices
and arcs are commonly called edges. In this book, we stick to the words nodes and
arcs for graphs, and restrict the use of the words vertices and edges to geometric
objects. We invite the reader interested in further investigation to refer to these
references if he or she should feel the need for it. Here, we content ourselves
with reviewing how balanced search trees can be used to efficiently implement
dictionaries and priority queues.
2.2. Balanced search trees 17
A branchof the tree is a path that stretches from the root to a leaf of the tree.
A tree is considered to be balanced if all its branches have approximately the
same length. This property, to be made precise below, ensures the efficiency of
the data structure but complicates the insertion and deletion operations. Indeed,
after each such operation, the structure must be rebalanced. There are several
kinds of balanced trees, such as AVL trees, 2-3 or 2-3-4 trees, or even red-
black trees. There are also many ways in which these variants can be used
to implement dictionaries and priority queues. All the performances of these
solutions are equivalent and optimal: if the set S stored in the data structure has
n elements, the data structure occupies O(n) space and any insertion, deletion, or
query takes O(log n) time. For instance, the next section describes how to achieve
these performances using red-black trees, and analyzes the corresponding cost of
these operations.
2.2.2 Red-black trees

A red-black tree is a complete binary tree, that is, each node has either two
children or none. The arcs, colored either red or black, satisfy the following
constraints:
1. The paths from the root to all the leaves have the same number of black
arcs.
2. All the leaves are related to their parent by a black arc.
3. There cannot be two consecutive red arcs along a path from the root to a
leaf.
All the nodes have a level, which is the number of arcs on the path from the root
to that node, and a black level, which is the number of black arcs on that path.
The number of black arcs on a path from the root to a leaf is called the black
height of the tree, since it does not depend on the particular leaf.
It is easy to see that a red-black tree is approximately balanced: the longest
branch cannot have more than twice as many arcs as the shortest.
We propose to show how such a data structure can be used to implement a
dictionary on a finite set S drawn from a totally ordered universe U. The red-
black tree is used as a searching data structure: to each node corresponds a key
and two pointers towards its children. The keys attached to the leaves serve to
represent the elements of S. The keys attached to the internal nodes serve as a
guide for the searching operations. The key attached to an internal node must
be greater than or equal to all the keys stored in its left subtree-the subtree
rooted at its left child, and smaller than all the keys stored in its right subtree.
For instance, the key attached to an internal node can be systematically set to
the greatest of the keys stored in its left subtree. A left-hand depth-first traversal
of the tree visits all the nodes of the tree in the following order: the root first,
then recursively the nodes in the left subtree, and finally the nodes in the right
subtree. Such a traversal visits the leaves of the tree in the order of the elements
of S.
Along with the key and the pointers to its children, the information stored at
a node contains a special field to mark the color, either red or black, of the arc
linking this node to its parent. To simplify the exposition, the color of an arc is
often transferred to the node as well, and so we call a node black if it is linked
to its parent by a black arc, and red if it is linked to its parent by a red arc. By
convention, the root of the tree is always colored black. From now on, we denote
by the same letter N, 0, P, Q, R, S.... both the node and the key stored at that
node.
Storage
Let S be a set of n elements, subset of the totally ordered universe. A red-black

tree representing S has n leaves, and therefore has n - 1 internal nodes and
2(n - 1) arcs. The space required to store such a structure is thus O(n).
Queries
To find out whether or not an element S of the universe U belongs to the set
S, we need only follow a branch of the tree. At each internal node N, the next
node in the branch is identified using a comparison between N and the key S
that we are searching for. If S < N, the search goes through the left child of
N; if S > N the search goes instead through the right child of N. The search
always ends up at a leaf S' of the tree: the answer is that S is present if S' = S,
and that S is missing from S if S' = S. In the latter case, S' is the element
of S that immediately precedes or follows S according to the order on U. The
following theorem shows that, if there are n elements in S, such a search visits
only E(log n) nodes of the tree, and therefore runs in E(log n) time.
Lemma 2.2.1 If a red-black tree has n leaves, any path from the root to a leaf
has at least 2 log n and at most 2 log n arcs.
Proof. The easiest proof of this result is to refer to a different kind of tree, the
2-3-4 tree. A 2-3-4 tree is a tree whose nodes have either 2, 3, or 4 descendants,
and all the paths from the root to the leaves have the same length, which is the
height of the 2-3-4 tree. From a red-black tree, it is easy to make a 2-3-4 tree by
merging all the nodes that are linked through red arcs (see figure 2.2). The height
2.2.Balaced
tres earc 1
AQ
XQ
QR
Figure 2.2. The correspondence between red-black trees and 2-3-4 trees.
In this and the subsequent pictures, the black arcs of red-black trees are
represented in bold, the circles stand for internal nodes, rectangles for the
leaves, and triangles stand for arbitrary subtrees.
h of this 2-3-4 tree is exactly the same as the black height of the corresponding
red-black tree.
The red-black tree and its associated 2-3-4 tree have the same number of
leaves, n, and the height h of the 2-3-4 tree satisfies
2h <rn < 4h
From this, it follows that the number h of black arcs on any branch is at least
2 log n and at most log n. The total number of arcs on such a branch cannot be
less than h, nor can it be more than 2h. El
Figure 2.3. Red-black trees: insertion.

The picture assumes that S < S'. Just invert the two leaves S and S' if the
converse is true.
Insertions
The insertion of an element S into the set S, represented by a red-black tree,

can be carried out in three stages.
First stage. We first find the location at which the new element must be
inserted into the structure. For this, we follow a branch of the tree, in the same
fashion as in the query explained above. The nodes along this branch are stored
on a stack as we go along, the node encountered last being on the top of the stack.
This stack will be used in the subsequent stages for rebalancing and recoloring
the tree in a permissible manner. By assumption, the query gives a negative
answer and ends on a leaf, the key S' of which is different from the key S of the
element to be inserted.
Second stage. This is the phase where the actual insertion is performed. We
replace the leaf S by an internal node R, with two newly created black leaves
as its children, whose keys are S and S' in the appropriate order. The node R
is linked by a red arc to the node Q that is stored on top of the stack. (This
was the last node to be visited, and was the parent of S prior to the insertion,
see figure 2.3.) The key of R is set to the smaller of the two keys of S and S'.
The node R is stacked above Q. This way, the first two structural constraints of
red-black trees are preserved; however, the last one is not satisfied if Q itself is a
red node.
Third stage. This stage is a rebalancing stage whose purpose is to enforce the
three structural constraints on red-black trees. During this stage, the algorithm
maintains the following invariant: the top of the stack stores a red node R, and
the two elements stored immediately before R in the stack are the parent Q and
the grandparent P of R; Q and R are the only nodes that don't comply with the
third constraint, therefore P itself is black, as well as the other child of Q and
the two children of R. The node P itself may have both red children, or only Q
is red and the other is black. The current step goes as follows:
1. Should P have both a black and a red child, then the third rule can be
enforced by one of the following transformations (as in figure 2.4): simple
(a) Simple left rotation
~~> PA
(b) Double left-right rotation
Figure 2.4. Red-black trees: rotations.
left (resp. right) rotation if Q and R are both right (resp. left) children;
double right-left (resp. left-right) rotation if Q is a right child and R a left
child (resp. Q is a left child and R is a right child). Figure 2.4 shows only
a simple left rotation and a double right-left rotation. We leave it to the
reader to represent the symmetric rotations.
2. Should P have two red children, then the algorithm colors both children
black and colors P red instead (see figure 2.5), unless P is the root of the
tree in which case it is left black and nothing else is done. If the parent
of P is black or at the root of the tree, then the third constraint has been
restored, and the whole rebalancing task is over. If the parent of P is red,
then the default in the third rule has been carried up two levels towards
the root of the tree. Nodes R and Q are popped from the stack and the
next step takes over with node P, its parent, and its grandparent.
The analysis of an insertion operation is almost immediate. The first stage

requires E(logn) operations if the set S to be searched has n elements. The
actual insertion in the second stage can be carried out in constant time. As to
Figure 2.5. Red-black trees: changing the colors.
) A
) A
>1 Mo.
Figure 2.6. Red-black trees: deletions.
the third stage, only O(log n) nodes may need to be recolored, and only as many
(simple or double) rotations may need to be performed. Red-black trees therefore
allow a new element to be inserted in time O(logrn).
Deletions
As with insertions, deletions can be performed in three stages.

First stage. A search for the key S to be deleted leads to the leaf that needs to
be deleted, just as explained for queries or insertions. If x is not found to belong
to S, then nothing else is done, otherwise the algorithm performs the second and
third stages below.
Second stage. This is where actual deletion is performed: the leaf S that
was located in the first stage, and its parent P, are removed from the tree; the
sibling S' of S (the other child of P) is linked directly to the parent Q of P (see
figure 2.6). If the arc linking P to Q is black, then the path leading from the root
to S' lacks one black arc after the removal, and the first structural constraint no
longer holds.
Third stage. We rebalance the tree obtained by the removal in the second
stage. This operation is carried out in steps. At the current step, the tree contains
one and only one short node: this is a node X such that the black height of the
subtree rooted at X is one arc smaller than that of other subtrees rooted at the
same black level in the tree. In the first step, the only short node is S'. Let X be
the current short node, Q its parent, and R the other child of Q. Node X being
the only short node, R cannot be a leaf of the tree.
1. Should R be black with two red children, then rebalancing can be obtained
by performing the rotation depicted in figure 2.7, case 1.
2. Should R be black with both a black and a red child, rebalancing can be
obtained by the double rotation depicted in figure 2.7, case 2.
3. Should R be black with two black children, two cases may arise. If node Q,
the parent of X and R, is red, then the tree can be rebalanced by changing
the colors as shown in figure 2.7, case 3a: Q is recolored in black and R in
red. If Q is black, the tree cannot be rebalanced in a single step. Changing
the colors as shown in figure 2.7, case 3b, makes the parent of Q become
the short node, and the next step takes over with this node as the short
node.
4. Finally, should R be a red node, the transformation explained in figure 2.7,

case 4, will yield a tree whose short node has a black sibling and therefore
can be taken care of by one of the transformations 1, 2, or 3a.
To summarize, the structural properties of a red-black tree can be restored with

at most O(log n) transformations of type 3b, one transformation of type 4 followed
by one of type 1, 2, or 3a, summing up as O(logn) elementary operations. The
first stage requires only O(logn) operations and the removal can be performed
with only a constant number of elementary operations. As a consequence, red-
black trees can be used to represent a set S of n elements from a totally ordered
universe U, allowing us to perform deletions in time O(log n).
As described above, red-black trees can be used to implement a dictionary. To
obtain a priority queue, it suffices to maintain a pointer towards the leaf storing
the smallest (or the biggest) element of S. To have an augmented dictionary,
one can add two pointers to each leaf, pointing to the previous and the next
element of the set. Maintaining these additional pointers does not modify the
complexity of the insertion and deletion operations, and allows the predecessor
or the successor to be found in constant time.
The following theorem summarizes the performances of red-black trees.
(1)>
(2)
(3a)>
(3b)>
(4)>
Figure 2.7. How to rebalance a red-black tree after a deletion.
Theorem 2.2.2 (Red-black trees) Let S be a set of n elements drawn from a

totally ordered universe U. A red-black tree can be built in space O(n) and time
O(n log n) to represent this set. The tree allows each of the following operations
to be performed in time O(log n): queries, insertions, deletions, minimum, max-
imum, and locations. Operationspredecessor and successor take constant time.
As an example of how to use such a data structure, we recall the problem of

sorting n real numbers. The solution we give uses red-black trees to store the
2.S. Dictionary on a finite universe 25
set of these n numbers. The tree can be built in O(nlogn) time, uses O(n)
space, and the elements can be enumerated in order by performing a left-hand
depth-first traversal of the tree. The only operation on the numbers used in this
algorithm is comparison. Taking into account theorem 1.2.1, we have proved the
following:
Theorem 2.2.3 (The complexity of sorting) The problem of sorting n real

numbers has complexity e9(n log n) in the comparison model.
2.3 Dictionary on a finite universe

When the totally ordered universe is finite, it is possible to implement dictionar-
ies and priority queues more efficiently by using more sophisticated structures.
Henceforth, we will not use these techniques except in chapter 6, and only to
show how to optimally implement dynamic randomized algorithms. Therefore,
in this section, we will only recall the performances of these structures, without
proof or further explanations. The reader may want to skip this section in a first
reading.
In this entire section, we assume that the set S we want to maintain is a subset
of a totally ordered universe U, which is finite and of cardinality u. If u is not
too big, the simplest way to implement a dictionary is to use an array of size
u which allows queries, insertions, and deletions to be performed in constant
time. To implement an augmented dictionary, one may have recourse to a data
structure commonly referred to as a stratified tree (see also exercise 2.3) whose
performances are given in the following theorem.
Theorem 2.3.1 (Stratified tree) Let S be a subset of a totally ordered uni-

verse of finite size u. The set S may be stored in a stratified tree, which uses
O(u log log u) space and can be built in time O(u log log u). Each of the insertion,
deletion, location, minimum, predecessor, and successor operations can then be
performed in time O(log log u).
Sometimes, the size of the underlying universe is just too big for this method
to be practical. Hashing methods can then be used as a replacement.
Perfect dynamic hashing is a method that stores the dictionary over a finite,
albeit huge, universe. In this method, random choices are made by the algorithm
during the execution of the insertion and deletion operations. Such algorithms
are called randomized below. The cost of these operations (insertions, deletions)
depends on the random choices made by the algorithm and can only be evaluated
on the average over all possible choices. Such an analysis is also said to be
randomized. Moreover, it is impossible to bound the cost of a single operation.
insertion minimum predecessor

storage deletion maximum successor
location
red-black tree n log n log n 1
stratified tree u log log u log log u log log u 1
perfect dynamic hashing n 1
stratified tree with
perfect dynamic hashing n log log n 1 1
Table 2.1. Implementations of dictionaries and priority queues.
The last three solutions assume that the universe is finite and of size u.
However, the cumulative cost of a sequence of m operations can be analyzed:

in this case, the analysis is said to be amortized. The cost of a sequence of
operations divided by the number of operations then stands as the amortized
cost of an operation. The performances of perfect dynamic hashing are given in
the following theorem (see also exercises 2.5 and 2.6).
Theorem 2.3.2 (Perfect dynamic hashing) Let S be a subset with n ele-

ments of a totally ordered and finite universe with, say, u elements. Perfect
dynamic hashing can be used to implement a dictionaryfor S using O(n) space.
Each query on the dictionary takes 0(1) time, and the amortized cost of an in-
sertion or deletion is also 0(1).
Finally, by combining both stratified trees and perfect dynamic hashing, one
may build a data structure that performs well on all the operations of an aug-
mented dictionary. Henceforth, this combination of data structures, a data struc-
ture in its own right, will be referred to as an augmented dictionary on a finite
universe. The theorem below summarizes its characteristics.
Theorem 2.3.3 (Augmented dictionary on a finite universe) Let S be a

subset with n elements of a totally ordered and finite universe with u elements.
An augmented dictionaryfor S may be built using O(n) storage with the following
performances: The minimum, predecessor, and successor operations can be per-
formed in time 0(1), and location in time O(loglogn). Insertions and deletions
run in amortized time O(loglogn) on the average.
Table 2.1 summarizes further the performances of the different data structures
discussed here that may be used to implement a dictionary or a priority queue.
2.4 Exercises
Exercise 2.1 (Segment trees) Segment trees were created to deal with a collection of
intervals on the one-dimensional real line. Intervals may be created or deleted, provided
2.4. Exercises 27
that the endpoints belong to a set known in advance. The endpoints are sorted, and
thought of as the integers {1, . . . , n} via a one-to-one correspondence that preserves the
order. The associated segment tree is a balanced binary tree, each leaf of which represents
an elementary interval of the form [i, i + 1]. Each node of the tree therefore corresponds
to an interval which is the union of all the elementary intervals associated with the leaves
of the subtree rooted at that node. Intervals of this kind will be called standardintervals,
and we will speak of a node instead of its associated standard interval.
The intervals of the collection are stored at the nodes of the tree. An interval I is
stored in the structure at a few nodes of the tree: a node V stores I only if its associated
standard interval is contained in I, but the standard interval of the parent of V is not.
1. Let 1, resp. r, be the left, resp. right, endpoint of I. Let VI be the standard
elementary interval whose left endpoint is 1, and let Vr be the standard elementary
interval whose right endpoint is r. Let Vf be the smallest standard interval containing
both VI and Vr. The node Vf is the nearest common ancestor to both V1 and Vr, and it
is called the fork of I. Show that the nodes which are marked as storing I are precisely
the right children of the nodes on the path joining Vf to VI in the tree, together with
the left children of the nodes on the path joining Vf to Vr in the tree. Deduce from this
that the nodes that store I correspond to a partition of I into O(log n) standard disjoint
intervals, with at most two intervals at each level of the tree.
2. At each node, a secondary data structure accounts for the set of intervals stored
by that node. According to the application, the data structure may list the intervals or
simply maintain in a counter the number of these intervals. To add an interval to the
segment tree simply consists of adding it to each of the secondary data structures of the
nodes storing this interval, or incrementing the counter at these nodes. Deletions are
handled similarly. Assume that only a counter is maintained. Show that an insertion or
deletion can be performed in time O(log n). Show that the segment tree can be used to
count the number of intervals containing a given real number x, in time O(log n).
Exercise 2.2 (Range trees) Given a set of n points S in Ed, we wish to build a data
structure to efficiently answer queries of the following kind: count the number of points
inside an axis-oriented hyper-rectangle, or report them. One solution consists of building
a range tree, a data structure particularly suited to this kind of query, which we describe
now.
* The first level of the structure is a segment tree T1 (see exercise 2.1) built on
the first coordinates of the points in S, that is on the set {xi(P) : P e S}. For
each node V of T1 , we denote by Sd(V) the set of those points P of S whose first
coordinate x 1 (P) belongs to the standard interval of V. The set Sd- (V) is the
projection of Sd(V) onto Ed-1 parallel to the x1 -axis.
. If d > 2, every node V of T1 has a pointer towards a range tree for the set of points
Sd-l(V) in Edl.
1. We first assume that the queries ask for the number of points in S inside a given
hyper-rectangle Rd (the counting problem). Let q(S, Rd) be the time it takes to answer a
query on the hyper-rectangle Rd. Let V1 stand for the collection of all the nodes storing
the projection of Rd onto the x1 -axis, and Rd-, be the projection of Rd parallel to the
x1 -axis. Show that
q(S,Rd) = O(logn) + E q(Sd-l(V),Rd-1).

VEVi
From this, show that the maximum amount of time that a query can take on a set of n
points in d dimensions is
q(n,d) = O(logn)q(n,d -1) = 0 ((log n)d) .
Show that a query in the reporting case can be answered in 0 ((log n)d + k) time if k is
the number of points to be reported.
2. Show that the preprocessing space requirement and time are both 0 (n(log n)d).
Exercise 2.3 (Stratified tree) Let S be a subset of a finite, totally ordered universe
U. Let u be the number of elements of U, and without loss of generality assume that
u = 2 k. For convenience, we identify the set of possible keys with {0, 1, . .. , 2k - 1}. A
stratified tree ST(U, S) that implements an augmented dictionary on S is made up of:
* a doubly linked list which contains the elements of U. Each record in this list
has three pointers sub, super, and rep, and a boolean flag marker to identify the
elements of S.
* a representative R with two pointers to the maximal and minimal elements in S,
and a boolean flag to detect whether S is empty or not.
* stratified trees ST(Uj, Si) for the sets Si = Sn Ui and the universes Ui =
2 [k/21
i 2[k/2j + {O,1, . . ., 2Lk/2j - 1}, with i ranging from 0 to 2 fk/21 -1. Depending on
the parity of k, each sub-universe Ui contains \/6 or vfi elements of U.
* A stratified tree ST(U', 1?) for the set of representatives of ST(Ui, Si). The rep-
resentative Ri of ST(Ui, Si) is the element whose key equals i in the set U' =
{0,1,..., 2 rk/21-1}. Depending on the parity of k, the size of U' is u or /u.
The trees ST(Ui, Si) and the tree ST(U', %) are called the sub-structures of ST(U, S).
In turn, ST(U, S) is called a super-structureof those trees. Pointers sub and super keep
a link between a record in the list and the corresponding record in the list of the sub-
structure (resp. super-structure). The pointer rep points toward the representative of
the structure.
1. Show that the stratified tree ST(U, S) can be stored in space O(uloglogu), and
can be built for an empty set S in time 0(u log log u).
2. Show that each operation: insertion, deletion, location, minimum, predecessor,
successor, can be performed in time 0(log log u).
Exercise 2.4 (Stratified trees and segment trees) Let U be a totally ordered, fi-
nite universe with u = 2 k elements. Consider a complete and balanced binary tree C3T
whose leaves are associated with the elements of U. Consider further the set {0, 1 ... , k}
of levels of that tree, and build a segment tree T on this set. Show that you can do it
in such a way so that each sub-structure of the stratified tree built on the universe U
corresponds to a standard interval on T.
2.4. Exercises 29
Exercise 2.5 (Perfect dynamic hashing) Let S = {xl,... , x} be a subset of a fi-

nite universe U. Let u be the number of elements of U, and without loss of generality
assume that u = 2 k. A hash function h is any function from U to a set of keys T of
size t = 21. We say that a class of hash functions is universal if, for any elements i, j in
U, the probability that h(i) = h(j) for a random h in X- equals 1/t. The class is almost
universal if the latter probability is 0(1/t).
1. Let U be the vector space IF2k of dimension k over the field IF2 with two elements,
and T be the vector space IF2 1. Let X- be the class of all injective linear maps h from U
to T, that is, such that h(a) = h(b) only if a = b. Show that X- is a class of universal
hash functions.
2. Let XH be the class of all functions h(x) = (kx mod p) mod t, for a fixed prime
number p > u and all elements k < p. Show that XH is an almost universal class of hash
functions.
Exercise 2.6 (Perfect dynamic hashing) Let S = {x 1 ,. .., x} be a subset of a fi-

nite universe U. Hash functions and universal classes of hash functions are defined in the
previous exercise. For convenience, we identify the universe with {0,1, ... ,u - 1} and
the set of keys with {0,1,. . ., t - 1}. Given a hash function h, we say that two elements
a and b in U collide at j if h(a) = h(b) = j. Given a set S of n objects in U, the problem
is to find a hash function that gives as few collisions as possible over S.
1. Given a random hash function h in a class X of hash functions, let Sj be the set
of elements in S that are mapped onto j, and let nj be the size of Sj. Let N be the
expected value, for a random element h of XH, of > (ni ). Show that N < 'Kl
if XH is
universal, and that N = O(n 2 /t) if X is almost universal.
From now on we assume that X- is a universal class of hash functions.
2. Note that two elements collide at j if and only if nj > 1. If t = n 2 , show that the
probability of having no collision over S for a random h in X- is at least 1/2. If t = n,
show that ,j_1 n2 < n with probability at least 1/2 for a random function h in 'H.
3. We now describe a two-level hashing scheme that has no collision with high proba-
bility. For any j in T such that nj > 1, we set tj = nj and pick a random hash function
hj in a universal class of hash function 7{j onto a set 7j of tj elements. The two-level
hashing scheme first maps an element x to j = h(x), and if nj > 1, maps x onto hj( )
in Tj. Assume that the sets T and 7j are all disjoint. Show that for a given S, the
two-level hashing scheme has no collision, uses space O(n), and has a query time 0(1)
with probability at least 1/2 over the choice of h and of the hj's.
4. In order to make this scheme dynamic we use the standard doubling trick: instead
of taking t = n, we take t = 2n so that the first table can accommodate twice as many
elements; similarly, for each j such that nj > 1, we take tj = 4nj. When a collision
occurs during the insertion of an element into a subtable, this subtable is rehashed.
Also, when the number of elements effectively present in a table or a subtable exceeds
twice (or falls below half) the number of elements present in the table during its last
rehash, the size of this table is doubled (or halved) and this table is rehashed. Show that
the expected number of times the table or subtables are rehashed is 0(1) if at most n
operations are performed in a table that stores n elements. Concrude that this dynamic
two-level hashing scheme uses space O(n), has a query time 0(1), and allows insertions
and deletions in time 0(1).
Exercise 2.7 (Persistent dictionary) A persistent dictionaryis a data structure that

maintains a subset S of a totally ordered universe while supporting insertions, deletions,
and locations in the past. The data structure thus obtained depends on the chronology
of the insertions and deletions: each insertion or deletion is given, as a date, its rank in
the sequence of insertions and deletions. A location in the past has two parameters: one
is the element 2 to search for in the universe and the other is the date at which the data
structure is to be queried. For such a query, the data structure answers the smallest
element greater than x that belongs to the set S at the date i, that is just after the i-th
insertion or deletion operation.
To implement a persistent dictionary, it is convenient to use a red-black tree in which
each node has k + 2 pointers, k being some non-negative integer constant. When during
a rotation the node must change one or both of its children, the pointers to the children
are not destroyed. Instead, the new child is stored in a new pointer among the k + 2. If
all these pointers are in use, then a copy of the node is created with two pointers, one
for each of the current children of the node. The parent of that node must also keep a
pointer to the new node, and the same mechanism is used: if all the pointers in the parent
node are used, a copy is made, and so forth. Each of these pointers has a time stamp
that remembers the date of its creation. When the root itself is copied, a new entry is
created in a dictionary of roots which is ordered chronologically. To perform a location
in the past, the algorithm begins by looking up the dictionary for the most recent root
prior to the requested date. Then location is performed in the normal fashion, taking
care to follow the pointer to the children that is the most recent prior to the requested
date.
Show that the space requirement of such a structure is 0(n) if n insertions or deletions
are performed, starting with an empty set. Show that the cost of a location-in-the-past
query is O(logn).
Hint: To estimate the storage requirements, an amortized analysis is useful. A node of

the structure is said to be active if it can be reached from the most current root, and
dead otherwise. The potential function of the data structure is defined as the number of
active nodes minus a fraction k 11 of the number of free pointers available in the active
nodes. The amortized cost of an operation (insertion or deletion) is simply the number
of created nodes minus the change in the potential function, as far as the storage is
concerned. Note that the potential is always positive, or zero for an empty structure.
The total number of nodes created is bounded above by the total amortized cost. The
amortized cost of copying is null, and to add a pointer to a node has an amortized cost
of k 1 1 Therefore, the amortized cost (in storage) of an insertion or deletion is 0(1).

Basic data structures are treated in any book on algorithms and data structure. The
reader is invited to refer to Aho, Hopcroft, and Ullman [6], Knuth [142], Sedgewick [200],
Cormen, Leiserson, and Rivest [72], or Froidevaux, Gaudel, and Soria [108] (in French).
The stratified tree (see exercise 2.3) is due to van Emde Boas, Kaas, and Zijlstra [216].
Our exposition is taken from Mehlhorn [163] and Mehlhorn and Nhher [166]. Segment
and range trees (exercises 2.1 and 2.2) are discussed in the book by Preparata and
2.5. Bibliographicalnotes 31
Shamos [192]. Persistent data structures as described in exercise 2.7 are due to Sarnak
and Tarjan [196]. Several geometric applications of persistent trees will be given in the
exercises of chapter 3.
The perfect dynamic hashing method (see exercise 2.6) was developed by Dietzfel-
binger, Karlin, Mehihorn, auf der Heide, Rohnert, and Tarjan [84] and the augmented
dictionary on a finite universe is due to Mehlhorn and Nhher [165]. See also the book by
Mehlhorn [163] for an extended discussion on hashing.
Chapter 3
Deterministic methods
used in geometry
The goal of this and subsequent chapters is to introduce the algorithmic methods
that are used most frequently to solve geometric problems. Generally speaking,
computational geometry has recourse to all of the classical algorithmic techniques.
Readers examining all the algorithms described in this book from a methodolog-
ical point of view will distinguish essentially three methods: the incremental
method, the divide-and-conquer method, and the sweep method.
The incremental method is perhaps the method which is the most largely em-
phasized in the book. It is also the most natural method, since it consists of
processing the input to the problem one item at a time. The algorithm initiates
the process by solving the problem for a small subset of the input, then maintains
the solution to the problem as the remaining data are inserted one by one. In
some cases, the algorithm may initially sort the input, in order to take advantage
of the fact that the data are sorted. In other cases, the order in which the data are
processed is indifferent, sometimes even deliberately random. In the latter case,
we are dealing with the randomized incremental method, which will be stated
and analyzed at length in chapter 5. We therefore will not expand further on the
incremental method in this chapter.
The divide-and-conquermethod is one of the oldest methods for the design of
algorithms, and its use goes well beyond geometry. In computational geometry,
this method leads to very efficient algorithms for certain problems. In this book
for instance, such algorithms are developed to compute the convex hull of a set of
n points in 2 or 3 dimensions (chapter 8), the lower envelope of a set of functions
(chapter 16), a cell in an arrangement of segments in the plane (exercise 15.9), or
even the Voronoi diagram of n points in the plane (exercise 19.1). In this chapter,
the principles underlying the method are outlined in section 3.1, and the method
is illustrated by an algorithm that has nothing to do with geometry: sorting a
sequence of real numbers using merging (the so-called merge-sort algorithm).
3. 1. The divide-and-conquer method 33
The sweep method, in contrast to the divide-and-conquer method, is deeply

linked with the geometric nature of the problems discussed thereafter. In 2 di-
mensions, numerous problems can be solved by sweeping the plane with a line.
Computing the intersection points of a set of n line segments in the plane is a
famous example of this kind of problem. In higher dimensions (3 and more),
sweeping the space with a hyperplane often reduces a d-dimensional problem to a
sequence of (d - 1)-dimensional problems. The sweep method is described in sec-
tion 3.2, and is exemplified by the problem of computing the intersection points
of a set of n line segments in the plane. Uses of the sweep method can be found
in chapter 12 as well, with the triangulation of a simple polygon, in chapter 15
for computing a single cell in an arrangement of line segments (exercises 15.8 and
15.9), and in chapter 19 for computing the Voronoi diagram of a set of points, or
of a set of line segments.
Of course, there are other methods which are intimately related to the geo-
metric nature of the problem at hand, and some are discussed and used in this
book as well. A common instance is the technique of geometrically transforming
a problem using some kind of polarity or duality, which converts a problem into
a dual problem. In chapter 7, we show using this method that computing the
intersection of a set of half-spaces is equivalent to computing the convex hull
of a set of points. In chapter 17, a geometric transform is shown that reduces
the computation of a Voronoi diagram to that of the intersection of half-spaces.
Another characteristic of geometric algorithms is that they sometimes depend on
decomposing the complex geometric objects that they process into elementary
objects. These objects can be stored using a constant number of memory units
and are handled more easily. Decompositions into simplices, or triangulations,
are discussed in chapters 11, 12, and 13. Section 3.3 describes the vertical de-
composition of a set of line segments in the plane. This decomposition refines
the subdivision of the plane induced by the segments, by decomposing each cell
into elementary trapezoids. Computing this decomposition serves as a running
example throughout chapter 5 to exemplify the design of randomized algorithms.
Such a decomposition can be seen as the prototype in a series of analog struc-
tures, also called vertical decompositions, which are of use in order to decompose
various shapes of the d-dimensional Euclidean space Ed into elementary regions.
For instance, vertical decompositions of polygons, of polyhedra, and of arrange-
ments of various surfaces (hyperplanes, simplices, curves, or even two-dimensional
surfaces) are presented throughout this book.
3.1 The divide-and-conquer method

3.1.1 Overview
The divide-and-conquer paradigm is a modern application of the old political
saying: "divide your enemies to conquer them all". Solving a problem by the
34 Chapter 3. Deterministic methods used in geometry
divide-and-conquer method involves recursive applications of the following

scheme:
Dividing. Divide the problem into simpler subproblems. Such problems have
a smaller input size, that is, if the input data are elementary, the input to
these problems is made up of some but not all of the input data.
Solving. Separately solve all the subproblems. Usually, the subproblems are
solved by applying the same algorithm recursively.
Merging. Merge the subproblem solutions to form the solution to the original
problem.
The performance of the method depends on the complexities of the divide and
merge steps, as well as on the size and number of the subproblems. Assume that
each problem of size n is divided into p subproblems of size n/q, where p and q
are some integer constants and n is a power of q. If the divide and merge steps
perform O(f(n)) elementary operations altogether in the worst case, then the
time complexity t(n) of the whole algorithm satisfies the recurrence
t(n) = p t (-) + f(n).
Usually, the recursion stops when the problem size is small enough, for instance
smaller than some constant no. Then k = [logq(n/no)] is the depth of the
recursive calls (logq stands for the logarithm in base q), and the recurrence solves
to
t(n) =0 ( +Epf
k-1 3
In this expression, the first term corresponds to the time needed to solve all
the elementary problems generated by the algorithm. The second term reflects
the time complexity of all the merge and divide steps taken together. If f is a
multiplicative function, i.e. such that f(xy) = f(x)f(y) (which in particular is
true when f(n) = n' for some constant a), then t(n) satisfies
t(n) = O )
or even, noting that n = qlogn/logq,
t(n) = E (nlogp/ log + n1ogf(q)/logq Z P)))

3.1. The divide-and-conquer method 35
If n is no longer assumed to be a power of q, it can nevertheless lie between

two consecutive powers of q. The above analysis applies to the powers of q
that bracket n and hence shows that t(n) obeys the same asymptotic behavior.
Summarizing, if f is a multiplicative function:
* If p > f (q), then t(n) = 0 (nlo0P/ Iogq).
* If p= f (q), then t(n) = 0 (nlogP/ logqlog n), and further if f (n) = n", then
t(n) = O(n log n).
* If p < f (q), then t(n) = 0 (n1og f (q)/1ogq), and further if f (n) n', then
t(n) = 0 (n').
3.1.2 An example: sorting n numbers using merge-sort

Recall that the problem of sorting a finite sequence X = (x 1, . . .,Ixn) of n real
numbers represented as, say, a list, consists of building a list Y formed of all the
elements of X sorted in ascending order. The merge-sort algorithm involves the
following stages:
Dividing. The list representing the sequence X is divided into two sublists that
represent the sequences X1 = (xl,... ,xm) and X2 = (xm+l,...,xn), with
m = Ln/2j.
Solving. Each of the sequences X1 and X2 is sorted recursively using the same
method. Let Y1 and Y2 stand for the two sequences resulting from the
recursive sorting of the subsequences X1 and X2 respectively.
Merging. The sequence Y that solves the sorting problem on X can be obtained
by merging the sequences Yi and Y2 in the following manner. We simul-
taneously go through the corresponding lists, maintaining a pointer to the
current element of each list. To start with, the current element of a list
is its first element. As we skip through, we compare the current elements
of both lists, and append the smaller at the end of the list representing
Y. The corresponding pointer is advanced to the next element in that list,
which becomes the new current element of that list.
We can perform the dividing stage in linear time, and the merging stage also
since each step involves making only one comparison and advances one element
forward in one of the two lists for a maximum total of 2n comparisons. Therefore
f (n) = ()(n) and p = q = 2 since the original list is divided into two sublists
of approximately equal size. From the preceding subsection, we can conclude
that the complexity of sorting n real numbers using merge-sort is E3(n log n).
Theorem 1.2.1 proves that this complexity cannot be improved by more than a
constant factor. The following theorem summarizes this result:
Theorem 3.1.1 The merge-sort algorithm sorts a sequence of n numbers in op-

timal E(n log n) time.
3.2 The sweep method

3.2.1 Overview
A sweep algorithm solves a two-dimensional problem by simulating a sweep of
the plane with a line. Let us agree on a fixed direction in the plane, say that
of the y-axis, which we will call the vertical direction. A line A parallel to this
direction sweeps the plane when it moves continuously from left to right, from
its initial position x = -oo to its final position x = +oo.
Algorithms that proceed by sweeping the plane can in fact be very dissimilar.
Their main common feature is the use of two data structures: one structure Y
called the state of the sweep and another X called the event queue. Though the
information stored in Y can vary from one algorithm to another, the following
characteristics are always true:
1. the information stored in Y is related to the position of the sweep line, and
changes when this line moves,
2. the structure Y must be modified only at a finite number of discrete posi-

tions, called events,
3. the maintenance of this structure yields enough information to build the

solution to the original problem.
The event queue X stores the sequence of events yet to be processed. This
sequence can be entirely known at the beginning of the algorithm, or discovered
on line, i. e. as the algorithm processes the events. The sweep algorithm initializes
the structure Y for the leftmost position x = -x of the sweep line, and the
sequence X with whatever events are known from the start (in increasing order of
their abscissae). Each event is processed in turn, and Y is updated. Occasionally,
new events will be detected and inserted in the queue X, or, on the contrary, some
events present in the queue X will no longer have to be processed and will be
removed. When the event is processed, the queue X gives access to the next
event to be processed.
When all the events are known at the start of the algorithm, the queue X may
be implemented with a mere simply linked list. However, when some events are
to be known only on line, the event queue must handle not only the minimum
operation, but also queries, insertions, and sometimes even deletions: it is a
priority queue (see chapter 2).
3.2. The sweep method 37
The choice of the data structure Y depends on the nature of the problem and
may be handled through multiple components. More often than not, each of these
components must handle a totally ordered set of objects, and the corresponding
operations: query, insertion, deletion, sometimes even predecessor or successor.
The appropriate choice is that of a dictionary, or an augmented dictionary (see
chapter 2).
The sweep method can sometimes be useful in three or more dimensions. The
generalization consists of sweeping the space Ed by a hyperplane perpendicular
to the xd-axis. The state of the sweep is stored in a data structure Y associated
with the sweep hyperplane, and the set of events is the set of positions of the
sweep hyperplane at which the state of the sweep Y changes. The data structure
Y often maintains a representation of a (d - 1)-dimensional object contained in
the sweep hyperplane. The sweep method in higher dimensions, therefore, often
consists of replacing a d-dimensional problem by a sequence of (d- 1)-dimensional
problems.
3.2.2 An example: computing the intersections of line segments

Let S be a set of line segments in the plane, the intersecting pairs of which we are
interested in computing, together with the coordinates of each intersection point.
The naive solution to the problem is to test all the n(n - 1)/2 possible pairs. The
resulting algorithm is therefore quadratic, i.e. runs in E(n 2 ) time. This is optimal
in the worst case, since the number of intersecting pairs of a set of n line segments
can be as high as Q(n 2 ). However, the algorithm performs Q(n 2 ) computations
regardless of what the actual set of line segments is, whereas the number of
intersection points is commonly much less than n2 . In those cases, an output-
sensitive algorithm is more desirable (see section 1.1.2). The algorithm we present
next possesses such a property: its running time complexity is 0 ((n + a) log n)
where a is the number of intersecting pairs in the set of line segments.
To simplify the description of the algorithm, let us assume that the line seg-
ments are in general position, which in this case amounts to saying that no three
segments have a common intersection. Moreover, let us assume that all the end-
points have distinct abscissae. In particular, there can be no vertical line segment
in the set S. Should these assumptions be violated, it would be easy but tiresome
to take special care of all the exceptions, and we leave the technicalities to the
careful reader.
The algorithm is based upon the following remark: if two segments S and S'
intersect, any vertical line A whose abscissa is close enough to that of S n S'
intersects both S and S', and these segments are consecutive in the vertically
ordered sequence of all the intersections of a segment in S and A (see figure 3.1).
The sweep algorithm stores in the data structure Y the set of segments of S
Figure 3.1. Computing the intersections of a set of line segments using the sweep method.
which intersect the vertical sweep line A. Such segments are said to be active at
the current position of the sweep line. The structure Y stores the active segments
in the order of the ordinates of their intersection point with the line A. The order
of the sequence, or the sequence itself, is modified only when the line sweeps over
the endpoint of a segment or over an intersection point.
1. If A sweeps over the left endpoint of a line segment S (that is to say, the
endpoint with the smaller abscissa), this segment S is added to the structure
Y.
2. If A sweeps over the right endpoint of a line segment S (that is to say, the
endpoint with the greater abscissa), this segment S is removed from the
structure Y.
3. If A sweeps over the intersection of two segments S and S', these segments
S and S' switch their order in the sequence stored in Y.
The set of events therefore includes the sweep line passing over the endpoints
of the segments of S, and over the intersections. The abscissae of the endpoints
are known as part of the input, and we wish to compute the abscissae of the
intersection points. A prospective intersection point I is known when two active
segments become consecutive in the sequence stored in Y. The corresponding
event is then stored in the event queue X. The state of the event queue is shown
for a particular position of A on figure 3.1: each event is marked by a point on
the x-axis.
At the beginning of the algorithm, the queue X stores the sequence of endpoints
of the segments in S ordered by their abscissae. The data structure y is empty.
., I
3.2. The sweep method 39
As long as there is an available event in the queue X, the algorithm extracts

the event with the smallest abscissa, and processes it as follows.
Case 1. the event is associated with the left endpoint of a segment S. This
segment is then inserted into Y. Let pred(S) and succ(S) be the active
segments which respectively precede and follow S in Y. If pred(S) and S
(resp. S and succ(S)) intersect, their intersection point is inserted into X.
Case 2. the event is associated with the right endpoint of a segment S. This seg-
ment is therefore queried and removed in the structure Y. Let pred(S) and
succ(S) be the active segments which respectively preceded and followed
S in Y. If pred(S) and succ(S) intersect in a point beyond the current
position of the sweep line, this intersection point is queried in the structure
X and the corresponding event is inserted there if it was not found.
Case 3. the event is associated with an intersection point of two segments S
and S'. This intersection point is reported, and the segments S and S' are
exchanged in Y. Assuming S is the predecessor of S' after the exchange,
S and its predecessor pred(S) are tested for intersection. In the case of
a positive answer, if the abscissa of their intersection is greater than the
current position of the sweep line, this point is queried in the structure X
and the corresponding event is inserted there if it was not found. The same
operation is performed for S' and its successor succ(S').
To prove the correctness of this algorithm, it suffices to notice that every in-
tersecting pair becomes a pair of active consecutive segments in Y, when the
abscissa of the sweep line immediately precedes that of their intersection point.
This pair is always tested for intersection at this point, if not before, therefore
the corresponding intersection point is always detected and inserted into X, to
be reported later.
It remains to see how to implement the structures Y and X. The structure Y
contains at most n segments at any time, and must handle queries, insertions,
deletions, and predecessor and successor queries: it is an augmented dictionary
(see section 2.1). If this dictionary is implemented by a balanced tree, each
query, insertion, and deletion can be performed in time O(log n), and finding
predecessors and successors takes constant time.
The event queue X will contain at most O(n + a) events, if a stands for the
number of intersecting pairs among the segments in S. This structure must
handle queries, insertions, deletions, and finding the minimum: it is a priority
queue (see section 2.1). Again, a balanced binary tree will perform each of these
operations in O(log(n + a)) = O(log n) time.
The global analysis of the algorithm is now immediate. The initial step that
sorts all the 2n endpoints according to their abscissae takes time O(n log n). The
structure X is initialized and built within the same time bound. Next, each of the
2n + a events is processed in turn. Each event requires only a constant number
of operations to be performed on the data structures X and Y and is therefore
handled in time 0(log n). Overall, the algorithm has a running time complexity
of 0((n + a) log n) and requires storage 0(n + a).
The algorithm can be slightly modified to avoid using more than 0(n) storage.
It suffices, while processing any of cases 1 to 3, to remove from the event queue
any event associated with two active but non-consecutive segments. In this way,
the queue X contains only 0(n) events at any time, and yet the event immediately
following the current position of the sweep line is always present in X. Indeed,
this event is associated either with an endpoint of a segment in S, or with two
intersecting segments which therefore must be consecutive in X. Some events
can be inserted into and deleted from X several times before they are processed,
but this does not change the running time complexity of the algorithm, as the
above scheme can be carried out using only a constant number of operations in
the data structures X and Y at each step.
Theorem 3.2.1 The intersection points of a set of segments in the plane can be
computed using the sweep method. If the set of n segments in general position
has a intersecting pairs, the resulting algorithm runs in 0((n+a)logn) time and
0(n) space.
3.3 Vertical decompositions

In this section we describe the vertical decomposition of a set of (possibly inter-
secting) line segments in the plane. A set S of segments induces a subdivision of
the plane into regions, or cells, which are the connected components of E2 \ S.
The complexity of each cell, that is, the number of segment portions appearing
on its boundary, is unbounded. The vertical decomposition of a set of segments
is obtained by subdividing each cell into elementary trapezoidal regions. This
decomposition of the plane can be considered as the prototype of a whole class
of analogous geometric decompositions, similarly called vertical decompositions,
presented in the remainder of this book.
3.3.1 Vertical decompositions of line segments
The vertical decomposition of a set of line segments in the plane is a structure

which depends upon the choice of a particular direction. Here we assume this
direction is that of the y-axis, which we call the vertical direction. When we want
to refer to this direction explicitly, we speak of a y-decomposition.
3.3. VeTtical decompositions 41
I I I I I i I
I I I I
I I I
I IPI
I I
I I I I
I I I I I I I I I I I I
i I I I I i I I I I I I
I I
I I
I I I
(a) (b)
Figure 3.2. (a)The vertical decomposition Dec(S) of a set of line segments S in the plane.
(b) Its simplified decomposition Dec, (S).
Let S be a set of n segments in the plane. As previously, we suppose that the

segments in S are in general position (meaning that no three segments have a
common intersection) and that the abscissae of their endpoints are all distinct.
In particular, this implies that no segment of S is vertical.
From each point P in the plane, we can trace two vertical half-lines both upward
and downward, A\1 p and A 2 P. Let Pi (i = 1, 2) be the first point of Aip distinct
from P where this half-line meets a segment of S. Should no such point exist, we
make the convention that Pi is the point at infinity on the line Aip. Segments
PP1 and PP2 are the walls stemming from point P. Hence, the walls stemming
from a point P are the maximal vertical segments that have P as an endpoint
and whose relative interiors do not intersect segments of S (see figure 3.2a).
We call vertical decomposition of the set S of segments, and we denote by
Dec%,(S), or more simply by Dec(S) when the vertical direction y is clearly un-
derstood, the planar subdivision induced by the segments and the vertical walls
stemming from the endpoints and from the intersection points of the segments
in S (see figure 3.2a). The vertical decomposition of S can be described as a
planar map whose vertices, edges, and regions subdivide the plane. The vertices
of this map are the endpoints and intersection points of the segments of S, and
the endpoints of the walls. Their number is O(n + a), if S has n segments with a
intersecting pairs. Edges are either walls, or pieces of segments between consecu-
tive vertices. Their number is therefore also O(n +a). Euler's theorem for planar
maps (see exercise 11.4) shows that the number of regions is also O(n + a). Each
region in the map has the shape of a trapezoid, the two parallel sides of which
are vertical. Some degenerate ones are triangular (with only one vertical side),
or semi-infinite (bounded at top or bottom by a segment portion with two semi-
infinite walls on both sides), or doubly infinite (a slab bounded by two vertical
lines on either side), or even a half-plane (bounded by only one vertical line).
It is easy to modify the above algorithm to compute not only the intersection
points, but also the vertical decomposition of the given set of line segments.
Theorem 3.3.1 A sweep algorithm builds the vertical decomposition of a set of

n segments in general position with a intersecting pairs in time O((n + a) log n)
and space O(n + a).
3.3.2 Vertical decompositions and simplified decompositions

Each region of a vertical decomposition is thus a trapezoid, or a degenerate one,
and its boundary has at most four sides', two of which are vertical. Each vertical
side of a trapezoid consists of one or two walls stemming from the same point.
The non-vertical sides of a trapezoid are respectively called the floor and ceiling
of the trapezoid. The floor or ceiling of a trapezoid is always included in some
segment of S and its endpoints are vertices of Dec(S). Neither the floor nor the
ceiling need be edges of the vertical decomposition Dec(S), however: they can be
made up of several (up to Q(n)) edges of the planar map Dec(S). Indeed, several
walls exterior to a trapezoid can butt against its floor or its ceiling, as is the case
for the shadowed cell in figure 3.2a. With this understanding, the boundary of
a region can be made up of Q(n + a) edges of the planar map Dec(S), and thus
have a non-bounded complexity.
From a slightly different point of view, we can describe a simplified decompo-
sition scheme for which each trapezoidal region has a bounded complexity. The
trick is to consider each segment as an infinitely thin rectangle. The boundary
of such a rectangle is made up of two vertical sides of infinitely small length and
two sides which are copies of the segment. The latter are called the sides of the
segment (see figure 3.2b). Each side of the segment completely ignores how the
walls abut against the opposite side, in other words the two sides of a segment
are not connected. The floor and ceiling of a trapezoid are included in sides of
two distinct segments. The simplified vertical decomposition of the set S can still
be viewed as a vertical decomposition of the plane. Simply, in addition to the
trapezoidal regions, there are (empty) regions included in the rectangles between
the two sides of a segment. The trapezoidal regions are unchanged, except that
they now have at most six edges on their boundaries: the floor and ceiling are
each made up of a single edge, and there can be up to two walls per vertical side.
'Here, as in the previous subsection, the word side is used in its usual geometric meaning: a
quadrangle is a geometric figure with four sides.
3.4. Exercises 43
3.4 Exercises
Exercise 3.1 (Union, intersection of polygonal regions) By a polygonal region,
we mean a connected area of the plane bounded by one or more disjoint polygons (a
polygonal region may not always be simply connected, and may have holes). Show how
to build the union or intersection of k polygonal regions using a sweep algorithm. Show
that if the total complexity of the regions (the number of sides of all the polygons that
bound it) is n, and the number of intersecting pairs between all the sides of all the
polygonal regions is a, the algorithm will run in 0((n + a) logn) time.
Exercise 3.2 (Detecting intersection) Show that to test whether any two segments
in a set S intersect requires at least time £Q(n log n). Show that the sweep algorithm can
be modified to perform this test in time 0 (n log n).
Exercise 3.3 (Computing the intersection of curved arcs) Modify the sweep al-
gorithm described in subsection 3.2.2 so as to report all the intersection points in a family
of curved arcs. The arcs may or may not be finite. We further assume that any two arcs
have only a bounded number of intersection points, which may be computed in constant
time.
Hint: Do not forget to handle the events where the arcs have a vertical tangent.
Exercise 3.4 (Arbitrary sets of segments) Sketch the changes to be made to the
sweep algorithm so that it still works on arbitrary sets of segments, getting rid of the
assumptions about general position. The algorithm should run in time O((n + a) log n)
where a is the number of intersecting pairs.
Exercise 3.5 (Location in a planar map) A planar map of size n is a planar subdi-
vision of the plane E2 induced by a set of n segments which may intersect only at their
endpoints. To locate a point in the planar map is to report the region of the subdivision
that this point lies in. Show that a data structure may be built in time 0(n log n) and
space 0(n) to support location queries in time O(log n).
Hint: The vertical lines passing through the endpoints of the segment divide the plane
into vertical strips ordered by increasing abscissae. The segments that intersect a strip
form a totally ordered sequence inside this strip, and two sequences corresponding to two
consecutive strips differ only in a constant number of positions. A sweep algorithm may
use persistent structures (see exercise 2.7) to build the sequence of such lists.
Exercise 3.6 (Union using divide-and-conquer) Consider the n polygonal regions

interior to n polygons. If any two such polygons intersect in at most two points, it can
be shown that the union of these polygonal regions has complexity O(n). Show that, in
this case, it can be computed in O(n log 2 n) time.
Hint: The algorithm proceeds by using the divide-and-conquer method. Each merge
step computes the union of two polygonal regions and can be performed using the sweep
method. Each intersection between the edges of these regions is a vertex of their union,
therefore there can be at most a linear number of such intersections.
Exercise 3.7 (Selecting the k-th element) Let S be a set of n elements, all belong-
ing to a totally ordered universe. A k-th element of S is any element S of S such that
there are at most k - 1 elements in S strictly smaller than S and at least k elements
smaller than or equal to S. Show that it is possible to avoid sorting S yet still compute
a k-th element in time 0(n).
Hint: The algorithm is as follows:

select(k, S)
if ISI < 50, sort S and return any k-th element
otherwise
1. Divide S into [i5.L subsets of 5 elements each, with at most 4
elements in an additional set.
2. Compute the median of each subset, and the median M of the
set M of all medians of these subsets by recursively calling
Select( 2o )
3. Let S,, S2, S3 consist of those elements of S respectively smaller

than, equal to, or greater than M.
if IS1, > k return Select(k,S,).
if IS,1 + IS21 > k return Select(k - ISI - IS21,S3).
otherwise return M.
To analyze the complexity of such an algorithm, observe that IS1 < 3n and that
IS21 < 3n. Then, if n > 50, show that the time complexity of the algorithm satisfies the
recurrence
t(n) > t(n/5) + t(3n/4) + cn,
where c is a constant. Show that this recurrence solves to t(n) = 0(n).
Exercise 3.8 (Union of parallel rectangles) Consider a set of axis-parallel rectan-

gles in the plane E2 . Propose an algorithm to compute the area, the perimeter, or even
the boundary of the union of these rectangles.
Hint: One may use a sweep algorithm that maintains the intersection of the union of the
rectangles with the sweep line, using a segment tree (see exercise 2.1). The perimeter or
the area can be obtained in time O(n log n), and the complete description of the boundary
in time 0((n + k) log n) if this boundary has k edges.

The sweep algorithm that computes the intersecting pairs of a set of n segments in the
plane is due to Bentley and Ottmann [23]. This output-sensitive algorithm is not optimal,
as the problem has complexity e(n + a) if, between the n segments, a pairs intersect.
Indeed, the output has size Q(n + a) and the result of exercise 3.2 shows that Q(n log n)
is also a lower bound on the complexity of the problem. Chazelle and Edelsbrunner [49]
give an optimal algorithm that computes, given a set of n segments with a intersecting
3.5. Bibliographical notes 45
pairs, the induced vertical decomposition in optimal O(n log n + a) time. In degenerate
cases, the number b of intersection points can be much lower than a, and Burnikel,
Mehlhorn and Schirra [40] have shown that it still is possible to compute the vertical
decomposition in O(n log n + b) time. In chapter 5, we describe a randomized algorithm
(that is, an algorithm which makes random choices during its execution) which runs in
time 0 (n log n + a) on the average over all possible random choices it can make.
Persistent data structures and the idea of using them for locating a point in a planar
map (as in exercise 3.5) are due to Sarnak and Tarjan [196]. Segment trees (see exer-
cise 2.1) are especially suitable for solving many problems on rectangles. The solution
to exercise 3.8 can be found in the book by Preparata and Shamos [192].
Chapter 4
Random sampling
The randomization method has proved useful in computational geometry. This

usefulness can be ascribed in large part to a few probabilistic theorems which
rely on combinatorial properties of certain geometric problems. The probabilities
involved in those theorems concern random samples from the set of data, and do
not involve statistical assumptions about the distribution of these data.
The goal of this chapter is to present those probabilistic theorems on which
the analysis of the randomized incremental method is based. This method is
described in chapters 5 and 6. We express these theorems in a framework gen-
eral enough to be adaptable to different geometric settings. All the randomized
algorithms presented below fit into the same framework as the one that we define
here.
The first part of this chapter recalls the necessary definitions and notation.
The second part proves the basic two theorems: the sampling theorem, and the
moment theorem. These theorems provide the main tools to analyze the average
performance of randomized algorithms.
4.1 Definitions
4.1.1 Objects, regions, and conflicts
In the framework presented here, any geometric problem can be formulated in
terms of objects, regions, and conflicts between these objects and regions.
Objects are elements of a universe 0, usually infinite. The input to some
problem will be a set S of objects of 0. The objects under consideration are
typically subsets of the Euclidean space Ed such as points, line segments, lines,
half-planes, hyperplanes, half-spaces, etc.
A region is a member of a set F of regions. Each region is associated with two
4.1. Definitions 47
sets of objects: those that determine it, and those that conflict with it.
The set of objects that determine a region is a finite subset of (9, of cardinality
bounded by some constant b. The constant b depends on the nature of the
problem, but not on the actual instance nor on its size. This restriction is required
for all the probabilistic theorems to be expressed within the framework.
The set of objects that conflict with a given region is usually infinite and is
called the domain of influence of the region.
Let S be a set of objects. A region F of F is defined over S if the set of objects
that determines it is contained in S. A region F is said to be without conflict
over S if its domain of influence contains no member of S, and otherwise is said
to have j conflicts over S if its domain of influence contains j objects of S.
For each geometric application, the notions of objects, regions, and conflicts
are defined in such a way that the problem is equivalent to finding all the regions
defined and without conflict over S.
Let us immediately discuss a concrete example. Let S be a set of n points
in the d-dimensional Euclidean space Ed. The convex hull of S is the smallest
convex set containing S; suppose we wish to compute it. Assume the points are in
general position'. The convex hull conv(S) is a polytope whose special properties
will be studied further in chapter 7. For now, it suffices to notice that, in order
to compute the convex hull, we have to find all the subsets of d points in S such
that one of the half-spaces bounded by the hyperplane passing through these d
points contains no other point that belong to S (see figure 4.1). In this example,
the objects are points, and the regions are open half-spaces in Ed. Every set of
d points determines two regions: the open half-spaces whose boundaries are the
hyperplane passing through these points. A point is in conflict with a half-space
if it lies inside it. To find the convex hull, one must find all the regions determined
by points of S and without conflict over S.
The preceding definitions call for a few comments.
Remark 1. A region is determined by a finite and bounded number of objects
and this restriction is the only fundamental condition that objects, regions, and
conflicts must satisfy. Nevertheless, we do not demand that all the regions be
determined by exactly the same number of objects. In the case of the convex hull
of n points in Ed, all the regions are determined by exactly d points. One may
envision other settings (as in the case of the vertical decomposition of a set a line
segments in the plane, discussed in subsection 5.2.2), where the regions can be
determined by a variable number i of objects, provided that 1 < i < b for some
constant b.
Remark 2. A region does not conflict with the objects that determine it. This
'A set of points is in general position if every subset of k + 1 < d + 1 points is affinely
independent, or in other words if it generates an affine subspace of dimension k.
48 Chapter 4. Random sampling
j.*. . . *.
* 0
Figure 4.1. Convex hull and empty half-spaces.
simple convention greatly simplifies the statements and proofs of the theorems
below, and does not modify their meaning. In the case of the convex hull, this
can be easily achieved by defining the domain of influence of a region as an open
half-space.
Remark 3. A region is characterized by two sets of objects: the set of objects
that determine it, and the set of objects that conflict with it. Regions determined
by different objects will be considered as different, even if they share the same
domain of influence. In this context, a set S of objects is in general position
precisely if any two regions determined by different subsets of S have distinct
domains of influence.
Remark 4. A set of b or fewer objects may determine one, or more, or zero
regions. Usually, the number of regions determined by a given set of (less than b)
objects is bounded by a constant. For instance, in the case of convex hulls, every
subset of d points determines exactly two regions. In this case, the total number
of regions defined over a set of cardinality n is 0(nb).
If S is a finite set of objects, say with n elements, we denote by J'-(S) the set
of regions defined over S and, for each integer j in [0, n], we denote by Fj(S)
the set of all regions defined over S that have j conflicts over S. In particular,
.Fo(S) is the set of those regions that are defined over S and without conflict over
S. Furthermore, we denote by ;F-k(S) the subset of regions defined over S that
have at most k conflicts over S.
When the regions are determined by a variable number i of objects (I < i < b),
the preceding notation may be refined to denote by Ej(S), P<k(S)7 Yik(S), the
subsets of those regions defined by exactly i objects of S, with (respectively)
exactly, at most, at least, k conflicts with the objects of S.
4. 1. Definitions 49
* F-
*i
Figure 4.2. Instances of regions.
4.1.2 Random sampling
Let 7? be a subset of S with cardinality r. This subset 7 is a random sample of

S if its elements are randomly chosen among all those of S, such that each subset
is equally likely to be chosen with probability 1/ (n). In what follows, we shall
call such a subset a random r-sample of the set S.
The notation defined in the previous subsection is valid over any subset 7? of S.
In particular, F(1R) is the set of regions defined over 7, FFj (R) is the set of regions
defined over 7? that have j conflicts over 7?, and F (7?) is the set of of regions
defined by exactly i objects of 7? that have j conflicts over 7?.. Since we may also
be interested in the conflicts over S of a region defined over 7?, or the converse,
we will avoid any ambiguities by setting up a special terminology. Henceforth,
by a region defined and without conflict over 7, we shall mean a region defined
over 7? and without conflict over 7?; these are the regions of Fo(7?). Likewise,
a region defined and with j conflicts over 1? is a region defined over 7? and that
has j conflicts over 7?; these are the regions of Fj (7).
In figure 4.2, the points of the subset 7? are enclosed by squares, the half-space
F+ belongs to F 6 (S) and to Fo(lR), while F- belongs to Fio(S) and to F3 (1Z).
From now on, we are primarily interested in the regions defined over a random
sample 7 from S. Generally speaking, if g(1R) is a function of the sample 7,
we denote by g(r, S) the expected value of g(7Z) for a random r-sample of S. In
particular, the following functions are defined: We denote by fj(7?.) the number
of regions defined and with j conflicts over a subset R of S (in mathematical
notation, fj(1?) = lFj(1?)I). Following our convention, fM(r,S) denotes the ex-
pected number of regions defined and with j conflicts over a random r-sample of
S. Likewise, fJ.(7?) stands for the number of regions defined by i objects of 7?
and with j conflicts over 7R (in mathematical notation, fj(1Z) = ~j(R?) I). Then
fjt(r, S) is the expected number of such regions for a random r-sample of S.
4.2 Probabilistic theorems
In this section, we prove two probabilistic theorems, the sampling theorem and
the moment theorem. These two theorems lay the foundations for our analysis
of randomized algorithms as described in chapters 5 and 6. The reader mostly
interested in the algorithmic applications of these theorems may skip this section
in a first reading. In order to understand the results, it would be enough to
memorize the definition of a moment, to look up lemma 4.2.5, and to admit
corollary 4.2.7.
The probabilistic theorems below are based on certain combinatorial properties
of the geometric objects. The probabilities involved concern mainly random
samples from the input data. In particular, these theorems do not make any
assumptions on the statistical distribution of the input data. The theorems are
stated in the formal framework introduced in the preceding section. Nevertheless,
to shape the intuition of the reader, we start by stating them explicitly for the
specific problem of computing the convex hull of a set of points in the plane.
Let S be a set of n points in the plane, assumed to be in general position,
let k be an integer smaller than n and let 7? be a random sample of S of size
r = Ln/k]. The sampling theorem links the number of half-spaces defined over
S and containing at most k points of S, with the expected number of half-spaces
defined and without conflict over 7Z, which is precisely the number of edges of
the convex hull conv(7Z). Let A and B be points of S. Segment AB is an edge of
the convex hull conv(IZ) if and only if A and B are points of 7? and also one the
half-planes HZB and HXB bounded by the line AB does not contain any points
of 7R. The sampling theorem relies on the fact that the segment AB joining two
points of S is an edge of the convex hull conv(IZ) with a probability that increases
as the smallest number of points in either H+B or HiB decreases.
The moment theorem concerns the number of points in S and in its sample 1?
that belong to some half-plane. If the size of 1Z is large enough, the sample is
representative of the whole set, and the number of points of 7? in a half-plane is
roughly the number of points of S in this half-plane scaled by the appropriate
factor r/n.
In fact, the moment theorem is a little more restrictive and concerns only
those half-planes defined and without conflict over the sample. Any edge E of
conv(7Z) corresponds to a region defined and without conflict over 7Z: the half-
plane H-(E) bounded by the line supporting E that contains no point of R?.
The first moment of R relative to S, or moment of order 1, is defined to be the
sum, over all edges E of the convex hull conv(7Z), of the number of points of S
lying inside H- (E). In other words, the moment of order 1 of 7Z with respect to
S counts each point of S \ 7? with a multiplicity equal to the number of edges
of conv(7R) whose supporting lines separate it from conv(7?) itself. Figure 4.3
4.2. Probabilistictheorems 51
Figure 4.3. Moment of order 1.
indicates the multiplicity of each point, and the first-order moment of the sample
is 16.
The moment theorem shows that, if the size of the sample is big enough, the
expected moment of order 1 is at most n - r.
4.2.1 The sampling theorem
The sampling theorem yields an upper bound on the number of regions defined
and with at most k conflicts over a set S of n elements. This bound depends
on the expected number of regions defined and without conflict over a random
Ln/kJ-sample of S. The proof of this theorem relies on the simple idea that,
the fewer objects in conflict with a region, the more likely this region is to have
no conflict with a random sample 1R of S. The proof uses the two fundamental
lemmas below.
Lemma 4.2.1 Let S be a set of n objects and F a region in conflict with j objects
of S and determined by i objects of S. If 7? is a r-sample of S, the probability
P3 ,k(r) that F be a region defined and with k conflicts over 7? is
pi k(r) = (
V k
r
Proof. Let 7? be a random r-sample of S. The region F of 4j(S) belongs to

Pk (7) if it is determined by i objects in 7? and conflicts with k objects in 7?. For
this to be the case, the i objects determining F must be part of 7?. The k objects
of 7? conflicting with F must be chosen among the j objects of S that conflict
with F. Finally, the r - i - k remaining objects in 7? not in conflict with F must
be chosen among the n - i - j objects in S that do not determine or conflict with

F. 0
We denote by pf (r) the probability pj 0 (r) that a region F of Fj(S) be defined

and without conflict over a random r-sample of S:
r -i-j
pi (r)iJ
Lemma 4.2.2 Let S be a set of n objects and 7Z a random r-sample of S. The

expected number fZ(r, S) of regions determined by i objects of Rt that conflict with
k objects of 7? is given by the formula
n-oin--j
fk(r, S) = , P3(S) I kJ r- kJ
j=o ( )
Proof. The expected number of regions in the set rk(IZ) is the sum, over all the
regions determined by i objects of S, of the probability that this region belongs
to the set P7k(7Z). This probability is given by the lemma 4.2.1 above. 0
Theorem 4.2.3 (Sampling theorem) Let S be a set of n objects and k an

integer such that 2 < k < b. Then
.F<k(S)I < 4(b+ 1)b kfo( Ln/kJ ,S).
where b is an upper bound on the number of objects that determine a region,

I2<k(S)I is the number of regions defined and with at most k conflicts over S,
and fo (In/k, S) is the expected number of regions defined and without conflict
over a random Ln/k] -sample of S.
Proof. For each i, 1 < i < b, we shall prove the following inequality bounding
the number of regions determined by i objects:
lprk (S) I < 4(b + 1)'k'fo'([nlkj S).
Then the theorem can be easily proved by summing over all the values of i between
1 and b.
Let k be an integer such that 2 < k < b and 1 a random sample of S, of

size r = [n/k]J. From lemma 4.2.2, we recall that the expected number A (r,S)
of regions defined and without conflict over 1R is
n-i (nij)(nik
fo (r, S) = E|() ( n) ' lk (S) ( )
The remainder of this proof is a mere computation on factorials, which shows

that for each k such that 2 < k < n , and r = Ln/kJ,
( n-i-k) 1
r-i > 1
n A 4(b + 1)iki
(r)
Indeed,
(( r-i
n)
J r!
(r -i)!
(n-i)! (n-r)! (n-i-k)!
n! (n-r-k)! (n - i)!
rJ
We compute
(n-r)! (n-i-k)! > (n-r-k+1 k
(n-r-k)! (n-i)! - in-i-k+1J
> (n-n/k-k+1)k
- ~~ nk - J
> (1 - l/k)k
> 1/4 (if 2 < k),
and
r! (n -i)! dr-l jjjtr+l-l
(r-i)!
- n! 1=0
n-I - 1=1
n
fn/kl > b)
> ;( nb > (1n)
> ki (b + )i (ifk b+l)'

proving the inequality stated by the theorem. 1:1
Remark 1. The sampling theorem deals with the numbers I.F<k(S)I of regions
with at most k conflicts, for values of k between 2 and b
For the case of regions without or with at most one conflict, however, it is
possible to prove the following bound
F I(S)I
I-Fo(S)I I< < In 2(S)| < 4(b+ 1)2 bfo(Ln/21 ,S),
valid whenever n > 2(b + 1).
Moreover, for values of k close to n, there is always the trivial bound
IY<k(S)I < I-F(S)l = O(nb)
if, as in remark 4 of subsection 4.1.1, we suppose that each subset of size at most
b determines at most q regions, for a constant number q that depends on the
interpretation of objects and regions.
Remark 2. The sampling theorem yields a deterministic combinatorial result
when an upper bound on fo( [n/k , S) can be derived. For instance, in chapter 14,
we will use an upper bound on the number of faces of a d-dimensional polytope
to yield, via the sampling theorem, an upper bound on the number of faces at
level at most k in an arrangement of hyperplanes.
The following corollary is very useful for analyzing the average performance of
randomized algorithms. It shows that the expected number of regions defined and
with one or two conflicts over a random r-sample of a set S is of the same order
of magnitude as the expected number of regions defined and without conflict over
such a sample.
Corollary 4.2.4 Let S be a set of n objects, with n > 2(b + 1). For each integer
r such that n > r > 2(b + 1), we have
fl (r, S) < 3fo( Lr/2j, S)
f 2 (r,S) < /fo(Lr/2J S)
where fj (r, S) is the expected number of regions defined and with j conflicts over
a random r-sample of S, and 3 is the real constant
/3 = 4(b + 1)b2b.
Proof. Let 7t be a subset of S of size r, such that 2(b + 1) < r. Applied to 7?,
remark 1 following theorem 4.2.3 yields
.F1(1Z)I < 4(b+ 1)b 2 bfo(Lr/2J ,1?).
The first inequality is obtained by taking expectations on the two members of
this equation. Indeed, fo ( Lr/2J , 7) is the expected number of regions defined and
without conflict over a random Lr/2]-sample of 7?, and the expectation of this
expected number when 1R itself is a random r-sample of S is simply fo( [r/2J , S).
The second inequality can be proved in much the same way. D
4.2.2 The moment theorem
Let S be a set of n objects and 71 be a subset of S. The moment theorem bounds

the total number of conflicts between the objects of S and the regions defined
and without conflict over 1Z.
Let k be a integer less than or equal to n. The moment of order k of 7z with
respect to S, denoted by mk(7Z, S), is the sum
mk(RS
= E (IS(F)l I
FEFo (7?)
where To7o(7Z) stands for the set of regions defined and without conflict over 7z,
and IS(F) is the cardinality of the set S(F) of objects in S that conflict with a
region F.
The moment of order 0, mo(1Z, S), is simply the number of regions defined and
without conflict over IZ:
mo(PZ,S) - lFo(R)I
The moment of order 1, ml(7Z, S), is the total number of conflicts between the
elements of S and the regions defined and without conflict over 7z:
ml(7Z,S) = E IS(F)I.
FEro(R)
The expectation of mk(Z, S) for a random r-sample 7Z of S is denoted by
mk(r, S). In particular, mo(r, S) = fo(r, S).
Lemma 4.2.5
b n-i
Mk(r, S) =I.F ji(S) I ( jk ) PJ~(r).
i=l j=O
Proof. Recall that p (r) stands for the probability that a given region F of
Yj (S) be defined and without conflict over a random r-sample of S, whence
mk(r, S) = b ni ( j )p.(r).
i=1 i=O FEFti(S)
Theorem 4.2.6 (Moment theorem) Let S be a set of n objects. The expec-

tation mk(r, S) of the moment of order k of a random r-sample of S is related to
the expected number fk(r, S) of regions defined and with k conflicts over a random
r-sample of S by the relation
mk (r, S) < fk(r, S) (n r + k)! (r- b -k)!
(n - r)! (r- b)!
where each region is determined by at most b objects.
Proof. According to the previous lemma 4.2.5, and to lemma 4.2.1 which gives
the expression for the probability p1 (r), we have
b n-iji
mk(r,S) = b S)
(t j (n)
(Ik n)
t-
- (n) - (n-j-r)! (r-i)!
< (n - r + k)! (r -b-k)!

(n -r)! (r -b)' lj=
__
IP3S)
(
) (i- ) (
(n
)
As proved by the same lemma 4.2.1, however, the factor
k n-
rA i- k
nA
rJ
is nothing else but the probability A(r)
Pk that a region F of PFJ(S) belong to
.k (7?), whence
mk(r, S) < fk(r, S) (n -r + k)! (r- b -k)!

(n -r)! (r -b)!
Corollary 4.2.7 Let S be a set of n objects. There exists a real constant -y and
an integer ro, both independent of n, such that for each n > r > ro,
ml(r,S) < -Y fo(Lr/2 ,-S)
M2(rS) < -,. 2)fo(Lr/21 ,S),
where mk(r, S) is the expected number of the k-th moment of a random r-sample
of S, and fo(r, S) is the expected number of regions defined and without conflict
over a random r-sample of S.
Proof. For k = 1, the moment theorem yields
ml(r,S) < fi(r,S) - br

4. 3. Exercises 57
and the upper bound is a consequence of corollary 4.2.4. The second inequality
can be proved very much the same way. E
4.3 Exercises
Exercise 4.1 (Backward analysis) In this exercise, regions are determined by at most
b objects of a set S. Let fi (r,S) be the expected number of regions defined and without
conflict over a random r-sample of S. Corollary 4.2.4 to the sampling theorem proves that
fi (r, S) = O(fo (r, S)). Backward analysis can be used to prove this without invoking the
sampling theorem.
Let 7? be a subset of S of cardinality r, and Jo (r -1, ?) the expected number of regions
defined and without conflict over a random sample of 1 of size r - 1. Show that
fo(r -1,) < !Ii(7?)I

rr + IFo(7Z)1 (4.1)
fo(r-1, R) > -F1 (7)I b (4.2)
r r
From this, show that fi(r,S) = O(fo(r,S)). Similarly, show that f 2 (r,S) = O(fo(r,S).
Hint: Backward analysis consists in observing that a random (r - 1)-sample 7?' of 1Z can
be obtained by removing one random object from R. Any region in Fo(7?') is defined
over RZ and belongs either to Fo (R) or to Fl (7). A region F that belongs to Fo (7)
determined by i objects is a region of Fo(7?') if the removed object is not one of the i
objects that determine F; this happens with probability r A region F that belongs to
FF1 (7) is a region of Fo (7') if the removed object is precisely the one that was removed
from 7?, which happens with probability 1. To show that fi(r,S) = O(fo(r,S)), it
suffices to take expectations in equation 4.2 over all r-samples of S and to assume that
fo(r, S) is a non-decreasing function of r.
Exercise 4.2 (The moment theorem, using backward analysis) Let R be a ran-
dom r-sample of a set S of n objects, and 0 a random object of S \ R?. Show that the
expected number of regions defined and without conflict over 1Z but conflicting with 0
is 0 ( $ fl(r + 1,S)). From this, show that the expected value ml(r,S) of the moment
of order 1 with respect to S of a random r-sample is O(n~r fl(r + 1,S)). From this,
deduce an alternative proof of the moment theorem by using the result of the previous
exercise or corollary 4.2.4 to the sampling theorem.
Hint: Note that 7? U {O} is a random (r + 1)-sample of S and that a region of Fo(7?)
that conflicts with 0 is a region of Fl (1Z U {0}) that conflicts with 0.
Exercise 4.3 (An extension of the moment theorem) A function w is called con-
vex if it satisfies, for all x, y in R and all a in [0, 1],
w(ax + (1 -a)y) > aw(x) + (1 - a)w(y).

We are interested in regions determined by at most b objects of a set S of n objects. For

each subset 7? of S and any convex function w, we define:
Wk(Z) = E (w(IS(F)I))
FE.FO()
where To(RZ) is the set of regions defined and without conflict over 7R and IS(F)l is the
number of objects in S that conflict with F. Let wk (r, S) stand for the expected value
of Wk(Z) for a random r-sample of S. Show that
wkr,) (n -r -k)! (r -b -k)! fk(r, S)\
Wk(r, S)<fo(r,S) w( (n - r)! (r - b)! fo(r, S)
Exercise 4.4 (Non-local subset of regions) We still work with the framework of ob-
jects, regions, and conflicts, each region being determined by at most b objects. In this
exercise, we are mostly interested, for a subset 1Z of objects in S, in a subset g0 (1?) of
regions defined and without conflict over R. The definition of g0 (Rz) is not necessarily
local, however: a region F of Fo (7) belongs to go (1) depending on all the elements of
7, not only those in conflict with F or that determine F. Nevertheless, suppose that the
subsets of the form g0 (1R) satisfy the following property: If F is a region of g0 (7Z), 7?'
a subset of 7?, and if 7t' contains the elements that determine F, then F is a region of
Let wk(r, S) be the expected value of the sum
E IS(F) Ik
FE 5 (7)
where IS(F)l is the number of objects of S in conflict with F. We are interested in

showing the moment theorem for the regions in 90(R), in other words that
4
wk(r,S) = o ( go(rS))
where go(r, S) is the expected number of regions in go(1R) for a random r-sample of S.
Hint: 1. Let p(r, F) be the probability that F be a region of go0(R) for a random r-sample
1 of S. Show that, for all t < r <n,
p(rF) < (r! (t- b) At
2. Let us propose an incremental algorithm to compute 5o(5). The probability that a

region F appear in go(7) precisely at step r is
b
-p(r, F).
r
The probability that it disappear from go(7Z) at the next step r + 1 is at least
nS(F)- p(r,F).
4.3. Exercises 59
Show that, for all r1 < r2,
r2b 2,)Sr + I
p(ri,F) + E -p(r,F) >p(rF).
rI+1 rl
3. Using the previous inequality, show by induction on k that
Vr,r/2 < ro < r EEy<

t-ro
Wk (t, S) • rr T go(r,S)
and that
Wk (r, S) < 'Yk go (r, S),
rk
where -y and -Yk are constants depending only on k.
Exercise 4.5 (Tail estimates) Let b be the maximum number of objects that deter-
mine a single region. Suppose again that a set of at most b objects determine at most
q regions, q being a constant, or that the number of regions determined by a set S of n
objects is 0 (nrb).
1. Let S be a set of n objects and R a random r-sample of S. Let a be a real constant
in ]0, 1[. Let 7ro(a, r) denote the probability over all samples 7? that some region defined
and without conflict over R have at least Fan] conflicts with S. Show that, for r big
enough,
tro(a, r) = 0 (r(1- n)r).
2. Show that for any constant A > b, the probability 7ro(A log r/r, r) that some region
F, defined and without conflict over X, have at least An log r/r conflicts with S decreases
to 0 as r increases.
Exercise 4.6 (Extension of the previous tail estimates) We propose to generalize

the tail estimates given in exercise 4.5. Again, let b be the maximum number of objects
that determine a single region, and suppose that the number of regions determined by a
set S of n objects is 0 (nrb).
Let S be a set of n objects, 1Z a random r-sample of S, a a real constant in ],1[,
and m a positive integer. Denote by 7rm(a, r) the probability that there exists a region
F defined over R with at most m conflicts over 1?, and at least [anl conflicts over S.
Likewise, denote by 7r+ (a, r) the probability that there exists a region F defined over 7?
with at least m conflicts over 7Z, and at most [an] conflicts over S.
Show that if the size r of the sample is big enough while still smaller than V'//2, then
if m <a(r -b), 7r (a),r)= O(r

() ( r.a3i, - a)r-3]
if m > a(r - b), 7i1 (a, r) = (1 (rb)

- - a)b [IZ(r) 3 j(1 -a
Then show that, if a(r) = Alog r/r and m(r) = log r/ log log r,
lii
l r -0 (a ()
m(r) r)) = 0.
Exercise 4.7 (An upper bound on fo(S)) Consider the set F(S) of regions defined
over a set S, each region being determined by at most b objects. Let fj (S) be the number
of regions defined and having j conflicts with S, and fo(n) be the maximum of fo(S) over
all sets S of n objects. Suppose that there is a relation between the number of regions
defined and without conflict over S on one hand, and the number of regions defined over
S and conflicting with one element of S on the other. Suppose further that this relation
is of the type
cfo(S) < f1 (S) + d(n) (4.3)
where c is an integer constant and d(n) a known function of n. Let t = b -c. Show then
that
fo(n) = ( 0+) )
In particular,
fo(n) = 0(nt) if d(n) = O(nt') for t' < t,

fo(n) = o(ntlogn) if d(n) =O(nt ),
fo(n) = 0(nt') if d(n) = O(nt') for t' > t.
Hint: Combining equation 4.2, written for a random (n-1)-sample of S, and equation 4.3
yields
n-b+
n__
cfo(S) = n-foS
n
+bfo
n
n -b 1
< -fo(S) +-(fi(S) + d(n))
n n
1
< fo(n- 1,S) + -d(n).
n
Exercise 4.8 (Union of parallel hypercubes) Consider a set of parallel hypercubes

in Ed, that is, hypercubes whose sides are parallel to the axes.
Show that the union of n hypercubes has at most O(nfd/21) faces for each d > 1.
Furthermore, show that the complexity of the union of n hypercubes of equal size is
O(nLd/2J) when d > 2 and remains O(n) in dimension 1.
Hint: Each vertex of the union belongs to a bounded number of faces of the union. Hence
it suffices to bound the number of vertices of the union to bound the total complexity.
The proof works by induction on d. The proof is trivial in dimension 1, and easy in
dimension 2.
In dimension d, each cube has 2d pairwise parallel facets. Let us denote by Fjt(C)
the facet of the cube C that is perpendicular to the xi-axis with maximal j-coordinate,
and by Fj- (C) the facet of the cube C that is perpendicular to the xj-axis with minimal
4.3. Exercises 61
j-coordinate. Let C be a set of axis-parallel cubes in Ed, and denote by U(C) the union of
these cubes and A(C) their arrangement, that is, the decomposition of Ed induced by the
cubes (see part IV for an introduction to arrangements). Each vertex of U(C) or of A(C)
is at the intersection of d facets of cubes, one perpendicular to each axis direction. Such
a vertex P is denoted by (Ci', C22,..., Cd") if at the intersection of facets F'j (Cj), for
=1,.. , d and ej = + or-. The vertex P is called outer if it belongs to a (d - 2)-face
of one of the cubes (then not all the cubes Cj are distinct). It is called an inner vertex if
it is at the intersection of d facets of pairwise distinct cubes. A vertex of A(C) is at level
k if it belongs to the interior of k cubes of C. The vertices of the union are precisely the
vertices at level 0 in the arrangement A(C). Let Wk (C) be the number of inner vertices of
A(C) at level k, and Vk(C) be the number of outer vertices at level k, and vk(n, d) (resp.
Wk(n, d)) the maximum of Vk(C) (resp. of 'Wk(C)) over all possible sets C of n axis-parallel
hypercubes in Ed.
1. The maximum number vo(n,d) of outer vertices of the union is 0(nrd/2l) (and
0(nLd/2]) when the cubes have same size). Indeed, any outer vertex of U(C) belongs to
a (d - 2)-face H of one of the cubes in C and is a vertex (either outer or inner) of the
union of all (d - 2)-cubes C n aff(H), where aff(H) is the affine hull of H. Consequently,
vo (n, d) < 2nd(d - 1)(bo(n -1, d -2) + woo(n - 1,d- 2)),
where 6O(n - 1, d - 2) and bo(n - 1, d -2) respectively stand for the maximum numbers
of outer or inner vertices in the union of n - 1 cubes in a (d - 2)-dimensional space lying
inside a given (d - 2)-cube.
2. Applying the sampling theorem (theorem 4.2.3) and its corollary 4.2.4, we derive a
similar bound on the maximum number v, (n, d) of outer vertices at level 1.
3. To count the number of inner vertices, we use the following charging scheme. For
each vertex P = (C11 2, ,Cdd) of U(C), and each direction j = 1,.. .d, slide along
.C2
the edge of A(C) that lies inside the cube Cj (this edge is nij FjiE (Ci)) until the other
vertex P' of this edge is reached.
If P' belongs to the facet F7Ej (Cj) of cube Cj, we do not charge anything. This case
cannot happen unless the cubes have different side lengths and Cj is the smallest of the
cubes intersecting at P.
If P' belongs to a (d - 2)-face of one of the cubes Ci (i 34 j) intersecting at P, P' is an
outer vertex at level 1, and is charged one unit for P. Note that P' cannot be charged
more than twice for this situation.
If P' belongs to another cube C' distinct from all the Ci intersecting at P, then P' is
an inner vertex at level 1, and is charged one unit for P. Any inner vertex P' of this type
may be charged up to d times for this situation. However, when it is charged more than
once, say m times, we may redistribute the extra m - 1 charges on the outer vertices at
level 0 or 1, and these vertices will only be charged once in this fashion.
In the case of cubes with different sizes, the induction is
(d - 1)wo(C) < wi(C) + 3v 1(C) + vo(C).
In the case of cubes with identical sizes, we obtain
dwo(C) < wi(C) + 3v1 (C) + vo(C).
It suffices to apply exercise 4.7 to conclude.

Randomized methods revolutionized computational geometry. Most of the material in
this chapter is taken from the ground-breaking work of Clarkson and Shor [71]. The
randomized incremental algorithms in the next two chapters are concrete applications
of the formalism developed in this chapter, and we invite the reader to consult the
bibliographical notes of these chapters for more references. Clarkson and Shor proved the
tail estimates and their extension as stated in exercises 4.5 and 4.6, which are the corner
stone on which all analyses of randomized divide-and-conquer algorithms rely. In their
article, they also prove the extension to the moment theorem proposed in exercise 4.3.
This extension will prove useful in exercise 5.8 for the analysis of an algorithm that
triangulates a simple polygon, due to Clarkson, Cole, and Tarjan [69].
The extension of the moment theorem to a non-local set of regions defined and with-
out conflict over a random sample (exercise 4.4) is due to de Berg, Dobrindt, and
Schwarzkopf [76]. The result stated in exercise 4.4 will be used in chapter 15 to an-
alyze the randomized incremental algorithm that builds a single cell in an arrangement
of line segments.
The backward analysismethod proposed in exercises 4.1 and 4.2 was used by Chew [59]
to analyze an algorithm that builds the Voronoi diagram of a convex polygon (see ex-
ercise 19.4). The method was used later in a systematic fashion by Seidel [203] and
Devillers [80].
The method used in exercise 4.7 to obtain an upper bound on the expected number
of regions defined and without conflict over a set of objects is due to Tagansky [212].
The analysis of the complexity of the union of parallel hypercubes in d dimensions (see
exercise 4.8) given by Boissonnat, Sharir, Tagansky, and Yvinec [34] illustrates the power
of this method.
Chapter 5
Randomized algorithms
A randomized algorithm is an algorithm that makes random choices during its

execution. A randomized algorithm solves a deterministic problem and, what-
ever the random choices are, always runs in a finite time and outputs the correct
solution to the problem. Therefore, only the path that the algorithm chooses to
follow to reach the solution is random: the solution is always the same. Most
randomized methods lead to conceptually simple algorithms, which often yield a
better performance than their deterministic counterparts. This explains the suc-
cess encountered by these methods and the important position they are granted
in this book. The time and space used when running a randomized algorithm de-
pend both on the input set and on the random choices. The performances of such
an algorithm are thus analyzed on the average over all possible random choices
made by the algorithm, yet in the worst case for the input. Randomization be-
comes interesting when this average complexity is smaller than the worst-case
complexity of deterministic algorithms that solve the same problem.
The randomized algorithms described in this chapter, and more generally en-
countered in this book, use the randomized incremental method. The incremental
resolution of a problem consists of two stages: first, the solution for a small subset
of the data is computed, then the remaining input objects are inserted while the
current solution is maintained. An incremental algorithm is said to be random-
ized if the data are inserted in a deliberately random order. For instance, sorting
by insertion can be considered as a randomized incremental method: the current
element, randomly chosen among the remaining ones, is inserted into the already
sorted list of previously inserted elements.
Some incremental algorithms can work on line: they do not require prior knowl-
edge of the whole set of data. Rather, these algorithms maintain the solution to
the problem as the input data are successively inserted, without looking ahead at
the objects that remain to be inserted. We refer to such algorithms as on-line or
semi-dynamic algorithms. The order in which the data are inserted is imposed
64 Chapter 5. Randomized algorithms
on the algorithm, and such algorithms cannot properly be called randomized, as

their behavior is purely deterministic. Nevertheless, we may be interested in the
behavior of such an algorithm when the insertion order is assumed to be random.
We may then speak of the randomized analysis of an on-line algorithm.
In the first section of this chapter, the randomized incremental method is
sketched in the framework introduced in the previous chapter, with objects, re-
gions, and conflicts. The underlying probabilistic model is made clear. At any
step, a randomized incremental algorithm must detect the conflicts between the
current object and the regions created previously. One way of detecting these
conflicts is to use a conflict graph. Algorithms using a conflict graph must have
a global knowledge of the input and are thus off-line. Another way is to use an
influence graph. This structure leads to semi-dynamic algorithms and allows the
objects to be inserted on-line. The conflict graph is described in section 5.2 and
the influence graph in section 5.3. In both cases, the method is illustrated by
an algorithm that builds the vertical decomposition of a set of line segments in
the plane. This planar map was introduced in section 3.3, and in particular one
can deduce from it all the intersecting pairs of segments. Both methods lead to
a randomized algorithm that runs in time O(n log n + a) on the average, where
a is the number of intersecting pairs of segments, and this is optimal. Finally,
section 5.4 shows how both methods may be combined and, sometimes, lead to
accelerated algorithms. For instance, we show how to decompose the set of line
segments forming the boundary of a simple polygon in time O(n log* n) on the
average (provided that the order of the edges along the boundary of the polygon
is also part of the input).
We give several randomized incremental algorithms, for instance to compute
convex hulls (chapter 8), to solve linear programs (chapter 10), to compute the
lower envelope of a set of segments in the plane (chapter 15) or of triangles
in three-dimensional space (chapter 16), or even to compute the k-level of an
arrangement of hyperplanes (chapter 14) or a cell in an arrangement of segments
(chapter 15) or of triangles (chapter 16).
5.1 The randomized incremental method

The problem to be solved is formulated in terms of objects, regions, and conflicts
in the general framework described in the previous chapter. The problem now
becomes that of finding the set .Fo(S) of regions defined and without conflict over
a finite set S of objects. The notation used in this chapter is that defined in
subsections 4.1.1 and 4.1.2.
The initial step in the incremental method builds the set Fo (7Zo) of regions that
are defined and without conflict over a small subset 7Ro of S. Each subsequent
step consists of processing an object of S \ RIo. Let 1? be the set of already
5.2. Off-line algorithms 65
processed objects and let us call step r the step during which we process the r-th
object.
Let 0 be the object processed in step r. From the already computed set of
regions defined and without conflict over 7?, we compute in step r the set of
regions defined and without conflict over 1? U {O}.
* The regions of .Fo(Z) that do not belong to Yo(R U {O}) are exactly those
regions in Fo(QR) that conflict with 0. These regions are said to be killed
by 0, and 0 is their killer.
* The regions of Fo(1Z U {0}) that do not belong to Fo(1Z) are exactly those
regions .Fo(1Z U {0}) that are determined by a subset of R U {0} that
contains 0. These regions are said to be created by 0.
A region created by 0 during step r is said to be created at step r. The set

of regions created by an incremental algorithm consists of all the regions created
during the initial step or at one of the subsequent insertion steps.
The chronological sequence is the sequence of the objects of S in the order in
which they are processed. The probabilistic assumption on which the randomized
analysis of incremental algorithms relies is that the chronological sequence is any
of the n! possible permutations of the objects of S with uniform probability. As
a consequence, the subset of objects R already processed at step r is a random
r-sample of S, and any subset of S is equally likely. The object 0 processed
during step r is a random element of S \ R. Equivalently, it is any element of
7? U {O} with a uniform probability.
5.2 Off-line algorithms

5.2.1 The conflict graph
The first task of each step in an incremental algorithm is to detect the regions of
.Fo(1Z) that conflict with the object 0 to be processed at this step. These are the
regions killed by 0. To check all the regions in Fo(7?) does not lead to efficient
algorithms. Instead, in addition to the set .Fo(R), we maintain a graph called
the conflict graph. The conflict graph is a bipartite graph between the objects of
S \ 7? and the regions of Yo (7). Arcs belong to Fo (7) x (S \ 7?), and are called
the conflict arcs. There is a conflict arc (F, S) exactly for each region F of Fo(7?)
and each object S of S \ 1Z that conflicts with F.
The conflict graph allows us to find all the regions killed by an object 0 in
time proportional to the number of those regions. Each step of the incremental
algorithm must then update the conflict graph. The conflict arcs incident to the
regions killed by 0 are removed and the new conflict arcs incident to the regions
created by 0 are found. The complexity of each incremental step is thus at least
bounded from below by the number of regions that are killed or created during
this step, and by the number of conflict arcs that are removed or added during
this step.
Update condition 5.2.1 (for conflict graphs) A randomized incremental al-

gorithm that uses a conflict graph satisfies the update condition if, at each incre-
mental step:
1. updating the set of regions defined and without conflict over the current
subset can be carried out in time proportional to the number of regions
killed or created during this step, and
2. updating the conflict graph can be carried out in time proportional to the
number of conflict arcs added or removed during this step.
Lemma 5.2.2 Let S be a set of n objects and F a region determined by i objects

of S that has j conflicts with the objects in S.
1. The probability p'j that F be one of the regions created by a randomized

incremental algorithm processing S is
,i. i!j!
(i+ j)!
2. The probability p' (r) that F be one of the regions created by the algorithm
during step r is
p~i(r) rp (r).
In these expressions, p (r) stands for the probability that a region F of JFj(S) be
defined and without conflict over a random r-sample of S, as given in subsec-
tion 4.2.1.
If we replace pj (r) by its expression in lemma 4.2.1, we obtain (see also exer-
cise 5.1) that the probabilities p', and p' (r) satisfy the relation
n
r~1
Proof. A region F of .Fj(S) is created by a randomized incremental algorithm if

and only if, during some step of the algorithm, this region is defined and without
conflict over the current subset. Such a situation occurs when the i objects that
determine F are processed before any of the j objects of S that conflict with F.
Since all permutations of these objects are equally likely, this case happens with
probability
i!j!
(i+j)!'
proving the first part of the lemma. Let 7? be the set of objects processed in the
steps preceding and including step r. For a region F to be created during step
r, we first require that F be defined and without conflict over 1R, which happens
precisely with probability pj (r). If so, F is created at step r precisely if the object
0 processed during step r is one of the i objects of 2 that determine F. This
happens with conditional probability i/r. El
Theorem 5.2.3 (Conflict graph) Let S be a set of n objects, and consider a

