Theory and Application of Density Matrix

Home
Search
Collections
Journals
About
Contact us
My IOPscience
Theory and applications of the density matrix
This content has been downloaded from IOPscience. Please scroll down to see the full text.
1961 Rep. Prog. Phys. 24 304
(http://iopscience.iop.org/0034-4885/24/1/307)
View the table of contents for this issue, or go to the journal homepage for more
Download details:
IP Address: 212.201.44.245
This content was downloaded on 06/04/2015 at 09:05
Please note that terms and conditions apply.
THEORY AND APPLICATIONS OF THE

DENSITY MATRIX
BY D. TER HAAR
T h e Clarendon Laboratory, Oxford
CONTENTS
3 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
General properties of the density matrix

Green function techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The description of statistical equilibrium by the density matrix. . .
Density matrix techniques applied to atoms, molecules and nuclei
The density matrix in solid state physics . . . . . . . . . . . . . . . . . . . . .
Non-equilibrium processes ; transport theory. . . . . . . . . . . . . . . . . . . . . . .
Polarization, scattering, and angular correlation experiments
Resonance and relaxation phenomena . . . . . . . . . . . . . . . . . . .
10. Thetheoryofmeasurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Concluding remarks and acknowledgments
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.
3.
4.
5.
6.
7.
8.
9.
PAGE
304
312
3 19
337
353
356
356
Abstract After a qualitative discussion of the advantages of the density matrix and of the
different ways to introduce it (the statistical, quantum mechanical and operational methods
of approach), $ 2 deals with the general properties of the density matrix, including a discussion
of pure cases and mixtures. A brief discussion is given of Green function techniques and of
the relation between Green functions and correlation functions. A discussion of recent
developments in the evaluation of partition functions concludes the first part of this article
dealing with the theory of density matrix techniques. Sections 5 to 9 discuss applications.
The first application is the quantum-chemical one to many-body systems in their ground
state, that is, systems at absolute zero, and it is shown how the density matrix fits into the
Hartree-Fock and Thomas-Fermi schemes. A brief discussion is given of the theory of
diamagnetism. This is followed by a discussion of non-equilibrium processes and of Kubos
approach to transport theory. After that the polarization of beams of electrons or of photons
is discussed and it is indicated how density matrix techniques can be used to treat scattering
processes. Section 9 concludes this part of the paper by a brief account of density matrix
theory applications to resonance and relaxation phenomena. Finally, the theory of measurement in quantum mechanics is considered.
3 1. I N T R O D U C T I O N
recent rapid growth of physics is producing not only an ever-growing
spate of papers-often
repeating what has been published independently
elsewhere-but
also an increasing degree of specialization. Different
branches of physics often use different terms for the same concept, and it takes
a long time before new techniques filter through from one branch of physics to
another. This is a great pity since in theoretical physics, for instance, techniques
which are useful in one field are often also applicable to other fields. Two such
techniques will be discussed in the present paper. T h e first-and main-one is
the use of the so-called density matrix which was introduced by von Neumann
(1927 a, b, c ; see also Dirac 1929, 1935, and von Neumann 1932), and was until
recently used mainly in statistical mechanics. T h e other one is the use of the
so-called Green functions which were originally used in field theory and which
HE
Theory and Applications of the Density Matyix
305
have recently been applied with great success to a great number of many-body
problems (for an extensive survey of Green function techniques applied to statistical
problems we refer to a paper by Zubarev (1960) ). I n this introduction we shall
first of all discuss why the density matrix was introduced and how its use fits in
easily with a description of different kinds of experiments. We shall then give
a brief outline of the subsequent sections. T h e aim of this introduction is that
anybody who wants to get acquainted with the use of the density matrix in various
branches of physics without being interested in details should get sufficient
information from the introduction. Those readers who wish to apply density
matrix techniques themselves and would thus be interested in a more detailed
account are referred to the sections dealing with the various applications. There
is also an extensive list of references, both of those quoted in the text and of some
related papers which are not referred to explicitly.
T h e density matrix was introduced by von Neumann to describe statistical
concepts in quantum mechanics. T o see how this can be done we must discuss
the fundamental ideas both of quantum mechanics and of statistical mechanics.
Most physical systems consist of so many particles, or possess so many degrees
of freedom, that it is impossible to specify completely the state of these systems.
Indeed, physicists are forced to make predictions about the behaviour of the
systems they study from a knowledge of a very small number of parameters. That
this is possible, and that those predictions in general will be in accordance with the
actual behaviour of physical systems, is because one can use statistical methods and
introduce so-called representative ensembles (see, for instance, Tolman 1938, ter Haar
1954, 1955, for a discussion of this point of view). Such representative ensembles
are collections of identical systems such that the values of those parameters which
are known are the same for all systems in the ensemble while the other parameters
will have values which are spread over a suitably chosen range for the different
systems. T h e methods by which such representative ensembles are chosen are
discussed in statistical mechanics, and we shall briefly return to possible choices
and the reasons why they are made in 5 4. Their success arises because it turns
out that the average value of any physical quantity, the average being taken in the
appropriate manner over the ensemble, will be equal to the actual value of that
quantity for the physical system under observation-at least in as far as we may
neglect fluctuations. This statistical aspect of physical systems is common to
systems described by classical mechanics and those described by quantum mechanics
and is due to our lack of knowledge about the details of the physical systems
observed in nature.
I n quantum mechanics there occurs another statistical aspect as the use of
wave functions itself introduces probability predictions : the value of a physical
quantity will, in general, not be well defined for a physical system in an arbitrary
state, but be subject to a probability distribution. This statistical aspect remains
even when we have as complete a knowledge about the physical system as is possible,
that is, even when we know its wave function. I n the classical case, complete
knowledge would entail knowledge of all parameters describing the system and all
physical quantities would thus have well-defined values. We see that statistical
methods appear thus much more naturally in quantum mechanics than in classical
physics. It must be emphasized, however, that for a quantum mechanical system
PO
306
D.ter Haar
usually both aspects will occur. In that case it is necessary to find the quantum
mechanical analogue to the classical ensemble density. As with all physical
quantities in quantum mechanics, this quantum mechanical ensemble density will
be an operator. I t is most easily defined by its matrix elements in some appropriately chosen representation and is known as the density matrix. From our
discussion it is clear that in quantum mechanics we would have this operator, even
if we had maximum knowledge about the system considered. Such a case is called
apure case, and the more general case a mixture. T h e reasons for these terms will
become clear presently. T h e density matrix can be used in the same way as the
classical ensemble density to define the average values of any physical quantity,
that is, one uses the normalized density matrix to construct averages. In the
classical case these averages were just ordinary weighted means, while in quantum
mechanics they are defined as the trace of the product of the density matrix and
the operator corresponding to the physical quantity considered. I n the classical case
the ensemble density is determined by the number of systems which have values
of the parameters defining the state of a system within given intervals. Similarly,
the density matrix will be determined by stating how many systems there are in
the quantum mechanical ensemble with given wave functions. T h e ensemble can
thus in general be considered to be a superposition, or mixture, of the states
corresponding to all the different wave functions represented in it. If all systems
correspond to the same wave function, that is, if we know all that can be known
about the system under consideration, there is no superposition, and we talk about
a pure case.
Before discussing in general terms the properties and uses of the density matrix
we must mention a different way of introducing the density matrix, and one where
perhaps the relation between the density matrix and the lack of detailed knowledge
is even more evident. I t is of interest to note that Dirac in his first paper (1929 ;
see also Dirac 1930 a) on the density matrix, where he uses the approach we have
just described, never introduces the actual term density matrix. He uses this
term, however, in two subsequent papers (Dirac 1930 b, 1931) where he approaches
the density matrix from a slightly different angle, namely the one we are about to
describe. I t should be mentioned that he does not discuss in these papers the
connection between the two kinds of approach and that the two kinds of density
matrix-both denoted by p-are normalized in a different way ; we shall return
to this question of normalization which has led to some rather misleading statements
in the literature (compare ter Haar 1960). I t is also interesting to note that
Landau and Lifshitz in their Quantum Mechanics (1958 a) use the second kind
of approach, but in their Statistical Physics (1958 b) the first kind. T h e recent
upsurge of density matrix techniques in quantum chemistry (see, for instance, the
April 1960 issue of Reviews of Modern Physics, and especially Golden 1960 a,
Lowdin 1960, McWeeny 1960) which will be discussed in 5 5 has used practically
exclusively the second kind of approach. We discussed earlier that in statistical
mechanisms one introduces the density matrix essentially by defining the probability
that the physical system under consideration is described by a given wave function.
T h e situation is then characterized by the superposition, or mixture, of a large
number of different wave functions. In quantum mechanics, however, one likes
to work with closed systems, that is, systems which are described by a wave function,
Theovy and Applications of the Density Matrix
307
the change in which is governed by a Hamiltonian. If we were interested in the