randomized incremental algorithm that uses a conflict graph to process S.
1. The expected total number of regions created by the algorithm is
0 ( fo(r, S))
r=1
2. The expected total number of conflict arcs added to the conflict graph by the
algorithm is
0 (n n; Mo r, S))
3. If the algorithm satisfies the update condition, then its complexity (both in
time and in space) is, on the average,
o (nE fo (r, S))
In these expressions, fo(r, S) denotes the expected number of regions defined and
without conflict over a random r-sample of S.
Thus, if fo(r, S) behaves linearly with respect to r (fo(r, S) = O(r)), the total
number of created regions is O(n) on the average, the total number of conflict arcs
is 0(n log n) on the average, and the complexity of the algorithm is 0(n log n) on
the average. If the growth of fo(r, S) is super-linear with respect to r (fo(r, S) =
O(r') for some a > 1), then the total number of created regions is O(n') on
the average, the total number of conflict arc is O(n') on the average, and the
complexity of the algorithm is O(n') on the average.
Proof.
1. We obtain the expectation v(S) of the total number of regions created by the
algorithm by summing, over all regions F defined over S, the probability that
this region F be created by the algorithm:
b n-i b n-i n
V~(S) Ip"
~~~ E JP
E|(S)| Iz p.(r).
i=1 j=O i=1 j=O r=1
In this expression, we recognize the expected number of regions defined and

without conflict over a random r-sample of S (lemma 4.2.2), so we get
v(S) =O (E A(r S))

r=1
2. Let e(S) be the expected total number of arcs added to the conflict graph.
To estimate e(S), we note that if a region F in conflict with j objects of S is a
region created by the algorithm, then it is adjacent to j conflict arcs in the graph.
Therefore,
bo n-i bo n-i n .
(S) ,E,|er 3(S) i PI';=
i=1 j=o
,E,£,|j(S)l 3lp(r)'
i=1 j=O r=1
Apart from the factor i/r, we recognize in this expression the moment of order 1
of a random r-sample (lemma 4.2.5). Applying corollary 4.2.7 to the moment
theorem, we get
e(A bE ml (r S)
r=l r-
=( r2fo(Lr/2i ,S)) O(nfE r2)
3. A given region is killed or created at most once during the course of the
algorithm and, likewise, a given conflict arc is added or removed at most once. A
randomized incremental algorithm which satisfies the update condition thus has
an average complexity of at most
v(S)+e(S) = ° r2S)
2
Exercise 5.2 presents a non-amortized analysis of each step of an algorithm that

uses a conflict graph.
5.2.2 An example: vertical decomposition of line segments

Let us discuss again the problem of finding all the a intersecting pairs of a set of
n line segments in the plane. The space sweep algorithm presented in chapter 3
solves this problem in time O((n +a) log n), which is suboptimal because a could
be as large as Q(n 2 ). The complexity of the problem is O(n log n + a), and
is matched by a deterministic algorithm (see section 3.5). The algorithm we
present here is randomized incremental, and its expected complexity also matches
the optimal bound. It is much simpler than the deterministic algorithm, and
generalizes easily to the case of curved segments (see exercise 5.5).
This algorithm builds in fact the vertical decomposition of the set of segments
S, or more precisely its simplified decomposition Dec8 (S). This structure is dis-
cussed in detail in section 3.3. Let us simply recall that the vertical decomposition
of S is induced by the segments of S and the vertical walls stemming from their
endpoints and intersection points (see figure 3.2 or figure 5.4 below). The de-
composition of the set S of segments with a intersecting pairs is of size O(n + a),
and can be used to report all the intersecting pairs of S in the same time bound
O(n + a).
To use the preceding framework, we define the set of objects 0 to be the set of
all segments in the plane, the universe of regions F being the set of all possible
trapezoids with parallel vertical sides, occasionally in some degenerate state such
as triangles, semi-infinite trapezoids, vertical slabs, or half-planes bounded by
a vertical line. A trapezoid F of F conflicts with a segment S if this segment
intersects the interior of F. A trapezoid F is determined by a minimal set of
segments X, the decomposition of which includes F. Each region in the simplified
decomposition Dec (S) is described using at most four segments of S. Indeed, the
floor and ceiling are each supported by a single segment in S, and each vertical
side consists of one or two walls stemming either from an endpoint of a segment,
or from an intersection of two segments. In the former case, the vertical side
consists either of one vertical wall stemming upward from an endpoint of the
floor, or stemming downward from an endpoint of the ceiling, or of two vertical
walls stemming in both directions from the endpoint of another segment. In
the latter case, one of the intersecting segments is necessarily supporting the
floor or ceiling of the trapezoid. In all cases, we see that four segments suffice
to determine the trapezoid. Three may suffice if the trapezoid is a degenerate
triangle. Partially infinite triangles may be determined by three, two, or even one
segment. The cardinality of the set X that determines a region is thus at most
four, and the constant b that upper-bounds the number of objects that determine
a region is thus set to b = 4.
Therefore, the set of trapezoids in the vertical decomposition of the set S of
segments is exactly the set of those segments defined and without conflict over
S. According to our notation, this set is denoted by Go(S).
l l
I I
Figure 5.1. Instances of trapezoids in the simplified vertical decomposition.
The algorithm
According to the randomized incremental scheme, the algorithm processes the

segments of S one by one in a random order. Again, we denote by 1 the set of
the already processed segments at a given step r. The algorithm maintains the
simplified vertical decomposition Dec (7?) of this set of segments, as well as the
corresponding conflict graph.
The decomposition Dec,(JZ) is maintained in a data structure, that encodes
its simplified description. This structure includes the list of all trapezoids of
Dec8 (R). Each trapezoid is described by the set of at most four segments that
determine it, and by the at most six edges in the simplified decomposition (floor,
ceiling, and at most four vertical walls) which bound it. The data structure also
describes the vertical adjacency relationships in the decomposition, that is the
pairs of trapezoids whose boundaries share a vertical wall. Such an adjacency is
stored in a bidirectional pointer linking both trapezoids. Note that each trapezoid
is vertically adjacent to at most four trapezoids.
The conflict graph has a conflict arc for each pair (F, S) of a trapezoid F of
Dec, (7) and a segment S of S \ 1R that intersects the interior of F. The conflict
graph is implemented by a system of interconnected lists.
* For each segment S of S \ 7?, the data structure stores a list £(S) repre-
senting the set of trapezoids of Dec,(7?) intersected by S. The list L:(S) is
ordered according to which trapezoids are encountered as we slide along S
from left to right.
* For each trapezoid F of Dec8 (7), the algorithm maintains the list £'(F) of
the segments in S \ 7? that conflict with F.
£(S)
L'(F)
Figure 5.2. Representation of the conflict graph used to build the

vertical decomposition of a set of line segments.
* The node that stores a segment S in the list C'(F) of a trapezoid F is

interconnected via a bidirectional pointer with the node that stores the
trapezoid F in the list C(S) of the segment S (see figure 5.2).
In the initialstep, the algorithm builds the decomposition Dec5 (R.) for a subset
7? of S that contains only a single segment. This decomposition consists of four
trapezoids. It also initializes the lists that represent the conflict graph. The
initial decomposition is built in constant time, and the initial conflict graph in
linear time.
The current step processes a new segment S of S \ I7: it updates the decom-
position and the conflict graph accordingly.
Updating the decomposition. The conflict graph gives the list £(S) of all
the trapezoids of Dec5 (1?) that are intersected by S. Each trapezoid is split into
at most four subregions by the segment S, the walls stemming from the endpoints
of S, and the walls stemming from the intersection points between S and the other
segments in R (see figure 5.3).
These subregions are not necessarily trapezoids of Dec, (1 U { S}). Indeed, S
intersects some vertical walls of Dec5 (1?), and any such wall must be shortened:
the portion of this wall that contains no endpoint or intersection point must be
removed from the decomposition, and the two subregions that share this portion
of the wall must be joined into a new trapezoid of Dec,(1Z U {S}) (see figure 5.4).
Thus, any trapezoid of Dec, (7? U {S}) created by S is either a subregion, or
the union of a maximal subset of subregions that can be ordered so that two
consecutive subregions share a portion of a wall to be removed. The vertical
adjacency relationships in the decomposition that concern trapezoids created by
S can be inferred from the vertical adjacency relationships between the subregions
and from those between the trapezoids of Dec,(7R) that conflict with S.
Updating the data structure that represents the decomposition Dec8 (R) can
therefore be carried out in time linear in the number of trapezoids conflicting
with S.
Updating the conflict graph. When a trapezoid F is split into subregions
Fi (i < 4), the list L'(F) of segments that conflict with F is traversed linearly,
and a conflict list V'(Fi) is set up for each of the subregions Fi. During this
72 Captr 5.Ranomied agorthm
I
So
Figure 5.3. Decomposing a set of segments: splitting a trapezoid into at most four new
trapezoids.
traversal, the list £(S') of each segment S' in £'(F) is updated as follows: each
node pointing to F in such a list is replaced by the sequence of those subregions
Fi that intersect S', in the left-to-right order along S'.
Consider now a sequence F 1 , F2 , .. ., Fk of subregions that have to be joined
to yield a trapezoid F' of Dec, (R U {S}) created by S. We assume that the
subregions are encountered in this order along S. To build the list £'(F'), we
must merge the lists L'(Fi) while at the same time removing redundant elements.
To do this, we traverse successively each of the lists V'(Fi). For each segment
S' that we encounter in £'(Fi), we obtain the entry corresponding to Fi in the
list L(S') by following the bidirectional pointer in the entry corresponding to
S' in the list £'(Fi). The subregions that conflict with S' and that have to be
joined are consecutive in the list C(S'). The nodes that correspond to these
regions are removed from the list L(S'), and for each entry Fj removed from
the list £(S'), the corresponding entry for S' in L'(Fj) is also removed. (This
process is illustrated in figure 5.5.) In this fashion, we merge the conflict lists of
a set of adjacent subregions while visiting each node of the conflict lists of these
subregions once and only once. Similarly, the corresponding nodes in the conflict
lists of the segments are visited once and only once. This ensures that the time
taken to update the conflict graph is linear in the number of arcs of the graph
that have to be removed.
Analysis of the algorithm

The preceding discussion shows that the algorithm which computes the vertical
decomposition of a set of line segments using a conflict graph obeys the update
5..Ofln loihs7
(a) (b)
I
.I
: I : . :
: : I .
.I.
(c) (d)
Figure 5.4. Decomposing a set of segments: the incremental construction

(a) The decomposition before inserting segment S.
(b) Shaded, the trapezoids that conflict with segment S.
(c) Splitting those trapezoids. Wavy lines indicate the portions of walls to be
removed.
(d) The decomposition after inserting segment S.
condition 5.2.1. We may therefore quote theorem 5.2.3 to show that the average
running time of the algorithm, given a set of n segments with a intersection
points, is
O fo (rS))
1<r<n
Here, fo (r, S) is the expected number of trapezoids in the vertical decomposi-

tion of a random r-sample of S. The following lemma estimates this number.
L'(F2 )
J'(F3 )
]L'(F4)
Figure 5.5. Decomposing a set of line segments: merging the conflict lists.
Lemma 5.2.4 Let S be a set of n segments, a pairs of which have a non-empty

intersection. The expected number fo(r, S) of trapezoids in the vertical decompo-
sition of a random r-sample of S is O(r + ar2 /n 2 ).
Proof. Let R be a subset of r segments in S, and denote by a(R) the number

of intersecting pairs of segments in 1R. The number of regions in the vertical
decomposition of R? is therefore O(r + a(R.)). If 7R is a random r-sample of
S. however, the expected number a(Z) of intersections is A 1) . ne
intersection point P between two segments of S is an intersection point of two
segments of 7? if the two segments of S that intersect at P belong to 7R, which
happens with probability
(zr-2) (r)
Therefore, we immediately derive the following theorem.
Theorem 5.2.5 The vertical decomposition of a set S of n line segments in

the plane with a intersecting pairs can be obtained by a randomized incremental
algorithm that uses a conflict graph, in expected time O(n log n + a).
Remark 1. The algorithm builds only the simplified representation Dec,(S) of

the decomposition. From this representation, however, we can derive the com-
plete representation Dec(S) of the decomposition in O(n + a) time. Exercise 5.3
shows that the complete representation can be directly computed by a variant of
the preceding algorithm, while still using no more than O(n log n + a) running
time.
5.3. On-line algorithms 75
Remark 2. The expected storage of the algorithm is O(n log n + a). In the
variant mentioned in the above remark, it is possible to simplify the conflict
graph: for each segment, we retain only a single conflict arc, for instance the
conflict with the trapezoid which contains the left endpoint of the segment. We
can still update the conflict graph in linear time, therefore the average running
time is unchanged and still O(n log n +a), but the expected storage is lowered to
O(n + a) (see exercise 5.4).
5.3 On-line algorithms
Algorithms that use a conflict graph are incremental but static, that is, they
require initial knowledge of all the segments to be inserted. In contrast, on-line
(or semi-dynamic) algorithms maintain the solution to the problem as the input
objects are inserted, with no preliminary knowledge of the input data. A possible
way to transform an algorithm that uses a conflict graph into an on-line algorithm
is to replace the conflict graph by a different kind of structure that can detect
conflicts between any object and the regions defined and without conflict over
the current set of objects. The influence graph is such a structure.
5.3.1 The influence graph
The influence graph is a structure that stores the history of the incremental con-
struction and depends on the order in which the objects have been processed
by the algorithm. This graph represents the regions created by the algorithm
during the incremental construction, and can be used to detect the conflicts
between these regions and a new object. When the algorithm uses a conflict
graph, the set of data is known in advance, and the algorithm may then com-
pute the objects in S that conflict with a given region. However, an on-line
algorithm does not assume any knowledge of the objects to be processed. Thus
it must be able to describe the entire domain of influence of a region which, as
we recall, is the subset of all the objects in the universe that conflict with this
region.
The influence graph is a directed, acyclic, and connected graph. It possesses
a single root, and its nodes correspond to the regions created by the algorithm
during its entire history. Therefore, a node corresponds to a region that was
defined and without conflict over the current set of objects at some point during
the execution of the algorithm. Properly speaking, this graph is not a tree: a
node might have several parents. Nevertheless, the terminology of trees will be
quite useful for describing it. In particular, a leaf is a node that has no children.
The influence graph must possess two essential properties.
Property 5.3.1 1. At each step of the algorithm, a region defined and without
conflict over the current subset is associated with a leaf of the influence
graph.
2. The domain of influence of a region associated with a node of the influence
graph is contained in the union of the domains of influence of the regions
associated with the parents of that node.
To shorten and simplify the terminology, we attach qualifiers normally used

for a node to its corresponding region and vice versa. This allows us to speak
of the domain of influence of a node instead of the domain of influence of its
associated region. Likewise, a region has children or parents. This slight abuse
in the terminology should not create any confusion.
The algorithm
The algorithm is incremental and maintains the set Fo(7) of regions defined
and without conflict over the current set RZ, together with the influence graph
corresponding to the chronological sequence of objects in R?.
The initial step processes a small set of ro objects. For instance, ro can be
the minimal number of objects needed to determine a region. The algorithm
computes the regions defined and without conflict over the set Ro of these ro
objects. The influence graph is initialized by creating a root node, corresponding
to a fictitious region whose influence domain is the universe 0 of objects in its
entirety. A node whose parent is the root is created for each of the regions of
F(Ro).
In the current step, the object 0 is added to R?. The work can be divided into
two phases: we first locate 0 and then update the data structures.
Locating. In this phase, we must find all the regions in Fo(1Z) that conflict
with the new object 0. Starting from the root of the influence graph, we recur-
sively visit all the nodes that conflict with 0, and their children. The regions of
.To(7?) that conflict with 0 are said to be killed by 0.
Updating. We now have to update the data structure that represents the set
of those regions defined and without conflict over the current subset of objects
(To( Z) becomes Fo(UU{O})). We also have to update the influence graph. A leaf
of the influence graph is created for each of the regions in To(7? U {0}) \ .Fo(R).
These are the regions created by 0. Each of these leaves is linked to enough
parents to satisfy property 2 of the influence graph. We never remove a node
from the graph.
The details of the implementation of these steps naturally depend on the prob-
lem. Typically, the set of regions created by 0 can be derived from the set of
regions killed by 0, and the parents of the leaves corresponding to created regions
may be chosen among the nodes corresponding to regions killed by 0. In this

case, the information needed to carry out the update is gathered in the locating
phase. In some cases, we may need to know some extra information, such as
adjacency relationships between regions. The influence graph is then augmented
with the required information.
Randomized analysis of the influence graph
A randomized on-line algorithm is not a randomized algorithm properly speaking.

Indeed, the order in which the data are processed is imposed on the algorithm, and
the algorithm makes no random choices whatsoever. The algorithm is therefore
perfectly deterministic. Nevertheless, we can analyze the average performance
of such an algorithm (in running time or storage) under the assumption that all
inputs are equally likely or, more precisely, that any permutation of the input
data is equally likely. The performances of the algorithm are valid for any input,
and in particular no assumption is made on the distribution of the input. In
this case, the analysis of the algorithm is said to be a randomized analysis. The
randomized analysis of an on-line algorithm which has currently processed a set
S of n objects assumes that the chronological sequence A, made up from the
objects of S in the processing order, is a random sequence.This means that all
n! permutations of S are equally likely to occur in E. At any step, the current
subset of objects already processed by the algorithm is thus a random sample of
S.
Each node of the influence graph corresponds to a region created at some
step of the algorithm. Such a region was, at this step, defined and without
conflict over the current subset of objects. The set of those regions created by
an incremental algorithm depends only on the order in which the objects are
inserted. In particular, it does not depend on whether the incremental algorithm
uses a conflict graph or an influence graph. The following theorem is thus an
immediate consequence of theorem 5.2.3.
Theorem 5.3.2 Let an on-line algorithm use an influence graph to process a set
S of n objects. The expected number v(S) of nodes in this influence graph is
(E (r, S))
In this expression, fo(r,S) denotes the expected number of regions defined and
without conflict over a random r-sample of S.
To carry the analysis further, we must also be able to bound the number of
arcs in the influence graph, since this number gives the time and storage taken
to update the set of regions without conflict and the influence graph itself, as
is done in the second phase of each insertion step of the algorithm. We also
need a special assumption to control the complexity of testing whether there is
a conflict between an object and a region.' The triple update condition stated
below is actually satisfied by a large class of practical problems.
Update condition 5.3.3 (for influence graphs) An on-line algorithm that

uses an influence graph satisfies the update condition if:
1. the existence of a conflict between a given region and a given object can be
tested in constant time.
2. the number of children of each node of the influence graph is bounded by a
constant, and
3. the parents of a node created by an object 0 are nodes that are killed by 0,
and updating the influence graph takes time linear in the number of nodes
killed or created at each step.
Theorem 5.3.4 (Influence graph) Consider an on-line algorithm that uses an

influence graph to process a set S of n objects. If the algorithm satisfies the update
condition 5.3.3, then:
1. The expected storage used by the algorithm to process the n objects is
o (E fo(r, S))
r=1
2. The expected time complexity of the algorithm is
o (En fo(rS))
3. The expected time complexity of the locating phase at step k is
0 (E fo(r2S))
4. The expected time complexity of the updating phase at step k is
0 (fo(k'S) + fo(L(k -1)/2j ,S))
1 Note
that such an assumption is implicitly contained in the update condition 5.2.1 when
the algorithm uses a conflict graph.
As always, fo(r, S) denotes the expected number of regions defined and without
conflict over a random r-sample of S.
Thus, the expected time complexity of an on-line algorithm that uses an influence
graph is identical to that of a similar incremental algorithm that uses a conflict
graph, as long as the respective update conditions are satisfied.
If fo(r, S) behaves linearly with respect to r (fo(r, S) = 0(r)), the complex-
ity of the algorithm is O(nlogn) on the average, and the expected storage is
O(n). Introducing the n-th object takes time O(logn) for the locating phase,
and constant time for updating the data structure and the influence graph.
If the growth of fo(r, S) is super-linear with respect to r (fo(r, S) = 0(r 0 ) for
some a > 1), the expected storage is O(n'). Introducing the n-th object takes
time 0(n'l) for the locating and updating phases.
Proof.
1. The upper bound on the expected storage is a direct consequence of theo-
rem 5.3.2, which bounds the number of nodes in the influence graph, and of the
second clause in the update condition, which bounds the number of children of
each node.
2. The contribution to the running time complexity of the updating phases is
proportional to the number of regions created, because of the third clause of the
update condition. From theorem 5.2.3, we know that this number is
0 (E fo(r, S)
kr=1
We still must evaluate the cost of the locating phases. From the first clause
of the update condition, we derive that the complexity of locating an object 0
is proportional to the number of nodes visited to locate 0. If every child has a
constant number of descendants (second clause in the update condition), however,
the number of nodes visited during the locating phase of 0 is at most proportional
to the nodes of the influence graph that conflict with 0. The overall cost of the
locating phases is therefore proportional to the total number of conflicts detected
during the algorithm.
Let F be a region of P.(S). If this region is created at some step of the al-
gorithm, the corresponding node in the influence graph will be visited j times
in the subsequent steps, and this happens each time we insert one of the j ob-
jects that conflict with F. For a given permutation of the input, an algorithm
that uses an influence graph will not only create the same regions as another
that uses a conflict graph, but will also detect a conflict with a given region
as many times as there are conflict arcs incident to this region in the conflict
graph.
As a consequence, the total expected complexity of the locating phases is pro-

portional to the expected number of conflict arcs created in the conflict graph,
and is given by theorem 5.2.3.
3. Finally, we can give a non-amortized analysis of each incremental step of
an on-line algorithm. At step k, the locating phase takes time proportional to
the number of nodes in the influence graph that conflict with Ok, the object
introduced in this step. The average complexity of this locating phase during
step k is thus proportional to w(k, S), the expected number of nodes in the
influence graph that conflict with Ok. From lemma 5.2.2, we know that a region
F in PFj (S) is created at step r with a probability p'i (r) = rpi (r). The conditional
probability that this region conflict with Ok, knowing that F is created prior to
step k, is j/(n - r). Consequently,
b n-i k-1
w(k,S) = EIv ( E - piZ(r) -
i=1 j=0 r=1 r n-
If we recognize the expression for the first order moment of a random r-sample
of S given in lemma 4.2.5, and bound the sum above by using corollary 4.2.7 to
the moment theorem, we obtain
b k-1
w(k, S) = Za,(r)m l(rS)
(E fo(Lr/2j S)) = (E fo(r.S))
4. Now, updating the data structure and the influence graph at step k takes time
proportional to the number of nodes created or killed by Ok. Let v(k, S) be the
expected number of regions created at step k. From lemma 5.2.2, we derive
b n-i
v(k,S) = ZZY(S) jpp(k)
i=i j=0
fo(k, S)
k
Let now v'(k, S) be the expected number of regions killed at step k. We denote
by Sk-1 the current subset immediately prior to step k. A region F in P, (S) is a
region killed at step k if it is a region of FO(Sk-1) that conflicts with Ok, which
happens with probability
pj (k -1) -+l.
Again, using lemma 4.2.5 and corollary 4.2.7, we get

b n-i
V'(k, S) = YEE IPji(S) I p,~(k - 1)n -k +1
i=1 j=0
mi(k- 1, S)
k-i
- fo( L(k-1)/2J S))
This completes the proof of theorem 5.3.4. E
Some remarks on the update condition
The update condition 5.3.3 is not mandatory and it is often possible to analyze
an on-line algorithm that does not satisfy all of its clauses.
1. For instance, if the first clause is not satisfied, the cost of testing the conflicts
may be added to the analysis. If this cost can be bounded, this bound appears
as a multiplicative factor in the cost of the locating phases.
2. The analyses of on-line algorithms developed above and in the remainder
of this section are still valid for less restrictive statements of the third clause.
We may assume only that the cost of the update phase is proportional to the
number of regions created or killed. We have preferred, however, to assume that
the parents of nodes created by some object are killed by the same object. This
assumption is satisfied by most of the algorithms given in this book, and it greatly
simplifies the analysis of dynamic on-line algorithms given in the next chapter.
3. Lastly, the second clause can also be relaxed. Indeed, in order to bound the
space needed to store the influence graph, it suffices to bound the total number of
arcs in the entire graph and not necessarily the out-degree of each node. We may
then generalize the analysis of the locating phase by using the notion of a biregion
(see exercise 5.7). In particular, such an analysis applies to the case when the
number of parents of a node is bounded, but not the number of children. We
illustrate this situation in the case of the on-line computation of convex hulls (see
exercise 8.5).
5.3.2 An example: vertical decomposition of line segments

Again, we discuss the problem of building the vertical decomposition of a set
of line segments in the plane, and this time we show how to compute it on-
line, using an influence graph. Each time a segment is inserted, the algorithm
updates the decomposition of the current set of segments, called the current
decomposition for short. The notions of objects, regions, and conflicts are defined
as in subsection 5.2.2.
The trapezoids in the current decomposition are the regions defined and with-
out conflict over the current set of segments and are linked to the corresponding
nodes in the influence graph. An internal node of this graph is associated with
a trapezoid which was in the current decomposition at some previous step of the
algorithm. In addition to the set of pointers that take care of the parent-child
relationships between the nodes, each node contains the following information:
* A description of the corresponding trapezoid and a list of the (at most four)
segments that determine it.
* At most four pointers for the adjacency relationships through the vertical
walls. As long as the node is a leaf of the influence graph, the corresponding
trapezoid F belongs to the current decomposition and is adjacent to at most
four leaves in the graph, each of which shares a vertical wall with F. When
the node corresponding to F becomes an internal node, these pointers are
not modified any more.
Therefore, a description of the simplified decomposition can be extracted from

the information stored at the leaves of the graph.
At each step, the new segment S is located in the influence graph, yielding the
list L(S) of trapezoids that it intersects. Each of these trapezoids is subdivided
into at most four subregions by S, by the walls stemming from the endpoints of
S, and by the walls stemming from the intersection points between S and some
previously inserted segment. In the influence graph, a temporary node is created
for these subregions. For each trapezoid F in the current decomposition that is
intersected by S, we create links to the temporary nodes for the subregions Fi in-
side F, and they become the children of F in the influence graph (see figure 5.3).
The list £(S) is not sorted, yet the vertical adjacency pointers allow a traversal
of the decomposition in the left-to-right order along S. This allows us to set
the vertical adjacencies of the subregions, and to identify which walls have to be
removed and which subregions have to be joined to obtain the simplified decom-
position Dec,(R. U {S}) after the insertion of S. We replace all the temporary
nodes that correspond to subregions to be joined by a single node that inherits
all the parents of the subregions. In this fashion, a leaf of the graph is created
for each trapezoid F created by S, and is linked to all the trapezoids F' in L(S)
which intersect the interior of F (see figures 5.6, 5.7, and 5.8). Properties 1 and 2
of the influence graph are thus maintained from step to step. Vertical adjacency
relationships between trapezoids created by S can be derived from those of the
subregions. This completes the description of the update phase in an incremental
step.
|- D
A -
a5D
E F H D
I
I I :
G I:
Figure 5.6. Vertical decomposition and influence graph (1).
Each internal node of the influence graph has at most four children, and the
running time needed to carry out all the operations described in the previous
paragraph is clearly proportional to the number of trapezoids in £(S). The
update condition is therefore satisfied.
From lemma 5.2.4, we know that the expected number of trapezoids in the
vertical decomposition of a random r-sample is O(n + ar2 /n 2 ), if n is the num-
ber of segments in S and a is the number of intersecting pairs of segments.
Theorem 5.3.4 therefore shows that the on-line algorithm just described has an
expected complexity of O(n log n + a) and uses expected storage O(n + a). The
average complexity of the n-th insertion is O(log n + a/n).
E :F A a
D
EFO I
,N . .:
Q
I I
Ir IVP
I IS w s I
oG
l l - l
5.4 Accelerated incremental algorithms

One of the problems encountered in solving a problem using the incremental
method, that is, in identifying the set Fo(S) of those regions defined and without
conflict over a given set S of objects, is the detection of conflicts between a
new object and a region defined and without conflict over the current subset.
Algorithms that use a conflict graph are static, as opposed to on-line algorithms
which use an influence graph. In this section, which may be skipped in a first
reading, we show how to combine both data structures to transform an on-line
algorithm into a static one that has a lower asymptotic average complexity.
5.4.1 The general method

Theorem 5.3.4 on the influence graph shows that, if the expected number of
regions defined and without conflict over S is 0(r), then the complexity of any
algorithm that uses an influence graph is dominated by the cost of the locating
phases in the incremental steps. (This cost is 0(n log n), whereas the cost of the
updates is only 0(n).) The idea is to lower the complexity of the locating phase
5.4. Accelerated incremental algorithms 85
54. Accelerated incremental algorithms 85
[ ' n '
i
I
.1 D
N
, SI Q
I : A , Q .
. I
E: F :AI S U Vaw D
I HI p
I, I -
I I '
X YA
i . l,
by using a conflict graph, in addition to the influence graph. We cannot afford

to maintain this conflict graph at every step, however, so we update it only at
certain steps of the algorithm.
Let Sk be the current subset immediately after step k. The conflict graph at
step k is the conflict graph that stores all the conflicts between the regions in
Fo(Sk) and the objects in S \ Sk. The following theorem shows how to use a
knowledge of this conflict graph at step k to speed up the subsequent locating
phases.
Theorem 5.4.1 If an on-line algorithm satisfies the update condition, a knowl-

edge of the conflict graph at step k can be used to perform the locating phase in
a subsequent step 1, with an average complexity of
(11 fo( Lrl2j , S))
r=k+l
In particular, if fo(r,S) is O(r), the cost of a locating phase at step 1 is

0 (log(k)) on the average.
Proof. The conflict graph at step k can be augmented, for each object 0 in
S \ Sk, by a list of pointers to the nodes of the influence graph which correspond
to a region of FO(Sk) that conflicts with 0. In order to locate the object 01
at step 1, the algorithm may start to traverse the influence graph not from the
root, but from the nodes of the influence graph which correspond to a region of
Fo(Sk) that conflicts with Oj. If the update condition is satisfied, the number
of children of each node is bounded by a constant, and the number of nodes
visited is proportional to the number of regions F created between steps k + 1
and 1 - 1 that conflict with 01. A region F in 4J(S) is created at step r with
probability 'pj(r). Given this, the conditional probability that F conflicts with
a given object 01 is g The expected number w(l,S) of nodes visited while
locating 01 is thus
I-1 b n-b9
w(l,S) = ZE ZYi(S)Ipi(r.i).
r=k+li=lj=O
In this expression we recognize the first order moment (lemma 4.2.5). Using
corollary 4.2.7 to the moment theorem, we obtain
w(lS) = 0 (Mt r(n-r) m(, S)) = 0 ( E r2 l)

r=k+l r~ )r=k+l
The accelerated algorithm proceeds as follows. At each step, the algorithm

updates the influence graph. At certain steps, numbered n 1 , n 2 ,.. ., nk,..., the
algorithm builds the conflict graph between the regions in YO(Snk) and the ob-
jects in S \ Snk. This conflict graph is used to speed up the locating phases in
the steps whose numbers range from (nk + 1) to nk+1
Of course, this design is useful only if the computation of the conflict graph
at steps n1 , n2, ... , nk, ... is not too unwieldy. The following theorem presents a
general situation when an on-line algorithm that runs in expected time 0(n log n)
can be speeded up into a static algorithm that runs in expected time 0(n log* n).
Theorem 5.4.2 Let S be a set of n objects, and 7Z be a random r-sample of

S. Suppose that the expected number fo(r, S) of regions defined and without
conflict over 7Z is 0(r). If the conflict graph at any step k can be built in ex-
pected time 0(n), then the randomized acceleratedalgorithm runs in expected time
0(n log* n).
Proof. The conflict graph is computed at steps nk = [n/ log(k) nj, for k = 1,...,
log* n. The conflict graph is therefore computed log* n times overall, accounting
5.4. Accelerated incremental algorithms 87
for an expected complexity of 0 (n log* n). The locating phases, between step nk
and step nk+1, have a total average complexity of
nk+1 nk+1
E 0(log( )) ( log( log n))

1=flk+1 l1flk+l
= (nk+l - nk)O (log log(k) n)

= O(n).
The total contribution of the locating phases to the running time is therefore, on
the average, O(n log* n). This fact combined with theorem 5.3.4 proves that the
average complexity of the accelerated algorithm is O(n log* n). cl
5.4.2 An example: vertical decomposition of a polygon

The vertical decomposition of a polygon P, which we denote by Dec(P), is the
decomposition Dec(Sp) of the set of line segments Sp that are the edges on the
boundary of the polygon P. The decomposition of a simple polygon is a very
interesting structure since we can derive from it a triangulation of the polygon,
as we explain in chapter 12. The previous method helps to compute the decom-
position of a polygon with n sides in expected time 0(n log* n), which therefore
leads to a randomized algorithm of better average complexity than most of its
deterministic counterparts.
The algorithm processes the segments in Sp in a random order, builds an
influence graph as described in subsection 5.3.2, and maintains the simplified
decomposition of the current set of edges. In accordance with the preceding idea,
it computes a conflict graph at steps nk = [n/ log(k) nJ, k = 1,. . , log* n, which
is then used to locate the subsequent edges between step nk and step nk+l.
The segments in Sp may intersect only at their endpoints. Lemma 5.2.4 shows
that the number fo(r, Sp) of trapezoids in the decomposition of a random r-
sample of Sp is O(r).
The conflict graph at step nk can be computed using the following method
in expected time O(n). Let DecS(S(nk)) be the current decomposition imme-
diately after step nk. A simplified decomposition will suffice for our purposes.
We first begin by locating a given vertex of the polygon in the decomposition
Dec,(S(nk)), using brute force and O(n) operations. We then follow the bound-
ary of the polygon, reporting all the conflicts between the edges and the trapezoids
of the decomposition Dec. (Sf(nk)). Immediately after step nk, an edge has either
been inserted already, or it conflicts with some trapezoids in the decomposition
Dec,(S(nk)). In the former case, it has been split into possibly many edges of
this decomposition, and the total complexity of following these edges is domi-
nated by the size O(nk) = O(n) of this decomposition. In the latter case, the
cost of following these edges is proportional to the number of conflicts between
these edges and the trapezoid of the decompositions Dec,(S(nk)).2 From the-
orem 4.2.6 and its corollary 4.2.7, the expected number of conflicts reported at
step nk is exactly the first order moment of the current subset of edges at step
nk. From corollary 4.2.7, this number is 0 (fo(Lnk/2j , S)), which is 0(n) for
non-intersecting segments, as is the case for the edges of a polygon.
The hypotheses of theorem 5.4.2 are thus satisfied, which yields:
Theorem 5.4.3 A randomized incremental algorithm can build the vertical de-
composition of a simple polygon with n edges in expected time O(n log* n).
Remark. The algorithm relies on two facts: the edges are connected, and do not
intersect except possibly at common endpoints. The same algorithm therefore
works as well in the more general cases of a polygonal line, or a connected set of
segments whose pairwise interiors are disjoint.
5.5 Exercises
Exercise 5.1 (Probabilities) Prove that
ii (n
n-i
i-i-j
J i!j!
E~srr n = (i + j)!
V rr
Then show that the probabilities p' and p'; (r) defined in section 5.2 satisfy the following
relation: n
plX ,pt'(r).
r=l1
Exercise 5.2 (Non-amortized analysis of the conflict graph) Consider a ran-

domized incremental algorithm that processes a set S of n objects by using a conflict
graph. Show that if the update condition is satisfied, then the complexity of step k is on
the average
( k k22fo( L s) + (k- 1)2fo(L 2 s))

2
We may recall that each trapezoid of the decomposition is adjacent to at most four trapezoids
through vertical walls and these adjacencies are encoded through additional pointers in the
influence graph. Since an edge may not intersect a floor or ceiling, we can trace its conflicts in
the current decomposition using constant time per conflict.
5.5. Exercises 89
Hint: The expected number of regions killed or created during step k of a randomized
incremental algorithm is estimated in the proof of theorem 5.3.4. It remains to estimate
the expected number of conflict arcs added or removed during step k.
Exercise 5.3 (Complete description of the vertical decomposition) Let S be a

set of line segments in the plane, with a intersecting pairs, and let It be a random r-
sample of S. We consider the complete description of the decomposition Dec(JZ) of 1Z,
and particularly the expected number of edges on the boundaries of the trapezoids of
Dec(JZ) which are cut by a random segment of S \ R. Show that this expectation is
0(1 + ar/n 2 ).
From this, show that the randomized incremental algorithm described in subsec-
tion 5.2.3 can be slightly modified to compute, within the same complexity bounds,
a complete description of the decomposition of the line segments, which in particular
includes all the adjacency relationships between the trapezoids.
Hint: We may redefine the notions of regions and conflicts as follows: A region defined
on 7t is a paddle with two components, a trapezoid F in the decomposition Dec(JZ), and a
wall butting on the floor or ceiling of F. A paddle is determined by at most six segments.
It conflicts with a segment if the interior of the trapezoid intersects the segment. The
problem is now to find an upper bound on the number of paddles defined and without
conflict with a segment of S \ RZ.
Exercise 5.4 (Storage) Consider the incremental algorithm that uses a conflict graph
as in subsection 5.2.2 in order to compute the decomposition of a set S of n segments.
Show that if a is the number of intersecting pairs, the storage needed by the algorithm
at step k is, on the average,
mi(k, S) =0 (n +a ).
Using the result of the previous exercise, show that we may reduce the storage to O(n)
by storing only one conflict for each non-inserted segment, say with the trapezoid that
contains its left endpoint, without affecting the running time of the algorithm.
Exercise 5.5 (Decomposing a set of curves) Show how to generalize the notion of
a decomposition for a set of curves supported by algebraic curves of bounded degree.
Two such curves intersect at only a constant number of points, which we assume may be
computed in constant time. Show that both algorithms given in subsections 5.2.2 and
5.3.2 may be extended to build the decomposition of a set of such curves.
Hint: Do not forget to trace walls from each point where the curves have a vertical
tangent.
Exercise 5.6 (Backward analysis) Backward analysis (see also exercises 4.1 and 4.2)
gives an alternative proof of the results of this chapter without using the explicit expres-
sions for p. (r) and p.(r).
For instance, we show how backward analysis can be used to estimate the number
v(k, S) of regions created at step k by an incremental algorithm. Note that if Sk is the
current subset immediately after inserting object Ok at step k, the regions created by Ok
during this step are the regions of FO(Sk) determined by a subset of Sk that contains
Ok. Since Ok, which has chronological rank k, may be any of the objects in Sk with
uniform probability k, a region of .FO(Sk) is created at step k with probability at most
b/k. Therefore, v(k, S) is at most the expectation of IFo(Sk)I over all possible Sk.
Similarly, a region that is killed during step k is a region of F1 (Sk) that conflicts with
Ok. Any region of -F (Sk) conflicts with Ok with probability I/k. The expected number
v'(k, S) of regions killed during step k is therefore at most the expectation of k 1F1(Sk)
over all possible Sk.
It is possible to compute in this fashion the expected numbers v(k,S) and v'(k,S)
of regions created or killed during step k. Show how to use backward analysis to
prove the other results in this chapter, for instance, to bound the number of conflict
arcs that are added to or removed from the conflict graph, or to bound the number
of conflicts detected during a locating phase by an algorithm that uses an influence
graph.
Exercise 5.7 (Biregions) The notion of biregion introduced in this exercise can be
used to analyze the average complexity of some algorithms that use an influence graph,
but do not satisfy the update condition 5.3.3. A biregion is pair of regions which can
have a parent-child relationship in the influence graph for at least one permutation of
the data. A biregion is determined by a set of at most 2b objects, those that determine
the parent region and those that determine the child region. Exactly one of the objects
that determine the child region conflicts with the parent region. We can extend the
notion of conflict to biregions in the following way: an object conflicts with a biregion
if it conflicts with at least one of its two regions and does not belong to the set of
objects that determine the biregion. A biregion can then be considered as a region in
the framework described in chapter 4.
1. Let S be a set of n objects. Show that a biregion, determined by i objects of S and
in conflict with j objects of S, is defined and with k conflicts over a random r-sample of
S with the probability p k(r) given by lemma 4.2.1.
2. From this, extend both the sampling theorem and the moment theorem to the case
of biregions.
3. In essence, the difference between biregions and regions resides in the following fact.
Let FF be a biregion determined by i objects and conflicting with j objects of S. For
FF to correspond to an arc in the influence graph built for S, it is not enough that the
i objects that determine FF be inserted before any of the j objects that conflict with
FF; it must also be the case that the i objects that determine i be processed in a certain
order. This order has to meet several criteria. These criteria depend on the algorithm.
At the very least, one of the objects that determine the child region, more precisely the
one that conflicts with the parent region, must be inserted after all the objects that
determine the parent region.
Show that the probability that FF correspond to an arc of the influence graph is eap'j,
where p"t is given
do in lemma
t 5.2.2, and a is a constant
t that satisfies 1 - o< a <
- and
that depends only on the particular criteria that the insertion order has to meet. Then
5.5. Exercises 91
show that the probability that the biregion FF correspond to an arc of the influence
graph that is created at step r is a p. (r), where p3 (r) is defined in subsection 4.2.1.
4. Our goal is now to give a randomized analysis of an on-line algorithm that uses an
influence graph in which a node can have arbitrarily many children. We thus forget about
the second clause in the update condition 5.3.3, and relax the third one by assuming that
the parents of a region created by 0 are either killed by 0, or still have no conflict after
the insertion of 0. In this way, regions defined and without conflict with the current
subset may not be leaves of the influence graph, but could have many children before they
are killed. The complexity of the update phase is still assumed to take time proportional
to the number of arcs added to or removed from the influence graph. For instance, the
algorithm that computes convex hulls described in exercise 8.5 meets these conditions.
Let ffo (r, S) stand for the expected number of biregions defined and without conflict
over a random r-sample of S. Show that the number of arcs in the influence graph built
for S is, on the average,
e (nffo(r, S))
Or= 1
Show that the cost of the locating phases is
0 (n E ffo (r, S) )
5. Assume now that the influence graph built for a random r-sample of S has an
expected number of arcs at most g(r,S), where g is a known function. For instance,
when each node of the influence graph has at most a bounded number of children, we
may choose
g(r, S) = 0( fo(aS))
j=l 3
Show that the n-th incremental step of the on-line algorithm has an average complexity
of
0 g(S) + n (r S)
O n +E
r=1
Exercise 5.8 (Decomposing a polygon) This exercise presents another randomized

algorithm that builds the vertical decomposition of a simple polygon with n edges in
expected time O(n log* n).
The algorithm is incremental but inserts a number of edges of the polygon at a time.
Let P be a polygon, and S the set of its n edges. Assume that the segments in S are
ordered in a random order, and let Si be the subset containing the first ri segments of
S, with ri = [n/ [log(') n1 |. The subset Si is thus a random ri-sample of S, and
S1 C S2 C ... C Slogs n = S-
The algorithm computes a simplified description of the decomposition Dec8 (P), using
log* n steps. Step i computes the decomposition Dec8 (Si) from Dec (Si- 1).
In the initial step, we build Dec,(Si) using the plane sweep algorithm of subsec-
tion 3.2.2, in time O(rjlogrl). (Any algorithm that runs in time O(rilogrl) would
do.)
In a subsequent step i, i > 1:
1. We locate the segments of S in Dec8 (Si-i). In other words, for each region F in
Dec8 (Si-i), we compute the set S(F) of segments in S which intersect F.
2. For each region F of Dec,(Si-I), we compute the decomposition of S(F) USi, and
the portion of it that lies inside F. To do this, we simply compute the total de-
composition Dec, (S(F) U Si), using the plane sweep algorithm of subsection 3.2.2.
(Again, any algorithm that runs in time O(m log m) for m segments would do.)
3. We obtain Dec,(Si) by putting together all the portions Dec,(S(F) U Si) n F
inside the trapezoids F of Dec,( S i-1), and merging the regions that share a wall
of Dec,(Si-1) which disappears in Dec,(Si).
Show that all three phases 1, 2, and 3 can be performed using O(n) operations. To
analyze phase 2, note that Si-, is a random ri-1-sample of Si, then use the extension
of the moment theorem given in exercise 4.3 for the function g(x) = xlogx.
Exercise 5.9 (Querying the influence graph) The influence graph built by an on-
line algorithm can be used to answer conflict queries on a set of objects. For instance, the
influence graph built for a vertical decomposition can answer location queries for a point
inside this decomposition. Show that, if n segments are stored in the influence graph,
answering a given location query takes time O(logn), on the average over all possible
insertion orders of the n segments into the influence graph. More generally, show that
the same time bound holds for any conflict query which, on any subset JZ of objects,
answers with a single region of Fo(7Z).

The first non-trivial (that is, sub-quadratic) algorithm that computes all the intersecting
pairs in a set of segments in the plane is that of Bentley and Ottmann [23], which
uses a plane sweep method. This algorithm, described in chapter 3, computes all a
intersecting pairs in O( (n + a) log n), which falls short of being optimal. About ten years
later, Chazelle and Edelsbrunner proposed in [48, 49] a deterministic algorithm that
runs in optimal time 0(n log n + a) to compute all the a intersections. The description
and implementation of their algorithm is rather complicated, however. At about the
same time, Clarkson and Shor [71] and independently Mulmuley [171, 173] proposed
randomized incremental algorithms for the same problem that have an optimal average
complexity.
The algorithm by Clarkson and Shor that uses a conflict graph is described in sec-
tion 5.2 in this chapter. In the same paper [71], they also set up the formalism of objects,
regions, and conflicts, and introduce the conflict graph in these terms; they give other
algorithms that use the conflict graph (computing the intersection of n half-spaces, the
diameter of a point set in 3 dimensions), and show how to compute a complete descrip-
tion of the decomposition of line segments and how to lower the storage requirements of
their algorithm (see exercises 5.3 and 5.4). Mulmuley's algorithm is very similar to that
of Clarkson and Shor, yet its analysis is based on probabilistic games and combinatorial
series, and is much less immediate.
The influence graph was first introduced in a paper by Boissonnat and Teillaud [31, 32]
where it was called the Delaunay tree, and was used there to compute on-line the
Delaunay triangulation of a set of points. Guibas, Knuth, and Sharir [117] proposed
a similar algorithm to solve the same problem. How to use the influence graph in an
abstract setting is described by Boissonnat, Devillers, Schott, Teillaud, and Yvinec in
[28] and applied to other problems, especially to compute convex hulls or to decompose
a set of segments in the plane. The method was later used to solve numerous other
problems. The influence graph is sometimes called the history of the incremental con-
struction.
The accelerated algorithm that builds the vertical decomposition of a simple polygon
is due to Seidel [204]. This method was subsequently extended to solve other problems
by Devillers [80], for instance to compute the Voronoi diagram of a polygonal line or of
a closed simple polygon (see section 19.2). The algorithm described in exercise 5.8 that
computes the decomposition of a polygon in time O(n log* n) is due to Clarkson, Cole
and Tarjan [69].
The method called backward analysis used in exercise 5.6 was first used by Chew in
[59] to analyze an algorithm that computes the Voronoi diagram of a convex polygon
(see exercise 19.4). It was subsequently used in a systematic fashion by Seidel in [203]
and Devillers in [80].
Mehlhorn, Sharir, and Welzl [167, 168] gave a finer analysis of randomized incremental
analysis by bounding the probability that the algorithm exceeds its expected perfor-
mances by more than a constant multiplicative factor.
Randomized incremental algorithms proved very efficient in solving many geometric
problems. The basic methods (using the influence or the conflict graphs) or one of
their many variants inspired much work by several researchers such as Mulmuley [172,
174], Mehlhorn, Meiser and 6 'Dinlaing [164], Seidel [205], Clarkson and Shor [71], and
Aurenhammer and Schwarzkopf [18].
There is a class of randomized algorithms which work not by the incremental method,
but rather by the divide-and-conquer paradigm. The subdividing step is achieved using
a sample of the objects to process. Randomization is used for choosing the sample, and
the method can be proved efficient using the probabilistic theorems given in exercises 4.5
and 4.6. Randomized divide-and-conquer is mainly used for building hierarchical data
structures that support repeated range queries. Typically, these queries can be expressed
in terms of locating a point in the arrangement of a collection of hyperplanes, simplices,
or other geometric objects. In a dual situation, the data set is a set of points and
the queries ask for those points which lie in a given region (half-space, simplex, ... ).
Haussler and Welzl [123] spurred new interest in the field with their notion of an e-net.
Later, Matousek introduced the related notion of c-approximations [150]. Chazelle and
Friedman [53] showed how to compute these objects in a deterministic fashion using
the method of conditional probabilities. The resulting deterministic method is called
a derandomization of the randomized divide-and-conquer method. This method was
then widely used, for instance by Matousek [150, 151, 152, 153, 154, 155], Matousek
and Schwarzkopf [156], or Agarwal and Matousek [4]. In his thesis [35], Br6nnimann
studies the derandomization of geometric algorithms and the related concept of the
Vapnik-Chervonenkis dimension. Randomized divide-and-conquer is also used by Clark-
son, Tarjan, and Van Wyk in [65] to build the vertical decomposition of a simple
polygon.
Last but not least, the book by Mulmuley [177] is entirely devoted to randomized
geometric algorithms, and serves as a very comprehensive reference on the topic.
Chapter 6
Dynamic randomized
algorithms
The geometric problems encountered in this chapter are again stated in the ab-
stract framework of objects, regions, and conflicts introduced in chapter 4. A
dynamic algorithm maintains the set of regions defined and without conflict over
the current set of objects, when the objects can be removed from the current set
as well as added. In contrast, on-line algorithms that support insertions but not
deletions are sometimes called semi-dynamic.
Throughout this chapter, we denote by S the current set of objects and use the
notation introduced in the previous two chapters to denote the different subsets
of regions defined over S. In particular, Fo (S) stands for the set of regions defined
and without conflict over S. To design a dynamic algorithm that maintains the
set Fo(S) is a much more delicate problem than its static counterpart. In the
previous chapter, we have shown how randomized incremental methods provide
simple solutions to static problems. In addition, the influence graph techniques
naturally lead to the design of semi-dynamic algorithms. In this chapter, we
propose to show how the combined use of both conflict and influence graphs can
yield fully dynamic algorithms.
The general idea behind our approach is to maintain a data structure that
meets the following two requirements:
* It allows conflicts to be detected between any object and the regions defined
and without conflict over the current subset.
* After deleting an object, the structure is identical to what it would have
been, had the deleted object never been inserted.
Such a structure is called an augmented influence graph, and can be imple-
mented using an influence graph together with a conflict graph between the re-
gions stored in the influence graph and the current set of objects. In some cases,
we might be able to do without the conflict graph.
96 Chapter 6. Dynamic randomized algorithms
In section 6.2, we describe the augmented influence graph and how to perform
insertions and deletions. The randomized analysis of these operations is given in
section 6.3. This analysis assumes a probabilistic model which is made precise
and unambiguous in section 6.1. The general method is used in section 6.4 to
design a dynamic algorithm that builds the vertical decomposition of a set of
segments in the plane.
This chapter also uses the terminology and notation introduced in the previous
two chapters. To ease the reading process, some definitions are recalled in the
text or in the footnotes.
6.1 The probabilistic model

The current set of objects, denoted by S, is the result of a sequence of insertions
and deletions. Due to the second requirement which we stated earlier, the data
structure does not keep track of the deleted objects. Consequently, at any given
time, the data structure depends only on S and on the order in which the ob-
jects in S were introduced. In fact, an object stored in the data structure may
have been inserted and removed several times, yet the current state of the data
structure only keeps track of the last insertion.
At any given time, the data structure only depends on the chronological se-
quence E = {O1, 02, ... , O°n} which is the sequence of objects in S in the order
of their last insertion.
The randomized analysis of a dynamic algorithm assumes that:
* If the last operation is an insertion, each object in the current set S is

equally likely to have been inserted in this operation.
* If the last operation is a deletion, each object present prior to the deletion
is equally likely to be deleted in this operation.
It follows from these two assumptions that the chronological sequence E is

random, and that every permutation of the objects in S is equally likely to occur
in E. Let the current set of object S be of size n, and let i be an integer
in {1, ... , n}. Each object in S is the object Oi of rank i in E with uniform
probability 1/n. Moreover, in a deletion, the object Oi of rank i in E is deleted
with uniform probability 1/n.
We let Si be the subset {01, 02, ... , OI} of the first i objects in the chrono-
logical sequence. The probabilistic model implies that for i, 1 < i < n, Si is a
random i-samplel of S and, for each pair (i, j) such that 1 < i < j < n, Si is a
'We may recall that a random i-sample is a random subset of size i of S. Its elements are
chosen in a way that makes all the ( n) possible subsets of size i of S equally likely.
6.2. The augmented influence graph 97
random i-sample of Sj.
6.2 The augmented influence graph

The augmented influence graph obtained after a sequence of insertions and dele-
tions that results in a set 5, is determined only by the chronological sequence
S of the objects in S and is denoted Ia(s). The augmented influence graph
-Ea(s)is connected, directed, and acyclic. It has the same nodes and arcs as the
graph built by an on-line algorithm which inserts the objects of the sequence E
in the order of E. There is a node in the graph for each region that belongs to
U> .Fo(Si), where Fo(Si) denotes the set of regions defined and without conflict
over Si. Let us recall that a region is characterized by two subsets of objects: the
subset of objects with which it conflicts, called the influence domain of the region,
and the subset of objects with bounded size that determine the region. In the
following, we call each object that belongs to this second subset a determinant
of the region. We call creator of a region the object of highest chronological rank
among the determinants of the region. As in the preceding chapter, we use the
terminology of trees to describe the structure of the augmented influence graph,
and often identify a node with the region that is stored therein. This lets us speak
for instance of the parent or children of a region, or of the influence domain of a
node. The arcs of the influence graph maintain the inclusion property that the
influence domain of a node is a subset of the union of the influence domains of
its parents. For dynamic algorithms, we demand that arcs of the augmented in-
fluence graph also ensure a second inclusion property stating that, apart from its
creator, the set of determinants of a region is contained in the set of determinants
of its parent regions.
In addition to the usual information stored in the influence graph, the aug-
mented influence graph stores a conflict graph between the objects in the current
set S and the regions stored in the nodes of the influence graph. This conflict
graph is represented as in the preceding chapter by a system of interconnected
lists: To each region F stored at a node of the influence graph corresponds a list
L'(F) of objects of S with which it conflicts. To each object 0 in the current
set S corresponds a list L(O) of regions stored in the entire influence graph that
conflict with 0. There is a bidirectional pointer between the entry corresponding
to a region F in the list L(O) of an object 0 and the entry corresponding to 0
in the list L'(F).
Inserting an object
Inserting an object O, into a structure built for a set Sn-_ is very similar to the
operation of inserting an object in an on-line algorithm that uses an influence
graph. The only difference is that, in addition to the insertion into the influence
graph, we must also take care of updating the conflict lists. This can be done in
two phases: a locating phase, and an updating phase.
Locating. The algorithm searches for all the nodes in the influence graph of
:Za(E) that conflict with On. Each time a conflict is detected, we add a conflict
arc to the conflict graph, add On to the conflict list of the region that conflicts
with it, and add this region to the list L(OA).
Updating. A node of the influence graph is created for each region in Fo(Sn)
determined by a set of objects that contains On. This node is also linked to
parent nodes so that the two inclusion properties hold.
We may recall that a region in Fo(Sn) is said to be created by On if it is
determined by a set of objects that contains On. Similarly, a region of FO(Sn-1)
is said to be killed by On if it conflicts with On. More generally, a region stored
in a node of the influence graph Ta(s) has a creator in E, and a killer if it is not
a leaf. The creatorof F is, among all the objects that determine F, the one that
has the highest rank in E. The killer of F is, among all the objects in E that
conflict with F, the one with the lowest chronological rank.
For the rest of this chapter, we assume that the augmented influence graph
satisfies the update condition 5.3.3. In particular, a node of the graph that stores
a region created by On is linked only to nodes storing regions killed by On.
Deleting an object
To simplify the discussion, assume that the current set S has n objects, and that
the current data structure is the augmented influence graph la(s) corresponding
to the chronological sequence E = {O1, - , On}. The object to be deleted is
.
Ok, the object that has chronological rank k. The algorithm must modify the
augmented influence graph to look as if Ok had never been inserted into E.
The augmented graph must therefore correspond to the chronological sequence
S = {O1i,. ,Ok-1, Ok+1.... On}.
For any integer 1, k < I < n, let us denote by S' the subset S1 \ {Ok} of S. In
particular, observe that Sk = Sk-1.
In what follows, an object is called a determinant of a region if it belongs to
the set of objects that determine that region. The symmetric difference between
the nodes of Ia(s) and those of Ia(E') can be described as follows.
1. The nodes of la(s) that do not belong to -Ea(V) are determined by a set
of objects that contain Ok. Therefore Ok is a determinant of those regions,
and we say that such nodes (and the corresponding regions) are destroyed
when Ok is deleted.
2. The influence graph Ta(V') has a node that does not belong to la(E) for
6.2. The augmented influence graph 99
each region in U1=k+1. nFo(SI) that conflicts with Ok. Let us say that
such a node is new when Ok is deleted, and so is its corresponding region.
A new region has a creator and, occasionally, a killer in the sequence E'. If
the region belongs to Fo(S'), conflicts with Ok, and is determined by a set
of objects that contain O, then it is a new region after Ok is deleted, and
its creator is Qi.
Nodes that play a particular role when Ok is deleted include of course the
new nodes as well as the destroyed ones, but the nodes killed by Ok also have
a special part to play. The nodes killed by Ok should not be mistaken for the
nodes destroyed when Ok is deleted. Nodes killed by Ok correspond to regions
of Fo(Sk-l) that conflict with Ok, whereas nodes destroyed when Ok is deleted
correspond to regions that admit Ok as a determinant. The latter nodes disappear
from the whole data structure when Ok is deleted. The former nodes are killed
when Ok is inserted but remain in the data structure (occasionally becoming
internal nodes), and they still remain after Ok is deleted.
Upon a deletion, the arcs in the influence graph Ta(s) that are incident to
the nodes destroyed by Ok disappear and the graph Ia(E') has arcs incident to
the new nodes. In particular, new nodes must be linked to some parents (which
are not necessarily new nodes). Moreover, a few nodes of Ia(s) that are not
destroyed witness the destruction of some of their parents. Let us call these
nodes unhooked. They must be rehooked to other parents.
Again, deletions can be carried out in two phases: a locating phase, and a
rebuilding phase.
Locating. The algorithm must identify which nodes of the influence graph
la(s) are in conflict with Ok, which nodes have to be destroyed, and which are
unhooked. Owing to both inclusion properties, this can be done by a traversal of
the influence graph. This time, however, we not only visit the nodes that conflict
with Ok, but also those which admit Ok as a determinant. The destroyed or
unhooked nodes are inserted into a dictionary which will be looked up during the
rebuilding phase.
Rebuilding. The first thing to do is to effectively remove all the destroyed
nodes. Those nodes can be retrieved from the dictionary, and all the incident
arcs in the graph are also removed from the graph. The conflict lists of the nodes
which conflict with °k are also updated accordingly. We shall not detail these
low-level operations any further, as they should not raise any problems. Next,
we must create the new nodes, as well as their conflict lists; we must also hook
these new nodes and rehook the nodes that were previously unhooked. The detail
of these operations depends on the nature of the specific problem in hand. The
general design is always the same, however: the algorithm reinsertsone by one,
and in chronological order, all the objects Qi whose rank I is higher than k and
that are creators of at least one new or unhooked region. To reinsert an object
involves creating a node for each new region created by QI, hooking this node
into the influence graph, setting up its conflict list, and finally rehooking all the
unhooked nodes created by O1.
To characterize the objects Oj that must be reinserted during the deletion of
Ok, we must explain what critical regions and the critical zone are. For each
I > k, we call critical those regions in .Fo(S -1) that conflict with Ok- We call
critical zone, and denote by Z- 1 , the set of those regions.
Lemma 6.2.1 Any object 01 of chronological rank 1 > k that is the creator of
a new or unhooked node when Ok is deleted conflicts with at least one critical
region in Z1 - 1.
Proof. If 01 is the creator of a new node, then there is a region F in Fo(Sl)

that is determined by 01 and conflicts with Ok* In the influence graph la(E'),
this region is linked to parents which, according to condition 5.3.3, are associated
with regions in Fo(S' 1-) which conflict with O1. Still according to this condition,
at least one of these nodes conflicts with Ok, which proves the existence of a
region G in Fo(Sl- 1 ) that conflicts with both 01 and Ok*
If 01 is the creator of a unhooked node, then there is a region F in Fo(sl) n
.Fo(Sl) whose determinants include Oj. The region F is linked in the influence
graph la(E') to parents, at least one of which is either new or killed by Ok
(otherwise the region does not need to be rehooked). Update condition 5.3.3
assures us that this parent conflicts with Oj, which proves that there is a region
G in Fo(SI -)1 that conflicts both with Ol and °k*
For each I > k, we must thus determine whether there is a critical region in
ZI-1 that conflicts with Qi. If so, then O1 is reinserted, and we must find all the
critical regions that conflict with 01. Dynamic algorithms are efficient mostly
when reinserting 01 involves traversing only a local portion of the influence graph
that contains all the critical regions which conflict with 01.
Before starting the rebuilding phase, the critical zone is initialized with those
regions killed by Ok. At each subsequent reinsertion, the critical zone changes.
To determine the next object that has to be reinserted and the critical regions
that conflict with this object, we maintain in a priority queue Q the set of killers,
according to the sequence ', of current critical regions. Killers are ordered within
Q by their chronological rank, and each one points to a list of the critical regions
that it kills.
The priority queue Q is first initialized with those regions killed by °k. For
each critical region F in Zk-l, we identify its killer in E' as the object, other
than °k, with the lowest rank in the list L'(F).
At each step of the rebuilding process, the object with the smallest chronological
rank O1 is extracted from Q, and we also get all the critical regions that conflict
6.3. Randomized analysis of dynamic algorithms 101
with 01. The object O1 is then reinserted, and the details of this operation depend
of course on the problem in hand. The main obstacle is that we might have to
change more than the critical zone of the influence graph. Indeed, the new regions
created by 01 always have some critical parents, even though they may also have
non-critical parents. Moreover, parents of an unhooked region are new, but the
unhooked region itself is not. To correctly set up the arcs in the influence graph
that are incident to new nodes, the algorithm must find in Ia(s) the unhooked
nodes and the non-critical parents of the new nodes. At this phase, the dictionary
set up in the locating phase is used. After reinserting 01, the priority queue Q is
updated as follows: the regions in Z 1 _1 that conflict with 01 are not critical any
more; however, any new region created by 0 belongs to Z1. Then for each of
these regions F, the killer of F in A' is identified as the object in L'(F) with the
smallest chronological rank. This object is then searched for in Q and inserted
there if it is not found. Then F is added to the list of regions killed by QI.
6.3 Randomized analysis of dynamic algorithms

The randomized analysis of the augmented influence graph and the insertion and
deletion operations are based on the probabilistic model described in section 6.1.
The first three lemmas in this paragraph analyze the expected number of elemen-
tary changes to be performed upon a deletion.
Lemma 6.3.1 Upon deleting an object, the number of nodes that are destroyed,
new, or unhooked is, on the average,
o ( Lfo (I S)
where, as usual, fo(l, S) stands for the number of regions defined and without
conflict over a random sample of size I from S.
Proof. We bound the number of destroyed, new, and unhooked nodes separately.
1. The number of destroyed nodes. A node in Ta(s) corresponding to a
region F in 4j (S) is destroyed during a deletion if the object deleted is one of the
i objects that determine the region F. Let F be a region in F(S). Given that F
corresponds to a node in the influence graph built for S, this node is destroyed
during a deletion with a conditional probability i/n < b/n. From theorem 5.3.2,
we know that the expected number of nodes in the influence graph is
n(fMlS)
so the number of nodes destroyed when deleting an object is, on the average,
1 E fo(1, S))
2. The number of new nodes. The regions that correspond to the new nodes
in the influence graph when Ok is deleted are exactly the regions created by 01,
for some I such that k < 1 < n, that belong to Fo(S') and conflict with Ok. Let
F be a region of 4j(S). This region F belongs to TFo(S1) with the probability
pj(1 - 1) that was given in subsection 5.2.2. Assuming this, F is created by 01
with conditional probability i/(l - 1), and F conflicts with Ok with conditional
probability j/(n - I + 1). Therefore, for a given k, the number of new nodes in
the influence graph upon the deletion Ok is, on the average (using corollary 4.2.7
to the moment theorem),
EL E AI(S)
P;(l -1) I = °( _ _
= O(E fo(L1/2 J S))
Averaging over all ranks k, the number of new nodes in the influence graph after
a deletion is
1 nn1 fo( Ll'/2j , S) 1 n-1 fo (1,S)
0n -EE 1'2 =0°VE
3. The number of unhooked nodes. Unhooked nodes are the non-destroyed
children of destroyed nodes. If condition 5.3.3 is satisfied, the number of children
of each node in the augmented influence graph is bounded by a constant. It
follows that the number of unhooked nodes is at most proportional to the number
of destroyed nodes. El
The update condition 5.3.3 assumes that the number of children of a node
is bounded by a constant. However, the number of parents of a node is not
necessarily bounded by a constant and the following lemma is useful to bound
the number of arcs in the influence graph that are removed or added during a
deletion.
Lemma 6.3.2 The number of arcs in the influence graph that are removed or
added during a deletion is, on the average,
0 (n1 fo(l, S))

Proof. The simplest proof of this lemma involves the notion of biregion encoun-
tered in exercise 5.7. A biregion defined over a set of objects S is a pair of regions
defined over S which can possibly be related as parent and child in the influence
graph, for an appropriate permutation of S. A biregionis determined by at most
2b objects, and the notion of conflict between objects and regions can be extended
to biregions: an object conflicts with a biregion if it is not a determinant of any
of the two regions but conflicts with at least one of the two regions. Biregions
obey statistical laws similar to those obeyed by regions. In particular, a biregion
determined by i objects of S which conflicts with j objects of S is a biregion
defined and without conflict over a random l-sample of S, with the probability
p'(I) given by lemma 4.2.1. A biregion defined and without conflict over a subset
S of S corresponds to an arc in the influence graph whenever the objects that
determine the parent region are inserted before those that determine the child
region and at the same time conflict with the parent region. This only happens
with a probability a E [0,1] (which depends on the number of objects determin-
ing the parent and the child, and the number of objects that at the same time
determine the child and conflict with the parent).
A biregion determined by i objects in S and conflicting with j objects in S
corresponds to an arc in the influence graph la(s) that was created by 01, with
a probability smaller than l pj(1) (see also exercise 5.7); this arc, created by 01,
conflicts with Ok with a probability smaller than
1 j
- pZ (I).
A computation similar to that in the proof of lemma 6.3.1 shows that the
expected number of arcs in the influence graph that are created or removed
during a deletion (which are those adjacent in the influence graph to new nodes
or to destroyed nodes) is
I n1E ffo (1, S)A
Vn 11 l
where ffo(i, S) is the expected number of biregions defined and without conflict
over a random 1-sample of S. It remains to show that ffo(l, S) is proportional to
fo(l, S). Let S1 be a subset of size 1 of S. The parent region in a biregion that
is defined and without conflict over SI is a region defined over SI that conflicts
with exactly one object in SI, and is therefore a region in F1 (SI). Conversely,
if the update condition 5.3.3 is true, every region in Fl(Sl) is the parent in a
bounded number of biregions defined and without conflict over S1. It follows that
ffo(l, S) is within a constant factor of the expectation fi(l, S) of the number of
regions defined and conflicting with one element over a random i-sample. From
corollary 4.2.4 to the sampling theorem, this expected number is O(fo(l, S)). E
Lemma 6.3.3 The total size of all the conflict lists attached to the nodes that
are new or destroyed when an object is deleted is, on the average,
0 (E fo(1 S))
Proof.
1. Conflict lists of destroyed nodes. A region F of 4j(S) corresponds to a
node of the influence graph Ta(s) with probability
n
P3; (1),
1=1
as implied by lemma 5.2.2. The conflict list attached to this node has length j
and this node is destroyed during the deletion of an object with probability i/n.
The total size of the conflict lists attached to destroyed nodes is thus
LEE~~I-i
l~(S) IP3
p(1)I- = ( Eml,)
1i=t1 3=111
= O( 2 (n l ) fo(LS2))
= 0 ( fo (1iS))
112
as follows from corollary 4.2.7 to the moment theorem.

2. Conflict lists attached to new nodes. A region F of J (S) is a new region
created by O1 when Ok is deleted, if it is a region of Fo(Si) determined by O1
that conflicts with Ok. The conflict list attached to the new node corresponding
to F has j - 1 elements. The total size of the conflict lists attached to new nodes
when deleting Ok is thus, on the average,
bo n n
E_ E I j(S)I pJ(l-1) n -
n-+)(1
(j-1)
i= =1 =k+l 11
Applying corollary 4.2.7, this size is
Av erm (a S)) ( nl fo(L JS)

Averaging over all ranks of k, the above sum becomes
(1 nE (n-I f0( L22 ) = (E; n (1 S)

Lastly, setting up the priority queue Q of killers of critical regions involves the
regions of the influence graph La(E) that are killed by Ok. The conflict lists of
these regions are traversed in order to set up the conflict lists of the new children
of these nodes. The following lemma is therefore needed in order to fully analyze
dynamic algorithms.
Lemma 6.3.4 The number of nodes in the influence graph la(E) that are killed
by a random object in S is, on the average,
o( Efo(1, S))
The total size of the conflict lists attached to the nodes killed by a random object
is, on the average,
o (I fo(1, S))
Proof. A node in la(E) that is killed by an object Ok of rank k corresponds to

a region in Fo(Sk-1) that conflicts with Ok. A region F in Fj7(S) is a region in
FO(Sk-1) that conflicts with Ok with probability
pj(k -1)
Hence, the average number of nodes in la(s) killed by a random object of S is
n EZZZ
k=2 i=1 j=1
i(S)I p,(k-1) 'k = E
k=2
- 0 (!foI:
(n kL
as can be deduced from corollary 4.2.7 to the moment theorem.
The total size of the conflict lists attached to nodes killed by a random object
is, on the average,
EA E 1-j-(S)I pj(kl (i-1) = °( E -k_))

kn = j=)
k=2 i=( = n -k+I-y -k+1 k=2
= E f o ( Lk/2j, S))
k=1
O (EMol, S))
The detailed operations required to insert or delete an object in an augmented

influence graph depend upon the problem under consideration. In particular,
deletions demand a number of operations (insertions, deletions, or queries) on a
dictionary of nodes, or even several dictionaries. To be able to present a general
analysis, we introduce here an update condition for dynamic algorithms that use
an augmented influence graph. This condition is similar to those introduced in
the previous chapter to analyze incremental algorithms, namely condition 5.2.1
for algorithms using a conflict graph and condition 5.3.3 for algorithms using an
influence graph. The condition we introduce here is a reasonable one, which will
be fulfilled by all the dynamic algorithms described in this book.
Update condition 6.3.5 (for augmented influence graphs) A dynamic al-

gorithm that uses an augmented influence graph satisfies the update condition
when:
1. The augmented influence graph satisfies the update condition 5.2.1 for algo-
rithms using an influence graph.
2. During a deletion:
a. The number of operations on a dictionary of nodes (insertions, deletions,

or queries) is at most proportional to the total number of nodes killed (by
the deleted object), destroyed, new, or unhooked.
b. The conflict lists of the new nodes are initialized using a time proportional
to the total size of the conflict lists of the nodes killed (by the deleted object),
destroyed, and new.
c. All the operations performed to update the augmented influence graph that
do not pertain to dictionaries, conflict lists, or the priority queue Q, are
elementary and their number is proportionalto the total number of destroyed
or new nodes, and of arcs incident to these nodes.
The complexity of a deletion depends partly on the data structures used to

implement the dictionaries and the priority queue Q of the killers of critical
regions. These data structures store a set of elements that belong to a finite,
totally ordered universe, whose cardinality is bounded by a polynomial in the
number n of current objects. Therefore, we can use the data structures described
in section 2.3, or more simply we may use a standard balanced binary tree. In
order to take all these cases into account, the analysis given below introduces two
parameters. The first parameter, t, is the complexity of a query, insertion, or
deletion performed on a dictionary of size 0(nc), where c is some constant. The
second parameter, t', is the complexity of a query, insertion, or deletion performed
on a priority queue of size O(n). Parameter t is O(logn) if a balanced tree is
used, and 0(1) if the perfect dynamic hashing method of section 2.3 is used.
Parameter t' is O(log n) if we use a balanced binary tree, but it is O(log log n)
if we use a stratified tree along with perfect dynamic hashing (see section 2.3).
Moreover, as we will see further on, if fo(l, S) grows at least quadratically, then
implementing Q with a simple array of size n will suffice, and t' can be ignored
in the analysis.
Theorem 6.3.6 This theorem details the performances of a dynamic algorithm

that uses an augmented influence graph and satisfies the update condition 6.3.5.
Let S be the current set of objects, and n the size of S.
1. The data structure requires an average storage of
o (n Ml S))
2. Adding an object can be performed in expected time
o ( fo(l, S)
3. Deleting an object can be performed in expected time
O (min (n t E fo(1,S)) + t E fo(1,S)+ E fo(1,S))
As always, fo(l,S) is the number of regions defined and without conflict over a
random I-sample of S, t is the complexity of any operation on a dictionary, and
t' is the complexity of an operation on the priority queue Q.
Proof.
1. The storage needed by the augmented influence graph fa(s) is proportional
to the total size of the conflict lists attached to the nodes of lTa(s). Each element
in one of these conflict lists corresponds to a conflict detected by an on-line
algorithm processing the objects in S in the chronological order of the sequence
E. The expected number of conflicts, for a random permutation of E, is thus
given by theorem 5.2.3 which analyzes the complexity of incremental algorithms
that use a conflict graph.
2. The randomized analysis of an insertion into the augmented conflict graph
is identical to that of the incremental step in an on-line algorithm that uses
an influence graph. Indeed, the two algorithms only differ in that one updates
conflict lists. Each conflict between the inserted object and a node in the current
graph is detected by both algorithms. In the dynamic algorithm, detecting such

a conflict implies adding the inserted object into the conflict list of the conflicting
node, which can be carried out in constant time. The expected complexity of an
insertion is thus given by theorem 5.3.4.
3. Let us now analyze the average complexity of deleting an object, say Ok*
When locating Ok in the augmented influence graph, the nodes that are visited
are exactly the destroyed nodes, and their children, and the nodes that conflict
with Ok* Since every node has a bounded number of children, the cost of the
traversal is proportional to the number of nodes destroyed or conflicting with
Ok. The number of nodes in the influence graph that conflict with Ok is, on the
average over all possible sequences A,
0 IS (L2 2,S
which we know from the proof of theorem 5.3.4. Averaging over the rank of the
deleted object, we get
o ( fo (l S))
From lemma 6.3.1, the latter expression is also a bound on the expected number
of nodes destroyed and thus on the global cost of traversing the influence graph.
If the update condition 6.3.5 is realized, lemmas 6.3.3 and 6.3.4 show that the
conflict lists of the new regions can be set up in time
o (E fo(1,S))
Lemma 6.3.1 and the update condition 6.3.5 (2a) assert that the term
o (t Eo ( IS))
accounts for the average complexity of all the operations performed on the dic-
tionaries of nodes.
Since t is necessarily Q(1), lemmas 6.3.1 and 6.3.2, together with condition 6.3.5
(2c), assert that the former term also accounts for all the operations that update
the augmented influence graph, not counting those on the conflict lists or the
priority queue.
It remains to analyze the management of the priority queue Q of critical region
killers. The number of insertions and queries in the priority queue is proportional
to the total number of critical regions encountered during the rebuilding phase.
These regions are either killed by the deleted object, or they are new regions.
Their average number is thus
0 1 Efo(l' 5))
as asserted by lemmas 6.3.1 and 6.3.4. The average number of minimum queries
to be performed on the queue Q is
0 (mi n, tnnfo(l )))
since the number of objects to be reinserted is bounded from above by n on the one
hand, and by the number of unhooked or new nodes (estimated by theorem 6.3.1)
on the other hand. [1
Consequently,
* If fo (I,S) grows slower than quadratically (with respect to 1), we use a

hierarchical structure for the priority queue, characterized by a parameter t'
which bounds the complexity of any operation on this structure (insertion,
membership or minimum query). Managing the queue has therefore the
associated expected cost
* If on the contrary, fo(l, S) grows at least quadratically, we use for Q a

simple array of size n. This allows insertions and deletions to be performed
into the queue in constant time. The cost of finding the minima during the
whole rebuilding phase is then O(n), and managing the queue has in this
case the associated expected cost
I fo (1,S))
* When fo(l,S) is 0(l), the expected number of destroyed or new nodes

visited during a deletion is 0(1). Updating the conflict lists costs 0(log n)
anyway. Both the priority queue and the dictionaries can be implemented
simply by balanced binary trees (t = t' = log n) to yield a randomized
complexity of 0(log n) for a deletion.
6.4 Dynamic decomposition of a set of line segments
The vertical decomposition of a set of line segments in the plane is a structure

defined in section 3.3. It can be built using a conflict graph by a static incremen-
tal algorithm, as explained in subsection 5.2.2, or by a semi-dynamic incremental
algorithm using an influence graph, as detailed in subsection 5.3.2. By combin-
ing both structures using the general method explained in section 6.2, we can
dynamically maintain the vertical decomposition of the segments under insertion
or deletion of a segment. The algorithm described here is a generalization of the
decomposition algorithms given in subsections 5.2.2 and 5.3.2. It is advisable to
thoroughly understand both these algorithms before reading further.
Let us first recall that, for this problem, the objects are segments and the
regions defined over a set S of segments are the trapezoids appearing in decom-
positions of subsets of S. A trapezoid is determined by at most four segments.
A segment conflicts with a trapezoid if it intersects the interior of the trapezoid.
Let S be the current set of segments and let = {=O,O2 )°...
,O0 be the
chronological sequence of segments in S. We may also denote by Sj the subset
of S consisting of the first I segments in E. The dynamic algorithm maintains an
augmented influence graph Ia(E), whose nodes correspond to the trapezoids that
are defined and without conflict over the current subsets SI, I = 1, . . . , n. The
nodes and arcs in this graph are identical to those built in the influence graph by
the on-line algorithm described in subsection 5.3.2. In addition, the augmented
influence graph includes a conflict graph between the trapezoids corresponding to
the nodes of the influence graph, and the segments in S. The conflict graph is im-
plemented using the interconnected list system, as explained in section 6.2. The
structure does not encode all the adjacency relationships between the trapezoids
but only those between the trapezoids in the current decomposition (correspond-
ing to leaves of the influence graph).
Adding the n-th segment does not create problems and can be carried out
exactly as explained in subsection 5.3.2. The only difference is that the conflict
lists are updated when the conflicts are detected during the locating phase when
inserting O,.
Let us now explain how to delete segment Ok, of rank k in the chronological
sequence E. As before, we denote by S' the subset SI \ {Ok} of S and by E' the
chronological sequence {O1, ** ,Ok-1, Ok+1.. . O*}.
The algorithm proceeds along the usual lines and performs the two phases:
locating and rebuilding. The locating phase detects the nodes of Ta(s) that
conflict with Ok, and the destroyed and unhooked nodes. In the rebuilding phase,
the algorithm processes the segments of rank 1 > k that are the creators of new or
unhooked nodes. For this, the algorithm manages a priority queue which contains,
at each step of the rebuilding process, the killers in E' of the critical regions. For
6.4. Dynamic decomposition of a set of line segments ill
each such object QI, a killer of a critical region, the algorithm builds the new
nodes created by 01 and rehooks the unhooked nodes created by 01. Figure 6.1
shows how the influence graph built for the four segments {°O, 02, 03, 04} is
modified when deleting 03. The reader may observe again how the graph was
created incrementally, in figures 5.6, 5.7 and 5.8. In this example, nodes B and H
are killed by 03, nodes J,K,L,M,N,O,P,Q,S,U,V are destroyed, nodes R,T,W
are unhooked (they are created by 04), and B' is a new node (its creator is 04).
The subsequent paragraphs describe in great detail the specific operations
needed.
Locating. This phase is trivial: all the nodes that conflict with the object
Ok to be deleted, or that are determined by a subset containing Ok, are visited
together with their children. The algorithm builds a dictionary D of unhooked
or destroyed nodes, which will be used during the rebuilding phase.
Rebuilding. The priority queue Q, which contains the killers of critical re-
gions, is initialized with the nodes in Ia(s) that are killed by Ok-
At each step in the rebuilding process, the algorithm extracts from the priority
queue Q the object 0 1 of smallest chronological rank. It also retrieves the list of
the critical regions that conflict with 01.
Each of these regions is split into at most four subregions by 01, and the walls
stemming from its endpoints and its intersection points. These subregions are
not necessarily trapezoids in the decomposition Dec(S'). Indeed, the walls cut
by 01 have to be shortened, keeping only the part that is still connected to the
endpoint or intersection point from which it stems. The other part of the wall
must be removed and the adjacent subregions separated by this part must be
joined. The join can be one of two kinds: internalwhen the portion of wall to be
removed separates two critical regions, and external when it separates a critical
region from a non-critical region (see figure 6.2).
To detect which regions to join, 2 the algorithm visits all the critical regions
that conflict with 01, and stores in a secondary dictionary DI, the walls incident
to these regions that are intersected by 01. Any wall in this dictionary that
separates two critical regions gives rise to an internal join, and any wall incident
to only one critical region gives rise to an external join.
In a first phase, the algorithm creates a temporary node for each subregion
resulting from the splitting of a critical region by 01 or the walls stemming from
Oj. The node that corresponds to a subregion Fi of the region F is hooked in the
graph as a child of F. Its conflict list is obtained by selecting, from the conflict
2
The algorithm cannot traverse the sequence, ordered by O, of critical regions for two rea-
sons: (1) it does not maintain the vertical adjacencies between the internal nodes of the influence
graph, and the adjacencies between either the trapezoids of the decomposition Dec(Sl-1) or the
critical regions of Zj-1 are not available, and (2) the intersection of 0Q with the union of the
regions in Zj-1 may not be connected (see for instance figure 6.4).
(a)
F D
(b) (c)
(d)
Figure 6.1. Deleting a segment.

6.4. Dynamic decomposition of a set of line segments 113
(b)
77
-I
.I--
I ! I : ;
(c) (d)
Figure 6.2. Internal and external joins:

(a) The decomposition Dec(SI) before deleting Ok.
(b) The decomposition Dec(S{.-,), with the critical zone Z1 _1 shaded.
(c) Reinserting 01. Splitting the critical regions and joining: internal join
G = G1 U G2 , external join F = G U Fd. Region H is unhooked.
(d) The decomposition Dec(S,) and the critical zone Z1.
list of F, the segments intersecting Fi. Then the algorithm processes the internal
and the external joins, as explained below.
1. Internal joins. Every maximal set {G 1 , Gh} of subregions, pairwise ad-
jacent and separated by walls to be removed, must be joined together into a single
region G. The algorithm creates a temporary node for G. The nodes correspond-
ing to G 1, G 2 ,. . . , Gh are removed from the graph and the node corresponding
to G inherits all the parents of these nodes. The conflict list of G is obtained by
merging the conflict lists of G 1 , G2 , .. ., Gh, removing redundancies. For this, we
use a procedure similar to that of subsection 5.2.2, but which need not know the
order along 01 of the subregions to be joined. By scanning the conflict lists of
these subregions successively, the algorithm can build for each segment 0 in S a
list LG(O) of the subregions that conflict with 0. A bidirectional pointer inter-
connects the entry in the list L'(Gi) that corresponds to an object 0 with the
entry in LG(O) corresponding to the subregion Gi. The conflict list of G can be
retrieved by scanning again all the conflict lists L'(Gi) of the subregions G1 , ....
Gh. This time, each segment 0 encountered in one of these lists is added to the
conflict list of G and removed from the other conflict lists, using the information
stored in LG(O).
Let us call auxiliary regions the regions obtained after all the internal joins.
These regions are either subregions that needed no internal join, or regions ob-
tained from an internal join of the subregions. An auxiliary region that does not
need to undergo any external join is a region of the decomposition Dec(S'). Let
H be such a region. This region is new if it conflicts with Ok, unhooked other-
wise. In the former case, the temporary node for H becomes permanent and the
killer of H is inserted into the priority queue Q. In the latter case, a node for H
already exists in the influence graph la(S). A simple query in the dictionary of
unhooked nodes retrieves this node, which can then be rehooked to the parents
of the auxiliary node created for H.
2. External joins. In a second phase, the algorithm performs the external joins.
An auxiliary region undergoes a left join if its left wall must be removed, and a
right join if its right wall must be removed, and a double left-right join if both its
vertical walls must be removed. Let G be an auxiliary region undergoing a right
join. For instance, this is the case for region G = GI U G2 in figure 6.2. The right
wall of G is on the boundary of the critical zone, since this is an external join. This
wall is therefore not cut by the deleted segment Ok. When the decomposition
of S is built incrementally according to the order in the sequence A, this wall
appears at a certain step and is removed when Oj is inserted. Thus, among all
the regions in la(S), there is one region Fd created by O1 that contains the right
wall of G.3 The region Ed is necessarily destroyed or unhooked: indeed, Ed is a
trapezoid in the decomposition Dec(S1), and has a non-empty intersection with
one or more critical regions in Z1 _1 . As every critical region in Z1 - is contained
in the union of the trapezoids of Dec(SI-1 ) of which Ok is a determinant, the
region Ed must intersect those trapezoids. Thus at least one of the parents of Ed
in the graph la(S) is a destroyed node. Similarly, if the left wall of G must be
removed, there is in Ia(E) one destroyed or unhooked region Fg created by 01
that contains the left wall of G. If the join is double left-right, Ed and Fg may
be distinct or identical (see figure 6.3).
Several auxiliary regions may be joined into the same permanent region (see
figure 6.4). Let {G1 , G2, . ., Gj} be the sequence ordered along 0l of the auxiliary
3
1t would have been more desirable to subscript F by 1 and r for left and right, but this
would have conflicted with the index I for Oj and created confusion. We have kept a French
touch with the indices g and r for the French gauche and droit, meaning respectively left and
right. (Translator's note)
V VK'1 2/Z
'
6<
Fg /
Fd 4
01:
Figure 6.3. External joins: double left-right joins.
regions 4 whose left wall is contained in the same region Fg of Ia(s) created by
O°. If j > 1, then the right walls of these auxiliary regions {G1 , G2 , . . ., Gj-1,
are also contained in Fg and must be removed as well. If the right wall of Gj is
a permanent wall (that does not have to be removed), the join results in a single
trapezoid of the decomposition Dec(S, ) that is the same as Fg U G1 U . .. U Gj =
Fg U Gj (see figure 6.4). If the right wall of G3 also has to be removed, then
we introduce the ordered sequence of auxiliary regions {Gj, Gj+l, . . , Gh}: this
sequence consists of regions whose right wall must be removed and which lie in
the same region Fd of Ia(s) created by O1. The left walls of the regions in
{ Gj, Gj+l,... , Gh} then also belong to Fd and have to be removed. The join
operates on the auxiliary regions {G1 , ... , GjC,..., GO} and results in a unique
trapezoid in Dec(S8) that is the same as F. U G1 U . . . U Gh U Fd = F9 U Gj U Fd.
We present below the operations to be performed in the latter case of a double
left-right join. The former cases can be handled in a similar manner. Suppose
for now that the auxiliary regions {G1 , . . . , Gj,. . . , Gh} as well as the regions Fg
and Fd in the decomposition Ia(s) that participate in the join are known to the
algorithm.
If the trapezoid resulting from the join F = Fg U Gj U Fd does not conflict
with Ok (see figure 6.3, right), it is a trapezoid in the decomposition Dec(Sl).
Necessarily, the regions Fg, Fd, and F are the same, and the corresponding node
in Ia(s) is unhooked. It then suffices to search for this node in the dictionary of
unhooked nodes, to remove the auxiliary nodes created for G1 , G2 , . . ., Gh, and
to rehook the node corresponding to F, with the critical nodes in the parents of
G1 , G2 , .. ., Gh as the parents of F.
If the resulting trapezoid F = Fg U Gj U Fd conflicts with Ok (see figure 6.3,
left), then it is a new region of la(E'), and the regions Fg and Fd in la(E)
4
We must emphasize that even though the given description of the region resulting from an
external join refers to the order of the joined auxiliary regions along 01, the algorithm does not
know this order, nor does it need it.
III
_ I I ,I
Uk I i
(a)
(a)
LD)
(c)
Figure 6.4. External joins.

(a) The decomposition before deleting Ok*
(b) Reinserting 01. The auxiliary regions.
(c) Joining auxiliary regions GI and G2 into F = F, U Gi U G 2 .
are destroyed. The auxiliary nodes created for G1 , G2 , .. ., Gh are removed, and
replaced by a single node corresponding to F. This node is then rehooked to the
parents of F, and Ed that are not destroyed, and to all the critical parents of G1 ,
G2 , ... , Gh. The conflict list of F is derived from those of F9 ,G1 , G 2 , ... , Gh
and Ed, as is the case for internal joins. Lastly, the killer of F is inserted into the
priority queue Q.
We now have to explain how to retrieve the unhooked or destroyed nodes
corresponding to the regions Fg and Fd involved in the join. Let G be an auxiliary
region whose left wall must be removed. The corresponding region Fg is either
destroyed or unhooked, created by QI, and the segments that support its floor
and ceiling 5 respectively support the floor and ceiling of G. Any region in the
decomposition of a given set of segments is identified uniquely by its floor, its
ceiling, and one of its walls. Below, we show that either we can find one of the
walls of F., or we can identify a destroyed region F' which is the unique sibling
of F9 in la(s).
* If G conflicts with Ok (as in figure 6.5a), the right wall of F. is determined

by Ok, and can be computed (by looking only at G and Ok).
* If G does not conflict with Ok, but its right wall is permanent (see fig-
ure 6.5b), then this right wall is also that of F9 .
* Lastly, if G does not conflict with Ok, and if both its walls must be removed
(see figure 6.5c), then segment 0Q intersects both walls of a critical region
that was subsequently split into G and G'. The other subregion G' also
conflicts with Ok but does not undergo any join. In Ila(s), exactly one
node F' has Ot for creator, is destroyed, and shares the same floor, same
ceiling, and same left wall as G'. This node F' has only one parent, and
this parent has two children, one of which is F' and the other the node F9
that we are looking for: indeed, the parent of F' corresponds to a trapezoid
in the decomposition Dec(S 1- 1) whose two walls are intersected by 0 e.
In either case, the region Fg, or its sibling F' is known through its creator, its
floor, its ceiling, and one of its left or right walls. This information is enough to
characterize it. Naturally, the same observation goes for Fd or its sibling Fd. We
can then use the dictionary D storing all the destroyed or unhooked nodes. This
dictionary comes in two parts, Dg and Dd. In the dictionary Dg, the nodes are
labeled with:
* the creator segment,

* the segment supporting the floor of the trapezoid,
* the segment supporting the ceiling of the trapezoid,
* the pair of segments whose intersection determines the right wall of the
trapezoid, or the same segment repeated twice if the wall stems from the
segment's endpoint.
Similarly, in its counterpart Dd, nodes are labeled the same way, except that in
the last component the right wall is replaced by the left wall. Any destroyed or
unhooked node is inserted into both dictionaries Dg and Dd.
5
We recall that the floor and ceiling of a trapezoid are its two non-vertical sides.
-77
(a) (b)
-i
A
(c)
Figure 6.5. External joins:

(a) G conflicts with Ok.
(b) the right wall of G is permanent.
(c) Double left-right join.
To analyze this algorithm, we first check that it does satisfy the update condi-
tions 6.3.5. The first condition is satisfied, since the augmented influence graph
has the same nodes and arcs as the influence graph built by the on-line algorithm
of subsection 5.3.2, which itself satisfies the update condition 5.3.3. Therefore,
we need only look at deletions.
1. Number of operations on the dictionaries. Each deletion involves a
two-sided dictionary D of destroyed or unhooked nodes, as well as a dictionary
D., for each reinserted segment Qi, of walls in the critical zone intersected by
Ot. A destroyed or unhooked node is inserted and queried at most once in D.
A critical region in Z 1 - has at most two walls which must be inserted into DJ,
and this region will not be a critical region any more after the reinsertion of 01.
The number of operations on all dictionaries DI is thus at most proportional to
the total number of critical regions encountered in the rebuilding phase. Any
critical region is either killed or new. The total number of operations is thus at
most proportional to the number of nodes that are killed, destroyed, unhooked,
or new.
2. Conflict lists of new nodes. The conflict list of a new node is obtained by
scanning the conflict lists of the auxiliary or destroyed regions of which it is the
union. Similarly, the conflict list of an auxiliary region is obtained by traversing

the conflict lists of the subregions of which it is the union, and the conflict lists
of those subregions are themselves obtained by consulting the conflict lists of the
critical regions cut by the reinserted object. During the rebuilding process, each
killed or new region appears at most once as a critical region which conflicts
with the reinserted object. Moreover, each subregion is involved in at most one
internal join, and each auxiliary or destroyed region in at most one external join.
From this, we derive that the conflict lists of the new nodes can be computed in
time proportional to the total size of the conflict lists of the nodes that are killed,
destroyed, or new.
3. Other operations. Apart from managing the priority queue Q, querying the
dictionaries, and setting up the conflict lists of the new nodes, all the remaining
operations are elementary. Their number is at most proportional to the number
of new or destroyed nodes, and to the number of incident arcs in the augmented
influence graph.
If S is a set of n segments, with a intersecting pairs, the mathematical expecta-
tion fo(l, S) of the number of regions defined and without conflict over a random
I-sample of S is 0(1 + ax), as given by lemma 5.2.4. Thus,
(- I f 1S) = (1 + a)'
O (If ( S) = O(logn + )
We can now use theorem 6.3.6 to state the following theorem, which summarizes
the results so far:
Theorem 6.4.1 Under the assumptions of dynamic randomized analyses, an

augmented influence graph can be used to maintain the vertical decomposition
of a set of segments with the following performances. Let S be the current set of
segments, n be the size of S, and a be the number of intersecting pairs in S.
* The expected storage required by the algorithm is
O(n log n + a).
* Inserting the n-th segment takes an average time
O(log n+ a).
n
* Deleting a segment takes an average time

0 (logn+ (1+ -)(t+t')),
where the parameters t and t' stand respectively for the complexities of the
operations on dictionaries and priority queues.
Therefore, if we use perfect dynamic hashing together with stratified tree, the
expected cost of a deletion is
0 (logn + (1 + ) loglogn) .
If we use balanced binary trees, it remains
O(logn+ (1 -) log n).
For the preceding algorithm, we have merely applied the general principles of
the augmented influence graph to the case of computing the vertical decomposi-
tion of a set of segments. In fact, in this specific case, we may derive a simpler
algorithm, yet one that uses less storage. This algorithm does not need to keep
the conflict lists and maintains a non-augmented influence graph. It is outlined in
exercises 6.1, 6.2 and 6.3, and its performances are summarized in the following
theorem:
Theorem 6.4.2 Under the assumptions of dynamic randomized analyses, an in-

fluence graph can be used to maintain the vertical decomposition of a set of seg-
ments with the following performances. Let S be the current set of segments, n
be the size of S, and a be the number of intersecting pairs in S.
* The expected storage required by the algorithm is
O(n+a).
* Inserting a segment takes an average time of

O(log n + a).
n
* Deleting a segment takes an average time of
o ((1 +-)(t + t')),
where the parameters t and t' stand respectively for the complexity of the
operations on dictionaries and priority queues.
Therefore, the expected cost of a deletion is 0 ((1 + a) log log n) if we use per-
fect dynamic hashing coupled with stratified trees. It remains 0 ((1 + a) log n)
if we use balanced binary trees.
6. 5. Exercises 121
6.5 Exercises
Exercise 6.1 (Dynamic decomposition) Let us maintain dynamically the decom-
position of a set of segments using an influence graph. Show that the creator of a new
trapezoid, or a trapezoid unhooked during a deletion, is also the creator of at least one
destroyed trapezoid.
Hint: The proof of this fact relies on the two additional properties possessed by the
influence graph of a decomposition:
1. The influence domain of an internal node is contained in the union of the influence
domains of its children.
2. If an object Ok is the determinant of an internal node, it necessarily is a determi-
nant of at least one child of this node.
Let Oj be a segment creating a new trapezoid, or a trapezoid unhooked during the

deletion of Ok. As in the entire chapter, the segments are indexed by their chronological
rank and Si stands for the set of the first i segments (in chronological order). To prove
our assertion, we investigate the addition of Ok and successively Qi to the decomposition
Dec(S' 1_). The decompositions obtained are then successively Dec(SI-1) and Dec(Si).
It can be shown that there is a region H in Dec(Si) determined by a set that contains
both Ok and 01.
Exercise 6.2 (Dynamic decomposition) Let us assume that we use an influence

graph to dynamically maintain the decomposition of a set S of segments. Here, we
consider the deletion of a segment Ok* We still use the notation of section 6.4. In partic-
ular, the segments are indexed by their chronological rank. Let 01 be the segment to be
reinserted during the deletion of Ok* Show that for any critical region F in the critical
zone Z 1 _1 that conflicts with 01, there is at least one destroyed region H, created by 01,
that intersects F, and satisfies at least one of the following conditions:
1. F contains a wall of H that stems from an endpoint of Oj, or an intersection point

on 0, and that butts against Ok (see figure 6.6a),
2. F contains a wall of H that stems from an endpoint of Ok, or an intersection point
on Ok, and that butts against Qi (see figure 6.6b),
3. F contains a wall of H that stems from the intersection 01 l Ok (see figure 6.6c),
4. F is bounded by two walls, one stemming from a point on Ok, the other stemming
from a point on 01, and both walls are contained within F (see figure 6.6d),
5. 01 and Ok support the floor and ceiling of H, both of which intersect F (see
figure 6.6e).
Exercise 6.3 (Dynamic decomposition) The aim of this exercise is to show how we
may dynamically maintain the decomposition of a set S of segments using a simple
influence graph, without the conflict lists.
01
> Ok
9k
/
01
- V
kA') (e)
Figure 6.6. Detecting conflicts in critical zone. Region F is shaded, and region H within
is emphasized.
The segments that must be reinserted during a deletion are the creators of destroyed
regions (see exercise 6.1) and can be detected during the locating phase.
Let 01 be one of the segments to be reinserted during the deletion of Ok. To retrieve
all the critical regions that conflict with O, the algorithm considers in turn the destroyed
regions H with creator 01, and selects the critical regions F related to H by one of the
five cases described in exercise 6.2.
For this, the deletion algorithm maintains an augmented dictionary A, storing the
sequence ordered along Ok of critical regions intersected by Ok. Let H be one of the
destroyed regions, created by 01. If H has a wall that stems from a point on 01 and
butts against Ok, or a wall stemming from 01 n Ok, this wall is located in the structure
A, and the critical region containing this wall is retrieved. If H has two walls stemming
from a point on Ok and from a point on 01, the region containing the wall stemming
from the point on Ok is searched for in A, and it is selected if it also contains the wall
of H stemming from the point on 01. Lastly, if 01 and Ok support the floor and ceiling
of H, the right wall of H is searched for in A, and any critical region that intersects the
floor and the ceiling of H is selected.
1. The selected region obviously conflicts with 01. As shown in exercise 6.2, any
critical region that conflicts with H is selected. Show that such a region can be selected
at most 16 times.
To speed up the locating phase, the algorithm maintains the lists of nodes killed by
each object stored in the structure. To perform the deletion, the algorithm proceeds
along the following lines.
Locating. The algorithm traverses the influence graph starting on the nodes killed
by Ok, and visits the destroyed or unhooked nodes. During this traversal, the algorithm
sets up a dictionary D that stores the destroyed and unhooked nodes, and a list C of the
creators of the destroyed nodes.
Rebuilding. The list £ is sorted by chronological order, for instance by storing the
elements in a priority queue, and extracting them in order. The redundant elements are
extracted only once. The dictionary A initially stores the regions killed by Ok*
The objects of £ are processed in chronological order. For each object 01, the critical
regions that conflict with Qi are selected as explained above. The remaining operations
are identical to those in the algorithm of section 6.4. The conflict lists of the new regions
do not have to be computed. On the other hand, the dictionary A must be updated.
2. Show that the performances of this algorithm are those given by theorem 6.4.2.
Exercise 6.4 (Lazy dynamic algorithms) In this exercise, we propose a lazy method
to dynamically maintain the decomposition of a set of segments. For simplicity, let us
assume that the segments do not intersect. The algorithm maintains an influence graph
in the following lazy fashion:
1. The graph is a mere influence graph, no conflict lists are needed.
2. During an insertion, the nodes corresponding to the new trapezoids are hooked to
the nodes corresponding to the killed trapezoids as in the algorithms described in
subsection 5.3.2 and section 6.4.
3. During a deletion, the nodes corresponding to the new trapezoids are hooked to
leaves of the graph that correspond to destroyed trapezoids. More precisely, a node
corresponding to a new trapezoid is hooked to leaves of the graph that correspond
to the destroyed nodes that have a non-empty intersection with the new trapezoid.
No node is removed from the graph.
4. The algorithm keeps the age of the current graph in a counter, meaning the total
number of operations (insertions and deletions) performed on this graph. Each
time the number of segments effectively present falls below half the number stored
in this counter, the algorithm builds the influence graph anew by inserting the
segments effectively present into a brand new structure.
1. Show that when 0(n) segments are stored in the structure, the expected cost of an
insertion or a location query is still O(logn).
2. The cost of the periodic recasting of the graph is shared among all the deletions.
Show that the amortized complexity of a deletion is still 0(log n) on the average. (Recall
that the segments do not intersect, by assumption.)
Hint: It will be noted that the number of children of a node in the influence graph is not
bounded any more. The analysis must then have recourse to biregions (see exercise 5.7)
to estimate the expected complexity of the locating phases.

The approach discussed in this chapter consists in forgetting deleted objects altogether,
and restoring the structure to the exact state which it would have been in, had this object
never been inserted. The first algorithm following this approach is that of Devillers,
Meiser, and Teillaud [81] which maintains the Delaunay triangulation of a set of points
in the plane. The algorithm by Clarkson, Mehlhorn, and Seidel [70] uses the same
approach to maintain the convex hull of a set of points in any dimension. The method
was then abstracted by Dobrindt and Yvinec [86]. A similar approach is also discussed
by Mulmuley [1761, whose book is the most comprehensive reference on this topic.
There is another way to dynamize randomized incremental algorithms. This approach,
developed by Schwarzkopf [198, 199], can be labeled as lazy. As outlined in exercise 6.4,
it consists in not removing from the structure the elements that should disappear upon
deletions. These elements are marked as destroyed, but remain physically present, and
still serve for all subsequent locating phases. Naturally, the structure may only grow.
When deletions outnumber insertions, the number of objects still present in the structure
is less than half the number of objects still stored, and the algorithm completely rebuilds
the structure from scratch, by inserting one by one the objects that were not previously
removed.
Finally, we shall only touch the topic of randomized or derandomized dynamic struc-
tures which efficiently handle repetitive queries on a given set of objects, while allowing
objects to be inserted into or deleted from this set. These structures embody the dy-
namic version of randomized divide-and-conquer structures, discussed in the notes of the
previous chapter. These dynamic versions can be found in the works by Mulmuley [175],
Mulmuley and Sen [178], Matougek and Schwarzkopf [153, 156], Agarwal, Eppstein, and
Matousek [3] and Agarwal and Matougek [4].
Part II
Convex hulls
Convexity is one of the oldest concepts in mathematics. It already appears
in the works of Archimedes, around three centuries B.C. It was not until the
1950s, however, that this theme developed widely in the works of modern math-
ematicians. Convexity is a fundamental notion for computational geometry, at
the core of many computer engineering applications, for instance in robotics,
computer graphics, or optimization.
A convex set has the basic property that it contains the segment joining any two
of its points. This property guarantees that a convex object has no hole or bump,
is not hollow, and always contains its center of gravity. Convexity is a purely affine
notion: no norm or distance is needed to express the property of being convex.
Any convex set can be expressed as the convex hull of a certain point set, that
is, the smallest convex set that contains those points. It can also be expressed
as the intersection of a set of half-spaces. In the following chapters, we will be
interested in linear convex sets. These can be defined as convex hulls of a finite
number of points, or intersections of a finite number of half-spaces. Traditionally,
a bounded linear convex set is called a polytope. We follow the tradition here, but
we understand the word polytope as a shorthand for bounded polytope. This lets
us speak of an unbounded polytope for the non-bounded intersection of a finite set
of half-spaces.
In chapter 7, we recall the definitions relevant to polytopes, their facial struc-
ture, and their combinatorial properties. We introduce the notion of polarity as
a dual transform on polytopes, and the notions of projective spaces and oriented
projective spaces to extend the above definitions and results to unbounded poly-
topes. In chapter 8, we present solutions to one of the most fundamental problems
of computational geometry, namely that of computing the convex hull of a finite
number of points. Chapter 9 contains algorithms which work only in dimension
2 or 3. Lastly, chapter 10 tackles the related linear programming problem, where
polytopes are given as intersections of a finite number of half-spaces.
Chapter 7
Polytopes
A polytope is defined as the convex hull of a finite number of points, or also as

the bounded intersection of a finite number of half-spaces. Section 7.1 recalls
the equivalence of these definitions, and gives the definition of the faces of a
polytope. Polarity is also introduced in this section. The polarity centered at 0
is a dual transform between points and hyperplanes in Euclidean spaces which
induces a duality on the set of polytopes containing the center 0. Simple and
simplicial polytopes are also defined in this section. Section 7.2 takes a close
interest in the combinatorics of polytopes. It contains a proof of Euler's relation
and the Dehn-Sommerville relations. Euler's relation is the only linear relation
between the numbers of faces of each dimension of any polytope, and the Dehn-
Sommerville relations are linear relations satisfied by simple polytopes. These
relations can be used to show the celebrated upper bound theorem which bounds
the number of faces of all dimensions of a d-dimensional polytope as a function
of the number of its vertices, or facets. Considering cyclic polytopes shows that
the upper bound theorem yields the best possible asymptotic bound. Linear
unbounded convex sets enjoy similar properties and are frequently encountered
in the rest of this book. Section 7.3 extends these definitions and properties to
unbounded polytopes. A simple method to enforce coherence in these definitions
is to consider the Euclidean space as embedded in the oriented projective space,
an oriented version of the classical projective space.
7.1 Definitions
7.1.1 Convex hulls, polytopes
Let A be a set of points in Ed. A linear combination of points in A is a sum of

the kind EkI 1 AiAi, where k is an integer, and .for all i = 1, ... , k, Ai is a real
128 Chapter 7. Polytopes
and Ai a point in A. A linear combination Ek=1 AAi is affine if

k
The set of affine linear combinations of points in A generates an affine subspace

of Ed called the affine hull of A. For instance, the affine hull of two distinct
points is the line going through these two points. More generally, k + 1 points
are said to be affinely independent if they generate an affine space of dimension
k.
A linear combination ZIk-1 AiAi is convex if
k
Ai I=1and Vi E {1,..., k}, Ai > .
i=1
A set A of points is convex if it is stable under convex combinations. Since

the set of all convex combinations of two points P and Q is the segment PQ,
the convexity of A is equivalent to the following geometric condition: for any two
points P and Q in A, the segment PQ is entirely contained in A. The intersection
of two convex sets is also convex.
The dimension of a convex set A is defined as the dimension of its affine hull
aff (A). If A is convex, its interior as a subset of the topological subspace aff(A)
is not empty. It is called the relative interiorof A.
Let A be a set of points in Ed. The convex hull of A, denoted by conv(A),
is formed by the set of all possible linear convex combinations of points in A.
Any convex set containing A must also contain its convex hull: the convex hull
of A is thus the smallest convex set in Ed that contains A, or equivalently the
intersection of all the convex sets that contain A.
The convex hull of a finite set of points in Ed is called a polytope. It is a closed
bounded subset of Ed. A polytope of dimension k is also called a k-polytope.
The convex hull of k + 1 affinely independent points is a particular k-polytope
called a simplex or also k-simplex. The convex hull of two affinely independent
points A and B is the segment AB; the convex hull of three affinely independent
points A, B, and C is the triangle ABC; finally, the convex hull of four affinely
independent points A, B, C, and D is the tetrahedron ABCD.
7.1.2 Faces of a polytope

Any hyperplane H divides the space Ed into two half-spaces situated on either
side of H. We write H+ and H- for these two open half-spaces, and H+ and
H- for their topological closure. Hence,
Ed = H+ UHUH-
7.1. Definitions 129
H+=H+UH, H-=H-UH.
Consider a d-polytope P. A hyperplane H supports P, and is called a supporting
hyperplane of P, if H n P is not empty and P is entirely contained in one of the
closed half-spaces H+ or H-. The intersection H n P of the polytope P with
a supporting hyperplane H is called a face of the polytope P. Faces are convex
subsets of Ed, with a dimension ranging from 0 to d - 1. To these faces, called
the proper faces of 'P, we add two faces called improper: the empty face whose
dimension is set to -1 by convention, and the polytope P itself, of dimension d.
A face of dimension j is also called a j-face. A 0-face is called a vertex, a 1-face
is called an edge, and a (d - 1)-face is called a facet of the polytope.
If F is a face of P and H a supporting hyperplane of P such that F = H n P,
H is said to support P along F.
Theorem 7.1.1 The boundary of a polytope is the union of its proper faces.
Proof. Consider a polytope P. It is easy to show that the union of the faces
of P is included in the boundary of P. Indeed, let F be a face of 'P, H a
hyperplane supporting P along F, and H+ the half-space bounded by H that
contains P. Any point X in F belongs to P and to H, and a neighborhood of
this point contains points that do not belong to P. The converse inclusion (of
the boundary in the union of the proper faces) results from a general theorem
on bounded closed convex sets of Ed, stated in exercise 7.5. It is a consequence
of this theorem that there is a supporting hyperplane passing through any point
of the boundary of a polytope 'P; thus every point of the boundary belongs to a
supporting hyperplane and hence to a proper face of P. n
Theorem 7.1.2 A polytope has a finite number of faces. Faces of a polytope are
also polytopes.
Proof. Consider a polytope 'P, the convex hull conv(X) of a finite set of points
X. The theorem can be proved by showing that every proper face of P is the
convex hull of a subset of X. Indeed, let H be a supporting hyperplane of P and
let X' be the subset of the points of X that belong to H. We first show that
H nP = conv(X'). That conv(X') C HfnP is immediate. To prove the converse,
we show that any point of P that does not belong to conv(X') does not belong
to H. Let H(Y) = 0 be an equation of H and assume that P is contained in the
half-space H+ = {Y E Ed: H(Y) > 0}. For any point X' in X' or in conv(X'),
we have H(X') = 0, and for any point X in X \ X' or in conv(X \ X'), we have
H(X) > 0. Any point Y in P is a linear convex combination of points in X. If
Y does not belong to conv(X'), at least one of the coefficients of the points in
X \ X' in this combination is strictly positive, and thus H(Y) > 0. 0
Hi Hi1
I /
II
Figure 7.1. For the proof of theorem 7.1.3.
Theorem 7.1.3 A polytope is the convex hull of its vertices.
Proof. Let P be a polytope defined as the convex hull of a finite point set X.
By successively removing from X any point Xi that can be expressed as a linear
convex combination of the remaining points, we are left with a minimal subset X'
of X such that P = conv(X'). Let us now prove that any point of X' is a vertex of
P. Let Xi be a point of X'. Since X' is minimal, Xi does not belong to the convex
hull conv(X' \ {Xi}) of the other points, and the theorem stated in exercise 7.4
shows that there is a hyperplane Hi' that separates Xi from conv(X' \ {Xi}) (see
figure 7.1). The hyperplane Hi parallel to Hi' passing through Xi supports P and
contains only Xi among all points of X'. Now theorem 7.1.2 above shows that
Hi n P = conv({Xi}) = {Xi}.
The following two theorems are of central importance. They show that a poly-
tope might equivalently be defined as the bounded intersection of a set of closed
half-spaces.
Theorem 7.1.4 Any polytope is the intersection of a finite set of closed half-
spaces. More precisely, let P be a polytope, and {Fi: 1 < i < m} be the set of
its (d - 1)-faces, Hi the hyperplane that supports P along Fi, and Hi+ the closed
half-space bounded by Hi that contains P. Then:
m
i=1
Proof. The inclusion P C ni=1 Hi is trivial, since P is contained in all the

half-spaces Hi. To prove the converse, we show that any point which does not
belong to P does not belong to the intersection ni=l1 Hi. Let X be a point
not in P, and Y a point in the interior of P but not in the hyperplane passing
through X and some d - 1 vertices of P. Such a point exists, since the interior
of P is of dimension d and cannot be contained in the union of a finite number
of hyperplanes of dimension d - 1. Segment XY intersects the boundary of P in
a point Z (see figure 7.2). This point necessarily belongs to a proper face of P
and, from the choice of Y, cannot belong to a face of P of dimension j < d - 1.
Thus Z belongs to one of the facets Fi of P. Then Z belongs to the hyperplane
Hi, Y to the half-space Hi+ and X to the opposite half-space Hi-. E
The following theorem is the converse of the previous one.
Theorem 7.1.5 The intersection of a finite number of closed half-spaces, if it is

bounded, is a polytope.
Proof. The proof goes by induction on the dimension d of the space. In dimen-
sion 1, the theorem is trivial. Let
m
Q n Hi
t=1
71
be the bounded intersection of a finite number of half-spaces in Ed.

For any j such that 1 < j < m, let Fj = Hj n Q. Fj is thus a bounded
intersection of half-spaces in the hyperplane Hj identified with an affine space of
dimension d - 1. By the inductive hypothesis, Fj is a polytope in Hj and thus
a polytope of Ed as well. Let Vj be the set of vertices of Fj and V be the union
Uj- 1 Vj. We can now show that Q is the convex hull of V. Indeed,
* any point X on the boundary of Q belongs to one of the polytopes Fj hence

to conv(V);
* any point X that belongs to the interior of Q belongs to a segment XoXj
which is the intersection of some line passing through X with Q. Since
both X0 and Xl are on the boundary of Q, they belong to conv(V) and so

does X.
From this we may conclude that Q C conv(V). And since Q is convex and
contains V, the opposite inclusion is trivial. 0
Remark. If the intersection Q is of dimension d and if its expression 7.1 is
minimal, that is, for any j = 1,.. ., m,
Q ln Hi
i54j
then Fj is a d - 1 face of the d-polytope Q. To prove this, it suffices to prove

that Fj is not empty and that its dimension is d - 1. The relative interior of Fj
inside Hj can be expressed as
H~i ( )
and is therefore not empty. Indeed, the intersection nij Hi is neither empty,
nor entirely contained in H-, because Q is not empty. But this intersection is
not contained in Ht3
either, otherwise Ht3
could be removed from expression 7.1
without changing the intersection Q.
Theorem 7.1.1 shows that the boundary of a polytope is the union of its proper
faces, and the preceding remark shows that the union of the (d - 1)-faces gives
the boundary of a polytope. The following theorem shows more precisely that
any proper face of a polytope is entirely contained within a facet of the polytope.
Theorem 7.1.6 A properface of a polytope P is a face of a (d -1)-face of P.

Conversely, any face of a face of P is also a face of P.
Proof. 1. Let F be a proper face of 'P, H a hyperplane supporting P along F,

and X a point in the relative interior of F. Point X belongs to the boundary
of P and is therefore in a (d - 1)-face of P, say F1 . Let H1 be the hyperplane
supporting P along F1 . Point X belongs to the relative interior of face F and to
hyperplane H1 which supports P along F1, therefore the whole face F is included
in H1 , and thus in the facet F1 .
Moreover, if H+ is the half-space bounded by H that contains P, then F1 is
entirely contained in the half-space H1 n H+ of H1 . And since F is contained in
Ft,
F=HnP=HnPnHi = (HnHI)n(PnHi) = (HnHi)nF1 ,

F2
which shows that H1 n H is a hyperplane of H1 that supports F1 along F.

2. Let P be a d-polytope, F1 be a proper face of P, and F2 be a face of F1 .
Let H1 be a hyperplane of Ed supporting P along F1 and K a hyperplane of H1
supporting F1 along F2 . To show that F2 is a face of P, we rotate the hyperplane
H1 around K to obtain a hyperplane H of Ed that supports P along F2 (see
figure 7.3).
More precisely, suppose that the origin 0 of Ed belongs to F2 , and hence to
F1 . There is a vector N1 in Ed such that
Hl=f{XEEd X NI = °},
Hj ={X E Ed X. N > °},
where Hj+ is the half-space containing P. There is a vector L in H1 such that
K={XEH 1 X.L=O},
K+= {X EH1: X L >O},
where K+ is the half-space containing F1 . Let V(P) denote the set of vertices of
P, V(F1 ) the set of vertices of F1 , and V(F 2 ) the set of vertices of F2 . Let
V-L
,10 = min{-V-N : V E V(P) \ V(F1 )}
and let 71 <7qo. The hyperplane H of Ed defined by

H ={X Ed: X.( 1 N1 +L)==O}
supports P along F2 . Indeed, the hyperplane H contains all the vertices of V(F2 ),
and the half-space
H+ ={X E Ed: X
X( 7N 1 + L) >O}
contains all the vertices in V(P) \ V(Fj) and all the vertices in V(Fi) \ V(F 2 ). El
Theorem 7.1.7 Let P be a d-polytope.
1. The intersection of a family of faces of P is a face of P.
2. Any (d - 2)-face of P is the intersection of two (d - 1)-faces of P.

3. For any pair (j, k) of integers such that 0 < j < k < d, a i-face of P is the
intersection of all the k-faces of P that contain it.
Proof. 1. Let {FI, F2 ,..., Fr} be a family of faces of the polytope P. Let F
be the intersection ni= Fi. If F is empty, F is trivially a face of P. Otherwise
we choose for the origin of Ed a point 0 in F. For i = 1, . .. , r, we let Hi be a
hyperplane that supports P along Fi, and Ni be the vector of Ed such that
Hi = {X E Ed: X. N, = °},
and
PCH+={XEEd: X.N>Ž°}.
If N = Eri=1 Ni, the hyperplane H defined by
H = {X Ed: X *N=0}
supports P along F.
2. Let {FI, F2 , .. ., Fm} be the facets of polytope P. Let {H1 , H2 , . . ., Hm} be
the hyperplanes that support P along these facets. Let F be a (d - 2)-face of P.
From theorem 7.1.6, F is a (d - 2)-face of a facet Fj of P. From theorem 7.1.5,
facet Fj can be expressed as
F3 = H3 n -P= H, n (ki&j
k)=n()nH
k54j
and any (d - 2)-face F of Fj can be expressed as