complete and detailed behaviour of such a system, we should be dealing with a
pure case and there would be no grounds for introducing the density matrix, as
there would be involved only one wave function and this wave function could be
used to determine the values of all physical quantities connected with the system.
In general, physical systems will possess many degrees of freedom and we are
often interested in the behaviour of only a few of them. T o fix our ideas let us
consider one of the basic problems of quantum chemistry : a many-electron system,
the total energy of which we should like to evaluate. We can usually assume that
our system-which may be an atom or a molecule-will be in a well-defined state,
normally the ground state. If we knew the exact form of the wave function we
could, of course, evaluate the energy using the known wave function and the
Hamiltonian. As this is practically never the case we must have recourse to
approximations. We could try to find approximations to the wave function, but
we can also use the fact that the Hamiltonian of a complicated atom or molecule
will contain only terms which involve the coordinates of either one or at most two
electrons : the kinetic energy terms and those potential energy terms referring
to nucleus-electron Coulomb energies involve one electron only, while the electronelectron interaction is a two-body Coulomb term. I n evaluating the energy we
are thus evaluating averages involving at most two electrons at the same time.
It is therefore tempting to introduce suitable averages of the many-body wave
function or rather of the product of this wave function with its adjoint over all
electrons but two. Such an average will be called a reduced density matrix (Husimi
1940). One can prove that, indeed, these reduced one-electron and two-electron
density matrices determine completely the energy of the system. Instead of looking
for approximate expressions for the total many-body wave function, one can thus
look for approximate expressions for the reduced density matrices.
We can generalize this approach to the density matrix as follows. Whenever
we are dealing with a large system which itself is described by a wave function, we
can by averaging over all degrees of freedom which are of no interest to us obtain
a density matrix describing the average behaviour of those degrees of freedom over
which we have not averaged. Of course, if the large system can be split into two
non-interacting parts, the one corresponding to the degrees of freedom over which
we average and the other corresponding to the other degrees of freedom, and if
the original state of the system is described by a product wave function, the factors
of which correspond to the non-interacting parts, we do not need a density matrix
to describe the sub-system in which we are interested, as it possesses a wave function
of its own. From this point of view we can say that the most general description
of a quantum mechanical system is by a density matrix and that only in very
favourable conditions is it possible to use a wave function. T h e extensive use of
wave functions in theoretical physics is due to the fact that the description by
means of a wave function is often a much simpler one.
Let us now consider the connection between the two ways of introducing density
matrices, and to fix our ideas let us again consider the one-electron density matrix
which we met a moment ago. I n the second method of introducing the density
matrix, it entered simply by averaging the product of the many-body wave function
and its adjoint over all coordinates except those of one electron. I n the first
308
D.ter Haar
method of approach we would have asked ourselves how a specially chosen electron
could be described (we must emphasize that although the indistinguishability of the
electrons plays an important role in determining the one-electron density matrix,
or the many-body wave function, it is irrelevant to our present discussion). This
would have meant that we should have constructed a representative ensemble.
This ensemble would have been constructed so as to be in accordance with our
knowledge about the electron, namely, the knowledge that the electron was part
of an atom in a particular quantum state. The density matrix would therefore
have to be constructed in such a way that the probability density function for the
electron-which in fact is the diagonal element of the density matrix in coordinate
representation as we shall see in the following-would be the same as the one
following from the many-electron wave function of the atom, that is, the average
of the absolute square of this wave function over all coordinates but those of the
one electron under consideration. This means that, indeed, in coordinate
representation the diagonal elements of the density matrix are uniquely defined.
One can prove similarly, as we shall do in the next section, that this uniqueness
extends also to the non-diagonal elements so that the density matrices defined by
the two different methods are, indeed, the same. This is less surprising if we
remember that a popular way of describing an ensemble, especially in classical
statistical mechanics, is by assuming the different systems in the ensemble together
to form a large closed system in which the original systems are interacting (see, for
instance, Schrodinger 1948). T h e electrons in the atom can thus be considered
to be the different systems, and their average behaviour is represented by the
ensemble which is described by the density matrix. I n using the one- or twoelectron density matrices we give up all hopes of getting information about threeor more-electron correlations, that is, we sacrifice all pursuit of more detailed
knowledge.
We must still discuss one more way of looking at the density matrix. So far
we have seen two different ways of obtaining the density matrix, which we could
call the statistical and the quantum mechanical methods of approach, and we have
made it plausible that they lead to the same results. T h e approach used by Fano
(1957) combines aspects of both methods. On the one hand it recognizes that
we have often insufficiently detailed data available about the physical systems
used in experiments so that we are forced to use ensembles but, on the other hand,
it emphasizes that we are usually not interested in a detailed description of these
systems, and that we can thus with profit use an approach which has already from
the beginning eliminated superfluous degrees of freedom from the description.
One might call this the operational approach. One tries to find a set of parameters
which define the physical system with as much accuracy as corresponds to the
experimental set-up (or the given or accessible data). T h e density matrix elements
can often be used as such parameters and as the density matrix is often more
convenient than the full quantum-mechanical description followed by the appropriate averaging process, it is used in this operational method. I n the following
we shall encounter examples of all three approaches. A typical example of experiments where the operational approach is suitable is that of polarization experiments
(see 8 8). T h e quantum mechanical approach is the logical one for many-body
problems such as the Hartree-Fock self-consistent field or the Thomas-Fermi-Dirac
309
statistical theory of the atom (see 5 5), and the statistical approach is the more
natural one for equilibrium problems (5 4) or transport problems (3 7).
From this discussion we see that there is really no simple answer to the question :
What is the density matrix? T h e answer depends completely on ones point of
view and can be either : It is the quantum-mechanical counterpart of the classical
distribution function (statistical point of view), or : It is the most general description of an open quantum mechanical system, that is, a system which cannot be
described by a wave function (quantum-mechanical point of view), or, finally :
It is the most convenient way to collect all parameters which are of interest for a
given experimental set-up and to describe their behaviour (operational point of
view).
I n the next section we shall discuss in more detail the general properties of the
density matrix, its normalization, the conditions to be satisfied for a pure state,
and other conditions to be satisfied by the density matrix elements. We shall
also prove the equivalence of the statistical and the quantum mechanical definitions
of the density matrix and derive the equations of motion to be satisfied by the
density matrix.
If one adopts the statistical point of view, one is often especially interested in
equilibrium properties. T h e discussion of the density matrix as a tool in statistical
mechanics falls to a large extent outside the scope of the present survey and can
be found in most textbooks on the subject (for instance, Tolman 1938, ter Haar
1954, Kittel 1958). There have, however, been recently various developments of
great importance in this subject and we wish to discuss briefly some of these
developments. They have been brought about largely by the use of the new
field-theoretical methods. That this is feasible can be seen as follows. I n modern
field theory one is concerned with the interaction of a particle with a field, or of
interactions between fields. This problem can be formulated by using second
quantization, that is, by putting it in terms of annihilation and creation operators.
Once this is done one can use the Feynman-Dyson diagram technique to evaluate
various physical quantities. I n quantum statistics one is dealing with systems of
interacting particles. This problem can also be expressed in terms of annihilation
and creation operators, and once this is done the problem is formally of the same
type as the problems encountered in field theory. A large part of the extensive
literature of recent years dealing with the statistical mechanics of systems with
large numbers of interacting particles has been using field theoretical methods.
As a result it became clear that the field theoretical propagators or Green functions
would be a useful tool. This line of attack has been developed recently, notably
in Russia, especially for the treatment of many-body problems such as superconductivity and plasmas (for instance, Bonch-Bruevich 1956 a, Fradkin 1959 a, b, c,
Abrikosov, Gorkov and Dzyaloshinskii 1959 ; see also Matsubara 1955 b, Martin
and Schwinger 1959). It has also been found to be very useful in treating transport
problems (see, for instance, Kubo 1957 a, Montroll and Ward 1959). I t may be
mentioned that Green functions were used in statistical physics before this, when
they were known as time-dependent correlation functions-another instance of the
difficulty of communications between one branch of theoretical physics and
another. Sections 3 and 4 will be devoted to a brief discussion of this kind of
problem. I n 3 we shall discuss Green function techniques and in 5 4 recent
3 =o
D.ter Haav
developments in the evaluation of the equilibrium density matrix of systems of