F=(H, HHi) ( n
k/+ {i~j
(Hj n Hk)) = Hi n H n (n
kjiji,j}
k+
or equivalently
F=Hn HI nP=FnF3.
3. Using theorem 7.1.6, a straightforward induction on k (from k = d - 1 down

to k = j) shows that a j-face of P is a face of a k-face of P for any j < k < d -1.
Using the second assertion in theorem 7.1.7, it is also easy to prove by induction
on j (from j = d - 1 down to j = 0) that any j-face (O < j < d - 1) of a polytope
P is the intersection of all the (d - 1)-faces of P that contain it.
Let then j and k satisfy 0 < j < k < d - 2. Consider a j-face F of polytope
P. The intersection of all the k-faces of P that contain F is also a face of P that
contains F. To show that this face is precisely F, it suffices to show that F is
the intersection of some k-faces of P. From what was said above, F is a face of
a (k + 1)-face G of P, and thus F is the intersection of all the k-faces of G that
contain it. But k-faces of G are also k-faces of P and therefore F is indeed the
intersection of some k-faces of P. [1
Incidences and adjacencies
Two faces F and G of a polytope P are called incident if one is included in the
other, and if their respective dimensions differ by one. Two vertices of a polytope
are said to be adjacent if they are incident to some common edge. Two facets of a
polytope are said to be adjacentif they are incident to some common (d- 2)-face.
7.1.3 Polarity, dual of a polytope

The polarity of center 0 is a geometric one-to-one transformation between points
in Ed (except for the origin) and hyperplanes in Ed that do not pass through 0.
Let A be a point in Ed distinct from the origin 0. The polar hyperplane of A,
denoted by A*, is defined by
A* = {X E Ed A *.X=1}.
Let H be a hyperplane that does not contain the origin 0. The pole of H is the
point H* that satisfies
H*.X=1, VXEH.
This double transformation point-to-hyperplane and hyperplane-to-point is an

involution. Indeed, it is a simple task to check that if A differs from 0, then A*
does not contain the origin. Similarly, if H does not contain the origin, then H*
differs from 0. It is then easy to show that
A** = A and H** = H.
Lemma 7.1.8 The polarity of center 0 reverses the inclusion relationships be-
tween points and hyperplanes: a point A belongs to a hyperplane H if and only if
the pole H* of H belongs to the polar hyperplane A* of A.
Proof.
A E H -t~A H* = I H* XA*
In this subsection, for any hyperplane not containing 0, we denote by H+

the half-space bounded by H that contains 0, and by H- the other half-space
bounded by H:
H+={XEEd: X.H*<1}
H-={X EEd: X H* > 1}
Lemma 7.1.9 The polarity centered at 0 reverses the relative positions of a

point and a hyperplane with respect to the center 0: a point A belongs to the
half-space H+ if and only if the pole H* of H belongs to the half-space (A*)+,
and A belongs to H- if and only if H belongs to (A*)-.
Proof.
AeH+ A H*<1 H*EA*+

A E H- A H* > 1 4=-H*E A*-
Generally speaking, we call a duality any bijection that reverses inclusion rela-
tionships. The preceding relation shows that polarity centered at 0 is a duality,
and the polar hyperplane A* is often called the dual of A. Similarly, the pole H*
is often called the dual of hyperplane H.
The notion of duality extends naturally to polytopes: a polytope Q is dual to
a polytope P if there is a bijection between the faces of P and the faces of Q that
reverses inclusion relationships.
The following theorems show that it is possible to define a polar image P# for
any polytope P whose interior contains the origin 0.
The polar transformationcentered at 0 is closely linked to the polarity defined
above, but it associates points with half-spaces and not with hyperplanes. Let A
be a point of Ed. The polar image A# of A is the half-space A*+ bounded by A*
that contains the origin:
A#=A*-+ = YeEd : Y.A<1}.

For any set A of points in Ed, we define the polar image A# of A as the intersection
of the polar images of its points:
A#=I{YEEd : y*X<1, VXeA}.
or I
Note that this formula allows the definition to be extended to the case where A
contains the origin 0.
The two following facts are immediate consequences of the above definition:
1. The polar image A# of a set A is convex.
2. If A and B are two sets such that A C B, then B# C A#.
In the rest of this subsection, P stands for a polytope of Ed whose interior
contains the origin 0, and P# denotes the polar image of P.
Theorem 7.1.10 The polar polytope P# of a polytope P whose interiorcontains

o is a polytope whose interior contains 0.
Proof. Point 0 is inside the polar polytope 'P# because P is bounded. Indeed, if
P is contained within a ball B(O, p) centered at 0 with radius p, its polar image
'P# contains the ball B(O, l/p) centered at 0 with radius 1/p.
A dual argument shows that the image P# of P is bounded because the interior
of P contains 0. In fact, P contains an entire ball B(O, e) centered at 0 with
radius e. Thus P# is entirely contained in the ball B(O, 1/e) centered at 0 with
radius 1/E.
If the polytope P is the convex hull of n points,
P = conv({Xi, , Xn.X
}),
then P# is the intersection of the n half-spaces Xi+ bounded by the polar hy-
perplanes Xi of the point Xi,
n
P#= nxi*+.
Indeed, the inclusion P# C ni=l Xi+ is trivial. To prove the converse, we show
that every point that does not belong to P# does not belong to i=l Xi*+. Let
Y be a point that does not belong to P#. There is a point X that belongs to P
such that Y .X > 1. Since X is a linear convex combination of {X, ..., Xr,}, its
existence implies that Y - Xi > 1 for at least one of the points Xi, and thus Y
does not belong to Xi*.
The polar set P# of polytope P is the bounded intersection of a finite number
of half-spaces. It is thus a polytope, by theorem 7.1.4. El
Theorem 7.1.11 The polar transformation is an involution on the set of poly-

topes whose interiors contain 0, that is for any such polytope P,
P## = P.
138 1Chapter 7. Polytopes
Proof. P is included in 7P##, by definition. To prove the converse inclusion, we

show that any point Z that does not belong to P does not belong to 'P##. Let
Z be a point that does not belong to P. Since P is the intersection nl;?L 1 Ht of
the half-spaces Ht bounded by the hyperplanes supporting the facets of 'P, one
of these hyperplanes must separate Z from 'P. Let us call this hyperplane Hk.
The polytope 'P is contained in the closed half-space H+, and Z lies within H .
The point Hk, pole of Hk, satisfies
VX E P, X Hk < 1 and Z *Hk* > 1.

Therefore Hk* belongs to 'P##, and Z does not belong to 'P##. 0
Lemma 7.1.12 If A is a point of Ed on the boundary of a polytope P, the polar

hyperplane A* of A is a hyperplane that supports the polar polytope P#.
Proof. For any point A of P, the closed half-space A*+ contains the polytope
'P#. Moreover, if A belongs to the boundary of P, it belongs to one proper face
F of 'P and there is a supporting hyperplane H of P that passes through A. The
pole H* of H is a point that belongs to both A* and P#. Thus A* n P# is not
empty and A* is indeed a supporting hyperplane of 'P#. 0
Theorem 7.1.13 There exists a bijection between the faces of 'P and those of
'P# which reverses inclusion relationships. This bijection associates the k-faces
of 'P with the (d-1 - k) -faces of 'P#,for all k = O. . . , d-1.
Proof. With each face F of P, we associate the set
F*=I{YEP#: Y*X=1, VXCEF}.

The set F* is a face of 'P#. Indeed, F* can be expressed as
F*= n(P#nX*)
XEF
where X* is the polar hyperplane of X. From lemma 7.1.12, if X is a point of

F, then the hyperplane X* is a supporting hyperplane of 'P#, and X* n P# is
thus a face of P#. Therefore, F* is the intersection of a family of faces of 'P#,
and theorem 7.1.7 proves that it is a face of P# as well.
By the definition of the polar image F* of a face F, if F1 and F2 are two faces
of P such that F1 C F2 , then F2* c Fr*.
To prove that the map from F onto F* that maps a face of 'P to a face of P#
is bijective, we show that it is in fact an involution, that is
F** = F.
This, property is proved below for the proper faces of P and P#. In order to
extend it to improper faces, we note that the images of P and fP# are empty sets
because both P and 7P# have non-empty interiors. Therefore we can make the
convention that the d-dimensional face of P (resp. P#) corresponds to the empty
face of P# (resp. 7P).
Let now F be a proper face of P. Then
F** = {X E P##: X Y = 1, VY e F*}.

Since 'P## is simply P, the inclusion F C F** is immediate. The converse
inclusion can be shown by arguing that any point X of P not in F does not
belong to F**. Let X be such a point of P not in F, and H any hyperplane
supporting P along F. Then X belongs to H+ and X *H* < 1. Nevertheless,
the pole H* of H lies within the face F*, which proves that X does not belong
to F**.
Finally, let us prove the assertion about the dimensions. Let F be a k-face of
P. If F is not a proper face,then k = -1 or k = d and the assertion is true. Let F
be a k-face of P, with 0 < k < d - 1. Then F contains k + 1 affinely independent
points, and F* is contained within the intersection of k + 1 hyperplanes whose
equations are linearly independent. The dimension of F* is thus at most d -1-k.
But since F** = F, this dimension must equal d - 1 - k exactly. 0
We have seen that the vertices of P correspond to facets of 'P#, and the converse
is true:
* If {P1i, P2 ,. . ., P.} are the vertices of P, then {PI*, P2*, . . ., P,} are the hy-
perplanes supporting 'P# along its (d - 1)-faces.
n
P = conv({P1,P2,...,P}), -P#= z=ln P*
* If {Hi, H 2 ,..., Hm}
, are the hyperplanes supporting P along its (d - 1)-
faces, then {H,*, H2*, . . ., Hm} are the vertices of P#.
m
,P=n Hj+ 4='P#= conv(IH*, H2*,.* , Hm}).
j=1
Finally, the following properties can be easily proved from the preceding ones.
1. If point A belongs to polytope P, the polar hyperplane A* avoids the polar

polytope P#.
2. If A is a point lying outside P, the polar hyperplane A* intersects the
interior of P#.
3. When A takes all positions in the relative interior of a face F of P, the

polar hyperplane A* describes the set of hyperplanes supporting 'P# along
F*.
4. If the origin 0 lies in the interior of two polytopes P1 and P2, then
(conv(Pi UP 2 ))# = ,P n '#
7.1.4 Simple and simplicial polytopes

We recall that a set of k + 1 points in Ed is affinely independent if its affine linear
combinations generate an affine subspace of Ed of dimension k. A finite set A of
points in Ed is in general position if any subset of A with at most d + 1 points is
affinely independent.
If A is a set of points in general position, no affine subspace of dimension j
may contain more than j + 1 points of A.
A set Kt of hyperplanes is in general position if, for any j < d, the intersection
of any j hyperplanes in X is of dimension d - j, and the intersection of any d + 1
hyperplanes in X is empty.
Let X be a finite set of hyperplanes, 0 a point not in any of the hyperplanes
in X, and X* the set of poles of the hyperplanes of Kd for the polarity centered
at 0. The set X of hyperplanes is in general position if and only if the set X* of
points is also in general position.
A k-simplex is the convex hull of k + 1 affinely independent points. Let S be a
k-simplex, the convex hull of a set A = {Ao, .. . , Ak} of k+ 1 affinely independent
points. Any subset A' of A of cardinality k' + 1 (0 < k' < k) defines a k'-simplex
conv(A') which is a face of S. Therefore, a k-simplex A has exactly ( +I
faces of dimension j (0 < j < k). A j-face of a k-simplex is incident to j + 1
(j- 1)-faces for any 0 < j < k, and to k-j (j +I1)-faces for any -1 < j < k-1.
A polytope is called simplicial if all its proper faces are simplices. By virtue of
theorem 7.1.6, it is enough to require that its facets have exactly d vertices, and
thus are (d - 1)-simplices. The convex hull of a set of points in general position is
a simplicial polytope. Note that this is a sufficient but not necessary condition:
the vertices of a simplicial polytope need not be in general position. Indeed, d
vertices may lie in the same hyperplane, provided that this hyperplane does not
support the polytope along a facet.
A polytope is simple if it is dual to a simplicial polytope. Therefore, a polytope
P is simple if any of its vertices belongs to exactly d facets.
Simplices are both simple and simplicial polytopes: they are the only polytopes
to possess this property (see exercise 7.6).
7.2. The combinatorics of polytopes 141
By using the bijection between the faces of polytope P and its dual, we can
easily prove the following lemma which will be useful in establishing the Dehn-
Sommerville relations (theorem 7.2.2) satisfied by any simple polytope:
Lemma 7.1.14 For any 0 < j < k < d - 1, any i-face of a simple polytope P is
a face of ( dj) k-faces of P.
7.2 The combinatorics of polytopes

Except in the case of simplices, or polytopes in dimensions no greater than 2, the
number of faces of a d-polytope is not entirely determined by the number of its
vertices, nor by that of its facets. The upper bound theorem, stated and proved
in subsection 7.2.3, gives an asymptotic upper bound of 0(nLd/2J) for the total
number of faces of a d-polytope that has n vertices, or respectively n facets. The
study of cyclic polytopes (in subsection 7.2.4) shows that this bound is optimal.
If P is a polytope, we denote by nk(P) (or simply nk when P is unambigu-
ously understood) the number of k-faces of P, for -1 < k < d. In particular,
n-1 = nd = 1. The proof of the upper bound theorem relies on a set of linear
relations satisfied by the numbers nk(P) of faces of a polytope. One of them
is Euler's relation (subsection 7.2.1), and is the only linear relation binding the
numbers nk and satisfied for any polytope (simple or not). The other relations
are known as the Dehn-Sommerville relations and are satisfied by all simple
polytopes (subsection 7.2.2).
7.2.1 Euler's relation
Theorem 7.2.1 (Euler's relation) The numbers nk(P) (O < k < d - 1) of

k-faces of a d-polytope P are bound by the relation
d-l
S-_1)knk(P) 1- (-)d,
k=O
or, if we also sum over the improper faces of 'P,

d
E (-1)knk((P) = 0.
k= -1
Proof. The proof we present here goes by induction on the dimension d of the
polytope. The base case is proved easily since, in one dimension, a polytope has
only two proper faces, namely its vertices, and thus satisfies Euler's relation.
Let P be a polytope of dimension d and let n = no(P) be its number of vertices.

We may always assume that the xd-coordinates of any two vertices are distinct, by
choosing the coordinate system appropriately. Therefore, a horizontal hyperplane
(that is, perpendicular to the xd-axis) contains at most one vertex of P.
Let {P1 , P2 , . .. , Pn} be the set of vertices of P sorted by increasing xd-coord-
inates. Consider a family of 2n - 1 horizontal hyperplanes {H1, H2 , . .. , H2n-l}
such that:
* the hyperplane H2 i- 1 (i = 1,...,n) goes through the vertex Pi of P,
* the hyperplane H2 j (i = 1,..., n- 1) passes between H 2 j j and H 2j+i.
For each face F of P and each hyperplane Hj, in this family, we define a
signature: Xj(F) = 1 if Hj intersects the relative interior of F, and Xj(F) = 0
otherwise.
Consider a face F, and call P, (resp. Pm) its vertex with minimal (resp. max-
imal) xd-coordinate. The horizontal hyperplanes intersecting the relative inte-
rior of F lie strictly between horizontal hyperplanes H2 1- 1 and H2 mi-that pass
through PI and Pm respectively. If face F is of dimension k > 1, then I and m are
distinct integers, and the number of hyperplanes with even indices that intersect
the relative interior of F is one more than the number of hyperplanes with odd
indices that intersect the relative interior of F, whence
2n-2
1 = E (- 1)'X(F).
j=2
Summing this relation over the set TFk(P) of k-faces of P, we get

2n-2
nk(P = E (-S)jXj(F),
ForFk(?) j=2
and summing over all k,

d-1 2n-2 d-1
Z(-l )knk(P) = E (-1)j 5(_)k E )Xi(F). (7.2)
k=1 j=2 k=1 FEFk(P)
Each hyperplane Hj for j = 2, .. ., 2n-2 intersects polytope P along a polytope

Pj = P n Hj of dimension d - 1, to which we can apply the inductive hypothesis:
d-2 d-1
Z(_-1)kknk pj) = E(- 1)k-nk-l(Pj) = 1 - (_l)d-1. (7.3)
k=O k=1
When j is even, any (k - 1)-face of 'Pj is the intersection of a k-face of P with

hyperplane Hj, and
nk-I(Pj) = E Xj (F), k = 1, . .. , d -1. (7.4

FE.k (P)
When j is odd, any (k - 1)-face of 'Pj is the intersection of a k-face of P with

hyperplane Hj, except for the vertex Pj that belongs to Hj. Therefore,
no(Pj) = 1 + E Xj(F), (7.5)

FEY1 (P)
nk-l(Pj) = E Xj(F), k = 2, .. . , d-1. (7.6)

Fe.Fk(P)
When j is even, we can use equations 7.3 and 7.4 to get

d-1 d-1
(-1)k E Xj(F) = Z(-1)knk-('Pj) = -1 + (-1 )d 1. (7.7)

k=1 FEFk(P) k=1
When j is odd, using equations 7.3, 7.5 and 7.6, we obtain

d-1 d-1
S (_I)k E Xj(F) = Z(-1)knk-l('Pj) + 1 = (_i)d 1. (7.8)
k=1 FECk(P) k=l
It now suffices to multiply relations 7.7 and 7.8 by (-1)i and to sum over
j = 1, . . . , 2n - 1 to obtain by use of equation 7.2, noticing that there are n - 1
even and n - 2 odd relations:
d-1
Z(-1)knk('P) = (_1)d-1 - (n -1).

k=1
Recall now that n is the number no(P) of vertices of 'P. In the last equation, we
may now recognize Euler's relation for polytope P. [1
In the case of a 2-polytope, Euler's relation, written as
no(P) - ni(P) = 0,
expresses the fact that a polygon has as many vertices as edges. In the case of a
3-polytope, the relation is a bit more interesting and can be written as
no('P) - n ('P) + n2(P) = 2.

7.2.2 The Dehn-Sommerville relations

Theorem 7.2.2 (Dehn-Sommerville relations) The numbers nj(P) of
j-faces of a simple d-polytope satisfy the d + 1 relations
E(-')i (
j=o
d-k )7(P) = nk(P), k = O.. .. ,d.
Proof. Let P be a simple d-polytope. The Dehn-Sommerville relation for k = d

is none other than Euler's relation for P. Suppose now that k < d - 1. Any
k-face F of P is a k-polytope, and thus satisfies Euler's relation
k
Z
j=-l
(-1)jnj(F) = 0.
Summing over the set .Fk(P) of k-faces of P, we get

k k
E E (- l)jnj (F) = E (- l)j E nj (F) = O.

F C-(F (P ) 1 j=-1 FEFk (P)
The sum
Z nj (F)
FEYk (P)
is exactly the number of pairs (F, G) of faces of P, where F is a k-face and

G is a j-face entirely contained in F. Since P is a simple polytope, for each
0 < j < k < d -1, any j-face of P is contained in exactly (d k-faces of
P (lemma 7.1.14). Therefore,
S
FcE-k (P)
nj(F)= d ) ni(P), (O< j < k < d-1).
Finally, we get
-nk(p) + Y,(-')i ( dk)njP=O

j=o
which is exactly the (k + 1)-st Dehn-Sommerville relation for polytope P.

The Dehn-Sommerville relations for a simple 3-polytope are
no = no
3no-ni = ni
3no-2n1 + n2 = n2
no-nl + n2-n3 = n3 .
-
The first relation is trivial, the following two are equivalent, and the last is pre-
cisely Euler's relation. They can be compacted into two linearly independent
equations binding the numbers no, ni, and n2 of proper faces of a simple 3-
polytope. Fixing the number n = n2 of facets, these relations may be expressed
as
no=2n-4, nl=z3n-6.
This proves the following theorem:
Theorem 7.2.3 A simple 3-polytope with n facets has exactly 2n - 4 vertices

and 3n - 6 edges,
and its dual counterpart:
Theorem 7.2.4 A simplicial 3-polytope with n vertices has exactly 2n -4 facets

and 3n - 6 edges.
The following subsection shows that the Dehn-Sommerville relations alone can
be used to derive an upper bound on the number of faces of any polytope as a
function of its number of vertices or facets.
7.2.3 The upper bound theorem
Theorem 7.2.5 (Upper bound theorem) Any d-polytope with n facets (or n
vertices) has at most O(n[d/ 2 i) faces of all dimensions and O(nLd/ 2 ]) pairs of
incident faces of all dimensions.
Proof. We are interested first in simple polytopes. If P is a simple polytope of

dimension d, the Dehn-Sommerville relations,
kdj
E(-W) d - Ok)nj (P) = nk (P), (k = O,. . .., d),
j=o
yield d + 1 linear relations between the d numbers no, n 1 ,. . . , nd-1 of proper faces
of polytope P. The first relation (obtained for k = 0) is trivial, and the others
are not all linearly independent. But one may prove easily that the odd relations
(those that correspond to odd values of k) are linearly independent. Indeed, the
coefficients of n2p+l in the equations obtained for k = 2q+ 1, with p and q ranging
from 0 to [4- 1 j, form a triangular matrix. Thus the Dehn-Sommerville relations
form a system of rank at least
r=L + 1j.
In fact, it can be shown that there are exactly r linearly independent relations
among the Dehn-Sommerville relations (see exercise 7.7). Moreover, it can be
shown that the Dehn-Somerville system can be solved for the variables nj, j =
0 ... ., r - 1, yielding an expression for these variables as a linear combination of
the nj's, j = r, . .. , d (see exercise 7.8). If the simple polytope has n facets, there
is a trivial bound, for j > r,
( n j )=O(n d-j)
on its number of j-faces. Indeed, lemma 7.1.14 shows that a j-face of a simple d-
polytope is the intersection of d - j facets. We conclude that a simple d-polytope
with n facets has 0(nLd/2j) faces. In a simple polytope, k-faces (k < d) are
incident to d - k (k + 1)-faces; thus the number of pairs of incident faces is also
0(0Ld/2j). We therefore have proved the theorem for a simple polytope with n
facets.
A dual statement of the theorem also shows that the theorem is true for sim-
plicial d-polytopes with n vertices. To extend this result to arbitrary polytopes,
it suffices to show that simple and simplicial polytopes maximize the number of
faces and incidences between faces. The following perturbation argument shows
that the numbers of faces and incidences of faces of a non-simplicial d-polytope
are less than those of some simplicial d-polytope obtained by slightly perturb-
ing the vertices. Let P be a non-simplicial d-polytope, and n be the number of
its vertices. Each face of P is the convex hull of its vertices and may therefore
be triangulated, or in other words decomposed into a union of simplices whose
vertices are the vertices of that face. In a triangulation,1 each face F of P is ex-
pressed as the union of simplices whose relative interiors form a partition of F. A
simple scheme to triangulate a d-polytope and its faces is to proceed recursively,
or equivalently in a bottom-up fashion, as follows. Let F be a (k + 1)-face of P.
To triangulate F, we choose a vertex A of F, and consider the (k + 1)-simplices
conv(A, T), where T ranges over the k-simplices in the recursively obtained trian-
gulation of the k-faces of F which do not contain A. The number of faces of the
triangulation is at least the number of faces of P. Slightly perturbing the vertices
of P while keeping the union of the simplices in the triangulation convex (and
this can always be done, see exercise 7.10) yields a simplicial polytope P' whose
faces are in one-to-one correspondence with the simplices in the triangulation of
P. The numbers of faces and incidences of P' are thus strictly greater than their
counterparts for P.
'Triangulations are studied at length in chapters 11, 12, and 13.

7.2.4 Cyclic polytopes
In this subsection, we prove that the bound given in the upper bound theorem
(theorem 7.2.5) is optimal. For this, we introduce a particular class of polytopes,
and show that their numbers of faces and incidences achieve the bound given in
the upper bound theorem.
The moment curve is the curve Md in Ed followed by a point MT parameterized
by a real number T:
Md {M(-T) = (T, T2,. .. ,Td),T E RI}.
Lemma 7.2.6 Any subset of k < (d + 1) points on the moment curve is linearly
independent.
Proof. Consider d+1 points {Mo, M1 , .. ., Md} on the moment curve, for the val-
ues {TO, Ti, . . ., rd} of the parameter. The determinant formed by the coordinates
of these points is
ro r2 __ od
To T02 ... Td
12ITi = (Td -Ti)
Td T2,,Tdd O<i<j<d
1rd Td ... d
the so-called Van Der Monde determinant, and does not vanish when the Ti's are
pairwise distinct. 5
A consequence of this lemma is that any hyperplane in Ed intersects the moment
curve in at most d points.
A cyclic polytope in Ed is the convex hull of n > d + 1 points on the moment
curve. By the above lemma, a cyclic polytope is simplicial.
Let now P be a cyclic polytope in Ed, the convex hull of n points {M 1 , M2 ,
... , Mn} of Md with respective parameters {T1 T2, ....., r}. Let I a subset of the
set of indices {1, 2,.. . , n}, of cardinality k < d/2, and consider the polynomial
7r1 (T) = H(T-T,)2-

iEI
This polynomial has degree 2k < d, and there is a point H* in Ed and a real ho
which may be used to interpret 1rz(T) in the form
7rz(r) = H* *M(T) - ho, M(T) G Md.

The polynomial 7Fr2T) is positive and vanishes exactly at T = Ti, i E I. The
hyperplane H defined by the equation
H* X = ho
is thus a hyperplane that supports P along the (k - 1)-face conv({Mi, i E I}).

We conclude that the convex hull of any subset of k < d/2 vertices of P is a face
of P, which proves the following theorem:
Theorem 7.2.7 (Cyclic polytope) A cyclic polytope with n vertices has

(n) (k - 1)-faces, for all 0 < k < d/2.
Considering the dual of a cyclic polytope, we can also prove:
Theorem 7.2.8 For any integer n > d + 1, there is a polytope in Ed that has n
facets and exactly ( n) (d - k)-faces, for all 0 < k < d/2.
7.3 Projective polytopes, unbounded polytopes

The study of polytopes shows the combinatorial equivalence that exists between
the convex hull of a finite set of points and the intersection of a finite set of half-
spaces, when it is bounded. In the following chapters (and especially in chapter 14
on hyperplane arrangements and in chapter 17 on Voronoi diagrams), convex sets
defined as the intersection of a finite set of half-spaces play a special role. These
intersections are not necessarily bounded, however. The projective space extends
the Euclidean space Ed by adding points at infinity. Using projective geometry
yields a unified treatment of bounded and unbounded subsets. One drawback of
projective spaces, however, is their lack of orientation: for instance, we cannot
define the notion of segment joining two points or the position of a point with
respect to a hyperplane without ambiguity. As a consequence of this lack of orien-
tation, the notions of convexity and of half-spaces have no meaning in projective
geometry. Oriented projective geometry allows us to salvage these notions and
at the same time keep a complete duality between convex hulls of a finite set of
points and intersections of a finite set of hyperplanes. A presentation that uses
oriented projective space makes for an easy extension of the combinatorial results
on polytopes to unbounded polytopes of Ed.
7.3.1 Projective spaces
We define a vector line, or line for short, as a one-dimensional subspace in a vector

space. Formally, a projective space of dimension d is the space of all vector lines
in a vector space Vd+1 of dimension d + 1. The projective subspaces are formed
by the subsets of those vector lines that belong to vector subspaces in Vd+l,. The
subset of those vector lines that belong to a vector subspace of dimension k + 1 in
Vd+1 is a projective subspace of dimension k, also called a projective k-subspace.
7.3. Projective polytopes, unbounded polytopes 149
L(X)
- I N
XEd
Figure 7.4. Projective subspaces.
More concretely, we can embed the d-dimensional space Ed in a projective space

pd by the following construction. Consider, in the (d+ 1)-dimensional space Ed+1,
the embedding of Ed as a hyperplane in Ed+l. In Ed+l, we set the origin at a
point Q that does not belong to the hyperplane Ed. The projective space pd
of dimension d is defined as the set of all lines in Ed+1 that pass through Q.
Alternatively, it can be defined as the quotient of Ed+1 \ {Q} by the equivalence
relation 1Z such that X 1Z X' if there exists a real A 7 0 such that X = AX'.
Any affine subspace of dimension k + 1 in Ed+1 that contains Q corresponds to a
projective subspace of dimension k in Pd, consisting of all the lines in this subspace
that pass through Q.
To avoid confusion, elements of Ed+' are systematically denoted by an under-
lined symbol, and elements of pd by a boldface symbol. Points in the projective
space pd are lines in Ed+1 that pass through Q. The space Ed itself is embedded
into Ed+l, so any point X in Ed corresponds to a point X in Ed+l, and the line
passing through Q and X is therefore a point X in pd. Reciprocally, any point
X in pd is a line L(X) in Ed+1 that passes through Q. If this line L(X) is
not parallel to the hyperplane Ed, X corresponds to the point X in Ed at the
intersection of Ed and L(X) (see figure 7.4). The points in pd which correspond
to lines in Ed+1 parallel to Ed do not correspond to points in Ed; they are called
the points at infinity. The points at infinity form a subspace of dimension d - 1
in pd, that is a hyperplane in pd called the hyperplane at infinity and denoted by
Hoo. If X is a point in Ed, we systematically denote by X the point in Ed+l that
corresponds to X and by X the corresponding projective point. For each point
X in pd, we denote by L(X) the corresponding line in Ed+l.
Any affine subspace F of Ed can be embedded as a projective subspace in Pd,
which is the family of lines passing through Q in the affine subspace aff({F, Q}) of
Ed+1. Any line of Ed can therefore be extended to a projective line by adding the
point at infinity, and similarly a subspace of dimension k of Ed can be extended
to a projective subspace of dimension k of pd by adding a (k - 1)-subspace at
infinity.
L(X
X /
Figure 7.5. Central projection.
Sometimes it can help to represent the projective space pd as the set of antipo-
dal (i.e. diametrically opposite) pairs of points on a sphere Sd in Ed+1 centered
at the origin Q. The point X of pd corresponding to the affine line L(X) in Ed+l
can be represented as the pair of points at the intersection of L(X) and Sd. In
this representation, k-subspaces of pd are represented by great k-spheres of Sd,
which are intersections of Sd with affine (k + 1)-subspaces of Ed+1 that contain
Q. The hyperplane at infinity Hoo corresponds to the great (d - 1)-sphere of Sd
in a hyperplane parallel to Ed. The function induced by this representation maps
a point X in pd not in Ho, to the point X = L(X) n Ed of Ed, and is commonly
referred to as the central projection (see figure 7.5).
Homogeneous coordinates
We must note at this point that the equivalence relation 'R used in the definition
of pd is compatible neither with the affine structure nor with the vector-space
structure of Ed+l. Indeed, if XI, X2, YE Y 2 are points in Ed+l, it may happen
that XN 11Z 2 and YlTZYI2 , yet (XI + Y 1 ) R. (X2 + Y2) does not hold. As
a consequence, the projective space pd is neither an affine space, nor a vector
space.
Nevertheless, any basis of Ed+1 can be used as a coordinate system for Pd:
we represent point X as a (d + 1)-tuple (X1 , . .., Xd+1) of reals, the coordinates
of some point in Ed+i on the line L(X). This (d + 1)-tuple (XI,... ,Xd+l) is
not uniquely defined, yet it is unique up to a non-null multiplicative factor, and
constitutes the homogeneous coordinates of X. Any projective hyperplane H
can be described as the set of projective points whose homogeneous coordinates
satisfy a linear equation
d+1
E hixi = 0
i=l
whose coefficients are unique up to a multiplicative factor.
Let 0 be a point of Ed chosen as the origin. If a basis of Ed+l is formed

by adding to a basis of Ed the vector QO, the point X in Ed with coordinates
(X, ... ,xd) is mapped to the projective point with homogeneous coordinates
(X,...., xd, 1). The hyperplane at infinity has equation Xd+1 = 0.
Below, we use the same notation X for a point of Ed and its coordinate vector
(xl, . . ., Xd) . Likewise, the notation X denotes either a projective point or (any
of) its homogeneous coordinate vectors (X 1, .. ., Xd+ ) .
Let k < d+1, and {Ao, A 1 ,..., Ak} be a set of k+I points in pd. These points
are said to be independent if the smallest projective subspace that contains them
has dimension k. Points {AoA 1 ,...,Ak4 are independent if their coordinate
matrix in some basis (which has dimension (k + 1) x (d + 1)) has rank k + 1.
Projective mappings
In a projective space, the hyperplane at infinity is like any other hyperplane and
plays no particular role. In general, the properties of a projective space pd are
invariant under any linear map X ) XT whose matrix T is non-singular.
Such a mapping is called a projective mapping. It transforms a k-dimensional
projective subspace into another projective subspace of the same dimension. The
hyperplane at infinity may be mapped onto any hyperplane of pd by a suitable
projective mapping.
Polarity, duality
Any hyperplane H in a projective space has a homogeneous equation of the kind
d+1
H=f{X: Zhixi= 01.
Let S be the (d + 1) x (d + 1) matrix

S- Rd O
- 0 -1 J
where Rd stands for the d x d identity matrix. Let H* be the projective point
(hi, .. ., hd, -hd+l). The homogeneous equation of H can be rewritten in matrix
form
H = {X: H*SXt = 0}.
Point H* is the pole of hyperplane H. Conversely, to any projective point P with
homogeneous coordinates (Pi, ... ,Pd+1) there corresponds a polar hyperplane P*
with homogeneous equation
P* = {X: PSX t = 0}.
This double correspondence between points and hyperplanes is called the polar-
ity centered at 0. It is exactly the extension to projective spaces of the polarity
centered at 0 described for Euclidean spaces in subsection 7.1.3. In a Euclidean
space, the polarity centered at 0 maps points other than 0 to hyperplanes that
do not pass through 0. In a projective space, the polarity centered at 0 maps
points to hyperplanes, in a one-to-one fashion without restrictions: the projective
point 0 (corresponding to the center 0) is mapped to the polar hyperplane at
infinity Hoo, and the pole of a hyperplane H that passes through 0 is the point
at infinity in the direction normal to the hyperplane H. In the projective space,
as in its Euclidean counterpart, the polarity centered at 0 is an involution, that
is,
P** = P and H** = H,
and reverses inclusion relationships, that is,
P E H -: H* E P*.
Polarity is therefore a duality.
More generally, for any symmetric non-singular (d + 1) x (d + 1) matrix AB,
we consider the mapping that maps a point P to the hyperplane P* satisfying
P*= {X : PAXt = 0, } and a hyperplane H to the point H* satisfying
H = {X : H*ABX t = O}. This mapping is an involution between points and
hyperplanes, therefore is one-to-one, and reverses inclusion relationships. The set
B of those projective points X that satisfy
XABXt = 0
corresponds to a quadric 1 in Ed, and the duality just defined is called the polarity
with respect to B. Using this terminology, the polarity centered at 0 is the polarity
with respect to the unit sphere Sd-i centered at 0. The signatureof the quadric
B is the set of signs of its eigenvalues. In fact, it can be shown that in a projective
space, two quadrics with the same signature or with opposite signatures can be
derived from one another by a projective mapping. The corresponding polarities
are called equivalent.
Besides the polarity centered at 0, one of the polarities most widely used in
computational geometry is that with respect to the unit paraboloid, Pd-1, with
Cartesian equation in Ed
d-1
Xd i
i=l
and homogeneous equation in pd
/ d-1 - 0 0 w
XA-pXt = O with A~p = | 0 -1/2 .
0 -1/2 0
The paraboloid Pd-i can therefore be derived from the unit sphere Sd-i by a
projective mapping sending the center 0 of Sd_1 to infinity along the xd-axis. The
polarity with respect to Pd-i is therefore projectively equivalent to the polarity
centered at 0. For more details on this polarity, see exercises 7.13 and 7.14.
7.3.2 Oriented projective spaces
Motivation
Projective geometry is a powerful and attractive framework for geometry and

algorithms. For instance, it allows us to ignore many particular cases arising
from the presence of parallel subspaces. Projective geometry also helps in giving
a unified presentation of conics and quadrics, and gives the concept of a projective
mapping which generalizes that of an affine transformation, while also adding
perspective mappings which swap points at infinity and points at a finite distance.
Such a bonus does not come without drawbacks, however. The most serious,
from the viewpoint of computational geometry, concern convexity and half-spaces,
which do not exist in projective geometry. Let us now expand a little on these
two points.
About convexity. In a projective space pd, there is no way to unambiguously
define the segment joining two points. Indeed, a projective space is neither a
vector space nor an affine space, and even the notion of linear combination of
projective points has no meaning. If P and Q are two projective points, we may
still let AP + MQ be the quotient of the set of points
{AP + :QP E L(P), Q E L(Q)}
by the relation JZ introduced in subsection 7.3.1. This set, however, is not an

equivalence class of this relation, but a union of such classes. If A and A are fixed
non-zero real numbers, AP + pQ is the 2-subspace of Ed+l which is the affine
hull of the lines L(P) and L(Q). The same set is obtained if we let A and I vary.
This remains true even if we require that A and , satisfy the convexity condition:
A > 0, A > 0 and A+ p = 1. This condition has no effect on the set generated by
AP+IQ.
The lines in aff (L(P), L(Q)) passing through Q do indeed form a projective
line, which is the smallest projective subspace generated by P and Q. But since
the convexity condition has no effect, there is no means to distinguish a subset
of this line which might be the segment PQ.
In the spherical model, the points P and Q are represented by two pairs of
antipodal points (P, -P) and (Q, -Q). The line joining them is a great circle of
Sd passing through P, -P, Q, -Q. The points P, -P, Q, -Q determine on this
circle four arcs pairwise diametrically opposite, or equivalently two projective
arcs. There is no way to identify one of these two projective arcs as being the
segment that joins P and Q.
Without segments, we certainly cannot define what it means for a set to be
convex, nor what the convex hull of a set of points is.
About half-spaces. Let us consider a hyperplane H in the projective
space pd, with homogeneous equation H*SXt = 0. If X does not belong to
H, the sign of the bilinear homogeneous form H*SXt is arbitrary and without
significance since the homogeneous coordinates are defined up to a multiplicative
factor (of either sign). It is therefore impossible to locate the point P on either
side of H. In fact, a projective hyperplane does not separate the space into two
disconnected half-spaces. In the spherical model, a hyperplane is represented by
a great (d - 1)-sphere of Sd. Each projective point P is represented as a pair
(P, -,P) of two antipodal points on Sd, and each of these points belongs to a
different hemisphere determined by H.
Oriented projective geometry remedies this situation while keeping the advan-
tages of projective geometry.
Definition
For each vector V of a vector space, the set {AV : A E R, A > 0} is an oriented
vector line. An oriented projective space of dimension d consists of oriented lines
of a vector space Vd+1 of dimension d + 1. A subspace of this space consists of
the oriented lines lying in a subspace of Vd+l.
More concretely, the oriented projective space pd that extends the affine space
Ed can be described in terms of the embedding of Ed in the space Ed+1. As before,
we let the origin Q be a point of Ed+1 not in the hyperplane that we consider as
Ed. The oriented projective space pd is the set of all rays cast from Q in Ed+l,
or equivalently the set of equivalence classes of the points in Ed+1 \ {Q} for the
relation 7?o defined by: X ZoX' if there exists A > 0 such that X = AX'. Thus, a
point in the projective space corresponds to two points in the oriented projective
space, which are then called opposite points. In the spherical representation, the
oriented projective space amounts to distinguishing the two points in an antipodal
pair of Sd.
When a basis of Ed+1 is understood, a point in the oriented projective space
has a vector of homogeneous coordinates which is defined up to a positive mul-
tiplicative factor. In the rest of this chapter, we denote by P either a point in
the oriented projective space or its vector (P1,P2, -,Pd+1) of homogeneous co-
. .
ordinates, and by -- P its opposite point. As in a projective space, k + 1 points

{Ao, A 1 , . .. , Ak} in pd are independent if their coordinate vectors are indepen-
dent. In particular, points P and -'P are not independent.
The subspaces of dimension k of pd are the subsets of points in pld that cor-
7.3. Protective polytopes, unbounded polytopes 155
respond to the rays cast from Q inside a subspace of Ed+l of dimension k + 1

that contains Q. In this manner, a subspace that contains a point P also con-
tains its opposite -UP, and the subspaces of IPd coincide with those of Pd. One
of the main advantages of working in oriented projective geometry is the possi-
bility of orienting the subspaces. Let F be a k-subspace of Pd, corresponding
to the k + 1-subspace F of Ed+l. Any set of vectors in Ed+l that forms a basis
for F gives a coordinate system in F.2 In such a coordinate system, a point
of F is represented by a (k + 1)-vector of homogeneous coordinates defined up
to a positive multiplicative factor. For two (k + 1)-tuple of independent points
in F, {Ao,Al,..., Ak} and {Bo, Bl,..., Bk}, we consider a (k + 1) x (k + 1)
matrix U that transforms the homogeneous coordinates of {Ao, A 1, ... , Ak to
those of {Bo, B 1 , . . ., Bk} in the same coordinate system. Two (k + 1)-tuples
{Ao, A 1 , . .. , Ak} and {Bo, B 1 ,. . . , Bk are called equivalent if the determinant
of U is positive. The sign of the determinant of U does not depend on the choice
of the coordinate system or on the choice of the vectors of homogeneous coordi-
nates used to represent each Ai or Bj. The (k + 1)-tuples of points of F fall into
two equivalence classes. To give an orientation to F is to choose one of these two
classes as the positive orientation. There are thus two possible orientations for an
oriented projective space. An oriented projective k-subspace is determined by a
(k + 1)-tuple which at once determines the subspace and its positive orientation.
From now on, projective subspaces are supposed to be oriented, and we denote
by F and -iF an oriented subspace and the same subspace with the opposite
orientation.
In particular, a projective oriented hyperplane H can be defined by a d-tuple
{Ao, A 1 , .-. , Ad-1} of affinely independent points, by the homogeneous equation
H = {X P [Ao A,, Ad-1, X = 0,
where [Ao, Al, . .- , Ad-1, X] is the determinant of the matrix whose coefficients
are the homogeneous coordinates of {Ao, A1 ... , Ad-L, X}. The coefficients
(hi, .. ., hd+
0 ) in the homogeneous equation Ed+' hixi = 0 defining H are thus
defined up to a positive multiplicative factor. It is possible to determine two
classes between the points in Pd \ H, and an oriented projective hyperplane
separates the space into two half-spaces
= {X
H+ H-= { E Pd
E Pd [AOdAl,...,Ad .,X]X] <
[Ao, Al,-..., Ad-1, > 0},
0}.
2
For instance, for any (k + 1)-tuple {Ao, A 1 ,..., Ak} of independent points in F, the vec-
tors generating the oriented vector lines of Ed+1 corresponding to {Ao,A 1 ,...,Ak} form a
coordinate system for F.
Duality
The notion of duality can be extended without problems to the oriented projective
space. The oriented projective point H* defined by
VX e Pd, H*SXt = [Ao, A, ... , Ad-1, X],
is the pole of the oriented projective hyperplane H for the polarity centered at
0. Likewise, H is the polar hyperplane of H*. The pole of the hyperplane -'H
with the opposite orientation is the point opposite to the pole of H. Polarity re-
verses the inclusion relationships between points and hyperplanes and, moreover,
reverses the relative positions of a point and a hyperplane, that is
P EH H*SPt = 0 PSH*t = 0 H* G P*
P e H+ H*SPt <O PSH*t < 0 H* E P*+
P E H- H*SPt >O PSH*t > 0 H* E P*-.
In the spherical model, a point in the oriented projective space pd is represented

by a single point of Sd, and an oriented projective hyperplane is represented by
an oriented great (d - l)-sphere. The two half-spaces determined by an oriented
projective hyperplane H correspond to the two hemispheres bounded on Sd by
this great sphere.
Remark. The points at a finite distance of a hyperplane H in pd, which are
the points in H \ (H n Hoc,), project onto a hyperplane H in Ed. The points
in the half-spaces H+ and H-, however, do not project onto a half-space of Ed.
In fact, the projective half-spaces determined by H each project onto the whole
Euclidean space Ed. Let us denote by H+ and H,;c the two half-spaces in pd
determined by the hyperplane at infinity:
H+ = {X E Pd : Xd+1 > O}
Ho,;=f{XEP: Xd+1 < °}
Now let H be a hyperplane of Pd which projects onto a hyperplane H in Ed. Each

of the subsets H+ n H+ and H- n Ho; projects onto one of the half-spaces of
Ed determined by H, say H+, and each of the subsets H+ n Hl7c and H- n H+
projects onto the other half-space bounded by H, namely H- (see figure 7.6).
7.3.3 Projective polytopes, unbounded polytopes

Projective simplices, convexity
We may also define the notion of simplex in an oriented projective space. Let
{Po, P1 , . .. , Pk4 be a set of k + 1 independent points in IPo. This set of points
+- n H,;
!+ni;
Figure 7.6. Projection onto Ed of an oriented projective half-space.
determines a k-subspace F, and we denote by [Po, Pi, . .. , Pk] the determinant

of the (k + 1) x (k + 1) matrix whose coefficients are the homogeneous coordinates
of {Po, P1 , . .. , Pk} in some coordinate system of F. Without loss of generality,
we may assume that this determinant is positive.
The interior of the simplex POP, ... Pk can now be defined as the set of those
points X of F for which the determinants [Po,... , Pi-1 , X, Pi+l,..., Pk], ob-
tained by substituting X for Pi for every i, 0 < i < k, are all positive. In this
manner, the points in the simplex PoP1 ... Pk are those whose homogeneous co-
ordinates can be derived from those of {Po, PI, ._. , Pk} by a linear combination
with non-negative coefficients.
In particular, the notion of a segment joining two points can be defined for any
pair of oriented projective points which are not opposite. In the spherical model,
the segment PQ is represented by the shortest arc joining P and Q on the great
circle of Sd passing through P and Q.
Theorem 7.3.1 The points on a projective segment can be mapped in a one-to-

one fashion to the points of a segment in Ed.
Proof. Let P and Q be two points in IP0, not opposite to one another. If P and
Q both belong to one of the half-spaces Ho or Hoo, then segment PQ of pd
projects onto segment PQ in Ed. Otherwise, since P is not opposite to Q, there
is a hyperplane H such that P and Q lie on the same side of H. The projective
mapping sending H to H, transforms segment PQ into a segment P'Q' which
projects onto segment P'Q' in Ed. 0
We may now redefine the notion of convexity in an oriented projective space:

* a subset X of Ipd is called quasi-convex if for each pair (P, Q) of non-

opposite points in X, the segment joining P and Q is entirely contained in
X;
* a subset X of Pod is called convex if it is quasi-convex, and does not contain

a pair of opposite points.
Defined this way, convex sets include segments, simplices, and open half-spaces;
antipodal pairs of points, closed half-spaces, projective subspaces, and the entire
oriented projective space are quasi-convex.
The notions of quasi-convexity and convexity are invariant under projective
mapping. The intersection of quasi-convex sets is quasi-convex, and the intersec-
tion of convex sets is convex.
Projective polytopes
The quasi-convex hull of a set of points in Pd is the smallest quasi-convex set
that contains that set. The quasi-convex hull is not always convex, but it is
convex when the set of points is contained within an open half-space of Pd. In
this case, we may speak of the convex hull of the set of points. Quasi-convex and
convex hulls consist of all linear combinations of the points with non-negative
coefficients. A projective polytope is the convex hull of a finite set of points which
is entirely contained in an open half-space.
Theorem 7.3.2 The points of a projective polytope can be put in one-to-one

correspondence with the points of a polytope in Ed.
Proof. Let P = conv({Po, PI,., Pn}) be a projective polytope, and H a

hyperplane of pd which bounds a half-space containing {Po, P 1 , ... ,Pn}. Any
projective mapping which transforms H into H,, transforms P into a projective
polytope 'P' which lies entirely on one side of Hoo, either H+ or Hos. This
polytope projects onto a polytope P' of Ed. If Pi' is the image of Pi under the
projective mapping, then
7'- conv({P ,P', . ,Pn}), I' = conv({PPPO, .P}).
The notions of supporting hyperplanes and faces can be carried over to the pro-
jective setting without problems. The above correspondence therefore establishes
a one-to-one correspondence between the faces of the polytopes P and P", which
also allows us to transfer to projective polytopes the combinatorial properties of
Euclidean polytopes. All the theorems in sections 7.1 and 7.2 can therefore be
stated for projective polytopes. Here, we only give the projective statements of
theorems 7.1.4 and 7.1.5, which concern the polar transformations.
A set 1t of projective hyperplanes in Pd is in generalposition if, for any j < d,
the intersection of any j of them is a projective subspace of dimension d -j,
and if moreover the intersection of any d + 1 of them is empty. The intersection
of m closed projective half-spaces nl 1 Ht is contained in an open projective
half-space if and only if there is a subset of d + 1 hyperplanes in general position
among the hyperplanes Hj bounding all the half-spaces: such an intersection is
called non-trivial. Theorems 7.1.4 and 7.1.5 can now be restated in a projective
setting:
Theorem 7.3.3 A projective polytope is the non-trivial intersection of a finite

number of closed projective half-spaces. Any non-trivial intersection of a finite
number of closed projective half-spaces is a projective polytope.
In the oriented projective space, the polarity centered at 0 (or any other po-
larity, for that matter) can be used to define an involutive one-to-one mapping
on the set of all projective polytopes, without anyjrestrictions. The polar image
A# of a polar point A is the closed half-space A*+ defined by
A# ={Y E Pd : ASYt < O} = A*+,
and the polar image A# of an oriented projective subset A is defined as the

intersection of all the images of points in A:
A# ={y Pd: ASYt < O, VA E A}.
Let now
'P = conv(Pi, . . ., Pn)
be a projective polytope. For each i = 1,...,n, we denote by Pi* the polar
hyperplane of Pi and by Pi+ = Pi# the polar half-space of Pi*. The polar
image of the polytope P is the intersection
n -
'P#nPi*+.
i=1
If P is a projective polytope of dimension d, the set {P 1 ,..., Pn} contains at

least d + 1 independent points and the intersection 7P# is non-trivial. Thus
P# is a projective polytope. The proofs of theorem 7.1.11, lemma 7.1.12, and
theorem 7.1.13 can now be stated almost verbatim for projective spaces and lead
to the following theorem:
(a) (b)
Figure 7.7. Projection onto Ed of a projective polytope.
Theorem 7.3.4 The polar transformationinduces an involutive one-to-one map-

ping defined on the set of all projective polytopes of dimension d in the oriented
projective space Pgd. There exists a bijection between the k-faces of polytope P
and the (d - k - 1)-faces of polytope p#, which reverses inclusion relationships.
Moreover, a projective point A is respectively inside, outside, or on the bound-
ary of a polytope P if its polar hyperplane A* respectively avoids, intersects, or
supports the polar polytope ?P.
In this manner, in oriented projective geometry, the polarity centered at 0 or,

more generally, any polarity, unambiguously defines a perfect duality between
convex hulls of finite sets of points and finite intersections of half-spaces.
It should not be forgotten, however, that this duality between convex hulls
and intersections of half-spaces is a duality between projective polytopes. The
corresponding faces P of p# are also projective polytopes, but they do not
always project onto polytopes in Ed. A projective polytope projects onto Ed as a
polytope only if it lies entirely within Ho; or H+, or equivalently if it does not
intersect the hyperplane at infinity Hoo (see figure 7.7).
All the other theorems in sections 7.1 and 7.2 can be restated verbatim for
projective polytopes. In particular, a projective polytope satisfies Euler's relation,
and the Dehn-Sommerville relations if it is simple, hence also the upper bound
theorem.
Unbounded polytopes
Let us comment again on the projection onto Ed of a projective polytope. Let P

be a polytope in Pdl, given as the non-trivial intersection
m
-P= nHt.
j= T
The projective hyperplane Hj is the projective extension of the hyperplane Hj

in Ed. The intersection Hf n H+ therefore projects as a half-space HII in Ed
3 - 3
(see figure 7.6). As a consequence, if polytope P is contained in H+, then it
projects onto the polytope
m
P= nHi+
j=1
in Ed. On the other hand, if 'P intersects Hoo, it projects onto the union of two
linear convex unbounded subsets
(n 7 U (nH-
In this case, the projective polytope
m
Q= ( H)fn Hi
j=1
projects onto the unbounded intersection of m half-spaces in Ed
Conversely, let
m
Q=fnHi
j=1
be an unbounded intersection of m half-spaces in Ed. Such an intersection is

called non-trivial if there is, among the hyperplanes bounding the half-spaces, a
subset of d hyperplanes in general position.
Theorem 7.3.5 Any unbounded and non-trivial intersection Q of closed half-

spaces in Ed is the projection onto Ed of a projective polytope.
Proof. Let 0 be a point lying inside Q, and 0 the projective point in Hi

whose projection onto Ed is 0. For every j =1,. . . ,m, let Hj be the projec-
tive extension of Hj and Ht the closed half-space of Pd bounded by Hj which
contains 0. Then Q (ni 1 Ha) fl H+ is a non-trivial intersection of closed
projective half-spaces, whose projection onto Ed is Q. E
This allows us to call any non-trivial unbounded intersection of a finite subset

of closed half-spaces, an unbounded polytope.
The notions of a supporting hyperplane and faces can therefore be extended
to unbounded polytopes. The facial structure and combinatorial properties of
unbounded polytopes can be derived from those of projective polytopes by central
projection.
7.4 Exercises
Exercise 7.1 (Radon's theorem) Show that any set X of at least d + 2 points of Ed
can be split into two subsets Xl and X2, such that conv(Xl) n conv(X 2 ) #' 0.
Hint: There exists a non-trivial linear relationship between the points in X
EAiXi = 0.
L=1
For Xl (resp. X2), choose the points Xi whose coefficients Ai in the above relation are
positive (resp. negative or zero).
Exercise 7.2 (Helly's theorem) Let {QC 1 , ./.. , Ir} be a family of r convex sets of Ed.
Show that if any d + 1 convex sets have a non-empty intersection, then so does the whole
family.
Hint: One possible proof goes by induction on r. By induction, we know that there is a
point Xi in the intersection nfl#i )Ci, for all i = 1,. . . , r. Then use Radon's theorem on
the set {Xi : i = 1, .. . , r} to construct a point X that belongs to all the sets ki.
Exercise 7.3 (Caratheodory's theorem) Show that the convex hull of a subset X of
Ed can be described as the set of all possible convex linear combinations of d + 1 points
of X. Use this to show that every polytope is a finite union of simplices.
Hint: Let conv(X) be the convex hull of a subset X of Ed and X a point in conv(X)
given by a minimal convex linear combination X = `=1 AiXi. If r > d + 1, the points
Xi are not independent. Use this to show that we may remove one of the points from
the combination, and that it is therefore not minimal.
7-4. Exercises 163
Exercise 7.4 (Supporting hyperplanes) Let IC be a closed bounded convex set in

Ed. A hyperplane H supports IC if and only if H n K is not empty but K is entirely
contained in one of the two half-spaces bounded by H.
1. Show that for any point X not in IC, there is a supporting hyperplane of IC which
separates K: and X, that is such that X E H- and A C H+.
2. Show that any closed bounded convex subset of Ed is the intersection of all the
half-spaces that contain it and that are bounded by its supporting hyperplanes.
Hint: First show that for any point X V AC, there is a unique point D(X) of K such that
d(X, 4(X)) = min{d(X, Y): Y E AC};
here, d(X, Y) denotes the Euclidean distance between X and Y.

For each X g AC, let D+ (,(X), X) be the infinite ray originating at D(X) towards X.
The hyperplane H passing through 4(X) and normal to D+(b(X), X) is a supporting
hyperplane of K and separates X from K.
Exercise 7.5 (Supporting hyperplanes) Let IC be a closed bounded convex subset

of Ed. Show that every point on the boundary of )C belongs to a supporting hyperplane
of A.
Hint: Consider the mapping 4Dfrom Ed to ACdefined by
VX EC, 4)(X) =X,
VX K, D(X) X',
where X' is the unique point of ACsuch that
d(X, X') = min{d(X, Y) : Y e K}.
1. Show that for every X and Y in Ed \ AC,
d(4)(X), D(Y)) < d(X, Y).
From this, deduce that 4) is continuous.

2. Let Sd-1 be a (d - 1)-sphere of Ed, the boundary of a ball in Ed that contains ).
Show that the image under 4Dof Sd-l is the whole boundary of AC.
3. Show that every point on the boundary of ACis the image under 4) of at least one
point of Sd-1- Use the previous exercise to show that there is a supporting hyperplane
of ACthrough this point.
Exercise 7.6 (Simplices) Show that simplices are the only polytopes which are both
simple and simplicial.
Exercise 7.7 (The Dehn-Sommerville relations) Prove the Dehn-Sommerville re-

lations for a d-polytope form a system of linear relations whose rank is exactly [d+lj
Hint: Consider a d-polytope P. The face-vector of P is the d-dimensional vector whose

components are {no(P),.. ., nd-l(P)I. Show that the face-vectors of cyclic d-polytopes
with d + 1, d +....
2 , d + [4-1J vertices are independent.
Exercise 7.8 (The upper bound theorem) Let P be a simplicial d-polytope and let
nj = nj (P) denote the number of j-faces of P. Show that the Dehn-Sommerville relations
on the numbers nk can be solved for the numbers nj, j = 0,..., [4-d1, yielding those
numbers as linear combinations of the numbers nj with j = |d] .... d.
Hint: Given integers r > 1, d > 2r -2, let D(r, d) be the r x r determinant
r d 1 d1> Id-r+l0
t r -1)Jtr1J
d
r --2) r-2
- (
-
r )
( d) ( do-1 ) ... ( d-0+1 )

. . .
Show by induction on r that D(r,d) = 1. This is trivial if r = 1. If r > 1, one may

subtract column i + 1 from column i for i = 1, . . . , r - 1 in that order. This yields
D(rd) D(r-l,d-1) C(r-ld-1) -D(r- 1, d-1)

0 1
for some (r - 1)-vector C(r -1, d - 1). This proves the inductive step.
Now let d be as in the exercise, and put 2r = d if d is even, d = 2r -1 if d is odd. For
each k = 1,. .. , r, there is a Dehn-Sommerville relation
d ) no- d- ni+.+ lr-1 d- -r +In-I L=nd-r+k,
r-kn rr- kn+ +(1 r-k J)-+k~drk

where Lk is some integral linear combination of nr, . . ., nd-r+k. Regard these relations as
a system of r linear equations for no, . . , n,- 1 . The matrix of coefficients of this system
is integral with determinant ±D(r,d) = +1,and therefore has an integral inverse. Hence
each of no,..., nr-1 can be expressed as an integral linear combination of nr,. .. , nd.
Exercise 7.9 (Euler's relation) Show that Euler's relation is the only non-trivial lin-
ear relation satisfied by the numbers nk(P) (O < k < d - 1) of faces of any d-polytope.
Hint: By induction on the dimension d, we show that any linear relation

d-1
E Aj nj (P) = Ad
j=O
satisfied by all d-polytopes P is proportional to Euler's relation. For this, it suffices

to consider any (d - 1)-polytope Q and, from it, to build two d-polytopes: a pyramid
P = conv(Q, P) where P is a point of Ed that does not belong to the affine hull aff(Q)
of Q, and a bipyramid 1Z = conv(Q, {P, P'}) where P and P' are two points on opposite
sides of aff(Q) such that the segment PP' intersects Q.
7.4. Exercises 165
Exercise 7.10 (Canonical triangulation) A triangulation of a polytope P is a set of

simplices whose relative interiors partition the faces of P. The canonical triangulation
Tf(P) of a polytope P is defined by the following bottom-up construction process:
* The 0-simplices of T7(P) are the vertices of P.
* For 0 < k < d, the k-simplices of 'T(P) are the simplices conv(VF, S) where VF is
the lexicographic smallest vertex of a k-face F of P and S is a (k -1)-simplex of
T,(P) included in the boundary of F and not including VF.
The aim of this exercise is to show that for 0 < k < d the k-simplices of Tec(P) are in one-
to-one correspondence with the k-faces of a simplicial polytope with the same number of
vertices as P. Therefore the number of simplices in the canonical triangulation T,(P) of
any d-polytope P with n vertices is bounded above by O(nr2J).
For this we describe a transformation on polytopes called pulling a vertex. Let P be a
d-polytope and V a vertex of P. Let V' be a point in Ed such that (i) V' belongs to the
intersection nVEF HF involving, for all the facets F of P that contain V, the half-space
HF bounded by the hyperplane supporting P along F and disjoint from P and (ii) the
segment VV' does not meet any of the hyperplanes supporting P along a facet except
in vertex V. The polytope P' = conv(V' U P) is said to be obtained from P by pulling
vertex V to V'.
1. Show that the k-faces of P' are (i) the k-faces of P which do not contain V and (ii)
faces of the form conv(V' U G) where G is a (k - 1)-subface not containing V of a k-face
F containing V.
2. Show that the number of k-faces of P, is less than the number of k-faces of P'.
3. Consider the polytope P, obtained from P by pulling successively, in lexicographic
order, the vertices of P. Show that P, is a simplicial polytope with the same number of
vertices as P and that, for all 0 < k < d, the number nk(P) of k-faces of P is less than
the number nk(Pc) of k-faces of Pc. Conclude.
Define the operation pushing a face of a polytope P as the dual of the operation pulling
a vertex and show that repeated applications of this operation allow us to build a simple
polygon P, with n facets from any polytope P with n facets, such that nk(P) < nk(P'c)
for each k, 0 < k < d.
Exercise 7.11 (Maximal polytope) Show that there exists a polytope with n vertices
on a sphere, or on a paraboloid, with maximal complexity Q(nLd/2] ).
Hint: In the Euclidean space Ed, when d is even (d = 2p), we consider the curve M'd on
the unit sphere, parameterized by
1
M' = {M(r) = - (sin(T), cos(r), sin(2r), cos(2r), .. ., sin(pr), cos(pr)), r E [0, 7r/2]}.
Using the identity:

1 cos(ro) sin(ro) ... cos(pro) sin(p,-o)
1 cos(rl) sin(Tl) ... cos(pTi) sin(p7l) = 2 jj sin (1 1-i))
O<i<j<d
1 cos(7 2 p) sin(r2p) ... cos(pr 2 p) sin(pr 2 p)
show that the convex hull of n points on this curve, parameterized respectively by
{r 1 , r2, ... , n}, is a polytope whose faces are in one-to-one correspondence with those of
a cyclic polytope (introduced in subsection 7.2.4). Conclude that it is possible to build
a maximal polytope whose vertices lie on the unit sphere of Ed, with exactly ( k)
(k - 1)-faces for any k, 1 < k < d/2.
By considering the projective mapping that sends the unit sphere of Ed onto the unit
paraboloid in Ed, with equation
d-1
Xd ZExi,
i=l
show that it is possible to build a maximal polytope whose vertices are on the unit
paraboloid of Ed.
Exercise 7.12 (Upper bound theorem) This exercise presents a very simple proof
of the upper bound theorem. This proof considers a polytope, given as the intersection
of n half-spaces in Ed bounded by hyperplanes in general position, and shows that the
number of vertices of this polytope is O(nJ14i).
1. Show that any vertex of the polytope is the vertex that has the minimal or maximal
Xd-coordinate in a k-face, for some k > [d] For this, we consider a vertex P of the
polytope. This vertex is incident to d edges, at least [d] of which are contained in the
half-space Xd > Xd(P) or the half-space Xd < Xd(P).
2. Note that a face has a unique vertex with maximal Xd-coordinate, and a unique
vertex with minimal Xd-coordinate. Recall the bound of (d-k) on the number of
faces of dimension k of a polytope given by the intersection of n half-spaces in Ed and
conclude.
Exercise 7.13 (Polarity with respect to a paraboloid) Consider the polarity with
respect to the unit paraboloid P with homogeneous equation
Ed-1 0 °
XApXt = 0 with ( 01 0 1/2
\ -1/2 0
where Ed-1 is the (d -1) x (d - 1) identity matrix. Show that the restriction of this
polarity to the Euclidean space maps a point P in Ed with coordinates (P1,P2, Pd)
to the hyperplane P* in Ed with equation
d-1
Xd = 2 PiXi -Pd,
and a non-vertical hyperplane H with equation Xd = 2 rd-1 hixi - hd to the point

H* = (hi, h2, . ,hd).
Show that this transformation is a one-to-one mapping between the points of Ed and
the non-vertical hyperplanes in Ed.
7.4. Exercises 167
Show that this bijection reverses inclusion relationships, that is,
PE H H*E P*.
For a non-vertical hyperplane H, with equation Ed 1 hii + hd+1 = 0, we denote

by H+ (resp. H-) the half-space above (resp. below) H, given by the inequality Xd >
£'1d hii- hd (resp. is < Ed 1' hixi - hd). Show that the polarity with respect to
paraboloid P reverses the relative vertical positions of a point and a hyperplane, that is,
P E H+ H* e P*+
P c H- H* E P*-.
Exercise 7.14 (Lower convex hull) Let {P1 , P2 ,... ., P } be a set of points in Ed and
O' be a point on the Xd-axis, with Xd > 0 large enough such that the facial structure
of conv(O', P1 ,P 2, ... ,PP) is stable as O' goes to infinity along the Xd-axis. We call
lower convex hull of {P1 , P2 ,. .. , Pn}, and we denote by convs(Pl, P2 ,... aPn), the set of
faces of conv(O', P1 , P2 , .. ., Pn) which do not contain 0'. Using the oriented projective
space and the polarity with respect to the unit paraboloid P studied in exercise 7.13, show
that there is a one-to-one correspondence between the faces of conv (P1 , P2 , ... P Pn) and
those of the unbounded intersection fn= Pi*+, where the half-spaces Pi*+ are defined as
in exercise 7.13.
Exercise 7.15 (Euler's relation) Show that Euler's relation for an unbounded poly-
tope of Ed can be expressed as
d
EZi)knk(P) = 0,
k=O
where nk(P) is the number of k-faces of the unbounded polytope P.
Exercise 7.16 (Half-space intersection) Let Q = n'l Hj be the intersection of m

half-spaces in Ed. Suppose that Q is not empty and that a point 0 inside Q is known.
1. Show that Q is bounded if and only if 0 is in the interior of the convex hull
conv({O, Ha,. . . ,H )
..
2. Show that if 0 is a vertex of conv({0, H*,..H ,m}),the faces of Q are in one-to-one
correspondence with the faces of conv({0, Hi,... , H }) that do not contain 0.
Exercise 7.17 (Zonotopes) The Minkowski sum A E B of two polytopes A and B is

the polytope defined by
A(DB={A+B: AeA,BCB}.
Clearly, this operator is associative. A zonotope is a polytope that can be expressed as

the Minkowski sum of a finite set of line segments. Let {S1 , ..., SnI be a set of n line
segments in Ed, and Z the zonotope S, E ... e Sn. A translation brings the midpoint
of each segment to the origin while simply translating the zonotope. The endpoints of
segment Si are denoted by Ai and -Ai.
1. Show that the polar transform Z# of Z is the polytope given by

n
Z# = {X: EIX Ail < 1}.
iil
2. Show that the faces of Z are of the form
F = Si, do .. Si, + ei, +lAir+l + + ei^ Ai,, (7.9)
where Eij =ij = r + I, .. , n.
There is a close relationship between zonotopes of Ed and arrangements of hyperplanes

in Ed+1 (see also exercise 14.8).

The reader interested further in polytopes will do well to turn to the books by McMullen
and Shephard (159], Griinbaum[114], Brbnsted [37], or Berger [24]. Each of these works
covers and goes well beyond all the material in sections 7.1 and 7.2. The reader will find
in particular solutions to exercises 7.1 to 7.7. The presentation of polytopes given in
sections 7.1 and 7.2 follows the same order as that by McMullen and Shephard.
The upper bound theorem 7.2.5 which we give here is an asymptotic version of the true
upper bound theorem, which gives an exact and optimal bound on the number of k-faces
of a polytope with n facets or vertices. Exercise 7.12 presents another proof of the upper
bound theorem that does not use Euler's or the Dehn-Sommerville relations. This proof
is due to Seidel. The oriented projective space was introduced by Stolfi in [210, 211].
Chapter 8
Incremental convex hulls
To compute the convex hull of a finite set of points is a classical problem in com-
putational geometry. In two dimensions, there are several algorithms that solve
this problem in an optimal way. In three dimensions, the problem is consider-
ably more difficult. As for the general case of any dimension, it was not until
1991 that a deterministic optimal algorithm was designed. In dimensions higher
than 3, the method most commonly used is the incremental method. The algo-
rithms described in this chapter are also incremental and work in any dimension.
Methods specific to two or three dimensions will be given in the next chapter.
Before presenting the algorithms, section 8.1 details the representation of poly-
topes as data structures. Section 8.2 shows a lower bound of Q(n log n + nLd/ 2 i )
for computing the convex hull of n points in d dimensions. The basic operation
used by an incremental algorithm is: given a polytope C and a point P, derive the
representation of the polytope conv(C U {P}} assuming the representation of C
has already been computed. Section 8.3 studies the geometric part of this prob-
lem. Section 8.4 shows a deterministic algorithm to compute the convex hull of n
points in d dimensions. This algorithm requires preliminary knowledge of all the
points: it is an off-line algorithm. Its complexity is O(n log n + nL(d+l)/ 2 i), which
is optimal only in even dimensions. In section 8.5, the influence graph method
explained in section 5.3 is used to obtain a semi-dynamic algorithm which al-
lows the points to be inserted on-line. The randomized analysis of this algorithm
shows that its average complexity is optimal in all dimensions. Finally, section 8.6
shows how to adapt the augmented influence graph method of chapter 6 to yield
a fully dynamic algorithm for the convex hull problem, allowing points to be in-
serted or deleted on-line. The expected complexity of an insertion or deletion is
O(logn + nL[d/2 -l), which is optimal.
Throughout this chapter, we assume that the set of points whose convex hull is
to be computed is in general position. This means that any subset of k + 1 < d + 1
points generates an affine subspace of dimension k. This hypothesis is not crucial
170 Chapter 8. Incremental convex hulls
Figure 8.1. A tetrahedron and its incidence graph.
for the deterministic algorithm (see exercise 8.4), but it allows us to simplify the
description of the algorithm and to focus on the central ideas. It becomes an
essential assumption, however, for the randomized analyses of the on-line and
dynamic algorithms.
8.1 Representation of polytopes

To compute the convex hull of a set of points amounts to setting up a data
structure that represents the polytope which is the convex hull of the set. A
polytope is generally represented by the incidence graph of its faces, which stores
a node for each face and an arc for each pair of incident faces. Recall that two
faces are incident if their dimensions differ by one and if one is contained in the
other. Figure 8.1 shows the incidence graph of a tetrahedron.
Using the upper bound theorem 7.2.5, the incidence graph of a d-polytope can
be stored using 0(nLd/2 j) space. This graph describes the entire combinatorial
structure of the polytope. In order to describe its geometric structure, some
additional information has to be stored: for instance, the node storing a vertex
contains the coordinates of that vertex, and the node storing a facet contains the
coefficients in an equation of the hyperplane that supports the polytope along
that facet.
Sometimes, it may be enough to store subgraphs of the incidence graph. The
j-skeleton of a polytope is the subgraph of the incidence faces of dimension at
most j. The 1-skeleton of a polytope is simply made up of the vertices and edges
of that polytope.
In a d-polytope, every (d - 2)-face is incident to exactly two (d - 1)-faces
8.2. Lower bounds 171
(theorem 7.1.7); two (d - 1)-faces of a polytope are said to be adjacent if they

are incident to a common (d - 2)-face. Thus, the incidence graph of a polytope
also encodes the adjacency graph, which has a node for each facet and an arc
for each pair of adjacent facets. The arcs of the adjacency graph are in one-to-
one correspondence with the (d - 2)-faces of the polytope. If the polytope is
simplicial, the full incidence graph can be retrieved from the adjacency graph in
time linear in the number of faces (see exercise 8.2).
8.2 Lower bounds

Theorem 8.2.1 The complexity of computing the convex hull of n points in d
dimensions is Q(n log n + nLd/2 i2).
Proof. Subsection 7.2.4 shows that the convex hull of n points in the Euclidean
space Ed may have Q(nLd/2 J) faces. In any dimension, Q(nLd/ 2j) is thus a trivial
lower bound for the complexity of computing convex hulls. In two dimensions, the
lower bound Q (n log n) is a consequence of theorem 8.2.2 proved below. Finally,
any set of points in E2 can be embedded into E3 , so the complexity of computing
convex hulls in E3 cannot be smaller than in E2 . [:
Theorem 8.2.2 The problem of sorting n real numbers can be transformed in

linear time into the problem of computing the convex hull of n points in E2 .
Proof. Consider n real numbers x1 , x2, .. ., X,, which we want to sort. One way
to do this is to map the number xi to the point Ai with coordinates (xi, x?) on
the parabola with equation y = x2 (see figure 8.2). The convex hull of the set
of points {Ai : i = 1,...,n} is a cyclic 2-polytope, and the list of its vertices
is exactly the list of the vertices {Ai : i = 1,...,n} ordered according to their
increasing abscissae. El
8.3 Geometric preliminaries

The incremental method for computing the convex hull of a set A of n points in
Ed consists in maintaining the succession of convex hulls of the consecutive sets
obtained by adding the points in A one by one. Each convex hull is represented
by its incidence graph. Let C be the convex hull of the current subset and P the
point to be inserted next into the structure, at a given step in the algorithm. The
problem is thus to obtain the incidence graph of conv(C U {P}), once we know
that of C. The following lemmas clarify the relations existing between these two
graphs.
172~ Chpe .Iceetlcne ul
45
XI X2 z3 X4 X5
Figure 8.2. Transforming a sorting problem into a convex hull problem in two dimensions.
Figure 8.3. The incremental construction of a convex hull.
Suppose that point P and polytope C are in general position, meaning that P
and the vertices of C form a set of points in general position. The facets of C
can then be separated into two classes with respect to P. Let F be a facet of C,
HF the hyperplane that supports C along F, and HF+ (resp. HF) the half-space
bounded by HF that contains (resp. does not contain) C. The facet F is red with
respect to P if it is visible from point P, that is if P belongs to the half-space
Hi. It is colored blue if P belongs to HF+. From the general position assumption,
it follows that P never belongs to the supporting hyperplane HF and therefore
every facet of C is either red or blue with respect to P.
Using theorem 7.1.7, any face of C is the intersection of the facets of C which
contain it. The faces of C of dimension strictly smaller than d- 1 can be separated
into three categories with respect to P: a face of C is red if it is the intersection
8.3. Geometric preliminaries 173
of red facets only, blue if it is the intersection of blue facets only, or purple if it is
the intersection of red and blue facets.
Intuitively, the red faces are those that would be lit if a point source of light
was shining from P, the blue faces are those that would remain in the shadow,
and the purple faces would be lit by rays tangent to C. In figure 8.3, the blue
faces of C are shaded, the red edges are outlined in dashed lines, and the purple
edges are shown in bold.
Lemma 8.3.1 Let C be a polytope and P a point in general position with respect
to C. Every face of conv(C U {P}) is either a blue or purple face of C, or the
convex hull conv(G U {P}) of P and a purple face G of C.
Proof. Note that if P belongs to C, all the facets of C are blue with respect to
C (theorem 7.1.4) and the content of the lemma is trivial.
In the other case, we first show that a blue face of C is a face of
conv(C U {P}). Let F be a facet of C that is blue with respect to P. Since
P belongs to the half-space H+, the hyperplane HF which supports C along F
also supports conv(C U {P}) and conv(C U {P}) n HF = F, which proves that
F is indeed a facet of conv(C U {P}). Any blue facet of C is thus a facet of
conv(C U {P}). Any blue face of C, being the intersection of blue facets of C, is
also the intersection of facets of conv(C U {P}): therefore a blue face of C is also
a face of conv(C U {P}) (theorem 7.1.7).
Next we show that, for any purple face G of C, G and conv(G U {P}) are
faces of conv(C U {P}). If G is a purple face of C, then there is at least one
red facet of C, say F1 , and one blue facet of C, say F2 , that both contain G
(see figure 8.4). Let H1 (resp. H2 ) be the hyperplane supporting C along F1
(resp. F2 ). Point P belongs to the half-space H+ which contains C, and since
Hlnconv(CU{P}) = G we have shown that G is a face of conv(CU{P}). Point P
also belongs to the half-space Hj- that does not contain C. Imagine a hyperplane
that rotates around H1 n H2 while supporting C along G. There is a position
H for which this hyperplane passes through point P. Hyperplane H supports
conv(C U {P}), and since conv(C U {P}) n H = conv(G U {P}), we have proved
that conv(G U {P}) is a face of conv(C U {P}).
Finally, let us show that every face of conv(C U {P}) is either a blue or a purple
face of C, or the convex hull conv(G U {P}) of P and of a purple face G of C.
Indeed, a hyperplane that supports conv(CU{P}) is also a supporting hyperplane
of C, unless it intersects conv(C U {P}) only at point P. As a consequence, any
face of conv(C U {P}) that does not contain P is a (blue or purple) face of C, and
any face conv(C U {P}) that contains P is of the form conv(G U {P}) where G
is a purple face of C. Note that the vertex P of conv(C U {P}) is also a face of
the form conv(G U {P}) obtained when G is the empty face of C. Indeed, when
H2
/
Figure 8.4. Faces of conv(G U {P}).
P does not belong to C, C necessarily has some facets that are blue and some
facets that are red with respect to P. The empty face, being the intersection of
all faces of C, is therefore purple. O
The following lemma, whose proof is straightforward, investigates the incidence
relationships between the faces of C and those of conv(C U {P}).
to C.
* If F and G are two incident faces of polytope C, either blue or purple with
respect to P, then F and G are incident faces of conv(C U {P}).
* If G is a purple face of C, then G and conv(G U {P}) are incident faces of
conv(C U {P}).
* Finally, if F and G are incident purple faces of .F, then conv(F U {P}) and
conv(G U {P}) are incident faces of conv(C U {P}).
Recall that two facets of a polytope C are adjacent if they are incident to the
same (d - 2)-face and that the adjacency graph of a polytope stores a node for
each facet and an arc for each pair of adjacent facets.1 We say that a subset
of facets of a polytope C is connected if it induces a connected subgraph of the
adjacency graph of C.
Lemma 8.3.3 Consider a polytope C and a point P in general position. The set
of facets of C that are red with respect to P is connected, and the set of facets of
C that are blue with respect to P is also connected.
'Two facets sharing a common k-face, k < d - 2, may be not adjacent, even though they
are connected as a topological subset of the boundary of the polytope. Such a situation is only
possible in dimension d > 3.
Figure 8.5. Isomorphism between the purple faces and the faces of a (d -l)-polytope.
Proof. If P belongs to C, the set of the red facets is empty, any facet is blue,
and the lemma is trivial. We will therefore assume that P does not belong to C.
The connectedness of the set of red facets can be proved easily in two dimen-
sions. Indeed, the polytope conv(C U {P}) has two edges incident to P. By
lemma 8.3.1, there are exactly two purple vertices of C with respect to P. Hence,
the adjacency graph of the 2-polytope C is a cycle that has exactly two arcs
connecting a blue and a red facet.
Let us now discuss the case of dimension d, and suppose for a contradiction that
the set of facets of C that are red with respect to P is not connected. Therefore,
we may choose two points Q and R on two facets of C that belong to two distinct
connected components of the set of red facets of C. Let H be the affine 2-space
passing through points P, Q, and R. This plane intersects polytope C along a
2-polytope It n H. The edges of C n H that are red with respect to P are exactly
the intersections of the red facets of C with H. The points Q and R belong to two
separate connected components of the set of red edges of C n H. Connectedness
of the set of red faces of a 2-polytope would then not hold, a contradiction.
Analogous arguments prove the connectedness of the set of facets of
conv(C U {P}) that are blue with respect to P. E
Finally, the lemma below completely characterizes the subgraph of the inci-
dence graph induced on the faces of C that are purple with respect to P.
to C. If C has n vertices and does not contain P, then the set of the properfaces of
C that are purple with respect to P is isomorphic, for the incidence relationship,
to the set of faces of a (d - 1)-polytope whose number of vertices is at most n.
Proof. From lemma 8.3.1, we know that the faces of polytope C that are purple
with respect to P are in one-to-one correspondence with the faces of conv(CU{P})
that do not contain P. Since point P does not belong to C, there must be
a hyperplane H which separates P from C (see exercise 7.4). Hyperplane H
intersects all the faces of conv(C U {P}) that contain P except for the vertex P,
and those faces only. Moreover, the traces in H of the faces of conv(C U{P}) are
the proper faces of the (d - 1)-polytope conv(C U{P}) n H, and the traces in H
of incident faces of conv(C U {P}) are incident faces of conv(C U{P}) n H. Thus,
the incidence graph of the (d - 1)-polytope conv(C U {P}) n H is isomorphic to
the subgraph of the incidence graph of conv(C U {P}) induced by the faces that
contain vertex P. Lemmas 8.3.1 and 8.3.2 show that this subgraph is isomorphic
to the subgraph of the incidence graph of P induced by the faces of C that are
purple with respect to P. Lastly, the vertices of polytope conv(C U{P}) n H are
the traces in H of the edges of conv(C U {P}) incident to vertex P, and their
number is at most n. L
8.4 A deterministic algorithm

In this section we describe an incremental deterministic algorithm to build the
convex hull of a set A of n points. The points in A are processed in increas-
ing lexicographic order of their coordinates. To simplify the description of the
algorithm, we assume below that the set is in general position. We denote by
{Al, A2 , .. ., An} the points of A indexed by lexicographic order. Let Ai be the
set of the first i points of A.
The general idea of the algorithm is as follows:
1. Sort the points of A in increasing lexicographic order of their coordinates.
2. Initialize the convex hull to the simplex conv(Ad+l), the convex hull of the
first d + 1 points of A.
3. In the incremental step: the convex hull of conv(Ai) is built knowing the
convex hull conv(Ai-1) and the point Ai to be inserted.
Details of the incremental step
Because of the lexicographic order on the points of A, point Ai never belongs to

the convex hull conv(Ai- 1 ), and is therefore a vertex of conv(Ai). The preceding
lemmas show that the subgraph of the incidence graph of conv(Ai-1) restricted
to the faces that are blue with respect to Ai is also a subgraph of the incidence
graph of conv(Ai). All the efficiency of the incremental algorithm stems from
the fact that the incidence graph of the current convex hull can be updated in
an incremental step without looking at the blue faces or at their incidences.
To perform this incremental step, we proceed in four phases:
8.4. A deterministic algorithm 177
Phase 1. We first identify a facet of conv(Ai-1) that is red with respect to Ai.
Phase 2. The red facets and the red or purple (d - 2)-faces of conv(Ai-1) are
traversed. A separate list is set up for the red facets, the red (d - 2)-faces,
and the purple (d - 2)-faces.
Phase 3. Using the information gathered in phase 2, we identify all the other
red or purple faces of conv(Ai-1). For each dimension k, d - 3 > k > 0O
a list lRk of the red k-faces is computed, as well as a list Pk of the purple
k-faces.
Phase 4. The incidence graph is updated.
Before giving all the details for each phase, let us first describe precisely the
data structure that stores the incidence graph. For each face F of dimension k
(O < k < d -1) of the convex hull, this data structure stores:
* the list of the sub-faces of F, which are the faces of dimension k - 1 incident
to F,
* the list of the super-faces of F, which are the faces of dimension k + 1
incident to F,
* the color of the face (red, blue, purple) in the current step, and
* a pointer p(F) whose use will very soon be clarified.
If F is a super-face of G, then a bidirectional pointer links the record for F in

the list of super-faces of G to the record for G in the list of sub-faces of F.
Phase 1. To find an initial red facet in conv(Ai-1), we take advantage of the
lexicographic order on the points in A. Because of this order, Ai-I is always
a vertex of conv(Ai-1) and there is at least one facet of conv(A4i-) containing
Ai- 1 which is red with respect to Ai. Indeed, let Fi-1 be the set of facets of
conv(A4i-) that contain Ai-, as a vertex. Let also H be the hyperplane whose
equation is x1 = x 1 (Ai- 1 ), and H+ the half-space bounded by H that contains
conv(Ai- 1 ), and H- the other half-space bounded by H. Since Ai- 1 is a vertex of
conv(Ai- ) with maximal abscissa, H+ U Ai-I contains the intersection of all the
half-spaces H1+ when F E Fi-1. Point Ai belongs to H-, and therefore cannot
belong to this intersection of half-spaces (see figure 8.6). Thus, at least one facet
F in FY j must be red with respect to Ai. All the facets of Fi- 1 were created
at the previous incremental step, so it suffices to store the list of facets created
during an incremental step and to traverse this list during the next incremental
step in order to find an initial red facet.
Phase 2. In the second phase, we use the connectedness of the set of red
facets (lemma 8.3.3). A depth-first traversal of the subgraph of red facets in
H-
Ai
4i-1
Figure 8.6. One of the facets of conv(Ai- 1 ) containing Ai-, must be red with respect to Ai.
the adjacency graph2 of conv(Ai-1), starting with the initial red facet that was
found in phase 1, visits all the facets visible from Ai, which we color red, and
their (d - 2)-faces, which we color red if they are incident to two red facets, or
purple if they are incident to a blue facet. The traversal backtracks whenever the
facet encountered was already colored red, or if it is a blue facet.
Phase 3. We now know all the red and purple (d - 2)-faces, and the red facets.
In this phase, all the remaining red and purple faces are colored, and their lists
are set up in order of decreasing dimensions. Assume inductively that all the red
and purple faces of dimension k' > k + 1 have already been identified and colored,
and that the lists lRk' and Pk, have already been set up. We process the k-faces in
the following way. Each sub-face of a face of Pk+, that has not yet been colored
is colored purple and added to the list Pk. Afterwards, each sub-face of Zk+1
that has not yet been colored is added to the list Rk.
Phase 4. To update the incidence graph, we proceed as follows. All the red
faces are removed from the incidence graph, and so are all the arcs adjacent to
these faces in the graph. The purple faces are processed in order of increasing
dimension k. If F is a k-face purple with respect to P, a new node is created for
the (k + 1)-face conv(F U {Ai}) and linked by an arc to the node for F in the
incidence graph. Also the pointer p(F) is set to point to the new node created
for conv(F U {Ai}). It remains to link this node to all the incident k-faces of
the form conv(G U {Ai}), where G is a (k - 1)-face incident to F. For each sub-
face G of F, its pointer p(G) gives a direct access to the node corresponding to
conv(G U {Ai}), and the incidence arc can be created.
2
The adjacency graph is already stored in the incidence graph, and need not be stored
separately (see subsection 8.1).
8.4. A deterministic algorithm 179
Phase 1 of each incremental step can be carried out in time proportional to the
number of facets created at the previous step. The total cost of phase 1 over all
the incremental steps is thus dominated by the total number of facets created.
At step i that sees the insertion of Ai, the cost of phase 2 is proportional to the
number of nodes visited during the traversal of the adjacency graph. The nodes
visited correspond to red facets of conv(Ai-1), and to the blue facets adjacent to
these red facets. The total cost of this phase is thus at most proportional to the
number of red facets of conv(Avi-) and of their incidences.
The cost of phase 3 is bounded by (a constant factor times) the number of
arcs in the incidence graph that are visited, and this number is the same as the
number of incidences between red or purple faces of conv(Ai-1).
Lastly, the cost of phase 4 is proportional to the total number of red faces and
of their incidences, plus the number of purple faces and of their incidences to
purple faces.
In short, when incrementally adding a point to the convex hull, the cost of
phases 2, 3, and 4 is proportional to the number of red or purple faces, plus the
number of faces incident to a red face, plus the number of incident purple faces.
Red faces and their incidences correspond to the nodes and arcs of the incidence
graph that are removed from the graph. The purple faces and the incidences
between two purple faces correspond to nodes and arcs of the incidence graph that
are added to the graph. The total cost of phases 2, 3, and 4 is thus proportional
to the number of changes undergone by the incidence graph. Since a node or arc
that is removed will not be inserted again (red faces will remain inside the convex
hull for the rest of the algorithm), this total number of changes is proportional to
the number of arcs and nodes of the incidence graph that are created throughout
the execution of the algorithm, which also takes care of the cost of phase 1. The
following lemma bounds this number.
Lemma 8.4.1 The number of faces and incidences created during the execution
of an incremental algorithm building the convex hull of n points in d dimensions
is O(n[(d+l)/ 21).
Proof. Lemma 8.3.1 shows that the subgraph of the incidence graph of conv(Ati)
induced by the faces created upon the insertion of Ai is isomorphic to the set of
faces of conv(A4i-) that are purple with respect to Ai. The number of incidences
between a new face and a purple face of conv(Ai-1) is also proportional to the
number of purple faces of conv(Ai-1). Finally, lemma 8.3.4 shows that the set of
purple faces of conv(A<i-) is isomorphic to a (d - 1)-polytope that has at most
i- 1 vertices. The upper bound theorem 7.2.5 shows that the number of these
faces and incidences between these faces, is O(iL(d-1)/2J). This is thus a bound on
the number of faces and incidences created upon inserting Ai. Summing over all
i, i = 1, . .. , n, the total number of facets and incidences created by the algorithm
is: n
0(iL(d 1)/2J) =(nL(d+1)-/2J
i~Ii
The storage needed by this algorithm is proportional to the maximum size of

the incidence graph stored at any step, which is 0(nLd/ 2 j). Taking into account
the initial sorting of the vertices, we conclude with the following result:
Theorem 8.4.2 The incremental algorithm builds the convex hull of n points in
d dimensions in time O(n log n + nL(d+l)/ 2 J ) and storage O(nLd/ 2 i).
This algorithm is optimal in the worst case when the dimension of the space is
even.
8.5 On-line convex hulls

Computing the convex hull of a set of points is one of the geometric problems to
which the randomization techniques developed in chapter 5 apply. Randomized
algorithms compute the convex hull of n points in optimal expected time, in any
dimension: O(n log n) in dimension 2 or 3, and O(nLd/ 2 J) in dimension d > 3. Let
us once again recall that the average value involved here is over all the possible
random choices of the algorithm, not over some spatial distribution of the points.
The only assumption we make on the points is that they are in general position.
The algorithm which we present here is an incremental on-line algorithm (or
semi-dynamic) that uses the influence graph method described in section 5.3, to
which we refer the reader if need be. The term "on-line" means that the algorithm
is able to maintain the convex hull of a set of points as the points are added one
by one without preliminary knowledge of the whole set. This algorithm is in fact
deterministic. Only the analysis is randomized and assumes that the order in
which the points are inserted is random.
Convex hulls in terms of objects, regions, and conflicts
This section applies the formalism described in chapter 4. In order to do so,

we must first recast the convex hull problem in terms of objects, regions, and
conflicts.
The objects are naturally the points of Ed. A region is defined as the union
of two open half-spaces. Such a region is determined by a set of d + 1 points in
general position. Let {P0 , P1 , ... , Pd- 1, Pd} stand for such a (d + 1)-tuple. Let
Hd be the hyperplane containing {Po, Pi,..., Pd-1} and Hi be the half-space
8.5. On-line convex hulls 181
bounded by Hd that does not contain Pd. Similarly let Ho be the hyperplane
containing {Pl, .. . , Pd- , Pd} and let Ho- be the half-space bounded by Ho that
does not contain Pd. The region determined by the (d + 1)-tuple is the union
of the two open half-spaces Hj and Ho-. A point conflicts with a region if it
belongs to at least one of the two open half-spaces that make up the region. In
this case, the influence domain of a region is simply the region itself.
With this definition of regions and conflicts, the convex hull of a set S of
n affinely independent points can be described as the set of regions defined and
without conflict over S. In fact, the regions defined and without conflict over S are
in bijection with the (d-2)-faces of conv(S). Indeed, let a region be determined by
the (d + 1)-tuple {Po, Pi, .. . , Pd- , Pd} of points in S. Because the points in S are
assumed to be in general position, if this region is without conflict over S, the two
d-1 simplices Fd = conv({Po, Pi, ..., Pd-1 }) and F0 = conv({PI,.. . ., Pd- , Pd})
are facets of conv(S), and the (d-2)-simplex G = FonFd = conv({Pi, .. ., Pd-,})
is the (d - 2)-face of conv(S) that is incident to both these facets. This region
will be denoted below by (Fo, Fd) or sometimes by (Ed, Fo). The set of regions
defined and without conflict over a set S therefore not only gives the facets of
conv(S), but also their adjacency graph. Using this information, it is an easy
exercise to build the complete incidence graph of conv(S) in time proportional
to the number of faces of all dimensions of conv(S) (see exercise 8.2).3
The algorithm
The algorithm is incremental, and in fact closely resembles that which is described
in section 8.4. The convex hull conv(S) of the current set S is represented by its
incidence graph. At each step, a new point P is inserted. The faces of conv(S)
can be sorted into three categories according to their color with respect to P,
as explained in section 8.3: red faces, blue faces, and purple faces. The on-line
algorithm, like the incremental algorithm, identifies the faces that are red and
purple with respect to P, then updates the incidence graph. The main difference
resides in the order with which the points are inserted. The on-line algorithm
processes the points in the order given by the input, and therefore cannot take
advantage of the lexicographic order to detect the red facets. For this reason, the
algorithm maintains an influence graph. As we may recall, the influence graph
3
1t would certainly be more natural to define a region as a open half-space determined by
d affinely independent points. In this case the region is one of the half-spaces bounded by the
hyperplane generated by these d affinely independent points, and a point conflicts with such a
region if it lies in this half-space. With these definitions, the facets of the convex hull conv(S)
of a set S of n points in Ed are in bijection with the regions defined and without conflict over S.
In fact, such a definition of regions is perfectly acceptable and so is an incremental algorithm
based on these definitions (see exercise 8.5). Such an algorithm, however, does not satisfy the
update conditions 5.2.1 and 5.3.3, and its analysis calls for the notion of biregion introduced in
exercise 5.7.
Figure 8.7. On-line convex hull: regions and conflicts.

The influence domain of region (AB, AC) is shaded, and the (d - 2)-faces
corresponding to regions conflicting with P are represented in bold.
is used mainly to detect the conflicts between the point to be inserted and the
regions defined and without conflict over the points inserted so far. The influence
graph is an oriented acyclic graph that has a node for each region that, at some
previous step in the algorithm, appeared as a region defined and without conflict
over the current subset of points. At each step of the algorithm, the regions
defined and without conflict over the current subset correspond to the leaves
of the influence graph. The arcs in this graph link these nodes such that the
following inclusion property is always satisfied: the influence domain of a node is
always contained in the union of the influence domains of its parents. 4 A depth-
first traversal of the influence graph can detect all the conflicts between the new
point P and the nodes in the graph. With a knowledge of the conflicts between
points P and the regions defined and without conflict over S, it is easy to find
the facets of conv(S) that are red with respect to P. Indeed:
* A region defined and without conflict over S that conflicts with P corre-
sponds to a red or purple (d - 2)-face of conv(S), since it is incident to two
(d - 1)-faces of conv(S), at least one of which is red (see figure 8.7).
* A region defined and without conflict over S that does not conflict with P
corresponds to a (d - 2)-face of conv(S) that is blue with respect to P.
In an initial step, the algorithm processes the first d+ 1 points that are inserted
into the convex hull. The incidence graph is set to that of the d-simplex formed
4Recall also that we frequently identify a node in the influence graph with the region that it
corresponds to, which for instance lets us speak of conflicts with a node, of the influence domain
of a node, or of the children of a region.
by these points, and the influence graph is initialized by creating a node for each
of the regions that correspond to the (d - 2)-faces of this simplex.
To describe the current step, we denote by S the current set of points already
inserted, and by P the new point that is being inserted. The current step consists
of a location phase and an update phase.
Locating. The location phase aims at detecting the regions killed by the new
point P. These are the regions defined and without conflict over S that conflict
with P. For this, the algorithm recursively visits all the nodes that conflict with
P, starting from the root.
Updating. If none of the regions defined and without conflict over S is found
to conflict with P, then P must lie inside the convex hull conv(S), and there
is nothing to update: the algorithm may proceed to the next insertion. If a
region corresponding to a (d - 2)-face of conv(S) is found to conflict with P,
however, then at least one of the two incident (d - 1)-faces is red with respect to
P. Starting from this red face, the incidence graph of conv(S) can be updated
into that of conv(S U {P}) by executing phases 2, 3, and 4 of the incremental
algorithm described above in section 8.4.
Its remains to show how to update the influence graph. Let us recall that
the nodes of the influence graph are in bijection with the (d - 2)-faces of the
successive convex hulls, and that the corresponding regions are determined by a
pair of adjacent facets, or also by the d + 1 vertices that belong to these facets.
To update the influence graph, the algorithm considers in turn each of the purple
(d - 2)-faces of conv(S), and each of the (d - 3)-faces incident to these faces.
1. Consider a (d - 2)-face G1 of conv(S) that is purple with respect to P, and
let (F1 , F1) be the corresponding region; F1 and F1' are two (d- 1)-faces of conv(S)
that are incident to G1 . We may assume that F1 is blue with respect to P and F1 is
red (see figure 8.8). The face GI is a (d-2)-face of conv(SU{P}) that corresponds
to the new region (F1 , F{'), where F{' is the convex hull conv(GI U {P}). A new
node of the influence graph is created for region (F1 , F{') and this node is hooked
into the influence graph as the child of (F1 , F,). In this way, the inclusion property
is satisfied. Indeed, let H1 and H' be the hyperplanes supporting conv(S) along
F1 and Ff, respectively. The hyperplane H7' supporting conv(S U {P}) along
F,' is also a hyperplane supporting conv(S) along G1 . As a consequence, the
half-space H`' that does not contain conv(S U {P}) is contained in the union
of the half-spaces H1 and H'7, which do not contain conv(S). The influence
domain of region (F1 , F{') is therefore contained within that of (F1 , F1).
2. Let K be a (d-3)-face of conv(S), purple with respect to P, and let G1 and
G2 be the purple (d - 2)-faces of conv(S) that are incident to K.5 Let (F1 , Ffl
and (F2 , F2) be the two regions corresponding to G1 and G2 , the faces F1 and F2
5 The set of purple faces of conv(S) being isomorphic to a (d - 1)-polytope (lemma 8.3.4),
any purple (d-3)-face of conv(S) is incident to exactly two purple (d -2)-faces (theorem 7.1.7).
/ .1
IFE
Figure 8.8. On-line convex hull: new regions when inserting a point P.
being blue with respect to P while faces Ff and F' are red (see figure 8.8). The
convex hull conv(K U {P}) is a (d - 2)-face of conv(S U {P}), and is incident
to the (d - 1)-faces Fj' = conv(Gi U {P}) and F2' = conv(G2 U {P})). In the
influence graph, a new node is created for the region (Ff', F2'), and hooked into
the graph to two parents which are the nodes corresponding to regions (F1 , F1)
and (F2 , F2). Let us verify that the inclusion property is satisfied. Indeed, the
influence domain of (Fl', F2') is the union H" - U H2' , where H"'- (resp. H2'-) is
the half-space bounded by hyperplane H1' (resp. H2') that supports conv(SU{P})
along Fj' (resp. F2') and does not contain conv(S U {P}). The half-space Hj'
is contained in the the influence domain of region (F1 , Ff), and similarly H2'-
is contained in the influence domain of (F2 , F2). Consequently, the influence
domain of (Ff', F2') is contained in the union of the influence domains of (F1 , Ff)
and (F2 , F2).
This description can be carried over almost verbatim to the case of dimension 2.
We need only remember that the polytope conv(S) has an empty face of dimen-
sion -1, incident to all of its vertices. If P is not contained within conv(S), the
empty face is purple and incident to the two purple vertices of conv(S) (see also
figure 8.9).
P
Figure 8.9. On-line convex hull in two dimensions.
Randomized analysis of the algorithm
In this randomized analysis, we assume that the points are inserted in a random
order. The performances of the algorithm are then estimated on the average,
assuming that all n! permutations are equally likely.
To apply the results in chapter 5, we must verify that the algorithm satisfies
the update condition 5.3.3 for algorithms that use an influence graph.
1. Testing conflict between a point and a region boils down to testing whether
a point belongs to two half-spaces, and can be performed in constant time.
2. The number of children of each node in the influence graph is bounded.

In fact, each node has d children, or none. Indeed, when inserting a point
P, the node corresponding to a purple (d - 2)-face G of conv(S) receives d
children: one for the (d-2) face G of conv(SU{P}), and d- 1 corresponding
to conv(KU {P}) for each (d- 3)-subface K of G. The nodes corresponding
to red or blue (d - 2)-faces of conv(S) do not receive children. The nodes
corresponding to the red or purple (d - 2)-faces of conv(S) are killed by P:
they no longer correspond to regions without conflict and will not receive
children after the insertion of P.
3. The parents of a region created by a point P are recruited among the regions
killed by P. From the analysis of phases 2, 3, and 4 of the incremental step
in section 8.4, we can deduce that updating the incidence graph takes time
proportional to the total number of red and purple faces of conv(S) and
of their incidences. If every (d - 2)-face of the convex hull is linked by a
bidirectional pointer with the corresponding node in the influence graph,
it is easy to see that updating the influence graph takes about the same
time as updating the incidence graph. The set of points being in general
position, the facets of conv(S) are simplices; thus the number of red or
purple faces and of their incidences is proportional to the number of red
facets of conv(S). Each of these red facets is incident to d - 1 red or purple
(d- 2)-faces of conv(S), each of which corresponds to a region that conflicts

with P. Each region defined and without conflict over S that conflicts with
P corresponds to a (d - 2)-face of conv(S) that is incident to one or two red
facets. As a result, the number of red facets of conv(S), and therefore the
complexity of the update phase, is proportional to the number of regions
killed by the new point P.
Since the update conditions are satisfied, the randomized analysis of the onl-line
convex hull computation can now be established readily by theorem 5.3.4 which
analyzes algorithms that use an influence graph. The number of regions without
conflict defined over a set S of n points in a d-dimensional space is exactly the
number of (d - 2)-faces of the convex hull conv(S), which is Q(nLd/21) according
to the upper bound theorem 7.2.5.
Theorem 8.5.1 An on-line algorithm that uses the influence graph method
to build the convex hull of n points in d dimensions requires expected time
O(nlogn + n Ld/ 2 J), and storage O(nLd/ 2 i). The expected time required to per-
form the n-th insertion is O(logn + nLd/ 2i-1).
8.6 Dynamic convex hulls

The previous section shows that it is possible to build on-line the convex hull of
a set of points in optimal expected time and storage, using an influence graph.
Such an algorithm is called semi-dynamic, since it can handle insertions of new
points. Fully dynamic algorithms, however, handle not only insertions but also
deletions.
The possibility of deleting points makes the task of maintaining the convex hull
much more complex. Indeed, during an insertion, the current convex hull and
the new point entirely determine the new convex hull. After a deletion, however,
points that were hidden inside the convex hull may appear as vertices of the new
convex hull. A fully dynamic algorithm must keep, in one way or another, some
information for all the points in the current set, be they vertices of the current
convex hull or not.
The goal of this section is to show that the augmented influence graph method
described in chapter 6 allows the convex hull to be maintained dynamically.
The algorithm which we now present uses again the notions of objects, regions,
and conflicts as defined in the preceding section. It conforms to the general
scheme of dynamic algorithms described in chapter 6, to which the reader is
referred should the need arise. Besides the current convex hull (described by the
incidence graph of its faces), the algorithm maintains an augmented influence
graph whose nodes correspond to regions defined over the current set. After
8.6. Dynamic convex hulls 187
each deletion, the structure is rebuilt into the exact state it would have been in,
had the deleted point never been inserted. Consequently, the augmented influence
graph only depends on the sequence E = {P 1 , P2 , . . ., Pn} of points in the current
set, sorted by chronological order: Pi occurs before Pj if the last insertion of Pi
occurred before the last insertion of Pj.
Let us denote by la s) the augmented influence graph obtained for the chrono-
logical sequence E. The nodes and arcs of Ia(s) are exactly the same as those of
the influence graph built by the incremental algorithm of the preceding section,
when the objects are inserted in the order given by E. We denote by SI the
subset of S formed by the first 1 objects in E. The nodes of la(E) correspond
to the regions defined and without conflict over the subsets SI, for 1 = 1, . . , n. .
The arcs of -;a(s) ensure both inclusion properties: that the domain of influence
of a node is contained in the union of the domains of influence of its parents, and
that a determinant of this node is either the creator of this node or is contained
in the union of the sets of determinants of its parents. Moreover, the augmented
influence graph contains a conflict graph between the regions that correspond
to nodes in the influence graph, and the objects in S. This conflict graph is
implemented by a system of interconnected lists such as that described in sec-
tion 6.2: each node of the conflict graph has a list (sorted in chronological order)
of the objects that conflict with the corresponding region; also, for each object
we maintain a list of pointers to the nodes in the influence graph that conflict
with that object. The record corresponding to an object in the conflict list of a
node is interconnected with the record corresponding to that node in the conflict
list of the object.
Insertion
Inserting the n-th point into the convex hull is carried out exactly as in the on-line
algorithm described in section 8.5, except that while we are locating the object in
the influence graph, each detected conflict is added to the interconnected conflict
lists.
Deletion
Let us now consider the deletion of point Pk. For l = k, . . . , n, we denote by S,
the subset Si \ {Pk} of S, and by E' the chronological sequence {P1 , . . ., Pk-1,
Pk+I .... Pn}. When deleting Pk, the algorithm rebuilds the augmented influence
graph, resulting in -Ta(VY). For this, we must:
1. remove from the graph la(s) the destroyed nodes, which correspond to
regions having Pk as a determinant, 6
6
Recall that an object is a determinant of a region if it belongs to the set of objects that
determine this region.
12
Figure 8.10. Convex hull: creator and killer of a region.

Points are numbered by chronological ranks. Unnumbered points have rank
greater than 12. Region (61,64) has point 6 as its creator and point 12 as its
killer.
2. create a new node for each region defined and without conflict over one of
the subsets S', 1 = k + 1, . . ., n that conflicts with Pk,
3. set up the new arcs that are incident to the new nodes. The new nodes must
be hooked to their parents which may or may not be new. The unhooked
nodes, which are nodes of Ia(s) that are not destroyed but have destroyed
parents, must be rehooked.
Before we describe the deletion algorithm, it is useful to recall a few definitions.

A region G of Ta(s) is created by PI or also Pi is the creator of G, if PI is among
all determinants of G the one with highest chronological rank. A region G of
Ia(s) is killed by PI or also PI is the killer of G if P1 has the lowest rank among
the points that conflict with G (see figure 8.10).
The deletion algorithm proceeds in two substeps: the location phase and the
rebuilding phase.
Locating. During this phase, the algorithm identifies the nodes in Ia(s) that
are killed by Pk, and the destroyed and unhooked nodes. For this, the algorithm
recursively visits all the nodes that conflict with Pk or have Pk as a determinant,
starting at the root. During the traversal, the algorithm removes Pk from the
conflict lists, and builds a dictionary of the destroyed or unhooked nodes for use
during the rebuilding phase.
Rebuilding. During this phase, the algorithm creates the new nodes, hooks
them to the graph, builds their conflict lists and rehooks the unhooked nodes.
For this, the algorithm considers in turn all the objects P1 of rank I > k that
are the creators of some new or unhooked node. A point Pl of rank I > k is the
creator of some new or unhooked node if and only if there exists a region defined
and without conflict over S8>- which conflicts with both P1 and Pk (lemma 6.2.1).
When processing P1, we call a region critical if it is defined and without conflict
over S' 1- but conflicts with Pk. The criticalzone is the set of all critical regions.
The critical zone evolves as we consider the objects PI in turn. At the beginning
of the rebuilding phase, the critical regions are the regions of Ia(s) that are
killed by Pk. Subsequently, the critical regions are either regions in Ta(s) that
are killed by Pk, or new regions in Ia(E'). At each substep in the rebuilding
phase, the next point to be processed is the point of smallest rank among all the
points that conflict with one or more of the currently critical regions. To find
this point, the algorithm maintains a priority queue Q of the points in E' that
are the killers of critical regions. Each point PI in Q also stores the list of the
current critical regions that it kills. The priority queue Q is initialized with the
killers in E' of the regions in Ia(E) that were killed by Pk.
At each substep in the rebuilding phase, the algorithm extracts the point Pi of
smallest rank in Q, and this point is then reinserted into the data structure. To
reinsert a point means to create new nodes for the new regions created by P1, to
hook them to the influence graph, and to rehook the unhooked nodes created by
Pi. The (d-2)-faces of conv(S>-) that are red or purple with respect to the point
Pk that is removed correspond to critical regions and are, below, called critical
faces. Unless explicitly stated, the color blue, red, or purple, is now given with
respect to the point Pi that is being reinserted. The regions that are unhooked
or new and created by P1 can be derived from the critical purple (d - 2)-faces
and their (d - 3)-subfaces, which will be considered in turn by the algorithm.
1. Processing the critical purple (d - 2)-faces
Along with point PI, we know the list of critical regions with which it conflicts.
These regions correspond to the critical red or purple (d - 2)-faces, and a linear
traversal of this list allows the sublist of its critical purple (d - 2)-faces to be
extracted.
Let G be a critical purple (d - 2)-face, and (F, F') be the corresponding region;
F and F' are (d- 1)-faces of conv(Sl-1), both incident to G, and we may assume
that F is blue with respect to P1 while F' is red (see figure 8.11 in dimension 3
and figure 8.12 in dimension 2.)
In the convex hull conv(S'), G is a (d - 2)-face that corresponds to (F, F"), a
region defined and without conflict over S', where F" is the convex hull
conv(G U {P 1}) (see figure 8.11 in dimension 3 and figure 8.12 in dimension 2.)
If region (F, F") conflicts with Pk (see figures 8.11a and 8.12a), then it is a new
region created by P1. In the augmented influence graph, a new node is created
for this region, with node (F, F') as parent. The conflict list of (F, F") can be
0 C 8. nP
.Pk .*Pk
/ / \/ /\
k-)
Figure 8.11. Deleting from a 3-dimensional convex hull: handling critical purple (d - 2)-
faces.
(a) (F, F") is a new region.
(b) (F, F") is an unhooked region.
set up by selecting the objects in conflict with (F, F") from the conflict list of
(F, F'). The killer of (F, F") in A' is inserted in the priority queue Q if it was not
found there. Finally, region (F, F") is added to the list of critical regions killed
by this point.
If region (F, F") does not conflict with Pk (see figures 8.11b and 8.12b), then it
corresponds to an unhooked node created by P1. This node is found by using the
dictionary D of destroyed and unhooked nodes, and hooked as a child of (F, F').
2. Handling the critical purple (d - 3)-faces
Critical purple (d - 3)-faces are subfaces of critical purple (d - 2)-faces. 7 For

each such (d - 3)-face, we must know the at most two critical purple (d - 2)-faces
incident to it. To find them, we build an auxiliary dictionary D' of the (d - 3)-
subfaces of critical purple (d - 2)-faces. Each entry in the dictionary D' for a
7
According to lemma 8.3.4, each critical purple (d - 3)-face is incident to two purple (d - 2)-
faces, at least one of which is critical.
Pk p. Pi Pk
G 6
(a) (b)
Figure 8.12. Deleting from a 2-dimensional convex hull: handling critical purple (d -2)-
faces.
(a) (F, F") is a new region.
(b) (F, F") is an unhooked region.
(d- 3)-face K has two pointers for keeping track of the critical purple (d-2)-faces
incident to K.
Let K be such a (d - 3)-face (see figure 8.13 in dimension 3 and figure 8.14 in
dimension 2). We denote by GI and G2 the two purple (d - 2)-faces incident to
K. At least one of them is a critical face, but not always both. We denote by
(F1 , Ff) and (F2 , F2) the regions corresponding to faces G1 and G2 of the convex
hull conv(S'-1). We may assume that facets F1 and F2 are blue, while F1 and
F2 are red.
The (d-2)-face conv(KU{Pj}) of conv(S') corresponds to some region (Ft', F2'),
where Fi' = conv(Gi U {Pi}) and F2' = conv(G2 U {P1}) (see figure 8.13; see also
figure 8.14, in dimension 2, in which K is the empty face of dimension -1, and
G1 and G2 are the two vertices of conv(S' 1-), both purple with respect to PI).
2.a If both GC and G2 are critical faces, the corresponding nodes in Ia(E')
may be retrieved through dictionary D'.
2.a.1 If region (Fl', F2') conflicts with Pk (see figure 8.14a), it is a new region
created by Pl; a node is created for this region, and inserted into the influence
graph with both (F1 , F) and (F2 , F2) as parents. The conflict list of (F',F2')
may be obtained by merging the conflict lists of (F1 , F1) and (F2 , F2), and then
selecting from the resulting list the objects that conflict with (Fl', F2'). Merging
the conflict lists can be carried out in time proportional to the total length,
because these lists are ordered chronologically. 8 The killer of (F',F2') in the
sequence E' is inserted into the priority queue Q if not found there, and region
(Ff', F2') is added to the list of critical regions killed by this point.
8
An alternative to this solution is to forget about ordering the conflict lists and to resort to
. Pi
. Pk
/ / p.\ \
I \PO
Figure 8.13. Deleting from a 3-dimensional convex hull: handling critical purple (d -3)-
faces.
2.a.2 If region (F',F"') does not conflict with Pk (see figure 8.14b), then this
region is an unhooked region created by Pi. It suffices to find the corresponding
node using dictionary D and hook it back to the nodes corresponding to (FI, F1)
and (F2 ,F2).
2.b When only one of the purple (d - 2)-faces GC and G2 incident to K is
critical, say G1 , the algorithm must find in the influence graph the node cor-
responding to G2 , the other purple (d - 2)-face incident to K. Lemma 8.6.1
below proves that, in this case, conv(K, P1 ) is a (d - 2)-face of conv(S1 ) which
corresponds to a destroyed or unhooked node of Xa(E), whose parents include
precisely the node corresponding to region (F2 , F2). To find (F2 , F2), we may
therefore search in the dictionary D of destroyed or unhooked nodes, created
by PI, corresponding to the (d - 2)-face conv(K, Pi) of conv(Si). This node is
uniquely known from this criterion, because we know not only the (d - 2)-face
conv(K, P1 ) of its corresponding region, but also its creator P1 .
Lemma 8.6.1 Let K be a (d - 3)-face of conv(Sl- 1 ) incident to two purple

faces G1 and G2 , only one of which is critical, say G1 . Then conv(K, PI) is a
the method used in section 6.4 for merging the conflicts lists of trapezoids.
, Pk
Pl / \
*I'F4' \ G
1 F1 - Pk
(a) (b)
Pk
L Pk / \
/ \\ /
,// I
/ \\
2 ~/\
G2
I 11
(C) ki) (e)
Figure 8.14. Deleting from a 2-dimensional convex hull: handling the critical purple (d-3)-
faces. Critical purple (d - 3)-face K here is the empty face of dimension -1.
G1 and G2 are its two purple vertices.
(a) G, and G2 are critical, (Fl',F2") is new.
(b) GI and G2 are critical, (F{', F2") is unhooked.
(c) G1 is critical, G2 is not, and GC is not a face of conv(S 1-1 ).
(d) G1 is critical, G2 is not, and GI is a face of conv(S 1- 1 ), but not purple
with respect to PI.
(e) G1 is critical, G2 is not, and G1 is a face of conv(S-1), this time purple
with respect to PI.
(d-2) -face of conv(Sl), its correspondingnode in la(s) is destroyed or unhooked,

and one of its parents is the region (F2 , F') that corresponds to the face G2 of
conv(S1- 1 ).
Proof. For the proof, imagine that Pk then P1 are inserted into S,-,: then we
obtain successively SI-1 and SI.
The (d - 2)-face K of conv(S' 1-) is purple with respect to Pk since it belongs
to a critical (d - 2)-face as well as to a non-critical (d - 2)-face. As a result, both
K and conv(K, Pk) are faces of conv(SI-1 ).
Since it is not critical, the (d - 2)-face G 2 is also a (d - 2)-face of conv(SI-'),
and its corresponding region is still (F 2 , F2), hence face G 2 of conv(Sj- 1 ) is purple
with respect to PI.
The (d - 3)-face K of conv(S 1- 1 ) is purple with respect to PI since it is incident

to G2 . As a result, conv(K, Pi) is a (d - 2)-face of conv(St), incident to the
(d - 1)-face conv(G 2 , Pl). In the graph la(E), one of the parents of the node
corresponding to the (d - 2)-face conv(K, P1 ) of conv(S 1), is region (F2 , F2).
We now have to show that the (d -2)-face conv(K, PI) of conv(Si) corresponds
to a region that is either destroyed or unhooked when Pk is deleted. For this, we
consider the face G1 of conv(S -1). The situation is one of three (see figure 8.14c,
d, e). If this face is red with respect to Pk (see figure 8.14c), then it is not a face
of conv(S 1 _1 ) any more. If this face is purple with respect to Pk, it remains a
face of conv(S 1- 1 ), but it may be blue (see figure 8.14d) or remain purple (see
figure 8.14e) with respect to PI.
In the first two cases, conv(K, Pk) is necessarily a (d -2)-face of conv(S1_J),
purple with respect to PI. Indeed, the set of those purple faces is isomorphic
to a (d - 1)-polytope, and since G1 is not purple with respect to P1, it must be
replaced by another (d - 2)-face incident to K which can only be conv(K, Pk).
Consequently, the region corresponding to the (d-2)-face conv(K, PF) of conv(Sl)
is region (conv(K, Pi, Pk), F2') which is destroyed during the deletion of Pk.
In the third case, F1'= conv(GI, Pl) must be a facet of conv(S1 ), and the region
that corresponds to the (d - 2)-face conv(K, P1) of conv(SI) is region (Fl', F2'),
which is an unhooked region created by P1. nl
Once the node corresponding to the (d-2)-face G2 of conv(Sl) has been found,
operations can resume as before, apart from a simple detail. If the region (Ff', F2')
that corresponds to the (d-2)-face conv(K, P1) of conv(S ) is new, then its conflict
list may be obtained by merging that of the critical region (FI, F{) and that of the
destroyed region (conv(K, Pi, Pk), F2'). (We do this in order to avoid traversing
the conflict list of region (F2 , F2) corresponding to face G2, which is neither new
nor destroyed.)
The algorithm is deterministic. Yet the analysis given here is randomized and
assumes the following probabilistic model:
* the chronological sequence E is a random sequence, each of the n! permu-

tations being equally likely;
* each insertion concerns, with equal probability, any of the objects present
in the current set immediately after the insertion;
* each deletion concerns, with equal probability, any of the objects present
in the current set immediately before the deletion.
8. 7. Exercises 195
Theorem 8.6.2 Using an augmented influence graph allows the fully dynamic
maintenance of the convex hull of points in Ed, under insertion or deletion of
points. If the current set has n points:
* the structure requires expected storage O(n log n + nLd/2 I),
* inserting a point takes expected time O(logn + nLd/ 2 ] -1),
* deleting a point takes expected time O(log n) in dimension 2 or 3 and time
O(tnLd/ 2 -1)l in dimension d > 3. The parametert represents the complexity
of an operation on the dictionaries used by the algorithm (t O(logn) if
balanced binary trees are used, t = 0(1) if perfect dynamic hashing is used.)
Proof. During the rebuilding phase in a deletion, the number of queries into the
dictionary of destroyed or unhooked nodes is at most proportional to the number
of destroyed or unhooked nodes. For each point P1 that is reinserted, the number
of updates or queries on the dictionary of (d - 3)-faces incident to critical purple
(d - 2)-faces is proportional to the number of these critical purple (d - 3)-faces.
Thus, the total number of accesses to the dictionaries is proportional to the total
number of critical faces encountered that correspond to new or killed nodes. The
conflict lists of new nodes can be set up in time at most proportional to the total
sizes of the conflict lists of new or killed nodes. All the other operations performed
during a deletion, except handling the priority queue, take constant time, and
their number is proportional to the number of destroyed, new, or unhooked nodes.
As a result, the algorithm indeed satisfies the update condition 6.3.5 for algo-
rithms that use an augmented conflict graph. Its randomized analysis is therefore
the same as in section 6.3, and is given in theorem 6.3.6 in terms of fo(l, S), the
expected number of regions defined and without conflict over a random I-sample
of S. For the case of convex hulls, since the number of such regions for any sample
is bounded in the worst case by Q(lId/2j) (upper bound theorem 7.2.5), so is their
expectation fo(1, S). This results in the performance given in the statement of
theorem 6.3.6. In dimension 2 or 3, the number of operations to be performed on
the dictionaries and on the priority queue is 0(1) whereas handling the conflict
lists always takes O(log n) time. Therefore, it suffices to implement dictionar-
ies and priority queues with balanced binary trees. In dimensions higher than
3, deletions have supra-linear complexity, and the priority queue may be imple-
mented using a simple array. °
8.7 Exercises
Exercise 8.1 (Extreme points) Extreme points in a set of points are those which are
vertices of the convex hull. Show that to determine the extreme points of n points in EE2
is a problem of complexity e (n log n).
Hint: You may use the notion of an algebraic decision tree: an algebraic tree of degree a
is a decision tree where the test at any node evaluates the sign of some algebraic function
of degree a for the inputs. Loosely stated, a result by Ben-Or (see also subsection 1.2.2)
says that any algebraic decision tree that decides whether a point in Ek belongs to some
connected component W of Ek must have a height h = Q(logc(W) - k), where c(W) is
the number of connected components of W.
Exercise 8.2 (Adjacency graph) Let a simplicial d-polytope be defined as the convex
hull of n points. Show that knowledge of the facets of the graph (given by their vertices),
along with their adjacencies, suffices to reconstruct the whole incidence graph of the
polytope in time linear in the size of the adjacency graph, which is Q(nLd/2J).
Exercise 8.3 (1-skeleton) This problem is the dual version of its predecessor. Let a
simple d-polytope be defined as the intersection of n half-spaces. Suppose that the 1-
skeleton is known, that is the set of its vertices and the arcs joining them. Each vertex
is given as the intersection of d bounding hyperplanes. Show that the whole incidence
graph of the polytope may be reconstructed in time 0(nLd/2J).
Exercise 8.4 (Degenerate cases) Generalize the incremental algorithm described in

section 8.4 to build the convex hull of a set of points which is not assumed to be in
general position.
Exercise 8.5 (On-line convex hulls) Give an algorithm to compute on-line the con-
vex hull of a set of points in Ed, by using an influence graph whose nodes correspond to
regions which are half-spaces. Give the randomized analysis of this algorithm.
Hint: Each region, or half-space, is now determined by a subset of d affinely independent

points that generates its bounding hyperplane. A point conflicts with a half-space if it
lies inside. The regions defined and without conflict over a set S are in bijection with
the facets, or (d - 1)-faces, of the convex hull conv(S) of S.
Upon inserting a point P into S, the regions killed by P correspond to the facets of
conv(S) that are red with respect to P, and the regions created by P correspond to the
facets conv(G U P) of conv(S U P) where G is any (d - 2)-face of conv(S) that is purple
with respect to P.
1. Let F1 and F2.be the facets of conv(S) incident to a (d- 2)-face G, which is purple
with respect to P. Show that the node of the influence graph that corresponds to the
facet conv(G U P) must have both nodes corresponding to F1 and F2 as parents.
2. In this manner, a node in the graph may receive a child without being killed,
therefore the number of children of a node is not bounded any more. The maximum
number of parents is two, however. For this particular problem, define and use the
notion of a biregion that was introduced in exercise 5.7, and show that the expected
complexity of the algorithm is 0(n log n + n 2 j).
2
Exercise 8.6 (Intersection of half-spaces) Give a randomized incremental algorithm

that uses a conflict graph to build the intersection of n half-spaces in Ed whose bound-
ing hyperplanes are in general position. Try to achieve an expected running time of
O(nrlogn + n L§i). Give an on-line version of the preceding algorithm that uses an influ-
ence graph.
Show that in the version of the algorithm that uses a conflict graph, the storage
requirements may be lowered if only one conflict is stored for each half-space.
Hint: Objects are half-spaces, regions are segments. A segment is determined by d + 1

half-spaces, or rather by the d + 1 hyperplanes which bound these half-spaces. The line
that supports the segment is the intersection of d- 1 hyperplanes, and the endpoints of the
segment are the intersections of this line with the two remaining hyperplanes. A segment
conflicts with a half-space if it has an intersection with the (open) complementary half-
space. The segments defined and without conflict over these half-spaces are precisely the
edges of the polytope, obtained as the intersection of the half-spaces.
When a new half-space H+ bounded by a hyperplane H is inserted, the conflict graph
identifies all the edges that lie in H-, which disappear, and those that intersect H. An
edge E that intersects H gives a shorter edge E' c E, and the conflict list of E' is set
up by traversing that of E. To obtain the new edges that lie in H, it suffices to follow,
for each 2-face F incident to each edge E that intersects H, the edges of F that conflict
with H+ until the second edge E' of F that intersects H is found. The new edge F n H
has vertices E n H and E' n H. Its conflict list can be obtained by traversing the conflict
lists of the edges of F killed by H+. Knowing the 1-skeleton, the whole incidence graph
of the intersection may be updated.

The incremental deterministic convex hull algorithm described in section 8.4 is due to
Seidel [201]. This algorithm is also described in detail in Edelsbrunner's book [89] where
degenerate cases are also handled (see exercise 8.4).
The first randomized algorithm to build convex hulls was due to Clarkson and Shor [71].
This algorithm uses a conflict graph and in fact solves the dual problem of computing
the intersection of half-spaces (see exercise 8.6). The on-line algorithm that uses an
influence graph is due to Boissonnat, Devillers, Teillaud, Schott and Yvinec [28]. The
dynamic algorithm presented in section 8.6 is due to Dobrindt and Yvinec [86]. Clark-
son, Mehlhorn, and Seidel [70] and independently Mulmuley [176, 177] proposed similar
solutions for dynamically maintaining convex hulls.
Chazelle [46] proposed a deterministic algorithm that is optimal in any dimension
greater than 3. This algorithm is a derandomized incremental algorithm, and uses the
method of conditional probabilities to determine which point must be inserted next.
Br6nnimann, Chazelle, and Matousek [36] and Br6nnimann [35] give a simpler version
which works in any dimension.
The lower bound proposed in exercise 8.1 to identify the extreme points in a set of
points in the plane is due to Yao [220]. The solution to exercise 8.1 can be found in the
book by Preparata and Shamos [192].
Chapter 9
Convex hulls
in two and three dimensions
There are many algorithms that compute the convex hull of a set of points in
two and three dimensions, and the present chapter does not claim to give a
comprehensive survey. In fact, our goal is mainly to explore the possibilities
offered by the divide-and-conquer method in two and three dimensions, and to
expand on the incremental method in the case of a planar polygonal line.
In dimension 2, the divide-and-conquer method leads, like many other methods,
to a convex hull algorithm that is optimal in the worst case. The main advantage
of this method is that it also generalizes to three dimensions while still leading
to an algorithm that is optimal in the worst case, which is not the case for the
incremental method described in chapter 8. The performances of this divide-and-
conquer algorithm rely on the existence of a circular order on the edges incident to
a given vertex. In dimensions higher than three, such an order does not exist, and
the divide-and-conquer method is no longer efficient for computing convex hulls.
The 2-dimensional divide-and-conquer algorithm is described in section 9.2, and
generalized to dimension 3 in section 9.3. But before these descriptions, we must
comment on the representation of polytopes in dimensions 2 and 3, and describe
a data structure that explicitly provides the circular order of the edges or facets
around a vertex of a 3-dimensional polytope.
The problem of computing the convex hull of a polygonal line is interesting
from the point of view of its complexity. Indeed, the lower bound of Q(nlogn)
on the complexity of computing the convex hull of n points does not hold if the
points are assumed to be the vertices of a simple polygonal line. In fact, any
simple polygonal line that links the points in a given set determines an order on
those points which is not completely unrelated to the order of the vertices on
the boundary of the convex hull. In section 9.4, we show how it is possible to
compute in time O(n) the convex hull of a set of n points given as the vertices
9.1. Representation of 2- and 3-polytopes 199
(a) (b)
Figure 9.1. Representation of a 2-polytope: (a) the incidence graph, (b) the circular list
of its vertices.
of a simple polygonal line, using an incremental algorithm that takes advantage

of the order of the points along the polygonal line.
As in chapter 8, and to simplify the presentation of the algorithms, we assume
that the sets of points to be processed are in general position. We leave it to the
reader to work out how to modify the algorithms so that they can handle any
sets of points without increasing the complexity.
9.1 Representation of 2- and 3-polytopes

Representation of 2-polytopes
The proper faces of a 2-polytope consist of its vertices and edges. Each edge
is incident to two vertices and each vertex to two edges. In fact, the incidence
graph of a 2-polytope is a cyclic graph that alternates vertices and edges (see
figure 9.1a). Without losing information, a 2-polytope may be represented by the
doubly-linked circular list of its vertices. Either direction in this list corresponds
to an order on the boundary of the polytope. If the plane that contains the 2-
polytope has an orientation, it induces an order on this boundary that is called
the direct (or counter-clockwise) order of the vertices, and the reverse order is
called the indirect (or clockwise) order of the vertices.
Representation of 3-polytopes
Generally speaking, any 2-face of a d-polytope is a 2-polytope and there is a

circular ordering on the set of edges and vertices contained in a given 2-face. By
duality, there is also a circular ordering on the set of (d - 1)- and (d - 2)-faces of
a d-polytope containing a given (d - 3)-face.
In particular, for a 3-polytope, there is circular order on the set of edges and
200 Chapter 9. Convex hulls in two and three dimensions
org(E)
sym(E
Figure 9.2. Representation of a 3-polytope.
vertices contained in any given facet, and also on the set of edges and facets
containing any given vertex. Let us agree that supporting hyperplanes are ori-
ented by the outward normal, pointing into the half-space that does not contain
the polytope. This orientation induces a circular order on the edges and vertices
contained in a facet, which we again call the direct (or counter-clockwise) order;
the other orientation induces the indirect (or clockwise) order.
Cycles of edges of a 3-polytope, around a vertex or a facet, are not stored in the
incidence graph of the polytope. These cycles are commonly used by algorithms
that deal with 3-polytopes, however, and for this reason an alternative data
structure is often preferred: the edge-list representation stores the order of the
edges incident to a given vertex or to a given facet of the 3-polytope.
In this structure, vertices and facets are represented by a single node, whereas
an edge is stored in a double node, one for each possible orientation of the edge.
To orient an edge is to choose an order on its two vertices: the origin is the first
vertex of the edge while the end is the last one. We can now make a distinction
between the two facets incident to an oriented edge: the facet incident on the left,
or left incident facet, is the one whose direct orientation traverses the edge from
origin to end, and the right incident facet is the one whose indirect orientation
traverses the edge from origin to end. In this data structure, each edge node
stores five pointers displayed in figure 9.2:
org(E) points towards the origin vertex of the oriented edge E,
left(E) points towards the facet incident to E on the left,
sym(E) points towards the node for the reverse edge. In this way, org(sym(E))
points towards the end of E and left(sym(E)) towards the facet incident
to E on the right,
9.2. Divide-and-conquer convex hulls in dimension 2 201
onext(E) points towards the edge E' that shares the same origin as E, and whose
facet incident on the right is the same as the facet incident to E on the left:
org(E') = org(E)
left(sym(E')) = left(E),
lnext(E) points towards the edge E" that follows E in the circular order of edges
on the boundary of left(E):
left(E")= left(E)
org(E") = org(sym(E)).
Conversely, each facet keeps a pointer to one oriented edge that has the facet
as its left incident facet. The entire edge cycle on the boundary of a facet may be
obtained in direct (resp. indirect) order by repeated applications of the functor
nextt) (resp. sym(onexto)). The time taken for this operation is constant per
edge on the boundary.
Each vertex node also keeps a pointer to one of the edges originating at that
vertex. We define the order of edges around a vertex as follows: All the edges
originating at that vertex may be obtained in direct (resp. indirect) order by
repeated applications of the functor onext() (resp. lnext(symo)). Again, the
time needed to enumerate these edges is constant per each edge.
9.2 Divide-and-conquer convex hulls

in dimension 2
Building the convex hull of a set A of points using the divide-and-conquer method
consists of dividing the set A into two subsets A1 and A 2 of equal size, recursively
computing the convex hull conv(Ai) and conv(A 2 ) of each set, and merging them
into a single convex hull conv(A) = conv(conv (Al) U conv(AA2)).
The divide-and-conquer algorithm which we present here divides the set A into
two sets Al and A2 separated by a vertical line. To efficiently split the subsets
in the recursive steps, the algorithm begins by sorting once and for all the set
A in order of increasing abscissae. Again, we may assume that A is in general
position.
The only subtlety of the algorithm lies in the method used to merge the convex
hulls conv(Al) and conv(A 2 ) into the convex hull conv(A). We assume that both
conv(Al) and conv(A 2 ) are represented by the doubly linked circular lists of their
vertices.
The edges of conv(Al) and conv(A 2 ) can be split into two categories, which it
is helpful to color again red or blue as follows:
* an edge of conv(Ai) is red with respect to conv(A2) if it is not an edge of

conv(A),
* an edge of conv(A 1 ) is blue with respect to conv(A 2 ) if it is an edge of

conv(A).
The color of an edge of conv(A 2 ) with respect to conv(Al) is defined symmetri-

cally. Intuitively, the red edges of conv(Ai) are those that would be lit if conv(A 2 )
was an extended source of light. Blue edges would remain in the shadow of
conv(Al). Using the terminology of chapter 8, an edge E of P is red with respect
to a point A if it is visible from A, meaning that A belongs to the half-plane
HE that is bounded by the line supporting P along E and that does not contain
P. Thus, an edge of conv(Al) is blue with respect to conv(A 2 ) if it is blue with
respect to all the vertices of conv(A2), and red with respect to conv(A 2 ) if it is
red with respect to at least one vertex of conv(A 2 ). The edges of conv(A2) that
are red or blue with respect to conv(Al) are defined symmetrically. In the rest of
this section, provided there is no ambiguity, we say red or blue for short instead
of red or blue with respect to conv(Ai) or to conv(A 2 ).
The vertices of conv(Al) can fall into one of the following three categories with
respect to conv(A2):
* a vertex of conv(Al) is red with respect to con v(A 2) if it is not a vertex of

conv(A),
* a vertex of conv(Al) is blue with respect to conv(A 2 ) if it is a vertex of

conv(A) that is not incident to a red edge,
* a vertex of conv(Al) is purple with respect to conv(A 2 ) if it is a vertex of

conv(A) that is incident to at least one red edge.
The color of a vertex of conv(A 2 ) with respect to conv(Al) is defined symmet-

rically. For convenience, we will say red, blue, or purple instead of red, blue, or
purple with respect to conv(Al) or to conv(A 2 ).
It follows from lemma 8.3.3 and section 8.4 that the set of edges of conv(A 1 )
that are red with respect to some vertex A of conv(A2) is connected among the
set of edges of conv(Ai) and contains at least one edge incident to the vertex
of conv(Al) that has maximal abscissa. As a consequence, the set of edges of
conv(A 1 ) that are red with respect to conv(A 2 ) is connected and not empty,
and the set of edges of conv(Ai) that are blue with respect to conv(A 2 ) is also
connected. If there is such a blue edge, then conv(Ai) has exactly two purple
vertices, each adjacent to a blue edge and to a red edge. If there is no blue
edge on the boundary of conv(Al), then conv(Al) has only one purple vertex
(see figure 9.3). Symmetrically, conv(A2) also has one or two purple vertices.
Figure 9.3. Blue, red, and purple faces.

Purple vertices are shown in bold.
Moreover, the total number of purple vertices on both convex hulls must be at
least three.
The edges of conv(A) that are neither edges of conv(Al) nor of conv(A 2 ) must
intersect the separating vertical line Ho, and there are exactly two such edges.
They must connect one purple vertex of conv(Al) to a purple vertex of conv(.A2):
they are the exterior bitangents to polytopes conv(Al) and conv(A 2 ). We call
the upper bitangentthe one that intersects the separating line Ho above the other,
which is called the lower bitangent. Much of the work in the merging process is
to identify these two bitangents.
Let Ak be the vertex of conv(Al) with the greatest abscissa, and Ak+1 be
the vertex of conv(A 2 ) with the smallest abscissa. The segment AkAk+1 lies
outside both conv(Al) and conv(A 2 ). Both vertices Ak and Ak+j are incident
to a red edge. To find the upper bitangent to conv(AI) and conv(A 2 ), the
merging step moves a segment U1 U2 upwards from position AkAk+l, while staying
outside conv(Al) and conv(A 2 ). The left endpoint U1 moves counter-clockwise
on the boundary of conv(Al), taking position at vertices of conv(Ai) that are
incident to a red edge. Likewise, the right endpoint U2 moves clockwise on the
boundary of conv(A 2 ), taking position at vertices of conv(A 2 ) that are incident
to a red edge. More precisely, let succ(U1 ) denote the successor of U1 along
the oriented boundary of conv(A4), and pred(U2 ) the predecessor of U2 along the
204~ ~ Chpe 9. Cove hl in tw an the imnin
Tl
U2
red(U 2 )
conv(Ai)
Figure 9.4. Divide-and-conquer convex hull in dimension 2.
oriented boundary of conv(A2). A move of U1 or U2 is computed by the following

procedure:
while U1 e or U2 H 7suc(Ul)
<i
if U1 e H U)U hen U2 = pred(U 2 )
else U1 = succ(Ul)
In this manner, the endpoints U1 and U2 of segment U(1U2 only traverse red edges
of conv(Al) and conv(A2 ), and both U1 and U2 keep in contact with a red edge.
It remains to show that, when the loop is exited, the line joining U1 and U2
is a line supporting both conv(Al) and conv(A 2 ), and therefore is the desired
upper bitangent. When U1 U2 has reached its final position, the edge U2 pred(U2 )
of conv(A 2 ) is blue with respect to U1 and the edge Ulsucc(Ul) of conv(Ai) is
blue with respect to U2. Without loss of generality, we may assume that the
last move was that of U1, on the boundary of conv(Al) (the proof is entirely
symmetrical in the converse situation). Then the edge pred(Ui)Ul of conv(Al) is
red with respect to U2, and as a result vertex U1 is purple with respect to U2, so
that line U1 U2 supports conv(Al) (lemma 8.3.1). It remains to show that U1 U2
also supports conv(A 2 ), or in other words that vertex U2 of conv(A 2 ) is purple
with respect to U1. The edge U2 pred(U2 ) of conv(A 2 ) is blue with respect to U1 ,
however, so we only have to show that edge succ(U2 )U2 is red with respect to
U1 . Let Ul be the position of U1 on the boundary of conv(A 1 ) during the last
move of the other endpoint on the boundary of conv(A2). The edge succ(U 2 )U2
of conv(A 2 ) is red with respect to Ulf. All the vertices between Ulf and U1 on
conv(Ai) lie on the same side of succ(U2 )U2 . As a result, the edge succ(U 2 )U2 is
also red with respect to U1.
We obtain the lower bitangent in the same fashion, only U1 moves on the
boundary of conv(Al), starting at position Ak and passing clockwise over the
vertices of conv(Al). Likewise, U2 moves on the boundary of conv(A 2 ), starting
at position Ak+j and passing counter-clockwise over the vertices of conv(A 2 ).
To analyze this algorithm, it suffices to notice that each test between a vertex
and an edge simply consists in evaluating the sign of a 3 x 3 determinant, and can
therefore be performed in constant time. Moreover, only two tests are performed
at each step, to follow a red edge of either conv(.Al) or conv(A 2 ), or to discover
that a bitangent has been found. A red edge of conv(Al) or conv(A 2 ) is not
part of the convex hull conv(A), and therefore will never be tested again in the
entire algorithm. The total time needed in the recursion for these operations is
therefore at most proportional to the number of edges of convex hulls created
by the algorithm. At each merging step, two new edges are created, and the
total number of these steps is O(n) if the size of the original set A is n. The
total number of edges created is thus linear, and the complexity of the operations
in the divide-and-conquer recursive calls is O(n). In two dimensions, the total
complexity of the algorithm is therefore dominated by the cost of the initial
sorting. Notice that if the points are sorted along one axis, the algorithm runs
in time O(n).
Theorem 9.2.1 A divide-and-conquer algorithm computes the convex hull of a

set of n points in E2 in optimal time 0(n log n).
9.3 Divide-and-conquer convex hulls

in dimension 3
The divide-and-conquer algorithm described in section 9.2 can be made to work
in dimension 3. Below, we describe this algorithm, which is deterministic and
optimal.
The algorithm in three dimensions follows the same paradigm as its counterpart
in two dimensions. The points are first sorted in order of increasing abscissae,
then the recursive procedure is performed.
Dividing. The set A is split into two almost equal-sized subsets Al and A 2 ,
separated by a vertical line.
Solving. Compute the convex hulls conv(Al) and conv(A 2 ) of each subset
separately, using recursive calls to the procedure.
Merging. The convex hull conv(A) of A can be computed by forming the

convex hull of the two polytopes conv(Al) and conv(A 2 ).
The only delicate step is the merging step, which amounts to computing the
convex hull of two polytopes separated by a plane. The discussion below shows
that this computation can be carried out in time 0(n) if n is the total number
of vertices of both polytopes. The analysis of the algorithm is then immediate:
the initial sorting takes time 0((n log n) and the complexity t(n) of the recursive
calls to the procedure satisfies
t(n) = 2t(n/2) + 0(n),
which solves to t(n) = 0(nlogn). This proves the theorem below. Notice that
in the three-dimensional case, the divide-and-conquer algorithm has complexity
E(n log n) even if the points are sorted along one axis.
Theorem 9.3.1 In three dimensions, a divide-and-conquer algorithm computes

the convex hull of a set of n points in optimal time O(nlogn).
Convex hull of two polytopes separated by a plane
Let C1 and C2 be two polytopes separated by a plane. We want to compute

their convex hull C = conv(Cl U C2 ). To keep things simple, we assume that
the interiors of C1 and C2 are both non-empty, and that their vertices are in
general position. Both assumptions are satisfied by the convex hulls conv(Ai)
and conv(A2) computed by the recursive procedure described above if we assume,
as usual, that the set A of points is in general position. We may also assume
that the two polytopes C1 and C2 are separated by the plane Ho with equation
x = 0, and that the abscissae of the vertices of C1 are negative, while those of C2
are positive.
Overview of the algorithm
As in the incremental algorithm of section 8.3, it helps to color the facets of

polytopes C1 and C2 red or blue, and the faces of dimension d - 2 or less either
red, blue, or purple. Intuitively, the faces of polytope C1 that are red with respect
to C2 are those that would be lit if C2 was an extended light source. The blue
faces of C1 are the ones in the shadow, and the purple faces are lit by tangent
beams. The colors of faces of C2 can similarly be explained by imagining that C1
is an extended light source. Throughout this section, we call a face of C1 (resp.
C2) red, blue, or purple, if it is red, blue, or purple with respect to the opposite
polytope C2 (resp. C1 ). Since the light sources C1 or C2 are polytopes and not
points any longer, the distinction between blue and purple faces is considerably
Figure 9.5. Convex hull of two polytopes separated by a plane.
more subtle than in the incremental algorithm. The colors can be attributed as
follows:
* A face (facet, edge, or vertex) of C1 or C2 that is not a face of C is red.
* A facet of C1 or C2 that is a facet of C is blue.

* An edge of C1 or C2 that is an edge of C is purple if it is incident to at least
one red facet, and blue otherwise.
* A vertex of C1 or C2 that is a vertex of C is purple if it is incident to at least
one red or purple edge, and blue otherwise.
It is very tempting to believe that the purple vertices and edges of C1 (resp.
C2) form a cycle in the incidence graph of C1 (resp. C2 ) as is the case for the set
of edges and vertices that are purple with respect to a point. This is not true,
however. Indeed, a purple vertex can be incident to an arbitrarily high number
of purple edges (for instance, vertex A in figure 9.6a is incident to three purple
edges). This number may even be zero when the purple vertex is the only non-red
face of polytope C1 (consider for instance vertex A in figure 9.6b). A purple edge
may be incident to two red facets: this happens for instance to an edge of C2
whose affine hull is a line that does not intersect C1, but whose incident facets
have their supporting planes intersecting C1 (for instance, edge AB in figure 9.6a).
The faces of C that are neither faces of C1 nor of C2 are the new faces, and they
necessarily intersect the separating plane Ho. A new edge is the convex hull of
a purple vertex of C1 and a purple vertex of C2. A new facet is a triangle, the
convex hull of a purple edge of C1 and of a purple vertex of C2, or conversely the
A A
Cl
C2
(a) conv(Cl UC2 )
A A
Cl
C2
conv(C, UC 2 )
(b)
Figure 9.6. Convex hull of two 3-polytopes. Peculiar cases.
convex hull of a purple edge of C2 and of a purple vertex of C1. Let Co be the
2-polytope formed by the intersection of C with the plane Ho (see figure 9.5).
The new edges of C intersect Ho at the vertices of Co, and the circular order on
the faces of the 2-polytope Co induces a circular order on the new faces (edges
and facets) of C.
The main idea is then to build the new facets of C in turn, in the order given
by the edges of Co. For instance, the plane Ho may be oriented by the x-axis, and
the boundary of the 2-polytope Co will be followed in counter-clockwise order.
As we will show below, the algorithm takes advantage on the order of edges of
a 3-polytope incident to a given vertex. So we choose to represent polytopes by
the edge-list structure, which explicitly encodes this order (see section 9.1).
The overview of the merging algorithm is then:
1. We first find a new edge of C.
2. The algorithm then discovers the other new faces (facets and edges) of C
in the order induced by Co. At the same time, the purple faces (edges and
vertices) of C1 and C2 are found.
3. In a third stage, all the red faces (facets, edges, and vertices) of C1 and C2
are found and the edge-list representation of C is built from those of C1 and
of C2.
Finding the first new edge
A first new edge of C can be found by applying the two-dimensional bitangent-

finding algorithm to the projections C' and C2 of polytopes C1 and C2 on the plane
z = 0. Vertex A1 of greatest abscissa of C1 projects onto vertex A' of greatest
abscissa of C'; starting from this vertex, its successor in the counter-clockwise
order on the boundary of Cl may be found by looking at all the vertices of C1
incident to A 1 . From one vertex to the next, we can build the vertices of C'.
The projection C' of C2 can be obtained in a similar way, starting at A', the
projection of the vertex A2 of C1 that has the least abscissa. Polytopes C' and
C2 are disjoint and separated by the line x = 0 in the plane z = 0. The two-
dimensional merging algorithm, starting with segment A'A' which lies outside
C' and C2, yields a bitangent U{U2 to both polytopes (U{ belonging to C' and U2
to C'). The vertices Ul and U2 of C1 and C2 that project onto U{ and U2 yield
a new edge U1 U2 of C. Moreover, we know that the vertical plane that contains
U1 U2 is a supporting plane of both C1 and C2.
Finding the other new faces
To find the other new faces of C, the algorithm uses the gift-wrapping method
which consists in pivoting a plane around the current new edge, so that it supports
both C1 and C2 , as if we were trying to wrap both C1 and C2 with a single sheet
of paper. More precisely, let A1 A2 , A1 G C1 , A2 E C2, be the new edge that
was most recently discovered. The algorithm knows a plane H1 2 that supports
C along its edge A1 A2 : take for H1 2 the vertical plane passing through A1 A2 if
this edge is the first edge found in the previous stage, or otherwise take the affine
hull of the most recently discovered new facet AA 1 A2 , which is incident to the
oriented edge A1 A2 on its left. At this point, the algorithm must discover the
new facet of C that is right incident to the oriented edge A1 A2 . This facet is a
triangle A1 A2 A' where A' is either a vertex of C1 or a vertex of C2. Consider
a plane H that pivots around edge A1 A2 , starting at position H1 2 and moving
counter-clockwise, meaning that its trace H1 2 n Ho in the separating plane Ho
pivots counter-clockwise around the vertex A1 A2 n Ho of Co. Vertex A' is the first
vertex of either C1 or C2 that is touched by H. We say that the winner of C1 for
the pivot A1 A2 is the first vertex A' of C1 that is touched by H. Similarly, the
winner of C2 for the pivot A1 A2 is the first vertex A' of C2 that is touched by H.
Necessarily, A is one of A' or A', and which one can be decided in the following
A2
C2
C1
Figure 9.7. How to find the winner for the pivot A1 A2 .
way. Let N be the unit vector normal to plane H 12 , and directed outside C.,
Likewise, for each i = 1, 2, let Hi' be the affine plane of the triangle AIA 2A', and
Ni' its unit normal vector directed outside C,. 2 Vertex A' is that A' (i = 1, 2) for
which the dihedral angle between H12 and Hi' is minimal, or equivalently that for
which the dot product N *Ni is minimal.
We must now explain how to find the winners A' and A'. These problems
being exactly symmetrical, we will restrict our attention the problem of finding
the winner of Cl.
If A' is a vertex of C, that is adjacent to Al, then we denote by pred(A')
and succ(A') the vertices of C, adjacent to A' that respectively precede and
follow Al in the counter-clockwise order around Al. Denote by HI the planar
affine hull of the triangle A1A'A 2 and H',+the half-space bounded by this plane
that is opposite to the wedge product -AlA2 A AA', which induces a clockwise
orientation of the triangle A1A 2 A1.
Lemma 9.3.2 The winner of C, for the pivot AlA 2 is the unique vertex A' of
C, adjacent to Al such that pred(A') and succ(A') both belong to the half-space
11 1
Proof. A' is the winner of pivot AlA2 if and only if triangle A'AlA2 is the face
of polytope conv(Cl U {A 2 }) incident to the oriented edge AlA 2 on its right. Then
HI is a supporting plane of conv(Cl U {A 2 }) and therefore of C,, so that AIA' is
an edge of C,, and both pred(A') and succ(A') belong to half-space H'+.
'The direction of vector N is the one that orients the triangle (Al, A2, A) counter-clockwise.
It is the direction of the wedge product AI A2 AA, A.
2 0n the other hand, the direction of vector Ni is the one that orients the triangle (Al, A , A%)
2
clockwise. It is the direction of the wedge product -A1 A2 AAAi.
Reciprocally, let A' be a vertex of C1 adjacent to Al such that both pred(AI)

and succ(AI) belong to half-space H'+. Then polytope C,, which is contained
in the intersection of two half-spaces bounded by the planes supporting C1 along
its facets AIA'succ(AI) and AA'pred(A'), is therefore contained in H'+. As
a result, Hl supports C1 and A'AIA 2 is the facet of polytope conv(Cl U {A 2 })
incident to the oriented edge A1 A2 on its right, which shows that A' is the winner
of the pivot A 1 A2 .
During the algorithm, we may occasionally encounter several pivots incident to
the same vertex Al of C1 . The following lemma shows that it is not necessary to
test each of the vertices of C1 adjacent to Al, for each of these pivots.
Lemma 9.3.3 When a vertex Al of Cl is incident to several pivots, the algorithm

encounters these pivots in such an order that their winners are ordered clockwise
around this vertex Al of C1 . Likewise, the pivots incident to a vertex A2 of C2
are encountered in an order such that their winners are ordered counter-clockwise
around this vertex A2 of C2 .
Proof. Here we prove the assertion concerning polytope C1. A proof for polytope
C2 is entirely symmetrical. Since A1 is a vertex of the convex hull C, there is a
plane H1 that separates vertex Al from all the other vertices of Ci and C2 (see
exercise 7.4). Such a plane intersects all the edges of C and C1 incident to Al
(see figure 9.8). Plane H1 intersects polytope C along the 2-polytope H1 n C and
polytope C1 along the 2-polytope H1 n C1 contained inside H1 n C.
Let us orient the plane H1 by a normal unit vector N1 directed towards the
half-space that does not contain A1 . The counter-clockwise order on the edges of
the 2-polytope H1 n C1 corresponds to the indirect order on the facets of C1 that
contain vertex A1 .
The order in which the new faces of C that contain Al are discovered is consis-
tent with the counter-clockwise order of the faces of the 2-polytope H1 nC. To see
this, it suffices to consider a plane H that pivots around HofnH 1 from Ho towards
H1 . As H pivots from Ho to H1 , the 2-polytope H n C changes. Nevertheless,
each new face of C that contains A1 always keeps a trace in H corresponding to a
face of H n C, and all these traces remain in the same order along the boundary
of H n C.
Any purple edge of C1 is also an edge of C, and the trace on plane H1 of a purple
edge of C that is incident to A1 is a common vertex of both polytopes H1 n C
and H1 n C1 . The trace on H1 of a pivot A1 A2 , however, is a vertex of H1 n C
but not of H1 n C1 . The trace on H1 of the plane that pivots around the edge
A1 A2 is a line L that pivots around the point A1 A2 n H1 . The pivoting plane
touches C1 at the winner of A' of C1 for the pivot A1 A2 whenever L becomes
a supporting line of the polytope H1 n C1 . The point along which L supports
Al
C1
Hi nAlA
Hi
Figure 9.8. Successive pivots incident to a vertex Al of Cl.
H1 n C1 is the trace on H1 of the edge A 1 A'. During the course of the algorithm,
the trace on H1 of the successive pivots incident to A1 moves counter-clockwise
on the boundary of H1 n C; as a result, the point at which the line L touches
H1 nC1 moves counter-clockwise on the boundary of H1 nC1 (see figure 9.8). The
edge AIA' therefore traverses the list of edges of C incident to A1 in indirect (or
clockwise) order. F
In order to find the winner in C1 for a pivot incident to a vertex Al, the algo-
rithm need only consider the edges of C1 incident to vertex A1 in clockwise order.
When the algorithm considers the first pivot incident to Al, the algorithm starts
searching at any edge incident to Al. If the algorithm has already encountered
one or more pivots incident to Al, however, then it starts the search at the winner
of C1 for the last encountered pivot incident to Al.
When the algorithm discovers a new face (A1 A2 A' or A1 A2 A') while pivoting
around edge A1 A2 , it also exhibits a new edge of C (A2 A' or A1 A'). The pivoting
process may be started again around this new pivot. Moreover, a purple vertex
A' (or A') and a purple edge A1 A' (or A 2 A) are also identified. The pivoting
process is terminated when the pivot is back at the initial edge U1 U2 . All the
new facets have been discovered, and the purple faces of C1 and C2 (vertices and
edges) have been sorted out. Note that a purple vertex can be enumerated several
times in the algorithm, and that some purple edges (those incident to two red
facets) may be enumerated twice.
Reconstruction of C
The third stage of the merging process must identify all the red faces of C1 and
C2 , and also build the edge-list representation of the convex hull C of C1 and
C2. For this, we may begin by traversing the list of purple edges and color the
incident facets: a facet of Ci (i = 1, 2) incident to a purple oriented edge AiA'
is red if there is a new facet of C incident to AiA' on the same side. It is blue
otherwise. The red facets are then colored by propagating the red color, as every
facet incident to a red facet and that is not colored blue must be a red facet as
well. The red edges and vertices are then easily determined: an edge incident to
a red facet must be red, unless it was colored purple in the previous stage (finding
the purple edges and vertices); likewise, a vertex incident to a red or purple edge
is red, unless it was colored purple in the previous stage.
When the new faces of C are discovered and the red and purple faces of poly-
topes C1 and C2 have been determined, it is easy to create the edge-list represen-
tation of C from those of C1 and C2.

During the initial stage, each edge of C1 or C2 is examined at most twice, once
for each endpoint, in order to build the projections C' and C2. In dimension 2,
the complexity of the algorithm that finds the tangent of two convex polytopes is
linear (see section 9.2), and hence the first new edge is found in time O(ni + n 2 ),
if n1 and n2 stand for the number of vertices of C1 and C2 respectively.
The complexity of the second stage is clearly proportional to the number of
edges of C1 and C2 that are considered when searching the winners of each pivot.
From lemma 9.3.3, the list of edges incident to a purple vertex is traversed at
most twice during the entire algorithm, once to find the winner of the first pivot
incident to this vertex, and another time to find the winners of all the other pivots
incident to this vertex. The complexity of this stage is therefore O(ni + n2) as
well.
Finally, finding the red faces of C1 and C2 in the third stage can be carried out
in time proportional to the number of these faces and of new faces of C, which is
again 0(ni + n2). We have proved:
Theorem 9.3.4 Let C1 and C2 be two polytopes in E3 separated by a plane, and

having respectively ni and n2 vertices. It is possible to compute the convex hull
C = conv(Ci U C2 ) in time O(n1 + n2).
9.4 Convex hull of a polygonal line

The reader will have noticed that, in dimension 2, the complexity of the incremen-
tal algorithm and that of the divide-and-conquer algorithm are both dominated
by the cost of the initial sorting. In fact, the costs of both algorithms are only
linear if the points are already sorted along some direction. Now a set of points
that is sorted along, say, the x-axis, forms a simple polygonal line, meaning that
it does not intersect itself. It is therefore tempting to conjecture that the convex
hull of any simple polygonal line in E2 can be obtained in linear time. It is the
goal of this section to show that this is indeed the case.
Simple polygonal lines and polygons are defined in a general way in chapter 12.
For this reason, here we only give a brief summary of definitions and results which
we will use in the forthcoming algorithm. A polygonal line is an ordered sequence
of points, called the vertices of the polygonal line. The segments that join two
consecutive vertices are called the edges of the polygonal line. The polygonal line
is simple if:
1. its vertices are all distinct, except perhaps the first and last which may be
identical, and
2. two edges do not intersect, except perhaps at a common endpoint.
A polygonal line is closed if the first and last vertices are identical. A simple
and closed polygonal line is also called a polygon. Thus a polygon may be defined
entirely by its circular sequence of vertices. A deep theorem of Jordan (a proof of
the theorem for polygons is given in exercise 11.1) states that any simple closed
curve separates the plane E2 into two connected components, exactly one of which
is bounded. Thus a polygon P separates the points in E2 \ p into two connected
regions, only one of which is bounded. This bounded region is called the interior
of polygon P, and denoted by int(P). The other (unbounded) region is called
the exterior of the polygon and is denoted by ext(P). Regions int(P) and ext(P)
are topological open subsets 3 of E2 , and the topological closure int(P) of int(P)
is the union int(P) U P. In this section, the Euclidean space E2 is oriented and
a polygon will be described as a circular list of vertices in direct (or counter-
clockwise) order. This defines an orientation on the edges. By convention, we
agree that the direct orientation of a polygon is such that the interior of the
polygon is to the left of each oriented edge.
Let A be a set of n points in E2 . One may wonder why knowing a simple
polygonal line £(A) joining these points would help in computing the convex
hull of conv(A). The following theorem shows a deep connection between the
3
For a brief survey of the topological notions of open, closed, and connected subsets, see
chapter 11.
9.4. Convex hull of a polygonal line 215
order of the vertices along the boundary of the convex hull conv(A) and the
order of these points along the polygonal line L(A).
Theorem 9.4.1 Consider two polygons P and Q, such that the interior of Q
is entirely contained inside the interior of P. The common vertices of P and Q
are encountered in the same order when both polygons are traversed in a counter-
clockwise order.
Let Ep and EQ and be the circular sequences of vertices of P and Q. Let E'
be the subsequence of Ep that corresponds to vertices common to both P and
Q, and similarly let X<? be the subsequence of EQ that corresponds to vertices
common to both P and Q. The theorem states that the two sequences E' and
E' are identical. In particular, if P is the convex hull of Q, then the theorem
states that the sequence of vertices of conv(Q) is a subsequence of the sequence
of vertices of Q.
Proof. Because the interior of Q is entirely contained inside the interior of P,
edges of P and Q cannot intersect in a point which is not a vertex of P or Q. In
the following, we first assume that the edges of P and Q intersect only in points
which are vertices of both P and Q, and then remove this assumption at the end
of the proof.
A chord of a topological 2-ball is a simple curve contained in the interior of
the 2-ball except for its endpoints which lie on the boundary. The proof of the
theorem relies on a consequence of Jordan's theorem that states that any closed
topological 2-ball is separated by a chord into two connected components. (A
proof of this consequence is given in exercise 11.2.) Both subsequences V and
E' have the same vertices, namely those common to both P and Q, but not
necessarily in the same order. To prove that they are in fact identical, it suffices
to show that two consecutive vertices in E' are also consecutive in E,. Let then
A1 and A2 be two consecutive vertices of E'. Let Q1 2 be the portion of Q that
joins Al to A2 in counter-clockwise order, and Q21 be the portion that joins A2 to
A1 in counter-clockwise order. Likewise, let P1 2 be the portion of P that joins Al
to A2 in counter-clockwise order, and P21 be the portion that joins A2 to Al in
counter-clockwise order (see figure 9.9). The vertices of P are distinct from those
of Q12 (except for the endpoints Al and A2 ) because Al and A2 are consecutive
in the subsequence EQ. Hence P1 2 and Q 1 2 cannot share a common vertex except
for Al and A2 . Furthermore, there are no intersections between edges of P and
Q except at a common vertex. Hence Q12 is a chord of the topological 2-ball
int(P). Moreover, Q21 is contained in int(P) and does not intersect the chord
Q12 except at its endpoints Al and A2 . Hence Q21 is entirely contained in one of
the two connected components of int(P) \ Q12- Clearly, P12 is entirely contained
in the other component except for its endpoints Al and A2 . Therefore, P1 2 and
Q21 cannot share a common vertex except for Al and A2 . This shows that Al
As
A2
11
*1
and A2 are consecutive in the subsequence E,, and proves the theorem when P
and Q intersect only at common vertices.
The case where some vertices of P can lie on edges of Q is handled by adding
those vertices of P as vertices of Q, splitting the corresponding edges of Q.
Similarly, when some vertices of Q lie on edges of P, we add those vertices of Q
as vertices of P and split the corresponding edges of P. The same vertices are
added to the subsequences Ep and EZ', producing two subsequences E" and E"
which are identical by the proof above. Thus E' and EQ, obtained by removing
the same elements, are also identical. El
The following corollary takes more interest in the convex hull of polygonal lines.
Let A be a set of n points in the plane and L(A) = (Al, A2 , ... , A.) be a simple
polygonal line joining the points in A. Consider the ranks in L(A) of the vertices
on the convex hull conv(A), and denote by Am and AM the vertices of conv(A)
with respectively lowest and highest rank (see figure 9.10).
Corollary 9.4.2 When the boundary of conv(A) is oriented counter-clockwise,

the vertices encountered between Am and AM form a subsequence of vertices of
L(A), whereas the vertices encountered between AM and Am form a subsequence
of the polygonal line that is the reverse of L(A).
Proof. We denote by C(A) the polygon that constitutes the boundary of the
convex hull conv(A), by CmM the portion of C(A) that joins Am to AM in counter-
clockwise order, and by CMm the portion of C(A) that joins AM to Am in counter-
clockwise order. To prove the first part of this corollary, it suffices to apply the
previous theorem to the polygons P and Q defined as follows. Polygon P is the
concatenation of CmM and a polygonal line PMm that joins AM to Am such that
L(A) and C(A) are both contained within int(P) (see figure 9.10). Polygon Q
is the concatenation of LmM (the portion of £(A) that joins Am to AM) and of
PMn. The second part of the theorem can be proved very similarly. El
9.4. Convex hull of a polygonal line 217
Figure 9.10. Convex hull of a polygonal line.
The remainder of this section presents an algorithm that builds the convex hull
of a polygonal line in E 2 in linear time.
Let A be a set of n points in E 2 and £C(A) a simple polygonal line whose vertices
are the points of A. The algorithm we present here is an incremental algorithm
that processes the points of A in the order of £(A).
Let (A 1 ,A 2 ,...,A.) be the sequence £(A) and Ai {A 1 ,A 2 ,...,Ai} be the
set of the first i points of C(A). The convex hull conv(Ai) of Ai is maintained as
a doubly connected circular list of vertices. The algorithm maintains a pointer
to the vertex of conv(Ai) with the highest rank in L(A).
The initial step builds a circular list for the triangle A1 A2 A3 and the pointer
points to A3 . The current step inserts point Ai in the structure, and updates
the data structure that stores conv(A-i) so that it represents conv(A2 ). The
algorithm works in two phases.
First phase. The algorithm determines whether point Ai belongs to the inte-
rior or exterior of conv(Ai-1). Lemma 9.4.3 below shows that this reduces to eval-
uating the signs of the two determinants [pred(AM)AMAi] and [AM SUCC(AM)Ai],
where AM has the highest rank among the vertices of conv(Ai- 1 ), and pred(AM)
and succ(AM) are respectively the predecessor and successor of AM in a counter-
clockwise enumeration of the vertices on the boundary of conv(A4i-) (see fig-
ure 9.11).
Lemma 9.4.3 Let H+ (resp. H+) be the half-space bounded by the line sup-
porting conv(A4i-) along the edge pred(AM)AM (resp. AMsucc(AM)), and that
contains conv(Ai-1). Point Ai is interior to polytope conv(Ai-1) if and only if
Ai belongs to the intersection of half-spaces H+ n H+.
Proof. The condition is obviously required. To show that it also suffices, we
1)
p Hp
H5
Figure 9.11. For the proof of lemma 9.4.3.
see that the simple polygonal line C(A) connects the vertices pred(AM) and
succ(AM) of conv(A2i-), and that the portion of L:(A) that joins pred(AM) to
succ(AM) together with the edges pred(AM)AM and AMSUCC(AM) of conv(Ai-1)
form a simple closed polygonal line that bounds a region D of E2 entirely con-
tained in conv(Ai-i). This region is shaded in figure 9.11. The simple polygonal
line L(A) also connects vertex AM to Ai, and the portion of L(A) that joins AM
to Ai cannot intersect the portion of L(A) that joins pred(AM) to succ(AM).
This guarantees that if point Ai belongs to both half-spaces H+ and H+,then it
also belongs to D and thus to conv(Ai-1).
Second phase. The algorithm now updates the convex hull if point Ai does
not belong to the interior of polytope conv(Ai-I). In this case, the previous
lemma shows that at least one of the edges pred(AM)AM and AMsucc(AM) is
red with respect to Ai, if we use the terminology of the incremental algorithm of
chapter 8. To update the convex hull, the algorithm need only perform steps 2,
3, and 4 of the incremental algorithm described in section 8.3.
Theorem 9.4.4 The algorithm described previously builds the convex hull of a
simple polygonal line in E2 in linear time.
Proof. For each vertex of the polygonal line L(A), phase 1 requires constant
time, since it only involves the computation of the signs of two 2 x 2 determi-
nants. As to the second phase, if it is performed, its complexity is shown in
section 8.3 to require time that is proportional to the number of the edges of
polytope conv(Avi-) that are red with respect to Ai. The total contribution of
this phase to the complexity of the algorithm is thus proportional to the number
of edges created by the algorithm, which is O(n). El
9.5. Exercises 219
9.5 Exercises
Exercise 9.1 (Common bitangents) Let Ci and C2 be two 2-polytopes, separated by
a vertical line A. Let C = conv(Cl U C2). Show that the edges of C intersecting A may
be found in time O(log n), where n is the total number of vertices of C1 and C2.
Hint: For each polytope C, let us denote by C+ the upper hull of C, which is the convex
polygonal line whose vertices are vertices of C and that joins the vertex of highest abscissa
to the vertex of lowest abscissa of C in the counter-clockwise order of the vertices along
the boundary of C (see exercise 7.14). Similarly, we define the lower hull C-. Note
that the boundary of C is a concatenation of C+ and C- and that any vertical line that
intersects C also intersects an edge of C+ and an edge of C-. If C is the convex hull
conv(Cl U C2) of two polytopes C1 and C2 separated by a vertical line A, we call bridges
the two edges of C intersected by A. We separately search for the upper bridge, which is
the edge of C+ intersected by A, and the lower bridge, which is the edge of C- intersected
by A. The upper bridge is an edge of C that joins a vertex of Cj+ to a vertex of C+. It is
possible to find it by a binary search on each of Cl+ and C,+. Indeed, consider a vertex U1
of C+ and a vertex U2 of C+, and look at the color with respect to U1 of the edges of C2+
incident to U2, and the color with respect to U2 of the edges of C+ incident to U1. There
are nine possible cases, and in each case, at least one of the four chains determined by
U1 and U2 on Cj+ and C2+ can be discarded from further consideration.
Exercise 9.2 (Dynamic convex hulls) We present an algorithm for maintaining the
convex hull of a set of points in the plane under insertion and deletion. In fact, the
algorithm maintains the upper and lower hulls separately (see exercise 9.1).
The data structure used to represent the upper hull is a balanced binary tree whose
leaves correspond to the points in the set ordered by increasing abscissae. At each internal
node N in the tree, we store a secondary data structure which allows us to efficiently
restore the convex hull of the points stored in the subtree rooted at N. More precisely,
this structure is a catenable queue which maintains a list under the following operations:
insertion and deletion of list items at either end of the list, splitting the list at a given
item, and concatenation of two lists. Each of these operations may be performed in time
logarithmic in the size of the lists. Let N be a node of the primary tree. We denote by
conv+(N) the upper hull of the points stored in the subtree rooted at N. The catenable
queue stored at N contains the portion of conv+(N) that is not on the boundary of the
convex hull conv+(M), where M is the parent of N in the tree. Moreover the position
of the first vertex of conv+(M) that does not belong to conv+ (N) is stored at node N
in an integer j(N). The catenable queue stored at the root maintains the upper hull of
all the points stored in the tree.
1. Show that the data structure requires a storage O(n) if n is the number of points
stored in the structure.
2. Show that each insertion or deletion takes time O(log 2 n) when the structure stores
n points.
Hint: Let us consider, for instance, the insertion of a point P into the structure that
maintains the upper hull. We follow the path in the tree that leads, from the root, to the
leaf that is the closest to P (in the x-order). At each node N on this path, we update
the catenable queues of N and of its sibling N' so that they store the upper convex hulls
conv+ (N) and conv+(N'), which may be done in time O(log n). Indeed, it suffices to split
the catenable queue of the parent M of N and N' at position j(N), and to concatenate
the sublists with those stored at N and N'. A node is created for P. Then, while going
the other way on this path, the primary structure is re-balanced and the correct chains
can be computed for each node. Let N be a node on the reverse path, and N' be its
sibling. We can compute the upper hull conv+ (M) of their parent M because the upper
hulls conv+ (N) and conv+ (N') are known, and the bridge joining them can be computed
in time O(logn) (see exercise 9.1). The lists conv+(N) and conv+ (N') may thus be split
at the bridge and we keep only the portion on the right for the left sibling and on the left
for the right sibling. All these operations may be carried out in time O(logn) for each
node on the reverse path from the new node P to the root of the primary tree. During a
rotation (simple or double) of the tree, only a constant number of nodes switch children
and a similar operation restores the correct chains stored in these nodes, in time O(log n)
for each node.
Exercise 9.3 (Onion peeling) Given a set S of n points in the plane, we consider the
subsets
SO = S,
Si = So \ {set of vertices of conv(S)}
Si= Si- \ {set of vertices of conv(Si)}
until Sk has at most three elements. Give an algorithm that computes the iterated
convex hulls conv(So), conv(S 1), . . . , conv(Sk),... in total time O(n log n).
Exercise 9.4 (Diameter, antipodal pairs) Let P be a set of points of Ed. The di-
ameter of S, denoted by d(S), is the maximal distance between two points of S. A pair
(Pi, Pj) of vertices of the convex hull conv(S) is said to be antipodal if conv(S) admits
two parallel hyperplanes supporting P along Pi and Pj respectively.
1. Show that if the diameter d(S) occurs for Pi and P,, that is d(Pi, Pj) = d(S), then
Pi and Pj are vertices of the convex hull conv(S), and (Pi, Pj) is an antipodal pair.
2. Derive an algorithm in E2 that enumerates all the antipodal pairs to find the
diameter in time O(n log n).
Exercise 9.5 (Hierarchical representation of a 3-polytope) Let P be a 3-poly-

tope with n vertices. The degree of a vertex is the number of edges incident to that
vertex. Two vertices of P are adjacent if they are incident to a common edge.
1. Show that P has at least n/3 vertices whose degree is less than 9. (Use Euler's
relation to count the number of adjacencies between edges and vertices of P.)
2. Show that any maximal subset of non-adjacent vertices of P with degree at most 8
has at least n/27 vertices.
Denote by V(P) the set of vertices of P. A hierarchical representation of P is a nested
sequence of polytopes (Po D P1 D ... D Pk) such that
* P0 -P7
* Pk is a simplex,
* for any 0 < i < k, V(Pi+l) c V(Pi) and V(Pi) \ V(Pi+1 ) is a maximal subset of
non-adjacent vertices of Pi that have degree at most 8 in Pi.
3. Show that k = O(log n) and that a hierarchical representation of P may be built in
time O(n).
Exercise 9.6 (Querying 3-polytopes) Let P be a 3-polytope with n vertices. Using

the hierarchical representation of P described in exercise 9.5, show that the following
queries can be answered in time 0(log n):
1. Computing the intersection with a line: compute the intersection of P with a line
L, or if this intersection L n P is empty, find the point P of P that minimizes the
distance to L.
2. Detecting the intersection with a plane: if the intersection of P with a plane H
is not empty, then output a point on the boundary of P n H, otherwise return a
point P of P that minimizes the distance to H.
3. Finding a plane of support: given a line L, and a plane H containing L that does
not intersect P, compute the first vertex of P encountered as H pivots around L.
4. Ray shooting: given a point 0 inside the polytope, and a direction U, find the face
of P intersected by the ray originating at 0 along the direction U.
Hint: In the first three cases, the algorithm can take advantage of the solution for Pi+,
to compute the solution for Pi, in constant time. Note that the fourth query type is dual
to the third.
Exercise 9.7 (Intersection of two convex polygons) Give an algorithm that com-
putes the intersection of two convex polygons in the Euclidean plane E2, in linear time
O(n + m) if m and n are the respective number of vertices of either polygon.
Exercise 9.8 (Union of tricolored triangles) Consider a set of n points in E2 and

the set of triangles having these points as vertices. Show that if each point is colored
with one of k given colors, it is possible to compute the union of all tricolored triangles
(no two vertices have the same color) in time O(nlogn).

There are several algorithms that compute the convex hull of a set of points, especially in
dimension 2. It is commonly accepted that the first optimal algorithm in dimension 2 was
given by Graham [111] in 1972. This algorithm sorts the points by polar angle around
a point inside the convex hull (for instance any convex combination of three original
points), then determines the edges of the convex hull by scanning this list. In fact,
Toussaint [215] shed new light on the history of the problem by pointing out a paper by
Bass and Schubert [21], published in 1967, that proposed an optimal algorithm for the
convex hull in dimension 2, although the paper had a slight error and gave no complexity
analysis. Another well known algorithm is Jarvis' "gift-wrapping" algorithm [130] which
computes the convex hull by successively finding all supporting hyperplanes. These
algorithms and others are detailed in the book by Preparata and Shamos [192], which
also gives a solution to exercise 9.4.
If the convex hull of n points in dimension d may have Q(nLd/2J) faces in the worst
case, the number of faces may be much less. It is thus important to have output-sensitive
algorithms, meaning that their complexities depend on the size of the output. Jarvis'
algorithm is output-sensitive, as it runs in time O(nh) for a set of n points whose convex
hull has h vertices. Kirkpatrick and Seidel [137] gave an algorithm in dimension 2 that
runs in time 0 (n log h). This algorithm uses a curious variant of the divide-and-conquer
algorithm, which may appropriately be called marriage-before-conquest,since the upper
and lower bridges connecting the convex hull of two subsets are computed before the
convex hull of either subset is known. Edelsbrunner and Shi [99] generalized the idea to
yield a convex hull algorithm in E 3 that runs in O(n log2 h).
The on-line algorithm presented in section 9.4 is due to Melkman [169]. The algo-
rithms by Lee [146] and Graham and Yao [112] are both correct and use only one stack.
The problem of dynamically maintaining the convex hull of a planar set of points (see
exercise 9.2) was solved by Overmars and Van Leeuwen [184] in time O(log 2 n) for each
operation. If only insertions or only deletions are to be performed, time O(logn) for
each operation may be achieved. The case of insertions was studied by Preparata [190]
and that of deletions by Chazelle [43] and Hershberger and Suri [125]. Chazelle's al-
gorithm [43] computes the onion peeling described in exercise 9.3. Hershberger and
Suri [126] also showed that a sequence of n insertions and deletions can be performed
in amortized time O(log n) for each operation if the sequence of operations is known in
advance. Moreover, their data structure can also handle queries (tangent line passing
through a given point, intersection with a line, finding a vertex, and more generally
any query that can be handled with binary search) in time 0(logn), and this after any
number of operations are performed.
The divide-and-conquer algorithm in dimension 3 was first proposed by Preparata and
Hong in 1977 [191], but the first entirely correct description of the algorithm was given
in 1987 by Edelsbrunner [89].
The hierarchical decomposition of 3-polytopes (see exercise 9.5) was invented by Dobkin
and Kirkpatrick [85] who used it to compute the distance between two polyhedra. Edels-
brunner and Maurer [94] also used it to answer different types of queries on 3-polytopes
(see exercise 9.6). Chazelle [45] also used the same kind of decomposition to compute
the intersection of two 3-polytopes in linear time.
The algorithm in exercise 9.8 that computes the union of tricolored triangles is due to
Boissonnat, Devillers, and Preparata [27].
Chapter 10
Linear programming
Several problems, geometric or of other kinds, use the notion of a polytope in

d-dimensional space more or less implicitly. The preceding chapters show how to
efficiently build the incidence graph which encodes the whole facial structure of
a polytope given as the convex hull of a set of points. Using duality, the same
algorithms allow one to build the incidence graph of a polytope defined as the
intersection of a finite number of half-spaces. It is not always necessary, however,
to explicitly enumerate all the faces of the polytope that underlies a problem.
This is the case in linear programming problems, which are the topic of this
chapter.
Section 10.1 defines what a linear programming problem is, and sets up the
terminology commonly used in optimization. Section 10.2 gives a truly simple
algorithm that solves this class of problem. Finally, section 10.3 shows how linear
programming may be used as an auxiliary for other geometric problems. A linear
programming problem may be seen as a shortcut to avoid computing the whole
facial structure of some convex hull. Paradoxically, the application we give here
is an algorithm that computes the convex hull of n points in dimension d. Besides
its simplicity, the interest of the algorithm is mostly that its complexity depends
on the output size as well as on the input size. Here, the output size is the number
f of faces of all dimensions of the convex hull, and thus ranges widely from 0(1)
(size of a simplex) to E)(nLd/2]) (size of a maximal polytope). It is therefore often
useful to use an algorithm whose running time depends on the number of faces
f effectively computed, rather than an algorithm whose complexity can only be
bounded by the worst-case complexity of a polytope. The algorithm presented in
section 10.3 runs in time O(n 2 + f log n) on a set of n points whose convex hull
has f faces, which is competitive in dimensions d higher than 5. This algorithm
relies on the existence of a shelling order of the polytope, which is defined and
proved to exist in section 10.3.
224 Chapter 10. Linear programming
10.1 Definitions
A linear programming problem consists of optimizing a linear function of d vari-
ables, where the variables must satisfy a given set of n linear constraints. The
linear function to be minimized may be written as a dot product:
f(X)=V X,
where V is a given vector of Ed and X a variable vector in the same space. The
linear constraints that X must satisfy may be written as
Ai-X <ai, 1< i =1,...,n,
where, for each i = 1, . .. , n, Ai is a vector in Ed and ai a real constant. Geo-
metrically, each constraint can be expressed by the fact that the point X lies in
a closed half-space. Denote by Hi+ the closed half-space that corresponds to the
i-th constraint:
Hi={XEEd : AiX<aai}.
The intersection f 1 Hi+ is called the feasible domain of the problem. If the
feasible domain is bounded and not empty, then it is a polytope P and the
solution to the linear programming problem is a vertex of 7P, or occasionally the
set of points on a higher-dimensional face of P. This face F of P is characterized
as follows: the set of supporting hyperplanes of P along F includes a hyperplane
normal to the direction V, and P is contained in the half-space bounded by
this supporting hyperplane that contains the vector V. If the feasible domain is
empty, the linear programming problem is termed unfeasible. On the contrary,
if the feasible domain is infinite in the direction -V, the linear programming
problem is termed unbounded. To have a bounded problem, one may always
restrict the feasible domain to a large box by adding 2d constraints. The size of
this box can be chosen in such a way that if the problem is bounded, the solution
is guaranteed to lie in the box (see exercise 10.1).
For example, the following location problem about convex hulls is a linear pro-
gramming problem in disguise: given n points {P,, P2 , .. ., P.} in Ed, and a query
point Q, the problem asks if Q belongs to the convex hull conv({P,, P2 , ... , ,}).
If an origin 0 is chosen inside this convex hull (for instance as the centroid of d+ 1
of the points), the polarity of center 0 (see section 7.1.3) lets the dual version of
the problem be expressed as follows: given n half-spaces {Pi*+ : i = 1,..., nI
whose intersection P# is bounded, and a query hyperplane Q*, determine whether
Q* intersects the polytope 'P#. The half-space Pi*+ is the half-space bounded by
the hyperplane Pi* polar to Pi and that contains 0, and Q* is the polar hyper-
plane of Q. Point Q lies inside P if and only Q* avoids the dual polytope P#
(see section 7.1.3). The equation of Q* is
Q *X = 1,
10.2. Randomized linear programming 225
and to locate point Q inside or outside P reduces to solving the following two
problems:
1. Minimize the function fi(X) = Q X with the constraints that X e P*+

for i = 1,..., n.
2. Minimize the function f 2 (X) -Q *X with the same constraints.
Point Q is inside polytope P if and only if 1 < ul or 1 > -u 2 where ul and u2
are the respective minima of fi (X) and f2 (X).
10.2 Randomized linear programming

Several algorithms have been devised to solve linear programming problems. The
deterministic ones with best asymptotic complexity perform the minimization in
time linear in the number n of constraints, but exponential in the number d of
variables. Some randomized algorithms have sub-exponential complexity in n
and d. The algorithm described here is a randomized algorithm that runs in
expected time O(d4 d!n). Its main advantage is its simplicity.
The algorithm
This algorithm is an incremental algorithm that adds the constraints one by one
while maintaining the solution to the current linear programming problem. For
the randomization, we assume that the constraints are inserted in random order.
To make the description simpler, assume for now that all linear programming
problems at any incremental step are feasible, bounded, and have a unique solu-
tion. These assumptions will be relaxed afterwards.
Here is the algorithm: In an initial step, compute the optimal vertex of the
polytope Pd, given as the intersection of the half-spaces that correspond to the
first d constraints:
d
Pd=nHj+
j=1
Subsequently, the constraints Hi are added one by one. The algorithm com-
putes the optimal vertex Xi of the polytope Pi = ni=l Ht knowing the optimal
vertex Xi-, of the polytope Pi-,. It proceeds as follows. If vertex Xi- 1 belongs
to the half-space Hi+, then Xi = Xi-,, and nothing else is done for this step.
Otherwise, vertex Xi necessarily belongs to hyperplane Hi, so we know one of
the hyperplanes incident to the optimal vertex. To find the other hyperplanes,
the algorithm recursively solves a (d - 1)-dimensional linear programming prob-
lem with d - 1 variables and i - 1 constraints. The optimizing function for this
problem is f'(X) = V X, where V' is the projection of V onto the hyper-

plane Hi, and the constraints are the half-spaces H'+ = H+ n Hi in Hi, for each
j = 1, .. , i - 1. The optimal vertex is therefore obtained by recursively solving
problems of smaller dimension. The recursion stops when d = 1, since the feasible
domain is an interval which may be computed easily in time linear in the number
of constraints.
Let us now indicate how to relax the assumptions stated at the beginning of
our description of the algorithm.
First of all, one of the sub-problems Pi may be unfeasible. But this will be
detected when adding the i-th constraint Hi, since at the bottom of the recursion
the feasible interval will be found empty. If so, the entire problem is also unfeasible
and we may stop the algorithm without even adding the remaining constraints.
We also assumed that the linear programming problems under consideration are
bounded. To guarantee this, we add constraints that bound a hyper-rectangular
domain To known to contain the solution if the problem is bounded (see exer-
cise 10.1). This limits the search space to the hyper-rectangle. In the initial step,
the algorithm computes the optimal vertex of the hyper-rectangle 'o, and the
algorithm continues with the constraints Hi+, i = 1, ... , n. If the optimal vertex
is found on the boundary of Po, then the problem is unbounded.
Finally, the algorithm demands that the solution to each sub-problem Pi be
unique, so that this solution does not depend on the order in which the constraints
are inserted. To enforce this, when the optimal solution is a face of Pi of dimension
k > 0, we let the algorithm systematically choose the vertex of this face which
has the highest lexicographic order: if the xd-coordinates of several vertices are
equal, we compare their xd-l-coordinates, etc., until the optimal vertex is found.
The analysis of the algorithm is also particularly simple. The order in which the
constraints are added is assumed to be random, so we evaluate the performances
of the algorithm on the average over all possible permutations. We prove that the
expected running time of the algorithm is 0(d 4 d!n) when the linear programming
problem has d variables and n constraints. More precisely, let t(d, i) be the
expected time to insert the i-th constraint into a linear programming problem
with d variables and i constraints. We prove by induction on d and i that t(d, i) is
O(d4 d!). Indeed, if d equals 1 each insertion is processed in constant time. When
d > 1, inserting the i-th constraint reduces to testing if Xi-1 belongs to the half-
space H+, except when Xi differs from Xi-1, in which case a linear programming
problem with d -1 variables and i - 1 constraints has to be solved. The latter
case occurs only when the i-th constraint is bounded by one of the d constraint
hyperplanes intersecting in Xi. Knowing the first i constraints, this happens
10.3. Convex hulls using a shelling 227
with conditional probability d/i. Furthermore, since this conditional probability

does not depend upon the subset formed by the first i constraints, d/i is simply
the probability that Xi differs from Xi-,. Testing whether Xi-1 belongs to the
half-space Hi entails computing the sign of a d x d determinant. The cost of this
operation, which we denote by f (d), is O(d3 ). Thus t(d, i) obeys the recursive
equation
d
t(di) = f (d) +7((i -1)t(d -1, i- 1)) ,
which yields
t(d, i) < f (d) + dt(d-1, i-1)

< f(d)+df(d-1)+...+d!f(1)
= O(d4 d!)
Note finally that the hyperplanes being in general position is not a require-
ment for the algorithm. This assumption is only needed in the analysis of the
algorithm. In fact, a perturbation argument clearly shows that when more than
d hyperplanes intersect in one point, the probability that the optimal vertex Xi
of sub-problem Pi is no longer the optimal vertex of sub-problem Pi+, is only
smaller, so that the average running time only decreases.
The following theorem summarizes the results of this section.
Theorem 10.2.1 A linear programming problem with d variables and n con-

straints may be solved by an incremental randomized algorithm in expected time
O(d 4 d!n).
10.3 Convex hulls using a shelling

In dimension d, the size of the convex hull may vary greatly between (d + 1)!
(number of faces of a d-simplex) and Q(nLd/ 2 J) (complexity of a cyclic polytope,
see subsection 7.2.7). It is conceivable that there might be a real advantage in
using an algorithm whose running time is a function of the size of the effectively
computed convex hull rather than the maximal size of this convex hull in the
worst case. The algorithm we present here computes the convex hull of a set
of n points in a d-dimensional space in time O(n2 + f log n) where f is the
number of facets of the convex hull. This algorithm is therefore of great interest
in dimensions higher than 5. The main notion involved in its design is that of a
shelling of a polytope by a line. This notion, which is of interest in its own right,
is described under the next heading. A shelling of a polytope is merely a linear
order on the facets of that polytope and the algorithm we propose here computes
the facets of a polytope in such an order. Let us call extreme points the vertices
of the convex hull. The algorithm needs, in a first phase, to identify the extreme
points among all the points in the set, and for each extreme point, to determine
the first facet that contains this point in the shelling order. This is where linear
programming helps.
Shelling of a polytope
Let P be a polytope. A shelling of P is a sequence (F1 , F2 ,..., Fm) enumerating

all the facets of P such that, for all i, 1 < i < m, the set {Fi, F2 , . . ., Fi} of the
first i facets is a connected set of facets.1
A line that intersects the interior of a polytope P is said to be acceptable for P
if it intersects all the hyperplanes supporting P along a facet at pairwise different
points. The following discussion shows that any oriented acceptable line defines
a linear order on the facets of P which is a shelling.
Let then P be a polytope, L an oriented line, acceptable for P, and 0 a point on
L and in the interior of P. Imagine a point V that moves along L, from 0 towards
infinity in the direction given by the orientation of the line, then reappearing from
the opposite infinity and back towards 0. The hyperplanes supporting P along
its facets are encountered in a certain order by V as it moves along L, and this
order is entirely determined since L is acceptable (see figure 10.1). This order
gives the shelling of P induced by L. Let U be the directing vector of L. Then
L may be parameterized by
L = {V(t) = U : t c ]-x,+oo[ \ {}}
which simulates the motion of V. To each facet Fi of P there corresponds a

parameter ti such that the hyperplane Hi supporting P along Fi is intersected
by L at V(ti). The oriented acceptable line orders the facets of P in increasing
order of their parameters ti.
Lemma 10.3.1 The sequence (F1 , F2 ,..., Fm) of facets of a polytope P ordered
by an acceptable oriented line is a shelling of the polytope P.
Proof. For each value t C ]tl, tm[ of the parameter, the point V(t) on L is exterior
to the polytope P. We may color the facets of P with respect to V(t) as in sec-
tion 8.3.2 If t is negative and belongs to ]ti, ti+l[, then the facets {FF,F 2 ,.. ., Fi}
'A set of facets of a polytope P is said to be connected if it determines a connected subgraph
of the adjacency graph of P.
2
1t is helpful to recall the terminology of section 8.3 here. Let P be a polytope, F a facet
of this polytope, H the hyperplane that supports P along F, H+ the half-space bounded by H
that contains the interior of P, and H- the other half-space bounded by H. Facet F of P is
blue with respect to a point V if V belongs to H+, and red if V belongs to H-.
Figure 10.1. The shelling of a polytope P induced by a line L.
whose corresponding parameters are smaller than t are red with respect to V(t),
whereas the facets whose corresponding parameters are greater than t are blue
with respect to V(t). If t is positive and belongs to ]ti,ti+l[, then the facets
{F1 , F2 , . . ., Fi} whose corresponding parameter is smaller than t are blue with
respect to V(t), whereas the facets whose corresponding parameter is greater
than t are red with respect to V(t). Therefore, the connectedness of the sets of
facets {F1 , F 2 ,.. ., Fi} for every i =1, . . ., m is a consequence of lemma 8.3.3. LI
Shelling for convex hulls
The algorithm described here builds the convex hull conv(S) of a set S of n points
in the space Ed. The underlying principle of this algorithm is to enumerate the
facets of P in turn, in the order given by the shelling of conv(S) induced by an
acceptable oriented line.
The set S = {Pi, P 2 , . . ., Pn} of points is supposed to be in general position.
Each facet of conv(S) is thus a (d - 1)-simplex and is entirely characterized by
the set of its d vertices. The algorithm not only builds the facets of conv(S) but
also their adjacency graph. From this graph, it is easy to reconstruct the entire
incidence graph of the polytope conv(S) (see exercise 8.2).
Before explaining the details of the algorithm, we must clarify how we discover
the facets of conv(S) at all, and moreover how this can be done in the order of the
shelling induced by a line L. Suppose therefore that there is a line L acceptable
for conv(S). As before, we agree that L is given a parametric representation:
L= {V(t) = t : t G ]-x,±+oo[ \ {0}}

Let (FI, F2 , . . ., Fm) be the sequence of facets of conv(S) in the shelling of conv(S)
induced by L. We also denote by ti the parameter of the intersection point
Hi n L of L with the hyperplane Hi that contains facet Fi. At a given stage of
the algorithm, the first i facets (F 1 , F2 , .. ., Fi) have been discovered, together
with their adjacency relationships. We call i-horizon (or horizon for short when
i is clearly understood) the boundary of the union U>=1 Fj of these facets. The
horizon is made of the (d - 2)-faces of conv(S) that are incident to only one of
the facets in (F1 , F 2 , .. ., Fi) together with the faces of all dimensions contained
in these (d - 2)-faces.
The algorithm uses the following lemma.
Lemma 10.3.2 Suppose that the first i facets (F1 , F2 ,..., Fi) of the shelling have
been discovered, together with their adjacency relationships.
* The horizon is made of the faces of conv(S) that are purple3 with respect
to V(t) for any t G Iti, ti+C[.
* The horizon is isomorphic to the set of faces of a (d - 1)-polytope with
respect to the incidence relationships among the faces of conv(S).
* For any (d - 2)-face G of the horizon and any point V(t) on L with param-
eter t E [ti,ti+l], the afJine hull of {GUV(t)} is a hyperplane that supports
conv(S) along G.
Proof. A face G of conv(S) belongs to the horizon if and only if it is contained

in at least one facet Fk with parameter tk < ti, and one facet F1 with parameter
t1 > ti+,. It follows from the proof of lemma 10.3.1 that G is purple with respect
to any point V(t) on L with parameter t E ]tk, t1[, and in particular with respect
to a point V(t) with parameter t E ]ti, ti+1[. The second statement in the lemma
is a direct consequence of lemma 8.3.4. Let us prove the third statement. Let
G be a (d - 2)-face G of the horizon. G is the intersection of two facets F1 and
Fk of conv(S) such that tk < ti < ti+1 < t1 and G is purple with respect to any
point V(t) of L with parameter t E Itk, t1[. Lemma 8.3.1 thus shows that, for any
t E ]tk, tZ [, conv({G U V(t)}) is a facet of conv(S U V(t)), which also proves that
the affine hull aff ({G U V(t)}) is a hyperplane that supports conv(S) along G. If
t = tk (resp. t = t1), conv({G U V(t)}) is exactly Fk (resp. F1) and the affine hull
aff({G U V(t)}) is a hyperplane that supports conv(S) along Fk (resp. Fl). E
Suppose therefore that the algorithm has already built the portion of the adja-
cency graph that contains the first i facets of the shelling of conv(S) induced by
L. Lemma 10.3.1 implies that the (i + 1)-st facet Fi+1 is necessarily incident to
one (or several) (d - 2)-faces of the horizon. Two cases may arise (see figure 10.2).
3
A face of a polytope P is purple with respect to a point V if it is contained in a blue facet
and in a red facet.
'N
/ V
/ N
/ h... -
(a) (b)
Figure 10.2. Shelling of a polytope:

(a) Facet Fj+j is incident to several (d - 2)-faces on the horizon.
(b) Facet Fj+1 is incident to only a single (d - 2)-face on the horizon.
First case. Facet Fi+j is incident to several (d - 2)-faces on the horizon. Let
G1 and G 2 be any two (d - 2)-faces on the horizon that are incident to
Fi+i. Since Fi+j is a (d - 1)-simplex, G1 and G 2 must be incident to a
common (d - 3)-face K = G1 n G 2 . Both faces G1 and G 2 therefore have
d - 1 vertices, of which d - 2 belong to K. Together, D 1 and G 2 have d + 1
distinct vertices. Facet Fi+j is the convex hull of G1 U G 2 .
Second case. The intersection between facet Fj+I and the union U>=, Fj of the
previous facets only yields a single (d - 2)-face on the horizon. Then facet
Fi+j is necessarily the convex hull conv(G U {P}) of G and a vertex P
of conv(S) that belongs neither to G nor to any of the already discovered
facets Fj, j < i. This implies that Fi+j is the first facet in the shelling that
has admits P as a vertex.
To summarize, either Fj+i is the convex hull of two (d - 2)-faces on the horizon
that share a common (d - 3)-face, or it is the first facet in the shelling that
contains some vertex P of S.
Suppose for now that, in an initial phase, the algorithm has determined for
every point P in S whether it is a vertex of the convex hull conv(S) and, if so,
which facet Fp is the first facet in the shelling that contains P. Each vertex
of cony(S) has a pointer to a record that contains this (d - 1)-simplex Fp, the
corresponding hyperplane Hp (the affine hull of Fp) and the corresponding value
tp of the parameter of the intersection point HpnL. Let JFj be the set of simplices
Fp whose parameter tp is greater than ti.
Since the horizon is isomorphic to a (d - I)-polytope (lemma 10.3.2), each
(d - 3)-face on the horizon is incident to two (d - 2)-faces of the horizon (theo-
rem 7.1.7). For each (d - 3)-face K on the horizon we store a pointer to a record
that contains the simplex FK which is the convex hull of the (d - 2)-faces on the
horizon that are incident to K, and also store the corresponding hyperplane HK
(the affine hull of FK) and the corresponding value tK of the parameter of the
intersection point HK n L. Let .F' be the set of simplices FK whose parameter
tK is greater than ti.
Lemma 10.3.3 The next facet Fi+I in the shelling is the simplex F* of Fi =
Y[ U .F'! that has the smallest parametert*.
Proof. The previous comments show that the face Fj+* is in Fi. Also, the
affine hull Hi+I of Fi+j is a hyperplane that supports conv(S) and, among all
the simplices of Pi whose affine hull supports conv(S), Fj+j is defined as the one
with the smallest parameter. Thus it suffices to show that the affine hull H* of
the simplex F* in .Fi whose parameter is minimal is a hyperplane that supports
conv(S). Now if F* is a simplex of .?T stored in a record that is pointed to
by a vertex of conv(S), then H* is indeed a hyperplane that supports conv(S),
by its definition. Thus suppose that F* is a simplex of .F7 pointed to by some
(d - 3)-faces on the horizon, and let K be such a (d - 3)-face. Let G1 and G2
be the two (d - 2)-faces of the horizon incident to K. From lemma 10.3.2, the
hyperplane aff({Gi U V(t) }) supports conv(S), for any t E [ti, ti+1 ]. In particular,
since ti < t* < ti+,, H* = aff({Gi U V(t*)}) supports conv(S). E
To build the facets of the polytope conv(S) in the order of its shelling induced
by a line L, the preceding lemma suggests that the set 9i of potential facets be
maintained and ordered according to their parameters.
But before this, we must explain how to determine, for each point P in S,
whether it is a vertex of the convex hull and, if so, how to find the first facet Fp
in the shelling that contains P. This is where linear programming comes in.
Lemma 10.3.4 To determine whether a point P of S is a vertex of the convex

hull conv(S) and, if so, to find the first facet Fp in the shelling that contains P
can be cast into a linear programming problem with n - 1 constraints and d - 1
variables.
Proof. Line L is acceptable for conv(S), so there is a point 0 in L inside conv(S).

Let H be a hyperplane in the space Ed that does not pass through the origin 0
and H* be the pole of H in the polarity centered at 0 (see section 7.1.3):
H ={X E Ed : H*.X=1}.
Then H passes through P if and only if H* satisfies:
H* P = 1, (10.1)
and H* supports P = conv(S) if and only if
VP' E S, H* PI < 1. (10.2)
Also, H intersects line L at a point with parameter t(H) defined by
t(H) = -U. H*.
Finding the hyperplane that supports P, contains P, and minimizes the pa-
rameter t of its intersection point with L reduces to finding the vector H* that
satisfies equation 10.1 and inequality 10.2, and that also minimizes the linear
functional t(H*). By changing the coordinate system, segment OP can be made
to lie on the xd-axis. Equation 10.1 determines the d-th component of vector
H* and the system given by inequalities 10.2 appears as a set of n - 1 linear
constraints over the remaining d - 1 components of H*. This system is thus a
linear programming problem with n - 1 constraints and d - 1 variables. If the
problem is unfeasible, then there is no supporting hyperplane that passes through
P and this means that P lies inside the convex hull P. If not, the general posi-
tion assumption means that the solution of the linear programming problem is
uniquely defined when line L is acceptable. This unique solution is therefore a
vertex of the feasibility domain that belongs to d - 1 hyperplanes corresponding
to d - 1 constraints in the system (10.2). Let Sp be the subset of d - 1 points in
S \ {P} that correspond to these constraints. The first facet Fp in the shelling
that contains P is then Fp = conv(Sp U {P}). [1
We can now explain the algorithm in its entirety.
The algorithm
First of all, we must choose an origin 0 in the interior of conv(S) and an oriented
line L that contains 0. We assume that L is acceptable for conv(S). If at any
stage of the algorithm it appears that L is not acceptable, the vector U directing
L can always be perturbed a little to make L acceptable without modifying the
result of the previous computations.
The algorithm builds the facets of the convex hull conv(S) in the order of the
shelling of the polytope conv(S) induced by L. It also builds the adjacency graph
of the facets. Each facet is described by the list of its vertices. In addition to the
adjacency graph of the current set of facets, the algorithm maintains the following
three data structures.
* The horizon graph denoted by It contains a node for each (d - 2)-face on

the horizon and an edge for each pair of (d - 2)-faces on the horizon that
share a common (d - 3)-face. Each (d - 2)-face on the horizon is incident
to exactly one facet that is already stored in the adjacency graph, so the
corresponding node in the horizon graph contains a pointer to this facet.
* A dictionary D contains, for each vertex P of conv(S), the (d - 1)-tuple Sp

of points of S such that Fp = conv(Sp U P) is the first facet of the shelling
that contains P. Each (d - 1)-tuple Sp corresponds to a (d - 2)-face of
conv(S) that will be on the horizon at some point during the execution of
the algorithm. Each item in the dictionary D contains a pointer that will,
at that time, point to the corresponding node in the horizon graph.
* A priority queue Q maintains the set Fi of facets that, when the first i
facets (F1,...,Fi) of the shelling have been computed, are candidates for
the next facet Fi+1. To each simplex of Ti corresponds a value of the
parameter that is the parameter of the intersection point of the affine hull
of this simplex with line L. The simplices in .Fi are ordered in the priority
queue by increasing values of their parameters. The priority queue stores
the d vertices of each simplex, and a pointer. If the simplex is the first
face Fp incident to a vertex P of conv(S), the pointer gives the entry in
the dictionary D corresponding to the (d - 1)-tuple Sp. If the simplex is
associated with a (d - 3)-face K on the horizon, the pointer gives the edge
of the horizon graph that corresponds to K.
Initialization. In a starting phase, the algorithm initializes the structures Q

and D. For this, it must solve the linear programming problem corresponding to
each point P. If the problem is unfeasible, then P is not a vertex of the convex
hull and can be discarded from further consideration. Otherwise, the solution to
the linear programming problem gives a facet Fp which is the first facet incident
to P in the shelling of conv(S) by L. This facet is the convex hull conv(Sp U {P})
of P and a subset Sp of S with d - 1 elements. The priority queue is initialized
with the facet Fp for each P E S and the subsets Sp are inserted in the dictionary
D.
First phase. Let ti be the smallest parameter of the facets present in the
priority queue Q immediately after the initial phase. This value t1 is achieved by
the simplices Fp corresponding to the points P in some subset SI of d points in S.
The convex hull conv(S 1 ) is the first face F1 in the shelling. The horizon graph
initially contains the complete graph on d nodes corresponding to all the (d - 2)-
faces of the (d - 1)-simplex Fl. Each subset of SI of size d - 1 is in the dictionary
D and corresponds to a node in the horizon graph. The item in the dictionary D
that corresponds to this subset is located and its corresponding pointer updated.
All the simplices of parameter t1 are retrieved from the priority queue.
Current phase. As long as the priority queue Q is not empty, the algorithm
extracts from Q the set of candidates with the smallest parameter t*, uses it to
determine the next facet in the shelling, and updates the adjacency graph of the
facets and the data structures 7i, D and Q.
If t* is the parameter of a simplex Fp that corresponds to a point P in S, the
subset Sp of vertices of Fp minus P itself is located in the dictionary D. The
pointer associated with this item allows the retrieval of the node in the horizon
graph that corresponds to Sp. The face Fp may then be added to the adjacency
graph. In the horizon graph, the node that corresponds to Sp is replaced by
a complete graph with d - 1 nodes, each of which corresponds to a subset of
Sp U {P} with d - 1 points other than Sp itself.
In the opposite case, t* is the parameter of one or more (d - 3)-faces on the
horizon. Let K be the set of (d - 3)-faces K for which tK = t*, and let g be the
set of (d - 2)-faces on the horizon that are incident to the faces of K. The facet
that must be added to the shelling is the (d - 1)-simplex FG, the convex hull of
the vertices of any two faces in g. This facet FG is adjacent to all the facets that
have already been built and that are incident to the (d - 2)-facets of g. Let g be
the number of faces in A9. The pair (5, K) is a complete subgraph with g nodes
in the horizon graph. This subgraph is replaced by the complete subgraph with
d - g nodes that correspond to the (d - 2)-faces of FG that do not belong to g.
In either case, all the hyperplanes of parameter t* are extracted from the prior-
ity queue. Each new edge in the horizon graph (resp. each edge that was removed
from the horizon graph) corresponds to a (d - 3)-face K' and the convex hull FKI
of the two (d - 2)-faces on the horizon that are incident to K' are inserted into
Q (resp. removed from Q) if its parameter t' is greater than t*. Also, for each
new node in the horizon graph, the algorithm checks whether its set of vertices
is stored in the dictionary D and, if so, updates the corresponding pointer.
When the priority queue Q is empty, the algorithm has discovered all the (d- 1)-
facets of the convex hull and their adjacency graph. Using this, it may build the
entire incidence graph of the polytope conv(S) (see exercise 8.2).

Theorem 10.3.5 Let S be a set of n points in general position in Ed. The algo-
rithm above computes the convex hull conv(S) in expected time O(n2 + f log n),
where f stands for the number of facets of conv(S).
Proof. If the dimension of the space is considered as a constant, the n linear

programming problems in the initial phase may be solved in time O(n 2 ). Let
us count the number of operations performed for each facet discovered by the
algorithm. The number of nodes added to or removed from the horizon graph is d,
and the number of queries and updates into the dictionary D is O(d). The number
of edges created or removed in the horizon graph is d(d - 1) and the number of
queries and updates into the priority queue Q is d(d - 1) as well. The size of
the dictionary D is at most 0(nLd/ 2 j) (by the upper bound theorem 7.2.5) and
the same bound is valid for the size of Q. Each operation on the data structures
can therefore be carried out in time O(logn). For each facet discovered by the
algorithm, the running time is therefore O(logn). Finally, the computation of
the incidence graph from the adjacency graph can be performed in time 0(f)
(see exercise 8.2).
10.4 Exercises
Exercise 10.1 (Unbounded linear programming problems) Consider a (possibly
unbounded) linear programming problem, with the constraints expressed as
X Ai <ai, 1< i =1, .. .,n,
where, for each i = 1, ... , n, Ai is a vector in Ed and ai a constant. By scaling if
necessary, one may assume that all the components of Ai, i = 1, . . . , n, are integers. Let
a = maxlsi•n(ai), and
A = mai<<< (Ai,<
)
if Aij denotes the j-th component of the vector Ai. Prove that, if a solution X to the
linear programming problem exists, any of its components Xj, j = 1, . . . , d, satisfies
Xi < dd/ 2 Ad-la
Hint: The point X is the solution of a d x d system of linear equations whose coefficients
are bounded by A on one side and by a for the constant side. The determinant D of this
system is a non-zero integer, and thus IDI is at least 1. Cramer's rules imply that X,
is the quotient by D of a d x d determinant that has the coefficients of d - 1 columns
bounded by A and the coefficients of one column bounded by a.
Exercise 10.2 (Separability) Let Si and S2 be two sets of points in Ed. Show that
there is an algorithm that decides in linear time whether the sets can be separated by
a hyperplane, that is whether there exists H such that SI is contained in one of the
hyperplanes bounded by H and S2 in the other.
Exercise 10.3 (Ray shooting) Let S be a set of points in general position in Ed.
Choose an origin 0 inside the convex hull conv(S), and a vector V in Ed. Show how to
compute in linear time the facet of conv(P) that is intersected by the ray originating at
0 in the direction V.
10-4. Exercises 237
=
Exercise 10.4 (Intersection of half-spaces) Let Q nl7 1 Ht be an intersection of
m half-spaces in Ed. Determine whether Q is empty and, if not, find a point 0 inside
the intersection Q in linear time.
Exercise 10.5 (Minimum area annulus) An annulus is the portion of the plane con-
tained between two concentric circles. Let n points in the plane be given. Find the
annulus of minimal area that contains all these points.
Hint: This can be shown to be a linear programming problem if we use the space of
spheres that is introduced in chapter 17. More directly, let {Pi(ui, vi) :1 < i < n} be
the set of n points in the plane. The problem can be expressed as deciding whether there
is a center (x, y) and two radii r, and r2 such that r2 - r2 is minimal subject to the 2n
constraints
2 2
r2 < (X - U) + (y - U) <2r.
This optimization problem can be cast into a linear programming problem if instead of
the variables X, y, r1 , and r2, we express the constraints in terms of the variables x, y,
x2 Y
-r2 and x2 + y 2 -rr2
Exercise 10.6 (Convex hull in time O(nlogh)) Let S be a set of n points in E2 .

Show that if S has only h extreme points, the following algorithm computes the lower
hull conv (S) of S in time O(n log h). (The lower hull is defined in exercise 7.14).
If n < 2, then return the segment conv(S).
Otherwise
1. Split S into two balanced subsets by a vertical line A.
2. Find the edge E of conv- (P) that intersects A (see exercise 10.3).
3. Recursively compute the lower hulls of each subset P n A- and P n A+, where A-
(resp. A+) is the closed half-space on the left (resp. on the right) of the vertical
line Al (resp. Ar) that passes through the left (resp. right) endpoint of E.
Exercise 10.7 (Bisection) We denote by LP(X, V) the linear programming problem

with d variables, where NH is the set of constraints and V is the vector of Ed defining
the minimization function. Assume that the set of constraints X has n elements. Show
that it is possible to know on which side of a hyperplane H parallel to V the solution to
LP(NH, V) lies (assuming that there is one), by solving at most five linear programming
problems with d - 1 variables and n constraints.
Hint: The first problem to be solved is the restriction LP(H nH1, V) to H of LP(7t, V).
If this problem is unbounded, then LP(X, V) itself is either unbounded or unfeasible
(the latter happens when some constraint in NH is parallel to H). If there is an optimal
solution to this linear programming problem, however, consider two hyperplanes H' and
H" parallel to H on either side of H. Solve the two lower-dimensional linear program-
ming problems LP(H' n Hopt, V) in H' and LP(H" n Nopt, V) in H", where NHopt iS
the minimal set of constraints that defines the solution to LP(H n X, V), and compare
the optimal values of these sub-problems. Finally, if LP(H n N, V) is unfeasible, then
consider the three following subsets of Ri in turn: the subset 'Ho of constraints whose
hyperplanes are parallel to V, the subset 1H+ of constraints whose half-spaces are un-
bounded in the direction of -V, and the subset 'X- of constraints whose half-spaces
are unbounded in the direction of V. The amount of unfeasibility of LP(H n A, V) can
be defined as the difference between the optimal value of LP(H n A-+, V) and that of
LP(H n e--, -V). It suffices to compare the amount of unfeasibility of LP(H' n t, V)
with that of LP(H" n AH, V).
Exercise 10.8 (Prune-and-search) Let LP be a linear programming problem with d

variables. Let V be the vector that defines the optimizing function and Ho a hyperplane
perpendicular to V. The constraints of LP are put into three categories: 'tH, 1 -+and
7i- as in the previous exercise.
1. Let HI and H2 be two hyperplanes corresponding to constraints of LP that are in
the same category. Let H 12 be the hyperplane parallel to V that contains H1 n H 2. Show
that if the solution of LP is known to lie on either side of H 12, then at least one of the
constraints may be pruned away.
2. In dimension 2, show that it is possible to prune half of the constraints by locating
the solution of LP with respect to a line H parallel to V, which is explained in the
previous exercise.
3. Generalization to dimension d > 2: show that a fixed fraction an of the constraints
may be pruned away (with a = 21-2 ) by locating the solution of LP with respect to
2 d- 1 hyperplanes parallel to V.
4. Using this, devise an algorithm that solves a problem LP with d variables and n
constraints in time c(d)n where c(d) = 0(20(2d)).
Hint: (For the second question.) Pair off the lines in 7-+, and do the same to the lines
in 7H-. Then project onto Ho the intersection of the two lines in each pair, as well as
the hyperplanes in 'Ho, and choose for H the line parallel to V that passes through the
median of the projections.
Exercise 10.9 (Minimum enclosing circle) Given n points in E2 , find the circle
with the smallest radius that contains all these points.
Hint: In a first step, show that the restricted problem where the center of the circle lies
on a given line can be solved in time O(n). Show also that it can be decided in time
O(n) on which side of this line the center of the (unrestricted) minimum enclosing circle
lies. Then apply to this problem the prune-and-search method described in the previous
exercise.
Exercise 10.10 (Maximum inscribed sphere) Let Q be a polytope given as the in-
tersection of m half-spaces in Ed. Determine the sphere inscribed in Q that has the
greatest possible radius.
Exercise 10.11 (LP-type problems) Let R be a finite set (whose elements are called
the constraints) and f a function on 2t that takes its values in a totally ordered set. An
optimization problem is to find the minimal subset B of t such that f(13) = f(R). Such
a problem is an LP-type problem if the following two conditions are true:
Monotonicity: For any two subsets F and G of E such that F C g C 7-, we have
f(F) < f(g).
Locality: For any two subsets F and 5 of 7- such that F C 5 C 7H and f(JF) = f(5),
and for any H E A, we have f(5) < f( U {H}) if and only if f(F) < f(Fu {H}).
Show that any linear programming problem is an LP-type problem. Show that the
minimum enclosing circle problem (see exercise 10.9) is also an LP-type problem, and so
is its generalization to minimum enclosing spheres in any dimension.

The concept of a linear programming problem was created around 1947 when Dantzig
invented the simplex algorithm to solve scheduling problems for the U.S. Air Force. This
method is still widely used, and performs well in practice. Nevertheless, it has been
proved to require in the worst case a time that is exponential in the number of the
variables and constraints [138]. Linear programming algorithms devised more recently
by Khachiyan [135] and Karmarkar [134] run in time polynomial in the number d of
variables, the number n of constraints, and the total number b of bits needed to express
the coefficients in the constraints. Khachiyan's algorithm is often referred to as the
ellipsoid method, and Karmarkar's is the precursor of the so-called interior-pointmethods.
Papadimitriou and Steiglitz [187] give a good introduction to linear programming and
combinatorial optimization. They give a solution to exercise 10.1.
In many cases and in particular in geometric problems, the number d of variables
is relatively small. The works of Megiddo [160, 161] and Dyer [87, 88] provide many
algorithms that take advantage of this: the corresponding complexities are linear in the
number n of constraints when the dimension d is considered a constant. The method
in Megiddo [160, 161] and Dyer [87, 88] is the so-called prune-and-searchmethod that
is outlined in exercises 10.7 and 10.8. The method consists in reducing the domain of
lEd in which to search for the solution (bisection of the problem), so as to prune away
a constant fraction of constraints that become redundant on the smaller domain. The
solution to exercises 10.7 and 10.8 is also presented in the book by Edelsbrunner [89].
The prune-and-search method may also be used to solve quadratic or convex optimization
problems. In particular, it can be used to solve the minimum enclosing circle problem
as indicated by exercise 10.9. The complete solution to this exercise can be found in a
paper by Meggido [160] or in another by Dyer [88].
The complexities of Meggido's and Dyer's algorithms depend doubly exponentially
on the number d of variables. This dependence was lowered to singly-exponential by
Clarkson [66]. Later on, faster and simpler randomized algorithms were devised by
Kalai [133], Clarkson [68], Sharir and Welzl [208], and Matougek, Sharir, and Welzl [157].
The one we present in section 10.2 is due to Seidel [205]. Welzl [218] adapts it to yield a
simple randomized algorithm for the minimum enclosing circle.
The existence of a shelling of a polytope was first proved by Bruggesser and Mani [39].
They used a shelling induced by a line as we did in section 10.3. The convex hull
algorithm described in section 10.3 is due Seidel [202]. More recently, Matousek and
Schwarzkopf [156, 153] showed how to preprocess the set of constraints of the n linear
programming problems solved by the algorithm in order to solve them more efficiently.
Their approach leads to an algorithm whose complexity is O(n2 1+ [d/2J +f ±flogn) for n
points in dimension d if the convex hull has f facets. The notation e stands for a constant
that can be made as small as wanted (albeit at the cost of increasing the constant in the
0( notation). In dimension 4 or 5, this complexity is O(n4 /3+E + f log n).
The convex hull algorithm in dimension 2 (described in exercise 10.6), whose complex-
ity O(n log h) depends on the number h of vertices of the computed convex hull, is due
to Kirkpatrick and Seidel [137]. In [99], Edelsbrunner and Shi generalized this algorithm
to 3 dimensions, obtaining an algorithm of complexity 0(n log 2 h).
Lastly, the LP-type problems defined in exercise 10.11 generalize the formulation in
terms of linear programming. Many geometric problems can be expressed as LP-type
problems, such as computing the smallest enclosing ellipsoid, the largest ellipsoid in-
scribed in a polytope, the smallest circle that intersects n convex objects, etc. The
randomized algorithm of Clarkson [68] or those of Sharir and Welzl [208] and Matousek,
Sharir, and Welzl [157] actually solve LP-type problems. Chazelle and Matousek [57]
even explicitly discussed under which conditions a deterministic algorithm to solve an
LP-type problem can be obtained by derandomization.
Part III
Triangulations
To triangulate a region is to describe it as the union of a collection of simplices

whose interiors are pairwise disjoint. The region is then decomposed into ele-
mentary cells of bounded complexity. The words to triangulate and triangulation
originate from the two-dimensional problem, but are commonly used in a broader
context for regions and simplices of any dimension.1
Triangulations and related meshes are ubiquitous in domains where the ambient
space needs to be discretized, for instance in order to interpolate functions of
several variables, or to numerically solve multi-dimensional differential equations
using finite-element methods. Triangulations are largely used in the context of
robotics to decompose the free configuration space of a robot, in the context
of artificial vision to perform three-dimensional reconstructions of objects from
their cross-sections, or in computer graphics to solve problems related to windows
or to compute illuminations in rendering an image. Finally, in the context of
computational geometry, the triangulation of a set of points, a planar map, a
polygon, a polyhedron, an arrangement, or of any other spatial structures, is
often a prerequisite to running another algorithm on the data. For instance, this
is the case for algorithms performing point location in a planar map by using a
hierarchy of triangulations, or for the numerous applications of triangulations to
shortest paths and visibility problems.
Triangulations form the topic of the next three chapters. Chapter 11 recalls
the basic definitions related to triangulations, and studies the combinatorics of
triangulations in dimensions 2 and 3. Chapters 12 and 13 are concerned with
algorithmic problems on triangulations in dimensions 2 and 3 respectively. Es-
sentially, two types of problem are studied. In the first kind of problem, the
vertices of the triangulation are given and we seek to decompose their convex
hull into simplices. In the second kind, the triangulation is required to include
'Although usage has consecrated that terminology, it would be more appropriate to use the
words to simpticiate and simpliciation.
242 Part III - Triangulations
several given simplices of positive dimension, such as those on the boundary of a

polyhedral region. For the latter, we speak of a constrained triangulation. Prob-
lems of this kind always have solutions in dimension 2 and algorithms that solve
them efficiently are given in chapter 12. In dimension 3, however, this may not
always be the case and chapter 13 exhibits a few pitfalls and describes ways of
avoiding them.
Chapter 11
Complexes and triangulations
The notion of a triangulation occurs naturally if we seek to describe the fun-

damental objects of linear geometry (polygons, polyhedra, etc.) in terms of
elementary objects that have a bounded complexity. Indeed, the simplest of all
such objects are simplices.
In section 11.1, we recall the definitions of a simplicial complex, of a cell com-
plex, and of a triangulation. In particular, this section emphasizes the relations
between these notions and the usual objects in linear geometry (polygons, poly-
hedra, polygonal and polyhedral regions, etc.). Section 11.2 studies the combi-
natorics of triangulations in dimensions 2 and 3. Finally, section 11.3 gives a
representation for these complexes and establishes a duality between them.
11.1 Definitions
11.1.1 Simplices, complexes
In this section, we work in the d-dimensional space Ed. Recall that a simplex
of dimension k for k < d, also called a k-simplex, is a k-polytope with k + 1
vertices, or equivalently the convex hull of k + 1 affinely independent points. Let
A = {Ao, . .. , Ak} be a set of k + 1 affinely independent points and S be the k-
simplex defined by A. Any subset of I + 1 < k + 1 points in A defines an l-simplex
which is a face of S. Simplices of dimension 0,1, 2, and 3 are respectively called
points, segments, triangles, and tetrahedra.
A complex C is a finite set of simplices that satisfy the following two properties:
1. any face of a simplex in C is also a simplex in C, and
2. two simplices in C either do not intersect, or their intersection is a simplex

of smaller dimension which is their common face of maximal dimension.
244 Chapter 11. Complexes and triangulation
For convenience, we consider that there is an empty simplex, whose dimension

is -1, which is a face of any simplex and hence of any complex. The complex
C is a k-complex, or a complex of dimension k, if the maximal dimension of the
simplices in C is exactly k.
The simplices that constitute a complex are called the faces of the complex. A
face of dimension l is called an 1-face. The faces of dimension 0 are called the
vertices and the faces of dimension 1 are called the edges. In dimension d, the
faces of dimension d and d- 1 are respectively called the cells and the facets. Two
faces of a complex are incident if one is included in the other and their dimensions
differ by one. Two vertices of a complex are adjacent if they share a common
incident edge, and two cells are adjacent if they share a common incident facet.
The 1-skeleton of a k-complex C is the 1-complex, subcomplex of C, consisting of
the faces of C of dimension at most 1. The 1-skeleton of a complex C is isomorphic
to a graph whose nodes are the vertices of C and whose arcs are the 1-faces
of C.
A k-complex C is homogeneous or pure if and only if any face of C is a face of
some k-simplex in C.
Figure 11.1. A 2-complex and a 3-complex. Neither is homogeneous.
To any complex C there corresponds a subset of the space Ed formed by the

points in Ed that belong to the simplices in C. This region is called the domain
of the complex and is denoted by dom(C). Sometimes, when this does not create
ambiguities, we abuse the notation and do not make a distinction between the
complex and its corresponding domain.
A k-complex C is said to be connected if its domain is connected. A k-complex
C is connected if and only if its 1-skeleton is connected.
Let C be a homogeneous k-complex. The boundary of C, denoted by bd(C),
is the homogeneous (k - 1)-complex consisting of all the (k - 1)-faces of C that
belong to only one k-simplex of C, and all the subfaces of these (k - 1)-faces. A
face of C is said to be external if it belongs to the boundary of the complex and

internal otherwise.
Complexes as they have just been defined are sometimes called simplicial com-
plexes since their faces are simplices. For instance, the set of proper faces of
a simplicial d-polytope is a canonical example of a simplicial (d - 1)-complex.
Sometimes, our investigations lead us to define a broader kind of complex, called
a cell complex, whose faces are polytopes or even unbounded polytopes. For in-
stance, it will be very useful to consider the set of faces of a Voronoi diagram (see
chapter 17) or of a hyperplane arrangement (see chapter 14) as a cell complex.
Formally, a cell complex C is a set of polytopes, bounded or unbounded, such
that
1. any face of a polytope in C is also a polytope in C, and
2. two polytopes in C either do not intersect, or their intersection is a polytope

of smaller dimension which is their common face of maximal dimension.
For instance, the set of faces of a d-polytope (except for the empty face) is
a d-dimensional cell complex, and the set of proper faces of a d-polytope is a
(d - 1)-dimensional cell complex. In the remainder of this chapter, we are mostly
concerned with simplicial complexes, which are at the core of the concept of
triangulation. Nevertheless, the definitions and properties that are stated for
simplicial complexes generalize easily to cell complexes.
11.1.2 Topological balls and spheres, singularities

Triangulations will be defined later on as pure connected complexes without sin-
gular faces, and thus we have to give a clear and precise definition of singularities.
Examples of singular faces appear in figure 11.2; vertex A of the complex on the
left and edge AB of the complex on the right are both singular faces. The notion
of singularity is essentially topological and its precise definition calls for that of
topological balls and spheres in Ed.
A topological k-ball is a set of points in Ed homeomorphic1 to the unit ball
of Ek defined as the set of points X = (Xl,.. ., Xk) such that i=lz1 < 1. A
topological k-sphere is a set of points in Ed+l homeomorphic to the unit sphere
in Ek+l defined as the set of points X = (X, Xk+l) such that ik+1 2
1. For instance, any d-polytope is a topological d-ball and its boundary is a
topological (d - 1)-sphere. With a slight abuse of terminology, we call hereafter
a topological ball (resp. topological sphere) a complex C whose domain dom(C)
'Two subsets of a topological space are homeomorphic if there is a continuous injection from
one to the other whose inverse bijection is also continuous.
246 Chapter 11. Complexes and triangulations
R G A
-
C
C
B
(a) (b)
Figure 11.2. Singular faces of a complex:

(a) A 3-complex showing a singularity at vertex A.
(b) A 3-complex whose edge AB is singular.
is a topological ball (resp. sphere). Note that the boundary of a topological k-

ball is a topological k-sphere, while the boundary of a topological k-sphere is
empty.
A O-complex consisting of two points is a topological O-sphere, and a O-complex
consisting of a single point is a topological O-ball. Let (Al, A2 , ... , An) be a finite
sequence of points in the plane. The set of segments which join two consecutive
points in the sequence forms a homogeneous 1-complex if two segments do not
intersect except at their common endpoints. Such a 1-complex is called a polyg-
onal line. Moreover, if the points are pairwise distinct, then the polygonal line
is simple and its boundary is a topological O-sphere formed by its two endpoints
{A 1, An}. If the points are pairwise distinct, except for A 1 and An which are the
same point, the polygonal line is a polygon and its boundary is empty. Simple
polygonal lines and polygons respectively describe all the topological 1-balls and
1-spheres.
To characterize the singularities of a complex, we introduce the star and shell
operations on a complex. Let C be a pure connected d-complex and F be a face
of C. The star of F in C is the sub-complex of C consisting of the d-simplices
of C that contain F and all their faces. The shell of F in C is the sub-complex
consisting of all the simplices in the star of F in C that have an empty intersection
with F. If C is a pure d-complex and F is a k-face of C, k < d, the shell of F in
C is a pure (d - k - 1)-complex. In the complex represented in figure 11.2a, the
shell of A is formed by the faces of the triangles BCD and EFG. In the complex
of figure 11.2b, the shell of edge AB is formed by the edges CD and EF and all
their vertices.
A face F of a complex C is singular if its shell in C is neither a topological ball
nor a topological sphere. For instance, in the 3-complex of figure 11.2a, vertex A
is singular because its shell is not connected, nor a topological O-sphere. For the
same reason, edge AB is singular in figure 11.2b.
11.1.3 Triangulations
Here and in the following, we consider d-complexes embedded in a space of di-

mension d', for some d' > d. Hence, d stands for the maximal dimension of the
simplices in the complex, not for the dimension of the space.
A d-triangulationis defined as a pure connected simplicial d-complex without
singular faces. The shell of any k-face of a d-triangulation is thus either a topo-
logical (d - k - l)-sphere, or a topological (d - k - 1)-ball. More precisely, the
following lemma shows that the shell of any internal face is a topological sphere
(whose boundary is empty) while the shell of any external face is a topological
ball (whose boundary is a sphere).
Lemma 11.1.1 Let C be a pure simplicial d-complex and F a k-face of C, k <

d - 1. The boundary of the shell of F in C is not empty if and only if F is an
external face of C.
Proof. Let F1 be a (d - 1)-face of C that contains the k-face F, and let Fl be

the (d - k - 2)-face of F1 that is disjoint from F. The face F, is a face of the
shell of F in C. For each d-simplex SI = conv(Fl U A) in C that contains F,, the
simplex S' = conv(F, U A) is a (d - k - 1)-simplex in the shell of F that contains
F,. As a result, F, is an external face of the shell of F if and only if F1 is an
external face of C. Note that this proof is perfectly valid for k = d - 1, taking
F = F, and F,' to be the empty face of dimension -1, which is a face of all the
faces of C and hence is in the shell of F in C. [1
The notion of a shell also helps to prove the following three lemmas on trian-
gulations.
Lemma 11.1.2 In a d-triangulation, any (d - 1)-face belongs to a most two

d-simplices.
Proof. Let T be a d-triangulation. If the domain dom(T) of the triangulation

is embedded in an affine space of dimension d, then the proposition follows di-
rectly from the definition of a complex. Nevertheless, it is conceivable that T
is embedded in a space of higher dimension. In any case, we consider, for any
(d - 1)-face F of T, the shell of F in T. This shell is a 0-complex formed by one
or two vertices of T, so that F belongs to either one or two d-simplices of T. E
Lemma 11.1.3 For any pair (T, T') of d-simplices of a d-triangulation, there is
a sequence T 1, T 2 ,... XTn of d-simplices such that T1 = T, T, = T', and Ti is
adjacent to Ti+1 for all i = 1,..., n - 1.
Proof. The lemma is trivial for 1-triangulations and can be proved by induction
on the dimension of the triangulation. Let A = Al, A2 , ... , An = A' be a path
in the 1-skeleton of T that joins a vertex A of T to a vertex A' of T'. Such a
path exists because the complex is connected. We show that for any vertex Ai,
i = 2,.. ., n - 1 on this path, and any d-simplices Ti and Ti+l in T that contain
respectively the edges Ai-,Ai and AiAi+l, there exists a sequence of adjacent
d-simplices in T that joins Ti to Ti+,. Let Fi and Fi+l be the (d- 1)-faces of Ti and
Ti+l that do not contain Ai. Then Fi and Fi+l are (d-1)-faces in the shell of Ai in
T, and the induction proves that there is a sequence Fi = G,, G2 , . . ., Gm = Fi+1
of adjacent (d- 1)-simplices in the shell of Ai in T that joins Fi to Fi+l. Therefore,
the sequence Ti = conv(Gi U Ai), conv(G 2 U Ai), . . ., conv(Gm U Ai) = Ti+i is a
sequence of adjacent d-simplices of T that joins Ti to Ti+,. Dl
Lemma 11.1.4 If T is a d-triangulation, the pure (d- 1)-complex bd(T) that is

the boundary of T has itself an empty boundary.
Proof. We first show that every (d - 2)-face of bd(T) belongs to two (d - 1)-
simplices of bd(T). Let G be a (d - 2)-face of bd(T). The shell of G in T is a
simple polygonal line whose boundary is formed by two points U and V. Since
U belongs to a single edge of the shell of G in T, the (d - 1)-simplex conv(G, U)
belongs to only one d-simplex in I and is thus a (d - 1)-simplex in bd(T). The
same argument applies to the (d - 1)-simplex conv(G, V). El
The notion of singularity can be extended to cell complexes. In the rest of

this book, unless explicitly mentioned, a complex will denote a pure connected
complex without singularities, be it either a simplicial or a cell complex.
11.1.4 Polygons and polyhedra
The 1-triangulations are precisely the simple polygonal lines and the polygons.
A triangulation T (of dimension 1 or 2) is said to be planar if its domain dom(T)
can be embedded in a space of dimension 2. A planar polygon P is a closed,
simple (that is, not self-intersecting), planar curve, and Jordan's theorem (see
exercise 11.1) states that this curves splits E2 \ p into two connected regions,
exactly one of them being bounded. The interior of P is the bounded region and
the exterior of P is the unbounded region.
More generally, we call a polygonal region any connected region in the plane
whose boundary is one polygon or the union of a finite number of disjoint poly-
gons. Depending on the context, we consider a polygonal region to include its
boundary or not. The edges and vertices of a polygonal region are the edges and
vertices of the polygons that bound the region.
A bounded polygonal region whose boundary consists of a single polygon is

a topological ball. Such a region is also called simply connected. A bounded
polygonal region whose boundary consists of k + 1 polygons is a polygonal region
with k holes. An unbounded polygonal region whose boundary consists of k
polygons has k holes.
By lemma 11.1.4, we know that the boundary of a 2-triangulation is a pure
1-complex whose boundary is empty. This complex is not necessarily connected
and can be the union of several disjoint polygons. The domain dom(T) of any
planar 2-triangulation T is thus a bounded polygonal region. Reciprocally, we
will see in the next chapter that any bounded polygonal region P is triangulable,
meaning that the domain it encloses can be decomposed into non-overlapping
triangles, or that it can be expressed as the domain of a 2-triangulation T whose
set of vertices is exactly the set of vertices of P.
We will not discuss here the topology of triangulations in dimensions 3 and
higher. We need only know that 2-triangulations can be orientable or non-
orientable. To orient a triangle is to choose a circular order on its three ver-
tices. The orientation of a triangle induces an orientation of its three edges. The
orientations of two adjacent triangles are consistent if they induce an opposite
orientation on their common incident edge (see figure 11.3). A 2-triangulation
is orientable if it is possible to orient each of its triangles in a way such that all
adjacent triangles have consistent orientations.
Figure 11.3. Consistent orientations of adjacent triangles.
In the 3-dimensional space E3 , an orientable 2-triangulation whose boundary

is empty is a polyhedron. A polyhedron may be a topological sphere, but need
not be. Generally speaking, it is homeomorphic to a sphere with handles (see
figure 11.4). The number of handles is called the genus. A polyhedron home-
omorphic to a topological sphere has genus 0. A polyhedron homeomorphic to
a torus is a handle and thus has genus 1. More generally, a polyhedron with h
handles is said to have genus h. A polyhedron P in E3 separates E3 \ P into two

connected components, exactly one of which is bounded. We call the bounded
component the interiorof the polyhedron, and the other component the exterior
of the polyhedron. The polyhedra as they are defined here are often called sim-
plicial polyhedra since their faces are simplices. We may extend this definition
by defining a polyhedron as an oriented cell complex of dimension 2 in E3 whose
boundary is empty.
More generally, we call polyhedral region any connected region of E3 whose
boundary is formed by one or several disjoint polyhedra. Depending on the
context, a polyhedral region is considered to include its boundary or not. We
call faces of a polyhedral region the faces (vertices, edges, and facets) of the
polyhedra that form the boundary of the region. A bounded polyhedral region
whose boundary consists of a single polyhedron of genus 0 is a topological ball.
A bounded polyhedral region whose boundary consists of a single polyhedron of
genus h is a polyhedral region with h handles. A bounded polyhedral region
whose boundary consists of k + 1 polyhedra is a polyhedral region with k holes.
The boundary of a 3-triangulation is a pure orientable 2-complex whose bound-
ary is empty. To prove the orientability, it suffices to orient any triangle according
to the normal that leaves the 3-simplex of the 3-triangulation that contains this
triangle. The boundary of a 3-triangulation is thus formed by one or several
polyhedra. Therefore, the domain of a 3-triangulation is a bounded polyhe-
dral region. A polyhedral region P is triangulable if there is a triangulation T
whose domain coincides with P and whose set of vertices is the same as that of
P. We will show in chapter 13 that there are polyhedral regions which are not
triangulable.
11.2 Combinatorics of triangulations

The complexity, also called the size, of a complex is the number of its faces of all
dimensions.
11.2.1 Euler's relation for topological balls and spheres
Our study is based on the existence of a linear relation, called Euler's relation,
between the faces of different dimensions in a complex. This relation admits an
elementary proof in the case of a 2-triangulation or for a 2-complex that can be
embedded in a space of dimension 2. Indeed, the 1-skeleton of the complex is then
a planar graph (see exercise 11.4). In higher dimensions, this proof does not work.
In fact, Euler's relation is one of the most famous results of homology, a theory
whose application goes well beyond the scope of this book. We limit ourselves here
to proving Euler's relation for topological spheres and balls, basing the proof on a
11.2. Combinatorics of triangulation 251
single result of the homology theory. (See the proof of theorem 11.2.1 below for a
statement of this result.) In the following subsection we show how to derive from
this result all the Euler's relations for 2-complexes, polyhedra, and polyhedral
regions in E3.
Let C be a d-complex and nk(C) the number of its k-faces, for k = 0, . .. , d. The
Euler characteristice(C) of the d-complex C, is defined as the alternating sum
d
e(C) = Z(_1)knk(C).
k=O
Theorem 11.2.1 (Euler's relation) If the d-complex C is a topological d-ball,

its Euler characteristicis 1:
d
1:(_-)k nk(C) = 1. (1. 1)
k=O
If the d-complez C is a topological d-sphere, its Euler characteristicis 1 + (-l)d:
d
Z(-1)knk(C) = 1 + (-I)d. (11.2)
k=O
Proof. The basic result from homology theory mentioned above is that two
complexes C and C' that have homeomorphic domains also have the same Euler
characteristic.
The set of faces of a d-polytope is a topological d-ball. By theorem 7.2.1, its
Euler characteristic is 1. By the definition, we know that any topological d-ball is
homeomorphic to the domain of a polytope. As a result, the Euler characteristic
of any topological d-ball is 1 and it satisfies equation 11.1.
Similarly, the set of proper faces of a d-polytope forms a topological (d - 1)-
sphere. By theorem 7.2.1, its Euler characteristic is 1 + (-l)d-'. As a result,
the Euler characteristic of any topological d-sphere is 1 + (-l)d, and it satisfies
equation 11.2. 0
11.2.2 The complexity of 2-complexes
In this subsection, we study the complexity of 2-triangulations and of cell com-

plexes of dimension 2 that can be embedded into a space of dimension 2, and
also of simplicial or cell polyhedra. For complexes that can be embedded into
E 2, the number of facets and edges is bounded by the total number of vertices
and the number of external vertices of the complex. For polyhedra, the number
of facets and edges can be bounded by the number of vertices and the genus of
the polyhedron.
For any 2-complex C, we denote by n(C) the number of vertices of C, by m(C)
the number of its edges, and by f (C) the number of its 2-faces. When the context
is clear, we drop the reference to C and simply write n, m, and f for n(C), m(C),
and f (C).
Theorem 11.2.2 (Euler's relation in dimension 2) Let C be a 2-complex

whose domain is contained in E 2. If the boundary of C is the union of k + 1
polygons (dom(C) is thus a polygonal region with k holes), then it satisfies Euler's
relation:
n-m+f = 1-k. (11.3)
Proof. Let us first note that dom(C) is a polygonal region with k holes because
C is as usual assumed to be pure, connected, and without singularities. We
first prove the theorem when the complex C is a triangulation E. If k = 0,
the triangulation T is a topological ball and the previous equation is simply
Euler's relation for topological 2-balls that was proved before. For k $? 0, we
invoke a result that we prove independently in the next chapter, showing that
any polygon can be triangulated. More precisely, for any polygon P there exists a
2-triangulation whose vertices are exactly the vertices of P and whose boundary
is the same as that of P. Let Pi, i = 1, ... , k, be the polygons forming the
boundaries of the holes of dom(T), and let 7 be a triangulation of Pi. Each
triangulation 7 is a topological ball, so it has an Euler characteristic of
e(fT) = n(Tf) - m(Tz) + f (Ti) = 1. (11.4)
The complex T' = T U (u = 0i)is also a topological ball and satisfies
e(T') = n(T') - m(T') + f(T') = 1. (11.5)
The faces common to T and to Uk=1 7 are also faces of Pi, so we can compute
the Euler characteristic of T' as
k k
e(T') = e(T) + ZeC h)-Ze(Pi) (11.6)
Since the Euler characteristic of a polygon is zero, we conclude that
e(T) = 1 - k.
Consider now a cell complex C of dimension 2. The 2-faces of such a complex are
polygonal regions, and can also be triangulated, using the same result as above.
11.2. Combinatorics of triangulation 253
By replacing each non-triangular 2-face of the complex by its triangulation, we

obtain a 2-triangulation that also satisfies Euler's relation 11.3. It is then easy
to see that this relation is also satisfied by the faces of the cell complex C: each
edge added while triangulating a 2-face of C adds an edge but also splits a 2-face
into two new 2-faces, so its leaves the linear combination n - m + f unchanged.
32
Corollary 11.2.3 Let C be a 2-complex in E2 , with n vertices, m edges, and f

2-faces and whose domain is a polygonal region with k holes. If C has ne external
vertices (on the boundary of C), then
f < 2(n-1+k)-ne,
m < 3(n-1+k)-n,.
Equality holds if and only if C is a triangulation.
Proof. Note that ne is also the number of external edges of C. Each external
edge of C is incident to a unique 2-face of C, while each internal edge is shared
between exactly two 2-faces. Also, each 2-face of C is incident to at least 3 edges.
Counting the number of incidences between an edge and a 2-face, we obtain
2m - ne > 3f,
and equality holds if and only if C is a triangulation. It now suffices to use Euler's
relation 11.3 to prove the corollary. [1
Theorem 11.2.4 (Euler's relation for polyhedra) If C is a polyhedron with

n vertices, m edges, f 2-faces, and genus h, then
n-m+f=2-2h. (11.7)
Proof. We provide a proof by induction on the genus h of the polyhedron. Let

C be a simplicial or cell complex. If the genus h of C is 0, then C is a topological
sphere and the relation sought is just Euler's relation for topological 2-spheres,
proved in theorem 11.2.1. Now if C is a polyhedron with genus h > 0, its Euler
characteristic can be derived from that of any polyhedron Ch obtained by adding
a handle to a polyhedron Ch-l of genus h - 1. To add a handle to Chl1, we
consider a polyhedron Co of genus 0 such that
Ch-l n Co = B u B',
where 13 and iS' are two disjoint topological 2-balls (which we call disks) such that
dom(Ch-l) n dom(Co) = dom(B3) U dom(B'),

(see figure 11.4). The polyhedron Ch can be obtained by removing the internal
faces of the two topological disks B and 1' from Ch-l U Co. Let P and P' be
the two polygons that form the boundaries of B and B' respectively. The Euler
characteristic e(Ch) of Ch is given by
e(Ch) = e(Ch-1) + e(Co) - 2(e(B3) + e(B')) + e(P) + e(P').

The Euler characteristic is 1 for each disk B and B', and 0 for each polygon P
and 7'P. We conclude that
e(Ch) = e(Ch-) -2,
and, by induction, that

e(Ch) = 2 - 2h.
This completes the proof of theorem 11.2.4.
Figure 11.4. A handle.
Corollary 11.2.5 Let C be a polyhedron of genus h, with n vertices, m edges,

and f 2-faces. Then
f < 2n-4+4h (11.8)

m < 3n-6+6h. (11.9)
and equality holds if and only if the polyhedron is simplicial.
Proof. By counting the number of incidences between edges and 2-faces of

polyhedron C, we obtain the inequality
2m > 3f
11.2. Combinatorics of triangulations 255
which becomes an equality when polyhedron C is simplicial. Using Euler's rela-

tion 11.7 for polyhedra, we derive both relations 11.8 and 11.9. E
Remark. From corollary 11.2.5, we may also infer that the expected number
of edges incident to a random vertex of a polyhedron is at most 6- 12(1-h)
n
Therefore, any polyhedron of genus h always has a vertex of degree at most 5
when h = 0, and of degree at most 6 if h = 1.
11.2.3 The complexity of 3-triangulations

Let T be a 3-triangulation in E3 . We denote by n(T) the number of vertices of
T, by m(T) the number of its edges, by f(T) the number of its triangles, and by
t(T) the number of its tetrahedra. When there is no ambiguity as to the complex
T, we simplify the notation by writing n, m, f and t for n(T), m(T), f(T), and
t(T).
Theorem 11.2.6 (Euler's relation in dimension 3) Consider a 3-triangul-

ation T whose boundary is a polyhedron of genus h. Euler's relationfor T states
that
n-m+f-t= 1-h. (11.10)
Proof. Again, we give a proof by induction on the genus h of the polyhedron that
bounds the complex T. If h = 0, then T is a topological 3-ball and the theorem
is a consequence of theorem 11.2.1. If h 7$0, then the Euler characteristic of T
is the same as that of any 3-triangulation Th = Th-1 U To, obtained by merging
a 3-triangulation Th-l whose boundary is a polyhedron of genus h - 1 and a
topological 3-ball To in such a way that
bd(Th- ) n bd(To) = 3 U Z1,
where B and 13' are two 2-complexes of disjoint topological balls such that
dom(Th-l) n dom(To) = dom(13) U dom(13').
The Euler characteristic e(Th) of Th is given by
e(Th) = e(Th- ) + e(To) - (e(B) + e(^')).
From this it follows that

e() = I1-h.
El
Corollary 11.2.7 Consider a 3-triangulationT whose boundary is a polyhedron

of genus h. The number t of its tetrahedrasatisfies
n2 3n
n-3+3h< t < - -- ne+3-3h.
2 2
Again, n stands for the number of vertices of T and n, for the number of its
external vertices.
Proof. Denote by ne, me, and fe the respective numbers of vertices, edges, and
triangles on the boundary of T, and by ni, mi, and fi the respective numbers
of internal vertices, edges, and triangles of T. Corollary 11.2.5 applied to the
boundary of T yields
Ae 2ne-4 + 4h (11.11)
me =3ne- 6+ 6h. (11.12)
Each tetrahedron of T is incident to four triangles. In turn, an internal triangle of

T is incident to two tetrahedra, while an external triangle is incident to a unique
tetrahedron. Using the same counting argument as in corollary 11.2.3, we obtain
4t = 2fi + fe. (11.13)

Eliminating fe, fi and f = fe + fi in equations 11.10-11.13, we get
t = m - n - ne + 3 - 3h,
which gives the number of tetrahedra in the triangulation as a function of the
number of its vertices and edges. The bounds on the number of tetrahedra
claimed by the theorem are then an immediate consequence of bounds on the
number of edges. On the one hand, the number of edges is trivially bounded
above by n(n - 1)/2. On the other hand, each internal vertex is incident to at
least four internal edges, each incident to at most two internal vertices, so the
number mi of internal edges is at least 2ni. Thus the total number m of edges
must satisfy
n(n-1) >m=me+mi > 3ne-6+6h+2ni

2 -
> 2n+ne-6+6h.
For a 3-triangulation whose boundary is a polyhedron of genus 0, the above

bounds can be written as
n -3 < t < - - - - ne + 3.
2 2
-
11.2. Combinatorics of triangulations 257
Both upper and lower agree when n = 4, and are thus optimal. Below, we
show that these bounds may also be matched for any n and at least some values
of n,. For this, we must exhibit a 3-triangulation whose number of tetrahedra is
quadratic (resp. linear) in the number of vertices. (See also exercise 11.6).
Lemma 11.2.8 For any integer n, there is a 3-triangulationwith n vertices, all

external, and with t tetrahedra, such that
n2 5n
t= - - - +3
2 2
Proof. Let us choose the vertices of the triangulation on the moment curve.
Recall from subsection 7.2.4 that the moment curve is parametrically described
as {(r,r 2 ,T 3 ), -OC < T < +oo}. Lemma 7.2.6 shows that any three points
A1 ,A2 ,A 3 on the moment curve are affinely independent. Moreover, if these
points have respective parameters T < T2 < r 3, any point M on the moment curve
parameterized by r lies on a given side of the hyperplane passing through A1 A2 A3
when r E] - I,71[ U ]T2, T3[, and on the opposite side if T E ]r 1 , T2[ U ]T3, +00[.
Let An = {Al, ... , An} be a set of n distinct points on the moment curve r,
parameterized respectively by r1 < ** < m. For each i = 1, ... , n, we denote
by Ai the set of the first i points in An. Consider the convex hull conv(Avn-).
Its facets consist of the 2(n -1) - 4 triangles: A 1 Ai, Ai+1 for i = 2, .. ., n - 2
and AiAi+jAn-1 for i = 1, . . ., n - 3. Indeed, the affine hulls of these triangles
leave all the points of An-l on some given side. For each i = 1, . . . , n - 3, the
affine hull of the triangle AiAi+ 1An- 1 separates point An from the other points
in An-, and, according to the terminology set up in chapter 8 (see section 8.3),
the triangle AjAi+,An-I is a red facet of conv(An-1) with respect to An.
Consider the complex Tn defined by the following induction:
T4 = {AjA 2 A 3 A 4 }
Tn = Tn-j U {AiAi+±An-An : i = .. - 3},
where each complex Tf is a pure 3-complex described by all the tetrahedra that
belong to it (adding all the 0-, 1-, and 2-faces of these tetrahedra). We now
show by induction on n that En is a triangulation. Indeed, for i = 1, . .. , n - 3,
the tetrahedron AiAi+iA,-iAn has no points in common with the tetrahedra
of Tn-j except for points in the triangles AiAi+,An- 1 on the boundary of Tn- .
The complex Tn is pure, by definition, and connected because its 1-skeleton is a
connected graph. Moreover, we can check easily that the shell in Tn of any vertex
Ai is a topological 2-sphere, that for i = 2, .. ., n the shell of any edge A1 Ai, is
a simple polygonal line, that the same holds for edges AiAn, i = 1,..., n - 1,
and finally that fore any 1 < i < j -1 < n - 1 the shell of the edge AiAj is a
simple polygon. This guarantees that En is indeed a triangulation. The domain
of Tn is the convex hull of A. The number Tn I of tetrahedra of TI is given by

the recurrence
IT41 =1 (11.14)
I-1= T.iI + n-3. (11.15)
Solving this recurrence yields ITnJ
n 2 /2 - 5n/2 + 3 which, by corollary 11.2.7,
is the maximum number of tetrahedra for a 3-triangulation with n vertices, all
external, and whose boundary is a polyhedron of genus 0. [E1
The following lemma shows the existence of linear triangulations.
Lemma 11.2.9 For any pair of integers (n, ne) such that 4 < ne < n, there
exists a 3-triangulation with n vertices, ne of which are external, and with t
tetrahedra, where
t=n-3+2(n- n).
Proof. Examining the proof of corollary 11.2.7, we note that the lower bound
n - 3 + 2(n - ne) for the number of tetrahedra can only be achieved when the
number of edges itself also achieves its lower bound, 3nT - 6 + 2(n - ne), and this
implies that there can be at most 2(n - ne) internal edges.
We can realize these conditions easily when all the vertices are external, namely
when n = ne. Indeed, we may build a triangulation without internal edges
incrementally, starting with the tetrahedron defined by the first four vertices.
In an incremental step, the next triangulation can be obtained by adding a new
tetrahedron adjacent to a single tetrahedron in the previous triangulation through
a single facet. For this, we choose the new vertex of the triangulation so that only
one facet is red. This implies that all the vertices and edges lie on the convex
hull of the set of vertices of the triangulation.
When n < ne, we can build a triangulation with ne external vertices and
edges using the previous construction. We then add n -Te internal vertices
incrementally. In an incremental step, the new vertex A is added inside an
existing tetrahedron. Let Fi, i = 1, . . ., 4, be the four facets of this tetrahedron
T. In order to make a triangulation of the new set of vertices, we replace T by
four new tetrahedra Ti = conv(Fi, A), as shown in figure 11.5. Each new internal
vertex therefore adds three tetrahedra to the triangulation, so there are exactly
ne - 3 + 3(n - nTe) = n- 3 + 2(n - ne) tetrahedra in the resulting triangulation.
11.3 Representation of complexes, duality

In the preceding section, we have already emphasized the combinatorial and
topological affinities between polytopes and complexes. It is thus natural to use
11.3. Representation of complexes, duality 259
Figure 11.5. Splitting a tetrahedron when adding an internal node.
similar data structures to describe and process these two kinds of objects.
The incidence graph
A d-complex can be encoded by means of its incidence graph, as we did for

polytopes. Recall that two faces of the complex are incident if one is entirely
contained in the other and their dimensions differ by one. The incidence graph
of a complex stores a node for each face of the complex and an arc for each pair
of incident faces.
The adjacency graph
The adjacency graph of a complex has a node for each cell and an edge for
each pair of adjacent cells (meaning that these cells are adjacent to a common
facet). Any internal facet is incident to exactly two cells in the complex, so the
adjacency graph may be built easily from the incidence graph. This definition
is also consistent with the one given for polytopes in section 8.1. Indeed, the
adjacency graph of a polytope, as defined in section 8.1, is exactly the adjacency
graph of the (d - 1)-complex formed by the proper faces of this polytope.
The incidence graph of a simplicial complex can be retrieved from its adjacency
graph in time linear in the number of faces (see exercise 11.3).
Duality
We may also generalize the concept of a duality from polytopes to complexes (see
subsection 7.1.3).
Let C be a d-complex. A d-complex C* is dual to C if there is a bijection
between the faces of C and those of C* which reverses inclusion relationships.
Such a bijection associates the k-faces of C with (d - k)-faces of C*, for any
k = 0, . .. , d (see figure 11.6 for an example).
Figure 11.6. A complex (solid edges) and its dual (dashed edges).
Note that in general the dual of a simplicial complex is not a simplicial complex.
A complex does not have a unique dual. Nevertheless, all the complexes dual
to a given complex C have isomorphic incidence graphs; we say that they are
combinatorially equivalent. Moreover, any complex (C*)* dual to the dual of C is
combinatorially equivalent to C itself.
The adjacency graph of a complex C is also the 1-skeleton of any dual of C. For
this reason, the adjacency graph is also called the dual graph of C.
11.4 Exercises
Exercise 11.1 (Jordan's theorem) A simple curve in the plane is the image in E2 of
the interval [0,1] under a continuous bijection f. The endpoints of the curve are f(0)
and f(1), and the curve is said to link its endpoints. If the mapping f is continuous
and bijective over ]0, 1[, and if f(0) = f(1), then the image f([O, 1]) is called a simple
closed curve. A region 7? in the plane is connected if any two of its points can be linked
by a simple curve entirely contained within R. Jordan's theorem states that if C is a
simple closed curve in E2 , then E 2 \ C has exactly two connected components whose
common boundary is C. This exercise presents a simple proof of Jordan's theorem when
the simple closed curve C is a polygon.
1. Let C be a polygon. Show that E2 \ C has at most two connected components. For
this, consider a disk D such that D n C consists only of a segment. If there are at least
three connected components in E2 \ C, then choose three points Q1, Q2, Q3 in distinct
components. Show that each of these points can be linked to a point of D by a curve
that does not intersect C. Then show that two of the points can be linked by a simple
curve entirely contained within E2 \ C.
11.4. Exercises 261
2. Let Q be a point in E 2 \ C and L any ray extending from Q towards infinity. The
intersection L n C has connected components which are either points or segments. Each
such component S of L n C (of zero length or not) is counted twice if C remains on
the same side of L just before S and just after S (we say that L touches C along S),
otherwise it is counted only once (L goes through C at S). Show that the parity of this
weighted intersection count does not depend on the direction of L, and that it is the
same for all points in the same connected component of E2 \ C. By considering a line L
that intersects C, show that both parities are possible and thus that E2 \ C has exactly