interacting particles. T h e techniques discussed in these two sections will also be
used in $ 7 where we discuss transport properties and in $ 6 where solid state
problems are discussed.
T h e statistical approach to the density matrix comes to the fore in our discussions
in $ 4 , but in a number of many-body problems such as atoms and nuclei the
quantum mechanical approach is more suitable. T h e problems we shall discuss in
$ 5 are the Hartree-Fock and Thomas-Fermi-Dirac approach to atomic systems
and to nuclei. We shall assume, as is usually done, that all forces are two-body
forces and are additive. This is a plausible assumption in as far as electrons are
concerned, but may be completely wrong for nucleons. If we do this, we can
express the energy of the system in terms of one- and two-particle density matrices,
and we can discuss various physical properties of the system under consideration
in terms of these density matrices. As one is usually especially interested in the
properties of the ground state of the system, one can use variational methods to
obtain the appropriate density matrices. One can also show that the Slater
determinants, normally used in Hartree-Fock theory, can be expressed in terms
of these density matrices. This means that one can, in the particular case where
the wave function can be expressed as a single Slater determinant, obtain even this
total wave function once the single-particle density matrix is known. I n the
Thomas-Fermi statistical model one can use the density matrix to obtain the von
Weizsacker correction to the energy of a nucleus (Naqvi 1959). We shall also
briefly discuss an ingenious method (suggested by Bopp (1959) ) to use the density
matrix to obtain atomic energy levels.
I n $0 6 and 7 we turn again to the statistical approach. In $ 6 we discuss some
equilibrium solid state problems. T h e density matrix is used to calculate the
diamagnetism, both of a free electron gas and of the conduction electrons in metals,
and a brief discussion is given of the Meissner effect in superconductors. In $ 7
we discuss the density matrix approach to irreversible processes, or rather, transport
processes. We shall briefly touch upon some aspects of the quantum mechanical
H-theorem, that is, the problem of the approach of a quantum mechanical system
to equilibrium, in $ 10, but only in as far as it is involved in our discussion of the
measuring process in quantum mechanics. For a more detailed discussion we
must refer to the literature (see, for instance, ter Haar 1955). The discussion of
irreversible processes in $ 7 is restricted to the response of equilibrium systems to
such external agents as electrical fields, magnetic fields, and temperature gradients.
One way to treat transport properties is to use a transport equation (see, for instance,
ter Haar 1954, Ch. 10, Wilson 1953). There are, however, many objections to
the use of a transport equation. This is true both for classical and for quantum
mechanical systems. T o see the difficulties involved in the usual approach, let US
fix our ideas and discuss the case of the electrical conductivity. Moreover, let us
simplify the discussion by considering the classical case. T o find the conductivity
we must evaluate the current produced by imposing upon the system an external
electrical field. This current is obtained by calculating the average velocity of the
charges in the system along the direction of the electrical field. T o do this we
need to know the one-particle distribution function-the classical counterpart of
the one-particle density matrix. This distribution function is the solution of the
Theory and Applications of the Density Matrix
311
Boltzmann transport equation. I n this transport equation there occur the electrical
field-it is usually assumed that it can be treated as a perturbation-and
also a
term measuring the influence of the interactions between particles (the so-called
collision term). This last term will contain the two-particle distribution function.
We see thus that we must either express this two-particle distribution function in
terms of the one-particle distribution function, or we must first solve a transport
equation for the two-particle distribution function. T h e first approach is usually
employed in elementary transport theory but, except in very simple cases, the
approximations used in simplifying the collision term are not very satisfactory.
If we try to apply the second approach we run up against the difficulty that the
transport equation for the two-particle distribution function contains the threeparticle distribution function, and once again we face the same decision, whether
to approximate here, or whether to look for the equation of motion for the threeparticle distribution function. This chain is broken only when we get to the
equation of motion for the n-particle distribution function (where n is the total
number of particles in the system). This distribution function is nothing but
the classical ensemble density, and its equation of motion is the so-called Liouville
equation. We may refer at this point to the many attempts to justify the use of
the Boltzmann transport equation by showing how and under what conditions it
follows from the (exact) Liouville equation. Many of these attempts were described
at the Brussels Conference on Transport Processes in Statistical Mechanics
(Prigogine 1958). As these attempts have not been an unqualified success, many
authors have used different modes of approach (Kubo 1957 a, Greenwood 1958,
Kohn and Luttinger 1957, Nakano 1956, Lax 1958, Edwards 1958 b, Chester and
Thellung 1959). I n these papers an exact expression is found for the conductivity,
or rather, an expression is obtained by approximating the exact equation of motion
for the density matrix in retaining only terms which are at most linear in the field
strength. We shall describe this way of evaluating transport coefficients in 5 7.
T h e next two sections use the operational approach. T h e first section deals
with polarization experiments, both with those dealing with particles having a
magnetic moment of their own, and with those dealing with polarized light. T o
fix the ideas, let us consider the polarization of a beam of electrons. I n this case,
we are only interested in the average value of the magnetic moment of the electrons.
T h e values of the other coordinates of the electrons are of no interest to us. There
are exactly three parameters which are of interest, namely the three components
of the polarization vector. One can introduce a two-by-two density matrixwhich because of its normalization contains three independent parameters-to
describe the system uniquely. If the beam traverses a region where there is a
magnetic field, the polarization will change. T o calculate this change in polarization by using the Dirac equation of an electron moving in a magnetic field and
averaging over the spatial coordinates of the electron is extremely tedious. On
the other hand, if one uses the equations of motion for the density matrix, and the
relation between the polarization and the density matrix, the equation of motion
for the polarization follows easily. I t turns out to be an example of the Ehrenfest
theorem which states that the equations of motion for a quantum-mechanical average
will be the same as the classical equations of motion for the corresponding classical
physical quantity. T h e polarization is the average of the magnetic moment-or
312
D.ter Haar
the spin-of the electrons, and its equation of motion is the classical equation for
the rate of change of a magnetic moment in a magnetic field. A similar analysis
can be carried out for particles with spin values larger than 4 and also for the
polarization of light.
In the last part of 8 8 we discuss angular correlation and scattering experiments.
Once again we are interested only in part of the parameters describing the system,
and the density matrix again has turned out to be a useful tool. I n $ 9 we return
to solid state problems, namely, the problem of the relaxation of magnetic moments
in a solid. This problem is of great interest in resonance experiments. In many
ways this problem-and
its solution-is similar to those involved in transport
processes which we discussed in $ 7.
I n the last section we discuss a topic of more academic interest, namely, the
theory and interpretation of measurement in quantum-mechanical systems. This
topic was first discussed extensively by von Neumann (1932) and is the subject of
an extremely lucid book by London and Bauer (1939). It is a topic which has
again been discussed recently by many authors (for instance, Ludwig 1954,
Green 1958, Durand, unpublished), but we shall base our discussion on the,
perhaps old-fashioned, point of view presented by London and Bauer. T h e
main aspects of the measuring process are the following ones. T h e system to
be observed must be coupled with a set of detectors. A definite change in the
state of the detectors will be brought about by this coupling. T h e measuring
set-up will depend upon the physical quantity to be measured and will be such
that different values of the physical quantity involved will lead to different detectors
being activated. After the interaction between the system and the detector the
two are decoupled, and from seeing which detector has been activated we can
conclude the value of the physical quantity which corresponded to the original
state of the system under observation. T h e measurement itself has, however,
changed the state of the observed system. Before the measuring process the
system might have been in a pure state, but this will no longer be the case after
the measurement.
I n this introduction we have attempted to give a rough idea of the problems
connected with the density matrix. In the limited confines of a survey article one
must of necessity restrict the discussion and in the next sections we shall try to
give a more quantitative discussion of the topics mentioned in the foregoing, but
we shall still have to leave out many details for which we must refer to the literature.
OF
S T H E D E N S I T YM A T R I X
$ 2 , G E N E R A LP R O P E R T I E
In this section we shall discuss the general properties of the density matrix.
We shall introduce the density matrix first of all through the statistical approach
and prove some of its properties. After that we shall introduce it through the
quantum-mechanical approach and show that the two density matrices are, indeed,
the same. We may refer to the following papers and books to supplement the
discussion in this section : Kemble 1937, Tolman 1938, Husimi 1940, ter Haar
1954, Fano 1957 and Hagedorn 1958.
In the statistical approach to the density matrix we are concerned with the
description of a physical system by an ensemble. This ensemble will often be a
313
grand ensemble, that is, the number of particles in the different constituent systems
will not necessarily be the same for all systems. We shall, however, in this section
not emphasize this aspect, but it will be important in the discussion in $4. Let
there be N systems in the ensemble and let these systems be described by normalized
wave functions $k (k = 1, ....N ) . It is convenient to introduce a complete orthonormal set p, in terms of which the i,bk can be expanded. T o simplify our discussion in the present section we shall neglect complications introduced by spin,
that is, we assume the t,P and p, to be scalars. It is easy enough to take spin into
account, and we shall do this, for instance, in $ 8. I n terms of the yn we have
$k =
C , C , ~ ~ , (k
, = 1, ....N ) .
......(2.1)
Originally we had defined our ensemble by the wave functions $k. I n the new
representation it is defined by the coefficients cnk.
Let A be a physical quantity, the value of which we wish to determine for the
system which is represented by our ensemble. This quantity will correspond
quantum mechanically to an operator A^ (all operators will be denoted by ") and its
average value for the kth system will be given by the equation
A,
$k*A$kdr,
. . . . . .(2.2)
where j...
d7 indicates integration over all arguments of gk. Taking the average
over the ensemble, which we shall denote by (. . .), we obtain the expectation value
of A^ for the system under consideration, and we get
(A^)
. . . . . .(2.3)
c,,
. . . . . .(2.4)
N-l
$k*A$kdT,
k-1
or, using Eqn (Ll),
<A^)
N-l ~k
n cmk* C n k A,,,
where A,, is the matrix element of A^ in the p,-representation :
=
A,,
J pm*Ap,dr.
......
Introducing the density matrix i; by its matrix elements in the p,-representation,

pmn = N-l
2 , cnk* clnk,
. . . . . .(2.6)
we can write Eqn (2.4) in the form
<A)= c,, n Amn Pnm = T r (?A^),
. . . . . .(2.7)
where T r indicates the trace, that is, the sum of the diagonal elements.
A few remarks should be made here about Eqns (2.3),(2.6) and (2.7). First of
all we notice that ( A ) is a double average : one average is the quantum-mechanical
average given by Eqn (2.2), and the other is the statistical average over the ensemble.
This means that two different kinds of probability considerations enter into our
discussion ; this point was specially emphasized by London and Bauer (1939).
T h e quantum-mechanical probability considerations enter even if we have as complete a knowledge about the physical system as is possible in quantum mechanics,
that is, if we know the wave function of the system ; they are a feature inherent in
quantum mechanics and are not related to any lack of (possible) knowledge. T h e
D.ter Haar
314
statistical probability considerations, on the other hand, are closely connected to our
lack of knowledge-as they are in classical statistics-and were introduced exactly
because our knowledge is incomplete. Secondly, we note that the density matrix,
or statistical operator as it is sometimes called, is defined by its matrix elements in
the particular representation in which we are working. We shall see presently
that if we change to another representation the density matrix will change according
to the usual rules of quantum mechanical transformation theory. Finally, we note
the intimate connection between the density matrix elements and the average values
of physical quantities. This connection is used by the operational approach to
define the density matrix in terms of averages. There are cases where the density
matrix is a finite matrix-for instance, when we are dealing with polarization
experiments-and if it is a matrix of rank M it is determined by 1%' - 1 independent
parameters (oide infra). If we can find the average values of M 2 - 1 physical
quantities A, these will suffice to determine the density matrix. I n the case of
the polarization of light, for instance, M will be 2, and the three components of the
polarization vector will just be sufficient to determine the density matrix completely,
as we shall see in 0 8.
From the definition (2.6) we see that i; is Hermitian,
Prim**
. . . . . .(2.8)
Applying Eqn (2.7) to the unit operator (d = ?) we find that i; is normalized
Pmn =
(to unity),
1 =(I)
= Trj3.l
. . . . . .(2.9)
Trj3.
This result could also have been obtained directly from Eqn (2.6) and the fact
that the t,!Jk are normalized
T r i;
N-lCk E n ank*
ank =
N-1
C k 1 = 1.
...... (2.10)
From Eqn (2.8) it follows that the diagonal elements of the density matrix, pnn,
are real, and from Eqn (2.10) it follows that they must satisfy the relations
C n p n n = 1, O G p n n 6 1 .
. . . . . *(2.11)
The physical meaning of the pnn is clear from Eqns (2.6) and (2.1) : it is the (normalized) probability that pfi is realized in the ensemble.
Let us now consider a change from the gon-representation to another representation, say, the Xp-representation. Instead of Eqn (2.1) we have now
. . . . . .(2.12)
C n C n k p n = Cpdpkxp*
The transformation from the pn- to the x,-representation will be characterized by
a unitary transformation matrix S,, such that
t,!Jk
X, = Cn Pn SnD,
while
so that
Sn,*
= (s-')pn,
Cn Snp*
s
n
q = apq,
where a,, is the Kronecker delta function.

From Eqns (2.12) and (2.13) it follows that
cnk Cq Snq d f l ,
. . . . . .(2.13)
. . . . . .(2.14)
. . . . . .(2.15)
. . . . . .(2.16)
315
and using Eqn (2.15) we get from this equation
. . . . . .(2.17)
dqk = Enenk Snp*.
If we denote the transformed density matrix by a prime, we have instead of

Eqn (2.6) the equation
PPq -N - l C d 4k * dPk
. . . . . .(2.18)
and from Eqns (2.17) and (2.14) we get
ppq =
N-l
X:k,m,n enk* Sng emk s m p *
C m , n ( S - l ) p m Pmn
or, in matrix notation,
p^
Snp,
* * * *
(2.19)
. . . . . .(2.20)
S-lPS,
which is the normal equation for the transformation of an operator.