two connected components, whose common boundary is C.
Exercise 11.2 (Jordan's theorem) The notion of a simple closed curve in the plane
is defined in the exercise above, which shows that it encloses a region that is a topological
2-ball, called a disk, and that the complement of this disk is connected. Prove that this
implies that a chord, meaning a simple curve that links two points on the boundary of
the disk and whose relative interior is entirely contained in the interior of the disk and
separates the disk into exactly two distinct connected components.
Hint: Since the complement of the disk is connected, one may join the two endpoints of
the chord by a simple curve that lies in the exterior of the disk. The concatenation of this
curve and of the chord is a simple closed curve, to which one may again apply Jordan's
theorem. One portion of the boundary of the disk lies in the interior, and the other
portion in the exterior. Concatenating the chord to these portions yields two simple
closed curves, to which we can again apply Jordan's theorem. The bounded regions
enclosed by these curves are exactly the two connected components of the disk.
Exercise 11.3 (Incidence graph) Show that the incidence graph of a simplicial d-
complex can be retrieved from its adjacency graph in time linear in the number of faces
of all dimensions.
Hint: Add all the (d -1)-simplices stored in the nodes of the adjacency graph, and for
each pair (F, G) of adjacent facets add a (d - 2)-face incident to F and G. Finally, for
k = d - 3,.. . , 0, add a node for each k-face of the already constructed (k + 1)-faces, and
merge nodes corresponding to identical k-faces, noticing that such nodes descend from a
common (k + 1)-face.
Exercise 11.4 (Planar maps) A graph G is said to be planar if it has a planar em-
bedding: the nodes correspond to points of E2 and the arcs to simple curves linking two
points corresponding to adjacent nodes, such that those curves intersect only at end-
points. The points and simple curves corresponding to the graph for a planar embedding
of the graph, and the induced subdivision of the plane is commonly called a planar map
g. The points are called the vertices of the map, the curves are the edges of the map,
and the connected components of E 2 \ 5 are the 2-faces (sometimes called regions) of the
map.
1. Let n be the number of vertices, m the number of edges, and f the number of
2-faces of a planar map g and let c be the number of connected components of graph G.
Prove Euler's relation:
n-m+f = 1+c.
2. Show that if the planar map has f' 2-faces whose boundary consists of only two
edges, then the number of edges is bounded by
m < 3n - 3 - 3c + f'.
Hint: Proceed by induction, while analyzing how the sum n - m + f - c varies when a
new vertex or a new edge is inserted into the map.
Exercise 11.5 (The Euler characteristic of an unbounded complex) By an un-

bounded complex we mean a simplicial or cell complex whose cells may be unbounded
polytopes (see section 7.3). Show that the Euler characteristic of an unbounded d-
complex whose domain is the whole of Ed is exactly (-l)d.
Exercise 11.6 (Quadratic triangulations) Show that for any pair of integers (n, ne)
such that 4 < ne < n, there exists a triangulation with n vertices, only ne of which are
external, and with t tetrahedra, where
m2 3n
t = 2- - - - n. + 3 -4(n - n)(n - 4).
2 2
Hint: For n = ne, see the proof of lemma 11.2.8. For n, = 4, simply choose n -2 points
A 1,..., A- 2 with respective parameters Ti < ... < n-2 on the moment curve r, and
two points Bo and B,_ 1 such that:
* For i = 2, .. ., n -3, Bo belongs to the half-space bounded by the affine hull of
triangle AjAiAi+l that does not contain any of the points Al,... .An-2
* For i = 1, . . , n - 4, Bn- 1 belongs to the half-space bounded by the affine hull of
triangle AiAi+lAn-2 that does not contain any of the points Al,.. .An-2-
* The interior of the tetrahedron BoA 1A- 2 Bn- contains the points Ai, i = 2,...,
n - 3.
Build a triangulation of vertices Al,-.. , A,,-2 as in the proof of lemma 11.2.8, then add
Bo and Bn-1. For 4 < ne < n, choose ne points on the moment curve, then triangulate
them as in the proof of lemma 11.2.8, and finally add the remaining n -ne points of a
smaller scaled moment curve inside one of the previously built tetrahedra.

The exposition of complexes given in this chapter is voluntarily kept to an elementary
level. The reader further interested in homology may find a good introduction in the
book by Giblin [110].
Jordan's theorem is a fundamental result; it is intuitively very easy, but its proof is
very involved. The simple proof for polygons that is suggested in exercise 11.1 is due to
C. Thomassen [214].
The bounds on 3-triangulations were proved by Edelsbrunner, Preparata, and West [95],
where the solution of exercise 11.6 can also be found.
Chapter 12
Triangulations in dimension 2
To triangulate a set of points A is to build a triangulation whose vertices are

exactly all the points in A and whose domain is the convex hull conv(A). In
dimension 2, if the number of points in A is higher than three and no other as-
sumption is made, the problem admits many solutions. It is therefore legitimate
to try to obtain the best possible triangulations with respect to some additional
criteria. For example, we may want to minimize the total edge-length of the
triangulation, or we may also want to avoid long narrow triangles. For the latter
criterion, the solution is given by the so-called Delaunay triangulations, which are
studied later on (see chapter 17). We may also require that certain given simplices
of non-zero dimension be part of the triangulation. We call the corresponding
triangulation constrained. Obviously, the given simplices must satisfy the inter-
section property, meaning that they intersect only along their common maximal
face, if at all. In dimension 2, constrained triangulations problems always have
solutions and there are efficient algorithms to find them. In the next chapter, we
will see that here lies a fundamental difference from the case of dimension 3: con-
strained triangulation problems do not always have a solution in E3 . The most
frequently encountered constrained triangulation problem is the polygon triangu-
lation problem: the set A of points is the set of vertices of a polygon P, and the
edges of the polygon are constrained to be part of the triangulation. By keeping
only the edges and triangles of the resulting constrained triangulation that lie
in the polygonal region that is the interior of P, we obtain a triangulation of
the polygon 7P, that is, a triangulation whose vertices are exactly those of P and
whose domain is the closed polygonal region interior to 'P.
Section 12.1 determines the complexity of computing a triangulation for a set
of points in E 2, and describes an optimal algorithm. In section 12.2, we prove the
existence of solutions for any constrained triangulation problem in E2. Finally, in
section 12.3, we study the more particular problem of computing a triangulation
of a given polygon. We clarify the complexity of this problem and describe
264 Chapter 12. Triangulationsin dimension 2
an algorithm that actually computes a vertical decomposition of the polygon,

a decomposition into monotone sub-polygons, then a triangulation. There are
many applications of triangulations in dimension 2, some of which can be found
in the exercises at the end of this chapter.
12.1 Triangulation of a set of points

12.1.1 The complexity of computing a triangulation
A triangulation T of a set A of n points is a triangulation (see previous chapter
for the precise definition of a triangulation) such that its vertices are exactly the
points in A and its domain is the convex hull conv(A). The relative interiors of
the simplices of this triangulation form a partition of the convex hull conv(A).
Any triangulation T(A) of A includes the edges of the convex hull conv(A) as
its external edges, and the vertices of conv(A) as its external vertices. Let ne
be the number of extreme points in A (these are the vertices of the convex hull
conv(A)). From corollary 11.2.3, we know that any triangulation T(A) of A has
exactly f triangles and m edges, where
f = 2(n-1)-ne
m = 3(n-1)-ne.
From a triangulation of A, we easily deduce the external edges, external ver-

tices, and the structure of the boundary, and therefore in O(n) operations we
obtain a complete description of the convex hull conv(A). The problem of com-
puting the convex hull of a set A of n points is thus transformable in time O(n)
to the problem of computing a triangulation of the same set A. Since the com-
plexity of computing the convex hull of n points is O9(nlogn) (see section 8.2),
the complexity of any algorithm that computes a triangulation in dimension 2
must be Q(nlogn). Anticipating slightly, we may also mention that there are
algorithms that compute a triangulation of a set of n points in time 0(nlogn),
proving the following theorem:
Theorem 12.1.1 Computing a triangulationof a set of n points in E2 is a prob-

lem whose complexity is E3(nlogn).
12.1.2 An incremental algorithm

The algorithm we present here uses the incremental method to compute the
triangulation of a set of points. This algorithm uses the same scheme as the
incremental algorithm that computes the convex hull of a set of points (see sec-
tion 8.4). The algorithm first sorts the points by increasing lexicographic order on
12.1. Triangulation of a set of points 265
(x, y), and then maintains a triangulation of the current set obtained by adding
the points one by one in that order.
Let A = {Al,..., A,n be a set of n points in the plane. To avoid lengthy
discussions, we assume as usual that the set of points is in general position.
Moreover, we assume that A has already been sorted by increasing lexicographic
order on (x, y), so that Al < A2 < ... < An.
The algorithm not only maintains the triangulation 7i-1 built for the subset
A1 = {A 1 , . . . , Ai-,} of points already processed, but also the boundary of this
triangulation, meaning the boundary of the convex hull conv(Avi-) of the set
Ai-1. The current triangulation is maintained as a data structure that stores the
incidence graph of the triangulation. The convex hull conv(Ai-1) is maintained
using the doubly linked circular list of its vertices, with a pointer p to the vertex
in the list that was last inserted.
In the initial step, we build the triangle formed by the first three points Al,
A2 , A3 and set the list L to {Al, A2 , A3}, with p pointing to the node that stores
A3 .
To describe the current incremental step when Ai is the point to be inserted
in the triangulation, we use the same terminology as that of section 8.3 for in-
cremental convex hulls. An edge F of conv(A4i-) is red with respect to Ai if the
line which is the affine hull of F separates Ai from conv(Ai-1), otherwise the
edge F is blue with respect to Ai. A vertex of conv(Ai-1) is red with respect
to Ai if it is incident to two red edges, blue if it is incident to two blue edges,
and purple it it is incident to both a red and a blue edge. Let us recall that the
vertex Ai-, is necessarily incident to at least one red edge (see phase 1 of the
algorithm described in section 8.3) and that the set of edges on the boundary
of conv(Ai-1) that are red with respect to Ai is also connected (lemma 8.3.3).
Starting at point Ai- 1 , the algorithm traverses the red edges of conv(ASi-),
and for each such edge, adds the triangle conv(F, Ai) to the current triangula-
tion. In L, the sub-list of red edges is replaced by the two edges AiAm and
AiA1 that connect Ai to the two purple vertices Am and Al in conv(Ai-1) (see
figure 12.1).
The analysis of the algorithm is immediate: Each incremental step of the

algorithms takes time proportional to the number of red edges at that step,
which is the same as the number of triangles added to the triangulation. A
triangle that is added to the triangulation at some step remains a triangle in
the subsequent triangulations, and the final triangulation has O(n) triangles.
The time needed to perform the n - 3 incremental steps is therefore O(n).
The initial step can be carried out in constant time. The complexity of the
algorithm is thus dominated by the cost of initially sorting the points, which is
O(n log n).
Figure 12.1. Incremental triangulation.
Theorem 12.1.2 The incremental algorithm described above requires O(nlogn)

operations to triangulate a set of n points in the plane. This is optimal in the
worst case.
Notice that the complexity of the incremental algorithm for a set of n points
is only O(n) if the points are sorted along some known direction.
12.2 Constrained triangulations

The preceding algorithm outputs a certain triangulation of the set of points.
There are many other ways that yield different triangulations, and we may even
ask for a triangulation that contains some given edges. The corresponding prob-
lem is called a constrained triangulationproblem.
Let A be a set of points in the plane. The following theorem shows that any set
of edges that do not intersect except at common endpoints and whose endpoints
are vertices in A can be completed into a triangulation of A. In other words, in
dimension 2, a constrained triangulation problem always has a solution.
Theorem 12.2.1 Let A be a set of points in the plane. Any maximal set of
segments that connect the points in A and have pairwise intersection only at
common endpoints is the set of edges of a triangulation of A, and the converse
is also true.
Proof. Let £ be a maximal set of segments that join the points in A and have
pairwise intersection only at common endpoints. The maximality of £ implies
that no segment may be added to £ while maintaining this property. The edges in
£ must include the edges on the boundary of the convex hull of A, since otherwise
adding any of them would contradict the maximality of £. Since the segments
in £ are all inside the convex hull of A, they determine a decomposition of this
convex hull into polygonal regions. Let us show that any such region must be
12.3. Vertical decompositions and triangulations of a polygon 267
FR
a triangle. Aiming for a contradiction, assume there is a region 5 that is not

triangular, and let P denote the polygon that forms the external boundary of
g. Consider any vertex A that is convex for P, meaning that the internal angle
of the polygon at vertex A is smaller than 7w (see figure 12.2). Note that any
polygon has at least two convex vertices, for instance the vertices of minimal and
maximal abscissae. Let AB and AC be the two edges on the boundary of P that
are incident to A. We claim that triangle ABC contains at least some point of A
in its interior. Indeed, since g is not triangular, either BC is not an edge of 8 or
the triangle ABC includes the boundary of some hole of g in its interior. In the
former case, BC has to intersect some edge in 8, otherwise the maximality of £ is
contradicted, and any edge of 8 that intersects BC must have an endpoint inside
the triangle ABC. Let us choose for X the point closest to A in the direction
parallel to BC, such that there is no point of A in the triangle defined by the
intersection of ABC and the half-plane bounded by a line passing through X and
parallel to BC which contains A. The edge AX has both endpoints in A and
does not intersect another edge of & by construction. Again, £ is not maximal, a
contradiction.
The converse statement simply states that the set £ of edges in a triangulation
is maximal, meaning that any segment having both endpoints in A and not
belonging to 8 must intersect the interior of some edge in 8. This is trivial, since
any segment that does not intersect the segments in E must lie in the interior of
a single triangle, but there cannot be such a segment with endpoints in A since
the interior of any triangle contains no point of A. [1
12.3 Vertical decompositions and triangulations

of a polygon
12.3.1 Lower bound
Let P be a polygon in the plane. From the preceding discussion, we know that
it is possible to compute a triangulation of the set of vertices of P constrained
to include the edges of the polygon 'P. Hence the obtained triangulation has,
besides the edges and vertices of 'P, interior edges and interior triangles that
are contained in the interior of 'P, and occasionally exterior edges and exterior
triangles contained in the exterior of P. The faces of P and the interior faces of the
triangulation form a triangulation whose domain is exactly the interior of P, and
whose boundary is the polygon P. Such a triangulation is called a triangulation
of the polygon P. The following theorem is a straightforward consequence of
theorem 12.2.1.
Theorem 12.3.1 Any polygon P in the plane can be triangulated, in other words
it can be described as the boundary of a 2-triangulationwhose vertices are vertices
of 7P.
If P is a polygon with n vertices, then any triangulation T of P has exactly
n vertices which are external. We can use corollary 11.2.3 to show that such a
triangulation has exactly f = n -2 triangles, m = 2n -3 edges, and n -3 internal
edges. Triangulating a polygon is therefore equivalent to finding the n -3 internal
edges that decompose the polygonal region into n - 2 triangles.
Incidentally, we note the following property:
Lemma 12.3.2 The dual graph of a triangulation of a planarpolygon is a tree.
Proof. This graph is obviously connected. Furthermore, it has no cycle. Indeed,
the existence of a cycle in the dual graph of a triangulation implies either the
existence of a hole in the polygonal region dorn(T), or the presence of an internal
vertex in the triangulation. E
Knowing a simple polygonal line that joins the points of A enables us to com-
pute the convex hull conv(A) in linear time (theorem 9.4.4). So the argument
that proves a lower bound of Q(n log n) on the complexity of computing a trian-
gulation for a set of n points does not apply to the set of vertices of a polygon.
We may legitimately suspect that computing a triangulation of a polygon is a
simpler problem than its counterpart for a set of points.
The complexity of computing a triangulation of a simple polygon remained
elusive for a long time. Classical algorithms only achieved time O(n log n) for the
general problems, while several algorithms were known to perform in linear time
on special kinds of polygons, such as convex, monotone, or star-shaped polygons,
or polygons visible from a single segment. In 1986, a deterministic algorithm
was proposed whose worst-case complexity is o(n log n), proving at least that
O(n log n) was not a tight bound. The problem was settled, at least theoretically,
in 1990 when a linear-time algorithm that computes the triangulation of any
simple polygon in the plane was given. This algorithm is too complex to be
presented in this book, or to be of any practical use. Its existence, however,
provides a proof of the following theorem:
Theorem 12.3.3 The complexity of computing a triangulation of a simple poly-

gon with n vertices in the plane is 1(n).
12.3.2 Triangulating monotone polygons
Classical methods used in computational geometry described in chapter 3, the

sweep and divide-and-conquer methods, both lead to simple and efficient algo-
rithms that triangulate a simple polygon. Here we choose to develop the method
that uses the decomposition of the polygon into trapezoids, which we call the
vertical decomposition of the polygon. This method leads to some of the cur-
rently most efficient deterministic algorithms, and also to the method mentioned
above that works in linear time. It is also through vertical decompositions that
randomization methods appear in triangulation algorithms.
In principle, this method begins by computing a vertical decomposition of the
polygon. From this decomposition, we deduce a decomposition of the interior
polygonal region into monotone polygonal sub-regions. Finally, the algorithm
relies crucially on a method that triangulates a monotone polygon in linear time.
We must begin by defining the monotonicity property of a polygon and how
it can be characterized, and describe an algorithm that triangulates a monotone
polygon in linear time. In the following discussion, we recall the notion of a
vertical decomposition for a polygon, and show how to build a triangulation from
the vertical decomposition in linear time, using a decomposition into monotone
polygons. Finally, we show how to simply compute the vertical decomposition of
a polygon using a sweep algorithm.
Definition and characterization of monotone polygons
A polygonal line C is monotone with respect to a given direction 6 if any line

perpendicular to the direction 6 intersects C in at most one point. In what
follows, the direction 6 is supposed to be that of the x-axis. Thus, a polygonal
line is monotone if it is the graph of a piecewise linear function f(x), or also if
the sequence Pi, P2 , ... Pn of its vertices is ordered by increasing abscissae:
X(Pl) < X(P 2 ) < ... < X(Pn),
or by decreasing abscissae:
X(Pl) > X(P 2 ) > ... > X(Pn).
A polygon is monotone with respect to a given direction if it can be obtained

as the concatenation of two polygonal lines which are monotone with respect to
that direction.
270 Chapter 12. Triangulations in dimension 2
Figure 12.3. Internal and external angles in a polygon.
Before we describe an algorithm that triangulates monotone polygons in linear

time, we first give a lemma that allows the monotonicity of a polygon to be easily
characterized. A few definitions are needed before we can state the lemma.
Let P be a polygon, P be a vertex of 1P, prev(P) and next(P) respectively the
predecessor and successor of P in the direct order along the polygon P. Then the
vertex P is said to be convex if the determinant [prev(P), P, next(P)] is positive,
reflex if this determinant is negative, and flat if the three consecutive vertices
are collinear. The notion of a convex vertex can also be defined in terms of
internal and external angles of the polygon. The internal angle at a vertex P
is the measure cei(P), in [0, 27r[, of the angle between the vectors Pnext(P) and
Pprev(P). The external angle at P is the measure a,(P) = 7r- a(P) of the
complementary angle between the vectors prev(P)P and Pnext(P)).
A vertex is convex (resp. reflex, flat) if the internal angle of the polygon at that
vertex is smaller than (resp. greater than, equal to) 7r.
A vertex P of the polygon P that breaks the monotonicity property in the
x-direction is necessarily a start vertex, that is, a vertex such that
x(P) < x(prev(P)) and x(P) < x(next(P)),
or an end vertex, that is, a vertex such that
x(P) > x(prev(P)) and x(P) > x(next(P)).
A vertex of P which is neither a start nor end vertex is called a monotone

vertex. For instance, points P,, P10 and P12 on the polygon in figure 12.4 are
convex start vertices, while P4 is a reflex start vertex; similarly, P3 and P5 are
convex end vertices, while P1, and P1 5 are reflex end vertices; all other vertices
on this polygon are monotone.
Lemma 12.3.4 A polygon P is monotone in the x-direction if and only if it has

no start or end reflex vertex.
Proof. This condition is necessary: indeed, a monotone polygon has only one
start vertex, which is the vertex with the minimum abscissa, and only one end
vertex, which has maximum abscissa. Both these vertices are convex.
Reciprocally, the following lemma shows that if a polygon has no start or end
reflex vertex, then it has only one start vertex and one end vertex, and it is
therefore monotone. [
Lemma 12.3.5 Consider any polygon P. We denote by c, the number of its

convex start vertices, by ce the number of its convex end vertices, by r, the number
of its reflex start vertices, and by re the number of its reflex end vertices. Then
CS = re + 1,
Ce = rs + 1.
Proof. Let P be a polygon and P1 , P2 , ... , Pn be the circular sequence of its

vertices in counter-clockwise order. The sum of all internal and external angles
of P is n7r, since a pair of internal/external angles at any vertex contributes 7r.
The sum of the internal angles is (n - 2)7r, because any triangulation of P has
n -2 triangles. The sum of the external angles is therefore 27r. Let Uj be the unit
vector directed along the vector prev(Pi)Pi. Consider a point P that follows the
boundary of the polygon in counter-clockwise order and a point U that describes
the unit circle centered at the origin. When P is on the edge prev(Pi)Pi, U is
U,, and when P is at a vertex Pi, then U describes the circular arc UiUi+l. Since
the sum of the external angles is 27r, when P has described all the boundary,
point U has gone through any oriented ray originating at 0 once more in the
counter-clockwise direction than in the clockwise direction. It is easy to see that
U crosses the y-axis counter-clockwise each time P is at a convex end vertex, and
clockwise each time P is at a reflex start vertex. Similarly, it crosses the opposite
(-y)-axis counter-clockwise when P is at a convex start vertex, and clockwise
when P is at a reflex end vertex (see figure 12.4). E
Incidentally, a polygon is said to be convex if its interior region is convex. Note

that a polygon is convex if and only if it has no reflex vertex, and hence if and
only if it is monotone for every direction.
Triangulating a monotone polygon
Let P be a monotone polygon in the direction of the x-axis. Then P is the con-
catenation of two monotone polygonal lines that connect the vertices of minimum
and maximum abscissae. The algorithm begins by sorting the vertices of P by
increasing abscissae, which can be done by merging the vertices of the upper and
P9
P1 0
pi!
7 Ui+ 1
(a)
I
Ui+1
(c) (d)
Figure 12.4. Different kinds of motion on the boundary of a polygon:

(a) convex start vertex (for instance Pi)
(b) convex end vertex (for instance P3 )
(c) reflex start vertex (for instance P4 )
(d) reflex end vertex (for instance Pu,).
lower monotone polygonal lines. Let Qo, Q1,. . ., Q,-i be the resulting ordered
sequence of vertices of P. In the course of the algorithm, the vertices of P are
visited in this order, one by one, and the algorithm adds to the edges of P the
internal edges of the triangulation. Each internal edge added by the algorithm
separates a triangle in the triangulation from the remaining polygon, whose num-
ber of vertices decreases by one at each step. Let Qo, Q1, . . . , Qi-1 be the vertices
already visited by the algorithm before the current step. The algorithm maintains
the following invariants:
1. The remaining polygon is monotone. It is the concatenation of the polyg-

onal line formed by the vertices that have not been visited yet, of abscissa
greater than x(Qi), and of a polygonal line (with increasing abscissae)
{V0, V1, . .. , Vt} whose vertices have already been visited by the algorithm.
2. If t > 1, the vertices {Vi, .. . , Vt- } are reflex vertices in the remaining
polygon.
The algorithm maintains a stack ordered by increasing abscissae, which contains

the vertices of {Vo,...,Vt}. These vertices are the vertices of the remaining
polygon that have already been visited. After the initial sort of the vertices of
P, the first vertices Qo and Q1 are put into a stack. The vertex Qi visited at the
current step is a vertex of the remaining polygon which is adjacent either to Vo
or to Vt (see figure 12.5). Only the last vertex Qn-a can be adjacent to both Vo
and Vt. The current step proceeds according to the following scheme.
1. If Qi is adjacent to Vo but not to Vt in the remaining polygon (see fig-

ure 12.5a), then we add the edges QiV1, QiV2 , ... , QiVt and the triangles
QiV0 V1 , QiV1 V2 , ... , QiVt-lVt to the triangulation. The stack is updated
so that it contains {Vt, Q}.
2. If Qi is adjacent to Vt but not to Vo (see figure 12.5b), as long as the stack
contains two vertices and the vertex Vt on top of the stack is a convex vertex
in the remaining polygon, the edge QiVt- 1 and the triangle QiVtit are
added to the triangulation, and vertex Vt is popped from the stack. Then
Qi is stacked.
3. If the vertex Qi is the last vertex Qn-_ in 7P, then it is adjacent to both Vo
and Vt (see figure 12.5c), and we add the edges QiV 1 ,QiV 2 , ... , Qi½t-l and
the triangles QiVoV1 , Qi~lV 2, ... , QiVt-,Vt to the triangulation.
A segment or a triangle is said to be interiorto a polygon if its relative interior

is contained within the interior of the polygon. It is easy to prove that each edge
and triangle added to the triangulation in the preceding scheme is interior to
the remaining polygon. It is also easy to check that the algorithm maintains the
invariants at each step. For instance, let us consider the first edge QiV1 and the
first triangle T = QiVoVl added in case 1. The vertices V2,..., Vt in the stack
do not belong to the triangle T because they are on one side of VoV, while Qi is
on the other side. This is guaranteed by the fact that the vertices {Vi, . . ., Vt-Il
are reflex in the remaining polygon. The vertices in P that have not yet been
visited do not belong to T either because their abscissae are greater than that of
Qi. Thus the triangle T does not contain any vertex of the remaining polygon,
except for its three vertices Qi, Vo and V1. The edges QiVo and V0o1 are edges
1A.
Il Qi Vt
k
Q-1
(b) (c)
Figure 12.5. Triangulating a monotone polygon:

(a) Qi is adjacent to Vo but not to Vt.
(b) Qi is adjacent to Vt but not to Vo.
(c) Qi is adjacent to Vo and to Vt.
of the remaining polygon, so no edge of this polygon can intersect the interior of
T. nor the edge QiV1. Vo is the vertex of minimum abscissa among the vertices
of the remaining polygon, it is therefore convex and so T and QiV1 are interior
to the remaining polygon.
The analysis of this algorithm is immediate. The initial sort can be performed
in linear time since it consists of merging two already sorted lists. Each step in
the algorithm can be carried out in time proportional to the number of vertices
added to and popped from the stack. Since each vertex is stacked and popped
only once, the algorithm has linear complexity, proving that:
Theorem 12.3.6 It is possible to triangulatea monotone polygon in linear time.
12.3.3 Vertical decomposition and triangulation of a polygon

From the vertical decomposition to the triangulation
The vertical decomposition Dec(P) of a polygon P consists of the vertical decom-

position of the set Sp of segments formed by the edges of the polygon P. This
structure is defined in section 3.3, so we only recall briefly how it is formed. The
vertical decomposition (or decomposition for short) depends on a direction in the
plane, which we assume is the direction of the y-axis. This direction is called
Figure 12.6. The vertical decomposition of a polygon.
vertical and the perpendicular direction is called horizontal. To each vertex P of

the polygon correspond two walls, stemming from P in both directions. These
walls are the maximal vertical segments PP1 and PP2 such that their relative
interiors do not intersect any edge of P. The vertical decomposition Dec(P) of
the polygon is the subdivision of E2 induced by the edges of the polygon and by
the walls stemming from its vertices. The decomposition Dec(P) can be consid-
ered as an (unbounded) cellular complex of size O(n) in the plane. Indeed, the
complex has at most 3n vertices: the vertices of P and the 2n endpoints of the
walls. The cells in the decomposition are trapezoids, with vertical parallel sides,
occasionally degenerated into triangles, infinite trapezoids, or half-planes. A wall
is said to be interior to the polygon if its relative interior is contained in the
interior of the polygon, and exterior in the opposite case. Similarly, a trapezoid
in the decomposition is interiorto the polygon if its relative interior is contained
in the interior of the polygon P, and exterior otherwise.
Each trapezoid in the vertical decomposition Dec(P) has two vertices of P on
its boundary, one on each vertical side. This simple observation allows us to
classify the trapezoids into two categories. Let P and Q be the vertices of P that
are on the boundary of a trapezoid (see figure 12.7):
* The trapezoid is said to be non-diagonal if P and Q are adjacent vertices

on the boundary of the trapezoid.
.0
P
P
F
0- 0
type 1 type2 type2
Figure 12.7. The two types of trapezoid in the vertical decomposition of a polygon.
* The trapezoid is said to be diagonal if P and Q are not adjacent vertices

on the boundary of the trapezoid. This happens when one of P or Q is not
a vertex of the trapezoid, or else if P and Q are opposite vertices of the
trapezoid.
When a trapezoid is diagonal, the relative interior of segment PQ is contained

in the interior of the trapezoid. Let us also call such a segment diagonal. The
set of diagonal segments PQ is a set of segments interior to the polygon P which
do not intersect except at their common endpoints. Once the vertical decompo-
sition of a polygon P has been computed, the trapezoids in the decomposition
may be examined one by one and the diagonal segments may be found in linear
time. The diagonal segment decomposes the polygonal region bounded by P into
polygonal sub-regions. We claim that each of these sub-regions is bounded by a
polygon which is monotone with respect to the horizontal direction. Indeed, such
a polygon cannot have any start or end reflex vertex, because such a vertex in
P is necessarily the endpoint of a diagonal segment, and the claim follows from
lemma 12.3.4. It may even be shown that the boundaries of these sub-regions
are unimonotone, meaning that either the upper monotone chain or the lower
monotone chain consists of a single edge. The total complexity of the resulting
monotone polygonal sub-regions is 0(n), and these regions can be triangulated
in linear time, which shows that:
Theorem 12.3.7 A triangulation of a polygon may be deduced from its vertical

decomposition in additional linear time.
Computing the vertical decomposition of a polygon by sweeping
It remains to show how to build the vertical decomposition of a simple polygon.

In fact, a randomized accelerated algorithm that builds the vertical decomposi-
tion of a polygon with n vertices was given in section 5.4 and runs in expected
time O(n log* n). The algorithm we describe here uses the sweep method, and is
a variant of the algorithm described in subsection 3.2.2 that computes the inter-
section points of a set of segments; it runs in time O(nlogn). We then describe
a lazy version of the same algorithm, whose complexity is lower for a large class
of polygons.
To decompose a polygon using the sweep method, we propose to sweep the
plane with a vertical line A from left to right. The state of the sweep, stored in a
structure Y, is the ordered list of active edges: these are the edges of the polygon
intersected by A, ordered according to their intersections along A. The structure
Y is implemented using a balanced binary tree, letting us insert, delete, or query
active edges in time O(log k) where k is the number of active edges. Moreover,
the nodes of the tree store two extra pointers that allow access in constant time to
the active edge immediately above or below the active edge E stored in this node.
The edge above E is denoted by above(E) and the edge below E by below(E).
The list Y of active edges changes only when A sweeps over a vertex of the
polygon. Thus the list of events to be processed is simply the list of vertices of P
sorted by increasing abscissae. Without loss of generality, we may assume that
no two vertices have the same abscissa.
The structure Y initially stores two fictitious edges that intersect the sweep
line at y = +oo and y =-co respectively. Processing the event corresponding to
a vertex Pi consists of the following operations:
1. Locate Pi in the structure Y according to its ordinate.
2. Create the walls stemming from Pi.
3. Update the structure Y.
Each of these operations depends on the type of vertex Pi: it can be either a
start, an end, or a monotone vertex.
If Pi is a start vertex (see figure 12.8), locating Pi in the list Y allows us to
retrieve the active edges E and E' that lie immediately below or above Pi on A.
The algorithm builds two walls starting at Pi and butting on E and E'. 1 The
edges incident to Pi are inserted in Y between E and E'.
If Pi is a monotone vertex (see figure 12.9), locating Pi in the structure Y allows
us to retrieve the active edge E1 that is incident to Pi. The walls stemming
from Pi and butting on the edges below(E1 ) and above(El) are inserted into the
decomposition, and edge E1 is replaced in Y by the other edge E2 of P that is
incident to Pi.
'Details of these operations depend on the particular representation of the vertical decom-
position used by the algorithm: simple list of walls, simplified or complete representation of
Dec(P) as described in section 3.3.
'- El
E2
I E,
Pi4
I iE
I
"41111111
q '
2
E
Figure 12.8. Vertical decomposition of a polygon by sweeping: a start vertex.
A ..A%12[I
abuveI' ) (LOUveh1~)
I a
F2
El F2E
I.
below(l 1 ) beloaw(E1)
Figure 12.9. Vertical decomposition of a polygon by sweeping: a monotone vertex.
Finally, if Pi is an end vertex (see figure 12.10), locating the vertex Pi in the
structure Y allows us to find both active edges E1 and E2 incident to Pi. Say that
E1 is above E 2 , then the walls stemming from Pi butt on the edges below(E 2 )
and above(El). The two edges E1 and E2 are both removed from the structure
Y.
Theorem 12.3.8 A sweep algorithm builds the vertical decomposition of a poly-

gon with n vertices in time O(nlogn).
Proof. Sorting the vertices of P to build the ordered list of events takes time
O(n log n). The number of active edges is always less than n. For each of the
n events, locating the current vertex in Y and updating the structure Y (which
involves at most two insertions or two deletions) require O(log n) operations, and
the remaining operations can be carried out in constant time. E
A Im DN I I I/
-oove-2) 20ove(PE2)
ED1
JeP El
El "llp"I
uetuwkrti) below(Ei)
Figure 12.10. Vertical decomposition of a polygon by sweeping: an end vertex.
Remark. The algorithm is based on the fact that the edges of the simple polygon
do not intersect except at common vertices. Nowhere do we use explicitly the
fact that the edges are connected. The algorithm can therefore be extended
straightforwardly to compute the vertical decomposition of a collection of several
disjoint polygons, or of polygonal region with holes.
Vertical decomposition of a polygon by lazy sweeping
In its basic version, an algorithm that computes a vertical decomposition using the
sweep method requires time Q(n log n) for any kind of polygon, even if it is convex
or monotone. In those cases, the vertical decomposition can be easily obtained
by other methods in only linear time. The analysis of this algorithm reveals
that its complexity is dominated by the cost of initially sorting the vertices of P,
the cost of locating the vertices in the structure Y, and the cost of rebalancing
the tree after each insertion or deletion of active edges. These operations are not
necessary for each vertex of 7P, however. Indeed, any polygon P can be considered
as the concatenation of monotone polygonal lines, the monotone chains whose
endpoints are the start and end vertices. The internal vertices of such chains are
monotone vertices and are ordered by increasing or decreasing abscissae along
the chain. Also, when the algorithm processes the event at a monotone vertex,
updating the structure Y only involves replacing an edge by another one, which
requires no rebalancing of the structure. If we manage to successively process
the events corresponding to monotone vertices on a chain, we no longer need to
locate these vertices in the structure Y.
In its lazy version, the algorithm only processes the event when the sweep line
A sweeps over start or end vertices. So the list of events is now only the list of
start or end vertices, sorted by their increasing abscissae, which can be obtained
in time O(n + slog s) where s is the number of these special vertices of P.
A p I
Al A
Figure 12.11. Vertical decomposition a polygon using lazy sweeping.
The structure Y is modified so that it can handle updates in a lazy fashion. This
structure is still implemented as a balanced tree, but each node now corresponds
to an active monotone chain, rather than to an active edge: a monotone chain is
called active if one of the edges on this chain is active, meaning that it intersects
the sweep line A.
The active edges or monotone chains subdivide the sweep line A into an ordered
sequence of segments which are alternately interiorand exterior. To simplify the
discussion, we only describe how to build the decomposition of the interior of the
polygon 7P, that is, to find the interior walls and trapezoids in the decomposition
of P. It thus suffices to consider the interior segments on the sweep line A.
Each of these segments is bounded by a pair (Ci, Ci+1) of active chains which are
consecutive along A. To each pair, we dedicate a local sweep line A'. The local
sweep line A' always lags behind A: the edge Ei of Ci intersected by Ai precedes
(in increasing x-order along the chain Ci) the edge Ei intersected by A; similarly,
the edge Ei,+ 1 of Ci+j intersected by Ah precedes the edge Ei+j of Ci+j intersected
by A. The information stored at the nodes of the tree Y corresponding to the
chains Ci and Ci+j is relevant only to the edges Ei and Ei+, intersected by the
local sweep line Ai.
The local sweep line Ai only advances and reaches the global sweep line A
when a node of Y corresponding to one of the chains Ci or Ci+j is visited in
order to locate a vertex P in the list of events. When visiting the node of Y
that corresponds to the chain Ci, the algorithm tests whether the active edge Ei
intersected by the local sweep line Ai intersects the global sweep line A. If not,
the local sweep line sweeps over the vertices of Ci and Ci+j that lie between the
12-4. Exercises 281
two lines Ai and A. A linear traversal of these two chains (analogous to merging
two sorted lists) then builds the wall from each vertex of Ci and Ci+i, in constant
time for each wall. When the local sweep line has caught up with A, the edge E'
of Ci intersected by the local sweep line Ai coincides with the edge Ei intersected
by the global sweep line A. The vertex P may then be compared with Ei and
the location of P in the structure Y may continue.
The process which advances the local sweep line Ai involves merging two sorted
lists of vertices and does not need to perform location queries in the structure Y
or to rebalance this structure. Such a process takes time linear in the number
of processed monotone vertices. If the number of start or end vertices is s, the
number of chains stored in Y is O(s); thus Y requires storage 0(s), and each
location or rebalancing takes time 0(log s). The total complexity of the lazy
sweeping is thus O(n + slog s). As is proved in lemma 12.3.5, for any direction
of the sweep line, the number of start and end vertices is at most r + 2 if r is
the number of reflex vertices in the polygon. The complexity of the lazy sweep
algorithm is thus O(n + rlogr).
It is possible to build the portion of the vertical decomposition that lies outside
the polygon P in a similar fashion. It suffices to maintain a local sweep line for
each segment of the global sweep line that lies outside the polygon. Finally, we
should also mention that the algorithm does not use explicitly the fact that the
edges are connected, so that it computes equally well the vertical decomposition
of a collection of disjoint polygons, or of a polygonal region with holes.
The following theorem summarizes the results of this paragraph:
Theorem 12.3.9 The lazy sweep algorithm computes the vertical decomposition
of a polygon (or of a collection of disjoint polygons) with a total of n vertices, of
which r are reflex vertices, in time O(n + r log r).
Note that if a polygon is monotone in some given direction, then the direction of
the sweep line may be appropriately chosen as perpendicular to this direction to
ensure that the number of start and end vertices for this direction is exactly two,
and the lazy sweep algorithm takes time O(n) in this case.
Theorem 12.3.10 The lazy sweep algorithm computes a vertical decomposition

of a monotone polygon in linear time.
12.4 Exercises
Exercise 12.1 (Decomposition into convex parts) Consider a polygon P with n
vertices and r reflex vertices.
1. Show that any decomposition of the interior of P into convex parts has at least
[r/21 + 1 regions.
2. We present a way to compute, knowing a triangulation of A, a decomposition of

the interior of P into a quasi-minimal number of convex parts. For this, we examine
each internal edge in the triangulation in turn, and remove it if the union of the two
incident convex regions is also convex. Show that the resulting complex has at most
2r + 1 convex cells interior to P, and that this is at most four times the minimal size of
such a decomposition.
3. Construct a polygon that cannot be decomposed into an optimal number of convex
polygonal parts having only vertices on the polygon. Therefore, reaching the optimum
may require points to be added inside the polygon.
Exercise 12.2 (Localization in a planar map) This exercise presents a proof of the
following result: given a planar map of size O(n), it is possible to build in time O(n log n)
a data structure of size O(n) that allows localization queries in the map to be performed
in time 0 (log n).
We first consider a planar triangulation T whose boundary consists of a single triangle.
Let n be the number of vertices of T.
1. Show that T has 3n - 6 edges and 2n - 5 triangles.
2. Consider a maximal set of internal vertices of T such that two vertices are not
adjacent and the degree of any of these vertices is at most d. Show that the size n' of
any such set is at least
n/ > (d- 5)n.
-d(d +i)
3. Show that such a set may be found in linear time.
From the triangulation T, it is possible to build a hierarchical structure that allows
efficient localization queries in the triangulation T. This structure is analogous to that for
3-polytopes which was described in exercise 9.5 and used in exercise 9.6 to answer several
kinds of queries on this polytope. The structure represents a sequence of triangulations
To = TOT,, ,Th,
such that Th has bounded complexity, and Ti+I can be deduced from T1 by removing a
maximal set of non-adjacent internal vertices of 71 with degree at most d. More precisely,
if Si is such a subset of vertices of T%, for each vertex P in Si, the triangles in the star
% (P) of P in <1 are replaced by a triangulation Ti'(P) of the boundary of this star. These
triangulations are merged to obtain l+v, so that
T+r = Ti - U %1(P) + U 7i7(P)

PESi PESi
The underlying data structure is a graph that has a node for each triangle that belongs
to one or several successive triangulations, and an edge for each pair (T', T) of triangles
such that, for some level i e [0, h[, T belongs to 71 but not to fi+i, T' belongs to T1+i
but not to Ti, and the intersection T n T' has a non-empty relative interior.
4. Show that h = O(logn), that 7+1 can be computed from 71 in linear time, and
that the graph described above can be built in time O(n) if the triangulation T is given.
5. Show that the graph can be used to perform localization queries of a point in the
triangles of T in time O(h) = O(logn).
12.4. Exercises 283
6. Extend the method to any planar map. If the map has no unbounded edges, it
can always be enclosed in a surrounding triangle and the resulting polygonal regions can
be triangulated. Otherwise, the regions in the map may be triangulated, by considering
points at infinity to triangulate the unbounded regions.
Exercise 12.3 (Triangulating a star-shaped polygon) A polygon P is called star-

shaped with respect to a point V if, for any point W belonging to the domain dom(P)
of the polygon, the relative interior of the segment VW is contained in the interior of
the polygon P. Devise a simple algorithm that triangulates a star-shaped polygon with
respect to a given point V.
Exercise 12.4 (Triangulating a polygon by sweeping) It is possible to devise a

sweep algorithm that triangulates a polygon without building the vertical decomposi-
tion into trapezoids. This algorithm merges the techniques of the sweep algorithm that
decomposes a polygon as described in subsection 12.3.3 with the algorithm that triangu-
lates a monotone polygon described in subsection 12.3.2.
The plane is swept by a vertical line, and the events are the abscissae of the vertices of
the polygon. The algorithm maintains the list of active edges (edges intersected by the
sweep line). The active edges subdivide the sweep line into segments alternately inside
and outside the polygon P. For each interior segment, the algorithm maintains a chain
V,, V2, .. ., Vk of vertices of the polygon such that V, and Vk are the right endpoints of
the edges that bound the interior segment, and if k > 2 the angle between the vectors
Vi Vi 1 and VV, +1 is greater than 7r for each i = 2, ... , k - 1. Finally, the algorithm
maintains a pointer towards the vertex of maximal abscissa on each chain.
1. Describe the current step of the algorithm.
2. Show that the complexity of the algorithm is O(n log n) if the polygon has n vertices.
3. Outline a lazy version of this algorithm whose complexity is O(n + r log r), where
n is the number of vertices and r the number of reflex vertices.
Exercise 12.5 (Shortest paths) Consider a polygon P with n vertices, and two points
P and Q that belong to the boundary or to the interior of the polygon P. We suppose
that a triangulation T(P) of P has already been computed. Show how to compute in
linear time the shortest polygonal line ir(P, Q) that links P to Q and remains in the
interior of the polygon.
Hint: Let Tp be a triangle of T(P) that contains P and TQ be a triangle of T(P) that
contains Q. The dual graph of T(P) is a tree, in which there is a unique path from Tp to
TQ. Thus, there is a sequence of adjacent triangles in T(P) that links Tp to TQ. Consider
the sequence El, E2 , ... , El of edges adjacent to two consecutive triangles on this path.
Let Ui and Vi be the two vertices of Ei. The algorithm computes the shortest paths
7r(P,Ui) and 7r(P,Vi) for increasing i. To compute 7r(P,Ui+1) and 7r(P,Vi+ 1) knowing
7r(P, Ui) and 7r(P, Vi), we use the following observations:
1. Either Ui = Ui+1 , or Vi = Vi+. It therefore suffices to compute the shortest path
7r(P, X) for the vertex X = Ui+l or X = Vi+l that does not belong to {UM, Vil}.
2. The shortest paths 7r(P, Ui) and 7r(P, Vi) are polygonal lines; they share an initial
polygonal line 7r(P, Ai), and then consist of two concave chains which are the shortest
284 Chapter 12. THangulations in dimension 2
Figure 12.12. The funnel of an edge.
paths 7r(Ai, Ui) and 7r(Ai, Vi) (see figure 12.12). The funnel of the edge UiVi, denoted
by Fp(UiVi), is the concatenation of the chains 7r(Ui, Ai) and 7r(Ai, Vi). The vertex Ai
is the origin of the funnel Fp(UiVi). The shortest path 7r(P, X) must also pass through
the origin Ai, and is the concatenation of 7r(P, Ai) with the shortest path 7r(Ai, X). This
shortest path is the segment AiX if it does not intersect the funnel Fp(UiVi). Otherwise,
there is a point Ai+l on Fp(UiVi) such that Ai+ 1 X is tangent to the chain that contains
Ai+,, and the shortest path 7r(Ai,X) is the concatenation of 7r(Ai,Ai+,) with Ai+ 1X.
To compute 7r(P,X) and the funnel Fp(Ui+ Vi+1), it therefore suffices to find Ai+ .
3. To find the point Ai+,, we follow Fp(UiVi) simultaneously starting at the two
endpoints Ui and Vi. The cost of the traversal is proportional to the number of Fp(UiVi)
visited. But half of the visited vertices of Fp(UiVi) do not belong to Fp(Ui+1 Vi+1 ), so
they will never be visited again. Show then that the algorithm takes linear time.
Exercise 12.6 (Shortest path tree) Consider a polygon P with n vertices and P a
point that belongs either to the boundary or to the interior of P. The shortest paths
that join P to the vertices of P do not cross, so their union forms a tree whose nodes are
vertices of P. Show that it is possible to compute the tree of shortest paths from P in
time O(n log n).
Hint: Let T(P) be a triangulation of P and Tp be a triangle of T(P) that contains P.

The dual graph 5(P) of this triangulation can be viewed as a binary tree rooted at the
node that corresponds to Tp. Each node of 5(P) has an incoming arc and one or two
outgoing arcs corresponding to internal edges in the triangulation T(P). For simplicity,
we make no distinction between the arcs of 9(P) and the edges in T(P). The algorithm
traverses the graph 5(P) in a depth-first fashion. For each node traversed, the algorithm
computes the funnel and the shortest path to the endpoints of the outgoing edges, as in
the previous exercise.
12-4. Exercises 285
The cost for each node is proportional to the number of nodes visited in the funnel
of the incoming edge. If the node has a single child, the number of visited vertices on
the funnel of the incoming edge is proportional to the number of vertices that do not
belong to the funnel of the outgoing edge, vertices that will not be visited later. The
total cost for these nodes is thus linear. If a node has two children and the funnel of its
incoming edge is of size m, the funnels of the outgoing edges have sizes ml + 1 and m2 + 1,
with ml + m2 = m. The cost for such a node is proportional to min(ml, M2 ). For this
exercise, say that the width of a subtree of 5(P) is the sum of the sizes of all the funnels
of the incoming edges of the leaves of this subtree. For each node T of G(P), denote
by m(T) the size of the funnel of the incoming edge, by e(T) the number of arcs of the
subtree rooted at this node, and by m'(T) the width of this subtree. It is easily shown
that m'(T) = m(T) + e(T), so that in particular m(T) < m'(T). From this, show that
the total cost c(m') of the binary nodes of a subtree of width m' satisfies the recurrence
c(m') = max (c(k') + c(m'-k') + O(min(k',m'-k'))
which solves to c(m') = O(m' log m').
Exercise 12.7 (Shortest path queries) Consider a polygon P with n vertices and
P a point that belongs either to the boundary or to the interior of P. Design a data
structure that allows us to find, for any point X on the boundary or in the interior of P,
the shortest path 7r(P, X) that links P to X inside P. Each shortest path query must
be answered in time O(log n + k) where n is the number of vertices of P and k is the
number of edges on the shortest path.
Hint: Build the shortest path tree from P as explained in exercise 12.6. The set of
regions D(E) bounded by an edge E of P and the funnel Fp(E) of this edge forms a
decomposition of the interior of P. Each region 4(E) can be subdivided further by
extending the edges of the funnel Fp(E) all the way until they meet E. Each of the
sub-regions in the decomposition induced by the tree and these extended edges is in fact
triangular, with an edge supported by E, and the opposite vertex is a vertex Q of Fp(E).
For each point X in this triangle, the shortest path 7r(P, X) is the concatenation of the
path in the shortest path tree that links P to Q with the segment QX. The problem is
now replaced by that of locating X in a planar map of size O(n) (see exercise 12.2).
Exercise 12.8 (Visibility polygon) Consider a polygon P with n vertices and P a

point that belongs either to the boundary or to the interior of P. A point X on the
boundary of P or in the interior of P is visible from P if the relative interior of the
segment PX is contained in the interior of P. Compute the polygon V(P, P) which
encloses the set of points visible from P.
Hint: Build the shortest path tree from P and the planar map formed by the edges of
P, the funnel Fp (E) of each edge E of P, and the edges that extend the edges of the
funnels as in the previous exercise. This induces a decomposition of 'P into sub-edges.
Follow the boundary of P, keeping track of whether each sub-edge is visible from P. The
visible edges linked in a proper way form the boundary of V(P, P).
Another solution is to sort the vertices of P by polar angle around P, and to build
the visibility polygon V(P, P) using a sweep ray that originates from P and sweeps the
plane by rotating with an increasing angle.
Exercise 12.9 (Art gallery) An art gallery is viewed as a polygonal region bounded
by a polygon with n vertices. The gallery must be watched over by a set of guards placed
at the vertices of the polygon. Each guard keeps an eye on the portion that is visible
from its location.
Show that it is always sufficient and sometimes necessary to place [n] guards to
completely watch over an art gallery with n vertices.
Hint: Let T be a triangulation of the polygon that encloses the art gallery. It is easy to
show that the vertices can be colored with three colors so that each edge is bichromatic,
or equivalently that each triangle has a vertex of each color. Simply take the color that
is attributed the least number of times, and place a guard at the vertices that have this
color.
To prove that [L] guards may be necessary, consider a comb-shaped polygon.
Exercise 12.10 (Hierarchical decomposition of a polygon) 1. Show that each bi-

nary tree with n nodes has an edge such that its removal creates two trees with at least
Ln+l] nodes each. Show how to compute such an edge in linear time.
2. Apply the result of the previous question to the dual graph G(T) of a triangulation
T of a polygon P. The edge that splits G(T) into two balanced subtrees corresponds to
an internal edge of T that splits P into two sub-polygons P, and P2 each containing at
most a fraction 32 of the triangles in T. By recursively splitting these polygons P, and
P2, we obtain a hierarchical decomposition with O(logn) levels.
Exercise 12.11 (Geodesic decomposition) Consider a polygon P that has n ver-

tices. A balanced geodesic decomposition of P can be obtained using the shortest paths
inside P that join vertices equally spaced on the boundary. More precisely, let P,, P2 ,
... I Pn be the vertices of P. In a first step, compute the shortest paths between P1 and
Pn ~j, between PL and PL ,rand between Pp.nj and P,. In a second step, connect
the pairs (P1 , PL[9 (PL[aIPLn2) etc., and iterate until the pairs consist of adjacent
vertices. The interior of P is then subdivided into regions whose boundaries consist of
three shortest paths joining three vertices of the polygon P. Apart from the first region
that was bounded by the three shortest paths computed in the first step, a region that
appears at step k of the process is bounded by two shortest paths that are computed at
step k and a shortest path that was computed at step k - 1. These shortest paths may
share common edges and the interior of a region (which may occasionally be empty) is a
pseudo-triangle bounded by three concave sub-chains of three shortest paths.
1. Show that all the shortest paths built in the process have exactly n - 3 distinct
edges interior to the polytope P, and that any segment contained inside P intersects only
O(logn) such edges.
2. Show that the subdivision may be built in overall time O(n log n).
Hint: To build the shortest path, first compute a triangulation of P and use the algorithm
of exercise 12.5. Note that the time required to compute a shortest path is proportional
to the number of triangles in T traversed by this shortest path, and that the total number
of intersections between the computed shortest paths and the edges of T is O(n log n).
Exercise 12.12 (Ray shooting) Consider a polygon P with n vertices. A ray shooting
query inside P consists of identifying the point of the boundary of P that is hit by a
ray originating from a given point Q inside or on the boundary of P in the direction of
a given vector U. Show that a balanced geodesic decomposition can be used to answer
such a query in time O(log 2 n).
Hint: To answer a ray shooting query (Q, U) involves locating the origin Q in the
geodesic decomposition, and following the ray (Q, U) in the decomposition. The location
can be performed in time O(logn) if a location structure has been precomputed (see
exercise 12.2). The ray (Q, U) intersects at most O(log n) edges in the geodesic decom-
position and the boundary of each region consists of three concave chains, allowing each
intersection to be located by binary search.

The algorithm that triangulates a monotone polygon in linear time, and the idea of using
a decomposition of a simple polygon into monotone polygons in order to obtain a trian-
gulation are due to Garey, Johnson, Preparata, and Tarjan [109]. In their article, they
obtain the decomposition into monotone polygons using a double sweep whose function
is to regularize the planar map induced by the edges of the polygon. This operation
consists in adding edges to this map so that each vertex of the map, except for the two
vertices with minimum and maximum abscissae, is adjacent to at least one vertex of
greater abscissa and at least one vertex of smaller abscissa. Later, Fournier and Mon-
tuno [107] showed that a decomposition into monotone polygons could be obtained from
a vertical decomposition in linear time. Chazelle and Incerpi [56] devised an algorithm
that computes the vertical decomposition of a polygon more efficiently than the simple
sweep algorithm. The complexity of their algorithm is 0 (n log c), where c is the sinuosity
of the polygon, a parameter that guarantees a certain shape of the polygon and whose
value does not exceed three for polygons usually encountered in the applications. The
first algorithm that showed that the complexity of triangulating a polygon is o(n log n)
was given by Tarjan and Van Wyk [213], and it computes a vertical decomposition of
the polygon in time (rn log log n) using the divide-and-conquer method. This algorithm
uses Jordan sorting and a few sophisticated data structures such as finger trees. Finally,
using vertical decompositions in triangulation methods was one of the key ingredients in
the linear-time triangulation algorithm designed by Chazelle [44].
Randomized algorithms can compute the vertical decomposition of simple polygons in
expected time O(n log* n). The algorithm by Clarkson, Tarjan, and Van Wyk is a nice
example of a randomized divide-and-conquer, but again it uses Jordan sorting. In con-
trast, the incremental randomized algorithm described in section 5.4, due to Seidel [204],
uses only simple data structures.
There are other triangulation algorithms that do not use vertical decompositions. For
instance, Chazelle [41] develops a triangulation algorithm that uses divide-and-conquer
and runs in time 0(n log n). This algorithm finds an internal edge that splits the polygon
into two balanced sub-polygons, and recursively triangulates them (see exercise 12.10).
We must also cite the algorithm by Hertel and Mehlhorn [128] that computes a triangu-
lation directly using the sweep method (see exercise 12.4) and also introduces the lazy
sweep method used in subsection 12.3.3. This method is also described in the book by
Mehlhorn [1621.
The applications of planar triangulations are so many that it is impossible here to
give a complete account. From the standpoint of computational geometry, certainly the
most important one is to provide a preprocessing step to the localization in a triangu-
lar planar map (see exercise 12.2) that was developed by Kirkpatrick [136]. Visibility,
shortest paths, and ray shooting problems tackled in exercises 12.5 to 12.12 also provide
a fertile application domain for triangulations. The algorithm described in exercise 12.5
that computes the shortest path between two vertices of a polygon is due to Lee and
Preparata [148]. In this article, they introduce funnels which are often used by others.
For instance, Guibas et al. [116] used it to compute shortest paths in a polygon from a
vertex or the sub-polygon visible from a given point or segment. Exercises 12.8 to 12.12
are borrowed from this article, but we must point out that the use of finger trees to rep-
resent funnels allows their algorithms to compute the shortest path tree in linear time.
The idea of using the hierarchical decomposition of a polygon (exercise 12.10) to solve ray
shooting problems is exploited by Chazelle and Guibas [54] and by Guibas et al. [116].
The solution that uses a geodesic decomposition of the polygon (see exercises 12.11 and
12.12) was developed by Chazelle et al. [50]. Again, we must point out that the use of
sophisticated data structures (weight-balancedtrees and fractional cascading) allows them
to answer a ray shooting query in time O(log n) for a polygon with n vertices. Finally,
the art gallery theorem (see exercise 12.9) and its numerous variants are discussed in the
book by O'Rourke [183].
Chapter 13
Triangulations in dimension 3
In dimension 3, the possible triangulations of a set of points do not all have the
same number of faces. In fact, there are some sets of points which admit trian-
gulations of both linear and quadratic sizes. Moreover, constrained triangulation
problems do not always have a solution in dimension 3. For instance, some poly-
hedra are not triangulable, meaning that the set of faces of the polyhedron cannot
be completed into a 3-triangulation so that the vertices of the triangulation are
exactly the vertices of the polyhedron. Yet several applications crucially rely on
our ability to decompose polyhedral regions into simplices. We must then design
a simplicial decomposition scheme. The simplicial decomposition of a polyhedral
region is a 3-triangulation whose domain is exactly the polyhedral region (as a
closed topological subset of E3); but this triangulation has additional vertices and
edges that are not faces of the polyhedral region, and the edges and 2-faces of the
polyhedral region may be split into several faces of the simplicial decomposition.
The size of the simplicial decomposition is crucial for subsequent operations, so
we aim at minimizing (exactly or approximately) the size of such decomposi-
tions. In this chapter, we show how to build a simplicial decomposition from
the vertical decomposition. The vertical decomposition of a polyhedral region is
the three-dimensional analog of the vertical decomposition of a polygonal region
introduced in the previous chapter.
Section 13.1 investigates triangulations of a set of points, and presents an algo-
rithm that builds a triangulation of linear size for any set of points such that no
three points are collinear. The remainder of the chapter considers constrained tri-
angulation problems. In section 13.2, we present first two unfeasible constrained
triangulation problems. Section 13.3 generalizes the notion of a vertical decompo-
sition to polyhedral regions and presents an algorithm that computes a simplicial
decomposition for polyhedral regions of genus 0. The resulting simplicial decom-
position is not minimal, but its size can be bounded by O(n+r2 ) for a polyhedron
with n vertices and r reflex edges.
2 2
1 1
3 3
1AB4 1A3C
12B4 A2B4 1 1A34 1AC4 A2BC

1234 < 1B30 1234 4 A23C<AB3C
1BC4 A2C4
(a) (b)
Figure 13.1. Examples of 3-triangulations.
13.1 Triangulation of a set of points
13.1.1 The size of a triangulation
To triangulate a set A of points in E3 involves finding a triangulation whose

vertices are the points of A and whose domain is the convex hull conv(At); the
relative interiors of the simplices in such a triangulation form a partition of the
convex hull conv(A). The main problem arising in dimension 3 is that the size of
a triangulation of a set A of points in E 3 , defined as the total number of simplices
of any dimension, is not entirely determined by the number of points in A and the
number of points of A on the boundary of the convex hull. There is no relation
that binds the numbers of tetrahedra, triangles, and edges of a 3-triangulation of
A as a function of the size of A and the size of the convex hull conv(iA). In fact,
the same set of points may be triangulated in several ways into triangulations of
different sizes. For instance, a set of seven points is shown in figure 13.1 along
with two triangulations, one with four tetrahedra and the other with five.
In section 11.2, it was shown that the size of a 3-triangulation of a set of
n points may vary between a linear bound of n - 3 and a quadratic bound of
n 2 /2 - 5n/2 + 3, and that both bounds may be achieved.
Linear triangulations
Some sets of points admit a triangulation of linear size, which we call linear tri-
angulation for short. In particular, this is true for any set of points in general
position, meaning that no three points are collinear and no four points are copla-
nar. Indeed, let A = {A1,. . . , An} be a set of n points in general position. We
may build a triangulation of A in the following way:
Triangulating the convex hull. Let us first consider the subset formed by the
ne points that are on the boundary of the convex hull conv(A). These points
are vertices of the convex hull and the 2-faces of the convex hull are triangles,
because of the general position assumption. Choose a vertex AO of conv(A),
and compute all the tetrahedra that can be obtained as conv(Ao, F) for
any 2-face F of the convex hull conv(A) that does not contain Ao. These
tetrahedra form a 3-triangulation of conv(A). Theorem 7.2.4 on simplicial
polytopes says that conv(A) has 2ne-4 facets, so if g is the number of facets
that contain vertex Ao, the number of tetrahedra in our 3-triangulation of
conv(A) is 2ne- 4 - g.
Adding the internal vertices. Let us now insert the ni = n e remaining

points of A. Those points belong to the domain of the triangulation built
above. Since A is in general position, each point must be contained in the
interior of some tetrahedron T in the current triangulation. We decompose
this tetrahedron into four smaller tetrahedra that have the newly inserted
point as a vertex, by adding the edges that connect this point to the four
vertices of T.
Each insertion therefore adds exactly three tetrahedra to the triangulation,
so the final triangulation has 2ne- 4 - g + 3ni < 3n - 11 tetrahedra. In
subsection 13.1.3, we will show how this method leads to an algorithm that
computes in time O(n log n) a linear triangulation of a set of points without
three collinear points.
Quadratic triangulations
There are also sets of points that have triangulations with a quadratic size, which
we call quadratic triangulations for short. Some sets of points even have no
triangulation with subquadratic size. Consider for instance the set A of 2n points
drawn in figure 13.2. This set of points has n points A1 , A2 , .. ., An situated (in
this order) on some given line in E3 , and another n points Bl, B2 , .. ., Bn on
another line that does not lie in the same plane as the first line. A triangulation
may be computed as follows: the convex hull of the points is a tetrahedron
Al, A,, B1, B,. Adding the n - 2 points A2 ,..., An- 1 splits this tetrahedron
B1
Al B,
AB
Bi
Bn
A,,
Figure 13.2. A set of points whose only triangulation has quadratic size.
into n - 1 tetrahedra AiAi+iBlBn, i = ,...,n - 1. Adding now the n - 1

points B2 , . . . B 1 on the other line splits each tetrahedron AiAi+iBjB, into
n-I tetrahedra AiAi+ BjBj+l,
1 j = 1,.. ., n-1. The resulting triangulation has
(n- 1)2 tetrahedra, and it is easy to see that this is the only possible triangulation
for this set of points.
Linear and quadratic triangulations

Finally, we show that there are sets of points which admit both linear and
quadratic triangulations. For instance, consider the set of 2n + 1 points ob-
13.1. Strangulation of a set of points 293
tained by adding a point C on the line AnBn in the preceding example. One way
to triangulate it is to consider that C splits the tetrahedron ABAnBn into two
tetrahedra AlBlAnC and ABCBn. Each of these two tetrahedra is then split
into n - 1 tetrahedra, by adding the points in {Ai : i = 2,.. . , n - 1} and in
{Bj : j = 2,.. ., n - 1} respectively into each tetrahedron. The resulting trian-
gulation has 2n - 2 tetrahedra. Another way is to simply add the point C to the
unique triangulation of the set {A 1, A2 , .. ., An, B 1, B 2 , .. , Bn. The addition of
C will merely split the tetrahedron A,- 1 ABn-IB, into two tetrahedra, so that
the resulting triangulation has size (n - 1)2 + 1.
Another example of a set of points that has both linear and quadratic trian-
gulations is a set of n points on the moment curve, for which we have shown the
existence of a quadratic triangulation in section 11.2. Since this point set is in
general position, the discussion above shows that it also admits a triangulation
with a linear number of tetrahedra.
Let A be a set of n points, {A 1 ,. . . , An}, in general position in E3 . In order to
triangulate this set of points, we may think of extending the incremental method
described in subsection 12.1.2 to 1E3. According to this method, the points are
sorted by lexicographic order of their coordinates x, y, z, then inserted one by one
into the triangulation. The triangulation initially consists of a single tetrahedron
formed by the first four points. At each incremental step, the algorithm main-
tains the convex hull conv({Al, A2 , . . , Ai-1}) of the points already inserted and
updates the triangulation by adding the tetrahedra conv(Ai, F) for all the facets
F of conv({Ai, A2 ,..., Ai- 1}) that are red with respect to Ai (recall that these
are the facets whose affine hull separates Ai from {Al, A 2 ,... ,Ai-I).
We used this method in section 11.2 in order to build a triangulation of the
set of points on the moment curve. A major drawback of this method is that, in
dimension 3, it can lead to triangulations with a quadratic number of tetrahedra,
although the set of points admits a linear triangulation. This is exactly what
happens for n points on the moment curve.
In contrast, the algorithm we present here finds a triangulation of a set A of
n points which has linear size if no three points in A lie on the same line. This
algorithm uses the divide-and-conquer method and relies on a theorem proved
in the next subsection. Loosely speaking, the theorem shows the existence of a
good splitter for the triangulation.
13.1.2 The split theorem

Even though we use the theorem below to triangulate a set of points in dimen-
sion 3, the split theorem is true in any dimension d, so in this subsection we
assume that the ambient space is Ed. Let A be a set of n points in general posi-
tion in Ed, whose convex hull is a d-simplex. We can always rename the points
so that the convex hull is the simplex S = AIA 2 ... Ad+1. Any point in A that
is not a vertex of S is an internal vertex of any triangulation of A, and is called
an internal point of A. The set A therefore has n' = n - (d + 1) internal points.
Any internal point X of A splits S = conv(A) into d + 1 simplices:
Si (X) = Al . .. Ai-IXAi+l . .. Ad+l, i = I, .. , d + 1.

An internal point X in A is a A-splitter of A if none of the interiors of the simplices
Si(X) contain more than An' points of A.
Theorem 13.1.1 (Split theorem) Any set of n points in general position in

Ed whose convex hull is a simplex contains a d/(d + 1)-splitter. Such a splitter
may be found in linear time, and the ratio d/(d+1) cannot be improved in general.
Proof. Consider the subset A' of the n' internal points of A. The proof consists
in successively removing the points in A' that are too close to a vertex of S
conv(A). The remaining points will be good splitters for A.
Let us put Qo = A'. We define a sequence of subsets Qo D Ql D ... D Qd+l in
the following way. For each i = 1, . .. , d + 1, let Ni be the normal vector to the
facet of S that does not contain Ai. By convention, Ni points away from S (see
figure 13.3). For each X C Qj-1, we define an i-ordinate si(X) = (X - Ai) - Ni.
Let Yi be the Fd[l1-th point of Qi-l with respect to the increasing order of
si(X). We split Qi-l into two subsets
'Pi = {X E Qi-l si(X) < sj(Yj)},
Pi = {X E Qi-l si(X) < SAW,
Qi = {X E Qi-l: si(X) > si(Yj)} = Qi-l \P i .
Let jPiI denote the size of Pi and IPil the size of Pi. By construction of Qi, we
have
PiI < [+1 < n'
- 1nd+ 1'
Using these inequalities, we can show that Qd+l is not empty, and that any
point Z in Qd+l is a d/(d + 1)-splitter for A. Indeed,
d+1
Qd+I = A'\ U Pi,
i=l
jQd+11 > n'- (d+ 1) n'_= .

Al
A4
A2
A3
Figure 13.3. A splitter.
Pick a point Z that belongs to Qd+1. For each i = 1, .. .,d+ 1, we denote by

Si(Z) the simplex A 1 ... Ai- ZAi+l... Ad+1 and by int(Si(Z)) the interior of this
simplex. Point Z belongs to all the subsets Qi, since Qd+1 is contained in Qd,
which is itself contained in Qd-l, etc. Therefore,
VX E int(si(z)) n A', (x -Ai) -Ni > (Z -Ai) -Ni > (xi-Ai) Ni,
and so we can bound the number of points in Si (Z) by
int(Si(Z))lnA' C A'\Pi,
jint(Si(Z))fnA'j < n'- +1 d+ln.
This proof of existence can be converted into a linear-time algorithm to find a

d/(d + 1)-splitter. Indeed, for i = 1, . .. , d + 1, we need only:
1. Find the normal vector Ni.
2. Compute si(X) for any point X in Qi-j and compute the [d~1 -th point
in Qi- with respect to increasing order of si(X).
3. Compute Qi.
For each value of i, step 1 takes constant time, step 3 takes linear time, and
step 2 requires the computation of n' dot products, which can be carried out
in time O(n'). The only delicate point is to select the d F 1-th value of these
products. This can nevertheless be performed in linear time (see exercise 3.7),
and so step 2 takes linear time as well. The overall cost is thus linear.
Now let us indicate why the ratio d/(d + 1) is optimal for a split theorem. It
suffices to show that there is a set of points that does not admit a A-splitter for
A2
Figure 13.4. A (2/3)-splitter.
A < d/(d + 1). Consider the set of points that is a generalization of the planar set
of points represented in figure 13.3. To build this set in Ed, we put d + 1 points
at the vertices of a regular simplex Al, A 2 , .. ., Ad+1 of center C, and m(d + 1)
internal points given by
Bi'= (2C+(1- ))Ai) i=1,...,d+1, j=l,...,m.
Clearly, the best splatters for this set are the points Bi,1 for i = 1, . .. , d + 1, and
these splatters are d/(d + 1)-splitters.
The algorithm we present here triangulates a set A of n points in E3 , and guar-

antees that the output triangulation has a linear number of vertices if A does
not contain three collinear points. The running time is O(n log n + t) where t
is the size of the output triangulation. The description of the algorithm given
below assumes at first that the set of points is in general position and that its
convex hull is a simplex. We then show how to remove the second assumption,
and finally how to triangulate any set of points.
Triangulating a set of points in general position,

whose convex hull is a simplex
Let us first consider a set A of n points in general position in E3 whose convex

hull is a tetrahedron. The split theorem 13.1.1 suggests the following algorithm
to obtain a linear triangulation of A:
1. Find a (3/4)-splitter for A, say Z, using the algorithm described in the

proof of theorem 13.1.1.
13.1. Triangulationof a set of points 297
2. Split the tetrahedron S = conv (A) into four tetrahedra Si(Z) (i = 1, . .. , 4).
For each of these tetrahedra Si (Z), recursively compute a triangulation for
the set Ai of points in A contained in Si(Z), if there are any.
The existence of a (3/4)-splitter is shown by the split theorem 13.1.1. Each

time a point is chosen as the splitter, four new tetrahedra are added to the
triangulation, and they replace one tetrahedron. The final triangulation therefore
counts 1 + 3(n - 4) = 3n - 11 tetrahedra.
The size of the set to be triangulated decreases by a factor of 3/4 from level to
level of the recursion. The depth of the recursion is thus at most log 3 / 4 n, and
the time needed to process a level of the recursion is linear. So the triangulation
is computed in overall time O(n log n).
Triangulating a set of points in general position
To triangulate a set A of points in general position whose convex hull is not

necessarily a simplex, we preprocess the set as follows:
1. Compute the convex hull conv(A) of A.
2. Triangulate the convex hull. For instance, pick the vertex of maximal degree
(the degree of a vertex is the number of incident edges). The facets of the
convex hull are triangles because of the general position assumption, hence
the collection of simplices of the form conv (Ao, F), where F ranges over
all the facets of conv(A) that do not contain AO, is a triangulation T of
conv (A) .
3. Locate the internal points of A in the triangulation T obtained in step 2.
Once this preprocessing is over, we are left with a collection of sets of points, each
contained in a tetrahedron of T, to which we can apply the algorithm above.
The convex hull conv(A) of a set A of n points in E3 can be computed in
time O(nlogn) (see chapters 8 and 9). The triangulation in step 2 can be ob-
tained in linear time once the convex hull is known. It has exactly 2n - 4 - go
tetrahedra if the vertex AO is incident to go edges of conv(A). To process the
location queries, we use the stereographic projection of the triangulation onto
a plane II to transform these queries into location queries in a triangular pla-
nar map. Let II' be a plane supporting conv(A) along the vertex Ao, and II
a plane parallel to 11' that does not intersect conv(A) but such that conv(A)
is contained in the slab between H and II' (see figure 13.5). The stereographic
projection centered at AO sends any point P in E3 \ {Ao} onto the point of
I that is the intersection of II with the line passing through AO and P. The
set of the projections of facets F of conv(A) that do not contain AO forms a
Ao
Figure 13.5. Stereographic projection onto H.
2-triangulation Tr,. A point X is contained in the tetrahedron conv(Ao, F) of

T if and only if its projection lies in the projection of F onto HI. The convex
hull conv(A) having 0(n) facets, there are 0(n) tetrahedra in T, so there are
also 0(n) triangles in fT 1 , which may be computed straightforwardly in time
0(n). The triangulation T11 may also be preprocessed to support point location
queries. The cost of this preprocessing can be made 0(n) (see exercise 12.2), and
each query takes time 0(logn). The global cost of the location queries is then
0(nlogn).
The complexity of preprocessing A is thus 0 (n log n), and the result is a decom-
position into subproblems of total size 0(n), each of which satisfies the assump-
tions of the preceding algorithm. Therefore, the set of points can be triangulated
in time 0(nlogn), and the triangulation has 2ne -4-90+3ni tetrahedra, if ne is
the number of vertices of conv(A), ni = n - ne is the number of internal vertices,
and go is the number of edges of the convex hull conv(A) that are incident to
AO.
Remark. The split theorem is true for any dimension d, so the preceding algo-
rithm can easily be extended to compute a triangulation of a set A of points in
general position in Ed. The only difficult step is the location of internal points
in the triangulation T of conv(A), for which the above method does not work
in dimension d. For this step we must use another method whose complexity
influences the overall complexity of the algorithm. If the size of the convex hull
is h, the resulting triangulation still has h + 3ni tetrahedra.
13. L Triangulation of a set of points 299
Triangulating any set of points
It now remains to show how to triangulate a set of points which are not in general
position. The general position assumption was used in two ways in the preceding
algorithm:
1. To guarantee that the convex hull conv(A) is a simplicial polytope.

2. To guarantee that each internal point in A that has not yet been chosen as
a splitter lies in the interior of a tetrahedron and never falls on a facet or
edge of the current triangulation.
If the polytope conv (A) is not simplicial, its 2-faces may always be triangulated,
and this triangulation may be used to derive a 3-triangulation of conv(A) as
before.
Then, to triangulate a point set in any position, we ignore in a first phase
the internal points that, during the location steps, fall on a facet or edge of a
tetrahedron in the current triangulation. Unlike the tetrahedra, which may be
split, the edges and triangles created by this algorithm will remain in all the
triangulations formed in the first phase. Each triangle and edge keeps a pointer
to a list that is initially empty, and whenever an internal point is located on a
triangle or edge, it is added to the corresponding list and nothing else is done
for that point until all the points have been processed. At the end of this phase,
the algorithm yields a linear-sized triangulation T' of a subset A' of A. The
tetrahedra in this triangulation do not contain any points of A in their interiors
but triangles and edges may. The points in A \ A' are stored in the lists of the
triangles and edges in whose relative interior they are contained.
In a second phase of the algorithm, the coplanar cases are taken care of. All the
triangles in T' which contain points of A in their relative interior are processed
in turn. For such a triangle F, the set of the mF points that are contained
in the relative interior of F is triangulated within F. The points which lie on
the incident edges are not taken into account yet. In this way, 2 mF + 1 new
triangles are created and the (at most two) tetrahedra adjacent to F are split
into 2mf + 1 tetrahedra each, by lifting these triangles towards the opposite vertex
of the tetrahedron. For each triangle F, the triangulation may be computed in
O(mF log mF) time and the number of tetrahedra increases by at most 2 (2 mF +
1-1) = 4 mF. This phase can thus be carried out in time O(FmFlogmF) =
O(n log n) and yields a triangulation T" with a linear number of tetrahedra.
If A has collinear points, the triangulation T" may still include edges with
points in their relative interiors. In a third phase, the algorithm processes these
edges in turn. For each non-empty edge E, the PE points in A contained in
this edge are sorted along E and each tetrahedron incident to this edge is split
into PE + 1 new tetrahedra. As an edge may be incident to a high number
of tetrahedra (up to O(n)), the triangulation may become quadratic during this
phase. The algorithm takes time O(EEPE10gPE) = O(n log n) to sort the points
along the edges, and each tetrahedron is created in constant time, so that the
overall complexity of the algorithm is 0(n log n + t) if t is the size of the final
triangulation.
The following theorem summarizes the characteristics of this algorithm.
Theorem 13.1.2 A set of n points in E3 may be triangulated in time

O(n log n + t) where t is the number of tetrahedrain the final triangulation. The
algorithm produces a linear triangulation if no three points are collinear.
13.2 Constrained triangulations

Theorem 12.2.1 does not generalize to dimension 3, and constrained triangulation
problems do not always have a solution. A particularly simple example of impos-
sible constrained triangulation is presented in figure 13.6. This example consists
of only three orthogonal segments A1 B1 , A2 B2 , and A 3 B3 whose endpoints are
situated as follows:
Al = (1, 0,0 ), B1 = (-1, 0,0 ),
A2 = (0,1, -e), B2 = (0,-1,-e),
A3 = (-e, -e, 1), B3 = (-e, -e, -1),
provided that e is a small enough constant compared to 1. The six points

Al, B 1, A2 , B2 , A3 , B3 define fifteen edges, and if e is small enough the convex
hull of the six points {A1 , B1 , A2 , B2 , A3 , B3} is an octahedron which has exactly
twelve edges A3 A1 , A 3 A2 , A3 B1 , A3 B2 , B3 A1 , B3 A 2 , B3 B1 , B3 B2 , A1 A2 , A 2 B1 ,
BlB2, and B2 A1 . No triangulation of this point set can include A1 B1 , A2 B2 ,
and A 3 B 3 as edges. Indeed, such a triangulation would have six external vertices
(n = n. = 6), three internal edges, and twelve external edges, for a total of fif-
teen edges (m = 15), with eight external facets (fe = 8). Yet the combinatorial
relations of subsection 11.2.3 show that the number of tetrahedra is given by
t =m - n - ne + 3 = 6,
and that there must be

fi = _(4t - fe) = 8 (13.1)
internal facets. Each internal edge of the triangulation must be incident to at

least three internal facets. Thus, the presence of the three non-coplanar internal
edges A1 Bj, A2 B2, and A 3 B 3 forces the existence of at least nine internal facets.
This contradicts equation 13.1 and shows that such a triangulation does not exist.
13-2. Constrained triangulation 301
13.2. Constrained triangulations 301
LZ
Figure 13.6. An impossible constrained triangulation problem.
C'
A' B'
Figure 13.7. Schbnhart's polyhedron.
It is possible to build a polyhedron, with edges A 1 Bj, A2 B2 , and A3 B3 as

above, that is not triangulable, meaning that no triangulation may have the set
of faces of the polyhedron as a boundary (see exercise 13.1). Another famous
non-triangulable polyhedron is Sch6nhart's polyhedron (see figure 13.7) which
can be built as follows. Let ABC be an equilateral triangle in the xy-plane.
Let A'B'C' be a translated copy of ABC along the z-axis. Sch6nhart's polyhe-
dron is obtained as a deformation of the prism ABCA'B'C' by slightly rotating
the triangle A'B'C' around its axis. The quadrilaterals ABA'B', BCB'C', and
CAC'A' cannot remain planar and the facets of the prism are folded towards the
interior of the prism along the new edges A'B, B'C, and C'A so that edges AB',
BC', and CA' remain outside the polyhedron. This polyhedron cannot be trian-
gulated, for any tetrahedron that is the convex hull of four of its vertices must
include at least one of the edges AB', BC', or CA', and these edges lie outside the
polyhedron.
Practical applications, however, rely crucially on the ability to decompose poly-
hedral regions into simplices. This problem is presented in section 13.3.
13.3 Vertical and simplicial decompositions

As we have just seen, in dimension 3, constrained triangulations problems do
not necessarily have a solution and some polyhedra cannot be triangulated. For
many applications however, it is very important to deal with a decomposition into
tetrahedra or at least into cells of bounded complexity. Thus the problem may
be formulated slightly differently to include so-called Steiner points, which are
not vertices of the polyhedral region but, when added to the constrained trian-
gulation problem, make the decomposition into tetrahedra possible. An optimal
decomposition could therefore be introduced as one having either the smallest
number of tetrahedra, or the smallest number of Steiner points. Unfortunately,
such optimal decompositions are very hard to obtain.
A first method to decompose a polyhedral region is to build a vertical decom-
position of this region. A vertical decomposition of a polyhedral region consists
of a decomposition into vertically cylindrical cells whose horizontal sections are
trapezoids. These cells have constant combinatorial complexity, yet they do not
form a cell complex. From the vertical decomposition, however, it is easy to
construct a 3-triangulation whose domain coincides with the polyhedral region.
Subsection 13.3.1 studies vertical decompositions in greater detail and describes
an algorithm that computes them.
The major drawback in using vertical decompositions is that the number of
simplices in the resulting decomposition may be quadratic even though the poly-
hedral region may be convex and so may admit a linear-sized triangulation. In
the second part of this section, we describe an algorithm which builds a decom-
position of a polyhedron with genus 0 into O(n + r2 ) tetrahedra, where n is the
number of vertices of the polyhedron and r the number of reflex edges: an edge
of a polyhedron is reflex if the dihedral angle between the two incident facets that
lies inside the polyhedron is greater than 7r. The reflex edges are responsible for
the polyhedron being non-convex.
13.3.1 Vertical decomposition of a polyhedral region

The notion of a vertical decomposition in dimension 3 derives from and generalizes
the notion of a vertical decomposition introduced in dimension 2 for segments and
polygons (see sections 3.3 and 5.4). The definition depends on the choice of a
vertical direction, here assumed to be that of the z-axis.
In what follows, the polyhedral region bounded by a polyhedron P is considered
as a topologically closed subset of E 3 . A vertical decomposition is built by drawing
a vertical segment on top of any point P on an edge of P. The segments extend
above and beneath P until they touch P again for the first time. Only the
segment lying in the interior of P will concern us here. A maximal connected
13.3. Vertical and simplicial decompositions 303
I
i
i i
... ....... I
I
Z - Z
------- IX
, -
(a) (b)
Figure 13.8. Vertical decompositions: (a) a cylindrical cell, (b) walls of type 2.
set of vertical segments stemming from points on the same edge and butting on
the same facet of P forms a vertical trapezoid called a 2-wall of type 1. These
2-walls of type 1 decompose the polyhedral region inside P into cylindrical cells
with vertical generator. These cells are called cylindrical cells. They each have
two non-vertical faces, a lower one called the floor and an upper one called the
ceiling. Each floor or ceiling lies in a unique facet of P (see figure 13.8a). The
floor and ceiling of a cell may have arbitrarily high complexity and may not
necessarily be convex or even simply connected. As a consequence, cylindrical
cells may be non-elementary (with an arbitrarily high number of vertical facets),
non-convex (with reflex vertical edges), or even have genus g > 0 (the horizontal
cross-sections have holes).
To obtain convex cells of bounded complexities, we decompose each cylindrical
cell C in turn. For this, we first consider the floor FI(C) of the cell, and we
decompose it. The decomposition of this polygonal region is described in sec-
tion 12.3, if we agree on a direction such as the projection of the y-axis on the
plane that supports Fl(C). The walls are then segments parallel to this direction
stemming from the vertices of Fl(C), contained in Fl(C), and maximal in Fl(C).
Call them 1-walls. On top of these 1-walls, we draw 2-walls as we did before to
construct the cylindrical cells. More precisely, from each point on a 1-wall on the
floor of C, we draw a maximal segment inside C which extends to the ceiling of
C. The set of those maximal vertical segments stemming from a single 1-wall
forms a vertical trapezoid called a 2-wall of type 2 (see figure 13.8b). Note that
the 2-walls of type 2 of a cylindrical cell C decompose the 2-walls of type 1 of
C into 2-walls to which we give type 1'. The 2-walls of type 2 decompose the
cylindrical cell C into cylindrical cells which are both convex and elementary:
each has a floor and a ceiling which is a trapezoid, occasionally degenerated into
304 Chapter 13, Triangulations in dimension 3
a triangle, and with at most four vertical walls which are also trapezoids. These
elementary cells are called prisms, and the set of all prisms forms the vertical
decomposition of the polyhedron P.
Decomposition and simplified decomposition
As described above, the vertical decomposition of a polyhedron P is a simplified

decomposition in the sense of section 3.3. Indeed, each cell C in the cylindrical
decomposition is processed independently, which amounts to considering the 2-
walls of type 1 that bounds these cells as infinitely thin but with two distinct sides.
Each side of a 2-wall of type 1 bounds a unique cylindrical cell, is cut into 2-walls
of type 1' by the 2-walls of type 2 inside this cell, and has no connection with
the 2-walls of type 2 that butt on the other side (see figure 3.2). The simplified
decomposition does not describe the infinitely thin cells included inside 2-walls
of type 1. As a result, it is a cell complex whose cells are elementary but whose
domain does not exactly coincide with the polyhedral region contained within
the polyhedron. In this view, two prisms incident to the two different sides of the
same 2-wall of type 1 are not adjacent.
To obtain a complete description of the polyhedron P, the two sides of a 2-wall
of type 1 are considered one and the same object. The 2-walls of type 2 in one
of the two cells C and C' incident to a 2-wall of type 1 decompose this wall
into 2-walls of type 1' incident to both C and C'. The result is a cell complex
whose domain coincides with the polyhedral region inside P but whose prisms
are not elementary any more: each has a floor and a ceiling, and arbitrarily many
2-walls of type 1'. Nevertheless, these prisms are convex and form the complete
vertical decomposition of P. It is easy to derive a simplicial decomposition by
triangulating each facet of the decomposition and by lifting the incident triangles
of each cell towards one of the vertices of that cell.
Complexity and construction of a vertical decomposition
The complexity of a vertical decomposition (simplified or not) is defined as the

total number of its faces (prisms, edges, facets, and vertices). The complexity of
the simplified decomposition is simply proportional to the number of prisms since
they are elementary. It is also easy to see that the complexity of a complete de-
composition is at most twice that of the corresponding simplified decomposition,
and that the complexity of the simplicial decomposition is also proportional to
the complexity of the simplified decomposition. The following lemma provides a
bound for this complexity.
Theorem 13.3.1 The complexity of the vertical decomposition of a polyhedron

with n vertices is O(n2 ) and can be obtained in time O(n2 log n).
Proof. The complexity of a simplified decomposition is proportional to the

number of prisms of that decomposition which is itself proportional to the number
of 2-walls of type 1 or 2. Indeed, each prism is incident to at least one 2-wall of
type 2 or to a 2-wall of type 1 which has not been split by a 2-wall of type 2, and
either wall is incident to two prisms. The number of 2-walls of type 2 is itself
proportional to the number of 2-walls of type 1 in the cylindrical decomposition.
Indeed, each 2-wall of type 2 is interior to a cylindrical cell and generated by a
1-wall of the decomposition of the floor of that cell. Each vertex on the floor of
a cylindrical cell generates one or two 1-walls in the decomposition of the floor.
The number of 2-walls of type 2 in each cylindrical cell is thus at most twice
the number of vertices on its floor. The number of vertices on the floor of a
cylindrical cell equals the number of edges of the floor and these edges are in
one-to-one correspondence with the 2-walls of type 1 that are incident to the
cylindrical cell. Each 2-wall of type 1 is incident to two cylindrical cells. As a
result, the number of 2-walls of type 2 in the decomposition is at most four times
the number of 2-walls of type 1 in the cylindrical decomposition, and it suffices
to count these walls only.
To count the number of 2-walls of type 1 in the cylindrical decomposition, we
count for each edge E of P the number of walls of type 1 formed by the vertical
segments originating from the points of E. Let HE be the vertical plane passing
through E and let SE be the set of segments formed by the intersections of the
facets of P with HE. The 2-walls of type 1 originating at E are exactly the 2-faces
of the 2-dimensional decomposition of SE that are incident to an edge contained
in E (see figure 13.9).
For each edge E, the size of SE is at most equal to the number of facets of 7P,
that is O(n). Two segments in SE do not intersect except at a common endpoint,
and the vertical decomposition of SE has a linear number of 2-faces. The number
of 2-walls of type 1 stemming from E is thus O(n). The polyhedron P having
O(n) edges, the total number of 2-walls of type 1 and the overall complexity of
the vertical decomposition of P are thus 0(n 2 ).
This analysis suggests the following algorithm to compute the walls of type 1.
For each edge E of P:
1. Compute the set SE of segment intersections of the facets of P with a

vertical plane HE that passes through E.
2. Build the vertical decomposition of SE and keep only the 2-faces of this
decomposition that are incident to a vertical edge that butt on E. (As we
will see in chapter 15, it is also possible to directly compute the 2-faces
of the decomposition of SE incident to E without computing the entire
decomposition of SE.)
306 Chapter 13. Thangulations in dimension 3
Figure 13.9. Walls of type 1 originating at edge E = AB.
The (complete or simplified) vertical decomposition of a polyhedron P is a

cell complex and can be represented by its incidence graph. Starting from the
2-walls of type 1, we first build the incidence graph that corresponds to the
decomposition of P into cylindrical cells. Then each cylindrical cell is decomposed
into elementary prisms by decomposing its floor, which also gives the incidence
graph of the simplified decomposition. It is easy to compute the incidence graph
of the complete decomposition in time linear in the complexity of the vertical
decomposition.
For each edge E of P, the set of segments SE is obtained in time O(n), and
its decomposition computed in time 0(nlogn). The 2-walls of type 1 are thus
computed in time 0(n 2 log n). Each cylindrical cell C is decomposed into prisms
in time 0(nc log n,) if nc is the number of vertices of its floor. The walls of type 2
are thus computed in time O(E, n, log n) = O(n2 log n).
Remark. By considering the vertical segments whose relative interior is con-

tained in the polyhedral region exterior to P, we can also build the vertical
decomposition of the exterior of P. The cylindrical regions formed by the walls
of type 1 can then be unbounded, with no floor and a ceiling, or no ceiling and a
floor, or neither floor nor ceiling. We can transform this problem into the previ-
ous one by adding a large surrounding box: the exterior of the polyhedron now
becomes the interior of the box and our polyhedron is now considered as a hole
in the box. The decomposition of the exterior of P can be obtained by removing

the floors and ceiling that belong to the boundary of the box, and extending
the prisms towards infinity. This shows that there is no real difference between
the interior and exterior of a polyhedron as far as vertical decompositions are
concerned. One may similarly decompose a polyhedral region whose boundary
consists of several polyhedra.
In the rest of this book, we will apply other 3-dimensional decomposition
schemes in other settings, such as the decomposition of a cell of an arrangement
of triangles (see section 16.2).
13.3.2 Simplicial decomposition of a polyhedron of genus 0

The method described in the previous subsection yields a decomposition into
elementary cells (prisms or simplices) of any polyhedral region. The complexity of
this decomposition is only guaranteed to be 0(n 2 ) for a polyhedral region with n
vertices, however, even though this region may be convex or may be triangulated
into 0(n) tetrahedra. The method we present in this subsection proceeds by
trimming the polyhedral region of its protruding parts before decomposing the
resulting polyhedral region. This method can be used for any polyhedral region
whose boundary is a polyhedron with genus 0. It yields a decomposition into
0(n + r2 ) elementary cells if the polyhedron has n edges out of which r are reflex
edges.
A few definitions
A few explanations and definitions are useful to describe the algorithm we present
below. Given a polyhedron P of genus 0, we seek a triangulation T whose do-
main coincides with the polyhedral region interior to P. The triangulation T
is not necessarily a triangulation of 'P: it may have additional vertices, called
Steiner points, which are not vertices of P. Moreover, the boundary bd(T) is a
2-triangulation whose domain coincides with that of P, but the triangulations P
and bd(T) do not necessarily contain the same triangles.
Consider an edge of P. The planes supporting the two incident facets define
a dihedral angle. If this angle is greater than 7r, the edge is called reflex, if it is
smaller than 7r the edge is called convex, and we say it is a flat edge if this angle
equals 7r. Similarly, a vertex is called reflex if it is incident to at least one reflex
edge, flat if the facets that contain it are contained in at most two planes, and
convex if it is neither flat nor reflex.
Consider a polyhedron P without flat vertices and V a convex vertex of P.
Let F be a facet of P that contains V and H(F) the plane that contains this
facet, that is, its affine hull. This plane H bounds two half-spaces. The one that
locally contains the interior of P, meaning that it contains the intersection of the
interior of P with a small enough ball centered at V, is denoted by H+(F). The
intersection of the closed half-spaces H+(F) for all the facets F of P that contain
V is an unbounded polytope called the conel of V.
We denote by K(V) the set of vertices of the polyhedron P which are inside the
cone of V or on its boundary, and by K'(V) the set K(V) \ {V} of these vertices
excluding V.
The cup of V can now be described as the difference of convex hulls
conv(K(V)) \ conv (C'(V)) (see figure 13.10). The cup of V is a polyhedral region
and its boundary is a topological 2-sphere which can be separated into two parts:
* The first part is called the dome of V and is formed by the facets of
conv(K'(V)) which are not facets of conv(KC(V)). These are the facets
of conv(K'(V)) which are red 2 with respect to V.
* The second part is called the lateral boundary of the cup. It is formed
by the facets of conv(K(V)) which are not faces of conv(JC'(V)). All the
facets on the lateral boundary are incident to V and the lateral boundary
is contained in the boundary of the cone of V.
The common boundary between the dome and the lateral boundary is a polygon
(although it may not always be contained in a plane), which we call the crown
of V. The crown of V is formed by the edges of conv(JC'(V)) that are incident
to a single facet of conv(/C'(V)) that is red with respect to V.3 The vertices and
edges of the dome which are not part of the crown are called internal. All these
definitions are illustrated in figure 13.10.
The following properties are best observed now and will be used later on.
1. The cup of V is star-shaped with respect to V. Its interior is contained in
the polyhedral region interior to P, since the segment VW joining V to any point
W inside this cup cannot intersect the polyhedron P in its relative interior.
'Since V is convex, the cone of V can also be described as the convex hull of the rays cast
from V and that contain the edges incident to V, or equivalently as the set of points given by
positive combinations V + i=Zai(Wi - V) where the Wi's are the vertices of P adjacent to V
and al, . ., ark are non-negative reals.
2
A facet F of a polytope C is red with respect to a point A if the hyperplane H that supports
C along F separates A from C: A belongs to the open half-space H- bounded by H which
does not intersect the polytope C (see chapter 8). On the other hand, if A belongs to the open
half-space H+ that contains the interior of C, then this facet F is blue with respect to A.
3
1f the set IC(V) of vertices is in general position, then the facets of conv(/C'(V)) are either
red or blue, and the edges and vertices of the crown are the purple faces of conv(C'(V)) with
respect to V, in other words the edges incident both to a blue and a red facet, and the vertices
incident to purple edges. If the set K(V) is not in general position, then V may belong to the
plane supporting a facet F of conv(K'(V)) which is neither red nor blue.
Figure 13.10. The cup of a convex vertex.
2. The lateral boundary of the cup of V is contained in the domain dom(P) of

P. Indeed, consider a facet Fi of P incident to V, and let Hi be the plane that
contains this facet and Fi' be the union of all the facets of the lateral boundary
that are contained in Hi. Let Ki(V) be the set of those vertices of P that belong
to Hi, and let KC'(V) be the set Ki (V) \ {V}. Then Fi' can be described as the
two-dimensional cup conv(lCi(V)) \ conv(K/'(V)) (see figure 13.11). As a result,
Fi' is star-shaped with respect to V and is contained in the union of the facets of
P contained in Hi.
3. The vertices of the cup of V are vertices of the polyhedron P. The internal
vertices of the dome are reflex vertices of P. The vertices of the crown are also
reflex vertices of P except for the vertices V' which are adjacent to V in P and
such that the edge VV' of P is incident to two non-coplanar facets. From this
reasoning it follows that a convex vertex V' may belong to the dome of another
convex vertex V only if V and V' are adjacent in P.
4. Unlike the vertices, the edges of the dome (internal edges or edges of the
crown) are not necessarily edges of P. The internal edges of the dome of V are
reflex edges of the cup of V, whereas the edges of the crown are convex edges of
the cup of V. The edges of the lateral boundary are edges of P, however, and they
are necessarily convex for the cup when they are incident to two non-coplanar
facets of the cup.
A convex vertex of the polyhedron P is called free if its dome does not include
any internal vertex, nor any internal edge that is an edge of P. Free vertices are
of special interest to us because removing the cup of a free vertex from P does
not create singularities, hence the remaining region is indeed polyhedral.
I----
Figure 13.11. A facet of the cup.
The algorithm
Let P be a polyhedron with n edges of which r are reflex edges. The algorithm
first normalizes the facets of this polyhedron so as to obtain a 2-triangulation
without flat vertices that has the same domain. For this, it suffices to merge
all the connected edges contained in a common line and all the adjacent facets
contained in a common plane into a polygonal region, possibly with holes. This
region can then be triangulated. The resulting polyhedron is 2-triangulated, still
has O(n) edges and the same r reflex edges, but no longer has flat vertices.
The algorithm then proceeds in two phases. In the first phase, or pull-off
phase, we remove the cups of certain free convex vertices with bounded degree.
The cups of these vertices can be triangulated easily and these triangulations
can be added back to the triangulation of the remaining polyhedron to yield a
triangulation of the original polyhedron. The algorithm keeps on pulling off the
cups of those vertices until the size of the remaining polyhedron is O(r). At this
point, the second phase of the algorithm computes the vertical decomposition,
as explained in the previous subsection, and this decomposition is triangulated
into O(r2 ) tetrahedra. So the only missing part is the description of the pull-off
phase, which we present now.
The pull-off phase is an iterative process. Let P, be the current polyhedron.
A set of vertices of P, is independent if its elements are pairwise not adjacent.
Recall that the degree of a vertex is the number of incident edges.The current
step consists of the following operations.
1. Compute a maximal independent set of vertices of 'Pc that are convex, free,
and of degree smaller than some constant g.
2. Remove those vertices and their attached cup from the polyhedron and its
corresponding polyhedral region.
To identify a maximal independent set of convex, free, and bounded-degree

vertices, we begin by putting all these convex, free vertices of degree smaller than
g into a list C. For this, we examine all the convex vertices of degree smaller
than g in turn. Let V be the vertex under examination, and consider the set
'X'(V) of vertices adjacent to V in the 2-triangulated polyhedron and the set
'X(V) = V'H(V) U {V}. We build the polyhedral region conv(Hl (V)) \ convt 1H'(V))
and we look for reflex edges of P, that intersect this region. If there are none, V
is free by definition and is put into C. Its cup is conv(7lt(V)) \ conv(H'(V)) and
has been computed already.
To obtain a maximal independent subset of the elements stored in L, we repeat
the following procedure until L is empty: pick any vertex V in C (for instance
the one stored in the first item of the list), put V into some list L', and remove
from L all the vertices incident to V as well as V.
Pulling off the cups of the selected vertices is done as follows. The cup of V
is triangulated using the O(g) tetrahedra conv(V, F) where F is a facet of the
dome of V. The remaining polyhedron is obtained by replacing the faces of Pc
that are contained in the lateral boundary of the cup of V, by the faces of the
dome of V.
At each step in the pull-off phase, the number of vertices and edges of the
current polyhedron decreases, yet the number of reflex edges is unchanged. For
reasons that will be made clear (see lemma 13.3.9 below), g must be chosen as
an integer greater than 6, so as to allow for another integer t such that t >
11 + 66/(g - 5). Let r be the number of reflex edges of the initial polyhedron
(which is also the number of reflex edges of the current polyhedron) and m the
total number of edges of the current polyhedron Po. The pull-off phase is iterated
as long as m > (1 + t)r.
Analysis
In the following string of lemmas, we prove that if r is the number of reflex edges
of the initial polyhedron P, then after the pull-off phase the resulting polyhedron
has size O(r) with exactly r reflex edges.
Lemma 13.3.2 If V and V' are two convex non-adjacent vertices of a polyhedron
1P, no vertex of P can be an internal vertex of the domes of both V and V'.
Likewise, no edge of P can be an internal edge of the domes of both V and V'.
Proof. Let us first show that, if V and V' are two non-adjacent convex vertices
of 1P, the intersection of their cups has an empty interior and is the intersection of
their domes. For this, we show that the polyhedral region I that is the intersection
of the cup of V with the cup of V' cannot have a vertex that does not belong to
both domes. Since V and V' are not adjacent, neither may qualify as a vertex
of I. Owing to the definition of a cup, the interior of a cup does not contain a
vertex of P, so that any vertex of I is the intersection of an edge E of the cup of
V with a facet F of the cup of V', or the converse. Since the lateral boundary of
a cone is contained in dom(P), however, the facet F and the edge E necessarily
belong to the domes of V and V', which proves our assertion.
Suppose now that a vertex P of the polyhedron P is an internal vertex for
both domes of some non-adjacent convex vertices V and V'. Then P must be a
reflex vertex for the cup of V and a reflex vertex for the cup of V', hence there
must exist a half-ball centered at P contained in the cup of V and a half-ball
centered at P contained in the cup of V' respectively. Since the two interiors of
the cups are non-intersecting, these half-balls have an empty intersection and so
are bounded by the same plane. This shows that the vertex P is a singular face
of the polyhedron P, which is not allowed by the definition of a polyhedron.
For the second part of the lemma, it suffices to place a dummy vertex on the
edge E of P that supposedly is internal to the domes of both V and V'. The
above discussion brings out the contradiction. n
Lemma 13.3.3 A vertex internal to the dome of a convex vertex is incident to

at least three reflex edges of P.
Proof. Let V be a convex vertex, and W be a vertex internal to the dome of

V. Then W is a reflex vertex of the cup of V and there is a half-ball centered at
W entirely contained in the cup of V and thus in the polyhedral region interior
to P as well. The relative interiors of the edges of P that are incident to W are
contained in the open half-space opposite to this half-ball, and because V is not
flat, the vertex W is necessarily a vertex of the convex hull of the set of points
formed by W and by the vertices of P adjacent to W. The edges of this convex
hull that are incident to W are edges of P and must also be reflex. n
Lemma 13.3.4 A reflex vertex of P is an internal vertex to the domes of at

most three convex vertices of 'P.
Proof. Let P be a reflex vertex of P, and assume for a contradiction that it

is an internal vertex of the domes of four convex vertices X, Y, Z, and T. The
four vertices X, Y, Z, and T must be adjacent (lemma 13.3.2) and P must
necessarily be included in the tetrahedron XYZT. Since P is internal to the
domes of X, Y, Z, and T, there is a ball centered at P that is included in each
of the four cones of X, Y, Z, and T. Since P is a vertex of 7P, however, this ball
must also intersect the polyhedral region outside 7P. We may choose a point Q
in this ball and exterior to 7P, such that it belongs to none of the planes passing
through P and through an edge of XYZT (see figure 13.12). The point P is
then interior to one of the four tetrahedra defined by Q and three of the points
in {X, Y, Z, T}, say XYZQ. Then P cannot belong to the dome of the fourth
vertex T, a contradiction. 0
Lemma 13.3.5 For any reflex edge PQ of the polyhedron 1P, there are at most
two convex vertices V and W of P such that P and Q belong to the crowns of V
and W respectively, and such that PQ is an internal edge of the domes of both
V and W.
Proof. Let PQ be a reflex edge of P. Assume for a contradiction that P and

Q belong to the crowns of three convex vertices U, V, and W, and that the edge
PQ is internal to the domes of U, V, and W.
Each of the segments PU, QU, PV, QV, PW, and QW is contained in dom(P),
and the interior of each triangle PQU, PQV, and PQW is contained in the
polyhedral region interior to P. As a result, the three affine half-planes bounded
by the line PQ that contain the triangles PQU, PQV, and PQW, respectively,
are pairwise distinct. Moreover, owing to lemma 13.3.2, U, V, and W are adjacent
vertices in P and each one belongs to the cups of the two others. Since PQ is
an internal edge to the dome of U, there exists a plane passing through PQ that
separates U from V and W. Likewise, there are two planes containing PQ that
separate V from W and U, and W from U and V. Hence the three tetrahedra
PQUV, PQVW, and PQWU have pairwise disjoint interiors and their union
contains a small enough ball centered at the midpoint of PQ. Since PQ is an
edge of 'P, this ball must intersect the polyhedral region exterior to P, and we
may choose a point Z in this ball, exterior to P, that does not belong to any
of the planes containing the triangles PQU, PQV, and PQW respectively (see
figure 13.13). Then PQ necessarily intersects one of the triangles UVZ, VWZ,
or WUZ, say UVZ. But then PQ cannot be an internal edge of the dome of W
because the relative interior of the triangle UVZ lies outside the dome of W. O
Lemma 13.3.6 A reflex vertex of the polyhedron P cannot at the same time
be internal to the dome of a convex vertex and on the crown of another convex
vertex.
Proof. Let P be a reflex vertex of P. If P is internal to the dome of a vertex

of P, then there is a half-ball centered at P contained in the polyhedral region
interior to P. If P is on the crown of a vertex V of P, then there is an internal
edge E of the dome of V that is incident to P. In the plane aff (V, E) the angle
at P internal to the polyhedron is less than 7r. This contradicts the existence of
a half-ball centered at P included in the region interior to 'P. El
Lemma 13.3.7 If P has r reflex edges, then at most 2r convex vertices are not
free.
Proof. The reflex edges of P may be put into one of three categories:
1. those whose vertices include an internal vertex of a dome,
2. those which are internal edges of a dome with both vertices on the crown
of that dome,
3. the others.
These three categories are disjoint by virtue of lemma 13.3.6. If rl, r 2 , r3 are
the respective sizes of these classes, then
rl + r 2 + r 3 = r.
The vertices of the edges in the first category are internal vertices of at most
2ri domes. Indeed, lemma 13.3.4 states that the vertices of an edge in the first
category are internal vertices of at most six domes, but lemma 13.3.3 shows that
each dome will account for at least three edges in the first category. Finally,
lemma 13.3.5 shows that the edges in the second category are internal edges of at
most 2r2 domes. The total number of convex vertices which are not free is thus
at most 2r1 + 2r2 < 2r. a
In what follows, we assume as before that the polyhedron does not have flat
vertices. The following two combinatorial lemmas prove the existence of a large
enough set of convex free vertices of bounded degree.
Lemma 13.3.8 Let P be a polyhedron of genus 0 with n vertices and m edges, r

of which are reflex. The number n'(g) of convex free vertices with degree at most
g of P is at least
n,(g) > m + 2-r2m - r
Proof. Euler's relation 11.2.4 and its corollary 11.2.5 state that
m
n > - +2.
__3
If the polyhedron P has r reflex edges, then at most 2r vertices are reflex and
the number n, of its convex vertices satisfies
m
nc> -+2-2r.
A convex vertex is only incident to convex edges, so the maximum number of

incidences between convex vertices and edges is at most 2(m - r). The number of
convex vertices with degree greater than g is thus at most 2(ml-r) and the number
nc(g) of convex vertices with degree at most g satisfies
nc(g) > 1 + 2-2r- (mlr)

3 g+1
Taking lemma 13.3.7 into account, the number n'(g) of free convex vertices with
degree at most g satisfies
n (g) > - +2-4r- (m-l)

3 g+1
n
Lemma 13.3.9 Let P be a polyhedron with n vertices and m edges, r of which

are reflex. Consider two integers g and t that satisfy g > 6 and t > 11 + 665. If
m > (1 + t) r, then P has at least s(m - r) free convex vertices of degree at most
g, where
t -11 2
3t g+1
Proof. If m > (1 + t) r, then the inequality proved by lemma 13.3.8 becomes
n4(9) > m-r- 11r 2(m-r)

3 3 g+ 1
> (m -r) (t 11 - 2)
Let us now come back to the analysis of the algorithm. Lemma 13.3.9 serves
to prove that each iteration in the pull-off process removes at least a constant
fraction of the vertices of the polyhedron. Indeed, if Pc, the current polyhedron,
has n vertices, m edges, and r reflex edges, we may assume that m > (1 + t)r
(otherwise the pull-off phase is over) and that P has at least s(m - r) free convex
vertices of degree at most g. Here g, t, and s are defined as in lemma 13.3.9. The
number n" of pulled-off vertices is thus at least
n" > + (m-r)
n" > t m
s t 3
> g--n = cn.
The last inequality follows from the fact that 2m > 3n since each vertex is incident
to at least three edges.
Let V be a vertex of P of degree at most g. Whether it is convex, and if so
the region conv(Hl(V)) \ conv(H'(V)), may be computed in constant time, and
its intersection with the reflex edges of P may be tested in time O(r). The set of
free convex vertices with degree at most g can thus be computed in time 0(nr),
and a maximal independent subset can be extracted in time O(n). The cup of
each pulled-off vertex has complexity 0(g), and can be triangulated in constant
time into 0(g) tetrahedra. The polyhedron can also be patched up in constant
time. The number r of reflex edges is constant throughout the algorithm, so
an iteration in the pull-off phase has a complexity of O(r) times the number of
vertices in the current polyhedron.
13-4. Exercises 317
The total complexity of the pull-off phase is

O(nr + (1 - c)nr + (1 - c) 2 nr +*) = 0(nr),
and it generates only O(n) tetrahedra.
After this phase is over, the remaining polyhedron is of size O(r). Lemma 13.3.1
shows that this polyhedron may be decomposed into O(r 2 ) tetrahedra in time
O(r 2 log r). The following theorem summarizes the performances of the algo-
rithm.
Theorem 13.3.10 A polyhedron of genus 0, with n vertices and r reflex edges,

may be decomposed into O(n + r2 ) tetrahedra in time 0(nr + r2 log r).
Remark. The size of the decomposition into tetrahedra produced by this al-
gorithm is optimal in the worst case up to a constant factor. Indeed, there are
polyhedra of size O(r) which cannot be decomposed into fewer than r2 convex
parts (see exercise 13.4).
13.4 Exercises
Exercise 13.1 (A non-triangulable polyhedron) Consider the six points {AI, B1 ,
A2, B2 , A3 , B3 } whose coordinates are given in section 13.2. Show that the eight triangles
AjBjA2 , AjBjA3 , A2 B2A1 , A2 B2 B3 , A3B 3B,, A3 B3 B2 , A1 B2 A3, and A2 B1 B3 define
a polyhedron, and that this polyhedron is not triangulable.
Exercise 13.2 (Pull-off) Show that the pull-off phase of the algorithm described in
section 13.3 may be implemented in time O((n + r 2 ) log r).
Exercise 13.3 (Flat vertices) State explicitly the algorithm that removes the flat ver-
tices of a polyhedron in linear time.
Exercise 13.4 (Chazelle's polyhedron) This polyhedron P can be described as a

rectangular parallelepiped whose top and bottom facets each have n + 1 notches. More
precisely, the polyhedron P is bounded by the six planes x = 0, x = 1, y = 0, y = 1,
z = 0, z = 1. The top facet of P in the plane z = 1 is split by n + 1 notches, each of
which is formed by a reflex edge parallel to the x-axis and incident to two quasi-vertical
facets (parallel to the xz-plane). The bottom facet of P in the plane z = 0 is split by
n + 1 notches, each of which is formed by a reflex edge parallel to the y-axis and incident
to two quasi-vertical facets (parallel to the yz-plane). The reflex edges of the notches on
the bottom facet are contained in the hyperboloid z = xy while the reflex edges of the
notches on the top facet are contained in the hyperboloid z = xy + e.
Show that if e < r, any decomposition of P into convex parts requires at least Q(n2 )
convex parts.
Hint: Show that any convex body contained in P intersects the volume between the
two hyperboloids z = xy and z = xy + e in a region of volume O(E).
318 Chapter 13. Thangulations in dimension 3
31 hpe 3 ragltosi ieso
Figure 13.14. Chazelle's polyhedron.

The algorithm that guarantees a linear triangulation for any set of points in E3 without
collinear points is due to Avis and El-Gindy [19]. In particular, the figures in section 13.1
are borrowed from their article. The split theorem was independently proposed by Avis
and El-Gindy [19] and by Edelsbrunner, Preparata, and West [95].
The decomposition into simplices of a polyhedron of genus 0 given in subsection 13.3.2
is borrowed from Chazelle and Palios [58]. They also give the solution to exercise 13.2.
The polyhedron described in exercise 13.4 was proposed by Chazelle [42]. This polyhe-
dron with O(n) edges and vertices which cannot be decomposed into fewer than Q(n 2 )
convex parts proves the optimality in the worst case of the decomposition described in
subsection 13.3.2. Bajaj and Dey [20] generalized the algorithm by Chazelle and Palios
to polyhedra with holes and handles: if the polyhedron has O(n) edges of which r are re-
flex, their algorithm outputs a decomposition into O(r 2 ) convex parts and their algorithm
runs in time 0(nr + r 2).
Part IV
Arrangements
By the arrangementof a finite set of curves or arcs in the plane, we mean the
decomposition of the plane induced by these curves or arcs. In Ed, we call arrange-
ments the decompositions induced by a finite set of hypersurfaces or portions of
hypersurfaces.
Arrangements play a central role and occur in many different applications. This
part is divided into three chapters. In chapter 14, we are interested primarily in
arrangements of hyperplanes. This interest is spurred mainly by two facts. A set
of points transforms into a set of hyperplanes by polarity, and the arrangement
of the hyperplanes contains several useful pieces of information on the set of
points. Also, arrangements of hyperplanes are particular cases of arrangements
of simplices, so the study of the former kind of arrangements provides interesting
combinatorial bounds for the latter.
The subsequent chapters investigate a few combinatorial and algorithmic no-
tions related to arrangements of line segments in the plane (chapter 15) and
arrangement of triangles in three-dimensional space (chapter 16). The central
problem in both chapters is to bound the combinatorial complexity of several
parts of these arrangements, and in particular to show that they may be of much
smaller complexity than the whole arrangement. Efficient algorithms to com-
pute these portions of arrangements are also sought. These studies are motivated
mainly by two applications: computing views and hidden surface removal in com-
puter graphics, and motion planning in robotics.
Chapter 14
Arrangements of hyperplanes
Hyperplane arrangements are the simplest arrangements one may think of. They
appear naturally in several applications and the results given below, which are of
mostly combinatorial nature, will be useful in the subsequent chapters.
The polarity that maps a set of points to a set of hyperplanes often transforms
a problem about points into a problem about hyperplanes. Many examples are
provided in the exercises. A preprocessing step for these problems often consists
of computing the arrangement of the corresponding set of hyperplanes. This
problem is discussed in section 14.4 which takes advantage of a combinatorial
result known as the zone theorem, given in section 14.3.
An interesting correspondence between hyperplane arrangements and a cer-
tain kind of polytopes, called zonotopes, also sheds more light on problems from
crystallography, architecture, or mixture design (see exercises 14.8 and 14.9).
Section 14.5 introduces the notion of levels in hyperplane arrangements, which
is central to our analysis of higher-order Voronoi diagrams, studied in chapter 17.
14.1 Definitions
Let 7i be a set of n hyperplanes in Ed. The intersection of a finite number of half-
spaces is a bounded or unbounded polytope, and so 7H induces a decomposition
of Ed into a collection of bounded or unbounded polytopes with pairwise disjoint
interiors. These polytopes and their faces form a pure cell complex of dimension
d which we call the d-arrangement of 7-, or more simply the arrangement of 7H if
d is clearly understood. This cell complex is denoted by A(X).
From now on, we often use the notions of a set of hyperplanes in general
position, or of a simple arrangement. A set Al of n hyperplanes is said to lie
in general position if the intersection of any k < d of them is an affine space
of dimension d - k, and if moreover the intersection of any d + 1 of them is
322 Chapter 14. Arrangements of hyperplanes
empty. The cells in the arrangement of a set of hyperplanes in general position

are simple polytopes. For this reason, such an arrangement is also called a simple
arrangement.
Using the polarity introduced in subsection 7.1.3, we may map a set of points to
the set of its polar hyperplanes. Since polarity preserves adjacency relationships,
the set of polar hyperplanes is in general position whenever the set of points is
in general position. Recall that a set of points is in general position if any subset
of k + 1 < d + 1 points generates an affine subspace of dimension k.
14.2 Combinatorial properties

This section establishes the combinatorial results related to arrangements that
will be used below. The proofs given here are direct; other proofs, however, may
be derived from the combinatorial properties of (spherical) polytopes as outlined
in exercise 14.1.
Henceforth, A(H) denotes the d-arrangement of XH, A(?\H) the d-arrangement
of the set 7H \ {H}, A(-t n H) the (d - 1)-arrangement in H of the intersections
of the n - 1 hyperplanes (Ht \ {H}) n H. We denote by nk(Ht) the number of
k-faces in the d-arrangement A(Hl), for k = O,. . . , d, and by nk(n, d) the maximal
number of k-faces in the arrangement of n hyperplanes in Ed.
The following lemma is trivial.
Lemma 14.2.1 Any k-face F in A(Ht \ H) that intersects H gives rise in A(7l)
to a (k - 1)-face, F n H, and to two k-faces, F n H+ and F n H-, where H+
and H- are the two half-spaces bounded by H. For a given H, all the k-faces of
A(-H) that are not faces in A4(H \ H) can be obtained in this way only once.
By counting separately the faces of A(X) that do not intersect H, those that
are contained in H, and those that intersect H but are not contained in H, the
previous lemma yields
no(R-) = no(H \ H) + no(1t n H) (14.1)
nd(7- n H) = 0 (14.2)
nk(R) = nk(Ht \ H) + nk(Hl n H) + nk-l(7H n H). (14.3)
Theorem 14.2.2 (Euler's relation) For any d-arrangement of a set 7' of n

hyperplanes of Ed, we have
d
EZ _l)knk(7H) = (1)d.
k=O
14.2. Combinatorialproperties 323
Proof. Let us begin with the following lemma.

Lemma 14.2.3 The number of k-faces in any simple d-arrangement of d hyper-
planes is
nk(d, d) = 2k ( ) < k < d.
More generally, nk(d, d) is also the number of k -faces containing a given vertex
in any simple d-arrangement.
Proof of the lemma. Consider the d-arrangement A(-t) of a set -t of d hy-
perplanes in general position. A k-face of A(Hi) is contained in the intersection
of d - k hyperplanes in 'H. Conversely, the intersection of a subset 'H' of d - k
hyperplanes of 'H is an affine space I of dimension k, and the restrictions to I
of the k hyperplanes in 'H \ 'H' form in I a k-arrangement, whose k-faces are
precisely the k-faces of A(1t) that are contained in I. Thus, we have
nk (d, d) = ( d) nk(k, k).
So we are left with the slightly simpler problem of counting the k-faces in a
simple k-arrangement of k hyperplanes. Under the general position assumption,
such an arrangement has precisely one vertex, S, at the intersection of the k
hyperplanes, and every face in the arrangement contains this vertex. This can
be shown by induction on the dimension of the faces. If an edge is incident to no
vertex, then it must be a line contained in k - 1 hyperplanes that has an empty
intersection with the k-th hyperplane, and the intersection of the k hyperplanes
is empty, contradicting the general position assumption. Suppose that the j-faces
in the arrangement all have S as a vertex, for j = 1, . . . , i - 1. Then any i-face F
that does not have a vertex cannot have a subface either, since these faces have
a vertex, namely S. Then F is an affine (k - i)-space contained in i hyperplanes
which does not intersect any of the other k - i hyperplanes. This again contradicts
the general position assumption and proves our statement.
By induction on k, we can show that nk(k, k) = 2 k. This is obviously true
for k = 1 and k = 2. Assume by induction that it is true for any k' < k and
consider a hyperplane H in 'H. Using the notation as above, the (k - 1)-faces of
A(1- n H) are contained in exactly two k-faces of A(X) and each k-face contains
only one (k - 1)-face of A(1- n H). Therefore, A(7i) has exactly twice as many
k-faces as A(1H n H) has (k - l)-faces. The first part of the lemma is thus proved
by induction. The second part is immediate if one considers the d hyperplanes
containing the given vertex. 0
The lemma can be used to verify Euler's relation for d hyperplanes:
Z( 1)k2k ( k ) = (1 -2)d = (_1)d.

k=O
This provides the base case for the inductive proof of Euler's relation, the induc-
tion progressing with an increasing number of hyperplanes. Relations 14.1 and
14.2 finish the proof. 1
The following lemma is analogous to lemma 7.1.14 for simple polytopes. It
serves in particular to show the Dehn-Sommerville relations (see exercise 14.2),
as was the case in chapter 7.
Lemma 14.2.4 Any i-face of a simple d-arrangementA(1t) is contained in ex-
actly
(d-i
j-faces in the arrangement, for any 0 < i < j < d.

Proof. The case when i = 0 is proved by lemma 14.2.3 and can be used as the
basis for a proof by induction on i. Let F be an i-face in the d-arrangement A(H).
Choose any hyperplane H that cuts F and that forms a simple arrangement with
the hyperplanes of 'H. Then H n F is an (i - 1)-face in the simple (d - 1)-
arrangement A(H-n H). The induction hypothesis on 'i implies that H n F is
contained in
( d-j)
(j - 1)-faces in A(Hi n H). These (j - 1)-faces are in one-to-one correspondence
with the j-faces of A(Ht) that contain F: indeed, each of these j-faces contains
F n H, hence also F since A9(X) is a complex. This proves the lemma. U
To each i-face F of A(H) (i = 1,... , d) corresponds a position vector defined

in the following way. For each hyperplane H E A, we denote by H+ one of the
two open half-spaces bounded by H and denote the other by H-. The position
vector of a point X with respect to 'H is the vector (£i,... , en) where ei is +1, 0,
or -1 according to whether X G Hi', X E Hi, or X e H,-. The points in a given
face have the same position vector, which is naturally called the position vector
of the face. If a face F is of dimension i, then it is the intersection of exactly
d - i hyperplanes (when the arrangement is simple) and so its position vector has
exactly d - i zero components. The position vectors of the cells that contain F
can be obtained by replacing all the zero components by +1 or -1. Lemma 14.2.4
shows that all possible replacements are realized in the arrangement.
The following theorem gives exact asymptotic bounds for the number of faces
and incidences between faces of an arrangement. These bounds are essential for
algorithms on arrangements.
Theorem 14.2.5 The total number of faces in a simple d-arrangement of n

hyperplanes is e(nd), and so is the number of incidences between faces. If the
arrangement is not simple, both numbers are still O(nd).
14.3. The zone theorem 325
Figure 14.1. A zone in an arrangement of lines.
Proof. For a simple arrangement, the number of k-faces that contain a given
vertex is
n, = 2 k ( ) 0
0< k < d,
as is implied by lemma 14.2.3. But we know that the number of vertices is

exactly ( n ) = 9(nd). Lemma 14.2.4 for j = i + 1 states that the number
of incidences between an i-face and an (i + 1)-face is exactly 2(d - i) times the
number of those faces, proving the theorem for the case of a simple arrangement.
A perturbation argument shows that the number of faces is maximized precisely
for simple arrangements. Indeed, if a subset of k < d + 1 hyperplanes intersect
along an affine subspace of dimension d - k' > d - k, then these hyperplanes can
be perturbed slightly so that the perturbed arrangement is simple. The number
of faces only increases, and so does the number of incidences.[
14.3 The zone theorem

Consider the d-arrangement A(N) of a set 1( of n hyperplanes and let Z be a
hyperplane that does not belong to Xi. The complex formed by the d-faces in
A(1H) that are intersected by Z and their subfaces is called the zone of Z in the
arrangement A(H-) (see figure 14.1).
The zone theorem shows that the complexity (the number of faces) of any zone
in a d-arrangement is an order of magnitude smaller than the whole arrangement.
'This is also true when k = d + 1, with the convention that an affine subspace of negative
dimension is the empty space.
This result is at the core of an optimal algorithm that builds arrangements of

hyperplanes, to be described in the next section.
Henceforth, rather than counting the faces in the zone, we will count the pairs
(F, C) where C is a d-face in the zone and F is a face of C. Such a pair is called a
side in the zone, and if F is a k-face we speak of a k-side. Observe that a face F
in the arrangement generally belongs to several d-faces in the zone. The number
of k-sides of a zone, however, is at most 2 d-k times the number of k-faces in the
zone.
Let us first prove the zone theorem when d = 2.
Lemma 14.3.1 The zone of a line Z in the arrangementA(X) of n lines in the

plane has complexity e)(n) if X- U {Z} is in general position. If not, it is still
0(n).
Proof. Let Z be the line that defines the zone. Without loss of generality, we
may choose a coordinate system such that the x-axis is supported by Z. We
may also assume that the lines in the arrangement and Z are in general position,
meaning that any two lines in this set intersect in exactly one point and no
three lines have a common intersection. A perturbation argument shows that
the complexity of the zone is maximized in this case (as is done in the proof of
theorem 14.2.5).
Let C be a 2-face in the zone of Z, F an edge of C, and (F, C) the corresponding
side of the zone. Let H(F, C) be the half-plane that contains C and is bounded
by the line that is the affine hull of F. The side (F, C) is called a left side if
H(F, C) contains the point (+oo, 0), a right side otherwise. Since no line in X
is parallel to the x-axis, a side may not be simultaneously left and right. We
show that the total number of left sides in a zone is at most 3n, and a symmetric
argument shows the desired result.
The proof goes by induction on the number of lines. The result is trivial for
n = 1. Let H be the line in the arrangement whose intersection with Z has the
greatest abscissa (see figure 14.2). By induction, the total number of left sides in
the 2-faces of the zone of Z in A(N \ H) does not exceed 3n - 3. Let C be the
2-face of this zone that contains the point (+ox, 0). The intersection F of C and
H is a line segment (with endpoints A and B) or a half-line (ending at A). When
adding H, (F, C) becomes a new left side and the left sides that contained A,
resp. B (if it exists), are both cut into two left sides each. Note that because H is
the line in the arrangement whose intersection with Z has the greatest abscissa,
(F, C) is the only left side supported by H in the zone of Z in A(7-). The overall
number of left sides increases by at most 3, proving the above statement and
the lemma. The bound is asymptotically tight whenever XHU {Z} is in general
position. z
Figure 14.2. Inserting H.
The preceding lemma serves as the basis for an inductive argument that proves
the zone theorem in the general case.
Theorem 14.3.2 (Zone theorem) The complexity of any zone in the d-arran-
gement A(N) of n hyperplanes in Ed is e(nd-l) if 'HU {Z} is in generalposition.
If not, this complexity is still O(nd-1).
Proof. The proof is structured as a double induction on the dimension and on

the number of hyperplanes. As in section 14.2, we denote by 'H the set of n
hyperplanes, by H a hyperplane in 'X, by A(- \ H) the d-arrangement of the set
'H \ H, and by A(N n H) the (d - 1)-arrangement in H of the set formed by the
n - 1 hyperplanes {X \ H} n H. We consider a hyperplane Z n H. We denote
by Z(H) the zone of Z in A(7), by Z(H \ H) the zone of Z in A(X \ H), and
by Z(QHnH) the zone of ZnH in A(H nfH).
We assume again that the set comprising Z and the hyperplanes in the arrange-
ment is in general position. A perturbation argument shows that it is indeed when
general position is realized that the complexity of the zone is maximized (see the
proof of theorem 14.2.5).
Let (F, C) be a k-side of Z(- \ H). We distinguish between three cases.
Casel: HnC=@
Then (F, C) gives rise to a single k-side of Z(H), namely (F, C) itself.
Case2: HnC#&Oand HnF=0

Consider the half-space HF bounded by H and that contains F, and let CF =
C n HF. If Z intersects CF, then (F, C) gives rise to a k-side of Z(XH), namely
(F, CF). Otherwise (F, C) does not correspond to a side in Z(-).
Case3: HnC#0 andHnFlF0

Consider the two half-spaces H+ and H- bounded by H, and let F+ = F n H+,
F- = FnH-, C+ = CnH+ and C- = CnH-. If Z intersects C+ but not
C- (resp. C- but not C+), then (F, C) gives rise to a k-side of Z(X), namely
(F+,C+) (resp. (F-,C-)). Otherwise, Z intersects both C+ and C- and (F,C)
gives rise to two sides of Z(H), namely (F+,C+) and (F-, C-). In the latter
case, we note that Z necessarily intersects C+ n C- = C n H. Thus C n H is a
cell of Z(X n H) and (F n H, C n H) is a (k - 1)-side of Z(H n H).
Let us denote by Zk(), Zk(H \ H), and Zk(H n H) the numbers of k-sides of
Z(7-), Z(- \ H), and Z(- n H) respectively, and by Zk(R, H) the number of
k-sides of Z(-) whose k-face is not contained within H. Any k-side of Z(H) that
is not contained within H is either a k-side of Z('1 \ H) that does not intersect
H, or a portion of a k-side of Z(i \ H) that intersects H. From the preceding
discussion, with the convention that zi ('X n H) = 0 if 1 < 0, it follows that
Zk(R, ( n H) for 0 < k < d.

H) < Zk(H \ H) + Zk-lN (14.4)
This result does not depend on the choice of the hyperplane H, so we may carry
out the same analysis for all the hyperplanes H in X, which yields
(n-d + k)Zk(H) < E (Zk(i \ H) + Zk-l( n H)). (14.5)

HER
Indeed, a k-side of Z(Ht) is counted each time H is not one of the hyperplanes
that contains its k-face, and this happens n - (d - k) times.
Let us denote by Zk(n, d) (or by Zk when n and d are clearly understood) the
maximum value of Zk(X) over all d-arrangements of n hyperplanes and all choices
of Z. Inequality 14.5 can be rewritten as
Zk(n, d) < - (Zk(n-1, d) + Zk-l (n-1, d-1)), (14.6)

n - d+ k
and holds for any 0 < k < d and n > d - k.

This recurrence can be solved more easily by introducing the quantities wk(n, d)
defined by
ne
Zk d) 14. ao ca(n, d).
Taking upper bounds, inequality 14.6 above can be rewritten as
Wk(n, d) < Wk(n- 1, d) + Wk-l(n-1, d- 1). (14.7)

Solving by induction on n, we obtain

n-1
Wk(n, d)< Wk(d - k+1,d) + E Wk-i (i, d-l1). (14.8)
i=d-k+l
An induction on d now yields the asymptotic bound on wk(n, d) and thus for
Zk(n, d). The base case is given by lemma 14.3.1 for the case d = 2 which can be
stated as
Zk(n, 2) = 0(n).
Suppose now that d > 2 and that Zk'(n d') = O(nd'-1) for all k' < d' < d,
the constant in the big-oh notation depending on d' and k'. Then wk'(n,d') =
O(nk' -1), and we infer from 14.8 that
n-1
Wk(n,d) <Wk(d-k+1,d)+ Z Q(ik-2).
i=d-k+1
Thus if k > 2, we have wk(n, d) = 0(nk-l) and hence zk(n, d) = 0(nd-l).

Unfortunately, for k = 1, this only yields zi (n, d) = 0(nd-l log n). Thus, let us
now examine the case of O-sides and 1-sides. Each d-face in the zone is a convex
polytope to which we can apply Euler's relation 7.2.1. For each bounded d-face,
the corresponding number nk of k-faces thus satisfies
d
Z(_1)knk = 1,
k=O
and for unbounded k-faces (see exercise 7.15), we have
d
S (-1)knk = 0.
k=O
Summing over all the cells in the zone, and assuming that n > d, we obtain
d
5 (-l)kZ, > 0,
k=O
whether the cell is bounded or not. From this we obtain
d
- zo < S(_1)k Zk = O(n d).
k=2
Each d-face in the zone is a simple polytope, so each vertex of this polytope is
incident to d edges. Therefore zo < 2zj, and so
(1- )Z < z - zo = O(nd1).

This provides the desired bound for 0-sides and 1-sides.

The bound is asymptotically tight when ?iU {Z} is in general position. Indeed,
the simple (d - 1)-arrangement of H fn Z is of complexity Q(nd- 1), and its k-faces
are in one-to-one correspondence with a subset of the (k + 1)-faces of Z(Hl)
(O < k < d).
14.4 Incremental construction of an arrangement

For the incremental construction of an arrangement, we insert the hyperplanes
one by one while maintaining the current arrangement. We assume henceforth
that the arrangement A(Hi) is simple and is represented by the incidence graph
of its 2-skeleton (that is, the set of its 0, 1, and 2-faces). These assumptions can
be removed (see exercises 14.4 and 14.6).
When inserting a new hyperplane Hi, the set of faces in the current arrangement
that are modified or whose incidences are modified is exactly the zone of Hi. The
zone theorem is thus at the core of the analysis of the incremental algorithm.
14.4.1 The case of dimension 2

Consider n lines H 1 , ... , Hn in general position in the plane, whose simple ar-
rangement we want to compute. Let Xi = {Hi,..., Hi-j} be the set of lines
already inserted and 5 the incidence graph of the corresponding arrangement. In
order to insert Hi, we first identify an edge E0 in the current arrangement A(Hi)
that intersects Hi. Pick any edge E in the arrangement. If this edge intersects
Hi, call it E 0. Otherwise, consider the edges in the arrangement A(Hi) that are
supported by the line that is the affine hull of E. Starting at E, we traverse these
edges until we find an edge E0 that intersects Hi. Let C be a cell in the current
arrangement that is incident to Eo; the other such cell is denoted by C'. We find
the other edge of C intersected by Hi (if it exists) by traversing all the edges on
the boundary of C. Let us call this edge E1 .
Let us assume that El exists (otherwise the description is only simpler). We
update g by creating two new vertices Io = E0 n Hi and 11 = E1 n Hi. We insert
the new edge II11 into 0, replace E0 by the two edges formed by the intersections
of E0 with the two half-spaces Hi and Hi7 bounded by Hi, and similarly for E1 .
We also replace C by the two cells C n Hi+ and C n Hi-. Finally, we update the
incidence relationships and the sorted lists of edges on the boundary of C n Htj
and C n Hi-.
If Eo is not the only edge of C intersected by Hi, then we consider the cell
other than C in the current arrangement that has E1 as an edge. From neighbor
to neighbor, we update all the cells in the current arrangement intersected by
14.4. Incremental construction of an arrangement 331
Hi
Figure 14.3. Updating G.
the ray supported by Hi originating at Io. The cells intersected by the opposite
ray are obtained similarly, starting with C' instead of C, and this completes the
update phase for 5. We easily verify that this update operation is carried out in
time proportional to the number of edges in the zone of Hi.
Lemma 14.3.1 therefore implies the following theorem.
Theorem 14.4.1 An arrangement of n lines in the plane may be computed in-
crementally in time O(n 2 ).
14.4.2 The case of dimensions higher than 2

Let A = A(-) be the arrangement of a set 7- of n hyperplanes in Ed. As
mentioned in the introduction to this section, we assume that this arrangement
is simple. The algorithm described below shows how to build the 2-skeleton of
A(X-) knowing the 2-skeleton of the d-arrangement A' = A(1H \ H) where H is
some hyperplane in 'H. We might also maintain the complete incidence graph
of the faces of any dimension, but it turns out to be simpler to compute the
incidence graph of the whole arrangement only after all the 0, 1, and 2-faces have
been computed (see exercise 14.4).
Let F be a k-face of A'. Since A is simple, F cannot be contained in H. F is
said to be active if F fl H $ 0. Note that a O-face cannot be active.
If F is active, F n H is a (k - I)-polytope whose (k - 2)-faces correspond to
the active (k - 1)-faces of F. This fact implies the following lemma.
Lemma 14.4.2 A k-face (k > 1) of A' is active if and only if it is incident to

an active (k - I)-face.
Let 52 be the 2-skeleton of A'. The active faces of g2 and the faces of 52 nH
are in one-to-one correspondence. But g2 n H is the 1-skeleton of the simple
(d - 1)-arrangement formed in H by the intersection of the hyperplanes in Nt
with H, hence 5 2 n H is connected. This implies the following lemma.
Lemma 14.4.3 The sub-graph of the incidence graph induced on the active 1-
faces and 2-faces of A' is connected.
The preceding two lemmas allow us to find all the active faces knowing only the
active edges. The algorithm can now be described. It proceeds in three phases.
We denote by H+ and H- the two half-spaces bounded by H.
Phase 1. Find an edge of A' that cuts H. For this, we start from any edge E
in A'. If E intersects H, we are done, otherwise we traverse the edges on the line
A that is the affine hull of E. Let A be the vertex of E that is the closest to H,
and E' the other edge on A that contains A as a vertex. Replacing E by E' and
iterating eventually leads to the edge on A intersected by H.
Phase 2. Mark the 1 and 2-faces of A' intersected by H as active. This can
be achieved by using a list L of active edges. The list initially contains the edge
found in phase 1. While L is not empty, extract an edge E from it. All its
incident 2-faces that have not yet been marked as active are considered in turn.
Consider such a face C. Using the incidence graph, the edges of C are traversed
until the other edge of C intersected by H is found. If no edge other than E
intersects H, then skip to another 2-face. If such an edge is found, however, then
it is inserted into C and marked as active. Phase 2 is over when the list L is
empty and no other 2-face is to be considered. Lemma 14.4.3 shows that this
traversal identifies all the active 1 and 2-faces.
Phase 3. Replace the active faces of A' and update the incidence relationships
according to lemma 14.2.1. More precisely, let F be an active 2-face of A', E and
E' the two incident active edges (the case where only one active edge is incident
is handled similarly). Denote by H+ and H- the two half-spaces bounded by
H (see figure 14.4). Create two new vertices E n H and E' n H, five new edges
Eo = FnH, E+ = EnH+, E- = EnH-, E'+ = E'nH+, E'- = E'nH-, and
update their incidence relationships. Create the 2-face F n H+ incident to the
edges of F contained in H+, to E+, E'+, and to E0 . Similarly, create the 2-face
F n H- incident to the edges of F contained in H-, to E+, E'+, and to E 0 .
It remains to create the 2-faces, which are intersections of 3-faces of A' and
H. These 3-faces are not represented explicitly in the data structure. It is
easy, however, to reconstruct the 2-faces of A' contained in H, as well as their
14.5. Levels in hyperplane arrangements 333
Figure 14.4. Updating the 2-skeleton of an arrangement.
incidence relationships, starting with the vertices and edges just created in H:
merely observe that these vertices and edges form the 1-skeleton of a (d - 1)-
arrangement (see exercise 14.4).
The analysis of this algorithm is based on the zone theorem. Phase 1 examines
only vertices contained in A and their incident edges. There are only n - d + 1
such vertices, and any vertex is contained in only 2d edges as was proved in lemma
14.2.4. The complexity of phase 1 is thus 0(n).
Phases 2 and 3 require time proportional to the number of faces traversed:
these are the 0-, 1-, and 2-faces of the zone of a hyperplane. The zone theorem
implies that the complexity of these phases is e9(nd-l). We have thus proved the
following theorem.
Theorem 14.4.4 The simple d-arrangement of n hyperplanes may be computed

incrementally in time E3(nd).
14.5 Levels in hyperplane arrangements

14.5.1 Definitions
Consider the arrangement A of n hyperplanes H 1 , . . ., Hn in Ed, assumed to be

simple, and let 0 be any point in Ed that does not belong to any hyperplane in
A. This point is called the reference point.
Let Hi- be the open half-space bounded by Hi that does not contain 0, and
let 7t- be the set of Hi-, i = 1,..., n. A point in Ed is said to be at level k if
it belongs to k half-spaces in X- (see figure 14.5). All the points in the relative
interior of a face are at the same level: this level is by definition called the level of
the face. The level of a face is in general different from the levels of its subfaces.
334 Chapter 14. Arrangements of h7/perplanes
3 2 2
Figure 14.5. Levels of facets in an arrangement of lines. The surface of level 1 is shown in
bold.
In particular, an i-face at level k is the intersection of d - i hyperplanes; hence,

a cell of A that contains F may be at any level between k and k + d - i.
The subcomplex of A formed by the cells at level k and by their faces is called
the k-level of A and is denoted by Ak. Note that a face on the k-level is not
necessarily at level k. The set of faces of A formed by all the faces whose level is
at most k is denoted by A<k; it is a sub-complex of A because the level of a face
never exceeds the level of the faces that contain it.
The faces on the boundary of A<k form a (d - 1)-complex called the surface of
level k. Note that the faces on the k-level are not necessarily on the surface of
level k. In fact, a j-face on the surface of level k has a level I that satisfies
k- (d-j) <1 < k

and conversely, a j-face of level 1 is on the surface of level k for some k that
satisfies
I < k < l+d-j.
The polarity as defined in chapter 7 gives a notion equivalent to levels in the
dual space. Each hyperplane Hi is mapped to a point Hi* and a point M is
mapped to a hyperplane M*. Let 7t* be the set of poles {Hi*}=1 ...... .From the
equivalence
M E Hi- * Hi* M
(see lemma 7.1.9), it follows that M belongs to A<k if and only if M* contains
at most k points of X-*. This dual version of the notion of a level is very useful
as it leads to the notion of a k-set (see exercise 14.18) and allows any result on
levels to be translated into a dual statement in terms of k-sets.
14-5. Levels in hyperplane arrangements 335
14.5.2 Combinatorial properties of levels

A rather direct application of the sampling theorem 4.2.3 bounds the number of
faces in the complex A<k. Let us recall the context and the notation used in
this theorem. Given a set S of n objects, regions each determined by a number
(bounded by b) of objects, and a conflict relationship between objects and regions,
the theorem bounds the number of regions 1F-k(S)J determined by the objects
in S that conflict with at most k objects, as a function of the expected number
fo(Ln/kj , S) of regions defined and without conflict over a random sample in S
of size Ln/kJ.
Consider the arrangement A = A(t) of a set X- of n hyperplanes in Ed. With-
out loss of generality, we suppose that the arrangement is simple, since this case
maximizes the number of faces in A<k. (This can be shown using a perturbation
argument analogous to the one given in the proof of theorem 14.2.5). The objects
are defined to be the hyperplanes, and the regions are d-tuples of hyperplanes.
Thus b = d and the regions defined over ?Y correspond to vertices of A. We say
that a hyperplane H conflicts with a region R if the intersection point of the d
hyperplanes in R belongs to the open half-space H- bounded by H that does
not contain 0.
Regions defined and without conflict over a set of hyperplanes correspond to
vertices of the polytope given as the intersection of the half-spaces that are
bounded by those hyperplanes and contain 0. If a polytope has r facets, how-
ever, the upper bound theorem 7.2.5 states that it has at most O(rLd/2 J) vertices.
Thus fo(Ln/kj ,S) is bounded above by 0((n/k)Ld/ 2 J). The sampling theorem
4.2.3 shows that the number of vertices of A at levels at most k is bounded by
0(nLd/2ikFd/2 1) for 2 < k < nd. For other values of k, see remark 1 after theo-
rem 4.2.3. Lemma 14.2.4 can then be used to extend this result to faces of higher
dimensions.
Theorem 14.5.1 The number of faces whose level is at most k (k > 1) in the
d-arrangement of n hyperplanes is bounded by O(nl d/2]kfd/21).
This bound is tight as is shown by exercise 14.20.
To bound the complexity of a single level is a much more demanding task. A
bound in the planar case is given in exercise 14.19.
14.5.3 Computing the first k levels in an arrangement

In this subsection, we present an on-line algorithm that computes the complex
A<k of the first k levels in the d-arrangement A of a set NH of n hyperplanes in
general position.
The algorithm we present now is an incremental on-line algorithm that uses the
influence graph method. The words on-line refer to the fact that the algorithm
maintains a representation of the arrangement each time a hyperplane is inserted

and that it assumes no preliminary knowledge about the hyperplanes to be in-
serted. This algorithm is deterministic. We give a randomized analysis assuming
that the order of insertion of the hyperplanes is random.
We choose to represent the arrangement using the incidence graph of the 0, 1,
and 2-faces of A<k, and maintain this representation under repeated insertions.
Using this description, it is easy to rebuild the total incidence graph and all the
faces of A<k (see exercise 14.4).
The framework
We wish to use the framework presented in chapter 4 for the construction of

the first k levels. For this, we redefine the problem in terms of objects, regions,
and conflicts in a way that differs slightly from that explained in the previous
subsection.
The objects are hyperplanes, and regions are line segments. A region is defined
over a set X-of hyperplanes if its affine hull is the intersection of d- 1 hyperplanes
of X and if each endpoint of the line segment is the intersection of d hyperplanes
of X. Note that assuming general position implies that each region is defined
by a unique subset of d + 1 hyperplanes. An object and a region conflict if they
intersect. Edges in the arrangement A correspond exactly to regions defined and
without conflict over the set X of hyperplanes.
The algorithm inserts the hyperplanes one by one. At the current step of the
construction, the complex A'k of the first k levels in the arrangement A' of the
current set of hyperplanes is available as the incidence graph of its 0, 1, and
2-faces. Moreover, each edge and each 2-face at level k are marked and a special
pointer K gives access to a face at level k.
We now describe what happens when a new hyperplane H is inserted into this
structure. The algorithm looks for all the edges of A<k that conflict with the
hyperplane H being inserted, and updates the incidence graph. To quickly find
the conflicting edges, the algorithm maintains an influence graph (see section
5.3). Recall that the influence graph is a structure whose goal is to detect the
conflicts between any object not necessarily in the current set and the regions
defined and without conflict over the current set. It is an oriented graph without
cycles that has a node for each region that was defined and without conflict
over the current subset of hyperplanes at some previous incremental step. The
influence domain of a node is the subset of the objects that conflict with the
region corresponding to that node. Arcs in the influence graph connect the nodes
so that the influence domain of any node is contained in the union of the influence
domains of its parents. In the influence graph we use here, there is a node for
each edge at level < k in the current arrangement. We observe that, in contrast
with other randomized algorithms given in this book, the algorithm is interested
in maintaining a subset only of the regions defined and without conflict (the edges
of level at most k).
In a first step, the first d hyperplanes are inserted, the incidence graph of the
0, 1, and 2-faces in the first k levels of their arrangement is computed, and the
corresponding influence graph is initialized accordingly.
In the current step, a new hyperplane H is inserted. The incremental step
consists of a location phase, and of an update phase which can itself be split into
three phases.
Locating. The location phase, using a simple traversal of the influence graph,
identifies all the nodes in the graph that conflict with H, which correspond to
the edges in A'<k intersected by H. If no edge in A'k is intersected by H, then
the hyperplane H does not contain a face in the current complex A<k of the first
k levels, nor in any of the subsequent complexes. So the algorithm may skip to
the next incremental step.
Updating. 1. Creating the new faces. In the location phase, all the edges
of A4 !<k intersected by H have been found. Each such edge is cut by H into two
parts and generates two new edges E n H+ and E n H-, where H+ is the open
half-space bounded by H that contains the reference point 0, and H- is the
opposite half-space (see figure 14.6). Let us say that these new edges are of type
1. Each 2-face F of A'<k intersected by H also generates a new edge, F n H.
Let us say that this new edge is of type 2. We compute such an edge in the
following fashion. Let E be an edge in A<k that is intersected by H, and A its
intersection point with H. Then all the 2-faces incident to E are intersected by
H and generate an edge of type 2. Consider such a 2-face F. Following the part
of the boundary of F that is contained in H-, we can look for the other edge E'
incident to F that is intersected by H. If E' does not exist, this means that the
edge of type 2 created by F is a semi-infinite ray originating at A. Otherwise,
we find an edge E' incident to F and intersecting H at a point B. Then the
edge of type 2 created by F is the segment AB. In either case, the incidence
relationships between these new edges and vertices are easily taken care of.
We must also update the 2-faces of A<k. We have just seen how to find the
2-faces intersected by H. Such a face F is replaced by the two new faces F n H+
and F n H-, which we may again call of type 1, and the incidence relationships
between these 2-faces and the new edges are updated correspondingly. More
tricky is the case of new 2-faces, which we may call of type 2, appearing as the
intersection of a 3-face with H. Even though the 3-faces are not stored explicitly
in the structure, it is not difficult (see exercise 14.4) to reconstruct the 2-faces of
type 2 and their incidence relationships, observing that their boundary is made
E' (type 1)
Figure 14.6. Updating.
up of edges that are contained in H and thus of type 2, and these edges have all
been created and their incidence relationships have been computed (see exercise
14.4).
Finally, the levels of the faces have changed by at most 1. We update the
markers for the edges and for the 2-faces at level k, and collect the created edges
that are at level k + 1 into a list Lk+1 of edges. More precisely, an edge E at level
k in A<k that is intersected by H generates two new edges of type 1 at levels
k and k + 1 respectively. The latter is inserted into Lk+l. Similarly, a 2-face
at level k in A'<k that is intersected by H generates two new faces of type 1, at
levels k and k + 1 respectively. They share a common incident edge (of type 2)
that is at level k and so is marked. Note that the new edges of type 1 obtained
as the intersection of H- and of edges at level k - 1 in A' are also at level k in
A. So are the edges at level k - 1 in A' entirely contained in H-. Nevertheless,
none have been marked yet. These marks will be given in the next phase.
2. Peeling faces at levels greater than k. The faces created in the previous
phase do not all belong to A<k. The peeling process removes all the faces at levels
greater than k from the incidence graph. Of course, this process is not needed if
the current subset of hyperplanes has fewer than k + d - 1 members. The faces to
be removed consist of faces that have been created during this incremental step,
or of faces of A<k that are contained in H-.
As we recall, the algorithm keeps a pointer K to some face at level k in A'<k.
The peeling process traverses the incidence graph of the set Ak+l of constructed
faces at levels greater than k. These faces are necessarily contained in H-. If
Lk+1 is empty and if the face pointed to by K is contained in H-, then Ak+j does
not intersect H, and its incidence graph is connected. The pointer K may be
used as the starting point for a traversal of Ak+l. Otherwise, the incidence graph
of Ak+1 may have several connected components. Nevertheless, each connected
component contains one of the new edges of type 1 stored in the list Lk+l, which
may be used as a starting point for a traversal of this connected component.
During the traversal, we can also mark the faces at level k that have not been
marked yet, since these faces are incident to at least one face of Ak+l1
3. Updating the influence graph. We must also show how to update the
influence graph. A new node is created in the graph for each new edge that
was not removed during the peeling process. A node corresponding to a new
edge of type 1 becomes the child of the node corresponding to the edge of A'k
that contains it. A node corresponding to a new edge of type 2, obtained by
intersecting a 2-face F in A'<k with H, has for parents all the nodes corresponding
to edges of F either contained in H- or intersecting H.
A randomized analysis of the algorithm
Let us first estimate the number of nodes in the influence graph. We first bound
the number of edges created by the algorithm, by bounding the average number
of vertices in the whole arrangement that were created by the algorithm at some
point, then using the fact that a vertex is incident to 2d edges.
Lemma 14.5.2 The probability pj that a vertex at level j in A is created by the

algorithm satisfies
if j < k, pj 1
if j > k, p3 = ( ) < (k+d

j +d
Proof. The proof is trivial if j < k, since the algorithm always creates all the
vertices at levels at most k. Let P be a vertex in the arrangement, let D be
the set of the d hyperplanes that intersect at P, and let C be the set of the j
hyperplanes that determine the level of P, meaning that they separate P from
the origin 0. If j > k, then P is created if and only if, among the hyperplanes
in D U C, the first k + d inserted by the algorithm all belong to D. This happens
exactly with the probability stated above. D
Let us denote by Sj, resp. S<j, the set of vertices at level j, resp. at most j, in
A. The expected number s of vertices created by the algorithm is thus
n-d
s = Elsjlpj
j=o
n-d
= 1SOlpO + Z(ISjI -1S<3j-)P
j=1
n-d-1
- E IS<jl(Pj-Pj+1)+IS<n-dlPn-d-
j=o
Theorem 14.5.1 and lemma 14.5.2 can then be used to bound the expected number
of vertices created by the algorithm, yielding
s = O(n d/2 ) jfd/ 21 pj ( j+ ) +O(nd) (knd\

j=k j+l±d ndJ
from which it follows that
s = krd/210(nLd/ 2 j) for any k 5 0.
Each time a vertex is created, at most 2d incident edges are created. Thus, the ex-
pected number of edges created by the algorithm and hence the expected number
of nodes in the influence graph are both bounded similarly by 0(nld/21krd/2 1).
To bound the number of arcs in the influence graph, we observe that when-
ever the node of a child corresponds to an edge of type 2, the level of the edge
corresponding to the parent node increases by 1. The level of a face cannot be
greater than k + 1 otherwise it is removed from the graph, so that a node in the
graph cannot have more than k children of type 2, and so not more than k + 1
children overall. The total space required to store the influence graph is thus
0(nLd/2J kOd/ 2 1+l).
Updating the incidence graph requires as many elementary operations as there
are arcs in the created incidence graph. This number is at most k times the
number of created edges, since a created edge is traversed at most k times and
has at most k children.
The number of faces traversed during the peeling phase is proportional to the
number of faces at level k + 1, which must have been created in a previous
incremental step.
Updating the influence graph has a complexity proportional to the number of
arcs in the final graph. Hence, the total combined cost of all the update operations
is bounded by the overall number of arcs created, which we have shown to be
0(nld/2i krd/21+1).
It remains to estimate the cost incurred by the location phases. Each node in
the influence graph having at most k + 1 children, the number of nodes visited
during the location phases is bounded by k + 1 times the number of conflicts
detected during these phases. To estimate the expected number of conflicts de-
tected by the algorithm, we first compute the probability Pji that an edge with
14-5. Levels in hyperplane arrangements 341
endpoints at levels j and j - i is created at some incremental step. Consider a

vertex S at level j in the arrangement A, and D one of the d lines supporting an
edge incident to S. This line D is the intersection of d - 1 hyperplanes among the
d hyperplanes that intersect at S. The ray Ds supported by D, that originates
from S towards the lower levels in the arrangement, intersects exactly j hyper-
planes H 1 , H2 ,. . ., Hj, ordered according to their intersections MI = Ds n H1
with Ds, I = 1, .. ., j. If the edge SMj is created during the execution of the
algorithm, the corresponding node in the graph will be found to conflict in sub-
sequent location steps at most i - 1 times. The condition for the edge SMj to be
created is that at some incremental step in the execution of the algorithm, one
of the d + 1 hyperplanes that determine this edge is inserted while hyperplanes
inserted at previous steps include the d other hyperplanes that determine SMi
but not the hyperplanes Hi' indexed by i' < i nor more than 1< k - 1 of the j - i
hyperplanes Hi' indexed by i' > i. With this characterization, the probability Pji
that SMj is created can be written as
if j < k, Pji = (d + )!(i 1)!
if j > k, Pji = E (ija) d+1

l+ d +1
k- jd I d+ 1
- =o (j+d
I+d 8j-1
J
Observing that
j- jai
(d + )! (i -1)! I)d+1
(d+i)! 1E0 (j+d)j-
and with the convention that (i i) vanishes for I > j -i, we can write in
either case
P'jz(k-1i+ 1 +d-ji)
Pji
1=0 ( j+ d j
+ d}
The average number c of conflicts detected by the algorithm is then
n-d i
c=E |Sj|I d (i -l)pji.
j=O i=1
We are led to compute gj = E 1l(i-1)pji.

-
i k-1
gm=E(i- 1)
i=1 1=0
An elementary computation shows that
i j-1
Z(i
ti 1
-1)
( j-i)
I J
= E(j-1-i)
i=o
j-1 m-1
(i)
= LE 1: m=O i=O
m=OY
- i( )
= 1+2J)'
so that
k-l d+1 ( 1+2)
( j+d )
1+d J
k-1
1) l-1 j! (I + d)!
= ld +
j+d (j+d-1)!(1+2)!
1=0
Put
k-I j (1 +d)!
gj' = E(d + 1) (j + d -1)! (I + 2)!
1=0
Since gj < g,.we can restrict our attention to gj, for which we have
k-It k +d )
d+ 1 1 (- 1 +d) = d+l ( d -1J
9i d- 1 jj+d- 1 E d -2J d-l1 j+d- 1 '
( d-1 J 1=01
14. 6. Exercises 343
and this yields finally

9, < d +1 (k+dd) d-l
We can perform an Abel transform analogous to that giving the expected num-
ber of vertices created during the execution of the algorithm, to obtain
n-d
c < SIdg'
I
j=0
n-d-1
< , lS<jI
d (gj -. j+q) + d IS<n-df an-d,
j=k
from which we derive, using theorem 14.5.1 once again, that the expected number
of conflicts detected by the algorithm is O(nk log(n/k)) if d = 2, O(nk 2 log(n/k))
if d = 3, and 0(nLd/2ikrd/21) if d > 4. The overall complexity of all the lo-
cation sub-phases in all incremental steps is thus O(nk2 log(n/k)) if d = 2,
O(nk3 log(n/k)) if d = 3, and O(nLd/2Jkrd/21+1) if d > 4.
The algorithm that computes the first k levels in the d-arrangement of n hyper-
planes, as described in this section, therefore has complexity Q(n[d/2J kfd/21+1).
This result may be somewhat improved as in exercise 14.21. A reference is
given in the bibliographical notes. This proves the following theorem.
Theorem 14.5.3 The first k levels in the simple d-arrangementof n hyperplanes

can be computed in time O(nld/2 Jkfd/2 1) if d > 4, and in time O(nkfd/21 log.k) if
d=2 or3.
14.6 Exercises
Exercise 14.1 (Projective arrangements and polytopes) Prove all the results in
section 14.2 using the combinatorial results on polytopes derived in chapter 11.
Hint: Let X be a set of hyperplanes in Ed. With each hyperplane in - we associate an

oriented projective hyperplane, and we denote by it the set of these oriented projective
hyperplanes. In the spherical model of the oriented projective space, each projective
oriented hyperplane is represented by a great (d - 1)-sphere in Sd. The arrangement
of these great (d - 1)-spheres on Sd is called the oriented projective arrangement of
Wt. The projective arrangement of X is obtained by identifying antipodal points on the
sphere. The oriented projective arrangement is a spherical polytope drawn on the sphere
Sd to which the same relations as a polytope apply. To each k-face in the projective
arrangement of X there corresponds two k-faces in the oriented projective arrangement,
so the formulae for projective arrangements can be obtained by halving the analogous
formulae for polytopes. In particular, Euler's relation for a projective arrangement is
d 1 + (_l)d
St-l)kpk(A) = 2(
k=O
if Pk is the number of k-faces in the arrangement.

The formulae for arrangements in Euclidean space can be obtained by observing that
the number of bounded k-faces is the same in Euclidean and in projective arrangements.
Unbounded k-faces, however, correspond to two faces in the Euclidean arrangement (one
on each side of the hyperplane at infinity), but also to the (k -1)-faces on the hyperplanes
at infinity. Reversing the argument, two unbounded opposite k-faces in the Euclidean
arrangement can be accounted for by the enclosing k-face in the projective arrangement
and by the (k - l)-face in the projective (d - 1)-arrangement, so that finally
n(d) = p(d) + (d-1)
Exercise 14.2 (Dehn-Sommerville relations) Show that the following relations are
satisfied for simple d-arrangements:
Z(-l)j2k i k ) ni = nk.
Hint: Use the correspondence described in exercise 14.1 between arrangements and
spherical polytopes. The exercise follows from an easy adaptation of the proof of theorem
7.2.2, using lemma 14.2.4 in place of lemma 7.1.14.
Exercise 14.3 (Number of faces) Unlike polytopes, the number of k-faces of a simple
d-arrangement of n hyperplanes depends only on n, d, and k, and not on the relative posi-
tions of the hyperplanes. Show that the number Pk (n, d) of k-faces in the d-arrangement
of n projective hyperplanes (see exercise 14.1) satisfies
Pk d( n ) k ( n-d-1+ k)
Give a similar formula for Euclidean arrangements.

It should not, however, be believed that all arrangements are combinatorially equiva-
lent (have isomorphic incidence graphs). Disprove this by drawing two simple arrange-
ments of lines in the plane or in the projective plane for which no bijection between the
two sets of faces preserves the incidence relationships.
Exercise 14.4 (Skeleton of an arrangement) Show that the entire d-arrangement

of n hyperplanes in general position may be reconstructed from only the 1-skeleton of
this arrangement, that is, only the vertices, edges, and their incidence relationships.
Show that this reconstruction can be achieved in optimal O(nd) time.
14.6. Exercises 345
Hint: Proceed by rebuilding the faces in order of increasing dimension. Under general
position assumptions, a k-face F incident to a (k - 1)-face G is obtained by relaxing a
hyperplane. To know on which side of this hyperplane F is contained, observe that F is
contained in the intersection of d - k half-spaces bounded by all but one of the d - k + 1
hyperplanes whose intersection contains G. When creating a k-face F, we update the
incidence relations between F and its sub-faces of dimension k - 1 (already created).
The incidence relations between F and the faces of dimension k + 1 will be taken care of
when processing the (k + 1)-faces.
Exercise 14.5 (Incidences along a line) Consider a set X- of n hyperplanes, and a

line L that is the intersection of d - 1 hyperplanes in 'H. Show that the number of edges
in the d-arrangement A(H) that are incident to vertices lying on L is O(nd-1).
Exercise 14.6 (Incremental construction) Generalize the algorithm of section 14.4

to compute non-simple arrangements within the same time bounds.
Exercise 14.7 (Canonical triangulation) The canonical triangulation of an arrange-

ment of hyperplanes is defined as follows. We triangulate the faces inductively in order
of increasing dimension. A k-face F is triangulated by joining its vertex SF of smallest
d-th coordinate to the (k - 1)-simplices in the triangulation of the (k - 1)-faces of F
that do not contain SF. Show that this triangulation has the same complexity as the
whole arrangement and that it can be computed in time 0(nd). Show that the number
of simplices in this triangulation that are crossed by any given hyperplane is O(nd- 1).
Exercise 14.8 (Zonotopes) A zonotope is a polytope obtained as the Minkowski sum

of a finite set of line segments (see exercise 7.17). Find a bijective transformation that
maps the faces of a zonotope Z in Ed, the sum of n line segments, to the faces of the
projective (d - 1)-arrangement of a set Ht of n hyperplanes in Ed-l and which preserves
incidence relationships. Using this, give an upper bound on the complexity of a zonotope
and an algorithm that computes it, knowing the n segments that define the zonotope. To
what does the zone of a hyperplane H in the arrangement correspond in the zonotope?
Hint: Let S 1 , ... , Sn be the n segments that define Z. Assume that the segments are
centrally symmetric through the origin, by translating them if necessary, so that Si has
endpoints Ai and -Ai. Denote by Z# the polytope polar to Z. From 7.17, deduce that
n
#= {X: IX Ail < 11.
i-l
Let F be the face defined by
F = Si, eD3
.. e Si, ± ±iAi + .+ + sAin (14.9)
where =i1, j r + 1, . , n. Show that the face F# is
F# = {X Z# X -XAi,
= 0 for j =1,...,r, and X (Ai,+, + + Aij =1}.
From this, it follows that the faces of Z# corresponding to faces of Z that contain a
translate of Si are themselves contained in the hyperplane Hi = {X E Ed: X *Ai = 0}.
The correspondence between d-arrangements of hyperplanes passing through the origin
and projective (d - 1)-arrangements finishes the proof.
Exercise 14.9 (The painter's problem) A painter has n buckets at his disposal, each
containing a mixture of some basic colors Cl,.. ., Cp. The i-th mixture is characterized
by the proportions of each color, which can be modeled as a point Si in a space EP. A
product is obtained by blending some mixtures together. Show that the set of feasible
products, meaning that they can be derived from the mixtures in the n buckets, can be
characterized by the zonotope Z (see exercise 14.8) built on the segments Si. Design an
algorithm that computes the feasibility of a product using only the mixtures in the n
buckets. Note that the solution is not unique in general. In the case of binary blends
(p = 2), there is an optimal way to compose a product M such that the residualset of
products, which are still feasible after the necessary quantities of mixtures have been used
to make M, is maximal with respect to inclusion. Devise an algorithm that computes
this optimal mixture.
Hint: After the product M is created, the residual set of feasible products is contained
in Zr= Z n (Z - M). If p = 2, then Zr is a zonogon (2-zonotope).
Exercise 14.10 (Ray shooting) Consider a set of n disjoint line segments in the plane.
A ray shooting query, given a point and a vector, asks for the first segment visible from
this point in the direction given by the vector. Equivalently, given a ray, the query asks
for the first segment that intersects this ray. Show how to preprocess a data structure in
time and space 0(n 2 ), so that ray shooting queries can be answered in time 0(logn).
Hint: Let D be the line that supports the ray. Using polarity, show that finding the
segments intersected by D is equivalent to locating the point D* dual to D in an arrange-
ment A of 2n lines. Show also that, for any point A* in a cell of A, the order in which
the segments intersect the dual line /\ is identical. Thus, to a cell A in A corresponds
an ordered list L(A) of segments. If we store this list in an auxiliary structure (such as
a dictionary or a balanced binary tree), then once the cell containing D* is found, the
ray shooting query can be answered in additional time 0(logn). For locating that cell,
we use the structure described in exercise 12.2. The size and preprocessing time of the
entire data structure is thus 0(n 3 ), for a query time of 0(log n). To save a factor n, we
must store the lists more compactly. Notice that these lists differ only by one element
for two adjacent cells. We may therefore use a persistent dictionary (see exercise 2.7).
Exercise 14.11 (Queries in the plane) Consider a set P of n points in the plane.
Show how to preprocess P into a structure using space 0(n), so as to retrieve a point
belonging to any half-plane H+, bounded by a query line H and containing the origin,
in time 0(log n) per query. Show that all these points may then be retrieved in time
proportional to their number.
Hint: Peel the consecutive layers Eo, . . £jk.of P (as is done in exercise 9.3), and build
their vertical decomposition. Polarity with respect to a point inside Sk transforms these
nested polygons into another set of nested polygons E ..... ,.£ and H into a point H*.
The point-location structure of exercise 12.2 can be used to find in logarithmic time the
greatest i such that H* does not belong to Si, which is also the smallest i such that H
intersects Si. A point on Si that belongs to H+ can be found in logarithmic time as well,
and the other points can be retrieved by following the boundary chain of a layer or by
using the vertical decomposition to go from layer to layer.
14.6. Exercises 347
Exercise 14.12 (Higher-dimensional queries) Given a set P of n points and a hy-

perplane H in Ed, explain how to build a data structure of size 0(nLd/2 +E) that finds
in logarithmic time a point in P that belongs to a half-space H+ bounded by a query
hyperplane H and containing the origin. Show how to retrieve all these points in time
O(klogn), if k is their number.
Hint: Consider again the dual problem, by using the polarity with respect to the point
(0, . . . , 0, +oo). The problem is thus to find the hyperplanes dual to the points in P
lying above the pole H* of H. Build a hierarchical structure with O(logn) layers. The
topmost layer represents the canonical triangulation (see exercise 14.7) of the cell at level
0 (with respect to the center of the polarity) in the arrangement of a small sample of
the dual hyperplanes. (These cells being unbounded, the d-simplices in the triangulation
are more appropriately cylinders based on (d -1)-simplices.) Using the tail estimates of
exercises 4.5 and 4.6, show that if H is the set of hyperplanes and 1? a random sample
of constant size r drawn from N, then with high probability no cylinder in the canonical
triangulation of the cell at level 0 in the arrangement of 1Z intersects more than 0( ' log r)
hyperplanes in 'H \ 1Z. Taking r big enough, show that the recurrence on the size of the
resulting structure solves to O(nLd/2j+,) (where E depends on r), and that the structure
has 0(logn) layers.
To answer a query, we traverse the structure from the first layer and recursively deter-
mine all the hyperplanes lying above H*. Let 7? be the sample attached to the current
layer and let S be the cylinder in the triangulation of the arrangement of 1Z that is
intersected by the vertical ray originating at H*. If H* belongs to S, then we recur-
sively determine all the hyperplanes crossing S that lie above H* in S. Otherwise, we
systematically test H* against the n hyperplanes, in time 0(n). Let k be the number of
hyperplanes lying above H*. Point H* lies in no simplex of the canonical triangulation
of the cell at level 0 in the arrangement of R if and only if 7? contains at least one of
the k hyperplanes lying above H*. This happens with probability O(k/n) since 1Z is of
size r = 0(1). Since the structure has 0(logn) layers, the expected time for a query is
O(k log n) as wanted.
Exercise 14.13 (Computing a zone) Show how to compute the zone of a hyperplane
in the arrangement of n hyperplanes in Ed in time 0(nd-1 + n log n) and space (nd-1 ).
Hint: Adapt the algorithm that computes a cell in the arrangement of segments or trian-
gles as described in section 15.4 and subsection 16.4.3, and use the canonical triangulation
of the arrangement defined in exercise 14.7.
Exercise 14.14 (Convex hull of an arrangement) Given n lines, show how to com-
pute the convex hull of all the vertices of their arrangement in O(n log n) time.
Hint: Only the two extreme vertices on each line may be on the convex hull, so this
convex hull has complexity 0(n). Computing the zone of the line at infinity (see exercise
14.13) and removing the infinite edges yields a simple polygonal line with 0(n) vertices,
whose convex hull is exactly the convex hull of the arrangement.
Exercise 14.15 (Sorting by polar angle) Consider a set of n points Ml,..., Mn in

the plane. We can define a partial order on the lines MiMj passing through any two
points Mi, Mj, in the following way: D = MiMj < D' = MiMk if the slope of D is
smaller than that of D'. Show that one can compute a total order on these lines that is
compatible with the partial order in optimal time O(n2 ).
Hint: We reason in the dual space where an oriented line D of equation y cos 0 - x sin 0 -
6 = 0 ( E t-7r, +7r[) is represented by the point (0,6). To each point Mi corresponds
the pencil of lines passing through this point, and the dual of that pencil is represented
by the line dual to Mi. Thus the duals of the points M 1 ,... , M,, form an arrangement
of dual lines. A breadth-first search traversal of the oriented graph of the vertices of this
arrangement whose arcs correspond to the edges of the arrangement oriented towards
increasing abscissae yields the desired total order.
Exercise 14.16 (Visibility graphs) Consider a set S of n segments in the plane with
disjoint interiors. The visibility graph of these segments is the graph whose nodes are
endpoints of segments and whose arcs join two nodes that are visible one from another,
meaning that the segment joining these two endpoints does not cross any of the other
segments. Show that the visibility graph of S can be computed in time O((n + k) log n),
where k is the size of the visibility graph. When k is high, show how to improve the
complexity to 0(n 2 ) which is optimal in the worst-case.
Hint: Compute the downward vertical decomposition of the segments by erecting walls
hanging below each endpoint. Then rotate the direction of the decomposition from -7r/2
to 37r/2 while maintaining the oriented decomposition. The visibility graph can be com-
puted in the process, since two visible endpoints will share a trapezoid for some orienta-
tion. In order to process the events during the rotation in the correct order, sort them in
a priority queue (as was done for computing segment intersections by a sweep algorithm
in subsection 3.2.2). In order to achieve time O(n 2 ), sort all the lines that connect two
endpoints by polar angle as in exercise 14.15.
Exercise 14.17 (Shortest paths) Let P be a polygonal region whose boundary is

made up of one or several polygons with a total of n edges. Let I and F be two points
of P. Show how to compute a shortest path that connects I to F and that is contained
inside P in time 0(n 2 ).
Hint: Use the visibility graph of exercise 14.16.
Exercise 14.18 (k-sets) Let M be a set of points. A subset of M is called a k-set if

it has exactly k points and if there exists a hyperplane H that separates its points from
the other points of M. Show that the number of k-sets of M equals the number of cells
at level k in the arrangement of the hyperplanes dual to the points in M. What does an
i-face in this arrangement correspond to in the original space?
Exercise 14.19 (Complexity of a level) Let A = {A 1 , . .. , An} be a set of n points

in the plane. Show that a set of n points in the plane has at most 0(nvk) k-sets, and
dually, that an arrangement of n lines in the plane has 0(nvik) faces at level k, for k > 0.
14.6. Exercises 349
Hint: We call a k-segment an oriented line segment AiAj such that the open half-plane
on the right of the line joining Ai to Aj contains exactly k points of A.
1. Let a line A sweep the plane parallel to the y-axis while maintaining the number
of k-segments intersected by A. Show that this number may increase or decrease by at
most one, each time A sweeps over a point of A.
2. Let A be an oriented line containing no points of A, and let Al and Ar be the
half-planes respectively to the left and right of A. Use the previous result to show by
induction on k that there are at most min(k + 1, L[J) k-segments crossing A that are
segments of the form AiAj, Ai E Al and Aj E Ar.
3. Draw n - 1 vertical lines D 1 ,..., Dn- 1 that divide the plane into strips each con-
taining one point of A. Among these lines, pick p - 1 and call them A 1, . .. , Ap-1, such
that each of the p vertical strips they define contains at most [n] points of A. The
number of k-segments that do not cross any of the Ai, i = 1,... ,p - 1 is at most
( Fn1 n2 + np
2 2p
The other k-segments must cut a line Ai and their number may be bounded using the
result in 2. The bound on the number of k-sets is obtained by optimizing the choice of
the parameter p.
Summing over all j < k, we obtain a bound of O(k'n) on the number of all j-sets for
0 < j < k. This bound is not tight, as shown by theorem 14.5.1.
Exercise 14.20 (A lower bound on the k-level) Show that the k-level of the d-
arrangement of n hyperplanes may have as many as Q(nLd/2 ]kfd/ 2 1-1) faces of all di-
mensions.
Hint: Consider the dual problem of bounding the number of j-sets for a set P of n
points in Ed (see exercise 14.18). To establish the lower bound, we place the points in
P on the moment curve M (see subsection 7.2.4). Any subset D of d points in M splits
M into d + 1 arcs Ml, . . . , Md+ and induces a decomposition of P into d + 1 subsets
Pi = P n Mi, i = 1, .d. , d + 1. The sets Pi are alternately on one or on the other side
of the hyperplane affine hull of D. The problem is now to count the subsets D of P such
that
Z 1P
21+11 =j or E P21 1=j.
1=1 1=1
For this, show that the number of ways to split an ordered set of s elements into r ordered
subsets is
1s+
Then deduce that the number of j-sets is (nLd/2J kfd/21 -1), and that summing over all
j < k yields a lower bound on the complexity of A~k which is identical to the upper
bound shown in theorem 14.5.1.
Exercise 14.21 (Lazy computation of the first k levels) Adapt the randomized
algorithm of subsection 15.4.2 to compute the first k levels in the arrangement of n
hyperplanes in expected time 0 (nkrd/ 21 log(n/k)) if d = 2 or 3, and 0(nLd/2ikrd/21) if

d > 4.
Hint: The peeling phase which is systematically processed in the algorithm of subsection
14.5.3 may be replaced by lazy clean-up operations between some well-chosen incremental
steps. Use the canonical triangulation (see exercise 14.7).
Exercise 14.22 (Easier computation of the first k levels) Show how to simplify
the algorithm that builds the first k levels in the d-arrangement of k hyperplanes when
all the hyperplanes contain a face at level 0.
Hint: It suffices to find the first conflict with the polytope at level 0 (a problem analogous
to that of finding the conflicts to build a convex hull), and then to use the adjacency
graph to detect the other conflicts.
Exercise 14.23 (Optimal separation) Let B be a set of n blue points and R be a

set of n red points in Ed. Given a hyperplane H, we define its separation defect as the
smaller of the two numbers JB n H+J + JR n H- I and Bl n H-1 + JR n H+f. Show how
to compute a hyperplane that best separates 1? from B in time 0(nd).
Hint: Without loss of generality, H- may account for points of B and H+ for 7R. A
blue point B (resp. red point R) will be accounted for by the wrong set if the pole H*
of H belongs to B*+ (resp. R*-). The question is now equivalent to finding the sum of
the levels of H* in the dual arrangements of B and 7?.

Combinatorial properties of arrangements and the solutions to exercises 14.1, 14.2, and
14.3 can be found in the book by Griinbaum [114]. Other combinatorial and algorithmic
results are presented in the book by Edelsbrunner [89], which is the authoritative refer-
ence on arrangements. The proof of the zone theorem given in section 14.3 is adapted
from an article by Edelsbrunner, Seidel, and Sharir [97]. Coxeter [73] and McMullen [158]
study zonotopes and the correspondence between zonotopes and arrangements of hyper-
planes. The application of zonotopes to mixtures is presented in the work by Lacolle,
Szafran, and Valentin [143], who give a solution to exercise 14.9.
A theorem analogous to the zone theorem holds for algebraic surfaces of bounded
degree in arrangements of hyperplanes. Aronov, Pellegrini, and Sharir [11] showed that
the complexity of such a zone is Q(nd-l log n).
The incremental algorithm that constructs the arrangement of n lines in the plane
uses 0(n 2 ) working space since it maintains a representation of the current arrangement.
Another time-optimal algorithm based on the topological sweep method by Edelsbrunner
and Guibas uses only 0(n) working space [92].
Theorem 14.5.1, which bounds the number of faces in the first k levels in arrangements
of hyperplanes, is due to Clarkson and Shor [71]. Lower and upper bounds on the
complexity of a single level (see exercise 14.19) are given by Erd6s, Lovasz, Simmons,
14. 7. Bibliographical notes 351
and Straus 1103) and generalized to higher dimensions in the book by Edelsbrunner [89].
The result in exercise 14.19 was slightly improved by Pach, Steiger, and Szemeredi [186].
The algorithm that computes the first k levels is due to Mulmuley [174] (see also [177)).
The algorithm as outlined in exercise 14.21 is fully described (and slightly improved in
dimensions 2 and 3) by Agarwal, de Berg, Matougek, and Schwarzkopf [2]. A determin-
istic optimal algorithm is given in the planar case by Everett, Robert, and van Kreveld
[104], who also give a better solution to exercise 14.23 in the planar case.
Query problems, for a large part unexplored in this book, have spurred a lot of research.
A recent account can be found in the books by Agarwal [1] or by Mulmuley [177]. Variants
of exercises 14.10 and 14.12 are solved in these books, and the solution to exercise 14.11
is due to Chazelle, Guibas, and Lee [55]. Bronnimann [35] explains how to achieve point
location in a polytope with preprocessing time O(nlogn + nLd/2j), storage 0(nLd/2l),
and query time O(log 2 n).
Computing visibility graphs and shortest paths have also motivated a lot of research.
Recent developments and references can be found in the articles by Pocchiola and Vegter
[188] and by Hershberger and Suri [127].
Chapter 15
Arrangements of line segments

in the plane
In an arrangement of n lines in the plane, all the cells are convex and thus have
complexity O(n). Moreover, given a point A, the cell in the arrangement that
contains A can be computed in time e(n log n): indeed, the problem reduces to
computing the intersection of n half-planes bounded by the lines and containing
A (see theorem 7.1.10).
In this chapter, we study arrangements of line segments in the plane. Consider
a set S of n line segments in the plane. The arrangement of S includes cells,
edges, and vertices of the planar subdivision of the plane induced by S, and their
incidence relationships.
Computing the arrangement of S can be achieved in time O(n log n+k) where k
is the number of intersection points (see sections 3.3 and 5.3.2, and theorem 5.2.5).
All the pairs of segments may intersect, so in the worst case we have k = 2(n2 ).
For a few applications, only a cell in this arrangement is needed. This is notably
the case in robotics, for a polygonal robot moving amidst polygonal obstacles by
translation (see exercise 15.6). The reachable positions are characterized by lying
in a single cell of the arrangement of those line segments that correspond to the
set of positions of the robot when a vertex of the robot slides along the edge of
an obstacle, or when the edge of a robot maintains contact with an obstacle at a
point. Since the robot may not cross over an obstacle, it is constrained in always
lying inside the same cell of this arrangement. It is therefore important to bound
the complexity of such a cell and to avoid computing the whole arrangement.
Among the cells of A(S), a few contain the endpoints of some segments, and the
others do not. The latter are naturally convex cells, their complexity is O(n) and
each can be computed in time 0(n log n). The complexity of the former cells,
however, is more difficult to analyze.
15.1. Faces in an arrangement 353
To conduct the combinatorial analysis, we introduce and study a certain class

of words over a finite alphabet, the so-called Davenport-Schinzel sequences (see
section 15.2). These words have a geometric interpretation that is both illumi-
nating and useful: lower envelopes of functions (see section 15.3). Section 15.4
bounds the complexity of a cell and gives an algorithm that computes it. We
first show that this complexity is almost linear, in contrast with the entire ar-
rangement which may have Q(n 2 ) edges in the worst case. The complexity of
the algorithm is shown to be roughly proportional to the complexity of the cell
it computes.
15.1 Faces in an arrangement

Let S be a set of n segments in the plane. To define a cell of their arrangement,
we need to distinguish the two sides of a segment, or equivalently to consider that
each line segment is a flat rectangle with an infinitesimally small width. The ar-
rangement of S is formed by cells, edges, and vertices of the planar subdivision
induced by S, and their incidence relationships. More precisely, the connected
components of E 2\ S are polygonal regions that may have holes (see figure 15.1):
the cells of the arrangement are formed by the topological closures of these re-
gions. The edges and vertices of this arrangement are the edges and vertices of
the polygons that bound the cell. The arrangement of S will be denoted by A(S).
15.2 Davenport-Schinzel sequences

Given an alphabet with n symbols, a word on this alphabet is an ordered sequence
of symbols in this alphabet, and a subsequence of a word ul ... u" is a word
Uil ... Uik for some indices 1 < il < ... < ik < n. Given two symbols a and b,
an alternating sequence of length s is a sequence ul ... us such that ui = a if i is
odd and ui = b is i is even. An (n, s)-Davenport-Schinzel sequence is a word on
an alphabet with n symbols such that:
1. Two successive symbols of this word are distinct.
2. For each two symbols a, b in the alphabet, the alternating sequence of length
s + 2 is not a subsequence of this word.
In other words, no two symbols can alternate more than s + 1 times.

So consider the phrase 'A DAVENPORT-SCHINZEL SEQUENCE',
considered as a word over the Roman alphabet. The reader will easily verify
that the longest alternating subsequence over two symbols is the subsequence
354 Chapter 15. Arrangements of line segments in the plane
Figure 15.1. A cell in the arrangement of line segments.
'ESESE' (the other subsequences 'ENENE' and 'ECECE' are also suitable).
The sequence 'A DAVENPORT-SCHINZEL SEQUENCE' is thus a
(26, 4)-Davenport-Schinzel sequence!
Denote by A,(n) the maximal length of an (n, s)-Davenport-Schinzel sequence.
First of all, it is not even clear that AS(n) is finite. In fact, it can be deduced from
the connection with lower envelopes (see section 15.3) that A,(n) < sn(n-l1) + 1.
The following theorem gives more precise bounds on Al, A2, and A3.
Theorem 15.2.1 The maximal length A,(n) of an (n, s)-Davenport-Schinzel se-

quence is bounded by:
AI(n) = n
A 2 (n) = 2n - 1
A3 (n) = e(na(n))
where a(n) is the very slow-growing inverse of Ackermann's function.'
Proof. The proof for s = 1 is trivial, since each symbol may appear only once.
For s = 2, we proceed by induction on n. The result is true for n = 1, so we
consider an (n, 2)-Davenport-Schinzel sequence (n > 1). Let a be its first letter,
'The definitions and order of magnitude of the inverse Ackermann function are given in
subsection 1.1.3.
15.3. The lower envelope of a set of functions 355
and put S = aS'. If a does not occur in S', then the induction applies for S' and
so
SI = 1 + jS'I < 1 + 2(n - 1) - 1=2n - 2.
Otherwise we can write S = aS1 aS2, where a does not occur in Si and ISI > 0.
If S2 is empty the length of aSla is smaller than (2n - 2) + 1 = 2n - 1, as
we have just shown. Otherwise, let k be the number of distinct symbols in S,.
By induction, IS11 < 2k - 1. Moreover, the definition of a Davenport-Schinzel
sequence ensures that no symbol b occurs both in SI and S2, otherwise abab is
a subsequence of S. Thus aS2 may contain at most n - k symbols (note that a
may occur in S2), and by induction we have IaS2 1 < 2(n - k) - 1. Hence
18 = IS11 + IaS2I+1 2n- 1.
To finish the proof for s = 2, we must also show that this bound is exact. This can
be readily seen by considering the sequence ablab2 a... abn- 1 a of length 2n - 1.
For s = 3, the proof goes into very technical details, so we will not prove the
announced result here. We can show, however, the simpler result that A3 (n) =
O(nlogn). Let S be a (n,3)-Davenport-Schinzel sequence, and S(a) be the
subsequence obtained from S by removing all the occurrences of a symbol a. In
S(a), there cannot be a subsequence bcbcb and identical consecutive symbols can
happen at most twice when the first and the last occurrences of a are surrounded
by two b's. Let us call S'(a) the sequence obtained by replacing in S(a) two
consecutive symbols b by a single b, whenever this happens. Then S'(a) is an
(n - 1, 3)-Davenport-Schinzel sequence, and
|S| < |S'(a)| +2+na < A 3 (n-1) +2+na
where na stands for the number of occurrences of a in S. Summing over all the
symbols a appearing in S, we obtain:
nISI < nA 3 (n- 1) + 2n + ISI.
This is true for any sequence S, so that
A3 (n) A 3 (n-1) 2
~+
n n-1 n-1
whence A3 (n) = 0 (n log n). l
15.3 The lower envelope of a set of functions

Consider n continuous functions fi(x),i = 1,...,n defined over R. The lower
envelope of the fit's is the graph of the function defined by
f (x) = min
i
fA()-
fl
fi f3 fi f2 fl
Figure 15.2. The lower envelope of a set of functions, and the corresponding Davenport-
Schinzel sequence.
The lower envelope is formed by a sequence of curved edges, where each edge is
a maximal connected subset of the envelope that belongs to the graph of a single
function fi(x). The endpoints of these edges are located at the intersections of
the graphs of the functions and are called the vertices of the envelope.
15.3.1 Complexity
Labeling each edge by the index of the corresponding function, we obtain a se-
quence of indices by enumerating these labels in the order in which they appear
along the envelope (see figure 15.2). If the graphs of the functions have pairwise
at most s intersection points, then this sequence is an (n, s)-Davenport-Schinzel
sequence. Indeed, let Ai and Aj be two edges appearing in this order along the
envelope, defined over two intervals I and J. The corresponding functions fi
and fj being continuous, they must intersect in a point whose abscissa is greater
than the right endpoint of I and smaller than the left endpoint of J. Having
an alternating subsequence of length s + 2 for the two symbols i and j implies
the existence of s + 1 intersection points between the graphs of fi and fj, a
contradiction.
The number of edges on the lower envelope is thus bounded above by the
maximal length A, (n) of an (n, s)-Davenport-Schinzel sequence.
Consider the case when the functions are defined over closed intervals and not
over the whole of R. The lower envelope is not continuous and the argument used
15.3. The lower envelope of a set of functions 357
Figure 15.3. Extending the function fi.
above to bound the number of its edges does not hold any more. This problem
may be overcome by extending the domain of definition of the functions fi to cover
the whole of JR. More precisely, pick a positive real number p/. If fi is defined over
[Ximin, XimajI, then we extend the graph of fi for x > Xijma by the semi-infinite ray
originating at (Xjmax, fj(xjmax)) whose slope is pi, and symmetrically for x < Ximin
by the semi-infinite ray originating at (Ximin Ifi (Ximin)) whose slope is -p (see
figure 15.3). Thus we have a set of functions gi which extend the functions fi
and are continuous. When p is large enough, the sequence of labels of the edges
on the lower envelope of the gi's is identical to that of the lower envelope of the
fj's, and this lower envelope can be easily constructed knowing that of the gi's.
It is readily verified that, for p large enough, gi and gj have at most s + 2
intersection points if the corresponding functions fi and fj intersect in at most s
points. It follows that the sequence of labels of the edges on the lower envelope
of gi, ... , g. is a (n, s + 2)-Davenport-Schinzel sequence.
The complexity of the lower envelope of the gi's is thus bounded above by
the maximal length AS+2 (n) of an (n, s + 2)-Davenport-Schinzel sequence. The
complexity of the lower envelope of the fit's is also bounded by As+2 (n).
Example. Consider the case of line segments. Two line segments intersect
in at most one point, so the sequence of labels on the lower envelope of a set
of segments is an (n, 3)-Davenport-Schinzel sequence. The complexity of this
lower envelope is thus O(na(n)). In fact, this bound is achievable and one may
actually construct line segments whose lower envelope has super-linear complexity
e(na(n)) (see the bibliographical notes at the end of this chapter).
Let us now consider the case when the functions fi are only defined over semi-
infinite intervals. We first consider the functions fi whose domains of definition
are intervals defined by x > Ximin. If we extend these functions by a half-line
starting at (XiminI fA(Ximin)) of slope -y for [z big enough, then we obtain func-
tions gi, defined over lR, whose graphs have pairwise at most s + 1 intersection
points if the graphs of the fit's had pairwise at most s intersection points. The
sequence of labels on the lower envelope 4r of the gi's is an (n, s + 1)-Davenport-
Schinzel sequence. The complexity of Lr is thus A,+,(n).
A similar result obviously holds for the lower envelope El of the functions f3
whose domains of definition are defined by x < xjmax, The lower envelope of the
n functions fA is the lower envelope of the union of Lr and El. Its complexity is
O(nr + n1) = 0(A,+,(n)) since both 4r and LI are monotone chains.
Example. The lower envelope of n half-lines has complexity 0(n).
15.3.2 Computing the lower envelope

We now present an algorithm that computes the lower envelope of n functions
fi, i = 1, ..., n, defined over R such that no two graphs of these functions have
more than s intersection points. The algorithm recursively computes the lower
envelope I, of f'(z),. . . , fLn/2j, and the lower envelope 12 of fLn/2J+1), .* , fn (x)-
Both envelopes are monotone chains of complexity A,(n) < As(n), as was shown
in the previous subsection. Monotonicity implies that we can compute the lower
envelope of the union of I, and 12 by sweeping the plane with a line parallel to the
y-axis, in a manner that is similar to merging two sorted lists (see section 3.1.2).
Let us call the current edges the two edges of Al and 12 intersecting the sweep line.
When the sweep line passes over a vertex of a current edge I, or 12 , this current
edge is replaced by the edge that follows on the corresponding lower envelope. If
this edge is part of the constructed lower envelope, a new edge is created for the
lower envelope. When the sweep line encounters an intersection point between
the two current edges, a new edge is created on the lower envelope. In either case,
the next intersection point between the two current edges is computed. Merging
the lower envelopes in this fashion takes time proportional to the total number
of edges on 11 and 12, and to the number of intersection points between -l and
12 which is O(As(n)) as was shown in the previous section. We have thus proved
that:
Theorem 15.3.1 The lower envelope of n functions fi, i 1, ...,n, defined over
R and whose graphs have pairwise at most s intersection points, has complexity
0(A,(n)) and can be computed in time 0(A,(n) log n).
15.4 A cell in an arrangement of line segments

Let S be a set of n line segments in the plane. In the arrangement of S, we
may distinguish between cells whose boundaries contain at least one endpoint
15-4. A cell in an arrangement of line segments 359
of a segment (the non-trivial cells) and the cells whose boundaries contain no
endpoints (the trivial cells). Trivial cells are convex and their complexity is
O(n). In section 15.3, we have seen that the complexity of the lower envelope
of a set of n line segments in the plane can be e(na(n)). So we can conclude
that Q(na(n)) is a lower bound on the worst-case complexity of a non-trivial
cell in the arrangement of n line segments. To show this, consider a set of n
line segments whose lower envelope has complexity O(na(n)). To S, we add 2n
segments, almost vertical, and long enough so that each of them stands above an
endpoint of a segment in S (see figure 15.3). We also add a horizontal segment
lying above all the segments in S while cutting all the almost vertical segments
that we added. The new set of segments S' has 3n + 1 segments, and the edges on
the boundary of the unbounded cell lying below all the segments are in one-to-
one correspondence with the edges of the lower envelope of S'. But the Q(na(n))
edges on the lower envelope of S also correspond to a subset of the edges on the
lower envelope of S'. It follows that the unbounded cell is at least as complex as
the lower envelope of S, so that it also has complexity Q(na(n)).
As we will see, this bound is also an upper bound, which shows that the com-
plexity of cells in the arrangement of line segments depends almost linearly on
the number of segments, while the total arrangement may have up to Q(n 2 ) edges
in the worst case. We will then explain how to efficiently compute such a cell.
15.4.1 Complexity
Consider a set S of n line segments in the plane. We will assume that these
segments are in general position, meaning that no three segments have a common
intersection and that any two segments intersect in at most one point. A standard
perturbation argument shows that the complexity of a cell is maximal in this case.
Indeed, if the segments are not in general position, one may perturb them slightly
so that they are in general position, without decreasing the number of edges or
vertices of the cell under consideration.
From now on, and as was done in section 15.1, we consider that each line
segment S is a rectangle of infinitely small width whose boundary is formed
by two copies of the segment S called the sides of S, and two infinitely short
perpendicular segments at the vertices. Under the general position assumption,
the boundary of the union of these rectangles is homeomorphic to the union of
all the segments. Henceforth, we will thus make a distinction between a segment,
considered as a infinitely thin rectangle, and a segment side. The number of sides
is 2n.
We orient the rectangles counter-clockwise, which induces a clockwise orienta-
tion for the connected components of the boundaries of each cell.
Let r be a connected component of the boundary of some cell C in the ar-
rangement A(S) of S. Note that a segment may contain several edges of F.
Lemma 15.4.1 Consider a segment S that contains at least one edge of f. The
edges of f contained in S are traversed on the boundary of r in the same order
as they are traversed on the boundary of S.
Proof. Consider the infinitely thin rectangle S and the region R bounded by r
that does not contain C. Then S is contained in R, and the result follows from
a slight adaptation of the proof of theorem 9.4.1. [1
We label each edge of F by the index of the side of the segment of S to which it
belongs. The sequence Er of these labels forms a circular sequence which we break
into a linear sequence by choosing some origin 0 on F. The number of distinct
labels in Er is at most the number of sides, which is 2n. Two successive labels are
distinct. Since two segments have only one intersection point, it is tempting to
conjecture that the sequence Er is a (2n, 3)-Davenport-Schinzel sequence. The
choice of 0 may induce some additional repeats, however. Indeed, if ababab is not
a subsequence of the circular sequence, it may not always be possible to choose
o so that the same is true for the linear sequence. For instance, consider figure
15.4: the linear sequence
Er = al c2 cl al a2 cl bi b2 Cl C2 b2 a2 al b2 bi
does contain the subsequence aiciaiciai. We solve this technical problem by

constructing another sequence E* on at most 3n symbols which is at least as
long as Er. Let L be a side of a segment that supports several edges along F.
These edges are naturally ordered by the orientation of L, so we let I be the first
point of L that belongs to F and F the last point of L that belongs to r. The
points I and F subdivide r into two chains ending at I and F. Denote by -ythe
oriented chain that contains the origin 0. The idea is to give a different label to
the edges on r that belong to L n y according to whether they are before 0 or
after 0. Then the new sequence E* is merely the linear sequence of these new
labels along the edges r. For instance, on figure 15.4, we now have
= al[ C2 cl al a2 Cl bt,' b2 Cl C2 b2 a2 a/ b2 b.
E* has at most 3n distinct labels, since only one side of each segment needs to
be relabeled.
Lemma 15.4.2 E* is a (3n, 3) -Davenport-Schinzel sequence.
Proof. We already know that E* has at most 3n distinct labels and does not
contain two identical consecutive elements. It remains to see that ababa is not a
subsequence of E* for any two symbols a $4b.
15.4. A cell in an arrangement of line segments 361
al
C2
--1 be
Figure 15.4. Circular and linear sequences.
We first show that, if abab is a subsequence of E*, the sides labeled a and b
must intersect. For this, let the subsequence abab correspond to the edges Ea,
Eb, Ea, Eb 1 on r. Let Sa be the side labeled a that contains Ea and Ea. Pick a
point Al in the relative interior of Ea and a point A2 in the relative interior of
Ea" (see figure 15.5). We define Sb, B1 and B2 similarly.
Let A be the union of the subchain F12 of F that joins Al to A2 and of the
simple polygonal chain contained in the interior 2 of Sa. Then A is a simple closed
polygonal chain. The bounded polygonal region A enclosed by A contains, in a
neighborhood of B1 , a portion of the segment B1 B2 . Indeed, if A is oriented by
the orientation induced by r, then in a neighborhood of Al the side Sa is on the
right of A and the cell lies to the left, and a similar statement holds for Sa in
a neighborhood of A2 and for Sb in a neighborhood of B1 . Moreover, A cannot
cross the portion of r that joins A2 to B2 , so that A cannot contain B2 . The
segment B1 B2 must therefore cross A. It cannot cross F1 2 , however, hence it
must cross A \ F12 , and therefore also A1 A 2.
Assume now for a contradiction that ababa is a subsequence of E*. In addition
to the notation above, let us pick a point A3 in the relative interior of Ea" that
is after A2 on Eat, and another point A4 in the relative interior of the third edge
Eas" labeled a, and so supported by Sa. From the preceding argument, we know
that A1 A2 and B1 B2 intersect, and similarly for B1 B2 and A3 A4 (simply consider
2
We assume that the segments are in general position, and that they are infinitely thin
rectangles.
r'2
the subsequence baba of Er). The two intersection points must be distinct since,
owing to the relabeling and to lemma 15.4.1, the points Ai are all distinct and
necessarily appear in the order A1 , A2 , A3 , A4 on Sal But this latter condition
implies that A1 A2 and A3 A4 cannot intersect, so that Sb must cut Sa twice. This
is impossible as two segments may only cross once. C
An immediate consequence of this lemma is:
Theorem 15.4.3 The complexity of a cell in the arrangementof n line segments
in the plane is O(na(n)).
As we mentioned in section 15.3, it is possible to place segments in the plane so
that the cell containing, say, the origin has complexity Q(na(n)), so the bound
in the theorem above is tight.
15.4.2 Computing a cell

The algorithm
Again let S be a set of n line segments in the plane, assumed to be in general
position, and pick a point A that does not belong to any of the segments in S.
Our goal is to compute the cell C(S) in the arrangement of S that contains A.
The algorithm we present here is a variant of the randomized on-line algorithm
that computes the vertical decomposition of S. The reader unfamiliar with that
algorithm is invited to refer to subsection 5.3.2 for more details. We will only
recall here the main definitions. The vertical decomposition is obtained by casting
a ray upwards and downwards from any endpoint of the segments. The ray stops
as soon as it encounters a segment in S (see figure 5.4a,d). The vertical segments
(sometimes half-lines) traced by the rays are called walls, and together with
the segments in S they decompose the plane into trapezoids that may degenerate
into triangles or unbounded trapezoids. The algorithm also computes the vertical
adjacencies of the trapezoids. 3
To apply the formalism of chapter 4, we defined the problem in subsection 5.3.2
in terms of objects, regions, and conflicts between objects and regions. For this
problem, an object is a segment. A region is a trapezoid in the decomposition of
a subset of the segments. Each region is determined by at most four segments.
There is a conflict between an object and a region if and only if the segment
intersects the trapezoid. Computing the vertical decomposition is thus the same
as computing the set of regions defined and without conflicts over S.
The algorithm we present to compute a cell C(S) in fact computes a vertical
decomposition of that cell (see figure 15.6). To generalize the algorithm of sub-
section 5.3.2 to compute only a single cell is not straightforward, however: the
regions that interest us are not all the trapezoids defined and without conflict
over the set S of segments, but only those contained in the cell C(S). Unfor-
tunately, whether a trapezoid is contained in the cell C(S) cannot be decided
locally by examining only that trapezoid and the segments that define it. This
forbids verbatim use of the formalism and results of chapters 4 and 5.
To avoid this difficulty, we proceed as follows. Let 7? be the subset of seg-
ments already inserted into the data structure, and let C(7?) be the cell in the
arrangement of R that contains A. We allow the algorithm to compute, in ad-
dition to the trapezoids in the decomposition of C(7?), other trapezoids in the
arrangement of 7R that are not trapezoids of 0(1?). In order not to degrade the
performances of the algorithm, at certain incremental steps we perform a clean-up
step, during which we remove the trapezoids that do not belong to the cell C(7?).
Only the trapezoids that belong to C(7?) will be subdivided during subsequent
incremental insertions. To distinguish between these trapezoids, we traverse the
connected component in the vertical adjacency graph 0 of the current vertical
decomposition that contains the trapezoid containing A. This latter trapezoid
is maintained throughout the incremental steps. The other leaves of the graph
that are not traversed are deactivated: they correspond to trapezoids in the cur-
rent decomposition that are not contained in the cell C(R). These trapezoids
will not be subdivided, and the corresponding leaves in the graph will not have
children in subsequent insertions. Figure 15.7 shows an intermediate situation in
the algorithm.
3
Recall that two trapezoids are vertically adjacent if they share a common vertical wall.
Figure 15.6. Vertical decomposition of the cell that contains A.
Between two clean-up steps, the algorithm is similar to the one described in
subsection 5.3.2, apart from a few details which will be noted below. For each
insertion of a new segment S, we locate S using the influence graph, then update
the decomposition by subdividing the active trapezoids intersected by S. In the
influence graph, this corresponds to creating new children for the active nodes
that conflict with S.
Between two clean-up steps, and inside each trapezoid which has not been de-
activated, we build the decomposition of the arrangement of the segments which
conflict with this trapezoid and are inserted between the two clean-up steps. Let
Tp be the set of nodes in the influence graph which were not deactivated during
the previous clean-up step p. To each node in Up we assign a secondary influence
graph. This secondary graph is rooted at T and its nodes are the descendants of
T created between step p and the next clean-up step. The secondary graph com-
puted just as in subsection 5.3.2 under the incremental insertions of the segments
inserted between step p and the next clean-up step. Its construction differs from
that of a usual influence graph in a minor detail: the removal of superfluous walls.
When inserting a segment S, if it intersects a wall, then only one of the two parts
of that wall intersected by S is a wall in the new arrangement, and the other
part must be removed and the two adjacent trapezoids must be merged. This
procedure is detailed in subsection 5.3.2, and we apply it here to adjacent trape-
zoids that belong to the same secondary influence graph and also to trapezoids in
different secondary influence graphs. A merge of the latter kind is called an ex-
Figure 15.7. Intermediate situation in the computation of the cell that contains A. The
shaded zone represents the final cell. The trapezoids which are neither entirely
nor partially shaded are deactivated.
ternal merge. The vertical adjacencies are updated accordingly. External merges
therefore introduce certain links between nodes in distinct secondary influence
graphs.
The clean-up steps ensure that not too many trapezoids are created. Never-
theless, they must not be so frequent that the algorithm becomes inefficient.
We perform clean-up steps after the insertion of the 2 t -th segment, for i =
1,..., [log nJ - 1. Note that the last clean-up step was performed at step pf
where pf is the greatest power of 2 such that 2 pf < n.
Suppose for now that n is a power of 2. We will analyze the complexity of the
algorithm between two clean-up steps p and 2p.
Denote by Si the set of segments inserted during steps 1, ... , 2'. Each trapezoid
T of C(Sp) is subdivided into trapezoids by the segments with which it conflicts.
Let ST stand for the set of segments of S2p that conflict with T, and let ET be
the corresponding chronological sequence. The portion of the decomposition of
the segments in ST that lies inside T has complexity (S1T82).
To make things simpler, we assume that the algorithm does not perform the
external merges. The number of nodes is only greater, so the location phase is
always more complex. The cost of the external merges is proportional to the
number of nodes killed (and hence visited) during the steps, so the external
merges are accounted for by the location phase. The bounds we obtain on the
complexity of the algorithm that does not perform the external merges will thus
still be valid for the algorithm that performs the external merges.
Subsection 5.3.2 provides us with a bound on the number of nodes and the
storage needed by the secondary influence graphs. For a trapezoid T the bound
is O(IS2,I2). To bound the storage required by all the sEcondary influence graphs
computed between steps p and 2p (and ignoring the external merges), we sum
this quantity over all the trapezoids of C(Sp). Here we need a moment theorem
analogous to theorem 4.2.6, but usable in a context where regions are not defined
locally. Such a theorem is stated in exercise 4.4 and bounds this sum by a function
of the expected complexity go (r, Z) of a cell in the arrangement of a random
sample of r segments in a set Z. (Note that this complexity is linearly equivalent
to the complexity of its vertical decomposition.) Therefore, the number of nodes
in all the secondary influence graphs computed between steps p and 2p is
O(IST 12) = ° ((-) go(p, S2p)) = O(p(p)).
The storage needed by the whole influence graph is thus
Llog nj-I
E 0(2 a(2t)) = O(na(n)).
The complexity of the algorithm can be accounted for by three terms that
correspond to the location phase, the update phase, and the clean-up steps. The
location phase is analyzed in much the same way as the storage. We first evaluate
the average number of nodes visited during step p, in the secondary influence
graph rooted at a node that corresponds to a trapezoid T in C(Sp). The node
is also denoted by T for simplicity. As before, we denote by ST (resp. ET ) the
subset (or the chronological sequence) of the segments that conflict with T and
that are inserted before step 2p. Denote by ST (resp. ET) the subset (or the
chronological sequence) of all the segments that conflict with T. Note that 2p
is a random subset of ST. Under the assumption above, it all happens as if we
were locating the segments in the sequence ST in the secondary graph rooted at
T. This graph is the influence graph corresponding to the decomposition of ST
inside the interior of T. A slight adaptation of the proof of theorem 5.3.4 yields an
upper bound on the expected number of nodes visited in the secondary influence
graph. Let fo(r, Z) be the expected number of trapezoids in the decomposition of
a random sample of r segments in a set Z. Then fo(r, Z) = O(r 2 ). If we assume
that S2TP is given, then we may use theorem 5.3.4 to bound the cost of inserting
the last object by
ISI Ifo( Lr/2], S2TP))
This expression bounds the cost of inserting all the segments in S2TP as well as the
cost of locating in the secondary graph rooted at T all the segments in ST \ S2p
that are inserted after step 2 p. The number of nodes visited in the secondary
influence graph during the successive insertions, averaged over all random samples
2p in ST, is thus
0 (E (IST I E 0 ( IST IE( IS2T I))
Hence, the expected number of segments inserted before step 2p that conflict
with T is
E(2S2p) = - IS ( T
since p < n. The average number of nodes visited in the secondary influence
graph rooted at T is finally 0 (nSTI 2 )
Summing over all the nodes of C(Sp), we then obtain a bound on the expected
number m of nodes visited in all the secondary influence graphs rooted at these
nodes:
m=O(PE1ZIST12)
Once again, we can use the adapted moments theorem in exercise 4.4, to obtain
m =O (Z (-) 2 P(p)) = O(na(p)). (15.1)
Summing over all the clean-up steps, we get

[log nj -1
Z O(na(2))= O(na(n)logn).
The update phases and clean-up steps are easily analyzed. Indeed, the up-
date phases require time proportional to the number of nodes created, which
is 0(na(n)). Identifying the trapezoids of C(Sp) during the clean-up step p re-
quires time proportional to the number of trapezoids in C(Sp), which is O(pa(p)).
To deactivate the trapezoids during the different clean-up steps takes time pro-
portional to the number of created nodes, which is again O(nco(n)). The total
cost of the clean-up steps is thus
Ltogn]-1
Z 0(2za(2 )) = O(na(n)).
i- 1
This finishes the proof of the theorem stated below when n is a power of 2. To
analyze the general case, we must also analyze the cost of inserting the segments
at steps 2 pf + 1, . .. , n. But this is word for word the same as the analysis above
and produces the same results (and notably equation 15.1) if we note that Pf > i
Theorem 15.4.4 A single cell in the arrangement of n line segments in the

plane can be computed in expected time O(na(n) log n) and storage O(nc(n)).
15.5 Exercises
Exercise 15.1 (Optimal computation of lower envelopes) Show that the lower
envelope of n line segments in the plane can be computed in optimal time 0(n log n).
Hint: The lower bound Q(nlogn) is proved by reduction to sorting. As for the upper
bound, first project the endpoints of the segments on the x-axis. They define 2n - 1
consecutive non-overlapping intervals. Build a balanced binary tree whose leaves are
the intervals in the appropriate order. To a node corresponds an interval which is the
union of the intervals at the leaves in the subtree. A segment S is assigned to the node
whose interval is the smallest that still contains the projected endpoints of the segments.
(This node is the first common ancestor of all the leaves covered by S.) Show that
the lower envelope of the segments assigned to a single node has complexity 0(m) and
not O(ma(m)), using that there exists a vertical line that intersects all these segments
and using also the result on half-lines mentioned in section 15.3. Observing that the
projections of two segments assigned to different nodes at the same level in the tree
do not overlap, show that the lower envelopes of the segments assigned to the nodes
on a given level of the tree also have linear complexity, and can be computed in time
0(n log n). These 0(log n) lower envelopes can be merged in time O(na(n) log log n),
which is 0(n log n).
Exercise 15.2 (Airport scheduling) Consider a set M of n points in Ed that move
along algebraic curves of bounded degree at given constant speeds. At each moment t,
we want to know the point M(t) in M that is the closest to the origin. Show that the
sequence E of points M(t) for t E [0, +oo[ is almost linear, and that it may be computed
in time 0((n + I3E) logn).
Exercise 15.3 (Computing a view) Consider a scene formed by n line segments in
the plane (not necessarily disjoint). Show that a view from a given point (defined as
the portions of segments visible from that point) may be computed in optimal time
0 (n log n). Similar question for more general objects.
15.5. Exercises 369
Hint: A projective transformation sends the origin to (- oo, 0), and the corresponding
problem is exactly that of computing the lower envelope of n segments, see exercise 15.1.
Exercise 15.4 (Convex hull of objects) Show that computing the convex hull of n
objects in the plane reduces to computing the lower envelope of n functions. If the objects
are convex and disjoint, the graphs of these functions have at most two intersection
points. Give bounds on the combinatorial and computational complexities of convex
hulls of curved objects, in particular circles, ellipses, etc.
Hint: For each object, consider the set of its tangent lines, and use polarity to work in
the dual plane (where points correspond to lines in the original plane).
Exercise 15.5 (Stabbing lines) Given n objects in the plane, compute the set of lines
that simultaneously stab all of them.
Hint: Use the same polarity as in the preceding exercise.
Exercise 15.6 (Motion planning of a polygon under translation) Consider a

polygon M with m sides that moves under translation within a polygonal region £ with
n sides.
1. Show that the set of translations that bring a vertex of M (resp. of E) in contact
with an edge of E (resp. of M) is a set C of mn line segments (identifying the vector OM
of the translation with the point M). Conclude that the set of feasible translations of
M in £ consists of one or several polygonal regions with total complexity O(m2 n2 ). If
M is convex, the complexity is only O(mn) (see exercise 19.8).
2. Show that the set of positions of M in & that are accessible from a given position
I is the cell in the arrangement of C that contains I. Conclude that it is possible to
determine whether two positions I and J are accessible one from the other, and if so,
compute a feasible path for M from I to J in time O(mna(mn) log(mn)).
3. Show that the complexity of the arrangement of C may be as bad as Q(m 2 n2 ) (see
figure 15.8).
4. Show that, in some cases, any path from I to J may have complexity P(mn). For
instance, consider a carpenter's folding rule with m segments, but in a semi-folded rigid
configuration, that tries to pass through n consecutive doors.
Exercise 15.7 (Non-trivial boundary) Consider two connected polygonal regions B

and R, which may have holes. A connected component of the intersection B n 1?
is called non-trivial if its boundary includes at least one vertex of B or R. The non-
trivial boundary of the intersection B n R is the union of the polygons that bound all
the non-trivial connected components of this intersection. Show that the complexity of
the non-trivial boundary of the intersection B n R is O(11I + 17ZI), where IBI and 17Z1
respectively stand for the number of sides of S and R. Hence any cell of the intersection
of two polygonal regions B and 1R has complexity O(B1f + IRl).
Figure 15.8. There are Q(m 2 n2 ) feasible positions of M that belong to the same number
of distinct cells of C.
Hint: Put V = B n 1Z for the intersection of B and R. Since R and B play entirely
symmetric roles, it suffices to look at the contribution of the boundary of B to the non-
trivial boundary of V. For this, follow the edges on the boundary of B, and count the
number of edges of V contained in each edge of B.
For each edge E on the boundary of B, count the edges of V contained in E. We
distinguish the first one along E from the others. Among the others, count separately
those that belong to the same connected component of V, those that do not belong to
the same connected component of the boundary of V, and the remaining edges.
Exercise 15.8 (Computing the non-trivial boundary) Show that the non-trivial
boundary of two polygonal regions B and 1t (see exercise 15.7) can be computed in time
O(m log m), if m is the total number of edges of B and RZ.
Hint: We sweep the plane with a line going in two directions, first going from left to
right and then from right to left. During the sweep, we maintain three structures which
respectively represent the segments of B, of 7?, and of the resulting non-trivial boundary
that intersect the sweep line. During the left-to-right sweep, we only create a new interval
for the result when the current event is a vertex of B contained in 7?, or a vertex of 7?
contained in B. We call such a vertex a remarkablevertex. Then we are assured that this
interval is contained in a non-trivial cell. We do not discover the entire non-trivial cell,
however, rather we only know the portion of this cell that can join a remarkable vertex
by a decreasing x-monotone path. This is why we need to sweep the plane in the other
direction, from right to left.
Exercise 15.9 (Computing a cell) Devise a deterministic algorithm that computes a

single cell in the arrangement of n line segments in the plane, in time O(n log2 n). The
cell is characterized by a point A that belongs to it.
Hint: Use the divide-and-conquer method. Split the set of n segments into two subsets
of roughly the same size to obtain two cells CIA and C2A in the sub-arrangements that
contain A. Merging these two cells can be done using a variant of the sweep method of
exercise 15.8. This variant in fact computes the non-trivial boundary of the intersection
CA n CA as well as the boundary of the cell in this intersection that contains A, even if
it does not belong to the non-trivial boundary. It remains to extract the description of
the cell CA in the current divide-and-conquer step that contains A.
Exercise 15.10 (Half-lines) Show that the complexity of a cell in the arrangement of
n half-lines is 0(n). Devise a deterministic algorithm that computes it in optimal time
0(n log n).
Hint: Applying a rotation if necessary, we may assume that no half-line is vertical.

Suppose that the cell is characterized by a point that belongs to it. Distinguish between
the subset £+ of the half-lines that intersect y = +00 and the subset £- of the half-
lines that intersect y = -oo. For each of these subsets, we explain how to compute
the unbounded cell that contains the origins of some half-lines. (The other cells can be
computed in a similar way.) For £-, we compute a left tree and a right tree by sweeping
the plane from top to bottom with a line parallel to the x-axis. At each intersection I
between two half-lines, we keep only the portion of the half-lines which lies to the left
(for the left tree) or to the right (for the right tree). The boundary of the unbounded
cell of £- is obtained by computing the boundary of the unbounded cell of the union of
both trees. Finally, exercise 15.8 can be used to compute the intersection of the cells of
£+ and £- that contain A.
Exercise 15.11 (Curved arcs) Bound the complexity of a cell of an arrangement of

curved arcs in the plane and devise an algorithm that computes the cell that contains
some given point.
Exercise 15.12 (Manipulator) A planar manipulator is formed by two rigid bodies

articulated in a point A. One body is fixed to the origin 0. The manipulator has two
degrees of freedom: a rotation around 0 and a rotation around A. The configuration of
the manipulator is parameterized by the corresponding two angles. Given some obstacles,
devise an algorithm that computes the set of configurations for which the manipulator
does not collide with an obstacle. Devise also an algorithm that determines whether two
positions are reachable one from the other and, if so, outputs a path that realizes this
change of configuration.
Hint: Express the constraints that limit the motion of the manipulator in the configu-
ration space (which has dimension 2) and use exercise 15.11.

The connection between lower envelopes of functions and Davenport-Schinzel sequences
was established in a paper by Davenport and Schinzel [74]. Atallah [13], then Sharir and
collaborators [1, 122, 206] proved bounds on the length of Davenport-Schinzel sequences.
Wiernik and Sharir [219] showed how to realize a lower envelope of n segments in the
plane that has complexity Q(na(n)). The solution to exercise 15.1 is due to Hershberger
[124].
The analyses of the complexity and of the computation of the unbounded cell in the
arrangement of line segments are given by Pollack, Sharir, and Sifrony [189]. Their result
is extended by Guibas, Sharir, and Sifrony [118] to the case of a cell in the arrangement
of curved arcs. Other results on curved arcs are given in [64, 90]. The complexity
and computation of m cells is studied by Edelsbrunner, Guibas, and Sharir in [91, 93].
Solutions to exercises 15.7, 15.8, and 15.9 can be found in their papers.
Alevizos, Boissonnat, and Preparata [7] study the arrangements of half-lines and give
a solution to exercise 15.10. These arrangements find applications in pattern recognition
[8, 33].
The randomized algorithm that computes a single cell described in this chapter is
due to de Berg, Dobrindt, and Schwarzkopf [76]. This algorithm can be generalized to
dimension 3, which is not the case for a previous algorithm due to Chazelle et al. [52].
A comprehensive survey of Davenport-Schinzel sequences and their geometric appli-
cations can be found in the book by Sharir and Agarwal [207].
Chapter 16
Arrangements of triangles
The questions of chapter 15 can also be asked in spaces of dimensions greater

than 2. Unfortunately, there are very few efficient algorithms for dimensions
greater than 3 and so this chapter studies problems in dimension 3 only. Known
extensions to higher dimensions and their corresponding references are given in
the exercises and in the bibliographical notes.
The notion of an arrangement of segments in the plane can be extended to
that of an arrangement of triangles in E3. Given a set T of n triangles in E3 ,
its arrangement is formed by the faces of dimensions 0, 1, 2, and 3 (respectively
called vertices, edges, facets, and cells) of the spatial subdivision of E3 induced
by T, and their incidence relationships.
It is easy to see that the complexity of an arrangement A of n triangles in E3
is O(n 3 ) in the worst case, which is the same as that of an arrangement of n
planes. In this chapter, we are mostly interested in the complexity analysis and
computation of only parts of A: the lower envelope of a set of triangles and the
cell of the arrangement that contains a given point.
The algorithm that computes the lower envelope of a set of triangles has an
important application in computer graphics since it allows us to compute an
orthographical view of a polyhedral scene. It is also useful in robotics (see exercise
16.12) and for computing Voronoi diagrams for L1 and Loo norms as is shown
in section 18.4. Computing a cell is essentially motivated by the computation
of the free space of a polyhedral robot moving by translation among polyhedral
obstacles.
The main problem that arises when trying to extend the results of the preced-
ing chapter to a space of dimension higher than 2 is the difficulty of defining a
decomposition scheme into elementary cells that has a complexity roughly com-
parable with that of the arrangement. This question is tackled in subsection 16.2
and again in subsection 16.4.2. Section 16.3 presents the known results concern-
ing envelopes of triangles. Finally, the complexity of a cell in the arrangements
374 Chapter 16. Arrangements of triangles
of triangles is bounded in section 16.4, and we describe an algorithm that com-

putes it which is similar to that presented in the previous chapter for the case of
dimension 2.
16.1 Faces in an arrangement

Given a set T of n triangles in E3, the arrangement of T is formed by faces of
dimension 0, 1, 2, and 3 (respectively called vertices, edges, facets, and cells) of
the subdivision of E3 induced by T, and by the adjacency relationships between
these faces. To define a cell in an arrangement of triangles, we need to distinguish
the two sides of a triangle, in other words to consider each triangle as an infinitely
thin flat prism whose boundary contains two parallel copies of the triangles at an
infinitesimal distance. We still speak of the interior of a triangle or face, and we
mean its relative interior.
More precisely, a cell in the arrangement is the topological closure of a con-
nected component of E3 \ T; a facet of a cell is the topological closure of a
connected component of the intersection of the boundary of the cell with a trian-
gle of T; the vertices and edges of the arrangement are the vertices and edges of
the polygons that bound the facets. An edge in the arrangement is contained in
the edge of a triangle of T or in the intersection of two triangles. A vertex of the
arrangement is either a vertex of a triangle in T, the intersection of a triangle in
T and the edge of another triangle in T, or the intersection of three triangles in
T.
16.2 Decomposing an arrangement of triangles

In several applications, it is useful to decompose the cells of an arrangement into
elementary cells that are convex with bounded complexity. In chapter 12, we
showed how to triangulate or decompose a polygonal region. For the case of
line segments in the plane, each cell in the arrangement can be considered as a
polygonal region, so the problem can be solved. The resulting decomposition has
the same complexity as the arrangement itself, up to some multiplicative factor.
In E3 , the situation is somewhat complicated by the fact that triangulating a
polyhedron with n vertices may require up to Q(n 2 ) tetrahedra in the worst case
(see exercise 13.4). The number of vertices in the arrangement of triangles can
be cubic, leading to a very onerous decomposition.
Subsection 16.2.1 presents a better solution obtained by generalizing the idea in
the decomposition of line segments and by using Davenport-Schinzel sequences.
The result is a vertical decomposition whose complexity is very close to that of
the arrangement itself. We have already used a similar construction to triangulate
16.2. Decomposing an arrangement of triangles 375
a polyhedron (see section 13.3). Subsection 16.2.2 then shows how to adapt the
vertical decomposition to subdivide all the non-convex cells in the arrangement of
n triangles into 0(n 2 ) convex cells. The vertical decompositions are very useful,
as will be shown in several algorithms described in this chapter. They also allow
arrangements of triangles to be triangulated, a very easy operation when the
vertical decomposition is known.
Henceforth, T will stand for a set of n triangles in E3. We assume that no
triangle in T is parallel to the z-axis.
16.2.1 Vertical decomposition

In this subsection, as before, each triangle has two sides: the upper side and the
lower side. This is equivalent to considering a triangle as an infinitely thin flat
prism.
The decomposition that we describe below is a very close relative of the decom-
position of polyhedral regions described in subsection 13.3.1. From each point P
on an edge of the arrangement, we draw two vertical segments (parallel to the
z-axis) upwards and downwards until they hit another triangle in T.
A connected set of vertical segments whose endpoints belong to a given edge in
the arrangement and to a given triangle in the arrangement constitutes a vertical
trapezoid which we call a 2-wall of type 1. These vertical walls decompose the
cells of the arrangement into cylindrical subcells with at most two non-vertical
faces: an upper face called the ceiling of the cell, and a lower face called its
floor. The decomposition process is not yet over, because the cylindrical subcells
may have many vertical faces and may not even be convex or may have holes
(see figure 16.1). To get convex cells of bounded complexity, we subdivide each
cylindrical subcell C again in the following manner.
1. We decompose the floor Fl(C) of C.1 This decomposition is obtained by
drawing, from every vertex of Fl(C) towards its interior, a line segment
of maximal length that is parallel to the projection of the y axis on the
non-vertical plane containing Fl(C). These segments form the 1-walls of
the floor and have both endpoints on the boundary of Fl(C). This decom-
position is shown in bold dotted lines in figure 16.1.
2. From each point P on a 1-wall of the floor of C, draw a vertical line segment
towards the interior of C until it meets the ceiling. The set of vertical
segments that pass through a given 1-wall of Fl(C) constitutes a trapezoid
which we call a 2-wall of type 2.
Each cylindrical cell is processed separately, so we must distinguish the two sides
of a 2-wall of type 1 (just as we did for the simplified decomposition of a set of
'The cell C is cylindrical, so decomposing the ceiling would yield exactly the same result.
,:
Figure 16.1. Decomposing an arrangement of triangles. The walls of type 1 are represented
by light lines, and the traces of walls of type 2 on the floor of the cell are
represented in dashed bold.
segments in section 3.3). Each side of a 2-wall of type 1 is subdivided by 2-walls

of type 2 into new walls which we call 2-walls of type 1'. All cells in the resulting
decomposition are convex and have bounded complexity: they have at most six
vertical sides and an upper and lower face, and all these faces are trapezoidal.
We call such cells prisms.
To estimate the number of prisms in this decomposition, we compute the num-
ber of 2-walls, and observe that any prism has at least one 2-wall and that a
2-wall belongs to at least two prisms. The number of walls of type 1' is at most
the sum of the numbers of walls of types 1 and 2. The number of 2-walls of
type 2 is at most twice the number of 2-walls of type 1. Indeed, each vertex of
the floor FI(C) of a cylindrical cell C is the endpoint of at most two 1-walls of
Fl(C), and the number of vertices of Fl(C) also equals the number of edges of
the vertical faces (2-walls of type 1) of C. Hence it is enough to count all the
vertical walls of type 1. Let E be an edge of some triangle or the intersection of
two triangles, and let 1? be the vertical strip generated by all the vertical lines
passing through a point of E. We denote by lZh the part of 1R that lies above
E and by iZi the part that lies below. Let Sh (resp. Si) be the set of segments
obtained as the intersections of triangles in T and RZh (resp. RI). The vertical
edges of the walls of type 1 that have an endpoint on E are precisely the vertical
16.2. Decomposing an arrangement of triangles 377
Figure 16.2. Vertical faces that intersect the edge E.
line segments that join a point of E and a vertex on the lower envelope (in 7?)
of Sh or on the upper envelope of S1. From theorem 15.3.1, we derive that the
number of 2-walls of type 1 that are incident to E is 0(na(n)). If these triangles
do not intersect, this number is 0(n).
Summing over all the segments E that are either edges of a triangle in T or the
intersection of two triangles in T, we find that the total number of walls of type 1
(and hence also the number of walls of types 2 and 1') is 0((n +p)na(n)), if the
number p of intersecting pairs of triangles is not zero. If it is, then the number
of walls of type 1 is 0(n 2 ). The theorem below summarizes this result.
Theorem 16.2.1 For any arrangementof n triangles in E3 , with > 0 intersect-

ing pairs of triangles, there exists a decomposition of complexity 0(n(n+p)a(n)).
If p = 0, the complexity of the decomposition is 0(n 2 ).
In the worst case, we have p = 0(n2 ) and the decomposition has complexity
0(n 3'(n)), which is very close to the optimum, since the arrangement itself has
complexity Q(n 3 ) in the worst case.
One may also bound the complexity of the decomposition by 0(n 2 a(n) log n +
t), where t is the complexity of the arrangement (see exercise 16.4). This bound
is better than that of the previous theorem when p = Q£(n log n).
16.2.2 Convex decomposition

In this section, we show how to decompose the non-convex cells in the arrange-
ment A of a set T of n triangles into 0(n 2 ) convex parts. Overall and in the
worst case, we obtain a decomposition into 0(n 3 ) convex parts for the whole
arrangement. The proof is constructive and is based on building a portion of the

vertical decomposition of the non-convex cells.
A face in the arrangement A is said to be outer if it is contained in a face on
the boundary of a triangle in E. Otherwise, it is an inner face. Note that the
boundary of a cell contains an outer face if and only if the cell is non-convex.
Theorem 16.2.2 In the arrangement of n triangles in E3, the union of the non-
convex cells can be decomposed into 0(n 2 ) convex parts. The union of all the
cells can therefore be decomposed into 0(n 3 ) convex parts.
Proof. As indicated above, only the first statement needs a proof. For the
second statement, simply note that the cells that do not contain outer faces are
convex, and that they form a subset of the cells of the arrangement of the planes
that support the triangles. They must therefore have overall complexity 0(n 3 ).
For the non-convex cells, we first show how to decompose the non-convex cells of
the arrangement of n line segments into 0(n) convex parts. This is achieved as
for the usual vertical decomposition of a set of line segments, except that we draw
only the walls from the endpoints of the segments and not from their intersection
points. The cells in the decomposition induced by the walls and the segments
are convex, since the only non-convex vertices of a cell in the arrangement of
line segments are the endpoints of the segments. There are at least one and
at most two walls incident to each non-convex vertex, so there are only twice
as many convex parts as there are endpoints, hence 0(n) convex parts in this
decomposition.
Consider now the case of dimension 3. We decompose the arrangement in a
way similar to the case d = 2. More specifically, we consider each edge on the
boundary of a triangle in T. Through each point P of such an edge we draw a
vertical line segment (parallel to the z-axis) upwards and downwards until it hits
a triangle in T other than that which contains P. In this way, we build 2-walls of
type 1. Note that, contrary to the decomposition described in subsection 16.2.1,
we do not build walls of type 1 on top of the edges of the arrangement that
are contained in the intersection of two triangles. Each cell in the resulting
decomposition is a vertical cylinder, bounded above and below by convex domes.
The cylinders are not necessarily convex since their horizontal projections are
generally not convex either. Yet the reflex edges must be vertical, and they
correspond to segments drawn on top of the vertex of some triangle, or on top
of the intersection of a triangle and an edge. We obtain convex parts by adding
more 2-walls of type 2. More precisely, if II(C) is the projection on the xy-plane
a non-convex cell C, then II(C) may be decomposed into convex parts as we did
for the two-dimensional case, by building 1-walls parallel to the y-axis for each
non-convex vertex of II(C). These vertices are vertices of the triangles, or the
intersection of a triangle and an edge. We then build vertical 2-walls of type 2
16.3. The lower envelope of a set of triangles 379
Figure 16.3: Walls in the convex decomposition induced by the triangles in a vertical plane H.
passing through these 1-walls. Each side of a 2-wall of type 1 is subdivided by

the 2-walls of type 2 into new 2-walls which we call of type 1'.
Each cell now reduces to a convex truncated cylinder whose generators are
vertical lines. The number of cells is proportional to the number of 2-walls.
We first estimate the number of 2-walls of type 1. In the vertical plane H that
contains an edge E of a triangle in T, the line segments formed by the intersections
of the triangles in T with H induce an arrangement AH. The 2-walls of type
1 which intersect H decompose the non-convex cells of AH incident to E into
convex parts (see figure 16.3). The 2-walls of type 1 contained in H are either
cells of this decomposition, or convex cells of AH incident to E. The former are
in number O(n) as we saw above, and the latter correspond to a subset of the
intervals in the decomposition of E by the triangles in T. The number of 2-walls
of type 1 in a plane H is thus O(n), accounting for O(n2 ) walls of type 1 overall.
(There are O(n) possible planes H.) There can also exist only 0(n 2 ) walls of
type 2, because the number of 2-walls of type 2 erected in a cell C is proportional
to the number of vertices of II(C), which is itself proportional to the number of
vertical edges in C and thus to the number of walls of type 1 in C. Finally, the
number of walls of type 1' equals the sum of the numbers of walls of types 1 and
2. This finishes the proof of the theorem.
16.3 The lower envelope of a set of triangles

We now consider a set T = Tl,... , T of n triangles in E3. For the sake of
simplicity, we will assume that the triangles are not vertical, so that any vertical
-f"
z
Figure 16.4. The lower envelope of five triangles, as seen from below projected onto the
xy-plane.
line intersects any triangle in at most one point.

Each triangle can be considered as the graph of a linear function z = Ti(x, y)
defined over a triangle in 1R2. We define the lower envelope £ of the triangles in
T as the graph of the function minl<i<, Ti(x, y) (see figure 16.4). The connected
components of the intersection of the envelope with a triangle of T are called
the facets of the envelope. These facets are polygonal regions whose vertices and
edges are vertices and edges of the lower envelope. The complexity of the lower
envelope is the total number of its faces.
The faces of the lower envelope can be projected into the xy-plane, and this
yields a planar map S whose cells correspond to the facets of the lower envelope.
We will assume that the triangles are in general position, by which we mean that:
1. two triangles do not overlap (their intersection cannot be two-dimensional),
2. an edge of a triangle intersects other triangles only in their relative interiors

(not on an edge or at a vertex),
3. a vertex of a triangle does not belong to another triangle,
4. the projections of any two edges of the triangles do not overlap, and
5. the projections of any three edges of triangles do not intersect in a common

point.
The general position assumption is not necessarily satisfied, especially when

the triangles are facets of a polyhedron. By a standard perturbation argument,
however, the triangles can be slightly perturbed into a general position while only
augmenting the number of faces on the lower envelope. Therefore, the upper
bounds we give below on the complexity of lower envelopes of triangles in general
position apply to degenerate configurations as well.
We note that, under this general position assumption, the number of vertices
(edges) on the envelope is at most twice the number of vertices (edges) of the
planar map £ obtained by projecting the faces of the lower envelope on the xy-
plane. This is because a vertex (edge) in this planar map is the projection of at
most two vertices (edges) of the lower envelope.
16.3.1 Complexity
This subsection is devoted to proving the following theorem, which bounds the
complexity of the lower envelope of triangles:
Theorem 16.3.1 The lower envelope of n triangles in E3 has worst-case com-

plexity E(n 2 a(n)). When the triangles do not intersect, this worst-case complexity
drops to e(n 2 ).
Proof of the upper bound. We first count the number s(,f) of vertices in the
planar map - obtained by projecting the faces of the lower envelope £ on the
xy-plane. For this, we consider the lower envelope £' of T \ {T}, T e T, and we
estimate the increase s(&) - s(£f) in the number of vertices of the map when T
is reinserted.
The new vertices of the envelope (which are vertices of £ but not of £') are either
on T or vertically below an edge of T. The triangles being in general position,
the new vertices in the map - are the projections of points of the following kinds:
1. a vertex of T,
2. the intersection of an edge of T with the interior of a triangle in fT\{T},
3. the intersection of the interior of T and an edge of a triangle in fT\{T},

4. a point on an edge of T that lies vertically above an edge of a triangle in
T \ {T},
5. the intersection of the interior of T and an edge of £' that is contained in

the intersection of two triangles in T\{T},
6. a point on an edge of T that lies vertically above an edge of £' that is

contained in the intersection of two triangles in T\{T}.
The vertices of types 1-4 do not create problems: there are at most O(n) of
them. When we insert T, each edge of £' is contained in the intersection of two
(a) (b)
(c) (d)
Figure 16.5. Inserting a triangle: an edge can be (a) hidden, (b) unchanged, (c) shortened,
or (d) split into two edges.
triangles, and can be (a) hidden by T, (b) unchanged, (c) shortened, or (d) split
into two edges of E (see figure 16.5). Only in case (d) does the number of vertices
in the projected map increase. More precisely, there are two new vertices and at
least one of them is of type 6. The number of vertices increases by at most twice
the number of vertices of type 6. Their number is O(na(n)). Indeed, if E is an
edge of T and H is the vertical plane strip formed by the vertical rays originating
from E and going upwards, the triangles in T \ T intersect H along segments,
and the vertices of type 6 contained in E are vertices of the lower envelope of
these segments in H. Theorem 15.3.1 shows that there are O(nce(n)) of them.
This is true for all three edges of T, so there are at most O(nce(n)) new vertices.
Therefore
s(E) = s(e') + O(na(n)).
By inserting all the triangles in T successively, and denoting by s(n) the maxi-
mum number of vertices in the planar map projection of the lower envelope of n
triangles, we get the recurrence
s(n) < s(n - 1) + O(na(n)),
which solves to s(n) = O(n2 a(n)). Under the general position assumption, each
vertex of the projected map £ has degree 2 or 3, so that the map has O(n2 a(n))
edges as well, and Euler's relation yields the same bound for the number of its
cells. As we saw above, the same bounds apply for edges and facets of S.
Figure 16.6. Configuration for the lower bound.
If the segments do not intersect, only cases (a) and (d) above are allowed, and
they account for O(n) new vertices. Hence a similar recurrence shows that the
complexity of the lower envelope is simply O(n 2 ).
Proof of the lower bounds. To construct the example, we use the fact men-
tioned above that the lower envelope of n segments may have Q(na(n)) edges
(see section 15.6). Now take 2 segments in the half-plane y = 0, z > 0, so that
their lower envelope has E)(na(n)) faces. Picking a point A far enough away on
the y-axis, we can construct ' triangles almost parallel to the y-axis by taking
the convex hull of these segments with the point A. The idea is to duplicate this
complexity e)(na(n)) a number of times: place ' disjoint, thin triangles in the
plane z = 0, long enough so that each intersects from side to side the vertical
projections of the '2 triangles constructed above. Then the lower envelope of
these n triangles has complexity E(n 2ae(n)).
A similar construction when the triangles do not intersect leads to the Q(n 2 )
bound.

We define the vertical decomposition Dec(g) of the lower envelope £ of a set T of
triangles as the portion of the entire vertical decomposition of T that lies below
the lower envelope.
The decomposition Dec(&) projects vertically onto the xy-plane along the de-
composition of the planar map £ whose faces are the projections of the faces of
the lower envelope £: the cells in this decomposition are trapezoids, sometimes
degenerate (unbounded or triangles, see section 3.3 and subsection 12.3.3). To
each trapezoid in the planar map corresponds a vertical cylinder whose horizon-
tal section is the trapezoid: such a cylinder is called a prism. Each prism is
unbounded in the direction of the negative z-axis, and is bounded in the opposite
direction by a facet contained in a single triangle of T. This facet is called the
ceiling of the prism. The collection of these prisms and of their faces constitutes
the vertical decomposition of the lower envelope of the triangles. If T consists
of n triangles, the complexity of this decomposition is O(n 2 oa(n)), and so is the
complexity of the lower envelope. Note that the ceiling of a prism C is a trapezoid
of the decomposition of a set Sc of segments contained in the triangle Tc that
contains the ceiling of the prism. This set SC is formed by
1. the edges of Tc,

2. the intersections of TC with other triangles of T,
3. the vertical projections onto Tc of edges or portions of edges of triangles
that lie below Tc.
Among the 2-walls in the decomposition of the lower envelope, we make a distinc-
tion between the 2-walls of type 1 hanging below the edges of £ and the 2-walls
of type 2 hanging below the 1-walls of the planar map S. By construction, the
non-vertical edges of the walls of type 2 (which are edges of the ceilings of the
prisms) all contain a vertex of E.
We note that a prism may be adjacent to many prisms since both sides of a
wall of type 1 may be subdivided differently by the abutting walls of type 2 (see
also the discussion in subsection 16.2.1). Nevertheless, the adjacency graph is
planar and its complexity is linear with respect to the number of prisms.
16.3.3 Computing the lower envelope

The algorithm we present here is incremental and uses the influence graph method
described in chapter 5. The analysis is randomized. To simplify the description
of the algorithm, we suppose that the plane z = +oo is a triangle in T and that
it is inserted first.
The triangles are inserted one after the other, and the algorithm maintains
the vertical decomposition of the current lower envelope as well as the adjacency
relationships of the prisms in the decomposition.
To fit the framework of randomized algorithms, we redefine the problem in
terms of objects, regions, and conflicts. Here, the objects are triangles and the
regions are prisms (vertical cylindrical cells with horizontal trapezoidal sections).
A region is defined by a set of triangles if it is a prism in the decomposition of the
lower envelope of a subset of triangles. A prism C is determined by the triangle
Tc that contains its ceiling, and by the at most four edges of Sc contained in
Tc that determine the edges on its ceiling. Each segment being either an edge
of TC, the intersection of TC and a triangle in T \ {Tc}, or the projection on Tc
of an edge of a triangle in T \ {Tc}, it follows that a region is determined by at
most five triangles.
An object conflicts with a region whenever they have a non-empty intersection.
Observe that the regions that are defined by the triangles of T and do not conflict
with these triangles are exactly the prisms in the decomposition of the lower
envelope of T.
Let T be a new triangle inserted into T, let £' be the current lower envelope
before inserting T, and let Dec(&') be the current decomposition. We first identify
the prisms in Dec(e') that intersect T, then we update the lower envelope and
its decomposition. Let us call £ the new lower envelope after updating and let
Dec(g) be its decomposition. To efficiently find the regions that conflict with
T, the algorithm also maintains an influence graph. We may recall that the
influence graph is a structure whose goal is to detect rapidly the conflicts between
a new object and the regions defined and without conflict over the current set
of triangles. The influence graph is an oriented acyclic graph that has a node
for each region that was a region defined and without conflict over a subset of
the set of triangles; this subset was the current set of triangles during a previous
incremental step. The arcs in the influence graph connect the nodes in such a way
that the influence domain of a node (the subset of objects that conflict with the
region that corresponds to this node) is contained in the union of the influence
domains of its parents. At each step of the algorithm, the regions defined and
without conflict over the current subset are stored in the leaves of the influence
graph.
We now describe how to perform the current incremental step by performing a
location phase and an update phase.
Locating. The location phase is used to retrieve all the leaves in the influence
graph that conflict with T. These leaves correspond to prisms in Dec(V') in-
tersected by T. A simple traversal of the influence graph that backtracks each
time it encounters a node that either was already traversed or that does not con-
flict with T identifies all the nodes that conflict with T. If no leaf in the graph
intersects T, then T does not appear on the lower envelope £ or on any lower
envelope that will subsequently be computed. The algorithm may skip updating
the structure and directly insert the next segment.
Updating. Among the prisms in Dec(&') that intersect T, we make a distinc-

tion between the prisms that are split by T into two distinct connected compo-
nents, and the others of which we say that they are pierced by T.
We first construct all the prisms in the decomposition Dec(E) of the new lower
Figure 16.7. The six sub-prisms whose ceiling is not supported by T, projected onto a
horizontal plane.
envelope £ whose ceiling is not supported by T. Consider a prism C pierced by

T. The portion of C that appears on £ is subdivided into at most six sub-prisms
with a trapezoidal section (see figure 16.7). A node in the influence graph is
created for each of the nodes corresponding to these six sub-prisms, and these
nodes become children of the node corresponding to C.
Some of these nodes are only temporary, as we will see shortly. Indeed, some
sub-prisms may share a wall of type 2 that has to be removed, and the prisms have
to be merged to obtain the decomposition of E. This merge process is analogous
to the merges performed by the incremental algorithm that builds the planar
vertical decomposition of segments (see section 3.3), and can be described as
follows. Each wall of type 2 in Dec(&') that is intersected by T is subdivided into
at most four sub-walls: a sub-wall above T, another below T, and occasionally a
sub-wall on each side of T (see figure 16.8). The first sub-wall must disappear, the
second will be processed later. As for the occasional last two sub-walls, neither
of their upper edges is contained in T and the one that does not contain a vertex
of the lower envelope must be removed, because it does not induce a wall in the
new subdivision. The two incident sub-prisms are merged and the adjacency
graph is updated accordingly. In the influence graph, the temporary nodes that
correspond to the sub-prisms are merged into a single node that inherits the
parents of all these sub-prisms.
In this way, we have computed all the prisms whose upper face is not contained
in T. It remains to compute the other prisms, whose union U is a portion of the
union of the prisms that conflict with T and whose ceilings are supported by T.
This union U is computed using the adjacency graph of the prisms. U is not
connected in general, as is shown for instance in figure 16.9.
Let T be a connected component of this union U. Then F corresponds to a
face F of the lower envelope £ that is supported by T. The prisms in the decom-
position of £ whose ceilings are supported by F result from the 2-decomposition
of F along the y'-axis in F that projects onto the y-axis in a horizontal plane. To
z
PI
Figure 16.8. Merging the adjacent temporary prisms: the shaded triangle is inserted and a
wall is subdivided into at most four sub-walls. The projected decomposition
of the lower envelope is represented above, and the walls to be removed are
shown in dashed lines.
;r
Y t
x
Figure 16.9. The union U of the prisms intersected by T, shaded and shown in projection
in the xy-plane. The vertical decomposition in the facets supported by T and
induced by the triangles is shown in dashed lines.
maintain the property that the influence domain of a child is contained within
the union of the influence domains of its parents, each prism should be attached
to all the prisms that conflict with T and intersect F; this would result in an
unbounded number of children for a node, and the usual randomized analysis
does not apply in this case. We remedy this drawback by creating a new node
N in the influence graph that corresponds to a prism whose ceiling is F. This
node is of a special kind and, unlike the other nodes, does not correspond to a
prism in the decomposition of the lower envelope. We attach N to the nodes in
the influence graph that correspond to prisms that conflict with T (either pierced
or split) and that intersect F. Face F (and hence the prism corresponding to
N) may have a large number of edges and may be not simply connected. It can
be decomposed using the planar randomized incremental algorithm (in the affine
hull of F) and we hang 2-walls vertically below the 1-walls of F. In this way,
we obtain a secondary influence graph that represents the planar decomposition
in the facet F, and hence a decomposition of the space lying below F. This
secondary influence graph is rooted at N.
We make a distinction between the primary nodes in the influence graph of the
triangles, which we also refer to as the primary influence graph, and the secondary
nodes in the secondary influence graphs. Primary nodes correspond to a region
defined and without conflict over a subset of the current set of triangles; this
subset was the current subset during some previous incremental step.
Recall that U is the union of the sub-prisms obtained by subdividing the prisms
that conflict with T and whose ceilings are supported by T. At the i-th step, the
number u of edges on the boundary of U is bounded by the number of primary
nodes that conflict with T. Moreover, lemma 5.2.4 implies that the average
number of secondary nodes created at step i is proportional to u. The expected
number of secondary nodes is thus proportional to the number of primary nodes.
Therefore it suffices to bound the number of primary nodes. These nodes are
in one-to-one correspondence with the prisms defined and without conflict over