We notice, by the way, that the averages (A^) will be unaffected by the transformation since
(A) = T r ?A = T r S-I p^SS-lAS= T r S-1 BAS
T r SS-l pÂ= T r pÂ= (A),

. . . . . .(2.21)
where we have used the property of the trace

Tr
= T r &?A,
. . . . . .(2.22)
and the fact that SS-l = 1.
Let us now assume that we have made a transformation to a representation in
which p^ is diagonal,
. . . . . .(2.23)
Pmn = Pm amn*
Y( = ( p ^ ) ) .
Consider now T r j
We have
. . . . . .(2.24)
( T r p^)2 = 1,
where we have used Eqn (2.9) and the fact that the pm are non-negative. As the
trace is invariant under a unitary transformation, we can write Eqn (2.24) in the
form
or,
E,,, Pmn Pnm = C%,mI Pmn 1 Q 1, * * * * (2.25)
T r P2 Q 1)
T r p2 = Cmpm26 (X,P,)~
where we have used Eqn (2.8). Inequality (2.25) imposes a limitation upon all
elements of the density matrix, that is, upon both the diagonal and the off-diagonal
elements.
If me are dealing with a finite matrix, say of rank M , there are altogether
M 2 complex matrix elements pmn, that is, 2 M 2 parameters. This number is
reduced by a factor 2 because of Eqn (2.8) and reduced by unity because of the
normalization condition (2.9) so that there are M 2- 1 independent parameters.
From the operational point of view this means that one needs M 2- 1 independently
measured quantities to fix the density matrix.
I n the preceding section we have discussed the difference between a pure state
and a mixture. We now wish to establish the condition that the density matrix
corresponds to a pure state. A pure state corresponds to the case where we have
the maximum information available about the physical system, that is, where all
systems in the ensemble possess the same wave function, #,, say. In that case there
is only one averaging process involved in obtaining ( A ) , namely, the quantum
mechanical one. T h e pure state is characterized by the existence of what Fano
D.ter Haar
316
(1957) calls a ' complete ' experiment ; this is an experiment which gives a result
predictable with certainty when performed on this state and gives this particular
result only for this particular state. T h e possibility of such an experiment is
apparent if we bear in mind that it is possible to find a Hermitian operator, corresponding to a physical observable, which possesses this particular state as a (nondegenerate) eigenstate. This complete experiment can be used as a filter. It should
be an experiment designed to measure the observable, of which y!Io is an eigenfunction. We may mention a Nicol prism as a possible apparatus which could
serve as a filter when we are dealing with density matrices describing the polarization
of light (compare 8 8).
We shall now prove that the necessary and sufficient condition that i; describe a
pure state is
i;2 = i;,
. . . . . .(2.26)
that is, i; should be idempotent. That Eqn (2.26) is a necessary condition follows
easily from the fact that Eqn (2.6) now leads to
pmn = C,(O)*
. . . . . .(2.27)
C,(O),
as the cnk are independent of k and are all equal to the expansion coefficients c n ( 0 )
of y!Io where y!Io is the wave function of all systems in the ensemble. As #o is
normalized we get immediately
(p),,
-&c(,o)
cl(o)*
CZ (0) cn(o)* = Cm(O) cn(0)* =
pmn.
. . . . . .(2.28)
T o prove that Eqn (2.26) is a sufficient condition we consider again a representation for which i; is diagonal. In that case i;2 is also diagonal, and Eqn (2.26) is
now equivalent to
pn2 = p n for all values of n,
. . . . . .(2.29)
or :
either pn
0 or
p n = 1.
. . . . . .(2.30)
From the normalization condition (2.9) and Eqn (2.30) it follows that one of the
pn, say p o , is equal to unity while the other p n vanish :
p o = 1 ; pn = 0,
n#0.
. . . . . .(2.31)
From Eqn (2.6) we then get

po = N-l&Jc0"Z = 1,
. . . . . .(2.32)
and as all cnkI2 must be not greater than 1, Eqn (2.32) can only be satisfied by
putting
lcOkl = 1, k = 1 , ... ,N,
. . . . . . (2.33)
or
#k =
exp (iak)yo,
olk
real,
. . . . . .(2.34)
which me ns that, apart from an irrelevant phase factor exp (iolk),all $k correspond
to the same wave function.
We shall now derive the equation of motion for the density matrix. We have
assumed that the systems in the ensemble are describable by a wave function and
we shall assume that the wave function satisfies the Schrodinger equation
HyP
= iA@,
. . . . . .(2.35)
317
where, as usual, the dot indicates differentiation with respect to time. T h e operator
to use Eqn (2.1) to obtain
the transformed Schrodinger equation
Z? is the Hamiltonian of the system. I t is convenient
. . . . . . (2.36)
i7iCnk = Z l H,, c,k,
and from Eqns (2.36) and (2.6) and the fact that A is Hermitean it follows straightforwardly that
itip,, = [Z?, i;]-,,,,,
or ;ti; = [A,?I--,
. . . . . .(2.37)
where [ , 1- indicates the commutator. We shall defer a discussion of Eqn (2.37)
until after we have considered the quantum-mechanical approach to the density
matrix, but there are one or two remarks we wish to make at this juncture.
We have so far been working in the Schrodinger representation, that is, the
operators are assumed to be time-independent and the time dependence is contained
in the wave function (compare Eqn (2.35) ). In the Heisenberg representation, on
the other hand, the time dependence is in the operators. Instead of a wave (or
Schrodinger) equation one has equations of motion for the operators. These
are of the same form as the equation of motion (2.37) for i; but with the opposite
sign. If we were to change to the Heisenberg representation we would find that i;
would be a constant operator (see, for instance, Hagedorn 1958).
T h e time dependence of (A) is, in the Schrodinger picture, invested in i; and
in the Heisenberg picture in A. For (A)we find from Eqns (2.7) and (2.37),
using the fact that
is time-independent in the Schrodinger picture
or,
;ti&
([A,A]-),
. . . . . .(2.38)
with the sign corresponding to the Heisenberg picture! Eqn (2.38) is, of course,
true in all representations.
We now turn to the quantum-mechanical approach to the density matrix. We
consider a system with many degrees of freedom. We shall denote those degrees
of freedom in which we are interested collectively by x and the other degrees of
freedom collectively by q. T h e total wave function of the system will be a function
of both q and x : Y(q, x). Consider now an operator A which is a function of the x
only. Its average value will be given by the equation
(4= jj-.f.*
(4, x)
x>dq dx,
. . . . . .(2.39)
where J ... dq and J ... dx denote integrations over the whole of q-space and the
whole of x-space.
If we define the density matrix i; by the equation
(3
I i; I x> =
Y(q, x) y q , x) 4,
. . . . . .(2.40)
where we have used for the moment Diracs notation for operators, Eqn (2.39) is
equivalent to
(A)=SS<x~i;lx}dx(xjdlx)dx = Tri;A,
. . . . . .(2.41)
D.tev Haar
3 18
where we have used the definition of the trace in coordinate representation.

From Eqn (2.40) and the fact that Y ( q , x ) is normalized we get easily
T r i; = (x I i; I x) dx = Y*(q,x) Y(q,x) dx = 1.
. . . . . .(2.42)
We see from Eqn (2.41) that as far as the degrees of freedom which are described
by x are concerned, i; describes the situation completely. As Eqns (2.41) and (2.42)
are the same as Eqns (2.7) and (2.9), respectively, the density matrices defined by
Eqns (2.40) and (2.6) must also be the same. We can pursue this point a little
further. T h e degrees of freedom corresponding to the q may be called the bath
in which the system described by the x is embedded. T h e averaging over the q
takes the place of the earlier ensemble average. T o know how i; will change in
time, we isolate the system from the bath and consider the Hamiltonian E? which
t ) the complete
describes the behaviour of the isolated system. Let $ ~ ~ ( x ,be
orthonormal set of time-dependent eigenfunctions of A. We can expand
(x I i; I x) in terms of the $n and write
(XI i; I x> = Em,,
amn $ n * ( ~ ,t ) $ m ( ~ t).
,
. . . . . .(2.43)
We then find for i; the following equation of motion.

i W x 1613)
Em,,
amnih[$n*(x, t ) 4m(x, t ) + & * ( X I ,
Em,,
amn
=s
=
x([
$n*(x, t ) <x I g
t )$ m ( ~ t)I
,
I x>dx$ m ( x n , t )
- ?Lm(x,t ) $hn*(x,t ) dx(x
IAlx)dx(xI p [ x) - ( X I
I x) dxxIAl
I H* [ x)
x)]
. . . . ..(2.44)
(xIAi;-i;RIx),
in accordance with Eqn (2.37).

If we can write
. . . . . .(2.45)
Y q , x) = W ) x ( x ) ,
we are dealing with a pure case. In that case we have
. . . . ..(2.46)
amn = aman*,
where the a, are the coefficients in the expansion of x(x) in terms of the &Jx, t).
From Eqn (2.40) it follows that
. . . . . .(2.47)
(x I i; 1x9 = x*(x)x(x),
and we get
(x I 8 2 I x)
(x I i; I x) dx(x I i; I x)
= /x*(r)
x(x) dxx*(x)~ ( x =) x*(x) x(x)
(x I i; [ x),
. . . . . .(2.48)
that is, the condition (2.26) for a pure case is satisfied.