the current subset at some previous incremental step.
Even though the situation does not exactly fit the framework of theorem 5.3.4,
because of the existence of secondary nodes, we can still apply its proof verbatim
to count the number of primary nodes created, and the number of primary nodes
visited during a location phase. So we conclude that the number of primary nodes
in the influence graph is
o (E /(r, T ))
where fo (r, T) is the number of prisms defined and without conflict over a random
sample of r triangles in T. Each node in the primary influence graph has a
bounded number of children, so this bound is also valid for the number of arcs in
the primary influence graph. Theorem 16.3.1 states that fo(r) = O(r 2 a (r)) and
16-3. The lower envelope of a set of triangles 389
so the total number of primary nodes created during the incremental insertions
of n triangles is 0(n 2 a(n)).
We now have to estimate the costs incurred by maintaining the secondary
influence graphs. The results of subsection 5.3.2 that refer to the construction
of the decomposition of a set of segments in the plane show that the location
in the secondary graph takes expected time O(logn). Likewise, the secondary
graphs can be computed in expected time O(nT log nT) where nT is the number of
primary nodes that conflict with T. Therefore the costs of updating and locating
in the secondary graphs are proportional by a factor of logn to the number of
primary nodes visited during an incremental step.
The number of primary nodes visited upon inserting the i-th triangle is given
by theorem 5.3.4:
0 (E f(Lr/2) T) = (n)).
We conclude that the costs of locating and updating the structures when inserting
the n-th triangle are both 0(na(n) log n).
0 2 ) and the cost of inserting
If the triangles do not intersect, we have fo(r) 0(r
the n-th triangle becomes 0(n log n).
We have not yet analyzed the cost of updating the adjacency graph. This
cost is not bounded easily since the algorithm keeps track of all the adjacencies
between prisms and yet a prism may be adjacent to several others. Nevertheless,
an argument similar to that presented in exercise 5.3 shows that the cost of these
updates is also 0(na(n)).
This finishes the analysis of the algorithm. Our findings are summarized in the
following theorem.
Theorem 16.3.2 The lower envelope of a set of n triangles in E3 can be com-

puted using an on-line algorithm that maintains an influence graph. The expected
time required for inserting the n-th triangle is 0(na(n) log n), or 0(n log n) if the
triangles do not intersect.
In several cases, fo(r, T) = 0(r) holds, for instance when studying certain
kinds of Voronoi diagrams (see section 18.4). Using this better bound in the
above discussion implies:
Corollary 16.3.3 Let T be a set of n triangles in E3 , such that the expected

number of faces of the lower envelope of a random sample of r triangles in T
has complexity 0(r). Then the lower envelope of T can be computed using an
on-line algorithm that maintains an influence graph. The expected time required
for inserting the n-th triangle is 0(log2 n).
16.4 A cell in an arrangement of triangles

16.4.1 Complexity
As we saw above, the complexity of a lower envelope of a set T of n triangles in 1E3

is bounded below by Q(n2a(n)) (see theorem 16.3.1). This bound is also a lower
bound on the complexity of a cell in the arrangement of n triangles in E 3. Indeed,
it suffices to add to each triangle T in T three planar facets obtained by casting
vertical rays in the direction of the increasing z-coordinates from each point of
the edges of T. The lower envelope of T corresponds to the non-vertical faces of
the unbounded cell that lies below all the triangles in T in the arrangement of T
and all the added triangles.
An upper bound that is very close to this lower bound is given by the following
theorem.
Theorem 16.4.1 The complexity of a single cell in an arrangement of n trian-

gles in E3 is O(n 2 log n).
The remainder of this subsection is devoted to proving a slightly worse bound

O(n2 log 2 n). Refining this analysis yields the bound stated in the theorem (see
also the bibliographical notes at the end of this chapter.)
The proof we present here is similar in spirit to the proof of the zone theorem
14.3.2. It is split into several lemmas that prove that the number of edges of a
cell in the arrangement of n triangles in E 3 is O(n 2 log2 n). Euler's relation then
finishes the proof.
Let T be a set of n triangles in E3 , and pick a point 0 in E3 that does not
belong to any triangle in T. Our goal is to bound the complexity of the cell C
in the arrangement of T that contains 0. We assume that the triangles are in
general position (see subsection 16.3.1), since a standard perturbation argument
shows that the complexity is maximized in that case.
As in subsection 16.2.2, we say that a face of C is outer if it is contained in
an edge on the boundary of a triangle in T. Otherwise, we say that the edge is
inner. We note that the outer faces are of dimension 0 or 1, and that an inner
k-face (k < 3) is contained in the intersection of the interiors of 3 - k triangles.
Lemma 16.4.2 The number of outer vertices and edges contained in a single
cell is 0(n 2 ).
Proof. Each edge of a triangle is subdivided by the other triangles of T into at

most n segments. Therefore, there can be no more than 0(n 2 ) edges. But there
are at most two outer vertices per outer edge. El
16.4. A cell in an arrangement of triangles 391
Figure 16.10. A cell in the arrangement of line segments. The definitions of outer, inner, and
popular are analogous to the three-dimensional case. The vertex V1 is outer,
while V2 and V3 are inner. V2 is not popular while V3 is popular. The edge EF
is popular, while E2 is not.
Henceforth, we will therefore be only interested in inner faces. They include

the facets, the edges contained in the intersection of the interiors of two triangles,
and the vertices contained in the intersection of the interiors of three triangles.
By the side of an inner k-face F, we mean one of the regions bounded by the
3 - k planes that support the triangles containing F. Lemma 14.2.4 states that
an inner k-face in the arrangement of T has 2 3-k sides.
We say that an inner face F of C is popularif all its sides are contained in C in
a neighborhood of F. The situation is depicted in two dimensions in figure 16.10.
The concept of popularity plays a central role in the proof. The following three
lemmas successively bound the numbers of popular edges, popular facets, and
edges on the boundary of non-popular facets.
Lemma 16.4.3 There are O(n 2 ) popular vertices and popular edges on the bound-
ary of a cell.
Proof. We count the number of lower endpoints of popular edges. The lower
endpoint A of a popular edge E has the smallest z-coordinate, and is either an
inner or an outer vertex. The number of outer vertices is O(n2 ), by lemma 16.4.2.
If A is an inner vertex, then it lies at the intersection of E with the interior of
a triangle T, and so it is the intersection of three planes: the affine hull of T
and the affine hulls of the two triangles that contain E. Among the sides of A,
there must be one, say R, for which A is the lower endpoint. Let us decompose
C into convex sub-cells as was done in subsection 16.2.2, and let us call C' the
sub-cell which intersects R in a neighborhood of A. Since C' is convex, it must
be contained in R and so A is the lower endpoint to C'. By theorem 16.2.2, the
number of convex sub-cells in this decomposition is O(n2 ). Since each sub-cell
has a unique lower endpoint, which must be incident to three popular edges since
the triangles are in general position, the result follows for popular edges.
Now a popular vertex is always incident to six popular edges when T is in

general position, so the number of popular vertices is proportional to the number
of popular edges. E
Lemma 16.4.4 The number of inner edges contained in popularfacets that are
on the boundary of a cell is O(n 2 log n).
Proof. Consider a triangle T in T and let C' be the cell in the arrangement of
T \ T that contains 0. When inserting T in this arrangement, we estimate the
increase in the number of inner edges in the cell that contains 0 and that belong
to popular facets. We let E be an inner edge in C' that is contained in a popular
facet F in C'.
Casel: TnE=0
Whether E is an edge of C or not, the number of edges that belong to E does
not increase when inserting T.
Case2: TnEO0
Let P be the intersection point of E and T. Then E is cut into two edges E1 and
E2 . If both E1 and E 2 are edges of C, and only in this case, the number of edges
that belong to E increases by one when inserting T. The edge E' of C that is
contained in F n T and incident to P must be popular when the number of edges
that belong to E increases. Therefore P must be the endpoint of a popular edge
of C.
Denote by q and q'(T) the number of inner edges of C and C' contained in
popular facets, and let q"(T) be the number of inner edges of C contained in T.
Moreover, denote by r(T) the number of vertices of popular edges of C contained
in T.
From what was said before, we know that
q - q"(T) < q'(T) + r(T). (16.1)
This result is independent of the choice of T. So we may carry out a similar

analysis for all the triangles in T, and by summing the resulting equations, we
count each inner edge E of a popular facet of C (n - 2) times (once for each
triangle that does not contain E), and we count each vertex S of C three times
(once for each triangle that contains S). Overall, we obtain
(n - 2)q < a, q'(T) + 3r. (16.2)

TET
Finally, if we denote by q(n) the maximum value of q over all possible arrange-
ments of n triangles and all possible choices for the origin 0, inequality 16.2
becomes
n
q(n) < q(n - 1) + O(n), (16.3)
by using the fact that the number of popular edges of C is O(n2 ) (see lemma
16.4.3). We solve this recurrence by putting
q(n) = n ) w (n)
Inequality 16.3 becomes
w(n) < w(n-1) + O(-). (16.4)
This implies that w(n) = 0(logn), which in turn shows that q(n) = O(n2 logn).
We can now finish the proof of the theorem, if we know a bound on the number
of inner edges in non-popular facets of C. This bound is provided by our last
lemma.
Lemma 16.4.5 The number of inner edges in the non-popular facets of C is

0(n2 log2 n).
Proof. The proof is similar to that of the previous lemma. Let T be a triangle
of T, C' be the cell of the arrangement of T \ T that contains the origin 0, and
E be an inner edge of C' that does not belong to any popular facet of C'. We
insert T into the arrangement and estimate the increase in the number of inner
edges in the non-popular facets of the new cell C that contains 0.
As in the previous lemma, the number of these inner edges that belong to E
increases (by one) only when T cuts E into two edges E1 and E 2 which are both
edges of F. Assuming this is the case, let P be the intersection point of E and T,
and FT be the facet of C contained in T and incident to P. Then FT is popular
and contains vertex P. Note that FT is the only popular facet of C that has P
as a vertex.
We now denote by q and q'(T) the number of inner edges in C and C' that are
contained in the non-popular facets of C that belong to T. Moreover we denote
by r(T) the number of vertices on the popular facets of C that belong to T, and
by r the number of vertices on the popular facets of C. Equations analogous to
equations 16.1 and 16.2 hold as well:
q-q"(T) < q'(T) + r(T),
(n -T2)q < E q'(T) +
Tefr
If we denote by q(n) the maximum of q over all possible arrangements of n

triangles in E2 and all the possible choices for the origin 0, using the fact that
r = O(n2 log n) as was proved in lemma 16.4.4, we derive the recurrence
n
q(n) < 2q(n - 1) + O(n log n), (16.5)
which solves to q(n) = O(n 2 log 2 n). E
Consider a set T of n triangles in E3 and its arrangement A, and pick a point 0

that does not belong to any triangle. In this subsection we bound the complexity
of the vertical decomposition of the cell C(T) that contains 0.
We first show that the complexity of the vertical decomposition of C(T) is
proportional to the sum of the complexity of C(T) and of the number of pairs
of edges of C(T) that are mutually visible, meaning that they can be connected
by a vertical line segment whose relative interior does not intersect any triangle
in T. For this, consider a line segment E, either an edge of a triangle or the
intersection of two triangles, such that E contains at least one edge of C(T). Let
1? be the vertical strip described by a vertical line passing through the edge E.
By R7a, resp. Rb, we denote the portion of 1t that lies above E, resp. below E.
We obtain a set of segments Sa (resp. Sb) by intersecting the triangles in T with
the half-strip R1a (resp. Rb). For reasons similar to those given in subsection
16.2.1 for bounding the complexity of the entire arrangement, the complexity of
the decomposition of a single cell is bounded by the total number of vertices on
the lower envelope (in the plane of R) of Sa and on the upper envelope of Sb (see
figure 16.2). Between the vertices of these envelopes that are contained in C(T),
we make a distinction between the intersection points of the envelopes and E, and
the others. This shows that the number of vertices on the envelope is accounted
for by the vertices of C(T) that belong to E and the edges of C(T) that are
visible from E. Summing over all the segments E that are edges of triangles or
intersections of two triangles, we see that our first statement is true.
We obtained a bound on the complexity of a cell in the previous section, so we
are only interested in pairs of mutually visible edges of C(T).
Consider two edges E1 and E2 of A. We say that E1 and E 2 are k-visible if
there exists a vertical line segment that connects them and whose relative interior
intersects C(T) and exactly k triangles in T. These triangles are said to obscure
the pair. A O-visible pair, or visible pair for short, always consists of two edges
of C(T).
As in the previous subsection, we make a distinction between the inner pairs of
edges for which each edge is contained in the intersection of two triangles, and the
other pairs called outer for which at least one edge is contained in the boundary
of a triangle.
Lemma 16.4.6 The number of 0- and 1-visible outer pairs is O(n 2 a (n)).
Proof. We use an argument very similar to that used in subsection 16.2.1,

which was recalled above. Since we are only interested in outer pairs, we need
only consider the edges of triangles and not the intersections of two triangles.
The number of outer visible pairs that involve some edge E of a triangle equals
the number of vertices on the upper and lower envelopes of two sets of segments,
which is O(na(n)). Since there are at most 3n such edges, the number of outer
visible pairs is O(n2 a(n)) The random sampling theorem 4.2.3 and its corollary
4.2.4 imply similar bounds for 1-visible pairs: here the objects are triangles, the
regions are vertical segments joining an edge of a triangle and either an edge of
another triangle or the intersection of two triangles, and a triangle conflicts with
a segment if they intersect. 2 EJ
We now count the number qo(T) of inner visible pairs. Let us consider two
inner edges E1 = T1 n T2 and E2 = T3 n T4 of C(T) that are mutually visible,
where the Ti's are triangles in T. Assume that E1 is above E2 and denote by
S the vertical segment that connects E1 to E2 . By hypothesis, the interior of S
does not intersect any triangle of T.
Consider an extensible vertical segment S', that occupies the same position
as S initially. We slide S' successively in four different directions. For the first
move, we constrain the upper endpoint of S' to belong to E1 , while the lower
endpoint belongs to T3 and S' intersects T4 . There is a single degree of freedom,
and we move S' along this direction until one of the following situations occurs
(see figure 16.11):
1. S' reaches the end of El; this endpoint is a vertex of C(T),
2. S' encounters an edge E of T3 or T4 ; then (E1 , E) is an outer visible pair if

E is an edge of T4 , or a 1-visible outer pair if E is an edge of T3 ,
3. S' meets the edge E of a triangle of T \ {Tl, T2 , T3 , T4 }, between E1 and T3 ;

then S' contains a vertical segment S" that connects the outer pair (E1 , E),
which is 1-visible if S" intersects T4 , or visible otherwise,
4. S' reaches an edge that is the intersection of T3 with some triangle T in

T \ {T 1 , T 2 , T 3 , T 4}.
2
Note that the proof also applies verbatim to pairs of edges in the arrangement that can be
connected by a vertical line segment intersecting zero or one triangle in T (but not necessarily
in 0(T)).
El ~El
Si .S T3
i . ' ./S - T3
T4
E T4
1 2
El ET
T S T3 eSt iS T3
T4
~T4
3a 3b
El
St IS T3
>:44
T
Figure 16.11. The four different kinds of events and their variants, represented in the vertical
plane that contains E1 . E2 intersects this plane in a single point, and T 3 and
T4 in two segments.
Likewise, for the second move, we slide S' while keeping its upper endpoint on
E1 and its lower endpoint on T4 , in the direction where S' intersects T3. We stop
S' in one of the analogous four cases (exchanging T3 and T4 ). Switching the roles
of E1 and E 2 gives the four different moves.
It is important to notice that each event above is encountered only once in all
the moves along a given direction. Indeed, when moving a vertical segment in one
direction, an inner visible pair is met again only after one of the above events.
We count the total number of events encountered during the moves correspond-
ing to each inner visible pair of edges of C(T).
1. A vertex of C(T) is encountered during at most six events of type 1. Indeed,
the guiding edge must be one of the at most six edges incident to this vertex.
The total number of events of type 1 is at most six times the number of vertices
of C(T), which is O(n2 log n) because of theorem 16.4.1.
2. The outer pair (E1 , E) is encountered during a single event of type 2 when
sliding along E1 . It follows that the total number of events of type 2 equals the
Figure 16.12. The 1-visible pair (E., Eb) is reached by two motions from the 0-visible pairs
(E,,E) and (Eb,E').
number of outer pairs that are visible or 1-visible. This number is O(n2 oa(n))
because of lemma 16.4.6.
3. The same reasoning holds for the events of type 3 whose number is also
O(n2a(n)).
4. The events of type 4 can be encountered at most four times. Yet the analysis
only works with a more precise account: the following lemma bounds the number
of events of type 4 that are encountered more than twice.
Lemma 16.4.7 There are at most O(n 2 logn) events of type 4 that are encoun-
tered more than twice.
Proof. Consider a 1-visible inner pair (Ea, Eb), and assume that Ea is contained
in the intersection of the two triangles Ta and Ta, that Eb is contained in the
intersection of the two triangles Tb and Tb, and that Ea lies above Eb (see fig-
ure 16.12). We denote by T the triangle that obscures the pair (Ea, Eb) and by
F the facet of the arrangement contained in T that is stabbed by the vertical
segment Sab that connects Ea and Eb.
If the pair is encountered more than twice during events of type 4, then it
must have been so after at least one motion along Ea and at least one motion
along Eb. So let us consider a visible pair from which we started one of the
motions, and assume it is a pair (Ea, E). (For pairs (Eb, E'), the situation is
entirely symmetrical.) Let Sa be the vertical segment that connects Ea and E
(see for instance figure 16.12 where we also show the vertical segment Sb that
connects Eb and E'.) The edge E is contained in the intersection of T with Tb

or Tb: otherwise, E would involve a triangle T' not in {T, Tb, Tb} and then either
T' would obscure (Ea, Eb) which would not be 1-visible, a contradiction, or the
motion of S along Ea from (Ea, E) would stop on an edge of the boundary of T'
before it reached (Ea, Eb), which is also impossible. So we may assume that E is
contained in T n Tb.
We note that E is an edge of F and that F is a popular facet of C(T). Connect
the points A = Sa n F and B = Sab n F by a line segment (contained in F) called
the trace on F of the motion from (Ea, E) to (Ea, Eb).
We connect B to the at most three other visible pairs from which we started
motions that ended on (Ea, Eb), and proceed similarly for all the inner 1-visible
pairs obscured by F that were encountered more than twice during events of type
4. We thus obtain a set S of segments contained in F (see figure 16.13) which are
the traces on F of all these different motions. A segment in S cannot intersect a
triangle in T. Moreover, two segments in S cannot intersect (except at common
endpoints), otherwise the intersection point would correspond to a 1-visible pair
obscured by F that would have stopped one of the motions ending at (Ea, Eb)
before it reached its endpoint.
By construction, any segment in S has exactly one endpoint on the boundary
of F. The endpoints that are not on the boundary of F correspond to events
encountered more than twice: these endpoints are thus incident to three or four
segments in S. Let us identify the endpoints of segments of S that belong to a
given edge on the boundary of F: we obtain in this way a planar graph 5 whose
external nodes are the edges of F, whose internal nodes are the endpoints of the
segments of S that are not on the boundary of F, and whose arcs correspond to
the segments in S.
Lemma 16.4.8 The number of internal nodes in the graph 5, hence the number
of events of type 4 corresponding to F that are encountered more than twice, is
O(IFI).
Proof. We first show that the graph 5 cannot contain cycles of length 2. Indeed,
each arc of g connects an internal node to an external node. An internal node,
that corresponds to the 1-visible pair (Ea, Eb), is connected to three or four
external nodes that represent the four distinct edges F n Ta, F n Ta, F n Tb,
F n Tb'. As before, we denote by Ta and Ta the triangles whose intersection
contains Ea, and by Tb and Tb those whose intersection contains Eb.
Therefore 5 is a planar graph without cycles of length 2. It has ne = O(IFI)
external nodes and internal nodes of degree 3 or 4, which are incident to external
nodes only. To prove the lemma, we construct from 5 another graph g' that
does not have internal nodes as follows. We distinguish the two sides of an arc
16-4. A cell in an arrangement of triangles 399
16.. A cell in an arrangement of triangles 399
Figure 16.13. The segments in S are shown in solid lines. The shaded face corresponds to
a cycle of length 2 in the graph G. Replacing these solid lines by the dashed
lines and identifying the vertices on a common edge of F yields the graph 5'.
of 5, and the three or four sides of a node of 5. We then replace two arcs that
are incident to a common internal node by a single arc in 5'. It is easy to see
that g' is planar and does not have a cycle of length 2. It has the same number
ne of nodes, and the same number of arcs, as G. Euler's relation shows that the
number of arcs of G', and hence of G, is proportional to the number of its nodes,
and hence is O(IFI).
The total number of internal vertices of G is thus O(IFI).
Summing over all the popular facets of C(T), we conclude that the total number
of events of type 4 that are encountered more than twice is O(n 2 log n) because
of lemma 16.4.4. El
The preceding discussion shows that the total number of events of type 4 is
bounded by
4qO (T) < 2qj (T) + O (n2 log n),
if we denote by qk (T) the number of k-visible inner pairs. It follows that
n q(T) <n qo(T) + -qi(T) + 0(nlogn). (16.6)

n n n
We claim that n-4 qo (T) + 1 qi (T) is bounded by the expected number of visible
inner pairs in the arrangement of a random sample of n-1 triangles in T. Indeed,
an inner visible pair in T is an inner pair visible in the sample if and only if
the removed triangle is not one of the four triangles that define the pair, which
happens with probability n-4. An inner 1-visible pair of f is an inner visible pair
in the sample if and only if the removed triangle is crossed by the segment that
connects the pair, which happens with probability 1. This provides the desired
upper bound. Observe that it is an upper bound but not an equality: indeed, it
does not account for the visible pairs in the sample that were visible in another
cell of the arrangement of T adjacent to C(T) through the removed triangle.
Denoting by qk (n) the maximum number of k-visible inner pairs in the arrange-
ment of n triangles in E3, we have
- qo(n) + q1(n) < qo(n -1),
from which and from inequality 16.6 we derive the recurrence equation
n-2 qo(n) < qo(n -1) + O(n log n),

n
whose solution is qo(n) = 0(n 2 log 2 n).
This finishes the proof of the following theorem:
Theorem 16.4.9 The complexity of the vertical decomposition of a single cell in

the arrangement of n triangles in E3 is 0(n 2 log 2 n).
16.4.3 Computing a cell
Let T be a set of n triangles in E3 , and pick a point 0 that does not belong to
any triangle. In this section, we give an algorithm that computes the cell C(T)
in the arrangement of T that contains 0.
The algorithm we present here is an extension of the incremental randomized
algorithm described in subsection 15.4.2 that computes a single cell in the ar-
rangement of line segments in the plane. It consists of inserting the triangles
in turn. Let 7? be the current subset of triangles, introduced in the previous
incremental steps, and C(7?) the cell in the arrangement of 7? that contains 0.
After inserting a triangle, we update the decomposition of the triangles (see sub-
section 16.2.1) without worrying that some prisms in this decomposition may lie
entirely outside C(1?). We thus allow the algorithm to compute, in addition to
the prisms in the decomposition of 7t, some other prisms that are not prisms of
0(7Z). To keep the complexity within reasonable limits, it is necessary to stop
the construction of the decomposition outside C(1?) at certain clean-up steps.
During those clean-up steps, we deactivate the prisms that do not belong to the
cell C(7?). Only the active prisms are subdivided in the subsequent incremental
steps.
Between two clean-up steps, the algorithm is similar to the algorithm presented
in subsection 16.3.3 to compute the lower envelope of triangles.
The algorithm maintains an influence graph similar to that which is used to

compute the lower envelope of triangles, and an adjacency graph. It also main-
tains a pointer to the prism in the current decomposition that contains the origin
0. The nodes in the influence graph represent the prisms computed by the algo-
rithm. If 1 is the current subset of triangles, the prisms in the decomposition of
the cell C(7?) correspond to the active leaves of the influence graph. Only those
leaves will have children in the subsequent steps. The nodes in the adjacency
graph G represent the active prisms and an arc in G connects two active prisms
that share a common vertical wall. We also maintain pointers between the nodes
of the influence graph and the node of the adjacency graph.
After an initial step which consists simply of creating a node that represents
the whole of E3 , the triangles are inserted each in turn in a random order. The
current incremental step can be subdivided into three phases: locating the prisms
intersected by the triangle T, updating, and occasionally cleaning up.
Locating. During the location phase, we retrieve the leaves of the influence graph
that intersect T by traversing the influence graph and backtracking whenever the
next node to visit is already visited or does not conflict with T. If no leaf in
the influence graph conflicts with T, then T does not appear on the boundary of
C(7?) or in any of the cells that are computed subsequently. So the algorithm
skips directly to the insertion of the next triangle.
Updating. Each active prism P is subdivided into a constant number of new
sub-prisms, which become children of P. (A case analysis shows that this number
is at most nineteen, to be precise.) As happens when decomposing line segments
in the plane or when computing the lower envelope of triangles, some of these
sub-prisms are not prisms in the new decomposition and they must be merged
or removed.
Among the new prisms, we make a distinction between those whose floor or
ceiling is contained in T and the others. Merging the latter kind of prisms does
not create particular problems and can be performed in a way that is very similar
to what was done in subsection 16.3.3.
As for the new prisms whose ceiling is contained in T, we build a secondary
influence graph as in subsection 16.3.3. We proceed similarly for the prisms whose
floor is contained in T.
Cleaning up. At special steps 2', for i = 1,..., Llognj - 1, we visit all the
prisms contained in the current cell C(R), by traversing the adjacency graph of
the current decomposition, starting from the cell that contains 0. The nodes
that are not visited in this process are deactivated.
The complexity analysis of the algorithm combines the analyses of subsections
15.4.2 and 16.3.3. We leave to the reader the careful examination of the com-
plexity, and the proof that it is O(IDec(n)llogn) where IDec(n)l denotes the
maximum complexity of a cell in the arrangement of n triangles in E3 . Theorem

16.4.9 finishes the proof.
Theorem 16.4.10 It is possible to compute a single cell in the arrangement of

n triangles in E3 in time O(n2 log 3 n) using storage O(n2 log2 n).
16.5 Exercises
Exercise 16.1 (The complexity of a lower envelope of simplices) Show that the
complexity of the lower envelope of n (d -1)-simplices in Ed is e(nd-'a(n)).
Exercise 16.2 (Deterministic computation of lower envelopes) Given a set T of

n triangles in E3 , show how to deterministically compute the lower envelope of T in time
O(n2 log n) if the triangles are disjoint, and O(n 2 a(n) log n) otherwise.
Hint: For a set S of triangles, we denote by A(S) the arrangement of the affine hulls
of the projected edges of the triangles in T in the xy-plane. We denote by £(S) the
planar map obtained by projecting the lower envelope of S, and by £ (S) the refinement
of £(S) obtained by superimposing £(S) and A(S). To use the divide-and-conquer
method, consider two subsets Tl and T2 of T of roughly the same size. Compute
9*(Ti) and £*(T 2 ) recursively. Superimposing E*(T 2) and A(T1 ) yields a planar map
£S and, likewise, superimposing 6*(T 2 ) and A(T1 ) yields a planar map £#1. Show
that lJE 1 I = O(n 2 o(n)) and that 19'0 I = O(n2 oa(n)) (think of inserting the lines in the
arrangements A(Ti) one by one). Compute A(T) and, for each cell in A(T), compute
the portion of the lower envelope whose projection onto z = 0 is this cell. This portion is
a convex dome, the intersection of two convex domes corresponding to a cell of £#9 and
a cell of £4 The factor logn can be removed; see the references in the bibliographical
notes.
Exercise 16.3 (Lower envelopes of surfaces) Devise a generalization of the algo-

rithm described in subsection 16.3.3, which computes the lower envelope of a set of
triangles, to the case of algebraic surface patches in E3.
Exercise 16.4 (Decomposition of an arrangement of triangles) Show that the

complexity of the vertical decomposition of the arrangement of n triangles in E 3 is
O(n 2oa(n) log n + t), if t is the complexity of the arrangement. Devise a randomized algo-
rithm that computes such a decomposition in expected time O((n2 a(n) logn + t) log n).
Hint: Adapt the proof of the theorem which bounds the complexity of the vertical
decomposition of a single cell.
Exercise 16.5 (Vertical decomposition of surfaces) Generalize the notion of a

vertical decomposition of an arrangement of triangles to the case of algebraic surfaces in
E 3 . Bound the complexity of such a decomposition and devise an algorithm to compute
it that is similar to the one presented in subsection 16.2.1.
16.5. Exercises 403
Exercise 16.6 (Computing a view) Consider a polyhedral scene with a total of n

triangular facets (which may intersect). Show that the view from a given point, which
is the portion of the facets visible from this point, has complexity 0(n2 ) if the relative
interiors of the triangles are disjoint, O(n 2 a(n)) otherwise. Show that these bounds are
best possible in the worst case. Give an algorithm that computes the view from a given
point with a cost proportional to these bounds by a factor log n.
Exercise 16.7 (Stabbing) Given n polyhedral objects in E3, compute the set of planes
that simultaneously stab them all.
Hint: Use polarity as in the planar analogue, exercise 15.5.
Exercise 16.8 (Levels in arrangements of simplices) Let S be a set of n (d - 1)-

simplices in Ed which we assume are in general position, and let A(S) be their arrange-
ment. To each (d - 2)-face of a simplex in S corresponds a vertical (d - 1)-face generated
by the vertical rays cast upwards from the (d - 2)-face. A simplex and these vertical
(d -1)-faces generated by its (d - 2)-faces bound a truncated cylinder with vertical gen-
erators. Let A'(S) be the arrangement of the corresponding cylinders. We note that the
vertical faces of A'(S) further subdivide the faces of A(S).
To maintain an analogy with what was said about levels in hyperplane arrangements,
we say that a point in E3 is at level k if it belongs to the interior of exactly k cylinders. All
the points in the relative interior of a face of A'(S) are at the same level. By definition,
this level is the level of the face. We denote by A'k(S) the sub-complex of A'(S) formed
by all the faces of A'(S) at level at most k.
Show that for any integer k, the number of faces at any level 0 < j < k is bounded by
O(knd 'a( i)).
Conclude that if T is such that any vertical line intersects at most k simplices, then
the complexity of the arrangement of T is 0(knd-l a()).
Hint: Use theorem 16.3.1, exercise 16.1, and the sampling theorem 4.2.3.
Exercise 16.9 (Computing the first k levels) Devise and analyze an algorithm
that computes the first k levels in the arrangement of n triangles in E3 .
Hint: Adapt the algorithm that computes the lower envelope of a set of triangles, and
the algorithm that computes the first k levels in an arrangement of hyperplanes (see
section 14.4).
Exercise 16.10 (A cell in higher dimensions) Show that the complexity of a single
cell in the arrangement of n (d - 1)-simplices in Ed is O(nd-1 logd 1 n).
Exercise 16.11 (Motion planning for a polyhedron) Consider a polyhedron M

with m facets that moves inside a polyhedral region £ with n facets. Without loss
of generality, we assume that the facets are triangular.
1. Show that the set of translations that bring a vertex of M into contact with a facet
of £, or a vertex of £ into contact with a facet of M, or an edge of M into contact with
an edge of £, is a set C of at most mn triangles and parallelograms (when we identify
the vector OM of the translation with its endpoint M).
2. Show that the set of positions of M inside £ that can be accessed from a given
position I corresponds to the cell that contains I in the arrangement of the triangles and
parallelograms of C. Conclude that it may be determined whether two positions I and
F in £ may be connected by a path along which M remains entirely inside C, and, if so,
compute such a path in time O(m 2 n2 log3 (mn)).
Exercise 16.12 (Flying saucers) Consider a polyhedral flying saucer that flies above
a terrain modeled by a function z(x, y) that is piecewise linear. Show that the set of
translations of E3 for which the flying saucer is strictly above the ground is characterized
as the region above the upper envelope of certain triangles in E3. Give a lower bound on
the complexity of such a set.

The bound on the complexity of the lower envelope of triangles was shown by Pach and
Sharir in [185]. The proof presented in subsection 16.3.1 is shaped after the proof given
by Halperin and Sharir in [120] for the case of algebraic surfaces. They show that the
lower envelope of n algebraic surface patches of bounded degree in Ed is O(nd-l+,) for
any E (but the constant increases when E goes to 0). The algorithm that computes the
envelope of triangles in E 3 described in subsection 16.3.3 and its extension to algebraic
surfaces (see exercise 16.3) are due to Boissonnat and Dobrindt [30].
The bound on the complexity of a single cell in the arrangement of triangles is due
to Aronov and Sharir [9, 10], who also generalize it to simplices in any dimension (see
exercise 16.10). This result was extended to algebraic surfaces by Halperin and Sharir
[119, 121] who showed that a single cell in the arrangement of n algebraic surface patches
of bounded degree in E3 has almost quadratic complexity. How to compute a cell in
dimension 3 is explained by de Berg, Dobrindt, and Schwarzkopf in [76].
The bound on the complexity of the vertical decomposition of a single cell in the ar-
rangement of triangles in dimension 3 given in subsection 16.4.2 and the bound on the
complexity of the entire arrangement (see exercise 16.4) are due to Tagansky [212]. A
slightly worse bound and a deterministic algorithm that computes the vertical decom-
position of the arrangement of triangles are given by de Berg, Guibas, and Halperin in
[77]. A bound of O(n4 log n) on the vertical decomposition of triangles in dimension 4 is
given by Guibas, Halperin, Matougek, and Sharir in [115]. Decomposing an arrangement
of algebraic surfaces is a problem studied by Clarkson et al. in [64] and by Chazelle et
al. in [51]. In the latter, they bound the complexity of their cylindrical decomposition
by O(n2d-3 (n)), where f(n) = 2a(n)e for a constant c that does not depend on n.
Many references and results concerning hidden surface removal in computer graphics
can be found in the book by de Berg [75]. Schwartz and Sharir [197] give a wide survey
of combinatorial and algorithmic techniques motivated by motion planning and robotics.
Part V
Voronoi diagrams
Voronoi diagrams are very useful structures, frequently encountered in several

disciplines because they model growth processes and distance relationships be-
tween objects: it is not surprising to see them appear in the study of crystal
growth or in studies on the great structures of the universe. In nature, they can
be observed in crystalline structures, on the shell of a turtle, or on the neck of a
reticulate giraffe.
Voronoi diagrams are very closely related to the geometric structures encoun-
tered so far: polytopes, triangulations, and arrangements. Their mathematical
properties are particularly numerous and interesting. Chapter 17 is entirely de-
voted to Voronoi structures with a Euclidean metric, whereas other metrics are
studied in chapter 18. Chapter 19 presents results specific to dimension 2 that
have no analogue in higher dimensions.
Voronoi diagrams can also be used as data structures to solve numerous prob-
lems: nearest neighbors and motion planning are two outstanding instances. Sev-
eral examples are given in the exercises and throughout chapter 19.
Chapter 17
Euclidean metric
This chapter is concerned with the simplest case of Voronoi diagrams, where the
objects are points and the distance is given by the usual Euclidean metric in Ed.
The cells in the Voronoi diagram of a set M of points are then the equivalence
classes of the equivalence relation "to have the same nearest neighbor in M". It
is possible to show (see section 17.2) that such cells can be obtained by projecting
the facets of a polytope in Ed+1 onto Ed, which enables us to use several results
concerning polytopes for Voronoi diagrams as well. Bounds can be obtained in
this way for the complexity of Voronoi diagrams and of their computation. In
section 17.3, we define a dual of the Voronoi diagram, the Delaunay complex,
that enjoys several properties which make it desirable in applications such as
numerical analysis in connection with finite-element methods. The last section
of this chapter introduces a first generalization of Voronoi diagrams (see section
17.4): the higher-order Voronoi diagrams. The cells in the diagram of order k are
the equivalence classes of the equivalence relation "to have the same k nearest
neighbors in M", a notion that is often very helpful in data analysis.
17.1 Definition
Let M be a set of n points in Ed , MI, Mn, which we call the sites to avoid
confusion with the other points in Ed. To each site Mi we attach the region
V(Mi) in Ed that contains the points in Ed closer to Mi than to any other point
in M:
V(Mi) = {X c TEd : 6(X, Mi) < 6(X, My) for any j # i}.
In this chapter, 6 denotes the Euclidean distance in Ed. Other distances will be
considered in chapter 18.
The set of points closer to Mi than to another site Mj is the half-space that
contains Mi and that is bounded by the perpendicular bisector of the segment
408 Chapter 17. Euclidean metric
Figure 17.1. The Voronoi diagram of a set of points in the plane.
MjMj: this is the hyperplane perpendicular to MjMj that intersects MiMj at

the midpoint of Mi and Mj. The region V(Mi) is thus the intersection of a finite
number of closed half-spaces, bounded by the perpendicular bisectors of MjMj,
j = 1, . .. , n, j =A i. This shows that V(Mi) is a convex polytope, which may or
may not be bounded. As we will see later, the V(Mi)'s and their faces form a cell
complex whose domain is the whole of Ed. This complex is called the Voronoi
diagram of M and is denoted by Vor(M) (see figure 17.1).
A first and useful interpretation of the Voronoi diagram (another interpretation
is given in the next chapter) views the cell V(Mi) as the set of centers of balls such
that the boundary of such a ball contains Mi and its interior does not contain
another site Mj, j =#i. In particular, this point of view leads to the interpretation
of a Voronoi diagram in Ed as a polytope in Ed+l, which also enables it to be
computed efficiently. This interpretation is developed in section 17.2 where we
represent spheres of Ed as points in Ed+l.
From now on, we say that the sites are in L2-general position if no sphere can
contain d + 2 sites on its boundary.
17.2 Voronoi diagrams and polytopes

17.2.1 Power of a point with respect to a sphere
Consider the Euclidean space of dimension d, Ed, and let 0 be its origin, and E
be a sphere of Ed centered at C with radius r. Its equation is given by E(X) = 0
17.2. Voronoi diagrams and polytopes 409
M N'
Figure 17.2. Power of a point with respect to a circle.
where
E(X) = XC 2 -_r 2 . (17.1)
By the interior of a sphere A, we mean the set of points X such that E(X) is
negative. The exterior is the set of points X such that E(X) is negative. A
point X is said to be on, inside or outside a sphere if it belongs to the sphere,
respectively to its interior, to its exterior. For any point X in Ed, E(X) is called
the power of X with respect to E. The power of the origin with respect to E is
also denoted by oa and we have
a, = r2(0) = C2 - r2 . (17.2)
If D is any line that contains X, and if M and N are the intersection points of
D with E, then
S(X) = XM - XN. (17.3)
This is obvious when D is the line connecting X and C. Otherwise let D' be the
line that contains X and C, and let M' and N' be its intersection points with E
(see figure 17.2). The triangles XMM' and XN'N are similar (the angles M'MN
and M'N'N are supplementary), which proves equation 17.3. In the case where
X belongs to the exterior of E and D is tangent to E at T, then M = N = T
and the previous equation can be rewritten
E(X) = XT 2 . (17.4)
17.2.2 Representation of spheres
Let X be the mapping that takes a sphere E in Ed, of center C and whose power
with respect to 0 is cr, to the point O(E) = (C,o-) in Ed+l. Using 0 enables us
to treat spheres in Ed just as points in Ed+1.
We embed Ed as the hyperplane in Ed+1 whose equation is xd+1 = 0. As usual,
the direction of the Xd+1-axis is called the vertical direction and we use the words
above and below in connection with this vertical ordering. We denote by X a

point of Ed or its coordinate vector (XI, ... , Xd) indifferently, and by X a point
in Ed+l or its coordinate vector (xI, ... , Xd+l). By the above embedding, O(E)
projects vertically onto C. Later on, we will often use vertical projections and,
unless mentioned otherwise, the word projection refers to the vertical projection
from Ed+l onto Ed.
We also use homogeneous coordinates and the matrix notation. We denote
by X = (xl, . .. ,Xd,t) (resp. X = (x,.. . ,Xd+1,t)) the homogeneous coordinate
vector of a point X in Ed (resp. a point X in Ed+l). The equation of the sphere
E can then be rewritten with homogeneous coordinates as
XEXt=O with E= (d C),
where Rd denotes the d x d identity matrix.
17.2.3 The paraboloid P
From equation 17.2, it follows that the images under ; of points in Ed, considered
as spheres of radius 0, belong to the paraboloid of revolution P with vertical axis
and equation
d
Xd+l=Zxix=X.X with X=(xl,...,Xd).
i=1
In a homogeneous system of coordinates P is given by
X Ap xt=O where AP=

( Ed
0
0
0 -1/2
0
0 -1/2 0
Identifying a point X and the sphere centered at X with radius 0 shows that X
maps any point X in Ed to the point +(X) in Ed+1 obtained by lifting X onto P.
The set of concentric spheres in Ed, centered at C, is mapped by 0 onto the
vertical line in Ed+l that contains C (and hence O(C)). Let E be such a sphere.
Equation 17.2 implies that the signed vertical distance from O(E) to O(C) equals
r2 (see figure 17.3). Thus, the real spheres, whose squared radii are non-negative,
are mapped by s to the points lying on or below the paraboloid, while the points
lying above the paraboloid are the images under X of the imaginary spheres,
whose squared radii are negative.
IXd+l
, (C)
O(E) -i2
7,E x
Figure 17.3. The paraboloid P (the coordinate system is not normed, so as to simplify the
representation).
17.2.4 Polarity
Consider a quadric Q in Ed+' defined by its homogeneous equation
Q(X) = XAQX t =O.

Henceforth, Q will be the paraboloid P, but we treat the case of any quadric
for generality as it introduces no additional difficulty. Two points X and Y are
conjugate with respect to Q if
Q(X, Y) = XAQY t = O.
The polarity with respect to Q described in section 7.3 is an involution between
points and hyperplanes in Ed+l which maps any point A to its polar hyperplane
A* of equation
AAQXt = 0
and maps any hyperplane H to a point H* whose polar hyperplane is H. The
point H* is called the pole of H.
Note that if Q is the paraboloid P and if we put O(E) = (C,a), the equation
of the polar hyperplane 0(S)* of 0(S) can be rewritten as
Xd+1 = 2C . X - a.
An essential property of polarity is that it preserves incidences (see section 7.3):
a point X belongs to a hyperplane H if and only if its polar hyperplane X*
contains the pole H* of H. Moreover (see exercise 7.14), when the quadric is the
paraboloid P, we have
X E H+ > H* c X*+
X E H- H* E X*-,
if we denote by H+ and H- the half-spaces bounded by H and that lie respec-

tively above and below H.
17.2.5 Orthogonal spheres
Two spheres El and E2 centered at C1 and C2 and with radii r1 and r2 are
orthogonal if
EI(C2) = r2, (17.5)
or equivalently if
E2(C1) = r.
A simple verification shows that, if the spheres are real, then they are orthogonal
if and only if the angle (IC1 , IC2) at any intersection point I of El n E2 is a right
angle, or equivalently, if and only if the dihedral angle of the tangent hyperplanes
at I is a right angle.
Expression 17.5 may be rewritten as
1
C1 *C 2 -- (01 +a2) = °, with vi = Ci2-r2 (i = 1, 2),
2 (i1,)
which shows that two spheres El and E2 are orthogonal if the two points O(El)
and O(E2) are conjugate with respect to the paraboloid P. This implies that:
Lemma 17.2.1 The set of spheres in Ed that are orthogonal to a given sphere
is mapped by q to the polar hyperplane O(E)* of d(S).
Let us now consider the points in Ed as spheres of radius 0. The set of spheres in
Ed that pass through a given point X E Ed is also the set of spheres orthogonal
to the sphere centered at X with radius 0. Therefore its image under 0 is the
hyperplane O(X)* polar to q(X) E P. This hyperplane must be tangent to P
and to O(X): indeed, the only sphere of radius 0 which is orthogonal to X is X
itself, and hence O(X)* intersects P in a single point O(X).
Let E be a sphere in Ed. The intersection of O(E)* with P is the image under
q of the set of spheres with radius 0 that are orthogonal to E, namely E itself
(considered as a set of points, or equivalently as a set of spheres of radius 0).
Consequently, O(E)* n P in Ed+l projects onto E in Ed. More generally, we have
the following result.
7P
t S(x)
Figure 17.4. Interpretation in E(X) in Ed+l.
Lemma 17.2.2 The intersection of the paraboloidP with a hyperplane H projects

onto Ed as a sphere 0-1 (H*) whose center is the vertical projection of H*. Con-
versely, the points of a sphere E of Ed lifted on the paraboloidP in Ed+l belong
to a unique hyperplane that intersects P exactly at these points. This hyperplane
is the polar hyperplane +(S)* of 0(S).
It follows from this lemma that the power of a point X with respect to a sphere E
equals the square of the radius of the sphere Ex orthogonal to E and centered at
X (Ex is imaginary if X is inside E). The power E(X) can be easily computed in
the space Ed+l that represents the spheres of Ed. Indeed (see figure 17.4), Ex is
mapped by X to a point I in Ed+l that is the intersection of the vertical line that
passes through X (which corresponds to the spheres centered at X) with the polar
hyperplane 0(S)* of q(E) (which corresponds to the spheres orthogonal to E).
The xd+l-coordinates of +(x) and I are respectively X 2 and Ex(O) = X 2 - (X)
since the square of the radius of Ex equals the power of X with respect to E. The
difference of these xd+l-coordinates is called the signed vertical distance. This
proves the following lemma.
Lemma 17.2.3 The power of X with respect to a sphere E equals the signed
vertical distance from the point O(X) to the hyperplane ¢(S,)*.
We thus have the following lemma:

Lemma 17.2.4 Let X and E be respectively a point and a sphere in Ed. If H is

a hyperplane in Ed+l, we denote by Hd- the half-space lying below H. Then:
X E E : O(X) G O(E)* 4 (S) E O(X)*

X E int(E) *.- (X) E 4)(S)*- A}= O(E) E O(X)*-
X E ext(E) . - ¢7O(X) E +(E)*+ A}> O(S) E +(X)*+
The equivalences on the left are consequences of the two preceding lemmas, and
the ones on the right are proved by the special properties of polarity (see subsec-
tion 17.2.4 and exercise 7.13).
Any point in the half-space that lies below 0(X)* in Ed+l is thus the image
under > of a sphere whose interior contains X. Likewise, any point in the half-
space that lies above +(X)* in Ed+l is the image under 0 of a sphere whose
exterior contains X, and the points on +(X)* are the images of the spheres
passing through X.
Remark. Lemma 17.2.3 shows that the squared distance IIXA112 separating
points X and A, which is also the power of X with respect to the sphere centered
at A with radius 0, equals the absolute value of the vertical distance between
O(A)* and 4(X). Points X and A play symmetric roles, so IIXA112 also equals
the absolute value of the vertical distance between O(X)* and O(A).
17.2.6 Radical hyperplane

Let El and E2 be two spheres in Ed. The set of points in Ed that have the
same power with respect to these two spheres is a hyperplane, called the radical
hyperplane and denoted by H1 2 , whose equation is given by
H1 2 : E 1 (X) - E 2 (X) = 0.
As we observed in subsection 17.2.5, the power of a point X with respect to a

sphere E equals the square of the radius of the sphere orthogonal to E centered
at X. A point has the same power with respect to El as with respect to E2 if it
is the center of a sphere orthogonal to both El and E2. Lemma 17.2.1 shows that
the set of spheres in Ed that are orthogonal to a given sphere E is mapped by 0
onto the polar hyperplane 0(S)*. The spheres orthogonal to El and E2 are thus
mapped by q to the affine subspace of dimension d - 1 that is the intersection of
O(E1)* and O(E2)*. The projection onto Ed of this affine subspace is exactly the
set of points that have the same power with respect to El and E2.
17.2.7 Voronoi diagrams

Let M f Ml, . . . , M } be a set of n points in Ed. As before, we embed the
Euclidean space Ed of dimension d into Ed+l as the hyperplane Xd+l = 0, and
Figure 17.5. The Voronoi polytope V(M).
we let q(Mi)* denote the hyperplane in Ed+l that is tangent to the paraboloid
P at the point O(Mi) obtained by lifting M vertically onto the paraboloid 7P,
for each i = 1, ... , n. The preceding discussion shows that the set of spheres
(real or imaginary) whose interiors contain no point of M is mapped by 0 to the
intersection of the n half-spaces lying above the hyperplanes i(Mi)*,..., i(Mn)*,
This intersection is an unbounded polytope which contains P. We call it the
Voronoi polytope and denote it by V(M) (see figure 17.5).
Theorem 17.2.5 The Voronoi diagram of M, denoted by Vor(M), is a cell

complex of dimension d in Ed whose faces are obtained by projecting onto Ed the
proper faces of the Voronoi polytope V(M).
Proof. The boundary of V(M) is a pure cell complex of dimension d, hence so is

Vor(M). Let A be a point on a facet of V(M) that is contained in the hyperplane
tangent to 'P at O(Mi). Then A is the image under 0 of a sphere EA that passes
through Mi and whose interior contains no other point of M (see lemma 17.2.4).
There cannot be a site in M closer to the center of EA than Mi. But this center
is exactly the projection A of A onto Ed. In other words, A belongs to the cell
V(Mi) of the Voronoi diagram. D
This theorem implies that the combinatorial properties of Voronoi diagrams

follow directly from those of polytopes as studied in chapter 7. In particular, if
the points O(Mi)* are in general position in Ed+l, then V(M) is a simple (d+ 1)-
polytope. Each vertex is thus incident to d + 1 hyperplanes. Expressed in terms
of Mi 's, the general condition assumption means that no d + 2 points in M lie on
the boundary of a sphere: this is exactly the L2-general position assumption. If
it is satisfied, Vor(M) is a complex whose vertices are all equidistant from some
d + 1 points in M and closer to these points than to any other point in M: they
are the centers of spheres circumscribed to (d + 1)-tuples whose interiors do not
contain any point in M. More generally, a k-face of Vor(M) is the projection of
a k-face of V(M). It is thus the set of points that are equidistant from d + 1 - k
points in M and closer to these points than to any other point in M.
Theorem 17.2.5 reduces the problem of computing the Voronoi diagram of n
points in Ed to the computation of the intersection of n half-spaces of Ed. The
algorithms described in this book that compute half-space intersections, be they
deterministic, randomized, static or dynamic, output-sensitive or not, can all be
used to compute Voronoi diagrams.
Corollary 17.2.6 The complexity (namely, the number of faces) of the Voronoi
diagrams of n points in Ed is E(n rd/ 21). We may compute such a diagram in
time O(n log n + nrd/2 1), which is optimal in the worst case.
Proof. The upper bounds on the complexity and running time of the algorithm
are immediate consequences of the upper bound theorem 7.2.5 and of results of
the previous sections.
That Q(nfd/2 1) is a lower bound on the complexity of the Voronoi diagram of n
points in Ed is a consequence of exercise 7.11, where it is shown how to construct
a maximal polytope whose vertices lie on the paraboloid, and of theorem 17.3.1
below.
That Q(n log n) is a lower bound on computing the Voronoi diagram in the plane
is a consequence of the fact that the unbounded edges of the Voronoi diagram of
a set M of points correspond to projections of the edges of the convex hull of M.
We also comment on this below. 0
17.3 Delaunay complexes

17.3.1 Definition and connection with Voronoi diagrams
Given a set of n points M = {M1 ,..., MnJ in Ed, we lift the points onto
the paraboloid P to {f(M 1 ), .O. , 0(Mn)}, and consider the unbounded polytope
V(M) that is the intersection of the n half-spaces that lie above the hyperplanes
O(Ml)*, . . ., (Mn)*, where O(Mi)* is tangent to P at O(Mi).
We denote by D(M) the convex hull of the points O(MI),.. ., O(Mn) and a point
O' on the Xd+1-axis, with Xd+1 > 0 large enough so that the facial structure of
17.3. Delaunay complexes 417
-Xd+l
P
W(M
4)
0(MI )
i \_ LZI 4i
4 >
M1 M2 0 | M3 M4
Figure 17.6. The polytope D(M).
that convex hull is stable as O' vanishes to infinity (see figure 17.6). The faces of
D(M) that do not contain O' form the lower envelope of conv(O(MI), . .. , O(Mn))
(see also exercise 7.14). Their projections onto Ed form a complex whose vertices
are exactly the Mi's. The domain of this complex is the projection of the convex
hull of the O(Mi)'s: it is therefore the convex hull conv(M) of the Mi's. This
complex is called the Delaunay complex of M and is denoted by Del(M). For
k = 0,..., d, the k-faces of Del(M) are thus in one-to-one correspondence with
the k-faces of D(M) that do not contain O'.
As shown in exercise 7.14, there exists a bijection between the faces of V(M)
and the faces of D(M) that do not contain O'. This bijection maps the facet
of V(M) containing O(Mi)* to the point q(Mi). More generally, the k-faces of
V(M) are in one-to-one correspondence with the (d - k)-faces of D(M) that do
not contain O'. Moreover, this bijection reverses inclusion relationships.
Owing to theorem 17.2.5, the k-faces of Vor(M) are also in bijection with
the k-faces of the unbounded polytope V(M). So we have a bijection between
the k-faces of Vor(M) and the (d - k)-faces of Del(M) that reverses inclusion
relationships. The Delaunay complex Del(M) is therefore dual to the Voronoi
diagram Vor(M).
Notice that the duality above maps a face of Vor(M), formed by the points
equidistant from m sites in M, to the face of Del(M) that is the convex hull of
these sites.
The preceding discussion leads to the following theorem:
Theorem 17.3.1 The Delaunay complex of n points M 1i... , Mn in Ed is a com-

plex dual to the Voronoi diagram. Its faces are obtained by projecting the faces of
the lower envelope of the convex hull of the n points 44(Mi),..., 4(Mn), obtained
by lifting the Mi 's onto the paraboloidP.
The preceding theorem reduces the computation of the Delaunay complex of n
points in Ed to the computation of the convex hull of n points in Ed+1. All the
convex hull algorithms described in this book, be they deterministic or random-
ized, static or dynamic, output-sensitive or not, therefore provide algorithms of
the same kind that compute Delaunay complexes.
Theorem 17.3.1 also gives a lower bound on the complexity of the Delaunay
complex. Indeed, exercise 7.11 exhibits polytopes in Ed+l with n vertices on the
paraboloid, whose complexity is E(nrd/2 1). The same bound therefore applies to
Delaunay complexes, and dually to Voronoi diagrams.
Corollary 17.3.2 The Delaunay complex of n points in Ed can be computed in

time O(nlogn + nFd/ 2 1), and this is optimal in the worst case.
17.3.2 Delaunay triangulations

Under L 2 -general position assumptions, V(M) is a simple polytope, D(M) is a
simplicial polytope, and Del(M) is a simplicial complex which we call in this
case the Delaunay triangulation(see figure 17.7). If there is a subset M' c M
l
of > d + 1 co-spherical points and if the interior of the sphere circumscribed
to M' (namely the sphere that passes through all the points in M') does not
contain points in M \ M', the Delaunay complex D(M) is not simplicial any
more since conv(M') is a d-face of the Delaunay complex and it is not a simplex.
Note however that this face may always be triangulated, and other non-simplicial
faces of the complex may be triangulated as well. There are many ways to trian-
gulate these faces, and any such triangulation is called a Delaunay triangulation.
Henceforth, we denote by Det(M) any such triangulation.
17.3.3 Characteristic properties

The Delaunay complex has remarkable properties, all due to the fact that it is
dual to the Voronoi diagram.
Theorem 17.3.3 Let M be a set of n points M1 , ... , Mn in Ed. Any d-face in

the Delaunay complex can be circumscribed by a sphere that passes through all its
vertices, and whose interior contains no point in M.
Proof. Let us pick a d-face T of the Delaunay complex. Then T is the convex hull
T = conv(Mi 0 , . . ., Mi,) of 1 co-spherical points Mi0 , ... ., Mil. (If the points are in
Figure 17.7. The Delaunay triangulation that corresponds to the Voronoi diagram shown
in figure 17.1.
L2-general position we have 1= d.) The convex hull conv(O(Mi,),. . . , 0(Mi)) is

a d-face F of the convex hull conv(b(MI), .. ., O(M,,)), because of theorem 17.3.1.
The intersection of the hyperplane HF that supports F and of the paraboloid
projects onto Ed as a sphere E circumscribed to conv(M1 o, ... , Mil), and its center
is the projection on Ed of the pole HF of HF (see lemma 17.2.2). H* is a
vertex of V(M), and more precisely is the intersection of the polar hyperplanes
(Mi* (M ). C is the vertex of the Voronoi diagram that is incident to
the cells that correspond to the sites MAO, ... , Mil, and the interior of E cannot
contain any other point in M. l
Our next theorem extends this result into a necessary and sufficient condition
for the convex hull of some points in M to be a face of the Delaunay complex of
M.
Theorem 17.3.4 Let M be a set of points in Ed, and Mk = {MiO,. .. ,Mik}
be a subset of k points in M. The convex hull of Mk is a face of the Delaunay
complex if and only if there exists a (d - 1)-sphere passing through Mi 0 .. ,ik
and such that no point in M belongs to its interior.
Proof. The necessary condition immediately results from the preceding theorem
and from the fact that a sphere circumscribed to a face is also circumscribed to
its subfaces. Assume that there exists a (d - 1)-sphere E that passes through
Mil, ... , Mik and whose interior contains no point in M. Let H be the hyper-
plane 0(S)* in Ed+1. This hyperplane contains the points I J
.(Mio),....
and the half-space H- lying below H does not contain points in ¢(M) (ac-
cording to lemma 17.2.4). Thus H is a hyperplane supporting D(M) along
conv(O(Mi.), . . .,(Mi)). Hence conv(O(Mio) .. . (Mi)) = H nD(M) is a face
of D(M). It follows from theorem 17.3.1 that Mi, ... ., Mi is a face of the De-
launay complex of M. El
Corollary 17.3.5 Any Delaunay triangulationof a set M of points in Ed is such

that the sphere circumscribed to any d-simplex in the triangulation contains no
point of M in its interior. Conversely, any triangulationsatisfying this property
is a Delaunay triangulation.
The next theorem now exhibits a local characterization of Delaunay triangu-

lations that will be put to good use later on. Let us consider any Delaunay
triangulation T(M) of a set M of points in Ed and let Si = M. MdMd+l
and S2 = Ml... MdMd+2 be a pair of adjacent d-simplices in T(M) that share
a common face F = Ml . . Md. The pair (Si, S 2 ) is called regular if Md+1 does
not belong to the interior of the sphere E2 circumscribed to S2 . If the sphere El
circumscribed to Si differs from E2, the regularity condition is equivalent to the
property that Md+2 does not belong to the interior of El. Indeed, Md+1 does not
belong to the interior of E2 if and only if E2(Md+l) > 0. But the hyperplane HF
that supports F is the radical hyperplane of El and E2- Since El(Md+l) = 0,
the half-space bounded by HF that contains Md+1 (resp. Md+2) consists of the
points whose power with respect to El is smaller (resp. greater) than their power
with respect to E2, and therefore
Y1(Md+2) > F2(Md+2) = 0,
which proves that Md+2 does not belong to the interior of El.
Theorem 17.3.6 Consider a triangulation T(M) of a set M of points in Ed.

Then f(M) is a Delaunay triangulation if and only if all the pairs of adjacent
d-simplices in f(M) are regular.
Proof. That the condition is necessary is a consequence of theorem 17.3.3. We

must now show that it suffices. To alleviate the notation, we denote by O(S) the
k-simplex in Ed+I whose vertices are the images of the vertices of a k-simplex
S in Ed, and by C the union of the O(S)'s for all the faces S of the Delaunay
triangulation T(M). The proof consists of proving that C is the graph of a convex
real-valued function over the convex hull conv(M).
As above, we consider two adjacent d-simplices SI = M1 ... MdMd+l and S2 =
Ml ... MdMd+2 in f(M) that share a common face F = M.. . Md. We denote
by El and E2 the spheres circumscribed to Sl and S2 . Owing to lemma 17.2.4, the

regularity condition is equivalent to q(Md+1) E O(E2)*+ and also to O(Md+2) E
O(E1 )*+ because of the discussion above. Therefore, if the pair (Sl, S2 ) is regular,
then the (d - 1)-face +(F) is locally convex, meaning that there is a hyperplane
that contains +(F) such that O(Sl) and O(S2) belong to the half-space lying above
this hyperplane. This is true for any (d - 1)-face of C incident to two d-faces,
and so C is locally convex at any point. Moreover, C is defined over a convex
subset of Ed, namely the convex hull of M. Therefore, C is convex and is the
lower envelope of the polytope D(M), which proves that T(M) is a Delaunay
triangulation of M. 2
17.3.4 Optimality of Delaunay triangulations

As we have seen in chapter 11, there exist several ways to triangulate a set of
points. Some are not very interesting in practice, and in many applications certain
criteria must be optimized, and an optimal triangulation is desirable. There
are several ways to define optimality. In this section, we show that Delaunay
triangulations maximize two criteria, compactness and equiangularity.
Compactness
The preceding theorem was concerned with spheres circumscribed to simplices in
the triangulation. The next theorem considers the smallest enclosing sphere for
each simplex S: this sphere is the circumscribed sphere of S if the center of the
latter belongs to S, or otherwise is a sphere centered on some k-face (k < d) of
S and passes through the k + 1 centers of this face.
As before, we consider a set M of points in Ed and T(M) a triangulation of
M. To T(M) corresponds a function ET(X) defined over conv(M) as the power
of a point X with respect to the sphere E circumscribing any d-simplex of T(M)
that contains X. By the results of subsection 17.2.6, ET(X) is well-defined when
X belongs to several cells.
Lemma 17.3.7 Let Det(M) be a Delaunay triangulation of M and T(M) be

any other triangulation of M. Then
VX E conv(M), EDet(X) > ET(X).
Proof. Consider a d-simplex T in T(M) that contains X, E its circumscribed

sphere, and q(T) the d-simplex of Ed+l whose vertices are the images under 4 of
the vertices of T. (Recall that these vertices are obtained by lifting the vertices of
T onto the paraboloid P.) Lemma 17.2.3 shows that ET(X) is the signed vertical
distance (here negative) from O(E)* to O(X). Notice that O(E)* is the affine hull
of O(T). For a given X, this signed vertical distance is maximized when +(T) is
a face of the convex hull of +(M) in Ed+l: in other words, when T is a simplex
of a Delaunay triangulation of M. El
Lemma 17.3.8 If T is a d-simplex and if ET is its circumscribed sphere, then
min ST(X) = ET(CT) =-rT

XET ~ (T
where CT and r' are respectively the center and the radius of the smallest sphere
enclosing T.
Proof. Let ET be the sphere circumscribed to T, CT its center, and rT its radius.
Then
ST(X) = XCT -rT
is minimized when X = CT and is therefore greater than -rT. If CT is contained
in T, the smallest enclosing sphere of T is ET, hence r+ = rT and the lemma
is trivial. Otherwise, the smallest enclosing sphere of T is centered on a k-face
(k < d), namely the face F such that the orthogonal projection of CT onto the
plane that supports F falls inside F. The radius r+ of this sphere is that of
the (k - 1)-sphere circumscribed to F. Its center CT minimizes the value of
XCT * XCT when X E T. Pythagoras' theorem then shows that
CTCT + rT = rT2
which finishes the proof. E]
Let T(M) be any triangulation of a set M of points in Ed. For each simplex T
in T(M), we let r' denote the smallest radius of a sphere that encloses T, and
the maximum min-containment radiusof T(M) is defined by
C(T(M)) = max rT.

TET(M)
The most compact triangulations are then defined as the triangulations that min-
imize the maximum min-containment radius.
Theorem 17.3.9 Delaunay triangulationsare the most compact among all the
triangulations of M.
Note that since the maximum min-containment radius C(T(M)) is defined only
by the simplices T of T(M) such that C(T(M)) = r', triangulations other than
Delaunay triangulations might also be most compact among the triangulations
of M.
Proof. Let T(M) be any triangulation of M and let Det(M) be any Delaunay
triangulation of M. We denote by XT the point X that minimizes E-(X) and

by Xet the point X that minimizes EDet(X). From lemma 17.3.8, we know that
Xf is the center of the smallest sphere that encloses the simplex in T(M) which
contains XT. We denote its radius by r'-. Likewise, XDet is the center of the
smallest sphere that encloses the simplex in Det(M) that contains XT and its
radius is denoted by r' ,et. The maximum min-containment radius of T(M) equals
r/ and that of Det equals r'et. Using lemmas 17.3.7 and 17.3.8, we obtain
2 2
ET(XT) = -rT <• T(XDet) < EDet(X~et) = -rvet
Equiangularity (d = 2)
We now restrict the discussion to triangulations of a set of points in the plane.

Given a triangulation T(M) of a set M of n points in the plane, we define its angle
vector as the vector Q(T(M)) = (al,.. ., O3t) where the ai's are the angles of
the t triangles of T(M) sorted by increasing value. We know that Et 1 ai = t7r.
Note that a triangulation that maximizes the angle vector for the lexicographic
order also maximizes the smallest of its angles. Such a triangulation is called
globally equiangular.
Theorem 17.3.10 A globally equiangular triangulationof a set M of points in

the plane is always a Delaunay triangulation.
Proof. We must prove that, among all the triangulations of M, the ones that
maximize the angle vector for the lexicographic order are always Delaunay tri-
angulations. Let us thus consider two triangles T1 = ABC and T2 = BCD in
some triangulation T(M), such that the union of T1 and T2 is a strictly con-
vex quadrilateral Q. (This means that A, B, C, and D are all vertices of the
convex hull conv(A, B, C, D).) In order to increase the equiangularity, we can
flip the diagonal as follows (shown in figure 17.8). If the triangles Tl = ABD
and T2 = ACD are such that Q(Tj',T2) > Q(Ti,T 2 ), then replace T(M) by a
triangulation Tl(M) which contains T, and T2 instead of T1 and T2 .
The previous rule may be dubbed a regularization rule since it transforms a
pair of adjacent triangles into a regular pair of triangles: if the two triangles do
not form a convex quadrilateral, then the pair is obviously regular, and the rule
does not apply; otherwise, T1 U T2 is convex and the pair is transformed into a
regular pair. Indeed, let El and E2 be the circles circumscribed to T1 and T2 .
We will show that the diagonal AD is flipped if and only if D is contained inside
the circle El. Let a, /, -y, and 6 be the angles at the vertices of the quadrilateral
ABCD, a, 31, and -1 the angles at the vertices of T1 , and /32, 'Y2, and 6 the angles
B B
D D
-- 3-y
A A
C- C
Figure 17.8. Flipping a diagonal to increase the equiangularity.
at the vertices of T2. Moreover, we denote by a', 3,and 6' the angles at the
vertices of T, and ca' y, and 6' the angles at the vertices of T2. The situation is
depicted in figure 17.8. If a is the smallest angle in T1 and T2 , then the diagonal
is not flipped. But then
6 = 7r-132 - '2 < 7r - a,
so that a+6 < 7r, which shows that A is not contained inside E2, and this implies
that D is not contained inside El. The situation is entirely symmetric when the
smallest angle is 6. When the smallest angle is 01, then we flip the diagonal only
if 6' is greater than /1,which only happens when D is contained inside El. Of
course, the cases when the smallest angle is -y1, /2, or 'Y2 are entirely similar, so we
have shown that the diagonal is flipped if and only if it transforms the irregular
pair (Ti, T2 ) into a regular pair (T1,T2).
Clearly, after a flip we have Q(T1 (M)) > Q(T(M)). Flipping the edges when-
ever possible progressively increases the angle vector of the triangulation. Since
there are only a finite number of triangulations, this process eventually reaches a
triangulation that has only regular pairs of adjacent triangles. This triangulation
is a Delaunay triangulation as is shown by theorem 17.3.6. D
Note that this local regularization always leads to a Delaunay triangulation.

When the points are in L2 -general position, there is only one Delaunay triangu-
lation: the result of the procedure described above therefore does not depend on
the starting configuration, nor on the order chosen to flip the diagonals.
When the points are not in L2 -general position, however, the theorem above
shows that flipping diagonals only reaches a Delaunay triangulation. Yet there are
several Delaunay triangulations, which may not all have the same angle vectors.
Still, there is an algorithm that can reach a globally equiangular triangulation
(see the bibliographical notes).
17.4. Higher-order Voronoi diagrams 425
17.4 Higher-order Voronoi diagrams
In this section, we define Voronoi diagrams of order k and show the connection
between these diagrams and the faces at level k in a hyperplane arrangement
in Ed+1. As usual, the Euclidean space Ed of dimension d is embedded in Ed+1
as the hyperplane Xd+1 = 0, and 4(M)* denotes the hyperplane in Ed+1 that is
tangent to the paraboloid P at the point O(M) obtained by lifting M vertically
onto the paraboloid.
In section 17.2, we established the connection between the Voronoi diagram
of a set M of n points M 1 ,.. ., Mn in Ed and the polytope in Ed+1 that is the
intersection of the n half-spaces O(Mi)*+ that lie above the hyperplanes O(Mi).
Equivalently, V(M) is the cell at level 0 in the arrangement A of the hyperplanes
O(M 1 )*,..., O(Mn)*, if the reference point is on the Xd+1-axis, sufficiently high
so that it is above all the hyperplanes. Let us recall that a point is at level k in
A if it belongs to exactly k open half-spaces O(Mi,)*-,...., 0(Mik)*-, such that
each (Mij)*- is bounded by (Mij)* and does not contain the reference point
(see section 14.5).
It is tempting to consider the cells at levels k > 0. We define below a cell
complex that spans Ed, called the Voronoi diagram of order k of M, and show in
theorem 17.4.1 that the cells of this complex are the non-overlapping projections
onto Ed of the cells at level k in the arrangement A.
Let Mk be a subset of size k of M. The Voronoi region of Mk is the polytope
Vk(Mk) of the points in Ed that are closer to all the sites in Mk than to any
other site in M \ Mk. Formally,
Vk(Mk) = {X: VMi E Mk, VMj E M \ Mk, 11XMiI • IIXMi II.
Let us consider all the subsets of size k of M whose Voronoi regions are not
empty. As proved in the theorem below, these polytopes and their faces form a
d-complex whose domain is Ed. This complex is called the Voronoi diagram of
order k of M (see figures 17.9 and 17.10). It is denoted by Vork(M). When
k = 1, we recognize the definition of the usual Voronoi diagram.
Theorem 17.4.1 The Voronoi diagram Vork(M) of order k of a set M =

{M1i....,Mn of n points in Ed is a cell complex of dimension d in Ed. The
cells of this complex correspond to the cells at level k in the arrangement A of
the hyperplanes o(Mi)*,..... , (Mn)* in Ed+l, when the reference point is on the
Xd+1-axis above all the hyperplanes. A cell of Vork(M) is obtained by project-
ing vertically onto Ed the corresponding cell in A. The 1-faces of Vork(M) are
obtained by projecting the l-faces common to cells at level k in A.
426 Chapter 17. Euclideari metmc
Figure 17.9. The Voronoi diagram of order 2 of the points in figure 17.1.
Figure 17.10. The Voronoi diagram of order 3 of the points in figure 17.1.
Proof. The proof relies on lemma 17.2.4. A sphere in Ed whose interior contains
k points is mapped by q to a point at level k in the arrangement .A of the
hyperplanes O(M1 )*, . . ., O(Mn)*.
More precisely, X belongs to the cell Vk(Mk) in the Voronoi diagram of order
k, if and only if X is the center of a sphere E whose interior contains the points
17.4. Higher-order Voronoi diagrams 427
Zdl I
to
V 2 (MiM 2 ) l V2 ( 2 M3 ) ' V 2 (M3 M 4 )

W >P, * . 1.I 1
Ed
_d
as=
42=
'14M
'11
Figure 17.11. The Voronoi diagram of order 2 is obtained by projecting the cells at level 2
in V(M).
of Mk and only those. Then +(S) belongs to the k closed half-spaces below the
hyperplanes O(Mj)* for the Mj's in Mk, and only to those half-spaces. The cells
of the Voronoi diagram of order k are obtained by projecting vertically the cells
at level k of A (see figure 17.11).
It is easily verified that any vertical line intersects at least one cell at level k in
A and does not intersect the interior of more than one cell at level k. It follows
that the i-faces, for I < d, of the Voronoi diagram of order k are obtained by
vertically projecting the i-faces common to several cells at level k. If the Mi's are
in L2 -general position, then the hyperplanes 0(Mi)*,.. ., O(Mn)* are in general
position. In that case, it was shown in section 14.5 that the cells of A that contain
an i-face F at level k have levels that vary between k and k + d + 1 - 1. Among
those, there is only one cell at level k and one cell at level k + d + 1 - 1, and
several cells at levels k < j < k + d + 1 - 1. It follows that the vertical projection
of F is an l-face of the Voronoi diagrams of orders k + 1, k + 2, .. ., k + d-1. D
Having computed the Voronoi diagram of order k of the sites, looking for the
k nearest sites of any point X in Ed can be performed by finding the cell of the
diagram that contains X (see exercises 17.2 and 17.4).
It follows from the construction that the total complexity of the Voronoi di-
agrams of all orders k, 1 < k < n - 1, is O(nd+1): indeed it is exactly the
complexity of the arrangement of the hyperplanes in Ed+l. Moreover, these dia-
grams can be computed in time 0(nd+1) (see theorem 14.4.4). An upper bound
on the complexity of the Voronoi diagrams Vorn(M),...,Vork(M) of orders
between 1 and k is provided by theorem 14.5.1, which bounds the complexity
of the first k levels in the arrangement of n hyperplanes in Ed+1. Moreover,

Vor<k(M) = {Vorl(M),. .. ,Vork(M)} can be computed by computing the first
k levels in this arrangement (see theorem 14.5.3 and exercise 14.22) and projecting
back onto Ed.
Theorem 17.4.2 The overall complexity of the first k Voronoi diagrams of a set
of n points in Ed is O(n[(d+1)/2JkF(d+1)/ 2 J). These k diagrams may be computed
in time O(nL(d+l)/ 2 ] kF(d+l)/2 1) if d > 3, or O(nk2 log n) if d = 2.
Let us close this section by observing that Vorn-i is the complex whose cells
consist of the points further from a particular site than from any other site. This
is why this diagram is sometimes called the furthest-point Voronoi diagram. The
vertices of this diagram are the centers of spheres circumscribed to d + 1 sites
and whose interiors contain all the other sites. Its cells are all unbounded. The
furthest-point diagram can be obtained by computing the intersection of the n
lower half-spaces bounded by the hyperplanes 4(Mi)*, i = 1, ... , n.
17.5 Exercises
Exercise 17.1 (Shortest edge) Denote by F and 5 two finite sets of points in Ed.
Show that the shortest edge that connects a point in F to a point in 5 is an edge of the
Delaunay triangulation of F U 5. From this, conclude that each point is adjacent to its
nearest neighbor in the Delaunay triangulation.
Exercise 17.2 (Nearest neighbor in the plane) Show that, given the Voronoi dia-
gram of a set M of points in the plane, it may be preprocessed in linear time to answer
nearest neighbor queries (that is, find the nearest site to a query point) in logarithmic
time. Same question for the set of k nearest neighbors (k fixed).
Hint: One may use the result of exercise 12.2.
Exercise 17.3 (On-line nearest neighbor) We place point sites in the plane and we
want to maintain a data-structure on-line so as to answer nearest neighbor queries on
the current set of sites (that is, find the nearest site to a query point). Devise a structure
that stores n sites using storage 0(n), such that under the assumption that the points
are inserted in a random order, the expected time needed to insert a new site is 0(log n),
and that answers any query in expected time 0(log2 n).
Hint: One may build a two-level data-structure in the following way. The first level
corresponds to a triangulation of the Voronoi diagram, obtained by connecting a site to
all the vertices of its Voronoi region. Build an influence graph for the regions defined as
the triangles in this triangulation (it is a variant of the influence graph used in exercise
17.10). Each triangle points to the site that kills it, and all the triangles created after the
insertion of a site are sorted in polar angle around this site and stored into an array: this
17.5. Exercises 429
is the second level of the data structure. Show that the query point belongs to O(logn)
triangles on the average, and hence that only O(log n) binary searches are performed in
the arrays of the second level.
Exercise 17.4 (Nearest neighbor) Consider a set of n point sites in Ed. Explain how
to design a data structure of size 0(nrd/211+6) that allows the nearest site to any query
point P to be found in logarithmic time.
Hint: Use the solution of exercise 14.12 in Ed+l. The exponent E can be removed, at
the cost of increasing the query time to O(log 2 n) (see the bibliographical notes at the
end of chapter 14).
Exercise 17.5 (Union of balls) Use lemma 17.2.4 to reduce the problem of computing
the union of n balls in Ed to that of computing the intersection of the paraboloid P with a
polytope in Ed+l. Conclude that the complexity of the union of n balls is O(n F21). Devise
an algorithm that computes the union of n balls in expected time E(n log n + n I61).
Exercise 17.6 (Intersection of balls The results of exercise 17.5 are also valid for
the intersection of n balls in Ed. In E , show that if the balls have same radius, the
complexity of the intersection is only O(n) and propose an algorithm that computes this
intersection in expected time E(nlogn).
Hint: Show that each face of the intersection is "convex", meaning that given any two
points in any face, there is an arc of a great circle joining these points which is entirely
contained in that face; then use Euler's relation. For the algorithm, use a variant of the
randomized incremental algorithm of section 8.3.
Exercise 17.7 (Minimum enclosing ball) Show that the center of the smallest ball
whose interior contains a set M of points in E2 is either a vertex of the furthest-point
Voronoi diagram (of order n - 1) of M, or else the intersection of an edge of this diagram
(on the perpendicular bisector of two sites A and B) with the edge AB.
Exercise 17.8 (Centered triangulation) Consider a triangulation T(M) of a set M

of points in Ed. Show that, if each simplex in T(M) contains the center of its circum-
scribed sphere, then T(M) is a Delaunay triangulation of M. Construct a counter-
example for the converse.
Hint: Show that any adjacent pair of d-simplices is regular.
Exercise 17.9 (Non-optimality of the Delaunay triangulation) Construct a set

of points in the plane whose Delaunay triangulation does not minimize the greatest
angle among all the possible triangulations. Same question to show that the Delaunay
triangulation does not necessarily minimize the length of the longest edge, nor the total
length of the triangulation (sum of the lengths of the edges).
Exercise 17.10 (Incremental algorithm) Let Del(M) be the Delaunay triangula-

tion of a set M of points in L2 -general position in Ed. Let A be a point of Ed distinct
from the points in M, let S be the sub-complex formed by the d simplices in Del(M)
whose circumscribed spheres contain A in their interior, and let F be the set of (d - 1)-
faces on the boundary of S. Show that if A belongs to the convex hull of M, then the
d-simplices in Del(M U {A}) are exactly the simplices in Del(M) that do not belong to
S and the simplices conv(A, F), F E F. Generalize this result to the case when A is
outside the convex hull of M, and derive an incremental algorithm that computes the
Delaunay triangulation. Show that this algorithm runs in time e(n i1 +1) in the worst
case. Show that if the points are inserted in random order, then the algorithm runs in
expected time O(n log n + n 121 ), which is optimal. Devise a dynamic algorithm that also
allows points to be removed.
Hint: Use a randomized algorithm with an influence graph. Objects are sites, regions
are the balls circumscribed to d+ 1 sites, and an object conflicts with a region if it belongs
to that region. Show that the ball circumscribed to any new simplex S = conv(A, F) is
contained in the union of the two balls circumscribed to T and V, the two d-simplices
that share the common facet F. Build an influence graph in which each node is the
child of only two nodes, namely the node corresponding to F is the child of the nodes
corresponding to T and V. The number of children of a node is not bounded, but the
analysis can be carried out using biregions (see exercise 5.7).
Exercise 17.11 (Flipping the diagonals) Devise an incremental algorithm to com-

pute the Delaunay triangulation of points in the plane which, at each step, connects the
new point to the edges of the triangle that contains it, and then regularizes the triangu-
lation as in the proof of theorem 17.3.10. Show that, if the points are inserted in random
order, the algorithm can be made to run in expected time O(n log n), which is optimal.
Exercise 17.12 (Flipping in higher dimensions) Generalize the local regularization

rule introduced in the proof of theorem 17.3.10 to the triangulation of point sites in E3
and in higher dimensional spaces. Show that this does not always result in the Delaunay
triangulation of the points, in contrast with the planar case.
Hint: As was done for planar triangulations (proof of theorem 17.3.10), local regular-
ization in E3 corresponds to replacing the upper facets of a simplex in E4 by its lower
facets. A simplex in E4 having five facets, local regularization in E3 leads to replacing
two adjacent tetrahedra T1 and T2 by three tetrahedra T3, T4, and T5 that are pairwise
adjacent (and have the same vertices as T1 and T2 ), or the converse. Show that the local
regularization rule cannot always be applied even though the triangulation is not regular
everywhere.
Exercise 17.13 (Flipping in higher dimensions) Show that if one adds a new point
P to a Delaunay triangulation Det(M) of a set M of points in Ed, the Delaunay triangu-
lation Det(M U {P}) can be obtained by splitting the simplex of Det(M) that contains
P into d + 1 new simplices, and then applying the generalized local regularization of
exercise 17.12. Show that if the n points in M are inserted in a random order, this
incremental algorithm computes Det(M) in expected time O(n logn + nJll), which is
optimal.
Hint: As in exercise 17.12, each regularization in Ed corresponds to replacing the

upper facets of a (d + 1)-simplex in E +1 by its lower facets. Show that any (d + 1)-
simplex S involved in any step of the local regularization has P as a vertex and that the
convex hull of its vertices other than P is a d-simplex of Det(M) that is destroyed by
the local regularization.
Exercise 17.14 (Complexity of the Voronoi diagram of order k) Show that the
complexity of the Voronoi diagram of order k of n points in E2 is always O(k(n - k))
and can be Q(k(n - k)) in the worst case.
Exercise 17.15 (Higher-order Voronoi diagrams and polytopes) Let M be a set

of n points M 1 ,..., M in Ed. With each subset Mk = {Mil,...,MikI of size k of
M, we associate its center of gravity G(Mk) = 1 Ek Mi3 , and the real number
a(Mk) = k >>1 M?. Show that the Voronoi diagram of order k of M is the pro-
jection of the polytope in Ed defined as the intersection of the half-spaces lying above
the hyperplanes polar to the points (G(Mk), a(Mk)), for all subsets Mk of size k of M.
Hint: From the fact that

k 2k k
:(X _Mij) _(X _Mij) =X2 _ ' i *X+ E tj
j=1 j=1 j=1
we infer that the k nearest neighbors of X are the points in Mk if and only Mk is the
subset for which X has the smallest power with respect to the sphere centered at G(Mk)
and whose power with respect to the origin is Or(Mk).
Exercise 17.16 (Euclidean minimum spanning tree) Consider a set M of n points
in L2 -general position in Ed. A Euclidean minimum spanning tree, or EMST for short,
is a tree whose nodes are the points in M and whose total edge length is minimal. Show
that such a tree is a subgraph of the Delaunay triangulation of M. For the planar case,
show that an EMST can be computed in time O(n log n). Consider the case where the
set of points is not in L 2 -general position any more.
Hint: Show that the following greedy algorithm produces a minimum spanning tree.
Denote by A the set of points of M that are already connected to the current tree. The
greedy algorithm picks the shortest segment that does not induce a cycle in the current
subtree. This edge connects a point of A to a point of M \ A. The latter point is added
to A, the edge is added to the tree, and so on until the tree spans M. Show that this
yields an EMST, even if the points are not in L2 -general position. Exercise 17.1 shows
that it can be completed into a Delaunay triangulation of M. Explain how to make the
algorithm run in time O(nlogn).

Voronoi diagrams have been used for a long time and in various disguises. Voronoi,
a Russian mathematician of the early twentieth century, was the first to give them a
precise definition and study them for their own sake, but they had already been used by
Wigner and Steiz in crystallography, by Thiessen in geography, and even by Descartes

in astronomy. Delaunay [78] established most of the fundamental properties about the
triangulation that bears his name.
More recently, the connection between Voronoi diagrams and polytopes was discovered
by Brown [38] and by Edelsbrunner and Seidel [96]. The presentation in this book is based
on the works of Boissonnat, C&r6zo, Devillers, and Teillaud [26] and of Devillers, Meiser,
and Teillaud [82]. The connection with polytopes answers exercise 17.4.
The optimality of the Delaunay triangulation was established by Rajan [193] for the
compactness, and by Lawson [144] for the equiangularity. Mount and Saalfeld [170] have
proposed an algorithm to compute a globally equiangular triangulation when the points
are not in L2 -general position. In the context of approximating surfaces by piecewise
linear patches, controlling the equiangularity serves to control the quality of the approxi-
mation even though it is more profitable to minimize the greatest angle as was shown by
Nielson and Franke [179]. The Delaunay triangulation does not generally minimize the
greatest angle, nor the total edge length, even though it often works well for practical
instances. Recent references on these topics can be found in the works of Edelsbrunner
and Tan [101], Edelsbrunner, Tan, and Waupotitsch [102], and Dickerson, McElfresh, and
Montague [83]. Rippa [195], and also Rippa and Schiff [194], gave other useful criteria in
the context of approximating surfaces for which the Delaunay triangulation is optimal.
Desnogues [79] provides a good survey of polyhedral approximation.
An incremental algorithm that computes the Voronoi diagram of a set of points (see
exercise 17.10) was given by Green and Sibson [113]. The Delaunay tree introduced by
Boissonnat and Teillaud [31, 32] improves on the average performance when the sites are
inserted in random order. This algorithm was made fully dynamic by Devillers, Meiser,
and Teillaud [81]. An algorithm that proceeds by flipping diagonals was proposed by
Lawson [144], then dynamized in the plane by Guibas, Knuth, and Sharir [117] who
present a randomized analysis and also solutions to exercises 17.3 and 17.11. Its general-
ization to higher dimensions was studied by Joe [131, 132], Rajan [193], and Edelsbrunner
and Shah [98], who provide a solution to exercise 17.13.
Lee proposed the first algorithm that computes Voronoi diagrams of higher orders in
the plane [145]. He also gave a solution to exercise 17.14. The connection between Voronoi
diagrams of order k and polytopes (exercise 17.15) was established by Aurenhammer [15].
Boissonnat, Devillers, and Teillaud [29] and also Mulmuley [174] proposed semi-dynamic
or even fully dynamic algorithms that compute the Voronoi diagrams of all orders up to k
in any dimension. Clarkson [67], Aurenhammer and Schwarzkopf [18], and also Agarwal,
de Berg, and Matousek [2] gave randomized algorithms that compute Voronoi diagrams
of a single order k, rather than all the diagrams of orders < k.
The connection between Delaunay triangulations and Euclidean minimum spanning
trees (see exercise 17.16) is discussed in the book by Preparata and Shamos [192], where
one can also find a linear time algorithm that computes the Euclidean minimum spanning
tree knowing the Delaunay triangulation. Conversely, Devillers [80] gave a randomized
algorithm that computes the Delaunay triangulation of a set of n points in the plane
knowing its Euclidean minimum spanning tree in expected time O(n log* n).
For other references on Voronoi diagrams, the reader is referred to the book by Okabe,
Boots, and Sugihara [182] or to the survey articles by Aurenhammer [16] and Fortune
[105].
Chapter 18
Non-Euclidean metrics
In the previous chapter, we established a correspondence between the points

in Ed and certain hyperplanes in Ed+1, namely the hyperplanes tangent to the
paraboloid P. It is tempting to define the analogue of V(M) for a more general
set of hyperplanes that may not necessarily be tangent to P. In that case, the
intersection of the n half-spaces lying above the hyperplanes is again a polytope
whose proper faces, projected onto Ed, form a cell complex that covers Ed en-
tirely. This complex generalizes the Voronoi diagram and can be considered as
the Voronoi diagram of a family of spheres, when the distance is defined as the
power of a point with respect to one of the spheres. This interpretation, to be
detailed in section 18.1, justifies the appellation power diagrams for such dia-
grams. These diagrams play a central role in several generalizations of Voronoi
diagrams: in particular, we explore affine diagrams, which are Voronoi diagrams
of point sites for a general quadratic distance (see section 18.2), and diagrams
for weighted distances (see section 18.3).
Not all Voronoi diagrams for different metrics can be cast into power diagrams.
For instance, polyhedral distances (and especially L1 and Loo) have important
applications and are studied in section 18.4, and an application of hyperbolic
Voronoi diagrams (see section 18.5) is given in the next chapter.
The representation of spheres introduced and used in the previous chapter is
again very useful for computing power diagrams and hyperbolic Voronoi dia-
grams. In addition to this representation of spheres, we introduce in this chapter
a new way of looking at Voronoi diagrams that is helpful for studying weighted
diagrams, L1 and Loo diagrams, and for the algorithms in the next chapter. In-
tuitively, the Voronoi diagram of a set M of points can be interpreted as the
result of a growth process starting with the points in M. Indeed, imagine crys-
tals growing from each point of M at the same rate in all directions. The growth
of a crystal stops at the points where it encounters another crystal, because of
the constraint that the crystals may not interpenetrate. The crystal originating
434 Chapter 18. Non-Euclidean metrics
at a point Mi in M covers the region that is reached by that crystal first, or in

other words the points that are closer to Mi than to any other point in M: this
is exactly the Voronoi cell of Mi.
This growth process in Ed can be visualized in Ed+l by adding another coordi-
nate, considered as the time elapsed since the start of the growth process. Thus
Ed corresponds to the hyperplane Xd+1 = 0 in Ed+1, and the isotropic growth of
a point Mi is a cone of revolution with vertex Mi and vertical (that is, parallel
to the Xd+1-axis) axis. The faces of the Voronoi diagram appear as the projec-
tions onto Ed of faces on the lower envelope of the cones. If the sites do not
start growing at the same time, the cones are translated vertically: this leads to
Voronoi diagrams with additive weights. If the sites do not grow at the same
rate, then the angles of the cones are different: the resulting Voronoi diagrams
have multiplicative weights. If the sites do not grow isotropically (namely at the
same rate in all directions), the cones are no longer cones of revolution: in this
way we can generate Voronoi diagrams for the L1 and Loo distances, and more
generally for polyhedral distances (see exercise 19.3).
Throughout this chapter, the "distances" we consider are not exactly distance
functions in the mathematical sense. In fact, we will only require that the distance
function is increasing.
18.1 Power diagrams

18.1.1 Definition and computation
Let S = {S,..., n} be a set of n spheres in Ed. To each Ei corresponds a
region P(Ei) of Ed, consisting of the points whose power with respect to Ei is
smaller than their powers with respect to the other spheres:
P(Ei) = {X E Ed : Vj + i, Ei(X) < Ej(X)}.
The region P(Ei) is the intersection of a finite number of half-spaces (bounded

by the radical hyperplanes Hij, j = 1, ... , n, j :$ i). It is therefore a convex
polytope, occasionally empty or unbounded. The P(Ei)'s and their faces form a
cell complex which covers Ed: this complex is called the power diagram of S and
we denote it by Pow(S) (see figure 18.1).
As in the previous chapter, we map a sphere E in Ed, centered at C and of
equation E(X) = 0, to the point O(E) = (C, E(0)) in Ed+1. The hyperplane polar
to O(E) with respect to the paraboloid P is denoted by O(E)*: if Ed is embedded
in Ed+l as the hyperplane Xd+1 = 0, then O(E)* is the hyperplane that intersects
P along the quadric obtained by lifting E onto P (see figure 17.4).
Let P(S) be the intersection of the half-spaces bounded below by the polar
hyperplanes O(i)*, to the points O(Ei), **, ,O(E.
.. ., 0(En)*
18.1. Power diagrams 435
Figure 18.1. A power diagram.
Theorem 18.1.1 The power diagram Pow(S) of a set S = {Yh, , n} of n

spheres in Ed is a cell complex in Ed+l. Its faces are obtained by projecting
the proper faces of the unbounded (d + 1)-polytope P(S), the intersection of n
half-spaces in Ed+l bounded below by the polar hyperplanes O(El)*, . . ., O(En)*
Proof. Let A be a point on a facet of P(S) which is supported by the polar

hyperplane O(Ei)*, and let A be its projection onto Ed. The power of A with
respect to Ei is the signed vertical distance from A to +(A) (lemma 17.2.3). Since
A belongs to a facet of P(S), the power of A with respect to Ei is less than or
equal to the power of A with respect to any other sphere in S. In other words,
A belongs to the cell P(Ei) that corresponds to Ei in the power diagram. 0
Note that when the hyperplanes O(Ei)* are in general position, P(S) is a simple
polytope in Ed+l, so each vertex is incident to d + 1 hyperplanes. In terms of
the spheres Ei, this general position assumption means that no subset of d + 2
spheres in S are orthogonal to a common sphere in Ed, or equivalently that no
point in Ed has the same power with respect to d + 2 spheres in S. In this case,
we say that the spheres are in general position. The power diagram Pow(S) is a
cell complex whose vertices have the same power with respect to d + 1 spheres in
S (and are therefore the centers of spheres orthogonal to d + 1 spheres in S), and
have a greater power with respect to the other spheres in S. More generally, a
k-face in Pow(S) is formed by the points that have the same power with respect
to d + 1 - k given spheres in S, and a greater power with respect to the other
spheres in S.
Corollary 18.1.2 The complexity of the power diagram of n spheres in Ed is

O(n d/2 l ). The diagram can be computed in time O(n log n + nrd/21), which is
optimal in the worst case.
Remark 1. In the case of Voronoi diagrams, all the hyperplanes O(Mi)* are
tangent to the paraboloid P, and each of them contributes a facet to V(M).
For power diagrams, however, a polar hyperplane O(Ei)* does not necessarily
contribute a face to P(S). Such a hyperplane is called redundant. In the power
diagram, it means that the cell P(Ei) is empty: Ei does not contribute a cell to
Pow(S).
Remark 2. There is no particular difficulty if some, even all, the spheres in S,
are imaginary. This fact is used in section 18.3.2.
Remark 3. Any polytope in Ed+l that is the intersection of upper half-spaces
corresponds to a power diagram: if H 1 ,... , H,n are the hyperplanes that bound
these half-spaces, then their upper envelope projects onto the power diagram of
the spheres 0- 1 (Hf*),...,4 -(H*).
18.1.2 Higher-order power diagrams

As was done for Voronoi diagrams in section 17.4, we may define power diagrams
of higher orders.
Let Sk be a subset of S of size k. We call the power cell of Sk the set P(Sk)
of points in Ed that have a smaller power with respect to any sphere in Sk than
to any sphere in S \ Sk:
P(Sk) = {X e Ed : V8i G Sk, V8j e S \ Sk, Ei(X) < Ej(X)}.
Consider all the subsets of size k of S whose corresponding power cell is not
empty. These regions and their faces form a cell complex that covers Ed entirely,
and that is called the power diagram of order k of S. We denote it by PoWk(S).
This fact is a consequence of the theorem below, whose proof closely resembles
that of theorem 17.4.1. This theorem clarifies the links between power diagrams
of order k in Ed and faces at level k in the arrangement of n hyperplanes in Ed+l.
As usual, the Euclidean space of dimension d is identified with the hyperplane
Xd+1 = 0 in the space Ed+l of dimension d + 1, and 0(E)* stands for the polar
hyperplane of +(S).
Theorem 18.1.3 Consider a set S = {E,l . , En} of spheres in Ed, and let
A be the arrangement of their polar hyperplanes O(El)*, .**, O(En)* The power
diagram of order k, Powk(S), is a cell d-complex in Ed. Its cells are the vertical
projections of the cells at level k in the arrangement A, the reference point being
on the Xd+ 1 -axis above all the hyperplanes O(i)*, i = 1,... ,n. The I-faces of
Powk(S), 1 < d, are obtained by projecting the l-faces common to at least two
cells of A at level k.
18.2. Affine diagrams 437
From theorems 14.5.1 and 14.5.3, we derive the following result.
Theorem 18.1.4 The complexity of the first k power diagrams of a set of n

spheres in Ed is Q(nL(d+1)/ 2 Jk[(d+l)/21). These k diagrams can be computed in
time O(nL(d+1)/2 JkF(d+l)/2 1) if d > 3, and in time O(nk2 log n) if d = 2.
18.2 Affine diagrams

The notion of a Voronoi diagram can be extended to more general sites or to non-
Euclidean distances. A particularly interesting extension occurs when the locus
of points equidistant from two sites is a hyperplane: in this case, the diagram
is called an affine diagram. Voronoi diagrams and power diagrams are affine
diagrams, and we will show that any affine diagram is a power diagram. Moreover,
certain non-affine diagrams can be derived from an affine diagram and therefore
from a power diagram: this is notably the case of diagrams with additive or
multiplicative weights studied in section 18.3.
18.2.1 Affine diagrams and power diagrams

An affine diagram is a diagram defined for object sites and for a distance such
that the set of points equidistant from two objects is a hyperplane. The cells of
such diagrams are thus convex polytopes and affine diagrams can be identified
with cell complexes.
To any affine diagram of n objects corresponds a set of ( 2 ) perpendicular
bisectors Hij, 1 < i < j < n. These hyperplanes must satisfy the relations
Hi, n Hk= Hij n Hik = Hikn Hjk = Iijk
for any 1 < i < j < k < n.

We say that the diagram is simple if the Iijk are disjoint and not empty.
Theorem 18.2.1 Any simple affine diagram in Ed is the power diagram of a set
of spheres in Ed.
Proof. We embed Ed in Ed+1 as the hyperplane Xd+1 = 0. The proof consists

of constructing a set of n hyperplanes Pi, . .. , Pn in Ed+l such that the vertical
projection of Pi n Pj for i < j is exactly Hij. Assuming these hyperplanes are
known, to each Pi corresponds a sphere Ei = 0-l(Pi*) whose polar hyperplane
is exactly Pi: Ei is also the projection on Ed of the intersection of Pi with the
paraboloid 'P. Hence Hij is the radical hyperplane of Ei and Ej for all i and j
such that 1 < i < j < n. It follows that the affine diagram is exactly the power
diagram of the spheres Ei, i = 1, . . . , n.
We now show how to build the Pi's. Denote by hij the vertical projection of
Hij onto Pi (note that i < j by the definition of Hij).
Let us take for PI any non-vertical hyperplane, and for P2 any non-vertical
hyperplane that intersects P1 along h 12 . For k > 3, we must take for Pk the
hyperplane that intersects P1 along h1k and P2 along h2k: such a hyperplane
exists because h1k and h2k intersect along the affine subspace of dimension d - 2
that is the projection of I12k onto P1, P2 , or Pk.
It remains to see that the vertical projection of Pi n Pj is exactly Hij. By
construction, this is true for PlnP 2 , PrnPj, and P 2 nPj, j > 3. For 3 < i < j < n,
we know that PinPjnP1 projects onto Ed along '1ij, and that PinPj n P 2 projects
onto Ed along I2ij. The diagram being simple, Ilij and I2ij must be distinct. The
projection of Pi n Pj must therefore contain I1ij and I2ij, and hence also their
affine hull which is nothing other than Hij.
Below, we rather use
Theorem 18.2.2 The affine diagram whose hyperplanes Hij have equations
-2(C,-C)&X++i-o- =0
is the power diagram of the spheres Ei, i = 1,... ,n centered at Ci and with
respect to which the origin has power ai.
Proof. We may simply check that the equation of Hij can be written as Ei(X) -
Ej(X) = 0, which is exactly that of the radical hyperplane of Ei and Ej (see
subsection 17.2.6).
18.2.2 Diagrams for a general quadratic distance
Consider two points X and A in Ed. By the general quadratic distance from A
to X, we mean the quantity
6Q (X, A) = (X - A) (X - A)t + p(A),
where A is a real symmetric d x d matrix, and where p(A) is a real number.

The diagrams encountered so far are all particular cases of Voronoi diagrams
for a quadratic distance:
* Standard Voronoi diagrams are obtained for A = Ed and p(X) = 0.

* Furthest-point Voronoi diagrams (introduced as diagrams of n points of
order n - 1) are obtained for A = -Ed and p(X) = 0.
18.3. Weighted diagrams 439
* Power diagrams correspond to A = Ed and p(X) #&0.

For any pair of points A and B, the set of points X that are equidistant from
A and B is the hyperplane HAB equation
HAB : 2(B - A)AXt + AAA t - BABt + p(A) - p(B) =0.
The Voronoi diagram of a finite set of points for a general quadratic distance is
thus an affine diagram by theorem 18.2.2.
Theorem 18.2.3 The Voronoi diagram of n points for an arbitrary general

quadratic distance in Ed has complexity E3(nrd/ 2 1). It can be computed in time
e(nlogn + n d/21).
18.3 Weighted diagrams

This section introduces two kinds of diagrams which are not affine. They are
defined for finite sets of point sites and for a Euclidean distance that is weighted
additively or multiplicatively. Each distance is appropriately defined in the sub-
section below.
These diagrams are not cell complexes those we have been studying so far.
Nevertheless, they can be given a facial structure that is similar to that of cell
complexes. Consider the equivalence relation shared by the points in Ed that have
the same nearest neighbors. The equivalence classes subdivide Ed into (open)
regions whose closures we call the faces of the diagram. The cells of the diagram
span Ed entirely and the intersection of two faces is a (possibly empty) collection
of lower-dimensional faces.
As we see below, the faces of these diagrams are not polytopes, and may not
even be connected. Nevertheless, these weighted diagrams can be derived simply
from power diagrams.
18.3.1 Weighted diagrams with additive weights

Let M = {M 1 , .. ., Mn} be a set of n points in Ed. To each Mi corresponds a real
ri called the weight of Mi. The additive weighted distance, or additive distance
for short, from a point X in Ed to M, is the quantity
6+(X, Mi) = IIXMi - ri.
The diagram of M with additive weights is defined like the Voronoi diagram
except that the distance used is not the Euclidean distance but the additive
distance defined above. This diagram is denoted by Vor+(M). An instance is
Figure 18.2. A diagram with additive weights. Sites are the centers and their correspond-
ing weights are the radii of the circles. In this example, the diagram of the
points with additive weights is also the Voronoi diagram of the circles for the
Euclidean metric.
shown in figure 18.2. We observe that it is the first example of a non-affine

diagram shown in this book.
Note that adding the same constant to all the points does not modify the
diagram. This lets us assume that all the ri's are non-negative.
The representation of spheres introduced in section 17.2 is not very helpful here
and we use another which shows a very natural correspondence between weighted
Voronoi diagrams and affine diagrams in dimension d + 1.
Consider the sphere Ei in Ed centered at Mi with radius ri, and let 4' be the
bijection that maps Ei to the point V)(Ei) = (Mi, ri) E Ed+l.
The spheres of zero radius correspond to the points in the hyperplane of equa-
tion Xd+1 = 0 in Ed+l.
The points at additive distance r from Mi can be considered as the centers of
the spheres of radius Irl tangent to Ei, that are inside or outside Ei according to
whether r is negative or positive. The images under 4' of these spheres generate
a cone of revolution C(S) of equation
C(E) : xd+l = IIXCII- r
which has apex (C, -r), is symmetrical to O(E) with respect to the hyperplane
Xd+1 = 0, and has an aperture angle of 74.' The vertical projection Ix of a point
X in Ed on the cone C(E) is the image under 0 of the sphere centered at X and
tangent to E. The signed vertical distance from X to Ix equals the additive
distance from X to C weighted by r.
To each sphere Ei, i = 1,. . . , n, corresponds the cone C(Ei), also denoted by
Ci. It follows from the discussion above that the projection of the lower envelope
of the cones Ci onto Ed is exactly Vor+(M).
The set of points in Ed that are equidistant (with respect to the additive dis-
tance) from two points of M is thus the projection of the intersection of two
cones. This intersection is a quadric contained in a hyperplane. Indeed, we have
C1 : (xd+l + ri)2 XM ++ r > 0,

C2 (xd+l + r2) = XM2, Xd+1 + r2 > 0.
The intersection of the two cones is contained in the hyperplane H1 2 whose equa-
tion is obtained by subtracting the two sides of the above equations:
H 12 : -2(Mi -M 2 ) -X-2(rl-r 2 )xd+l+M? -r -M2+r2 =0.
This and theorem 18.2.2 show that there exists a correspondence between the di-
agram Vor+ (M) and the power diagram of the spheres Ei in Ed+l (i = 1, . .. , n),
where Ei is centered at V)(Ei) and has radius riXV (see figure 18.3). More pre-
cisely, the cell of Vor+(M) that corresponds to Mi is the projection of the inter-
section of the cone Ci with the cell of the power diagram corresponding to the
sphere Ei. Indeed, X is in Vor+(Mi) if and only if the projection Xi of X onto
Ci has a smaller Xd+l-coordinate than the projections of X onto the other cones
Cj, j 4 i. In other words, the coordinates (X, Xd+1) of Xi must obey
(Xd+l +ri)2 = XMz
(xd+l + rj)2 < XM? for any j i,
and by subtracting both sides, it follows that V (Xi) < E (Xi) for all j.
The additive diagram can be computed using the following algorithm:
1. Compute Vi, for i = 1, . . ., n.

2. Compute the power diagram of the Ei's.
3. For all i = 1, . .. , n, project onto Ed the intersection with the cone Ci of the
cell of the power diagram that corresponds to V,.
The power diagram of the E4 can be computed in time (n [d/2j+1). The inter-
section involved in step 3 can be computed in time proportional to the number
of faces of the power diagram of the V's, which is 0(nLd/2i+1). We have thus
proved that:
H12
Figure 18.3. Any Voronoi diagram for the additive distance can be derived from a power
diagram in Ed+l.
Theorem 18.3.1 The Voronoi diagram of a set of n points in Ed with additive

weights has complexity O(nLd/ 2 ]+l) and can be computed in time O(nLd/ 2J+l).
This result is optimal in odd dimensions, since the bounds above coincide with
the corresponding bounds for the Voronoi diagram of points under the Euclidean
distance. It is not optimal in dimension 2, however, as we now show. We also
conjecture that it is not optimal in any even dimension.
In the plane, we have seen that additive diagrams can be thought of as the
projection onto E 2 of the lower envelope of cones with vertical axis and aperture
angle 7. Therefore, each cell is connected. Moreover, the vertices of the diagram
are incident to exactly three edges, under the general position assumption, and
these edges are arcs of hyperbolas, each of which is the projection of the inter-
section of two cones. Euler's relation shows that the diagram has complexity
O(n). A perturbation argument shows that the general position assumption is
not restrictive, since allowing degeneracies only merges some vertices and makes
some edges disappear. In section 19.1, it is shown that such a diagram can be
computed in optimal time O(n log n).
18.3.2 Weighted diagrams with multiplicative weights
Let M = {M 1, . .. , Mn} be a set of n point sites in Ed. To each Mi corresponds a

positive real number p(Mi) called the weight of Mi. To simplify the presentation,
we suppose that the p(Mi)'s are all distinct, but the extension to the more general
case presents no additional difficulties.
Figure 18.4. A diagram with multiplicative weights. Sites are represented by small disks,
and the weight of a site is inversely proportional to the diameter of the disk.
By the distance with multiplicative weights, or multiplicative distance for short,

from a point X to a point Mi, we mean the quantity
a* (X, MO) = pAM) VAX~ 11
The Voronoi diagram of M for the multiplicative distance is defined like the
Voronoi diagram, except that the distance is not the Euclidean distance but
rather the multiplicative distance. We denote this diagram by Vor*(M) (see
figure 18.4). Observe that a cell of the diagram need not be connected.
The set of points at equal multiplicative distance from two sites Mi and Mj is
a sphere Eij of equation
Pi (X _Mi)2 =pj(X _Mj)2
with pi = p(M,) 2 . In normalized form, we obtain
x2 - 2PiMi- pjMj .X + PiM -pMj = o.

Pi -Pj Pi - Pj
Using the representation of section 17.2, this sphere is represented in Ed+1 as the
point
) PiM _-p3 Mj PiM 2_pj 3 M2
V PtSj)
PiP pi -Pi
444 Chapter 18. Non-Euclidean metric
Its polar hyperplane Hij with respect to the paraboloid P has equation
Hii(X, xd+1) = (Pi - Pj)Xd+1- 2piMi X + 2pjMj *X + piMi2 - pjM? = 0.
.
The hyperplanes Hij are the radical hyperplanes of spheres Ei in Ed+l (i

1, ... ,n). The sphere Ei, possibly imaginary, is centered at (piMi, - i), and
with respect to it the origin has power piMi2 . This establishes a correspondence
between the diagram Vor*(M) and the power diagram of the spheres Ei. More
precisely, the cell V1(Mi) in Vorn (M) that corresponds to Mi is the projection of
the intersection of the paraboloid P with the cell P(Ei) that corresponds to Ei
in the power diagram of the Ei's. Indeed, if X is a point in Ed and O(X) is its
vertical projection onto the paraboloid P of equation Xd+1 = X2 , then we have
2
X EV.(Mi) p i(X-Mi)
< p 3 (X-Mj) 2 Vj 0 i
- Hij(X,X 2) <O Vji i
: Ei(0X)) < Ej(0X)) vi 7&i
• 9(X) E P(i).
An algorithm that computes the diagram of M with multiplicative weights is:
1. Compute Ei, for i = 1,...,n.
2. Compute the power diagram of the Ei's.
3. For i = 1, . . . , n, project the intersection of the cell that corresponds to EA
in the power diagrams of the si's with the paraboloid P.
This proves the following theorem.
Theorem 18.3.2 The Voronoi diagram of a set of n points in Ed with multiplica-
tive weights has complexity O(nLd/ 21+1) and can be computed in time O(nLd/ 2J+1).
Figure 18.5. An instance of a quadratic multiplicative diagram in dimension 2: n points

are put on a given vertical line and are given the same weight, while 11
2 other
points are aligned on a horizontal line and have the same weight, which is
much larger than the weight given to the points in the first half.
This result is optimal in odd dimensions, since in that case these bounds match
those of the Voronoi diagram of n points in Ed for the Euclidean distance. It is
also optimal in even dimensions (see exercise 18.4). Figure 18.5 shows a quadratic
multiplicative diagram in dimension 2.
18-4. Li and L,,, metric 445
18.4. L1 and L metrics 445
Figure 18.6. Co-cubes in dimensions 2 and 3.
18.4 L1 and Loo metrics

The L 1 distance of a point X = (x1 ,...,xd) in Ed to a point M = (ml, ... ,md)
in Ed is defined as
d
61 (X, M) = E xi - Mil.
i=1
The points at a given distance r from M are thus on a polytope whose vertices
are given by their coordinates xi = mi i r and xj = mj if i #&j, for j = 1, . .. , d.
In dimension 2 this polytope is a tilted square, and in dimension 3 it is a regular
octahedron (see figure 18.6). This polytope is dual to the cube and we call it a
co-cube. Henceforth, a co-cube always means a polytope dual to a cube whose
edges are parallel coordinate axes.
Let M = {M 1, ... , Mn} be a set of n point sites in Ed. The Voronoi diagram
of M for the L1 distance is defined similarly to the Voronoi diagram, except that
the distance used in the definition of the cells is not the Euclidean distance but
the L1 distance. It is denoted by VorL, (M) (see figure 18.7).
We can define a facial structure for this diagram by using the equivalence rela-
tion R shared by the points in Ed that have the same subset of nearest neighbors.
The equivalence classes of R subdivide the space Ed in open regions whose clo-
sures are called the faces of the diagram. The faces of the diagram are piecewise
affine.
If the points in Ed are identified with the hyperplane Xd+1 = 0 in Ed+l, then,
in a way similar to what was explained in subsection 18.3.1, to each point Mi
there corresponds a pyramid Pi of equation
Xd+1 = 6 1 (X, M) -
Figure 18.7. Diagram for the L1 metric.
Let us consider the lower envelope of the Pi's, that is, the graph of the func-
tion minl<i<n 61 (X, Mi). The portion of the lower envelope that belongs to any
Pi projects onto the hyperplane xd+1 = 0 as the cell of the diagram VorL, (M)
that corresponds to Mi. The facets of the Pi's form a collection of d-pyramids.
The lower envelope of these pyramids is a collection of d-faces, and their lower-
dimensional faces include all the lower-dimensional faces of the lower envelope of
the Pi's. The vertical projections onto Xd+1 = 0 of the d-faces of the lower en-
velope of the pyramids form a refinement of the faces of the diagram VOrL, (M).
The complexity of the diagram VorL, (M) can thus be bounded by combining the-
orem 16.3.2 and exercise 16.1, which bound the complexity of the lower envelope
of n d-simplices in Ed+1. This yields
IVorL,(M)I = Q(nda(n)).
This bound is almost tight for certain sets of points that are not in general
position (see exercise 18.9). We conjecture, however, that for points in general
enough position, this bound is not attained and that these diagrams have the
same complexity as their Euclidean counterparts. Later on, we show that this is
indeed the case in dimension 2, for which we give a linear bound. It is also the
case in dimension 3 (see exercise 18.10). If d = 2, the bisector for the Ll-distance
of two points is, in general, a polygonal line formed by three linear pieces; if the
line connecting the two points is parallel to one of the main bisectors, however,
18.4. L1 and Loo metrics 447
I %
Iez
-I in
I
I
I
I-
I~ \%
Figure 18.8. Bisectors for the L1 distance. If the line connecting the two points is parallel
to one of the main bisectors, the bisector is not a polygonal line.
the L1 -bisector is no longer a polygonal line and contains two faces of dimension
2 (see figure 18.8).
We say that two points are in Li -general position if no two points are connected
by a line parallel to one of the main bisectors, and if no four points belong to
a common co-cube. In this case, the bisectors are polygonal lines formed of
three line segments, and VorL,(M) contains n connected cells: indeed, for any
i E {1,. . . , n}, the cell V 1(Mi) that corresponds to a point Mi is star-shaped with
respect to Mi (meaning that if X G V1 (Mi), then the segment XMi is contained
in V1 (Mi)), and is therefore connected. Moreover, each vertex in the diagram
is incident to two or three edges because of the Ll-general position assumption.
The diagram VorLi (M) is therefore a planar map with n cells whose vertices have
degree two or three and whose edges consist of at most three segments. Euler's
relation then shows that the complexity of the diagram is O(n).
If the points are not in Li-general position, then some regions may correspond
to pairs of points (see figure 18.8) and some vertices may be of degree higher than
3. This second complication can be straightened out by simply perturbing the
diagram so as to replace each vertex of degree k > 3 by a small polygonal chain
with k - 2 vertices of degree 3 and k - 3 edges. The number of faces does not
increase in the process, and the number of vertices increases by the same amount
as the number of edges; hence Euler's relation still guarantees that the complexity
of the diagram is O(n). The first complication, however, is more serious and may
allow the size to grow up to quadratic: exercise 18.9 presents such an example
and a way to avoid this problem. The example generalizes to higher dimensions
and the lower bound Q(nd) may be shown to hold for the complexity of Voronoi
diagrams of n points in Ed for the L1 distance.
If the points are in Li-general position, the complexity of the diagram is thus
O(n) in dimension 2 and the algorithm that computes the lower envelope of n
triangles in space (see subsection 16.3.3 ) can be used to compute this diagram
in time 0(n log 2 n) (see corollary 16.3.3). An optimal algorithm exists that com-
putes such a diagram in time 0(n log n) (see exercise 19.2).
The situation for the L., distance is very similar to the one just described for
the L1 distance. Its complexity in dimensions higher than 3 is easier to analyze,
however. The Loo distance of a point X = (X1, ... ,xd) in Ed from a point
M= (mI,. .. ,md) in Ed is given by
6""(X, M) = max [xi-mil.

i=l,...,d
The points at a distance r from M are thus on a cube centered at M whose facets
are parallel to the coordinate axes, and whose side is 2r.
The Voronoi diagram of M for the Loo distance is denoted by VorL. (M). An
instance is shown in figure 18.9.
The cells of this diagram can be obtained by projecting onto the hyperplane
Xd+1 = 0 in Ed+1 the cells on the lower envelope of the n pyramids Qi of equation
Xd+I = bo (XI Mi) -
The facets of the Q1 's form a collection of d-pyramids. The faces on the lower
envelope of these pyramids form a refinement of the faces on the lower envelope
of the Qi's. Hence, the vertical projections onto the hyperplane Xd+1 = 0 of the
faces on the lower envelope of these pyramids form a refinement of the faces of
the diagram VorL. (M). The complexity of the Voronoi diagram VorL", (M) is
thus bounded by the complexity of a lower envelope of n simplices in Ed+1:
IVorL.(M)| = 0(nda(n)).
This bound is almost tight for certain sets of points that are not in general
position (see exercise 18.9). If the points are in so-called L,,-general position,
then it is possible to show that the complexity of Voronoi diagrams for the L,,
metric is the same as that for Euclidean Voronoi diagrams, namely O(n [d/21) (see
exercise 18.10). We show this for the case d = 2. When d = 2, VorL. (M) can
18.5. Voronoi diagrams in hyperbolic spaces 449
Figure 18.9. Diagram for the L. metric.
be identified with the diagram VorLI (M) studied previously by simply rotating
the coordinate system by an angle of '. The points are in LOO-general position,
if no two points are connected by a line parallel to the coordinate axes, and no
four points belong to a common cube whose facets are parallel to the coordinate
axes. If so, then the complexity of VorL. (M) is O(n) in dimension 2, and the
algorithm described in subsection 16.3.3 that computes the lower envelope of
triangles can be used to compute this diagram in time O(nlog2 n) (see exercise
19.2 for a better algorithm).
The distances considered here, L1 and L., are particular cases of polyhedral
distances, so-called because their unit ball is a polytope. Voronoi diagrams for
polyhedral distances are studied in exercise 19.3.
18.5 Voronoi diagrams in hyperbolic spaces

18.5.1 Pencils of spheres
A pencil of spheres in Ed is a set S of spheres that are affine combinations of two

given spheres Ej and S2:
.F={E :3A E R, VX E Ed, E(X) = AE(X) + (1 -A)2(X)

If we apply to spheres the mapping ¢> introduced in section 17.2, we map the
spheres in Ed to points in Ed+l. From the results of section 17.2, it follows that
the image under 4 of a pencil F is the line 0(F) in Ed+l that connects the points
O(El) and O(E2)-
We may distinguish between four kinds of pencils, according to whether the
line that is the image under 4 of the pencil intersects the paraboloid in one point
(transversally), in two points, is tangent to P, or does not intersect P (see figure
18.10).
* If 4)(F) intersects P transversally in only one point, then F contains a single

sphere of zero radius, and +(F) is a pencil of concentric spheres.
* If the line 0(F) intersects P in two points, F contains two spheres of radius
zero, called the limit points of the pencil.
* If the line O(F) is tangent to P, then F may be considered as a pencil

whose two limit points are identical, or as a pencil supported by a sphere
that reduces to a point. Such a pencil is called a tangent pencil.
* If the line 0(F) does not intersect P, there exists a family of hyperplanes
tangent to P that contain +(F). Let O(Ey) be the set of points of P at
which these hyperplanes are tangent to P. Then O(.F) is the image under
4 of the set Er of points that belong to all the spheres in the pencil F.
Coming back to the definition of a pencil, we have E(X) = 0 for all values of
A, and this implies that E1 (X) = E2 (X) = 0 and that EJF can be identified
with the (d - 1)-sphere El nF2. All the d-spheres in the pencil F intersect
along the (d - 1)-sphere obtained as the intersection of any two spheres in
the pencil. For this reason, E.F is called the supporting sphere of the pencil.
The very definition of a pencil of spheres implies that any point in the radical
hyperplane H1 2 of two spheres El and E 2 in the pencil has same power with
respect to any sphere in the pencil. We may therefore define the radicalhyperplane
of a pencil of spheres as the radical hyperplane of any two spheres in the pencil.
The radical hyperplane of a pencil supported by a sphere is the affine hull of the
supporting sphere. A concentric pencil has no radical hyperplane. The radical
hyperplane of a pencil with limit points is the perpendicular bisector of these two
points. The radical hyperplane of a tangent pencil is the hyperplane tangent to
all the spheres in the pencil.
18.5.2 Voronoi diagrams in hyperbolic spaces

The Poincar6 model of the hyperbolic space of dimension d is the half-space
Hd = {X E Ed: Xd > 01. We will not define the hyperbolic distance precisely.
Concentric pencil Pencil with two limit points
Tangent pencil Pencil with supporting sphere
Figure 18.10. The four kinds of pencils.
LI..
-U
Figure 18.11. B is closer to A for the hyperbolic distance than C is.
The interested reader will find a more precise account in the classical references
on the topic ([22] for instance). To define the hyperbolic diagram, it suffices to
decide, given three points A, B, and C in Ed, whether B or C is closer to A.
For this, we consider the pencil .FA of spheres with limit points A and A', where
A' denotes the symmetric of A with respect to the hyperplane Ho of equation
Ho
0
Figure 18.12. A hyperbolic Voronoi diagram in the Poincar6 half-space.
Xd = 0. (Note that Ho is the radical hyperplane of .FA.) We say that B is closer

to A for the hyperbolic distance if the sphere in FA that passes through B has a
smaller radius than the sphere in .FA that passes though C (see figure 18.11).'
Given a set M = {M, ... , Mn} of n points in the Poincar6 half-space Hd, there
corresponds a region Vh(Mi) in HEd to each point Mi in M. This region consists
of the points in d that are closer to Mi than to any other point in M:
E
Vh(Mi) = {X E dbh(X, Mi) < bh(X, Mj) for any j = i}
The Voronoi diagram for the hyperbolic distance of M, also called the hyper-
bolic diagram of M, is the subdivision of the Poincar6 half-space induced by
the equivalence relation shared by the points that have the same nearest neigh-
bors for the hyperbolic distance. The faces of the diagram are the closures of
the equivalence classes. The Vh(Mi)'s form the cells of the diagram (see figure
18.12).
Vh(Mi) is the set of points X E d that have Mi as a nearest neighbor. Since
E
the locus of points in at a given hyperbolic distance from a given point A E d

Ed
is a sphere of the pencil FA, it follows that, for any point X in Vh(Mi), the interior
of the sphere in the pencil Fx that passes through Mi contains no point of M.
We can also embed into Ed+l by identifying it with the half-hyperplane
Ed
Xd+1 = 0, Xd > 0. The hyperplane Ho is therefore identified with the subspace

{Xd+1 = Xd = 0}. The pencil Fx is mapped by 0 into a line in Ed+l parallel to
the xd-axis. Indeed, if X' is the symmetric of X with respect to Ho, the pencil Fx
has limit points at X and X' that are mapped by 0 to A(X) and to O(X'), and
both these images are symmetric with respect to the hyperplane Xd = 0 in Ed~l.
This implies that a point X belongs to Vh(Mi) if and only if the ray parallel to the
'It is tempting to define the hyperbolic distance from A to a point B as the radius of the
sphere in .FA that passes through B. This "distance" is not symmetric, however, and is not the
true hyperbolic distance defined for instance in [22I. Nevertheless, in what follows, taking the
pseudo-distance to be this radius or indeed any other increasing function of this radius leads to
the same diagram.
Xd+ 1 UP
0(-Fx)
Figure 18.13. X belongs to Vh(A) (d-1).
Xdaxis in Ed+1 originating at ¢7(X) (which is entirely contained in the paraboloid

P) and directed towards Xd > 0 intersects the hyperplane O(Mi)* polar to 4,(Mi)
before any of the other polar hyperplanes O(Mj)* (see figure 18.13).
This observation has important consequences:
1. The bisecting surface of two points for the hyperbolic distance is a half-
sphere: indeed, a point X is equidistant from A and B if and only if Fx contains
a sphere that passes through A and B, that is, if and only if O(Tx) inter-
sects +(A)* n ¢(B)*. In other words, ¢(X) belongs to F, the projection of
¢(A)* n ¢(B)* parallel to the Xd-axis onto the paraboloid P (more exactly, the
half of the paraboloid that is in the half-space Xd > 0). But F is the intersection
of P with a hyperplane H in Ed+l parallel to the Xd-axis. Its vertical projection
onto Xd+1 = 0 is a sphere EAB (lemma 17.2.2), centered on Ho by symmetry.
Moreover, EAB belongs to the pencil with limit points A and B. Indeed, the
spheres in this pencil are mapped by 4 to the points
7(E) = A>¢(A) + (1 - A)4(B).
The corresponding polar hyperplanes have equations
¢7(S)* = >¢>(A)* + (1 - )(B)*
and they are all the hyperplanes that contain ¢(A)* n ¢(B)*. H is thus a hyper-
plane polar to a sphere in the pencil FAB that has two limit points A and B.
But H is the hyperplane polar to C(EAB) (see lemma 17.2.2). As a result, EAB
belongs to -FAB. Finally, EAB is the unique sphere in FAB that is centered on
Ho.
2. A point X is equidistant from d + 1 points AO, .. ., Ad if and only if 0(X)
is the projection of nl=L0 O(Ai)* parallel to the Xd-axis, onto the half-paraboloid.
The point at equal hyperbolic distance from d + 1 points is the limit point of
the pencil that contains the sphere circumscribed to the d + 1 points of radical
hyperplane Ho.
3. The hyperbolic Voronoi diagram can be obtained by projecting the poly-
tope V(M) = nl_ O(Ai)*+ parallel to the xd-axis onto the half-paraboloid, then
projecting the result vertically onto the hyperplane Xd+1 = 0. Note that the
projection parallel to the Xd-axis does not map all the points of V(M) onto the
half-paraboloid. This double projection establishes an injective correspondence
between the Euclidean and the hyperbolic Voronoi diagrams of M. More di-
rectly, these two projections can be avoided by performing the single following
transformation. Replace the planar (d - 1)-faces of the Euclidean diagram that
are (at least partly) contained in the half-space Xd > 0, by the corresponding
portions of spheres (hyperbolic bisectors limited to Xd > 0); a k-face (k < d - 1)
of the Euclidean diagram is the intersection of d - k + 1 planar (d - 1)-faces, and
is replaced by the portion of surface that is the intersection of the d - k + 1 corre-
sponding spherical faces. From the injective correspondence between Euclidean
and hyperbolic diagrams, we deduce the following theorem.
Theorem 18.5.1 The complexity of the hyperbolic Voronoi diagram of n points

in the hyperbolic Poincar6 half-space Hd is E3(nld/ 2 1). Such a diagram can be
computed in time E3(n log n + n[d/ 21).
18.6 Exercises
Exercise 18.1 (Greatest empty rectangle) Let X and A be two points in E2 . The
quadratic distance 6Q(X, A) is defined as
6Q(X,A)=(X-A)A(X-A) with A= ( I )
Show that 6Q(X, A) is the area of the rectangle whose sides are parallel to the coordinate
axes and of which A and X are two opposite vertices. Given a set S of points in the
plane, show that its diagram for this quadratic distance function can be used to compute
the rectangle of greatest area whose sides are parallel to the coordinate axes, whose sides
each contain at least one point of S, and whose interior does not contain any point of S.
Hint: To find a greatest empty rectangle, use a divide-and-conquer algorithm. The

merge step consists in finding the greatest empty rectangle intersected by the separating
18.6. Exercises 455
line. The greatest empty rectangle for which three points of contact lie on one side of
the separating line (and the fourth on the other side) can be found easily. Two points of
contact on one side of the separating line define a corner of the rectangle, and the corners
of empty rectangles are the so called maxima and can be found in O(n log n) time. The
greatest empty rectangle with two points of contact on either side of the separating line
is defined by an opposite pair (A, B) of maxima such that the segment connecting them
is an edge of the affine diagram defined for the generalized quadratic distance 6Q (A, B).
The complexity of the merge step is 0(n log n), hence the total algorithm runs in time
O(n log2 n).
Exercise 18.2 (Lower envelope of cones) Show that the lower envelope of n vertical
cones of revolution in Ed has complexity O(nLd/2j+1) and can be computed in time
O(nLd/ 2 J+1). If the vertices of the cones are all contained in a given horizontal hyperplane,
and if their angles are all identical, then the complexity of the lower envelope drops to
0(n L2 ) and it can be computed in time O(n log n +n[ ).
Exercise 18.3 (Spheres and disks) According to the general definition of Voronoi
diagrams, we may define the Voronoi diagram of a set of disks Dl,. .,Dn as usual,
where the distance of a point X from a disk Di centered at Ci and of radius ri is defined
by
ds(X, Di) = max(O, IIXCiI - ri).
Show that the Voronoi diagram of n disks in Ed, where d > 3, has complexity O(nLd/2J+1)
and that it can be computed in time O(nLd/2i+1). If d = 2, show that these bounds are
0(n) and 0(n log n) respectively.
Hint: To each Ci, give a weight ri and compute the diagram of the disks knowing the
additive diagram of their centers. In the discussion of subsection 18.3.1, the cone Ci,
i = 1, . .. , n must be replaced by the same cone truncated by the halfspace Xd+ 1 > 0.
Exercise 18.4 (Diagrams with multiplicative weights) Show that Q(nLd/2J+1) is

a lower bound on the complexity of the Voronoi diagram of n points in Ed with multi-
plicative weights.
Hint: Generalize the example of figure 18.5.
Exercise 18.5 (Regular complex) Let C be a d-complex in Ed. We say that C is

regularif it can be obtained as the vertical projection of a polytope in Ed+l. Show that
C is regular is and only if it is a power diagram. Show that any simple complex (meaning
that its cells all consist of simple polytopes) is regular (the best-known examples are