Instead of using the $Jx, t ) to expand (x I i; I x) we could have used the complete
orthonormal set pn(x) which we used in Eqn (2.1). In that case we would have got
(3
I 8 I x) = Em,,
cmn pn*(x> ~
m(x),
. . . . . .(2.49)
Theovy and Applications of the Density Matvix
39
where the c,,, possess the same properties as the pmn at the beginning of this section.
Indeed, one can prove from Eqns (2.41), (2.42) and (2.44) that the c, satisfy
Eqns (2.7), (2.9) and (2.37) where all matrix elements are again in the p,-representation. I t is known from quantum mechanical transformation theory that the
pn(x) can be used as a transformation matrix to change over from the discrete
n-representation (which we previously called the pn-representation) to the
continuous x- or coordinate-representation. We see thus that the density matrix
introduced in the quantum mechanical method is just the coordinate-representation
of the density matrix introduced in the statistical approach.
T o conclude this section we shall discuss briefly the case of compound systems
(we shall return to this subject in Q 10). Let our system consist of two sub-systems
A and B, and let superscripts AB, A and B denote quantities referring respectively
to the complete system and to the two sub-systems. T h e density matrix p A B will
now have two sets of indices, one (m,m) referring to A and the other (n, n) referring
to B. It is important to define averages over the sub-systems. We have
(A^)*
,.-
Tr,,(AIBPAB) = E,,
where the density matrix
i;*
Am,,
En%,a,,
p&Bn.,
PA for the sub-system

=
Tr,
PdB
or
p&,,
Tr,Aî;
. . . . . .(2.50)
A is defined by
E,
pmnmtn,
AB
. . . . . .(2.51)
and where A^ is an operator acting upon the sub-system A only,

If the two sub-systems are completely independent, we must have for any two
operators A^ (acting upon A) and B (acting upon B)
. . . .(2.52)
and the density matrix must satisfy the condition
. . . .(2.53)
where x indicates a direct product.
Q 3 . G R E E NF U N C T I OTNE C H N I Q U E S
I n this section we shall briefly consider the definition and properties of the
temperature-dependent Green functions which are playing such an important role
in recent developments of quantum statistical mechanics. We cannot enter into a
detailed discussion here, and refer to the steadily growing literature, especially
to the papers by Matsubara (1955 b), M a A n and Schwinger (1959) and Zubarev
(1960). Other relevant papers are those by Koppe (1951), Salam (1953), Kinoshita
and Nambu (1954), Bonch-Bruevich (1955, 1956 a, b, c, 1957, 1958 a, b,
1959 a, b, c, d), Van Hove (1955, 1957), Watson (1956), Ezawa, Tomozawa and
Umezawa (1957), Klein and Zemach (1957), Kubo (1957 a), Migdal (1957),
Belyaev (1958 a, b), Galitskii (1958), Galitskii and Migdal (1958), Gorkov (1958),
Klein and Prange (1958), Kobelev (1958 a, b, c, d, e, 1959), Kraichnan (1958 a, b),
Landau (1958), Montroll and Ward (1958, 1959), Prange and Klein (1958),
Abrikosov, Gorkov and Dzyalozhinskii (1959), Bogolyubov and Tyablikov (1959),
Bonch-Bruevich and Glasko (1959), Chen Chun-syan (1959), Falk (1959 a, b),
Fradkin (1959 a, b, c, d, e), Fujita (1959), Hugenholtz and Pines (1959), Kanazawa
D.ter Haar
320
and Watabe (1959), Klinger (1959 a), Kogan (1959), Pitaevskii (1959), Vedenov
(1959), Vedenov and Larkin (1959), Balian and De Dominicis (1960 a), BonchBruevich and Kogan (1960), Bonch-Bruevich and Mironov (1960), Dzyub (1960),
Fujita and Hirota (1960) and Sawicki (1960).
We mentioned in the introduction that the use of Green functions came about
quite naturally on two accounts. First of all, the resemblance between the main
statistical problem-that of evaluating the thermodynamic properties of a system
of interacting particles- and the main field-theoretical problem-that of considering
interacting fields-when both problems are stated in terms of annihilation and
creation operators leads naturally to the use of Green functions in quantum
statistics, as Matsubara (19.55 b) was the first to note. Secondly, the close relation
between the correlation functions of kinetic theory and the two- or more-particle
Green functions (vide inf~cz)niakes their occurrence also a natural one. In this
section we shall briefly introduce the Green functions and briefly consider some of
their properties which will be of interest for the discussions in later sections.
There are many ways of defining Green functions, and it is perhaps expedient
to dwell briefly on the connection between different Green functions. There are
two kinds of Green functions in which we shall be interested, namely, the
generalized double-time (causal) Green functions discussed by Zubarev (1960 ; it
was pointed out by Zubarev that it is often more convenient to use the advanced or
the retarded Green functions, because of their analytical properties, but this
distinction is too subtle for our present discussion), and the propagators of field
theory. T h e generalized Green function of two operators A^ and l? is defined by
the equation
. . . . . .(3.1)
GJ,B ( t ,t) = (-;/ti) (TA(t) B(t)),
where A^ and
are time-dependent operators, and where T indicates Wicks
chronological or the T-product of the two operators. One can write this T-product
in the form
. . . . . .(3.2)
TA(t)B(t) = 0(t-tf)A(t>B(t)+y0(t-t)B(t)A(t),
where 8 is the step function
q t ) = I,
t>o ;
e(t) = 0, t < 0,
. . . . . .(3.3)
and where y = + 1 or - 1 according to whether d and B are Bose- or Fermioperators. T h e generalization of the usual Green function of field theory consists
of two parts. Firstly, we have not yet made any restrictions on the operators
A^ and I?, while in field theory A^ and B are normally the second-quantized wave
functions, that is, the wave functions which are themselves operators by virtue of
their expansion in creation and annihilation operators. Secondly, the average
in Eqn (3.1) is over a statistical ensemble, in accordance with our definition of
the ( ) average. It is usual to choose a canonical or grand canonical ensemble
for this average, that is, an equilibrium ensemble, but in our definition this restriction has not yet been introduced. I n field theory, the average is normally over a
pure state, namely, the ground state. There is also one other difference, namely,
that the Green functions which we have introduced depend on two time arguments
only, while in field theory there occur multiple-time Green functions.
321
T h e propagator Green function occurs in field theory when one discusses

diagram techniques (see, for instance, Bogolyubov and Shirkov 1959 or Hamilton
1959) and it is then shown that, indeed, it coincides with the usual definition of a
Green function in the theory of differential equations. This definition is the
following one (for instance, Morse and Feshbach 1953). If we wish to obtain the
field due to a source distribution, we calculate the effect of each part of the source
distribution and add them up. If G(x,t ; x, t) is the field at x at time t caused by
a unit point source at x at time t, G ( x ,t ; x, t) is called the Green function of
the problem considered. If the field equation is given by the equation
. . . . . .(3.4)
a* = f ,
where # is the field function (for instance, the components of the electromagnetic
four-potential), s2 a differential operator (in the case of the electromagnetic field
the dalembertian), and f the source density (in the electromagnetic case proportional to the current-charge four-vector components), the Green function will be
a solution of the equation
G ( x ,t ; x, t) = U
~ ( X x)
. . . . . .(3.5)
8(t - t),
where 6(x) and 8 ( t ) are the three- and one-dimensional Dirac delta-functions.
T h e constant 01 depends on the exact form of the right-hand side of Eqn (3.4).
We shall try to make it plausible how these Green functions play the role of
propagators by considering the Fokker-Planck equation. This is the diffusion
equation for Brownian motion. T h e ordinary diffusion equation has a field operator
Q given by the equation
. . . . . .(3.6)
R = v2 - a2 apt,
where a is a constant involving the diffusion coefficient.
Green function equation
T h e corresponding
(aGli?t)-~-~VG
= 8(~-~)8(t-t),
. . . . . .(3.7)
is the Fokker-Planck equation for the probability that a particle at time t at x will
at time t be at x through Brownian motion. T h e Green function thus describes
the propagation of the Brownian particle through the medium.
For our further discussion it is instructive to see how the Green function (3.1)
for the special choice of A = $, = $7 satisfies Eqn (3.5). T h e $ and $t are the
wave functions in second quantization, and the t indicates, as usual, the Hermitian
conjugate, which means that if $ contains creation operators, $t will contain
annihilation operators and vice versa. T h e and 1% are thus operators and they
satisfy the commutation relations
1%
$(x,~)$?(x,t ) - q $ t ( ~ , t ) $ ( ~ , t=
) 8 ( ~ - x ) 8(t-t).
. . . . . .(3.8)
1%
If describes a free field, that is, if there are no interactions, it satisfies the equation
of motion
. . . . . .(3.9)
= ik(a$/at) + (k2/2m)v2$ = 0.
n$
From Eqns (3.1), (3.2), (3.3)) (3.8) and (3.9) it follows in a straightforward manner
that
. . . . . .(3.10)
LIG$,$+= S(X - x) 8 ( t - t),
21
D.ter Haar
322
where we have used Eqn (2.9) and the fact that

-
aqt - tyat = q
t -t y a t
8(t - t),
. . . . . .(3.11)
It is of interest for statistical applications of Green functions to consider also

the case of interacting particles. Instead of Eqn (3.9) we then have the equation
of motion
where we have assumed two-body interactions, and where u ( x , x ) is the potential

energy corresponding to these interactions. If we now proceed as before we find
the following equation of motion for the Green function G$,$t,
ik aGG1$it+ ??
! V2G$l,;lt
~
at
2m
- - d3 X a ( x , x) G+l$2,;;$,t = S(x - x ) 8(t - t).
. . . . . . (3.13)
We note that the equation of motion for the one-particle Green function G$,;t
involves the two-particle Green function G$l$l,$l+$lt.In turn the equation of motion
for the two-particle Green function will involve the three-particle Green function,
and so on. This situation is similar to the situation in the theory of transport
processes (compare 3 7 ) and in the theory of liquids where we encounter similar
chains of equations for the correlation functions.
Let us briefly consider this connection between the Green functions and the
correlation functions FJ,; which are defined by the equation
. . . . . . (3.14)
F ~ , g ( t)
t , = (A(t),B(t)).
Note that, in general, F;,fi =l=
F$,a, just as Gi,$ + G s i , and also that F - Z ,is~ defined
for t = t, which is not the case for G 2 , j . The close connection between the F;,;
and the G i g is apparent from Eqns (3.14), (3.1) and (3.2). T h e correlation functions (and thus the Green functions) describe the microscopic properties of physical
systems. As statistical mechanics is interested in deriving the macroscopic properties from a knowledge of the microscopic behaviour, we see still another reason why
Green function techniques have become important. We shall not discuss one of
the most important applications of Green functions-an application for which it is
essential to use the time-dependent Green functions-namely,
the derivation of
the energy spectrum of the elementary excitations, that is, of the quasi-particles
which originate from the many-body interactions in the system.
T h e Green functions will occur in the next section as propagators, and also in
$9 6 and 7 as correlation functions.
9 4. T H ED E S C R I P T I OONF
S T A T I S T I C AELQ U I L I B R I U M
D E N S I T YM A T R I X
In this section we shall first of all discuss the form of the density matrix which
corresponds to thermodynamic equilibrium (see Tolman 1938, Husimi 1940,
ter Haar 1954, 1955). We also discuss the relation between this density matrix
and the various thermodynamic functions of the system under consideration. We
shall then consider some of the methods developed recently for the evaluation of
BY THE
323
the grand canonical or canonical partition function. Among these methods we may
mention the Feynman path method (Feynman 1953, Yaglom 1956, Husimi, Kitano
and Nishiyama 1958, Brush 1961, Siegert 1960), Feynman graph methods
(Matsubara 1955 b, Riesenfeld and Watson 1956, Thouless 1957, 1959, Bloch
and De Dominicis 1958, 1959 a, b, Montroll and Ward 1958, Brout 1959, Glassgold,
Heckrotte and Watson 1959, Lee and Yang 1959, Balian and De Dominicis 1960 a,
Levine 1960, Stillinger and Kirkwood 1960), and second quantization methods
(Blatt and Matsubara 1958, Fujita 1959, Hubbard 1959, Balian and De Dominicis
1960 b, c, Gaudin 1960 b). For other developments, especially perturbation
theory methods, we can refer to papers by Goldberger and Adams (1952), Kubo
(1952), Chester (1954), Takabayasi (1954), Zubarev (1954), Butler and Friedman
(1955), Friedman and Butler (1955), Kaschluhn (1955 a), Kotani (1955), Nakajima
(1955), Watson (1956), Mazur and Oppenheim (1957), Oppenheim and Mazur
(1957), Oppenheim and Ross (1957), Schafroth, Butler and Blatt (1957), Yvon
(l957,1960a, b), Klein andPrange (1958), Pavlikovskii and Shchuruvna (1958,1959),
Siegert and Teramoto (1958), Lundquist (1959), Balian, Bloch and De Dominicis
(1960), Fujita and Hirota (1960) and Gaudin (1960 a), and also to many of the
papers quoted in the previous section. We may remark here that a number of
authors have applied graph methods to classical partition functions with great
success (for instance, Salpeter 1958, van Leeuwen, Groeneveld and de Boer
1959).
If the system is in a state of thermodynamic equilibrium, the density matrix
must be time-independent, and we see from Eqn (2.37) that it must satisfy the
equation
. . . . . .(4.1)
PI- = 0,
[a
which is most easily attained by letting p^ be a function of I?. We have mentioned