arrangements of hyperplanes in general position). Devise an algorithm that determines
whether a complex is regular and, if so, computes the corresponding polyhedron.
Hint: Use theorem 18.2.1. For hyperplane arrangements, the connection with zonotopes
is particularly helpful (see exercise 14.8).
Exercise 18.6 (The inverse problem) Show that it is possible to determine whether
a complex is a Voronoi diagram and, if so, to compute the corresponding sites in time
linear in the total complexity of its cells.
Exercise 18.7 (Spider webs) By a spider web, we mean the 1-skeleton of a 2-complex
that covers E2 . Show that if the spider web is the skeleton of a power diagram, then we
can assign a tension to each edge such that each vertex is in an equilibrium state.
Hint: For the tension of an edge, take the length of the dual edge. An edge and its
dual edge are perpendicular, and the dual edges of the edges incident to a vertex S form
a cycle that we orient counter-clockwise. At a vertex S, the sum of the tensions equals
the sum of the vectors of the dual edges, so that the total tension vanishes at the vertices.
Exercise 18.8 (Cubes and co-cubes) Show that in E3 , several homothetic cubes or
co-cubes may pass through four points even though these points are in LOO-general po-
sition.
Exercise 18.9 (Degenerate positions for L1 and Loo distances) Show that in E2 ,
the Voronoi diagram for the L, metric of points that are along one of the main bisectors
is quadratic. Show that if the bisector of two points on a line parallel to one of the main
bisectors is redefined as the Euclidean perpendicular bisector, then the complexity of
the diagram becomes linear, and a cell is formed by the set of points that share exactly
one common nearest neighbor for the L, distance (but do not necessarily have the same
subset of nearest neighbors). Generalize the example above to show that Q(nd) is a lower
bound on the complexity of a Voronoi diagram of n points in Ed for the L1 metric. Also
give similar results for the Loo metric.
Exercise 18.10 (Complexity of VorL-) Show that the complexity of a Voronoi dia-
gram for the Loo metric of a set M of n points in Ed in Lc,,-general position is o(nrd/21 ).
Hint: It suffices to bound the number of so-called maximal placements of a maximal

cube whose facets are perpendicular to the coordinate axes, and whose interior contains
no point of M. A contact is a pair formed by a facet of such a cube and by a point in M. A
placement realizes a contact of multiplicity k at a point if this point belongs to k facets of
the corresponding cube. If k = 1, the contact is said to be simple. For a given maximal
placement, the sum of the multiplicities of the points of contact is d + 1. First show
that any maximal placement realizes two contacts, called parallel contacts, whose facets
are parallel. We say that a maximal placement is reducible if at least one of its parallel
contacts is simple and if the other does not have multiplicity d. Show then that it suffices
to bound the number of irreducible maximal placements. For this, charge a reducible
placement to an irreducible one by applying the following procedure as many times as
needed: Scale up the cube by a homothety centered at one of its vertices that lies on a
facet involved in some parallel contact. In this way we obtain a smaller cube contained
in the preceding one but whose multiplicity is increased for at least one of the contacts.
Show that an irreducible placement is charged by at most 0(1) reducible placements.
Finally show that the number of irreducible placements is 0(nrd/21). For this, notice
that the centers of such placements belong to some affine subspace of dimension at most
d - 3. In this subspace, the centers of maximal placements correspond to the vertices of
a union of n cubes of same size, so we may use the result of exercise 4.8.
Exercise 18.11 (Simplicial distance) Let S be a (d + 1)-simplex in Ed that contains

the origin 0. We denote by AS the image of this polytope under the homothety centered
at 0 and of ratio A. The simplicial distance 6s(X, A) from point X to point A is defined
as the smallest real A > 0 such that X - A belongs to AS. Show that the complexity of
a Voronoi diagram for a simplicial distance of a set M of n points in Ed is O(n rd/21).
Hint: We define a reducible placement of S as in exercise 18.10: it is a placement that

has several simple contacts. The number of irreducible placements is 0(n Fd/21) and we
can also show that the same bound holds for the reducible placements.
Exercise 18.12 (Hyperbolic bisector) Show that the equation of the hyperbolic bi-
sector EAB of two points A = (a,.. ., ad) and B = (b,... .,bd) is
(ad -bd)X - X + 2(bdA-adB) * X-bdA A + adB - B = 0.
Exercise 18.13 (The Poincare disk) Rather than using the Poincar6 half-space H2
as a model of the hyperbolic space, we introduce the Poincar6 disk 1Dwhich can be derived
from H2 by a homographic transformation. More precisely, if the Poincar6 half-space is
identified with the complex half-plane {z E C, Imz > 0}, the homographic map defined
by
h(z) =
Z + i
2
is a bijection from H11into D. Show that the edge that joins two points remains a circle
centered on the boundary of 1D, and also that the points at equal distance from A are
on a circle that belongs to the pencil that has A as a limit point and that contains the
boundary of 1D. From this, explain how to compute the Voronoi diagram of a set of points
in H2 .
Exercise 18.14 (Dual of a hyperbolic diagram) Show that we may dualize the hy-
perbolic Voronoi diagram of a set of points M in Hd by projecting the convex hull of
O(M) parallel to the Xd-axis onto the half-paraboloid, and then projecting the result of
this first projection onto the hyperplane Xd+1 = 0. Show that this dual is in bijection
with a sub-complex of the Delaunay complex.

Power diagrams were studied by Aurenhammer [14] and by Imai, Iri, and Murota [129].
Affine diagrams are defined in [17] by Aurenhammer and Imai, who also show their con-
nection with power diagrams and diagrams with additive and multiplicative weights. So-
lutions to exercises 18.5 and 18.6 are due to Aurenhammer. The solution to exercise 18.1
is adapted from that given by Chazelle, Drysdale, and Lee [47] and Aurenhammer. Spi-
der webs were already analyzed by Maxwell in the nineteenth century. Recent references
can be found in the article by Ash, Bolker, Crapo, and Whiteley [12].
Diagrams for the L, and Lo, metrics in the plane were studied by Lee and Wong [149]
then by Lee and Drysdale [147]. The generalization to general convex distances is tackled
by Chew and Drysdale [61]. Voronoi diagrams for the L1 and Loo metrics in dimensions
3 and higher (see exercise 18.10) and also simplicial distances (see exercise 18.11) are
treated by Boissonnat, Sharir, Tagansky, and Yvinec [34]. In the plane, Klein proposes
a notion of abstract Voronoi diagram (139] and Klein, Mehlhorn, and Meiser describe a
randomized algorithm that computes such diagrams [141].
Diagrams for the hyperbolic distance axe studied by Boissonnat, C6rezo, Devillers, and
Teillaud [26], who present an application to shape reconstruction from plane sections.
Chapter 19
Diagrams in the plane
In the two preceding chapters, we have shown how to compute several types
of Voronoi diagrams in Ed by computing the upper envelope of hyperplanes in
Ed+l or Ed+ 2 . This often leads to optimal algorithms: this is notably true for
diagrams of points under a general quadratic distance and for power diagrams of
spheres. In contrast, we have seen that for diagrams with additive weights, such
an approach does not lead to optimal algorithms in dimension 2.
Section 19.1 describes an algorithm that computes the Voronoi diagram of a
set of points in the plane. This algorithm is remarkable for several features: it
uses the sweep method, it is simple and optimal, and it can be generalized in a
number of ways. For instance, it can be adapted to compute the Voronoi diagram
of a set of segments or to use other metrics such as the L1 or Loo distances.
Voronoi diagrams of line segments have important applications, such as the
motion planning of a disk (see subsection 19.2.5). In dimension 2, they are
well understood and we present, in addition to the generalization of the sweep
algorithm, a randomized algorithm and its accelerated version when the set of
segments is connected, and the segments intersect only at common vertices.
Section 19.3 studies an instance of the problem when the points belong to two
planes in E3 . This is a particular instance of three-dimensional diagram for which
an algorithm is presented that is output-sensitive and optimal.
19.1 A sweep algorithm

In this section, we present a sweep algorithm that computes the Voronoi diagram
of a set M = {M1 ,. . ., Mj} of n points in E2 . To simplify the presentation, the
points are supposed to be in L2 -general position, meaning that no four points are
co-circular.
As shown in subsection 18.3.1, if E2 is identified with the xy-plane z = 0 in
460 Chapter 19. Diagrams in the plane
y
XI
Figure 19.1. The projection used in the sweep algorithm.
E3 , we may place cones Ci on top of each Mi: Ci is the upward vertical cone of
revolution of angle 7 that has apex Mi; its equation is
z = |IXMill.
Then the Voronoi diagram of M is the projection onto the plane z = 0 of the
lower envelope of the cones Ci, i = 1, . . . , n.
The algorithm computes this lower envelope. Rather than using a vertical
projection, however, we project parallel to a line that generates the cones. More
precisely, the plane is swept by a line parallel to the x-axis that moves along the
increasing y-axis, and the direction of the projection onto this plane is given by
the vector (0,1, -1).
In this way, we map a point X = (x, y) to another point
o(X) = (xy+ min (IIXMiI1)).

MiEM
This point is obtained by first lifting X onto the lower envelope of the Ci's and
then projecting the result onto the plane z = 0 parallel to the direction (0, 1, -1).
The map a is depicted in figure 19.1. As usual, V(Mi) stands for the cell that
corresponds to Mi in the Voronoi diagram. Because the lower envelope is con-
tinuous, the map a is also continuous. Moreover, its restriction to a line D that
19.1. A sweep algorithm 461
is parallel to the y-axis and that does not contain a point in M is injective. The
images of the points on D are also on D, and if X1 = (x, yi) and X2 = (x, y2)
are two points on D with yi < Y2, the triangle inequality shows that
FVi, IIXMiII < IIX1X 2 II+ IIX2 MivII
This shows that the n maps that return the value y + IIXMiII, given the ordinate
y of a point X on D, are continuous functions that increase with y. Therefore the
minimum of these functions is also an increasing function, and so a is injective on
D. Now if D contains a point Mi, the inequality above still holds if the ordinate
of X2 is greater than that of Mi and so a is still injective. If both X1 and X2
have smaller ordinates than Mi, then
IIX1MiII = IIX1X 2 II + IIX2 MiII.

Any point X e V(Mi) that has the same abscissa as, and a smaller ordinate than
Mi is mapped by a onto Mi, so Mi is invariant. From the preceding discussion,
it follows that the restriction of a to the edges of the Voronoi diagram of M is
injective, since any ray cast from Mi intersects the boundary of the convex cell
V(Mi) at only one point.
The map a deforms the Voronoi diagram Vor(M) into a new diagram Vor'(M).
Since a is injective on the edges of Vor(M) and is continuous, the cell V(Mi)
is mapped by a onto a simply connected cell V'(Mi) which we call a cell of
Vor'(M). The edge V(Mi, Mj) = V(Mi) n V(Mj) in Vor(M) that is supported
by the perpendicular bisector of two points Mi and Mj is mapped by a onto a
curved arc V'(Mi, Mj) which we call an edge of Vor'(M). Finally, a maps the
vertices of Vor(M) onto the endpoints of the edges of Vor'(M): we call these
points the vertices of Vor'(M).
The curved arc V'(Mi, Mj) is in fact contained in a branch of the hyperbola,
that is, the projection of the intersection of the cones Ci and Cj. This branch
passes through Mi or Mj, whichever has the greatest ordinate, and this point is
the point of smallest ordinate on this branch. The hyperbola degenerates into a
line parallel to the y-axis if Mi and Mj have identical ordinates. The cell V'(Mi)
is delimited by hyperbolic arcs and the point with smallest ordinate in V'(Mi) is
Mi (see figure 19.2). This property is crucial for the sweep algorithm. Observe
that all the hyperbolic arcs that form the boundary of V'(Mi) are y-monotone,
except for the one that contains Mi. This one can be split into two y-monotone
sub-arcs that have Mi as the endpoint of smallest ordinate.
The sweep line A is parallel to the x-axis and moves towards increasing ordi-
nates. To simplify the presentation, we assume that no two points in M have
identical ordinates. Like any sweep algorithm, this algorithm maintains two struc-
tures. The first structure is a dictionary D of the edges of Vor'(M) that intersect
A, in increasing order of their intersections along A, and the intervals induced
Figure 19.2. Vor(M) and Vor'(M).
by these points on A. Each interval is the intersection of A with a cell of the

deformed Voronoi diagram. If this corresponds to the site Mi, the interval is
labeled by i. As we observed above, the edges of Vor'(M) that contain a site
are not y-monotone, and they intersect A in two points. This slight problem is
easily remedied by splitting the edge at Mi into two arcs as already mentioned.
The second data structure maintained by the algorithm is a priority queue Q
that represents the events (sorted by increasing ordinate) for which A sweeps over
a vertex or over a site, at which point the first data structure must be updated.
At the start of the algorithm, A is the line parallel to the x-axis that passes
through the point of smallest ordinate. Only a single unbounded region exists
and the dictionary D stores a single interval. As A moves towards increasing
ordinates, the structure D must be updated for two kinds of events. In a site
event, the sweep line sweeps over a site, say Mi. A new region and its two
bounding hyperbolic arcs appear in the diagram, giving rise to a new interval
with two new endpoints on A. More precisely, if Mi belongs to an interval labeled
j, the hyperbola that supports the two arcs to be added to the diagram is the
image under ae of the perpendicular bisector of Mi and Mj. The interval labeled
j is replaced in the structure D by three intervals labeled j, i, and j in that order.
In the second kind of event, a circle event, A sweeps over the intersection point
of two arcs. This situation corresponds to a vanishing interval on A whose two
endpoints are the intersections of A and the two meeting hyperbolic arcs. Let j
be the label of that interval, and i and k be the labels of the adjacent intervals
in D. These two arcs are the images under a of two incident edges that bound
V(MiMj) and V(MjMk). Their meeting point is the image of a vertex of the
Voronoi diagram, the center of the circle passing through Mi, Mj, and Mk (hence
the name of this kind of event). The two arcs are replaced by the image of the
edge of V(MiMk) that is incident to this vertex.
19.1. A sweep algorithm 463
As happens when computing the arrangement of a set of line segments using the
sweep algorithm described in section 3.2, each time two arcs become consecutive
in D, we test whether they intersect beyond A and, if so, we insert this intersection
point into Q. When deleting an arc, we must also remove the two corresponding
entries defined by this arc in Q. In fact, for each arc Q must contain only an
entry that stores the intersection point of smallest ordinate defined by this arc
(beyond this point, the arc cannot exist), and for each arc in D, we maintain a
pointer to this entry. Each time an arc is considered, its pointer is updated, and
the entry in Q is removed when the arc disappears.
We must point out that computing the hyperbolas is not needed. In fact,
hyperbolas are the images under a of the perpendicular bisectors of two of the
sites, and their intersections are the images of the intersections of the two cor-
responding perpendicular bisectors. Location in D can be performed by using
the perpendicular bisectors rather than the hyperbolas. Location in Q can be
carried out by computing the ordinates of the images of the intersection of two
perpendicular bisectors. The Voronoi diagram is computed directly during the
sweep.
The complexity of the algorithm can be estimated very simply. The sweep line
stops over sites and vertices of Vor'(M) so the number of events processed by the
sweep algorithm is no more than the size of Vor'(M), namely O(n). During each
event, only 0(1) locations, insertions, or deletions are performed in D and 0(1)
events are added to or removed from Q. The sizes of D and Q are thus O(n) at
any event throughout the algorithm. That the size of D is O(n) can also be seen
by noticing that A intersects any edge of Vor'(M) at most twice. Each update
operation (insertion, deletion, or location) can therefore be carried out in time
0(log n), for a grand total of 0(n log n) operations. This is optimal as shown by
corollary 17.3.2, and this proves the following theorem.
Theorem 19.1.1 The Voronoi diagram of n points in the plane can be com-
puted using a sweep algorithm in time 0 (n log n), using storage 0(n), and this is
optimal.
Application to diagrams with additive weights

Let Ml, . . . , Mn be n points in the plane, endowed with the weights r, . .. , rn.
The distance with additive weights (additive distance for short) from a point X
to Mi is defined as
6+(X,Mi) = IIXMi -ri.
The additive diagram is defined as the Voronoi diagram for this additive distance.
As shown in subsection 18.3.1, the additive diagram of a set of n points in the
plane is the projection onto this plane of vertical cones with aperture angle 74r.
Its complexity is 0(n). The aperture angles of the cones being identical, the
sweep algorithm performs in a way that is strictly analogous to the Euclidean
case. Only a few differences deserve to be pointed out.
First of all, the transformation a introduced above (see figure 19.1) becomes
a(X) = (xy) = (xy+ mi (JJXMijj -ri))
and the Mi's are no longer invariant. Secondly, some sites may have an empty
corresponding region. This can be detected during the sweep: when a new site
Mi is encountered, it is contained in the region of a point Mi. of weight ri. in the
additive diagram, and its region is non-empty if and only if its weight ri satisfies
AMMe. || - ri. > -ri.
The remaining details of the algorithm are strictly analogous to those in the
Euclidean case, and hence:
Theorem 19.1.2 The Voronoi diagram of n points in the plane for the additive
distance can be computed in time 0(nlogn) by a sweep algorithm, and this is
optimal.
19.2 Voronoi diagram of a set of line segments

19.2.1 Definition and basic properties
Let S = {S1,...,Sn} be a set of n disjoint line segments in the plane. The
distance of a point A from a segment Si, denoted by 6(A, Si), is defined as the
smallest distance between A and any point in Si. The segments are said to lie in
L2 -general position if no point is equidistant from more than three segments in S.
Let us consider the equivalence relation R, shared by two points if and only
if they have identical subsets of nearest neighbors in S. The equivalence classes
of R subdivide E2 into (open) regions whose topological closures form the faces
of the Voronoi diagram of S. The interior of a cell in this diagram is formed
by the points closer to some segment Si in S than to the others, and this cell is
denoted by V(Si) (i = 1, . . . , n). The edges are the regions formed by the points
equidistant from two segments in S and closer to these segments than to any
other segment in S. The vertices are the points that are equidistant from at least
three segments in S and closer to these segments than to any other. The cells,
edges, and vertices and their incidence relationships form the Voronoi diagram
of the set S of segments, denoted by Vor(S) (see figure 19.3).
The two following lemmas characterize the cells V(Si) in such a Voronoi dia-
gram.
19.2. Voronoi diagram of a set of line segments 465
Figure 19.3. The Voronoi diagram of a set of line segments.
Lemma 19.2.1 Let V(Sj) be the cell of Si in the diagram, X a point in V(Sj),
and X' its closest point in Si. The ray originating at X' that contains X is
either entirely contained in V(Sj) or intersects V(Sj) along a line segment whose
endpoints are X' and the unique intersection point of the ray with the boundary
of V(SO).
Proof. With the notation of the lemma, any point Y on the segment XX' is
closer to X' than to any other point in S, so that Y belongs to V(S1 ). The
segment XX' is therefore contained in V(S2 ), proving that V(Sj) is star-shaped.
The ray X'X thus intersects V(S1 ) along a connected subset of the ray that
contains X', which can only be the ray itself or a line segment whose endpoints
are X' and a point Z on the boundary of V(Sj). It remains to see that in the
latter case, this point Z is uniquely defined as the intersection of the ray with
the boundary of V(Sj) (this intersection could contain several points, possibly
infinitely many).
For this we show that the relative interior of X'Z does not intersect the bound-
ary of V(S1 ). Towards a contradiction, assume that there is a point Y on the
relative interior of the segment X'Z that is on the boundary of V(Sj). This im-
plies that Y is equidistant from Si and another segment Sj in S. Let us denote by
Y' the point on Sj that is the closest to Y. The disk centered at Z of radius ZX'
contains the disk centered at Y of radius YX', and only X' belongs to both their
boundaries. So the first disk must contain Y' in its interior. Therefore Z is closer
to Y' and hence Sj than to Si, and it does not belong to V(Si), a contradiction.
It follows that if the ray X'X intersects V(Si) along a line segment, only the
endpoint other than X' of this segment belongs to the boundary of V(Si), whereas
if the entire ray is contained in V(Si), then no point of this ray belongs to the
boundary of V(Si).
Lemma 19.2.2 The interior of any cell of the Voronoi diagram of S is simply
connected.
Proof. Let us first see that V(Si) is connected. Indeed, considering two points
X and Y in V(Si), denote by X' and Y' their closest point in Si. It follows from
the previous lemma that the segments XX' and YY' are contained in V(Si).
Obviously, Si is contained in V(Si), and X'Y' is also contained in V(Si) since it
is a subset of Si. Thus X and Y can be connected by the polygonal line XX'Y'Y
which is entirely contained in V(Si). Note that its interior is also connected since
the segments are disjoint and thus Si is contained in the interior of V(Si).
Let us now show that the interior of V(Si) is simply connected. Assuming the
contrary, there exists a point X that does not belong to V(Si) but is contained
in the interior of a closed Jordan curve r that is entirely contained in V(Si). If
X' is the closest point to X on Si, the ray originating at X' that contains X
intersects r in at least two points (see exercise 11.1), one of which is further from
X' than from X; call this point I. The interior of the disk centered at I and of
radius IIX' I does not intersect any segment in S, hence neither does the interior
of the disk centered at X of radius IIXX'II. So X is closer to Si than to any other
segment of S, hence belongs to the interior of V(Si), a contradiction. 0
The edges of the diagram are formed by the points that are equidistant from
two segments and closer to these segments than to the others. An edge is thus
contained in the bisector of two segments, which is the set of points at equal
distance from two segments. In general, such a bisector can be split into several
components: line segments, contained in the perpendicular bisector of two end-
points or in the bisector of the lines supporting the two segments, and parabolic
arcs formed by the points at equal distance from an endpoint of one segment and
the line supporting the other segment. The following lemmas explore the nature
of these bisectors more precisely.
Lemma 19.2.3 The bisector of two disjoint line segments is a simple curve that
disconnects the plane into two connected components and that can be split into at
most seven line segments and parabolic arcs.
Proof. Consider two disjoint line segments Si and S2 whose endpoints are Al,
B1 and A2 , B2 respectively. At least one of these segments (say Si) is contained
in one of the two half-spaces bounded by the line that supports the other (S2).
Each segment is identified with a flat rectangle oriented counter-clockwise, and
its four elements are conceptually separated: its endpoints Ai, Bi, and its two
oriented sides Ei = (Ai, Bi) and Fi = (Bi, Ai). Each arc of the bisector D1 2 is
the locus of points at equal distance from two elements, one of S1 and the other
of S2, and closer to these than to any other element. The arcs of the bisector D1 2
are labeled by the two corresponding elements.
When following the boundary of the region V(Si) counter-clockwise, the labels
of the arcs of D12 are enumerated in counter-clockwise order for those that belong
to Si, and in clockwise order for those that belong to Sj (i, j = 1, 2 and i $ j).
Indeed, consider two points X and Y in D1 2 that are labeled by two distinct
elements of Si, say LX and Ly. Let X' be the closest point to X in LX and let
Y' be the closest point to Y in Ly. It is easy to see that the relative interiors of
XX' and YY' cannot intersect. Thus the labels of Si are seen in counter-clockwise
order along the boundary of V(Si) oriented counter-clockwise.
Let us construct the following oriented graph G (see figures 19.4 and 19.5). A
node in the graph represents a pair formed by an element of S and an element
of S2. There is an arc in G between the nodes N = (L1 , L2 ) and N' = (Li, L 2 )
if and only if L'1 follows L1 in counter-clockwise order along Si. Similarly, there
is an arc in G between the nodes N = (L1 ,L 2 ) and N' = (L1 ,L') if and only
if L' follows L2 in clockwise order along S2. The graph G can be drawn on a
torus (the topological product of the boundaries of Si and S2). The bisector of
SI and S2 is represented by an oriented path in this graph. We distinguish two
cases according to whether the entire segment Si appears on the boundary of the
convex hull of SI and S2, which is the convex hull of A1 , A2 , B1 , and B2 . (Note
that S2 is always entirely contained in this boundary.)
Case 1. The boundary of the convex hull of S1 and S2 is the polygonal line A1 ,
B1 , B2 , A2 , in clockwise order (see figure 19.4). The bisector of SI and S2 has two
infinite branches supported by the perpendicular bisectors of A1 and A2 , and of
B1 and B2 . From the preceding discussion, it is clear that the bisector of SI and
S2 corresponds to an oriented path in the graph G that connects the two nodes
(A1 , A2 ) and (B 1 , B2 ). The number of arcs on this bisector equals the number of
vertices of such a path, namely five.
Case 2. Only one endpoint of S1 , say A1 , appears on the boundary of the convex
hull conv(Sl, S2 ), which is the polygonal line A1 , B2 , A2 in clockwise order (see
figure 19.5). The bisector of S and S2 has two infinite branches supported by
the perpendicular bisectors of A1 and A2 , and of A1 and B2 . The bisector of SI
and S2 corresponds to an oriented path in the graph G that connects the two
nodes (A 1 , A2 ) and (A 1 , B2 ). The number of arcs on this bisector is thus seven.
El
A2
16-
Al E1 B1 Fl Al
Figure 19.4. A bisector of sepgments with five arcs.
B2 E2 A2
-4
Al El B1 F1 Al
Figure 19.5. A bisector of segments with seven arcs.
Lemma 19.2.4 Let Si, S2 , and S3 be three disjoint segments. Any two of the
three bisectors of these segments intersect in at most two points.
Proof. Consider three segments Sl, S 2 , S3, and denote by V(Si), V(S 2 ), and
V(S 3 ) the three cells in the Voronoi diagram of the three segments. Let us assume
that two of the three bisectors, say D1 2 and D1 3 , have three points in common.
Then these points also belong to D2 3 : they are at the same distance from Sl and
S2 , and from Sl and S3, so they are also at the same distance from S2 and S3 .
This shows that the three bisectors have three points in common. Then V(Si) has
at least three vertices 1I, I2, and I3, say, in clockwise order along its boundary,
each belonging to the intersection of V(S 1 ), V(S 2 ), and V(S 3 ). Two successive
edges of V(Si) cannot belong to the same bisector, otherwise the cells V(S1 ),
V(S 2 ), and V(S 3 ) would all contain the endpoint common to both edges, and
since they cover the plane entirely, one would not be simply connected, violating
lemma 19.2.2. The edge E1 of V(S1 ) that precedes I1, and the edge E3 of V(Sl)
that connects 12 to I3, both belong to the same bisector, say D1 2 . Similarly, the
edge E2 of V(S 1 ) that connects I, to 12 and the edge E4 of V(Sl) that follows
I3 both belong to the same bisector D1 3 . The situation is shown in figure 19.6.
r2 r3
. 2 D 12 3
Now pick a point Ai on Ei (i = 1,...,4). Since the interiors of V(S 1 ), V(S 2 ),

and V(S 3 ) are simply connected, there exists a path r, connecting Al to A3 that
is entirely contained in the interior of V(S1 ) (except for its two endpoints Al
and A3 ). Similarly, there exists a path F2 connecting Al to A3 that is entirely
contained in the interior of V(S 2 ) (except for Al and A3 ), and there exists a
path r 3 connecting A2 to A4 that is entirely contained in the interior of V(S 3 )
(except for A2 and A4 ). Now the union of r 1 and F2 is a closed simple curve
r 12 which splits the plane into two connected components (by Jordan's theorem),
one containing A2 and the other containing A4 . Therefore r 3 must intersect F12,
which contradicts the fact that the cells have disjoint interiors. 0
Remark. The bisector of two segments may contain an arbitrarily high number
of edges of Vor(S).
Lemma 19.2.5 The Voronoi diagram of n segments has complexity O(n).
Proof. Let S be a set of n segments. Lemma 19.2.2 assures us that the diagram
has exactly n connected cells. Moreover, each vertex in the diagram has degree
three when the segments are in L2 -general position. The Voronoi diagram of S is
thus a planar map with n cells and vertices of degree three. Lemma 19.2.3 implies
that each edge in the diagram has at most seven arcs, so Euler's relation can be
used to show that the number of faces of this map is O(n). The complexity of the
diagram, defined equivalently (within constant factors) as the number of arcs, or
the number of faces, is thus O(n). It is easy to see that the L2-general position
assumption does not matter for the result. Indeed, the segments in S may be
slightly perturbed so as to achieve L2 -general position, and perturbed back into
their primary positions. During this second perturbation, no face is created; on
the contrary, some vertices are merged and the zero-length edges that join them
disappear. The complexity of the diagram only decreases from that of a diagram
of segments in L2-general position. O
19.2.2 A sweep algorithm

Consider n segments Si = A 1 Bi,... , Sn = AnBn that are disjoint and in L2 -
general position. This means that no point is at equal distance from four seg-
ments. By convention, Ai is the endpoint of Si that has the smallest abscissa.
If we don't care which endpoint Ai or Bi of Si we consider, we use the notation
Mi. We define and use a mapping a, that is analogous to the mapping a used
in section 19.1:
a (X)=x, y + min6 (X, Si))
Its geometric interpretation is identical to that of a, except that Ci (i = 1,... , n)

now denotes the union of the vertical cones Ci(X) of angle 7 whose vertices are
the points X on segment Si. An argument similar to that used in section 19.1
shows that the restriction of the application a, to the edges of the diagram is
continuous and injective.
This mapping deforms the Voronoi diagram Vor(S) into a new diagram Vor'(S)
whose faces are in one-to-one correspondence with the faces of Vor(S). The cell
V(Si) is transformed into a cell V'(Si) whose points all have a greater ordinate
than Ai. An arc of Vor(S), contained in the parabolic locus of the points at
equal distance from an endpoint Mi of Si and the line Lj that supports Sj, is
transformed by as into an arc contained in a parabola that is the projection of
the intersection of Ci and of the dihedron Dj (the union of the cones Ci(X) for
X E Lj). An arc of Vor(S) contained in the bisector of the lines supporting two
segments Si and Sj is mapped by a, onto a line segment contained in the projec-
tion of the intersection of the two dihedra Di and Dj. Finally, as is the case for
the Voronoi diagram of points, an edge of Vor(S) contained in the perpendicular
bisector of the endpoints Mi and Mj of two segments is mapped onto an arc
contained in the branch of the hyperbola that is the projection of the intersection
of the cones Ci and Cj. This branch contains the points Mi or Mj, whichever
has the greatest ordinate, and this point has the smallest ordinate among all the
points on this branch. The hyperbola degenerates into a line parallel to the y-axis
if Mi and Mj have the same ordinate y.
The sweep line A is parallel to the x-axis and moves towards increasing or-
dinates. To simplify the presentation, we assume that all the ordinates of the
endpoints are distinct. In particular, no segment is parallel to the x-axis. The
sweep uses data structures similar to those used for computing the diagram of
points. There are now three kinds of events. The first kind happens when A
sweeps over an endpoint of smallest ordinate Ai, and is analogous to a site event
for the sweep algorithm that computes the diagram of points. The only signifi-
cant difference arises because the bisector of two segments consists of several arcs.
These arcs are computed when the bisector is encountered for the first position
of the sweep line, and their endpoints correspond to as many events of the third
kind that must be inserted into the event priority queue. An event of the second
kind is an intersection point of two arcs of the transformed diagram. It can be
handled in very much the same way as a circle event for the sweep algorithm
that computes the diagram of a set of points. Processing the events of the third
kind is straightforward: we must simply change the description of the arc when
A sweeps over the endpoint of greatest ordinate of that arc, and also update the
event in the priority queue that corresponds to the disappearance of the edge
that contains that arc.
Theorem 19.2.6 The sweep algorithm described above computes the Voronoi
diagram of a set of n line segments in the plane in time O(nlogn), and this is
optimal.

This section presents an incremental algorithm that computes the Voronoi dia-
gram of a set of n line segments. This algorithm is an extension of the incremental
algorithm for points that is given in exercise 17.10. Before we describe the algo-
rithm, we must point out that any algorithm that computes the Voronoi diagram
of n points (and a fortiori segments) incrementally may require time Q(n 2 ). In-
deed, as is shown in figure 19.7, the Delaunay triangulation of n points on the
moment curve in E2 contains the triangles connecting the point of smallest ab-
scissa to the edges of the convex hull. Adding these points in order of decreasing
abscissae will create i - 2 new triangles at the i-th step, so Q(n 2 ) triangles must
be created overall in the worst case. If the points are inserted in random order,
however, the expected complexity of an incremental algorithm is O(nlogn) as
was shown in chapter 5.1
The algorithm we present now is an on-line incremental algorithm that uses an
influence graph (see section 5.3). The term "on-line" means that the algorithm
maintains a representation of the Voronoi diagram under insertions of arbitrary
segments. This algorithm is deterministic, and if the n segments are inserted
in random order, a randomized analysis shows that its expected complexity is
O(n log n). Again, we point out that the algorithm does not assume any geometric
distribution of the segments: the expectation is taken for a given set of n segments
and only the insertion order is random. Our only assumption is that the segments
are in L 2 -general position.
To fit the framework of chapter 4, we must redefine the problem in terms of
objects, regions, and conflicts.
The objects are naturally the line segments. The regions are the edges in the
Voronoi diagrams of any subset S' of S. A region E is defined by a subset S' of
'This is the case for the randomized algorithm that computes the convex hull in E3 of the
points lifted on the paraboloid, but is also the case for the algorithm described in this section.
472Chpte 1. iagam i th pan
x
0
Figure 19.7. A difficult instance for the incremental algorithm.
Si
S4
53
Figure 19.8. A region E, in bold, and its domain of influence DE, shaded.
segments if and only if E is an edge in the Voronoi diagram of S' but not in the
diagram of any subset of S'. A segment in S' is called a determinant of E. A
region is determined by at most four segments: the two line segments Si and S2
whose bisector contains E, and occasionally one or two segments S3 and S4 that
determine the endpoints of E.
To a region E corresponds a domain of influence DE that is the union of the
open disks centered on E and not intersecting the segments defining E (see figure
19.8). An object S and a region E conflict if S and the interior of DE have a non-
empty intersection. The edges of the Voronoi diagram are precisely the regions
defined and without conflict over S, that is, the regions determined by segments
of S that do not conflict with any segment in S. The edges determined by a
segment S that are without conflict over S are the edges of the Voronoi cell V(S)
and all the edges incident to a vertex of V(S).
The algorithm proceeds by inserting the segments one by one. At an incre-
mental step, the Voronoi diagram of the current subset Sc of already inserted
segments is stored in the influence graph. Each edge of the current diagram is
a region without conflict over the current subset of segments, and has pointers
towards the segments that define it. Updating the diagram upon inserting a
segment S involves removing all the edges or portions of edges in Vor(S,) that
are contained in the Voronoi cell V(S) of the new diagram Vor(S, U {S}), and
adding the new edges that are on the boundary of V(S). We note that the edges
of Vor(S,) that intersect V(S) are the regions that conflict with S. The following
lemma will be very useful later on:
Lemma 19.2.7 The set A of edges or portions of edges in Vor(S,) that are
contained in the cell V(S) of the new diagram Vor(S, U {S}) is connected.
Proof. Assume the contrary. Then there is a path r contained in V(S) that
connects two points X and Y on the boundary of V(S), does not intersect the
edges in A, and subdivides V(S) into two connected components that each contain
a connected component in A. Since F does not intersect the edges in A, it is
entirely contained in a single cell of Vor(S,), say the cell V,(R) that corresponds
to segment R E S,. But then X and Y belong to the boundary of the cell V(R)
of R in the new diagram Vor(S, U {S}), so there must exist a path F' contained
in V(R) with endpoints X and Y. The union of F and IF' is a simple closed curve
in V,(R) whose interior contains a connected component of A. Thus V,(R) is not
simply connected, which contradicts lemma 19.2.2. 0
To find the conflicting edges rapidly, the algorithm also maintains an influence
graph. We may recall that the influence graph is a structure used for detecting
conflicts between the new segment and the regions defined and without conflict
over the current subset of segments S,. The influence graph is an oriented acyclic
graph that contains a node for each region that was defined and without conflict
over the current subset at some previous incremental step. At each step of the
algorithm, the regions defined and without conflict over S, are stored in the leaves
of the influence graph. The arcs in the influence graph connect two nodes in such
a way that a segment that conflicts with the region stored at a node also conflicts
with the region stored in one of this node's parents. Thus the influence domain
DE of a region E stored at a node is contained in the union of the influence
domains DEj of the regions E1 stored at this node's parents.
The algorithm proceeds in two phases, first locating the edges or portions of
edges to be removed, and then updating the current Voronoi diagram and the
corresponding influence graph.
Locating. The location phase aims at retrieving all the leaves in the influence
graph that conflict with the new segment S inserted during the current incremen-
tal step. This can be achieved by traversing the influence graph from the root
only through the nodes that conflict with S.
Updating. The reader is referred to figure 19.9. All the regions that conflict
with S, found during the location phase, correspond to edges in the Voronoi
Figure 19.9. Updating the diagram.
diagram that are modified or disappear in the new diagram. Consider such an
edge E, that belongs to the bisector of two segments S1 and S2. E intersects
the region V(S) in the Voronoi diagram of S, U {S}. If E does not intersect
the boundary of V(S), then E disappears from the new diagram. Otherwise, the
intersection points between E and the boundary of V(S) are new vertices in the
Voronoi diagram: they are the vertices that are at equal distance from S1, S2,
and S. Owing to lemma 19.2.4, there are at most two of them for each edge E,
and they can be computed in constant time. The at most two portions of E that
belong to V(S) disappear from the new diagram while the at most two portions
outside V(S) become two new edges. Each of these portions is a new region that
becomes a child of E in the influence graph.
It remains to compute the new edges that form the boundary of V(S) in the
diagram. These edges have their endpoints at the vertices that we have just
computed and they are contained in the edges of V(S, Si) for some already in-
serted segments Si. These edges are computed in counter-clockwise order along
the boundary of V(S) by the following operations. Let V be a new vertex that
belongs to an edge that is equidistant from S and S,. V is the endpoint of a new
edge E to be created. From V, the next vertex V' of E is found by following
the edges in V(S 1 ) that conflict with S in the previous diagram. (Lemma 19.2.7
sit
Figure 19.10. The algorithm is correct.
proves that these edges are connected.) V' is equidistant from S, S,, and S3. We
then create the new edge E; the corresponding node in the influence graph has for
parents all the edges visited between V and V'. Starting from V' and following
the boundary of V(S 3 ), we discover a new edge and we repeat this procedure
until V is encountered again. Then all the edges on the boundary of V(S) have
been created, and the update phase is over.
It is easy to see that the update procedure described above creates all the
edges in the Voronoi diagram. It remains to show that it correctly updates the
influence graph. For this, we show that a segment that conflicts with a node of
the influence graph conflicts with at least one of its parents. Consider for instance
a new edge E and a segment L that conflicts with E (see figure 19.10). There
exists a maximal disk D centered on E whose interior does not intersect any of
the segments inserted, including S, but that intersects L. D is tangent 2 to S and
to another segment S'. Let D' be the disk tangent to S' at the same point as
D, maximal among those that cut S but no other segment in the current set of
segments Sc. Then D' contains D and its center C' belongs to an edge E' of the
cell V(S') of the Voronoi diagram of Sc that conflicts with S. Thus L conflicts
with E'. Moreover E' intersects V(S) and C' belongs to V(S). It follows that
the portion of E' n V(S) that contains C' was traversed during the update phase,
and so E received E' as a parent. This finishes the proof of the correctness of
the update phase.
To use the results of the analysis of randomized incremental algorithms, we
must verify the three clauses of the update condition 5.3.3: detecting a conflict is
performed in constant time (condition 1), the number of children of a node must
2
We say that a disk is tangent to a segment if their intersection consists of a single point. It
could either be tangent to the line that supports the segment or contain only one of its endpoints.
be bounded by a constant (condition 2), and the update phase can be performed
in time proportional to the number of conflicts between S and the edges of the
Voronoi diagram of S, (condition 3).
Condition 1 is clearly satisfied.
As we have seen, an edge of the current Voronoi diagram is split into at most
three pieces by the new region V(S): two pieces outside V(S) and one piece
inside, or two pieces inside and one piece outside, or one piece inside and one
piece outside. To a portion of E outside corresponds a new node in the influence
graph that is a child of E. A portion of E inside V(S) belongs to two cells in the
previous diagram. It is therefore traversed twice during the update phase and
two children are attached to E. The maximum number of children of E is thus
five, which shows that condition 2 is satisfied.
For condition 3, we note that the first part of the update phase requires only
constant time per conflicting edge. For the second part of the update phase,
during which the new edges are created, the incidence graph of the edges in the
Voronoi diagram is traversed, and each edge is visited only four times (since each
edge has at most two portions in V(S), each visited at most twice). This shows
that condition 3 is satisfied.
Finally, the results of theorem 5.3.4 apply, and we must estimate the maximum
number fo(n, S) of edges defined and without conflict over a set of n segments.
This number is simply the number of edges in the Voronoi diagram, which is
0(n). Theorem 5.3.4 shows that the expected cost of inserting the n-th segment
is O(logn).
Theorem 19.2.8 The Voronoi diagram of n segments in the plane can be com-
puted by a randomized on-line algorithm in expected time 0(n log n). The expected
cost of inserting the n-th segment is 0(logn).
19.2.4 The case of connected segments

Up to now, we have required that the segments are disjoint. It is very natural to
ask for the Voronoi diagram of a polygon, or of a collection of polygons, or more
generally of segments that do not intersect except possibly at their endpoints. In
such a situation, the set of points at equal distance from two segments may be a
region of the plane (see figure 19.11a). We modify the definition of the Voronoi
diagram in one of two ways, in order to ensure that the bisectors remain curves.
One possibility is to consider the interiors of the segments and their common
endpoint as three sites. This leads to diagrams of the kind depicted in figure
19.11b. The other possibility is to consider the interiors of the segments but not
their endpoints: this leads to diagrams of the kind depicted in figure 19.11c. This
latter definition allows us to consider only disjoint segments and to extend the
distance from Si and S2
(b) (c)
Figure 19.11. The two possible ways to define the Voronoi diagram of a set of connected
segments.
previous results to the case of non-intersecting segments that may share their
endpoints. The two diagrams differ, but one may be computed from the other in
linear time.
In this section, we show that the general technique of accelerated randomized
algorithms (see section 5.4) can be used to compute the Voronoi diagram of a con-
nected set S of n segments with disjoint relative interiors in time O(n log* n). The
same algorithm can be used to compute the Voronoi diagram of a simple polygon
P, which is the portion of the Voronoi diagram of the edges of P contained in
the interior of P. Such a diagram is depicted in figure 19.12.
The algorithm is essentially the same as that presented in the previous subsec-
tion. The main difference is that, at certain incremental steps in the construction,
the algorithm computes the conflict graph between the regions, i.e. edges of the
current Voronoi diagram, and the remaining objects, i.e. segments yet to be in-
serted. (We recall that the conflict graph is a bipartite graph that stores an arc
between a region and the remaining objects that conflict with it.) This graph is
used to speed up the subsequent locations. Section 5.4 explains the operations
for such algorithms in detail.
The general analysis presented in section 5.4 applies, if we can show how to
compute the conflict graph at step r in time O(n), meaning that we must compute
all the conflicts between the regions defined and without conflict over the subset
R of segments inserted before or during step r and the segments in S \ 1Z. This
Figure 19.12. The Voronoi diagram of a simple polygon.
can be carried out by traversing the incidence graph of the segments in S. Pick
any segment in the sample, say So, and find all the edges of Vor(7?) that are
determined by So, in time 0(n). Then traverse the incidence graph starting at
So, say, in a depth-first fashion, and for each segment that has not yet been
inserted, enumerate the edges of Vor(7?) with which it conflicts.
Denote by S. the current segment in this traversal of the graph, and let SP
be a segment incident to S, that has been already visited. Either Sp belongs to
7?, or we already know the edges of Vor(7?) that conflict with Sp. If S, belongs
to 7?, the traversal proceeds to the next incident segment not yet visited (case
1). Otherwise, we seek an edge of Vor(R.) that conflicts with S,. If Sp belongs
to 7? (case 2a), we identify, among the edges of Vor(R?) determined by Sp (the
edges on the boundary of V(Sp)), an edge R that conflicts with S,. If Sp does
not belong to 7? (case 2b), among all the edges of Vor(7?) that conflict with Sp,
we look for an edge R that conflicts with S,. In either case, it is easily checked
that we find such an edge R. The other edges of Vor(7?) that conflict with S,
are connected by lemma 19.2.7, so they can be found by traversing the incidence
graph of the edges of Vor(7R), starting at R. At the end of the traversal, all
the conflicts between the segments in S \ 7? and the edges in Vor(7?) have been
identified.
The cost of processing cases of type 1 is bounded by r < n. Each edge of Vor(7?)
is determined by at most four segments in 7?, so it is examined at most four times
during the search for a first conflict in cases 2a. The cost of processing cases 2a is
thus 0(r). The cost of finding the remaining conflicts or a first conflict in cases
2b is proportional to the number of conflicts detected. Indeed, each vertex of
Vor(RZ) is incident to three edges, assuming L2-general position, and we suppose

that the incidence relationships are stored explicitly, or at least can be retrieved
in constant time so that it takes constant time to find one conflict from another.
Theorem 4.2.6 and corollary 4.2.7 show that the expected number of conflicts
detected is 0 (!r!fo(r,S)), which is 0(n).
The hypotheses of theorem 5.4.2 are satisfied, and we may conclude that:
Theorem 19.2.9 There is an incremental algorithm that computes the Voronoi

diagram of a connected set of n segments that have disjoint relative interiors in
expected time 0(n log* n).
19.2.5 Application to the motion planning of a disk

Let £ be a bounded polygonal region, possibly with polygonal holes. Let n denote
the total number of edges in £. By the Voronoi diagram of C, we mean the portion
of the Voronoi diagram of the edges of C that is contained in the interior of E.
If the Voronoi diagram of E is precomputed, then a greatest disk contained in E
can be found in linear time. Indeed, such a disk is tangent to three edges in £ (it
has only one intersection point with each edge), and hence it is a vertex of the
Voronoi diagram.
If our goal is to plan the motion of a disk D of radius r inside £, then the
Voronoi diagram can also be used efficiently. Intuitively, the edges of £ are the
obstacles, and the edges of the Voronoi diagram are the positions that are the
farthest possible from any obstacle. We may restrict the motion of the disk to
these positions rather than the entire domain C: a geometric problem is thus cast
into a graph-theoretic problem.
Let X be a point inside E. We map it to a point v(X) as follows. If X belongs
to an edge of the Voronoi diagram of C, then we put v(X) = X. Otherwise, X
belongs to a single cell in this Voronoi diagram, say V(Sx), attached to a segment
Sx. Let X' be the closest point in Sx to X, and draw the ray from X' to X.
It follows from lemma 19.2.1 that the segment X'X is contained in the cell, and
the ray intersects the boundary of V(Sx) in at most one point. Since the cells
are bounded, this point always exists and is our choice for v(X). Thus v(X) is
well-defined.
Equivalently, v(X) is the center of the greatest disk contained in E in the pencil
of disks tangent to Sx at X'.
Any point X is mapped onto a single point v(X), and the boundary of V(Sx) is
connected by lemma 19.2.2. As a consequence, the restriction of v(X) to a single
cell of the diagram is continuous. As the boundaries of the cells are invariant
under v, v is continuous over its entire domain of definition.
To each point X inside C corresponds the distance from X to E, which is the
Figure 19.13. Motion planning of a disk.
same as its distance to X' defined above. It also equals the radius p(X) of the
greatest disk centered at X that is contained in S.
We say that an arc r that connects two points A and B is an admissible path
if D moving along this path remains entirely inside E.
Lemma 19.2.10 If there exists an admissible path -y that connects A to B then

there exists an admissible path -y' that consists of the segment Av(A), a path from
v(A) to v(B) contained in the edges of the Voronoi diagram, and the segment
v(B)B.
Proof. Let 'y" be the path obtained by applying v to all the points in 'Y. Then
," connects v(A) to v(B) and is contained in the edges of the Voronoi diagram.
Moreover, we have p(X) < p(v(X)), since p is non-decreasing on the segment
from X to v(X). Therefore D remains inside E when moving along Oy", showing
that Oy" is admissible. If A and B are admissible positions for D, then so are the
paths Av(A) and v(B)B. E
The algorithm that computes an admissible path from a position A to another

position B proceeds as follows (see figure 19.13). We first test in time O(n)
whether A and B are admissible positions, meaning that the disk D centered at
A or B is entirely contained inside S. If this is not the case, then there is no
19.3. The case of points distributed in two planes 481
solution to the problem. Otherwise, we compute the Voronoi diagram of & in

time O(n log n), and remove the portions of the edges for which p(X) < r in time
0(n). Then we compute v(A) and v(B) and we find a path -y" in the incidence
graph of the remaining edges of the Voronoi diagram. This also takes time 0(n).
Theorem 19.2.11 Let D be a disk and £ be a polygonal region with a total of

n edges. Knowing the Voronoi diagram of E, one can in time 0(n) either find a
path from a position A to a position B along which D remains entirely inside £,
if it exists, or otherwise correctly conclude that D cannot reach B from A while
remaining entirely inside E.
19.3 The case of points distributed in two planes

Studying the Voronoi diagram of points distributed in two planes in E3 presents a
level of difficulty intermediate between the planar case and the three-dimensional
case. We will see that such a diagram may have quadratic size, but that it can
be computed optimally in time that depends on the size of the input and of the
output. This is very interesting in practice, since the quadratic bound is rarely
achieved.
19.3.1 The two planes are parallel

Let P1 and P2 be two parallel planes in E3 , M 1 a set of nl points in P1 , and M 2
a set of n2 points in P2 . We put M = M 1 U M 2 . To simplify the explanations,
we assume that the points are in L2 -general position, meaning that no five points
in M are co-spherical. This also implies that no four points of Ml or M2 are
co-circular.
The following lemma shows that the planar Delaunay triangulations Del(Mi)
of M 1 and Del(M2 ) of M 2 can be computed easily from the three-dimensional
Delaunay triangulation Del(M) of M. Conversely, we show later how to compute
Del(M) knowing Del(M1 ) and Del(M2 ).
Lemma 19.3.1 The intersection of Del(M) with Pi, i = 1, 2, yields exactly

Del(Mi).
Proof. Del(M) is a simplicial 3-complex whose vertices are the points of Ml

(which belong to Pi) and the points of M 2 (which belong to P2 ). Its intersection
with Pi is thus a 2-complex whose vertices are the points of Mi (i = 1, 2). A face
F of Del(M)nPi is a face of Del(Mi). Indeed, it follows from theorem 17.3.4 that
there exists a sphere SF passing through the vertices of F whose interior contains
no point in M. The intersection of SF with Pi therefore is a circle that contains
the vertices of F and whose interior contains no point of Mi. Theorem 17.3.4,
this time used in the plane Pi, shows that F is a face of Del(Mi). Conversely, if
F is a face of Del(Mi), then F can always be circumscribed by a sphere whose
interior contains no point of M: simply place the center of this sphere far away
from the plane Pj, j k i. Theorem 17.3.4 then shows that F is a face of Del(M).
C
Notice that this result does not depend on the distance between the two planes,
but it is important that they are parallel for proving the second statement in the
lemma.
We now show how to compute Del(M) knowing Del(M1 ) and Del(M2 ). The
previous lemma shows that we already have the faces of Del(M) contained in PI
and P2 . The following lemma characterizes the others.
Lemma 19.3.2 The faces of Del(M) that are not faces of Del(Ml) orDel(M2 )
are in one-to-one correspondence with the 2-complex obtained by projecting the
edges of Vor(Ml) onto Vor(M 2 ) orthogonally to P1 and P2 -
Proof. Consider two points CG in P1 and C2 in P2 that belong to a line orthogonal

to Pi and hence to P2 (since P1 and P2 are parallel). We denote by Ei the greatest
circle in Pi centered at Ci and whose interior does not contain any point in Mi
(i = 1, 2). There exists a unique sphere E12 that intersects P1 along El and P2
along E2. By construction, the interior of this sphere does not contain any point
in M.
We must distinguish between several cases, according to whether Ci (resp. C2)
belongs to a cell, to an edge, or is a vertex of Vor(Ml) (resp. Vor(M2)).
Case 1. Ci belongs to the interior of a cell of Vor(Ml), say V(A 1 ), and C2
belongs to the interior of a cell of Vor(M 2 ), say V(A 2 ). Then El passes through
A 1 , F22 passes through A2 , and hence E12 contains both A1 and A2 . Theorem
17.3.4 shows that A1 A2 is an edge in Del(M).
Reciprocally, let A1 A2 be an edge in Del(M). By theorem 17.3.4, there is a
sphere 212 that passes through A1 and A2 and whose interior does not contain
any point in M. This sphere intersects P1 and P2 along two circles l1and E2
whose interiors do not contain points in M 1 or M 2 . The center of El thus belongs
to V(A 1 ) and that of E 2 belongs to V(S 2 ).
As a consequence, the orthogonal projection of V(A 1 ) onto P2 and V(A 2 ) have
a non-empty intersection if and only if A1 A2 is an edge of Del(M). There exists
a bijection between the edges of Del(M) that connect a point of M 1 to a point
of M2 and the cells in the planar subdivision obtained by projecting Vor(Ml)
orthogonally onto Vor(M 2 ) in P2 .
Let us denote by C this planar 2-complex in P2 . Thus C is the overlay of the
orthogonal projection of Vor(Ml) onto P2 and of Vor(M 2 ). In the rest of this
proof, the only projection used will be the orthogonal projection onto P2.
Case 2. C1 belongs to an edge of Vor(Ml), say V(AlBl), and C2 belongs to

the interior of a cell of Vor(M 2 ), say V(A 2 ). E12 thus passes through the two
points A1 and B1 in M 1 , and through the point A2 in M 2 . It does not contain
any point of M in its interior by construction, so A1B1 A2 is a face of 79el(M).
The converse can be shown in a manner almost identical to the first case. As
a result, the edges of C that are intersections of the projection of an edge of
Vor(Mi) with a cell of Vor(M2 ) correspond bijectively to the faces in Del(M)
that have two vertices in M1 and one in M 2 .
Case 3. Cl belongs to the interior of a cell of Vor(Mi), and C2 belongs to
an edge of Vor(M 2 ). This case is entirely symmetric to the second case, and
establishes a bijection between the edges of C obtained by intersecting an edge
of Vor(M2 ) with the projection of a cell of Vor(Mi), and the faces in Del(M)
that have two vertices in M 2 and one in M 1 .
Case 4. Ci belongs to an edge of Vor(M 1 ), say V(AlBl), and C2 belongs to
an edge of Vor(M2 ), say A2 B2 . Then E12 passes through the two points A1 and
B1 in M 1 , and through A2 and B2 in M 2 . By construction, E12 does not contain
points in M, and thus A1 BlA2 B2 is a tetrahedron in Del(M).
The converse is also true, showing a bijection between the vertices of C obtained
by intersecting the projection of an edge of Vor(M 1 ) with an edge of Vor(M 2 ),
and the tetrahedra in Del(M) that have two vertices in M1 and two vertices in
M2 .
Case 5. Ci is a vertex of Vor(Ml), say V(A 1BiCi), and C2 belongs to the
interior of a cell of Vor(M2 ), say V(A 2 ). Then E12 passes through three points
in M 1 and one point in M 2 , and contains no point in M by construction. Thus
A 1 BlClA2 is a tetrahedron in Del(M).
The converse is also true, showing a one-to-one correspondence between the
vertices of C that are the projections of vertices of VorA(M 1 ) and the tetrahedra
in Del(M) that have a facet in P1 .
Case 6. Ci belongs to the interior of a cell of Vor(Mi), and C2 is a vertex of
Vor(M 2 ). This case is entirely symmetrical to the fifth case, and so provides a
bijection between the vertices of C that are vertices of Vor(M 2 ) and the tetrahe-
dra in Del(M) that have a facet in P2 . El
Figure 19.14 shows a simple example. There are three points in P1 , A1 , B1 , and
Cl, and three points in P2 , A2 , B2 , and C2. Vor(Ml) and Vor(M 2 ) have each
one vertex and three edges. Vor(M) contains a vertex corresponding to case 5,
namely the center of the sphere circumscribed to A 1 B1 ClA 2 , a vertex correspond-
ing to case 6, namely the center of the sphere circumscribed to A 1 A 2 B2 C2 , and
three vertices corresponding to case 4, the centers of the spheres circumscribed
to A 1 ClA2 C2 , A 1 ClB2 C2 , and B 1ClB2 C2 .
V(AiBi)
B2 C2 )
V(BICI)
V(AlCi)
V(A 2 C2 )
Figure 19.14. Construction of Vor(M) knowing Vor(Mi) and Vor(M42). The triangulation
Del(M) consists of five tetrahedra Si = AiBiCiB2 , S 2 = A 1 A 2 B 2C2 , S3 =
BiCiB2 C 2 , S4 = AiCiB2 C2 , and S5 = AlC 1A 2C2.
Lemma 19.3.2 gives a method for computing Vor(M) knowing Vor(M1) and
Vor(M 2 ). In fact, it reduces the problem to that of computing the 2-complex
C that is the overlay of the orthogonal projection of Vor(Ml) onto P 2 and of
Vor(M 2 ). We may compute this complex by using any of the randomized al-
gorithms that compute the intersection of a set of line segments presented in
subsections 5.2.2 and 5.3.2, or a more sophisticated deterministic algorithm (see
the bibliographical notes of chapter 3). The complexity of these algorithms is
0(m log m + t) if m is the number of segments and t the number of intersection
points. Here, m = O(n) is the total number of edges in the planar Voronoi dia-
grams Vor(Mi) and Vor(M 2 ), and t is the number of vertices of Vor(M). This
finishes the proof of the following theorem.
Theorem 19.3.3 Given a set M of n points that belong to two parallel planes,
we can compute its Voronoi diagram Vor(M) in time 0(nlogn + t), where t is
the number of tetrahedra in Vor(M). This is optimal as a function of the input
and output sizes.
19.3.2 The two planes are not parallel

In this section, we are now interested in a set M of points in E3 that belong to
two planes P1 and P2 that intersect along a line L. We write Mi = M n Pi,
i = 1, 2, and for simplicity we assume that the points in M are in L2 -general
position.
Let us first assume that the points in M1 are contained in a half-plane H1 of
P1 bounded by L, and that the points in M 2 are also contained in a half-space
H2 of P2 bounded by L. Each plane has an orthonormal system of coordinates,
by choosing a common origin 0 and a common unit vector contained in L, and
the other unit vectors in P1 and P2 perpendicular to L contained in H1 and H2 .
Thus the equation of Hi is yi > 0 i = 1, 2.
Lemma 19.3.1 still holds and its proof is unchanged. Lemma 19.3.2 can be
generalized as shown in this section. As in section 17.2, we represent the circles
in the planes P1 and P2 by the points in a three-dimensional Euclidean space,
which we denote by £ in order not to confuse it with the Euclidean space that
contains M. More precisely, the circle Ei in the plane Pi (i = 1 or 2) whose
center has coordinates Ci in Pi and with respect to which the origin has a power
of vi (in that plane), is mapped to the point O(Ei) = (Ci, vi) in E. We identify
the x-axes of P1 and P2 and the first axis of 8, which we also call the x-axis of 8,
so that there is no confusion between the notation x for the points in 8 and in E3.
Also, we identify the yl-axis of P1 , the y2-axis of P2 , and the second axis of E,
which we call the y-axis. The third axis of 8 is naturally called the z-axis. Note
that, with the identification of x and y in PI, P2 , and 8, M 1 and M 2 induce two
point sets in the xy-plane of 8, which we also denote by M 1 and M 2 .
We denote by V(Mi) (i = 1,2) the Voronoi polytope of Mi. Recall from
subsection 17.2.7 that V(Mi) is the polytope in £ defined as the intersection of
the half-spaces lying above the hyperplanes tangent to the paraboloid P in 8
at the points of M lifted onto P. V(Mi) projects vertically onto Pi along the
Voronoi diagram of Mi in the plane Pi.
Remark. Under the assumption that the points in M 1 are contained in the
half-plane H1 and those of M 2 in H2 , the points of M 1 and M 2 are lifted onto
the portion of the paraboloid contained in y > 0. Hence the projection parallel
to the y-axis onto the plane y = 0 of the polytope V(M 1 ) covers the entire plane,
and so does the projection of V(M 2 ). (This projection map plays an important
role below.)
Lemma 19.3.4 The faces of Del(M) that are not faces of Del(M1) orDel(M2)
can be put in one-to-one correspondence with the faces of the 2-complex C obtained
by projecting onto the plane y = 0, parallel to the y-axis, the edges of the polytopes
V(Mi) and V(M 2 ) in 8.
'I
,21G
Ill
Figure 19.15. The pencil of spheres F2 .
Proof. The proof is analogous to that of lemma 19.3.2, so we only mention how
to construct the bijection between the 0-faces of C and the tetrahedra of Del (M).
A vertex of C is either the projection of a vertex of V(M 1 ) or V(M 2 ), or the
intersection of the projections of two edges of V(Mi) or V(M 2 ).
Case 1. Consider a vertex Sl of V(Ml). S1 is the image under 4 of a circle
El in P1 that passes through three points Al, B1 , Cl in M 1 , and the interior of
El contains no point in Ml. Let F be the pencil of spheres in E3 that intersect
along El (see figure 19.15).
This pencil of spheres intersects P2 along a pencil of circles F2 in P2 whose
image under 4 is the line 0(F 2 ) parallel to the y-axis that contains Si. Indeed,
the radical axis of the circles in -F2 is just L = PinP2 since Pi is the radical plane
of F. The line that supports the centers of the circles in -F2 is thus orthogonal
to L and so 0(:F2 ) is contained in a plane perpendicular to the x-axis. Since
o belongs to L, it has the same power with respect to all the circles in F2 , so
O(F2 ) is also contained in a plane perpendicular to the z-axis. Moreover, q(F 2 )
contains S1, since the center of El has the same abscissa as the centers of the
spheres in the pencil :F2 and since the power of 0 with respect to any circle in
.F2 equals the power of 0 with respect to any sphere of F and hence with respect
to El. This shows that 4(.F 2 ) is the line parallel to the y-axis that contains S1.
The line O(.F2 ) intersects V(M 2 ) in exactly one point. This point is the image
under 0 of the sphere E2 in F2 that contains a point of M 2 , say A2 , but contains
no point of M 2 in its interior. There exists a unique sphere that intersects P1
along El and P2 along E2, and this sphere belongs to F. This sphere passes
through the four points Al, B1 , C1 in M 1 and A2 in M 2 . Its interior contains
no point of M 1 or M 2 . It follows from theorem 17.3.4 that A1 BlClA2 is a
tetrahedron in the Delaunay triangulation Del(M).
Case 2. An analogous discussion establishes a map from the projections of

the vertices of V(M 2 ) to the tetrahedra in Del(M) that have three vertices in
P2 and one in PI.
Case 3. Consider now a vertex S of C that is the intersection of the projection
of two edges, E1 in V(M 1 ) and E2 in V(M 2 ). The line A that contains S and is
parallel to the y-axis intersects E1 in exactly one point, say Sl, and intersects E2
in another point S2. Si (i = 1, 2) is the image under 0 of a circle El in Pi that
passes through two points of M 1 , Ai and Bi, and whose interior in Pi contains
no point in Mi. The powers of 0 with respect to El and E2 are equal, so there
exists a unique sphere E that intersects Pi along l1and P2 along E2. This sphere
passes through the four points A 1 , B1 , A2 , B2 and its interior contains no point
in M 1 or M 2 . Thus A1 BlA2 B2 is a tetrahedron in the Delaunay triangulation
Del (M).
Converse of cases 1, 2, 3. Let T be a tetrahedron in Del(M). It is circum-
scribed by a sphere E that intersects P1 along a circle El and P2 along a circle
E2. Neither circle contains points of M 1 or M2 in its interior. Both circles are
mapped by 0 onto points that belong to a line A parallel to the y-axis, since all
the points on the x-axis have the same power with respect to El and E2. Note
that A intersects V(Mi) in only one point (i = 1, 2).
The tetrahedron T has either three vertices in P1 and one in P2 (first case),
three vertices in P2 and one in P1 (second case), or two vertices in each of P1
and P2 (third case). In the first case, El passes through three points in M 1
and E2 passes through only one point in M2 . So A contains a vertex of V(M 1 )
and intersects the interior of a facet of V(M 2 ). The second case is entirely
symmetrical. In the third case, Ei contains exactly two points in Mi (i = 1, 2) so
that A intersects the relative interiors of two edges, one of V(M1 ) and the other
of V(M 2 ). 0
In the preceding discussion, we assumed that the points in M 1 were contained
in a half-plane Hj+ of P1 bounded by L, and that similarly the points in M 2
were contained in a half-plane H2+ of P2 bounded by L. If this is not true,
lemma 19.3.1 collapses and the triangles in Del(Mi) are not necessarily all faces
in Del(M), since some triangles in Del(Mi) may admit no circumscribed sphere
whose interior contains no points in Mj (j # i). Nevertheless, we can salvage our
construction by considering successively the four dihedra defined by P1 and P2 .
More precisely, denote by V(Mi)+ (resp. V(Mi)-) the portion of V(Mi) visible
from (O,+oo,O) (resp. (O,-oo,O)). Each pair of the kind (V(Ml)*,V(M 2 )*),
where *, * E { -, +}, corresponds to a dihedron. The previous construction for
this dihedron applies to this pair with almost no modification. The only difference
comes from the fact that the projection of V(Mi)* parallel to the y-axis may
not cover the plane y = 0 entirely, but only a region R*, and similarly the
projection of V(M 2 )* parallel to the y-axis may cover only a region R* of the
plane y = 0. (See also the remark preceding lemma 19.3.4). The previous lemma
still holds if the complex C is restricted to the region R* n R*. It is also worth
observing that a vertex of V(Mi)+ or V(Mi)- that does not project onto R+ or
R17 (j =A i) corresponds to a triangle of Del(Mi) that is not a face of Del(M).
A vertex of the resulting internal subdivision is in bijection with a tetrahedron
of some dihedron, meaning that the center of its circumscribed sphere belongs
to that dihedron. Proceeding similarly for all four dihedra retrieves the entire
triangulation Del (M).
This finishes the proof of the following theorem.
Theorem 19.3.5 Given a set M of n points that belong to two planes, we can
compute their Voronoi diagram Vor(M) in time O(n log n + t), if t is the size of
this diagram. This is optimal both in the size of the input and in the size of the
output.
The previous construction is the strict analogue of the construction for points
distributed in two parallel planes, when embedded in the space that represents
circles. The reader will also notice similarities with section 18.5 devoted to hy-
perbolic diagrams. Exercise 19.11 explains the reasons for these similarities.
19.4 Exercises
Exercise 19.1 (Divide-and-conquer) Adapt the divide-and-conquer method to com-
pute the Voronoi diagram of n points in the plane in optimal time O(n log n).
Exercise 19.2 (Optimal Li diagrams) Devise an optimal deterministic algorithm

that computes the Voronoi diagram of n points in the plane for the L1 distance in
time 0(n log n). Assume that the points are in Li-general position.
Hint: Use either the divide-and-conquer method, or the sweep method.
Exercise 19.3 (Polygonal distance) Let P be a polygon that contains the origin 0.
We denote by AP the image of this polygon under the homothety centered at 0 and
of ratio A. The polygonal distance e6p(X,A) from point X to point A is defined as the
smallest real A > 0 such that X - A belongs to AP. Show that the Voronoi diagram of n
points under this polygonal distance has complexity 0(np) if p is the number of vertices
of P and that it may be computed in time 0(nplog(np)).
Exercise 19.4 (Convex polygon) Show how to compute the Voronoi diagram of a
convex polygon in the plane with a randomized algorithm in expected time 0(n).
Hint: Maintain the convex hull of the polygon while pulling off the vertices one by one
and in a random order. Then insert the points back into the polygon in the reverse
order, while maintaining the Voronoi diagram. During each insertion, we already know
19-4. Exercises 489
a region (half-plane) that conflicts with the vertex to be inserted. The other conflicts
can be retrieved and the structure updated, without paying the logarithmic cost of the
location phase in the usual incremental algorithm.
Exercise 19.5 (Dual of the Voronoi diagram of segments) Consider a set of seg-
ments in the plane, assumed to be in L 2 -general position. A vertex of the Voronoi
diagram is incident to two arcs if the arc is contained in the relative interior of an edge,
or to three arcs if it is an endpoint of an edge. For each vertex contained in two arcs,
the maximal circle centered on this vertex that does not properly intersect the segments
is tangent to two segments, and we draw the edge that joins the two points of tangency.
Show that these edges partition the convex hull of the segments into O(n) regions which
are either triangles or trapezoids.
Exercise 19.6 (Dynamic algorithm) Devise a dynamic randomized algorithm to

compute the Voronoi diagram of a set of segments in the plane.
Exercise 19.7 (Constrained Delaunay triangulation) Consider a set S of segments

in the plane with disjoint relative interiors (some may be reduced to a point). We denote
by E the collection of their endpoints. A constrained triangulation of S is a triangulation
of &that contains all segments in S as edges. This triangulation is a constrainedDelaunay
triangulationif the interior of a circle circumscribed to some triangle T contains no point
that is visible from all three vertices of T. Two points are visible if the segment that
connects them does not intersect the relative interior of any segment in S. Show that
such a triangulation exists for any S, and devise an algorithm that computes it in time
O(n log n).
Exercise 19.8 (Motion planning of a polygon) Let P be a convex polygon with m

edges. Extend the notion of a Voronoi diagram for the polygonal metric defined by P
(see exercises 19.3 and 19.2) to the metric defined by a bounded polygonal region 7?. If 7?
totals n edges, show how to compute the greatest homothet of P contained in 7? in time
O(mnlog(mn)). If I and J are two points in 7?, show how to determine whether there
exists a path from I to J such that the polygon P moving on this path by translation
remains entirely contained in 7?. If so, show how to compute this path.
Exercise 19.9 (Segments in E 3) Characterize the surfaces bisecting two segments in

E3.
Exercise 19.10 (Segments in two planes) Show that the complexity of the Voronoi
diagram of a set of n segments distributed in two planes is t = O(n 2 ). Show how to
compute such a diagram in time 0(n log n + t). Extend these results to polygonal regions
distributed in two planes.
Exercise 19.11 (Two non-parallel planes) Consider a set M of points in E 3 dis-

tributed on two half-planes P1 and P 2 bounded by a common line L. Denote by Ml and
M 2 the sets of points in either plane. Show that the tetrahedra of Del(M) circumscribed
by a sphere that does not intersect L are in one-to-one correspondence with the vertices
of the subdivision obtained by overlaying the hyperbolic diagram Vorh(M 2 ) onto the
hyperbolic diagram Vorh(Ml), with a rotation about the axis L that maps P2 onto P1.
Hint: Adapt the discussion in section 19.3 (for two non-parallel planes) using argu-
ments borrowed from section 18.5.
Exercise 19.12 (The case of several planes) Show how to compute the Delaunay
triangulation of a set M of n points in E3 distributed in k given planes in time O(kn log n).
Hint: Compute the triangulation by a greedy method, finding the tetrahedra one
by one. If F is a facet that belongs to a single computed tetrahedron T, we seek the
tetrahedron T' that shares the facet F with T. Notice that the spheres circumscribed
to T and T' are the two greatest spheres in the pencil of spheres that contain the circle
circumscribed to F and whose interiors contain no points of M. This pencil intersects
each of the k planes along a pencil of circles. In each plane Pi, we seek the greatest circle
in the corresponding pencil whose interior contains no point in Mi = M n Pi. This
problem reduces to finding the intersection between a line and a polytope in E3, which
can be solved in logarithmic time (see exercise 9.6). Among all k candidates, pick the
one that gives the fourth vertex of T'.
Exercise 19.13 (Conformal triangulation) Consider a set S of n segments in the

plane, and let £ denote the set of their endpoints. Show that by adding a finite set P of
points on the segments of S, each segment becomes an edge of the Delaunay triangulation
of £ U P. Show that one may have to add Q(n 2 ) points to £ in the worst case.

The sweep algorithm described in section 19.1 and its extension to the case of additive
weights and line segments is due to Fortune [106]. Its generalization to L1 and Lo
distances is described by Shute, Deneen, and Thomborson in [209].
The incremental algorithm that computes the Voronoi diagram of a set of line segments
in the plane is presented by Boissonnat, Devillers, Schott, Teillaud, and Yvinec [28], and is
dynamized by Dobrindt and Yvinec [86] (see exercise 19.6). The analysis of the case when
the segments are connected is due to Devillers [80]. The case when the segments are the
edges of a convex polygon is treated by Aggarwal, Guibas, Saxe, and Shor who propose
a deterministic algorithm [5]. The remarkably simple randomized solution presented in
exercise 19.4 is due to Chew [59].
Optimal algorithms that compute the constrained Delaunay triangulation of a set of
points (see exercise 19.7) are described by Chew [60] and by Wang and Schubert [217].
A randomized linear time algorithm by Klein and Lingas computes the constrained tri-
angulation of a polygon [140].
Conformal triangulations were used first by Boissonnat [25]. Edelsbrunner and Tan
[100] give a polynomial solution to exercise 19.13.
O'Dinlaing and Yap were the first to use the Voronoi diagram for planning the motion
of a disk. The case when the robot is polygonal rather than circular, and can move
by translation and rotation in the plane (with three degrees of freedom), is studied by
6
'Dunlaing, Sharir, and Yap in [180, 181] and by Chew and Kedem in [62].
Bounding the complexity of the Voronoi diagram of a set of line segments in E3 is a

long-standing open question. Related results for the L1 and Loo metrics were established
by Chew, Kedem, Sharir, Tagansky, and Welzl [63].
The case of points distributed in two parallel planes has been studied by Boissonnat in
the context of three-dimensional shape reconstruction from cross-sections in tomography
[25] and generalized to two non-parallel planes and to a (small) number of planes (see
exercise 19.12) by Boissonnat, Cerezo, Devillers, and Teillaud [26].
References
[1] P. K. Agarwal. Intersection and Decomposition Algorithms for PlanarAr-

rangements. Cambridge University Press, New York, NY, 1991.
[2] P. K. Agarwal, M. de Berg, J. Matousek, and 0. Schwarzkopf. Constructing

levels in arrangements and higher order Voronoi diagrams. In Proc. 10th
Ann. ACM Symp. Comp. Geom., 67-75, 1994.
[3] P. K. Agarwal, D. Eppstein, and J. Matousek. Dynamic half-space report-

ing, geometric optimization, and minimum spanning trees. In Proc. 33rd
Ann. IEEE Symp. Found. Comp. Sci., 80-89, 1992.
[4] P. K. Agarwal and J. Matousek. Ray shooting and parametric search. In

Proc. 24th Ann. ACM Symp. Theory Comp., 517-526, 1992.
[5] A. Aggarwal, L. J. Guibas, J. Saxe, and P. W. Shor. A linear-time algorithm

for computing the Voronoi diagram of a convex polygon. Discrete Comp.
Geom., 4(6):591-604, 1989.
[6] A. V. Aho, J. E. Hopcroft, and J. D. Ullman. The Design and Analysis of

Computer Algorithms. Addison-Wesley, Reading, MA, 1974.
[7] P. Alevizos, J.-D. Boissonnat, and F. P. Preparata. An optimal algorithm

for the boundary of a cell in a union of rays. Algorithmica, 5:573-590, 1990.
[8] P. D. Alevizos, J.-D. Boissonnat, and M. Yvinec. Non-convex contour re-

construction. J. Symbolic Comp., 10:225-252, 1990.
[9] B. Aronov and M. Sharir. Triangles in space or building (and analyzing)

castles in the air. Combinatorica, 10(2):137-173, 1990.
[10] B. Aronov and M. Sharir. Castles in the air revisited. In Proc. 8th Ann.
ACM Symp. Comp. Geom., 146-156, 1992.
[11] B. Aronov, M. Pellegrini, and M. Sharir. On the zone of a surface in a

hyperplane arrangement. Discrete Comp. Geom., 9(2):177-186, 1993.
References 493
[12] P. Ash, E. Bolker, H. Crapo, and W. Whiteley. Convex polyhedra,

Dirichlet tessellations, and spider webs. In Marjorie Senechal and George
Fleck, editors, Shaping Space: A Polyhedral Approach, chapter 17, 231-250.
Birkhduser, Boston, MA, 1988.
[13] M. J. Atallah. Dynamic computational geometry. In Proc. 24th Ann. IEEE
Symp. Found. Comp. Sci., 92-99, 1983.
[14] F. Aurenhammer. Power diagrams: properties, algorithms and applications.
SIAM J. Comp., 16:78-96, 1987.
[15] F. Aurenhammer. A new duality result concerning Voronoi diagrams. Dis-
crete Comp. Geom., 5:243-254, 1990.
[16] F. Aurenhammer. Voronoi diagrams: A survey of a fundamental geometric
data structure. ACM Comp. Surv., 23:345-405, 1991.
[17] F. Aurenhammer and H. Imai. Geometric relations among Voronoi dia-
grams. Geom. Dedicata, 27:65-75, 1988.
[18] F. Aurenhammer and 0. Schwarzkopf. A simple on-line randomized incre-
mental algorithm for computing higher order Voronoi diagrams. Internal.
J. Comp. Geom. Appl., 2:363-381, 1992.
[19] D. Avis and H. El-Gindy. Triangulating point sets in space. Discrete Comp.
Geom., 2:99-111, 1987.
[20] C. Bajaj and T. K. Dey. Polygon nesting and robustness. Inf. Proc. Lett.,
35:23-32, 1990.
[21] L. J. Bass and S. R. Schubert. On finding the disc of minimum radius
containing a given set of points. Math. Comp., 12:712-714, 1967.
[22] A. F. Beardon. The Geometry of Discrete Groups. Graduate Texts in
Mathematics. Springer-Verlag, 1983.
[23] J. L. Bentley and T. A. Ottmann. Algorithms for reporting and counting
geometric intersections. IEEE Trans. Comp., C-28:643-647, 1979.
[24] M. Berger. Geometrie 3. Convexes et Polytopes, Polyedres R1guliers, Aires
et Volumes. Fernand Nathan, Paris, 1974.
[25] J.-D. Boissonnat. Shape reconstruction from planar cross-sections. Comp.
Vision Graph. Image Proc., 44(1):1-29, October 1988.
[26] J.-D. Boissonnat, A. C&rezo, 0. Devillers, and M. Teillaud. Output-sensitive
construction of the Delaunay triangulation of points lying in two planes.
Internat. J. Comp. Geom. Appl., 6(1):1-14, 1996.
494 References
[27] J.-D. Boissonnat, 0. Devillers, and F. Preparata. Computing the union of

3-colored triangles. Internat. J. Comp. Geom. Apple, 1(2):187-196, 1991.
[28] J.-D. Boissonnat, 0. Devillers, R. Schott, M. Teillaud, and M. Yvinec.

Applications of random sampling to on-line algorithms in computational
geometry. Discrete Comp. Geom., 8:51-71, 1992.
[29] J.-D. Boissonnat, 0. Devillers, and M. Teillaud. A semidynamic construc-

tion of higher-order Voronoi diagrams and its randomized analysis. Algo-
rithmica, 9:329-356, 1993.
[30] J.-D. Boissonnat and K. Dobrindt. On-line construction of the upper en-
velope of triangles and surface patches in three dimensions. Comp. Geom.
Theory Appl., 5:303-320, 1996.
[31] J.-D. Boissonnat and M. Teillaud. A hierarchical representation of objects:

The Delaunay tree. In Proc. 2nd Ann. ACM Symp. Comp. Geom., 260-268,
1986.
[32] J.-D. Boissonnat and M. Teillaud. On the randomized construction of the

Delaunay tree. Theor. Comp. Sci., 112:339-354, 1993.
[33] J.-D. Boissonnat and M. Yvinec. Probing a scene of non-convex polyhedra.

Algorithmica, 8:321-342, 1992.
[34] J.-D. Boissonnat, M. Sharir, B. Tagansky, and M. Yvinec. Voronoi diagrams

in higher dimensions under certain polyhedral distance functions. In Proc.
11th Ann. ACM Symp. Comp. Geom., 79-88, 1995.
[35] H. Brdnnimann. Derandomization of Geometric Algorithms. Ph.D. thesis,

Dept. Comp. Sci., Princeton University, Princeton, NJ, May 1995.
[36] H. Br6nnimann, B. Chazelle, and J. Matousek. Product range spaces, sen-

sitive sampling, and derandomization. In Proc. 34th Ann. IEEE Symp.
Found. Comp. Sci., 400-409, 1993.
[37] A. Brdnsted. An Introduction to Convex Polytopes. Springer-Verlag, New

York, NY, 1983.
[38] K. Q. Brown. Voronoi diagrams from convex hulls. Inf. Proc. Lett., 9:223-
228, 1979.
[39] H. Bruggesser and P. Mani. Shellable decompositions of cells and spheres.

Math. Scand., 29:197-205, 1971.
References 495
[40] C. Burnikel, K. Mehlhorn, and S. Schirra. On degeneracy in geometric

computations. In Proc. 5th ACM-SIAM Symp. Discrete Algorithms, 16-23,
1994.
[41] B. Chazelle. A theorem on polygon cutting with applications. In Proc. 23rd

Ann. IEEE Symp. Found. Comp. Sci., 339-349, 1982.
[42] B. Chazelle. Convex partitions of polyhedra: a lower bound and worst-case

optimal algorithm. SIAM J. Comp., 13:488-507, 1984.
[43] B. Chazelle. On the convex layers of a planar set. IEEE Trans. Inform.
Theory, IT-31:509-517, 1985.
[44] B. Chazelle. Triangulating a simple polygon in linear time. Discrete Comp.

Geom., 6:485-524, 1991.
[45] B. Chazelle. An optimal algorithm for intersecting three-dimensional convex

polyhedra. SIAM J. Comp., 21(4):671-696, 1992.
[46] B. Chazelle. An optimal convex hull algorithm in any fixed dimension.

Discrete Comp. Geom., 10:377-409, 1993.
[47] B. Chazelle, R. L. Drysdale, III, and D. T. Lee. Computing the largest

empty rectangle. SIAM J. Comp., 15:300-315, 1986.
[48] B. Chazelle and H. Edelsbrunner. An optimal algorithm for intersecting

line segments in the plane. In Proc. 29th Ann. IEEE Symp. Found. Comp.
Sci., 590-600, 1988.
[49] B. Chazelle and H. Edelsbrunner. An optimal algorithm for intersecting

line segments in the plane. JACM, 39:1-54, 1992.
[50] B. Chazelle, H. Edelsbrunner, M. Grigni, L. Guibas, J. Hershberger,

M. Sharir, and J. Snoeyink. Ray shooting in polygons using geodesic tri-
angulations. Algorithmica, 12:54-68, 1994.
[51] B. Chazelle, H. Edelsbrunner, L. Guibas, and M. Sharir. A singly-

exponential stratification scheme for real semi-algebraic varieties and its
applications. Theor. Comp. Sci., 84:77-105, 1991.
[52] B. Chazelle, H. Edelsbrunner, L. Guibas, M. Sharir, and J. Snoeyink. Com-

puting a face in an arrangement of line segments and related problems.
SIAM J. Comp., 22:1286-1302, 1993.
[53] B. Chazelle and J. Friedman. A deterministic view of random sampling and

its use in geometry. Combinatorica, 10(3):229-249, 1990.
496 References
[54] B. Chazelle and L. J. Guibas. Visibility and intersection problems in plane

[55] B. Chazelle, L. J. Guibas, and D. T. Lee. The power of geometric duality.
BIT, 25:76-90, 1985.
[56] B. Chazelle and J. Incerpi. Triangulation and shape-complexity. ACM
Trans. Graph., 3(2):135-152, 1984.
[57] B. Chazelle and J. Matousek. On linear-time deterministic algorithms for
optimization problems in fixed dimension. In Proc. 4th ACM-SIAM Symp.
Discrete Algorithms, 281-290, 1993.
[58] B. Chazelle and L. Palios. Triangulating a non-convex polytope. Discrete
Comp. Geom., 5:505-526, 1990.
[59] L. P. Chew. Building Voronoi diagrams for convex polygons in linear ex-
pected time. Technical Report PCS-TR90-147, Dept. Math. Comp. Sci.,
Dartmouth College, Hanover, NH, 1986.
[60] L. P. Chew. Constrained Delaunay triangulations. Algorithmica, 4:97-108,
1989.
[61] L. P. Chew and R. L. Drysdale, III. Voronoi diagrams based on convex
distance functions. In Proc. 1st Ann. ACM Symp. Comp. Geom., 235-244,
1985.
[62] L. P. Chew and K. Kedem. Placing the largest similar copy of a convex
polygon among polygonal obstacles. In Proc. 5th Ann. ACM Symp. Comp.
Geom., 167-174, 1989.
[63] L. P. Chew, K. Kedem, M. Sharir, B. Tagansky, and E. Welzl. Voronoi
diagrams of lines in 3-space under polyhedral convex distance functions. In
Proc. 6th ACM-SIAM Symp. Discrete Algorithms, 197-204, 1995.
[64] K. Clarkson, H. Edelsbrunner, L. Guibas, M. Sharir, and E. Welzl. Com-
binatorial complexity bounds for arrangements of curves and spheres. Dis-
crete Comp. Geom., 5:99-160, 1990.
[65] K. Clarkson, R. E. Tarjan, and C. J. Van Wyk. A fast Las Vegas algorithm
for triangulating a simple polygon. Discrete Comp. Geom., 4:423-432, 1989.
[66] K. L. Clarkson. Linear programming in O(n3d2 ) time. Inf. Proc. Lett.,
22:21-24, 1986.
[67] K. L. Clarkson. New applications of random sampling in computational
References 497
[68] K. L. Clarkson. A Las Vegas algorithm for linear programming when the
dimension is small. In Proc. 29th Ann. IEEE Symp. Found. Comp. Sci.,
452-456, 1988.
[69] K. L. Clarkson, R. Cole, and R. E. Tarjan. Randomized parallel algorithms
for trapezoidal diagrams. Internat. J. Comp. Geom. Apple, 2(2):117-133,
1992.
[70] K. L. Clarkson, K. Mehlhorn, and R. Seidel. Four results on randomized
incremental constructions. Comp. Geom. Theory Appl., 3(4):185-212, 1993.
[71] K. L. Clarkson and P. W. Shor. Applications of random sampling in com-
putational geometry, II. Discrete Comp. Geom., 4:387-421, 1989.
[72] T. H. Cormen, C. E. Leiserson, and R. L. Rivest. Introduction to Algo-
rithms. The MIT Press, Cambridge, Mass., 1990.
[73] H. S. M. Coxeter. A classification of zonohedra by means of projective
diagrams. J. Math. Pure Appl., 41:137-156, 1962.
[74) H. Davenport and A. Schinzel. A combinatorial problem connected with
differential equations. Amer. J. Math., 87:684-689, 1965.
[75] M. de Berg. Ray Shooting, Depth Orders and Hidden Surface Removal,
volume 703 of Lecture Notes in Computer Science. Springer-Verlag, Berlin,
Germany, 1993.
[76] M. de Berg, K. Dobrindt, and 0. Schwarzkopf. On lazy randomized in-
cremental construction. In Proc. 26th Ann. ACM Symp. Theory Comp.,
105-114, 1994.
[77] M. de Berg, L. J. Guibas, and D. Halperin. Vertical decompositions for
triangles in 3-space. Discrete Comp. Geom., 15:35-61, 1996.
[78] B. Delaunay. Sur la sphere vide. A la memoire de Georges Voronoi. Izv.
Akad. Nauk SSSR, Otdelenie Matematicheskih i Estestvennyh Nauk, 7:793-
800, 1934.
[79] P. Desnogues. Triangulations et Quadriques. These de doctorat en sciences,
University de Nice-Sophia Antipolis, France, 1996.
[80] 0. Devillers. Randomization yields simple O(n log* n) algorithms for diffi-
cult Q(n) problems. Internal. J. Comp. Geom. Appl., 2(1):97-111, 1992.
[81] 0. Devillers, S. Meiser, and M. Teillaud. Fully dynamic Delaunay triangu-

lation in logarithmic expected time per operation. Comp. Geom. Theory
Appl., 2(2):55-80, 1992.
498 References
[82] 0. Devillers, S. Meiser, and M. Teillaud. The space of spheres, a geometric

tool to unify duality results on Voronoi diagrams. Report 1620, INRIA
Sophia-Antipolis, Valbonne, France, 1992.
[83] M. T. Dickerson, S. A. McElfresh, and M. Montague. New algorithms and

empirical findings on minimum weight triangulation heuristics. In Proc.
11th Ann. ACM Symp. Comp. Geom., 238-247, 1995.
[84] M. Dietzfelbinger, A. Karlin, K. Mehlhorn, F. Meyer auf der Heide,

H. Rohnert, and R. E. Tarjan. Dynamic perfect hashing - upper and lower
bounds. In Proc. 29th Ann. IEEE Symp. Found. Comp. Sci., 524-531, 1988.
[85] D. P. Dobkin and D. G. Kirkpatrick. A linear algorithm for determining

the separation of convex polyhedra. J. Algorithms, 6:381-392, 1985.
[86] K. Dobrindt and M. Yvinec. Remembering conflicts in history yields dy-

namic algorithms. In Proc. 4th Ann. Internat. Symp. Algorithms Comp.
(ISAAC 93), volume 762 of Lecture Notes in Computer Science, 21-30.
Springer-Verlag, 1993.
[87] M. E. Dyer. Linear time algorithms for two- and three-variable linear pro-
grams. SIAM J. Comp., 13:31-45, 1984.
[88] M. E. Dyer. On a multidimensional search technique and its application to

the Euclidean one-centre problem. SIAM J. Comp., 15:725-738, 1986.
[89] H. Edelsbrunner. Algorithms in Combinatorial Geometry, volume 10 of

EATCS Monographs on Theoretical Computer Science. Springer-Verlag,
Heidelberg, West Germany, 1987.
[90] H. Edelsbrunner, L. Guibas, J. Pach, R. Pollack, R. Seidel, and M. Sharir.

Arrangements of curves in the plane: Topology, combinatorics, and algo-
rithms. Theor. Comp. Sci., 92:319-336, 1992.
[91] H. Edelsbrunner, L. Guibas, and M. Sharir. The complexity of many cells

in arrangements of planes and related problems. Discrete Comp. Geom.,
5:197-216, 1990.
[92] H. Edelsbrunner and L. J. Guibas. Topologically sweeping an arrangement.

J. Comp. Syst. Sci., 38:165-194, 1989. Corrigendum in 42 (1991), 249-251.
[93] H. Edelsbrunner, L. J. Guibas, and M. Sharir. The complexity and con-

struction of many faces in arrangements of lines and of segments. Discrete
Comp. Geom., 5:161-196, 1990.
References 499
[94] H. Edelsbrunner and H. A. Maurer. Finding extreme points in three di-

mensions and solving the post-office problem in the plane. Inf. Proc. Lett.,
21:39-47, 1985.
[951 H. Edelsbrunner, F. P. Preparata, and D. B. West. Tetrahedrizing point
sets in three dimensions. J. Symb. Comp., 10(3-4):335-347, 1990.
[96] H. Edelsbrunner and R. Seidel. Voronoi diagrams and arrangements. Dis-
crete Comp. Geom., 1:25-44, 1986.
[97] H. Edelsbrunner, R. Seidel, and M. Sharir. On the zone theorem for hyper-
plane arrangements. SIAM J. Comp., 22(2):418-429, 1993.
[98] H. Edelsbrunner and N. R. Shah. Incremental topological flipping works

for regular triangulations. Algorithmica, 15:223-241, 1996.
[99] H. Edelsbrunner and W. Shi. An O(n log 2 h) time algorithm for the three-
dimensional convex hull problem. SIAM J. Comp., 20:259-277, 1991.
[100] H. Edelsbrunner and T. S. Tan. An upper bound for conforming Delaunay
triangulations. In Proc. 8th Ann. ACM Symp. Comp. Geom., 53-62, 1992.
[101] H. Edelsbrunner and T. S. Tan. A quadratic time algorithm for the minmax
length triangulation. SIAM J. Comp., 22:527-551, 1993.
[102] H. Edelsbrunner, T. S. Tan, and R. Waupotitsch. An 0(n 2 log n) time
algorithm for the minmax angle triangulation. SIAM J. Sci. Statist. Comp.,
13(4):994-1008, 1992.
[103] P. Erd6s, L. LovAsz, A. Simmons, and E. Straus. Dissection graphs of planar
point sets. In J. N. Srivastava, editor, A Survey of Combinatorial Theory,
139-154. North-Holland, Amsterdam, Netherlands, 1973.
[104] H. Everett, J.-M. Robert, and M. van Kreveld. An optimal algorithm for
the (< k)-levels, with applications to separation and transversal problems.
Internat. J. Comp. Geom. Apple , 6:247-261, 1996.
[105] S. Fortune. Voronoi diagrams and Delaunay triangulations. In D.-Z. Du
and F. K. Hwang, editors, Computing in Euclidean Geometry, volume 1 of
Lecture Notes Series on Computing, 193-233. World Scientific, Singapore,
1992.
[106] S. J. Fortune. A sweepline algorithm for Voronoi diagrams. Algorithmica,
2:153-174, 1987.
[107] A. Fournier and D. Y. Montuno. Triangulating simple polygons and equiv-
alent problems. ACM Trans. Graph., 3(2):153-174, 1984.
500 References
[108] C. Froidevaux, M. C. Gaudel, and M. Soria. Type de Donnees et Algo-

rithmes. McGraw-Hill, Paris, 1990.
[109] M. R. Garey, D. S. Johnson, F. P. Preparata, and R. E. Tarjan. Triangu-

lating a simple polygon. Inf. Proc. Lett., 7:175-179, 1978.
[110] P. J. Giblin. Graphs, Surfaces and Homology. Chapman and Hall, London,
1977.
[111] R. L. Graham. An efficient algorithm for determining the convex hull of a

finite planar set. Inf. Proc. Lett., 1:132-133, 1972.
[112] R. L. Graham and F. F. Yao. Finding the convex hull of a simple polygon.
J. Algorithms, 4:324-331, 1983.
[113] P. J. Green and R. R. Sibson. Computing Dirichlet tessellations in the

plane. Comp. J., 21:168-173, 1978.
[114] B. Grfinbaum. Convex Polytopes. Wiley, New York, NY, 1967.
[115] L. J. Guibas, D. Halperin, J. Matousek, and M. Sharir. On vertical decom-

position of arrangements of hyperplanes in four dimensions. In Proc. 5th
Canad. Conf. Comp. Geom., 127-132, Waterloo, Canada, 1993.
[116] L. J. Guibas, J. Hershberger, D. Leven, M. Sharir, and R. E. Tarjan. Linear-

time algorithms for visibility and shortest path problems inside triangulated
simple polygons. Algorithmica, 2:209-233, 1987.
[117] L. J. Guibas, D. E. Knuth, and M. Sharir. Randomized incremental con-

struction of Delaunay and Voronoi diagrams. Algorithmica, 7:381-413, 1992.
[118] L. J. Guibas, M. Sharir, and S. Sifrony. On the general motion planning

problem with two degrees of freedom. Discrete Comp. Geom., 4:491-521,
1989.
[119] D. Halperin and M. Sharir. Near-quadratic bounds for the motion planning
problem for a polygon in a polygonal environment. In Proc. 34th Ann. IEEE
Symp. Found. Comp. Sci. (FOCS 93), 382-391, 1993.
[120] D. Halperin and M. Sharir. New bounds for lower envelopes in three di-
mensions, with applications to visibility in terrains. Discrete Comp. Geom.,
12:313-326, 1994.
[1211 D. Halperin and M. Sharir. Almost tight upper bounds for the single cell
and zone problems in three dimensions. In Proc. 10th Ann. ACM Symp.
Comp. Geom., 11-20, 1994.
References 501
[122] S. Hart and M. Sharir. Nonlinearity of Davenport-Schinzel sequences and

of generalized path compression schemes. Combinatorica, 6:151-177, 1986.
[123] D. Haussler and E. Welzl. Epsilon-nets and simplex range queries. Discrete
Comp. Geom., 2:127-151, 1987.
[124] J. Hershberger. Finding the upper envelope of n line segments in O(n log n)
time. Inf. Proc. Lett., 33:169-174, 1989.
[125] J. Hershberger and S. Suri. Applications of a semi-dynamic convex hull

algorithm. In Proc. 2nd Scand. Workshop Algorithm Theory, volume 447
of Lecture Notes in Computer Science, 380-392. Springer-Verlag, 1990.
[126] J. Hershberger and S. Suri. Offline maintenance of planar configurations.

In Proc. 2nd ACM-SIAM Symp. Discrete Algorithms, 32-41, 1991.
[127] J. Hershberger and S. Suri. Efficient computation of Euclidean shortest

paths in the plane. In Proc. 34th Ann. IEEE Symp. Found. Comp. Sci.
(FOCS 93), 508-517, 1993.
[128] S. Hertel and K. Mehlhorn. Fast triangulation of simple polygons. In Proc.

4th Internat. Conf. Found. Comp. Theory, volume 158 of Lecture Notes in
Computer Science, 207-218. Springer-Verlag, 1983.
[129] H. Imai, M. Iri, and K. Murota. Voronoi diagrams in the Laguerre geometry
and its applications. SIAM J. Comp., 14:93-105, 1985.
[130] R. A. Jarvis. On the identification of the convex hull of a finite set of points
in the plane. Inf. Proc. Lett., 2:18-21, 1973.
[131] B. Joe. Construction of three-dimensional Delaunay triangulations using

local transformations. Comp. Aided Geom. Design, 8(2):123-142, May 1991.
[132] B. Joe. Delaunay versus max-min solid angle triangulations for three-
dimensional mesh generation. Internat. J. Num. Methods Eng., 31(5):987-
997, April 1991.
[133] 0. Kalai. A subexponential randomized simplex algorithm. In Proc. 24th

Ann. ACM Symp. Theory Comp., 475-482, 1992.
[134] N. Karmarkar. A new polynomial-time algorithm for linear programming.

Combinatorica, 4:373-395, 1984.
[135] L. G. Khachiyan. Polynomial algorithm in linear programming. USSR

Comp. Math. and Math. Phys., 20:53-72, 1980.
502 References
[136] D. G. Kirkpatrick. Optimal search in planar subdivisions. SIAM J. Comp.,

12:28-35, 1983.
[137] D. G. Kirkpatrick and R. Seidel. The ultimate planar convex hull algo-
rithm? SIAM J. Comp., 15:287-299, 1986.
[138] V. Klee and G. J. Minty. How good is the simplex algorithm? In 0. Shisha,
editor, Inequalities III, 159-175. Academic Press, 1972.
[139] R. Klein. Concrete and Abstract Voronoi Diagrams, volume 400 of Lecture
Notes in Computer Science. Springer-Verlag, 1989.
[140] R. Klein and A. Lingas. A linear-time randomized algorithm for the
bounded Voronoi diagram of a simple polygon. Int. J. Comp. Geom. Apple,
6:263-278, 1996.
[141] R. Klein, K. Mehihorn, and S. Meiser. Randomized incremental construc-
tion of abstract Voronoi diagrams. Comp. Geom. Theory Appl., 3(3):157-
184, 1993.
[142] D. E. Knuth. Fundamental Algorithms, volume 1 of The Art of Computer
Programming. Addison-Wesley, Reading, MA, 1968.
[143] B. Lacolle, N. Szafran, and P. Valentin. Geometric modelling and algo-
rithms for binary mixtures. Internat. J. Comp. Geom. Appl., 4:243-260,
1994.
[144] C. L. Lawson. Software for C' surface interpolation. In J. R. Rice, editor,
Mathematical Software III, 161-194, New York, NY, 1977. Academic Press.
[145] D. T. Lee. On k-nearest neighbor Voronoi diagrams in the plane. IEEE
Trans. Comp., C-31:478-487, 1982.
[146] D. T. Lee. On finding the convex hull of a simple polygon. Int. J. Comp.
Inform. Sci., 12:87-98, 1983.
[147] D. T. Lee and R. L. Drysdale, III. Generalization of Voronoi diagrams in
the plane. SIAM J. Comp., 10:73-87, 1981.
[148] D. T. Lee and F. P. Preparata. Euclidean shortest paths in the presence of
rectilinear barriers. Networks, 14:393-410, 1984.
[149] D. T. Lee and C. K. Wong. Voronoi diagrams in L, (L,,,) metrics with
2-dimensional storage applications. SIAM J. Comp., 9:200-211, 1980.
[150] J. Matousek. Approximations and optimal geometric divide-and-conquer.
In Proc. 23rd Ann. ACM Symp. Theory Comp., 505-511, 1991. Also to
appear in J. Comp. Syst. Sci.
References 503
[151] J. Matousek. Cutting hyperplane arrangements. Discrete Comp. Geom.,

6:385-406, 1991.
[152] J. Matousek. Efficient partition trees. Discrete Comp. Geom., 8:315-334,

1992.
[1531 J. Matougek. Linear optimization queries. J. Algorithms, 14:432-448, 1993.
[154] J. Matougek. Randomized optimal algorithm for slope selection. Inf. Proc.
Lett., 39:183-187, 1991.
[155] J. Matousek. Reporting points in halfspaces. Comp. Geom. Theory Appl.,

2(3):169-186, 1992.
[156] J. Matousek and 0. Schwarzkopf. Linear optimization queries. In Proc. 8th

[157] J. Matousek, M. Sharir, and E. Welzl. A subexponential bound for linear

programming. In Proc. 8th Ann. ACM Symp. Comp. Geom., 1-8, 1992.
[158] P. McMullen. On zonotopes. Trans. Amer. Math. Soc., 159:91-109, 1971.
[159] P. McMullen and 0. C. Shephard. Convex Polytopes and the Upper Bound
Conjecture. Cambridge University Press, Cambridge, England, 1971.
[160] N. Megiddo. Linear-time algorithms for linear programming in R3 and

related problems. SIAM J. Comp., 12:759-776, 1983.
[161] N. Megiddo. Linear programming in linear time when the dimension is

fixed. JACM, 31:114-127, 1984.
[162] K. Mehlhorn. Multi-dimensional Searching and Computational Geometry,

volume 3 of Data Structures and Algorithms. Springer-Verlag, Heidelberg,
West Germany, 1984.
[163] K. Mehlhorn. Sorting and Searching, volume 1 of Data Structures and

Algorithms. Springer-Verlag, Heidelberg, West Germany, 1984.
[1641 K. Mehlhorn, S. Meiser, and C. O'Dinlaing. On the construction of abstract

Voronoi diagrams. Discrete Comp. Geom., 6:211-224, 1991.
[165] K. Mehlhorn and S. Ndher. Bounded ordered dictionaries in O(log log n)

time and 0(n) space. Inf. Proc. Lett., 35:183-189, 1990.
[166] K. Mehlhorn and S. Ndher. Dynamic fractional cascading. Algorithmica,

5:215-241, 1990.
504 References
[167] K. Mehlhorn, M. Sharir, and E. Welzl. Tail estimates for the space com-
plexity of randomized incremental algorithms. In Proc. 3rd ACM-SIAM
Symp. Discrete Algorithms, 89-93, 1992.
[168] K. Mehlhorn, M. Sharir, and E. Welzl. Tail estimates for the efficiency of
randomized incremental algorithms for line segment intersection. Comp.
Geom. Theory Appl., 3:235-246, 1993.
[169] A. Melkman. On-line construction of the convex hull of a simple polyline.
Inf. Proc. Lett., 25:11-12, 1987.
[170] D. M. Mount and A. Saalfeld. Globally-equiangular triangulations of co-

circular points in O(nlogn) time. In Proc. 4th Ann. ACM Symp. Comp.
Geom., 143-152, 1988.
[171] K. Mulmuley. A fast planar partition algorithm, I. In Proc. 29th Ann.
IEEE Symp. Found. Comp. Sci., 580-589, 1988.
[172] K. Mulmuley. An efficient algorithm for hidden surface removal. Comp.
Graph., 23(3):379-388, 1989.
[173] K. Mulmuley. A fast planar partition algorithm, II. JACM, 38:74-103,
1991.
[174] K. Mulmuley. On levels in arrangements and Voronoi diagrams. Discrete

Comp. Geom., 6:307-338, 1991.
[175] K. Mulmuley. Randomized multidimensional search trees: Dynamic sam-
pling. In Proc. 7th Ann. ACM Symp. Comp. Geom., 121-131, 1991.
[176] K. Mulmuley. Randomized multidimensional search trees: Lazy balancing
and dynamic shuffling. In Proc. 32nd Ann. IEEE Symp. Found. Comp.
Sci., 180-196, 1991.
[177] K. Mulmuley. Computational Geometry: An Introduction Through Ran-
domized Algorithms. Prentice Hall, Englewood Cliffs, NJ, 1994.
[178] K. Mulmuley and S. Sen. Dynamic point location in arrangements of hy-

perplanes. In Proc. 7th Ann. ACM Symp. Comp. Geom., 132-141, 1991.
[179] G. M. Nielson and R. Franke. Surface construction based upon triangula-
tions. In R. Barnhill and W. Boehm, editors, Surfaces in Computer-Aided
Geometric Design, 163-177. North Holland, 1983.
[180] C. O'Dinlaing, M. Sharir, and C. K. Yap. Generalized Voronoi diagrams for
a ladder: I. Topological analysis. Commun. Pure Appl. Math., 39:423-483,
1986.
References 505
[181] C. 6'Difnlaing, M. Sharir, and C. K. Yap. Generalized Voronoi diagrams for

a ladder: II. Efficient construction of the diagram. Algorithmica, 2:27-59,
1987.
[182] A. Okabe, B. Boots, and K. Sugihara. Spatial Tessellations: Concepts and

Applications of Voronoi Diagrams. John Wiley & Sons, Chichester, Eng-
land, 1992.
[183] J. O'Rourke. Art Gallery Theorems and Algorithms. Oxford University

Press, New York, NY, 1987.
[184] M. H. Overmars and J. van Leeuwen. Maintenance of configurations in the

plane. J. Comp. Syst. Sci., 23:166-204, 1981.
[185] J. Pach and M. Sharir. The upper envelope of piecewise linear functions and
the boundary of a region enclosed by convex plates: combinatorial analysis.
Discrete Comp. Geom., 4:291-309, 1989.
[186] J. Pach, W. Steiger, and E. Szemeredi. An upper bound on the number of

planar k-sets. Discrete Comp. Geom., 7:109-123, 1992.
[187] C. H. Papadimitriou and K. Steiglitz. Combinatorial Optimization: Algo-

rithms and Complexity. Prentice Hall, Englewood Cliffs, NJ, 1982.
[188] M. Pocchiola and G. Vegter. Computing the visibility graph via pseudo-
triangulations. In Proc. 11th Ann. ACM Symp. Comp. Geom., 248-257,
1995.
[189] R. Pollack, M. Sharir, and S. Sifrony. Separating two simple polygons by a

sequence of translations. Discrete Comp. Geom., 3:123-136, 1987.
[190] F. P. Preparata. An optimal real-time algorithm for planar convex hulls.

Commun. ACM, 22:402-405, 1979.
[191] F. P. Preparata and S. J. Hong. Convex hulls of finite sets of points in two
and three dimensions. Commun. ACM, 20:87-93, 1977.
[192] F. P. Preparata and M. I. Shamos. Computational Geometry: An Introduc-

tion. Springer-Verlag, New York, NY, 1985.
[193] V. T. Rajan. Optimality of the Delaunay triangulation in Rd. In Proc. 7th

[194] S. Rippa and B. Schiff. Minimum energy triangulations for elliptic prob-
lems. Computer Methods in Applied Mechanics and Engineering, 84:257-
274, 1990.
506 References
[1951 S. Rippa. Minimal roughness property of the Delaunay triangulation.

Comp. Aided Geom. Design, 7:489-497, 1990.
[196] N. Sarnak and R. E. Tarjan. Planar point location using persistent search
trees. Commun. ACM, 29:669-679, 1986.
[197] J. T. Schwartz and M. Sharir. Algorithmic motion planning in robotics. In

J. van Leeuwen, editor, Algorithms and Complexity, volume A of Handbook
of Theoretical Computer Science, 391-430. Elsevier, Amsterdam, 1990.
[198] 0. Schwarzkopf. Dynamic maintenance of geometric structures made easy.

In Proc. 32nd Ann. IEEE Symp. Found. Comp. Sci., 197-206, 1991.
[199] 0. Schwarzkopf. Dynamic Maintenance of Convex Polytopes and Related

Structures. Ph.D. thesis, Fachbereich Mathematik, Freie Universitiat Berlin,
Berlin, Germany, June 1992.
[200] R. Sedgewick. Algorithms. Addison-Wesley, Reading, MA, 1983.
[201] R. Seidel. A convex hull algorithm optimal for point sets in even dimensions.
M.Sc. thesis, Dept. Comp. Sci., Univ. British Columbia, Vancouver, BC,
1981. Report 81/14.
[202] R. Seidel. Constructing higher-dimensional convex hulls at logarithmic cost

per face. In Proc. 18th Ann. ACM Symp. Theory Comp., 404-413, 1986.
[203] R. Seidel. Backwards analysis of randomized geometric algorithms.

Manuscript, ALCOM Summerschool on efficient algorithms design, Arhus,
Denmark, 1991.
[204] R. Seidel. A simple and fast incremental randomized algorithm for com-
puting trapezoidal decompositions and for triangulating polygons. Comp.
Geom. Theory Appl., 1:51-64, 1991.
[205] R. Seidel. Small-dimensional linear programming and convex hulls made

easy. Discrete Comp. Geom., 6:423-434, 1991.
[206] M. Sharir. Improved lower bounds on the length of Davenport-Schinzel

sequences. Combinatorica, 8:117-124, 1988.
[207] M. Sharir and P. K. Agarwal. Davenport-Schinzel Sequences and their

Geometric Applications. Cambridge University Press, New York, 1995.
[208] M. Sharir and E. Welzl. A combinatorial bound for linear programming and
related problems. In Proc. 9th Symp. Theoret. Aspects Comp. Sci., volume
577 of Lecture Notes in Computer Science, 569-579. Springer-Verlag, 1992.
References 507
[209] G. M. Shute, L. L. Deneen, and C. D. Thomborson. An O(nlogn) plane-

sweep algorithm for L1 and Loo Delaunay triangulations. Algorithmica,
6:207-221, 1991.
[2101 J. Stolfi. Oriented Projective Geometry. In Proc. 3rd Ann. ACM Symp.
Comp. Geom., 76-85, 1987.
[211] J. Stolfi. Oriented Projective Geometry: A Frameworkfor Geometric Com-

putations. Academic Press, New York, NY, 1991.
[212] B. Tagansky. A new technique for analyzing substructures in arrangements.

Unpublished manuscript, Tel-Aviv University, 1994.
[213] R. E. Tarjan and C. J. Van Wyk. An O(nloglogn)-time algorithm for

triangulating a simple polygon. SIAM J. Comp., 17:143-178, 1988. Erratum
in 17 (1988), 106.
[214] C. Thomassen. The Jordan-Sch6nflies theorem and the classification of

surfaces. Amer. Math. Monthly, 99:116-130, 1992.
[215] G. T. Toussaint. A historical note on convex hull finding algorithms. Pat-

tern Recog. Lett., 3:21-28, 1985.
[216] P. van Emde Boas, R. Kaas, and E. Zijlstra. Design and implementation
of an efficient priority queue. Math. Syst. Theory, 10:99-127, 1977.
[217] C. A. Wang and L. Schubert. An optimal algorithm for constructing the

Delaunay triangulation of a set of line segments. In Proc. 3rd Ann. ACM
Symp. Comp. Geom., 223-232, 1987.
[218] E. Welzl. Smallest enclosing disks (balls and ellipsoids). In H. Maurer,

editor, New Results and New Trends in Computer Science, volume 555 of
Lecture Notes in Computer Science, 359-370. Springer-Verlag, 1991.
[219] A. Wiernik and M. Sharir. Planar realizations of nonlinear Davenport-

Schinzel sequences by segments. Discrete Comp. Geom., 3:15-47, 1988.
[220] A. C. Yao. A lower bound to finding convex hulls. JACM, 28:780-787,

1981.
Notation
This appendix offers a collection of the notational conventions used in the book.
Apart from a few exceptions, we have tried to abide by the following rules: Lower
case italic letters represent integers and lower case Greek letters represent real
numbers. Upper case letters (whether italic or Greek) represent elementary ge-
ometric objects (points, lines, etc.), and upper case script letters represent sets
thereof, geometric structures, or data structures. Bold upper case letters repre-
sent objects, sets, and structures in projective spaces.
Mathematical symbols
Ed Euclidean space of dimension d

Ed hyperbolic space
11 identity matrix
Ed identity matrix of order d
pd projective space of dimension d
pod oriented projective space
aff() affine hull
conv( convex hull
int( open interior
int( closed interior
ext( exterior
dot product
l lcardinality
Euclidean norm
F1 upper integer part, also called ceiling
LI lower integer part, also called floor
Note that for any integer d, [d/2J + Fd/21 = d and that
L(d + 1)/2= Fd/21.
log* number of iterations of the log function necessary and
sufficient to reach a value smaller than 1
( nk) number of distinct subsets of size k in a set of size n
Notation 509
Notaton 50
Lower case italic

a number of intersecting pairs of a set of segments
b maximum number of objects that define a region
number of distinct intersection points of a set of segments
c integer constant
d dimension of the space
e(C) Euler characteristic of the complex C
f number of faces or facets of a polytope, triangulation,
or arrangement
fo(r, S) expected number of regions defined and without conflict
over a sample of size r of a set S
f3 (r, S) expected number of regions determined by i objects, defined
and with j conflicts over a sample of size r of a set S
9 degree of a vertex
h height of a tree
i index variable, number of objects that define a region
i index variable, number of objects that conflict with a region
k index variable, rank of item in a list, level in an arrangement
index variable, level in an arrangement
m number of edges of a polytope
mk(7Z, S) moment of order k of a subset 7? of S
mk(r, S) expected moment of order k of a random sample
of size r of S
n number of objects, cardinality of a set of objects
nk(C) number of k-faces of a complex C
p probability
Pj,k(r) probability that a region in 4j (S) be a region defined
and with k conflicts over a random sample of size r of S
pj (r) probability that a region in 4(S) be a region defined
and without conflict over a random sample of size r of S
p i probability that a region F in 4.(S) be created
by a randomized incremental algorithm
p' (r) probability that a region F in PFj(S) be created
at step r of a randomized incremental algorithm
r size of a sample, radius of a sphere
S maximum number of alternations of two symbols in a
Davenport-Schinzel sequence
t number of triangles, or of simplices of maximal dimension
in a triangulation
x, y, z index variables for the coordinates in E2 or E3
(xi, X2, ...* Xd) real coordinates of a point in Ed
510 Notation
510 Notation
Upper case italic

A vertex, point, or vector
AB segment, vector
A 1 A 2 ... Ad+l d-simplex
[AlA 2 . Ad+1 determinant
C vertex, point, vector, center of a sphere
cylindrical cell, cell in an arrangement
D line
E edge
F region, face (generally of dimension d - 1)
G region, face (generally of dimension d - 2)
H hyperplane
H+,H- open half-spaces bounded by H
H+,H- closed half-spaces bounded by H
H* pole of a hyperplane H
K hyperplane, face of dimension d - 3
L line
M,N vertex, point, vector
0 object, origin
OA point or vector
O() asymptotic upper bound
P,Q,R vertex, point, vector
P* hyperplane polar to a point P
p# half-space bounded by the hyperplane polar to a point P
S object, segment, simplex
T triangle, simplex of maximal dimension in a complex
U,VIW vertex, point, vector
V(M) Voronoi cell of a point M.
Vk(Mk) Voronoi cell of order k of a subset Mk
X,Y,Z vertex, point, vector
X(x1, X 2 , * **, Xd) point or vector in Ed
X(xTlX 2, ** Xd+1) point or vector in Ed+1
Upper case script

A arrangement
A(X-) arrangement of a set of hyperplanes XH
Ak sub-complex of A consisting of all cells
at level k and of their subfaces
A<k sub-complex of A consisting of all the faces
at level at most k
Notation 511
Notation 511
B quadric
C polytope, complex
D dictionary
TD(M) the Delaunay polytope, dual to V(M)
Del(M) the Delaunay complex
Dec(S) vertical decomposition of S
Dec8 (S) simplified vertical decomposition
£ set of edges, envelope
F pencil of spheres
F(S) set of regions defined over a set S of objects
Fo(7Z) set of regions defined and without conflict
over a set S of objects
4j(S) set of regions defined by i objects and with j conflicts
FOk(S) set of regions defined and with at most k conflicts
g graph
horizon graph
set of hyperplanes
set of indices
list, lower envelope
M set of points
Md moment curve in Ed
polytope, polyhedron, paraboloid
polytope dual to a polytope P
Pow(S) power diagram
Q priority queue, quadric
sample of size r
S set of objects, of segments, of sites, of spheres
T triangulation
T (M) triangulation of a set M of points
U universe
V(M) Voronoi polytope of a set M
Vor (M) Voronoi diagram of a set M
Vork (M) Voronoi diagram of order k of a set M
Vor+(M ) Voronoi diagram with additive weights
Vor* (M) Voronoi diagram with multiplicative weights
VorL1 (M) Voronoi diagram for the L1 norm
VorLc,O (M) Voronoi diagram for the Lx norm
512 Notation
Lower case Greek

o(n) Ackermann function
6+ (X, Mi) distance with additive weights
6* (X, Mi) distance with multiplicative weights
61 (X, Mi) L1 distance
6 (XI MO) Lo<> distance
6Q(X, Mi) generalized quadratic distance with respect to Q
E real constant that can be made arbitrarily small
A, (n) maximal length of a Davenport-Schinzel sequence
p radius of a sphere
a0 power of the origin with respect to the sphere E
0 bijective mapping from spheres in Ed onto Ed+1
Upper case Greek

r curved arc
A sweep line
AQ matrix of the quadric Q in homogeneous coordinates
0( ) asymptotic equivalent
E chronological sequence, sphere
E (C, r) sphere of center C and radius r
E(X) power of a point X with respect to a sphere E
Q( ) asymptotic lower bound
Bold upper case

H projective hyperplane
P projective point
-1P point antipodal to P in an oriented projective space
P projective polytope
-P# polar transform of a projective polytope P
Index
k-set, 334, 348 analysis,randomized, 6, 65, 77, 101,

k-th element (selection of), 44 185, 194
annulus of minimum area, 237
accelerated algorithm, 84, 477 arrangement, of triangles, 373
acceptable line, 228 arrangement,convex hull of, 347
Ackermann's function, 7, 354 arrangement,of hyperplanes, 321, 353
adjacency, 135, 244 arrangement,of segments, 352
adjacency,graph, 170, 196, 259 arrangement,projective, 343
admissible path, 480 arrangement,simple, 321
affine,combination, 128 arrangement,zone, 325, 347
affine,diagram, 437 art gallery problem, 286
affine,hull, 128
affine,independence, 128 backward analysis, 57, 89
algorithm, accelerated, 477 ball,intersection, 429
algorithm,accelerated, 84 ball,topological, 245
algorithm,adaptive, 5 ball,union, 429
algorithm,divide-and-conquer, 33, 201, biregion, 90, 103, 196
358, 370, 402, 488 bisection, 237
algorithm,dynamic, 95, 186, 219, 430, bisector, of line segments, 466
489 bisector,L 1 , 446
algorithm,incremental, 176, 264, 296, bisector,hyperbolic, 453, 457
330, 331, 345 bisector,of line segments, 489
algorithm,lazy, 123 bisector,perpendicular, 407
algorithm,off-line, 65 bitangent, 203, 219
algorithm,on-line, 75, 335, 362, 400, boundary,non-trivial, 369, 370
430, 471 boundary,of a complex, 244
algorithm,output-sensitive, 37, 237
algorithm,randomized, 63 canonical triangulation, 165, 345
algorithm,semi-dynamic, 63 Carath6odory's theorem, 162
algorithm,sweep, 36, 470 cell,complex, 245, 321, 407, 437
alternating sequence, 353 cell,of an arrangement, 358, 390, 403
amortized analysis, 6 co-cube, 445
analysis,amortized, 6 complex, 243
analysis,backward, 57, 89 complex,Delaunay, 416
514 Index
complex,homogeneous, 244 decomposition,of a polygon, 87, 91,

complex,of cells, 245, 321, 407, 437 274
complex,of simplices, 245 decomposition,of a polyhedral region,
complex,pure, 244 302
complex,regular, 455 decomposition,of line segments, 40,
complexity, average-case, 4 69, 81, 110
complexity,output-sensitive, 5 decomposition,of line segments,complete,
complexity,worst-case, 4 40, 89
conflict, 46 decomposition,of line segments,dynamic,
conflict,graph, 65 121
conflict,list, 70 decomposition,of line segments,lazy,
conformal triangulation, 490 123
constrained triangulation, 267, 300, decomposition,of line segments,simplified,
489 42
convex hull, 125, 128 decomposition,of surfaces, 402
convex hull,divide-and-conquer, 205 decomposition,of triangles, 374, 394,
convex hull,dynamic, 186, 219 402
convex hull,incremental, 176 decomposition,simplicial, 307
convex hull,lower, 167 Dehn-Sommerville,relations, 144, 163,
convex hull,of a polygon, 214 164, 344
convex hull,of an arrangement, 347 Delaunay,complex, 416
convex hull,of objects, 369 Delaunay,triangulation, 418
convex hull,of two 3-polytopes, 206 Delaunay,triangulation,flipping, 423,
convex hull,on-line, 180 430
convex hull,output-sensitive, 229, 237 determinant, 155, 165
convex hull,upper, 167 determinant,of a region, 97, 472
convex,edge, 307 determinant,Van der Monde, 147
convex,function, 57 diagram,affine, 437
convex,set, 128 diagram,for the L1 metric, 456
convex,vertex, 270, 307 diagram,for the Loo metric, 456
curve, 43, 89, 371 diagram,higher order power diagram,
curve,of moments, 147, 257 436
cyclic polytope, 147 diagram,power, 434, 437
diagram,simple, 437
Davenport-Schinzel sequences, 353 diagram,weighted, 439, 442, 455
decision tree, 10 diameter, 220
decomposition,geodesic, 286 dictionary, 16, 25
decomposition,hierarchical, 286 dictionary,augmented, 16, 25
decomposition,in convex parts, 281, dictionary,persistent, 30, 346
377 disk, 261
decomposition,of a lower envelope of disk,motion planning, 479
triangles, 383 distance,L 1, 445
Index 515
Index 515
distance,LOO, 448 extreme points, 195, 227

distance,in hyperbolic space, 450
distance,point to segment, 464 face,inner, 390
distance,polygonal, 488 face,of a complex, 243
distance,polyhedral, 457 face,of a polytope, 128, 129
distance,quadratic, 438 face,of a Voronoi diagram, 439, 446,
divide-and-conquer, 33, 402 448, 452, 464
divide-and-conquer,a single cell, 370 face,of an arrangement, 344, 353, 374
divide-and-conquer,convex hull, 201, face,outer, 390
205 face,popular, 390
divide-and-conquer,lower envelope of face,proper, 129
functions, 358 face,singular, 245
divide-and-conquer,lower envelope of facet of a polytope, 129
triangles, 402 feasible domain, 224
divide-and-conquer,union of polygons, function,Ackermann, 7, 354
function,convex, 57
43
divide-and-conquer,Voronoi diagram, function,hash function, 29
488 general position,Ll-general, 456
domain, feasible, 224 general position,L 2 -general, 408, 464
domain,of a complex, 244 general position,Lx-general, 448, 456
domain,of influence, 47 general position,for hyperplanes, 140,
duality, 136, 259 321
duality,hyperbolic diagram, 457 general position,for points, 140
dynamic,algorithm, 95 general position,for spheres, 435
dynamic,convex hull, 186, 219 geodesic decomposition, 286
dynamic,decomposition of segments, gift-wrapping, 206
121 graph, 16
dynamic,perfect hashing, 25, 29 graph,of adjacencies, 170, 196, 259
dynamic,Voronoi diagram, 430, 489 graph,of conflicts, 65
graph,of incidences, 170, 259
edge,convex, 307 graph,of influence, 75
edge,of a complex, 243 graph,of influence (augmented), 97
edge,of a polytope, 129 greatest empty rectangle, 454
edge,reflex, 302, 307
EMST (Euclidean minimum spanning hashing,hash functions, 29
tree), 431 hashing,perfect dynamic, 25, 29
Euler,characteristic, 250, 262 Helly's theorem, 162
Euler,relation, 141, 164, 167, 250, 261, hierarchical,decomposition, 286
262 hierarchical,representation, 220, 282
event queue, 36 history, 93
exterior,of a polygon, 248 homogeneous complex, 244
exterior,of a polyhedron, 249 hull,affine, 128
516 Index
hull,convex, 125, 128 level in arrangements,of triangles, 403

hull,lower, 167 level in arrangements,surface of, 334
hull,upper, 167 linear programming, 223
hyperbolic, space, 457 list, 14
hyperbolic,bisector, 457 localization in a planar map, 282
hyperbolic,duality, 457 lower envelope, of functions, 355
hyperbolic,space, 450 lower envelope,of cones, 455
hyperbolic,Voronoi diagram, 450, 489 lower envelope,of half-lines, 358
hyperplane,arrangement, 321 lower envelope,of line segments, 357,
hyperplane,radical, 414 368
hyperplane,supporting, 129, 163 lower envelope,of simplices, 402
lower envelope,of surfaces, 402
incidence, 135, 244 lower envelope,of triangles, 379
incidence,graph, 170, 259, 261 lower hull, 167
incremental,arrangement, 330, 331, 345 LP-type problems, 238
incremental,convex hull, 176
incremental,triangulation, 264, 296 mapping,projective, 151
independence,affine, 128 maximal polytope, 165
influence,domain, 47 maximum inscribed sphere, 238
influence,graph, 75 minimum enclosing circle, 238, 429
influence,graph (augmented), 97 Minkowski sum, 167
inner face, 390 moment, 55
inner vertex, 60 moment,curve, 147, 257
interior,of a polygon, 248 moment,theorem, 55
interior,of a polyhedron, 249 monotone polygon, 269
interior,of a polytope, 128 motion planning,of a disk, 479
interior,of a sphere, 408 motion planning,of a manipulator, 371
intersection,miscellaneous, 221 motion planning,of a polygon, 369,
intersection,of balls, 429 489
intersection,of convex polygons, 221 motion planning,of a polyhedron, 403,
intersection,of half-spaces, 167, 196, 404
237
intersection,of line segments, 37 nearest neighbor, 428, 429
intersection,of polygonal regions, 43 nearest neighbor,on-line, 428
nearest neighbor,queries, 428
Jordan's theorem, 260 non-local subset of regions, 58
non-trivial boundary, 369, 370
lazy,decomposition of segments, 123
lazy,sweeping, 279 object, 46
level in arrangements,of hyperplanes, off-line,algorithm, 65
333, 349, 350 on-line,< k-levels, 335
level in arrangements,of lines, 348 on-line,a single cell, 362, 400
level in arrangements,of simplices, 403 on-line,algorithm, 75
Index 517
on-line,convex hull, 180 polygonal,distance, 488

on-line,decomposition of line segments, polygonal,line, 248
81 polygonal,region, 43, 248
on-line,Delaunay triangulation, 430 polyhedral,distance,, 457
on-line,Voronoi diagram, 416, 430, 471 polyhedral,region, 250
onion peeling, 220 polyhedron, 248, 249
orientation,clockwise and counter-clockwise, polytope, 127, 128
199, 200 polytope,cyclic, 147
orientation,direct and indirect, 199, polytope,maximal, 165
200 polytope,projective, 158
orientation,in a projective space, 154 polytope,representation, 170
orientation,of a 2-triangulation, 249 polytope,simple, 140
orthogonal spheres, 412 polytope,simplicial, 140
outer face, 390 polytope,unbounded, 161, 167
outer vertex, 60 polytope,Voronoi, 414
output-sensitive,complexity, 5 popular face, 390
output-sensitive,convex hull, 229, 237 power diagram, 434, 437
output-sensitive,segment intersections, power diagram,of higher orders, 436
37 power of a point, 408, 434
output-sensitive,triangulation, 300 priority queue, 16, 25
projective,arrangement, 343
paraboloid, 166, 410 projective,mapping, 151
pencil of spheres, 449 projective,oriented space, 153
perpendicular bisector, 407 projective,polarity, 151
persistent dictionary, 30, 346 projective,polytope, 158
planar map, 261 projective,simplex, 156
Poincar6, disk, 457 projective,space, 148
Poincar6,half-space, 450 proper face, 129
polarity,image, 136 prune-and-search, 238
polarity,polytope, 137 pure complex, 244
polarity,projective, 151
polarity,transformation, 135 queue, 14
polarity,with respect to a paraboloid, queue,event, 36
166, 411 queue,priority, 16, 25
polygon, 248
polygon,convex hull of, 214 radical hyperplane, 414
polygon,decomposition of, 87, 91 Radon's theorem, 162
polygon,intersection, 43 random sampling, 49
polygon,monotone, 269 randomized,algorithm, 63
polygon,star-shaped, 283 randomized,analysis, 6, 65, 77, 101,
polygon,union, 43 185, 194
polygon,Voronoi diagram of, 477 range tree, 27
518 Index
518 Index
ray shooting, 236, 287 space,projective, 148

rectangle (greatest empty), 454 spanning tree, 431
reflex,edge, 302, 307 sphere,orthogonal, 412
reflex,vertex, 270, 307 sphere,representation of, 409
region,determined by some objects, sphere,topological, 245
46 split theorem, 293
region,polygonal, 248 stabbing,lines, 369
region,polyhedral, 250 stabbing,planes, 403
relation,Dehn-Sommerville, 144, 344 stack, 14
relation,Euler, 141, 164, 167 star of a face, 246
representation,hierarchical, 220, 282 Steiner points, 302
representation,of a complex, 258 stratified tree, 28
representation,of a polytope, 170, 199 supporting hyperplane, 129, 163
representation,of spheres, 409 surface, 402
surface,of level, 334
sampling,random, 49 sweep, 36, 459
sampling,theorem, 52 sweep,lazy, 279
segment tree, 26 sweep,non-trivial boundary, 370
selection, 44 sweep,polygon, 276, 283
semi-dynamic algorithm, 63 sweep,segment intersection, 37
separability, 236 sweep,Voronoi diagram, 470
sequence,alternating, 353 sweep,Voronoi diagrams, 459
sequence,Davenport-Schinzel, 353
shell of a face, 246 tail estimates, 59
shelling of a polytope, 227 theorem,augmented influence graph,
shortest path, 283, 348 107
simple,arrangement, 321 theorem,Caratheodory, 162
simple,diagram, 437 theorem,conflict graph, 67
simple,polytope, 140 theorem,Helly, 162
simplex, 128, 140, 163, 243 theorem,influence graph, 78
simplex,projective, 156 theorem,Jordan, 260
simplicial,complex, 245 theorem,moment, 55
simplicial,decomposition, 307 theorem,Radon, 162
simplicial,polytope, 140 theorem,sampling, 52
singular face, 245 theorem,split, 293
skeleton,of a complex, 244 theorem,upper bound, 145, 164, 166
skeleton,of a polytope, 170, 196 transformation,duality, 136
skeleton,of an arrangement, 344 transformation,hyperbolic duality, 457
sorting, 10, 24 transformation,of problems, 11
space, hyperbolic, 457 transformation,polarity, 135, 411
space,hyperbolic, 450 traversal of a graph, 17
space,oriented projective, 153 tree, 16
Index 519
tree,2-3-4, 18 vertical decomposition,of line segments,

tree,balanced, 16 40
tree,decision, 10 visibility, 285
tree,range, 27 visibility,art gallery, 286
tree,red-black, 17 visibility,computing a view, 368, 402
tree,segment, 26 visibility,graph, 348
tree,stratified, 28 visibility,ray shooting, 236, 287, 346
triangulation, 247, 263, 289 Voronoi diagram, of a polygon, 477
triangulation, quadratic, 262 Voronoi diagram,dynamic, 430
triangulation,canonical, 165, 345 Voronoi diagram,for the L1 distance,
triangulation,conformal, 490 445
triangulation,constrained, 267, 300, Voronoi diagram,for the L,, distance,
489 445
triangulation,Delaunay, 418 Voronoi diagram,furthest-point, 428
triangulation,linear, 291 Voronoi diagram,higher-order, 425
triangulation,of a polygon, 274 Voronoi diagram,hyperbolic, 450, 489
triangulation,optimality, 421 Voronoi diagram,of disks, 455
triangulation,output-sensitive, 300 Voronoi diagram,of higher order, 431
triangulation,quadratic, 291 Voronoi diagram,of line segments, 464,
489
unbounded polytope, 161, 167
Voronoi diagram,of points, 407, 414
union,miscellaneous, 221
Voronoi diagram,on-line, 471
union,of balls, 429
Voronoi diagram,sweeping, 459, 470
union,of parallel hypercubes, 60
Voronoi polytope, 414
union,of parallel rectangles, 44
union,of polygonal regions, 43 weighted diagram, 439, 442
update condition,augmented influence
graph, 106 zone in an arrangement, 325, 347
update condition,conflict graphs, 66 zonotope, 167, 345, 346
update condition,influence graph, 78
upper bound theorem, 145, 164, 166
upper hull, 167
vertex,convex, 270, 307
vertex,inner, 60
vertex,of a complex, 243
vertex,of a polytope, 129
vertex,outer, 60
vertex,reflex, 270, 307
vertical decomposition,of a polygon,
274
vertical decomposition,of a polyhedral
region, 302
In all areas of computing the optimal design of algorithms
is essential forthe efficient running of programs. The design
and analysis of geometric algorithms has, in particular, seen
remarkable growth in recentyears, due totheir application
in, for example, computer vision, graphics, medical imaging
and CAD. The subject itself is built on three pillars: geometric
data structures, algorithmictechniques and results from
combinatorial geometry.
The goal of this book istwofold: first to provide a coherent

and systematic treatment of the foundations; second to
present algorithmic solutions that are amenable to rigorous
analysis and are efficient in practical situations. As far as
possible, the algorithms are presented in their most general
d-dimensional setting. Specific developments are given for
the 2-or 3-dimensional cases when this results in significant
improvements. The presentation is confined to Euclidean
affine geometry, though the authors indicate wheneverthe
treatment can be extended to curves and surfaces. The pre-
requisites for usingthe book are few which make it ideal for
teaching advanced undergraduate or beginning graduate
courses in computational geometry.
The illustrations, P-486-1( from

the series Laserglyphs, used on the
cover ofthis book are reproduced
with the permission of the artist,
/ Manfred Mohr.
CO-,AMBRIDGE
UNIVERSITY PRESS
ISBN 0-521-56529-4
91111111111111

Algorithmic Geometry

Uploaded by

Copyright:

Available Formats

Algorithmic Geometry

Uploaded by

Document Information

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

Algorithmic Geometry

Uploaded by

Copyright:

Available Formats

J-D.

Translated by Herv6 Bronnimann

First published in French by Edisciences, 1995 as Geom6trie Algorithmique

English edition 0 Cambridge University Press 1998

This book is in copyright. Subject to statutory exception

First published in English 1998

Printed in the United Kingdom at the University Press, Cambridge

ISBN 0 521 56322 4 hardback

Translator's preface xix

Part I - Algorithmic tools 1

Chapter 1 - Notions of complexity 3

Chapter 2 - Basic data structures 13

Chapter 3 - Deterministic methods used in geometry 32

3.1.2 An example: sorting n numbers using merge-sort ......... 35

Chapter 4 - Random sampling 46

Chapter 5 - Randomized algorithms 63

Chapter 6 - Dynamic randomized algorithms 95

Part II - Convex hulls 125

Chapter 7 - Polytopes 127

Chapter 8 -Incremental convex hulls 169

Chapter 9 - Convex hulls in two and three dimensions 198

Chapter 10 - Linear programming 223

Part III - Triangulations 241

Chapter 11 - Complexes and triangulations 243

Chapter 12 - Triangulations in dimension 2 263

Chapter 13 - Triangulations in dimension 3 289

Part IV - Arrangements 319

Chapter 14 - Arrangements of hyperplanes 321

14.2 Combinatorial properties ................................... .. 322

Chapter 15 - Arrangements of line segments in the plane 352

Chapter 16 - Arrangements of triangles 373

Part V - Voronoi diagrams 405

Chapter 17 - Euclidean metric 407

17.2 Voronoi diagrams and polytopes ........................... 408

Chapter 18 - Non-Euclidean metrics 433

Chapter 19 - Diagrams in the plane 459

19.3.2 The two planes are not parallel ...................... 485

Contents and layout of this book

Goals and limits of this book

of geometric data structures for queries and multidimensional searching only in

From algorithms to programs

How to read this book

the unalienable rights of the reader, as defined by Daniel Pennac. 1 In particular,

A translation is an excellent opportunity to enhance the text with added refer-

Computational geometry aims at designing the most efficient algorithms to solve

1.1 The complexity of algorithms

1.1.2 Notions of complexity

Worst-case and average-case complexity

Each instance of a problem (be it geometric or not) is specified by a set of data

often the average-case complexity, or complexity on the average. This is a func-

The complexity of randomized algorithms

Preprocessing, queries, amortized analysis

1.1.3 Asymptotic behavior, notation

log* n = i{•> log( ')n > 1

is thus an extremely slow-growing function. In fact, ae(n) is at most 4 for all

* By f(n) = O(g(n)), to be read "f(n) is a big-oh of g(n)," we express the

Vn > no, f(n) < cg(n).

* By f(n) = Q(g(n)), to be read "f(n) is a big-omega of g(n)," we express

Vn > no, f(n) > cg(n).

* By f(n) = E3(g(n)), to be read "f(n) is a big-theta of g(n)," we express