before that one way of looking at an ensemble is by considering one of the systems
in the ensemble to be embedded in a bath formed by the other systems. Just as
the distribution function of a molecule in a gas will be given by the MaxwellBoltzmann distribution which is produced by the interaction of the molecule with
the other molecules in the gas, so the density matrix of an ensemble in thermodynamic equilibrium will be given by the canonical or grand canonical expression
i;
exp[-q+vfi--pA],
. . . . . .(4.2)
I n Eqn (4.2) q, v and /3 are c-numbers, and fi is the number operator. We have
given here the formula for the grand canonical ensemble, that is, an ensemble where
the constituent systems may have different ndmbers of particles. One can give
a more detailed proof that Eqn (4.2) corresponds to an equilibrium ensemble (for
instance, ter Haar 1955). We must also refer to the literature for a proof of the
statement that p is related to the absolute temperature T by the equation
p-'
= kT,
....
.(4.3)
where k is Boltzmann's constant, and that v = pp, where p is the chemical or

thermal potential. The q-potential, finally, is determined from Eqn (2.9), or
exp q = T r exp ( v f i - PE?).
. . . . . .(4.4)
D.ter Haar
324
We also note, without proof, that the grand potential q determines all physical
quantities for a grand canonical ensemble just as the free energy does for a canonical
ensemble. T h e quantity Z defined by the equation
. . . . . .(4.5)
Z = exp q,
is called the grand partition function and is the quantity which is usually discussed
in the papers quoted at the beginning of this section.
From Eqn (4.2) it follows that i; satisfies the so-called Bloch equation (Bloch
1932)
. . . . . .(4.6)
apiap = -Hi;.
T h e fact that this equation is formally the same as the Schrodinger equation, with
Feynmans path
integral method (Feynman 1948) was applied by Feynman himself to liquid helium
(Feynman 1953) and has recently been discussed by Brush (1961). Watson (1956)
has used this analogy to apply scattering theory for an evaluation of the second
virial coefficient. Let g(E)be the Laplace transform of e d H ( = $(p) ) :
,k3 replacing itik, has been the basis for recent developments.
(4.7)
and let the Hamiltonian be split into the kinetic energy
energy P,
? and the potential
. . . . . . (4.8)
H = T+P.
We then have from Eqn (4.6), which is also satisfied by $(p),
. . . . . .(4.9)
(E-T)X=l+PR,
where we have used the fact that $(O) = 1.
From Eqn (4.9) we get the equation
(E- T)-I+(E-
T)-1
PR.
. . . . . .(4.10)
This equation reduced to the equation of stationary state scattering
Psi,
. . . . . .(4.11)
d(E) = E ( E )( E - F ) .
. . . . . .(4.12)
d = 1+ ( E -
T)-1
if we write
T h e equation for the density matrix is thus reduced to the scattering equation and
quantum-mechanical scattering theory (for instance, Brueckner theory : see Bethe
1956) can be applied. We have not got the space to discuss the many theories
which have, indeed, used Brueckner theory to evaluate 2.
We note from Eqn (4.10) that is the Green function for the operator E - H
(see Koppe 1951).
We shall now briefly discuss the Feynman diagram methods, but once again
we cannot go into a detailed discussion. We refer especially to Gaudins paper
(1960 b) which is based upon the earlier work of Matsubara (1955 b) and Bloch
and De Dominicis (1958); he seems to have given the clearest expos6 of this rather
complicated method. For many details we must refer to textbooks on quantum
field theory (for instance, Bogolyubov and Shirkov 1959, Hamilton 1959).
325
We shall consider a system of bosons which interact through two-body forces.

T h e Hamiltonian is then
A=Ao+P,
. . . . . .(4.13)
where gocorresponds to non-interacting particles and P t o the interaction. If we
describe the system in terms of creation and annihilation operators,
and a",,
where a",t creates a particle of energy E, and a", annihilates such a particle, we have
. . . . . . (4.14)
lrfo = E, Ek.,+dk,
V = ' 4C
and
r,s,ni,n (YS
I v I ma> a",?
a"st
a",
4,
. . . . . .(4.15)
where we have once again used Dirac's notation for the matrix elements of the
two-body interaction potential z'(x,x'). As we are dealing with bosons the a",
and a",? satisfy the commutation relations (compare Eqn (3.7) )
[&,,
We now write
a",y- =
[a",+,
a"k,+]-
[ah.,~ 7 ~ , ' ] _
= ~37~2.
. . . . . .(4.16)
. . . . . .(4.17)
. . . . . .(4.18)
e-pa,
. . . . . .(4.19)
$o(P> = exp ( - PQO).
while
where
0,
$(PI = $O(P)8(P),
where again
As
$(p) satisfies the Bloch equation, we find that 8(P)must satisfy the equation
. . . . . .(4.20)
- W P = P(P) &P),
P(P) = exp (PI?,) P exp ( -/?go).
. . . . . .(4.21)
I t also satisfies the boundary condition
8(0)= 1,
. . . . . .(4.22)
and we find the following solution of Eqn (4.20) for

&?)
"=cl
( - l)"/pdPl/%/3z.
0
.J'"-&
0
8,
P(P,) . . P(Pn).
*
. . . . . .(4.23)
For the grand partition function 2 we have

2 = T r [exp (v.&'=
PA)]
T r [exp ( ~ f i - p l ? ~ .#@)I
)
. . . . . .(4.24)
T r [exp kfi- PA011( f l ( P D 0 ,
where (. ..)o indicates an average over an ensemble with A = go,
that is,
=
(fl(P)>o
T r [exp ( - 4 0 + vfi - PfJo)
.fl(P)I
{Tr [exp (vfi-PZ?o) .fl(p)]} {Tr [exp (~fi-/3lrf~)]}-~, . . . . . .(4.25)

We see that we have obtained here an expression of the partition function in a
power series in P, as fl itself is expressed as such a power series. Using the normal
field theoretical techniques we can write Eqn (4.24) in the form
=
2=2
2 ( - 1>"(n!>-' jo'dpl/o'dPz
0 n=O
. . . / o ' d ~ n < ~ ~...

( ~P(pn)>o,
l)
. . . . ..(4.26)
326
D.ter Haar
where T indicates again the chronological product and where
. . . . . .(4.27)
2, = T r [exp(vfi-pgo)].
Using Eqns (4.26), (4.15) and (4.21) we have finally
I t is this expansion for 2 which is evaluated by diagram techniques. It must

be emphasized here that using diagram techniques does not solve the problem ;
they only serve as an aide-mkmoire so that it is more difficult to forget terms, and
also it is often possible to combine a large number of terms and sum those. If
one can convince oneself that those terms are, indeed, the only important ones, one
hopes to have obtained a reasonable approximation to the exact partition function.
Fig. 1.
T o conclude this section we shall show how one can attribute diagrams to the
various terms in Eqn (4.28). One first of all uses Wick's theorem (1950 ; see also
Gaudin 1960 a) to sort out the various terms of nth order according to whether the
indices of the creation and annihilation operators are different or the same. One
is finally left with a product of ' contractions ' of the form (Ta"l,+(P)BI,.(P'))o
and
these are equal to the Green functions (compare Eqn (3.1) ),
( Ta"kt(P)
a"k'(P')>O
= G5 Qa,t,;,@,
P').
. . . . . .(4.29)
T o find all diagrams corresponding to the nth order term one proceeds as follows
(compare Goldstone 1957 and Hugenholtz 1957 ; for the sake of simplicity we
shall take n = 3). Draw 3 horizontal wavy lines (see Fig. 1) numbered 1, 2 and 3,
corresponding to the interactions P(P1), P(PJ and P(P,). On the left one draws
two lines, one joining from below corresponding to dm and one joining from above
corresponding to a"?+ ; similarly on the right two lines corresponding to a", and
a",+. This corresponds to a scattering where two particles in states E , and E , are
scattered into the states
and E ~ . We shall forget for the sake of simplicity the
fact that there is a symmetry between m and n, and between Y and s (for a discussion
of this point see, for instance, Gaudin 1960 b). The only terms which will contribute to the sum are those where there are no loose ends, and neglecting the symmetry
327
effects, just mentioned, we see that only diagrams of the kind depicted in Fig. 2
contribute to the term of third order. We do not wish to pursue this discussion
any further, but refer to the literature.
AE
P PS
LIED TO ATOMS,
$ 5 . D E N S I T YM A T R I XT E C H N I Q U
M O L E C U L EASN D N U C L E I
Until now we have been concerned with the properties of the density matrix,
rather than with the density matrix as a means of tackling physical problems.
This will be the subject of $9 5 to 9. We start with a discussion of many-body
systems in this section and the next. I n the present section we shall discuss the
application of density matrix techniques to atoms, molecules and nuclei, that is,
their application in the Hartree-Fock and Thomas-Fermi-Dirac theories. I n that
case we are dealing with temperature independent density matrices, or, perhaps
more exactly, with density matrices applied to systems at absolute zero, where the
systems are in their ground state. Quantum chemists have recently become
interested in the use of density matrices, especially the so-called reduced density
matrices, and as a result the literature is now growing fast. We may refer here to
papers by Dirac (1930 b, 1931), Husimi (1940), Corson (1951), Kinoshita and Nambu
(1954), Lowdin (1954, 1955 a, b, c, 1959, 1960), Kaschluhn (1955 b), Macke
(1955 a, b), Matsubara (1955 b), Mayer (1955), McWeeny (1955, 1956 a, b, 1957,
1959 a, b, 1960), Blatt (1956), GombAs (1956), Chirgwin (1957), Golden (1957 a, b,
1960 a, b), Koppe (1957), Mizuno and Izuyama (1957), Tredgold (1957), Ayres
(1958), Blatt and Matsubara (1958), Husimi, Kitano and Nishiyama (1958),
Kobelev (1958 a, e), March and Young (1958), McConnell (1958), SarolCa and
Koppe (1958), Bopp (1959), Ehrenreich and Cohen (1959), Fradkin (1959 c),
Goldstone and Gottfried (1959), Hubbard (1959), Martin and Schwinger (1959),
Naqvi (1959), Peacock and McWeeny (1959), Shull (1959), Alfred (1960), ter Haar
(1960), Hall (1960), McWeeny and Mizuno (1961), McWeeny and Ohno (1960),
Penrose (1960), Thouless (1961) and Young and March (1960). I n writing this
section we have also used repeatedly unpublished notes of lectures given by
D.ter Haar
328
L. Rosenfeld at Copenhagen in 1958-59. For a general discussion of the HartreeFock and the Thomas-Fermi theories we refer to the literature (for instance,
Gombhs 1956, Hund 1956, Lowdin 1956, March 1957, ter Haar 1958).
In the present section we consider again systems of particles interacting through
two-body forces so that the Hamiltonian is again of the form
where we now write
A = Bo+P,
. . . . . .(5.1)
A, = -&A,i
. . . . . .(5.2)
,.
P = +c..v...
2.3
kJ
. . . . . .(5.3)
T h e problem we shall consider in this section is the determination of the ground

state energy and wave function. T h e Schrodinger equation is
HY
. . . . . .(5.4)
ET,
where y. is a function of the coordinates (including the spin coordinates) of all N

particles in the system which is suitably symmetrized according to whether we are
dealing with bosons or fermions, If y. is the exact wave function, the energy is
given by the equation
. . . . . .( 5 . 5 )
Y*HAYdrAy.
Using Eqns (5.1) to (5.3) and the fact that Y: is completely symmetric or antisymmetric in all particles, we can write Eqn (5.5) in the form
&A7 Y!*[A,+A2+ (N-1)
q2]Yd~~y.
. . . . . .(5.6)
We can now introduce the one- and two-particle density matrices through the
equations (compare Eqn (2.40) and bear in mind that both q and x stood for sets
of coordinates)
n
(x,IjPIx,)
( x , , x 2I ~
=J Y * ( x 1 , x 2 ,.... x s ) Y ( x l , x 2 , .... x s ) d 3 x2 . . . d3xAy,

1
( 2 )x,,
x z ) = [ v y x l r , x z , X 3 , ....
. . . . . .(5.7)
~ ( x , x, 2 ,x 3 , .... XLv) d3X,
... d3XAy,
. . . . . .(5.8)
where here and henceforth we shall omit explicit reference to spin coordinates ;
if necessary J ... d 3x will be assumed to include summation over spin coordinates
and x will be assumed to stand for both spatial and spin coordinates.
From Eqns (5.6), (5.7) and (5.8) we get
E
or
where
N T r $(,)A,
+ ;N(N- 1) T r jY2) PI2,
+iVT r jY2) A,
A = A, +A,+ ( N -
1>q2.
. . . . . .(5.9)
. . . . . .(5.10)
. . . . . *(5.11)
We see that the energy is completely determined by the two-particle density

matrix, provided there are no three- or more-body forces.
329
So far we have not made any assumptions about the form of Y, and all equations
are exact. T h e first application of our equations will be one discussed by Bopp
(1959). We note that we have expressed the energy exactly in terms of the twobody energy operator (5.11), and one might ask whether one could reduce the
AT-particle problem to a two-body problem. Let us assume that we know the
complete orthonormal set of the eigenfunctions, +, of and the corresponding
eigenvalues, E,,
h^+n(xl,~ 2 =) 6% +,(XI, ~ 2 ) .
. . * . * . (5.12)
We can use these 4, to expand Y and also to expand
=
Xn+n(x1,XA~n(x3,
...>x.\),
(x1, x:! I P Z 1)XI, xz)

where the p,,
p,,
$(z),
c,,,,
4n(X1, X d
+n2*(Xl1,Xz)P,m,
. . . . . .(5.13)
. . . . . . (5.14)
are given by the equation (compare Eqn (5.8) )

= Jxn(x3,
..., xv) X,*(X3, ... ,X-y) d3X3 ... d3X-v.
. . . . . .(5.15)
Substituting expression (5.14) into Eqn (5.8) and using Eqn (5.12) we see that only
the terms with p,, remain in the expression for E and we find
&NC,p,,
E,.
. . . . . .(5.16)
We have, of course, not yet gained anything. We may know the E , from solving
the two-body problem (5.12) but the pn, can only be found from a knowledge of
the N-body wave function (see Eqn (5.15)). From Eqns (5.13) and (5.15) we
see that p,, is the probability amplitude of the state 4, in Y. One can, however,
find a good approximation to E by using an argument which is similar to the one
used in explaining the electronic structure of the elements in the periodic system.
I n the latter case one fills up the lowest N levels in the self-consistent field.
Similarly one can argue that one should fill up the lowest & N ( N - 1) pair-states,
that is, put
p,, = 2[N(N- 1)]-1, n = 1, ..., $ N ( N - 1) ;
. . . . . .(5.17)
Pn, = 0,
n > * N ( N - l),
which satisfies the normalization condition
CnP,,
1.
A more detailed discussion (see Bopp 1959) shows that the p,,
condition
P,, 2 [ N ( N - 1)1-l,
=
. . . . . .(5.18)
must satisfy the
. . . . . .(5.19)
so that the choice (5.17) will lead to a lower bound for the energy as the lowest
$ N ( N - 1) pair-levels are over-occupied and the higher pair-levels underoccupied . We get thus the following relation for the energy
E 3 (A- l ) - l c
E,
. . . . . .(5.20)
where the prime on the summation sign indicates that only the lowest & N ( N - 1)
levels are involved.
D.ter Haar
330
The next step in Bopps argument is the use of experimentally known two-body
energy levels to evaluate E for specially chosen atoms. For a nucleus with charge
Ze surrounded by N electrons, the operator is given by the equation
A
62
--(V
2m
12+V22)-Ze2
1 1 (Ar-l)e2
-++--.
i*,*2j
. . . . . .(5.21)
*12
Changing the variables by the substitution

xi =
. . . . . .(5.22)
(N- l ) - l ~ i , 2= ( N - l)-lZ,
we get
(*:-7+1r, 1+?
. . . . . .(5.23)
which, apart from the factor ( N - 1)2, is the helium atom problem with a nuclear
charge 2. T h e energy levels of A are thus known for integral values of 2and
we can use this fact to evaluate E for appropriately selected ions. For instance :
Be- has Z = 4, N = 3, so that 2= 2, and
3
2 C {(He).
i-1
This leads to a value of 3 171 038 cm-1 for E which is less than 1% larger than the
observed value. For the case of Os+ we have Z = 8, N = 3, 2= 4 so that
3
ei(Be2+)
i=l
which leads to E = 14 107 380 cm-1 as against the observed value of 14 104970 cm-l.
We turn now to a discussion of the Hartree-Fock and Thomas-Fermi theories
of atoms (and nuclei). T h e density matrix is especially suitable to discuss the
Thomas-Fermi theory, as it is a statistical model. I n it one assumes the particles
to be independent, but moving within a constant potential, the value of which is
adjusted self-consistently using Poissons equation. In the Hartree-Fock theory
one tries to describe the system by ascribing to each particle a one-particle wave
function. This wave function is obtained by writing down for each particle a
Schrodinger equation with a potential energy which is due to all the other particles.
T h e wave functions are varied until the whole set of equations is self-consistent.
I n this case we are thus also essentially describing the system in terms of independent
particles, or rather, quasi-particles. If to each particle we can assign a one-particle
wave function, the wave function of the whole N-particle system will be of the form
1
V
CCp7jP .JJ$ ! p i )
. . . . . .(5.24)
a=1
where the t , ! ~ ~ ~ indicates a wave function corresponding to a one-particle state ki

which is a member of a complete orthonormal set, i stands for all coordinates of
the ith particle, Pi for a permutation of the N i, 9 is + 1 or - 1 according to whether
331
we are dealing with bosons or fermions, the summation is over all possible permutations of the N i , and C is a normalization factor which is given by the equation
c-2 = N ! ,
if all states ki ( i = 1, ...,N ) are different.
find for the one-particle density matrix
(xI
. . . . . .(5.25)
I n that case we can use Eqn (5.7) to

N
I x ) = N-l ic= l +k,*(X)# k , ( X )
where we have used the orthonormality of the y!rk,

that
(XI
. . . . . .(5.26)
From Eqn (5.26) we find easily
[i;(1)]21x)= ~ ( x l P ( l ) j x j d S X I ( xP(1)Ix)
~
=
or
N-yx
[i;(l)]Z
IX I ) ,
I
N-1
P(l)*
. . . . . .(5.27)
We notice that this result is independent of the statistics, but is a consequence solely
of the fact that all ki were different. T h e physical significance of Eqn (5.27) is
clear : it means that
has eigenvalues 0 and N - l , that is, that there are N states
which each have an equal probability of being occupied, and that the other oneparticle states are not occupied. I n as far as the only possible wave function (5.24)
for a many-fermion system is the one where all the k, are different (Slater determinant) while there are other wave functions possible for many-boson systems, there
is a certain amount of justification for calling Eqn (5.27) a Pauli condition (for
instance, Young and March 1960). It is rather unfortunate that in the literature
of quantum chemistry the reduced density matrices p Q C ( I ) have become the normally
used ones. They are related to our P ( l ) by the relation
PQCl)=
N p ,
. . . . . .(5.28)
so that they satisfy the equations
T r pac(1) = N ,
[ j 3 Q C ( 1 ) ] z= & ( I ) ,
. . . . . .(5.29)
and the latter condition is sometimes wrongly interpreted as indicating a pure state.
T h e N-body system is, of course, in a pure state, but not the one-particle systems
described by the one-particle density matrix. I t is correct to say that Eqn (5.27)
is the necessary condition that y. can be described by Eqn (5.24) with all k, different,
or for the special case of fermions that the system can be described by a Slater
determinant. I t is interesting t o note that the possible confusion between $ ( l ) and
P Q C ( l )can be traced back to Diracs original papers (1929, 1930 a, b, 1931 ; for a
discussion of this point see ter Haar 1960).
From Eqn (5.8) we get for the two-particle density matrix, again assuming all
ki to be different,
N ( N - 1) ( X I , x 2 I p ( 2 ) I x l , x2) = ( x l I
I x l ) (XJ I x2)
+ q ( x l I jY1)I x 2 )( x 2I I x l ) .
. . . . . .(5.30)
We note first of all that jY2) possesses the correct symmetry properties. This
follows also immediately from Eqn (5.8). Secondly, we note that for the case of
D.ter Haav
332
the wave function (5.24) the two-particle density matrix (and, indeed, all other
many-particle density matrices) can be expressed in terms of the one-particle
density matrix. T h e quantum-chemical two-particle density matrix pQc(2)is given
by the equation
p Q c ( 2 ) = & N ( N - 1) p w .
. . . . . .(5.31)
We note one more property of the single-particle and two-particle density matrices.
Their diagonal matrices have the following physical meaning. From Eqn (5.26)
it follows that
. . . . . .(5.32)
I x> = N -l I hi(x) 12,
(x
ci
I x) is just the normalized particle density in the system.

and we see that (x I
Similarly (xl, x2 p(2)I xl, x2) denotes the probability density for finding simultaneously a particle at x1 and a particle at x2.
We shall from this point onwards in this section restrict our discussion to
fermion systems (7 = - 1) where the wave function (5.24) corresponds to a Slater
determinant. Moreover, we shall be interested in one particular state of the
N-particle system, namely the ground state. We have seen earlier in this section
that the energy can be expressed in terms of the two-particle density matrix and
one method of approach is by considering possible expressions for
This
method was first suggested by Mayer (1955) and has since then been explored by
various authors (Koppe 1957, Tredgold 1957, Mizuno and Isuyama 1957, Young
and March 1960). On the other hand, we have also seen that for a Slater determinant the two-particle density matrix can be expressed in terms of the one-particle
density matrix and we might try to find a suitable expression for jY1)(see Golden
1957 a, b, Naqvi 1959). We shall briefly consider both methods.
From Eqns (5.10) and (2.9) we see that we have for the energy of the system
E
where
. . . . ..(5.33)
&NTr&(2),
A is given by Eqn (5.11) while p(2) must satisfy the normalization condition
Trp^(2)= 1
. . . . . .(5.34)
and the antisymmetry condition
I p ( 2 ) I Xll,
(xl, x2
I p ( 2 ) I x21,xl)
x21) = - (XI, x2
= - (xz, XI
I p ( 2 ) I XI1, x21).
. . . . . .(5.35)
Taking care to see that Eqns (5.35) are satisfied at all times one can use Eqns (5.33)
and (5.34) for a variational method to evaluate p ( 2 ) and then E.
I n the other method we consider only $(l). From Eqns (5.9) and (5.30) we
have in terms of
E = Trf?p^(l),
. . . . . .(5.36)
where
. . . . . .(5.37)
( x l A l x l ) = N(XIf?JX)+4(XI Blx/),
(XI
qxl) =
(XI
- (x
q 2 ~ X ) d ~ X ( X ~ p h6(x-x)
(~~~X~)
I q2I
XI)
(x
1 XI).
. . . . . .(5.38)
333
and bear in mind the condition

If we vary expression (5.36) with respect to
through 0, we find the following
(5.27) on $(I) and the fact that A contains
equation to be satisfied by $(l),
[A,$]- = 0,
. . . . . . (5.39)
A = NQl+ D.
with
. . . . . .(5.40)
If we had varied the $lc which occur in the Slater determinant instead of
we would have found for the $k the Schrodinger equation
A y k = Ek*h.,
$(l),
. . . . . .(5.41)
which is just the Hartree-Fock equation.

I n the case of the Thomas-Fermi theory we are interested in very large systems.
If the system were infinite, ( x I jY1) I x) should depend on x - x only, and we may
assume that this is still correct for the actual systems studied, Moreover, one
assumes in the Thomas-Fermi method that N is very large. If N were infinite,
the off-diagonal elements of $(l) would be zero (this follows from Eqn (5.26), and
the fact that we would then sum over the complete set : see, for instance, Kramers
1957, p. 129). It is thus plausible to assume that for the actual system, the offdiagonal elements of fi(l) will be small and can be expanded in a Taylor series in
x - x. T h e method to be used now is the following one. We replace the exact
Hamiltonian by a model Hamiltonian Qb1 (compare the use of a harmonic oscillator
well in shell model discussions), and require jY1) to satisfy the following conditions :
(ii) the normalization condition (2.9),
(i) Eqn (5.27),
(iii)
[AlI,$I-
0,
. . . . . .(5.42)
and (iv) the requirement that the model energy EbIgiven by the equation
E&l= TrGbI$t1)
. . . . . .(5.43)
be a minimum. Once we have found $(I) the total energy is given by Eqn (5.36).
T o a first approximation we can use for the t,blCin Eqn (5.26) plane waves. We
then get from conditions (i) to (iv) an energy density E(X) which is given by
22 12
E = -- ( 6 ~ 9 ~ n513
1 3 + n&(x),
. . . . .(5.44)
8m 5
where I& is the potential energy in the model Hamiltonian (assumed to a first
approximation to be a constant) and n the particle density,
. ..
*
* .(5.45)
n(x) = A\(x I $(I) I x),
where No is the number of particles per unit volume. I n the next approximation
one uses the fact that N is finite and one introduces the variation with x of both
KI and n, and requires the total energy Je(x) d 3 x to be a minimum. T h e result is
then the following equation for n
where E,, is the energy of the most energetic particle in the system. Eqn (5.46) is
the so-called von Weizsacker equation (1935). We refer to Naqvis paper (1959)
for details.
D.ter Haar
334
$ 6 . T H ED E N S I T YM A T R I XI N S O L I DS T A T EP H Y S I C S
Both density matrix and Green function techniques have been used in recent
years to discuss many solid state problems, both equilibrium and non-equilibrium
problems, I n the present section we shall be concerned with equilibrium properties
and in the next section with non-equilibrium problems. Once again we can only
discuss a few selected topics and must refer to the literature for a more comprehensive
coverage. For this section we may mention the work on many-electron systems
(Zubarev 1954, Mayer 1955, Koppe 1957, Mizuno and Izuyama 1957, Tredgold
1957, Edwards 1958 a, Ehrenreich and Cohen 1959, Goldstone and Gottfried 1959,
Kanazawa and Watabe 1960, Kanazawa, Misawa and Fujita 1960, Young and
March 1960), on diamagnetism of a system of free electrons (Sondheimer and Wilson
1951, Nakajima 1955, Nitsovich 1959, Kanazawa and Matsudaira 1960), on diamagnetism of conduction electrons (Nakajima 1956, Enz 1960 a, b, Hebborn and
Sondheimer 1960), on electron-phonon systems (Matsubara 1955 a, b, Nakajima
1955, Ichimura 1956), on semiconductors (Bonch-Bruevich 1956 c, Krivoglaz and
Pekar 1957 a, b, Osaka 1959), on superconductors (Salam 1953, Schafroth, Butler
and Blatt 1957, Koppe and Muhlschlegel 1958, Valatin 1958), on ferromagnetism
(Bogolyubov and Tyablikov 1959, Brout 1959), and on surface tension in liquids
(Ono 1958). I n this section we shall restrict ourselves to a discussion of the diamagnetic susceptibility of a system of electrons, including a brief discussion of
Schafroths treatment (1951) of the Meissner effect.
I n discussing the diamagnetic susceptibility we are dealing with an equilibrium
aspect and one could do this by evaluating the partition function. We s%win 4
(Eqns (4.4) and (4.5) ) that this can be done by finding the trace of e-flH. This
is usually done by finding all energy levels of the system and writing (we restrict
ourselves here to petit ensembles)
. . . . . . (6.1)
=
exp ( -PE,),
z c,
where the summation is over all energy levels of the system. There is, however,
no reason whatever to restrict oneself to this particular form of the trace. If we
introduce once again the operator $(P) by the equation
e+k,
. . . . ..(6.2)
@pp = -H$,
. . . . . .(6.3)
. . . . . .(6.4)
which satisfies the Bloch equation (4.6),

we have
T h e magnetic susceptibility
Tr$(P).
x follows from the usual formula

=
(p~~-lazjaz,
. . . . . .(6.5)
where 2 is the magnetic field, which we shall assume to be uniform.

We shall assume the particles (electrons) in our system to be independent, so
that the many-particle partition function is the Nth power of the single-particle
partition function. If we take x to be the susceptibility per particle, we can still use
Eqns (6.2) to (6.5), but with
being the single-particle Hamiltonian which is of
the form
.(6.6)
335
where A is the vector potential,

A = $[%A
. . . . . .(6.7)
x],
- e the charge of the electron, and m the electron mass. It will be noted that we
are not introducing the electron spin and are thus restricting the discussion to
Boltzmann statistics. We refer to the paper by Sondheimer and Wilson (195 1) for
a discussion of the Fermi-Dirac case (we may also refer to the discussion in 5 9.4
of ter Haar 1954 ; the I?v,Bo of that section is related to the diagonal element of
the (x I$(P) I x) we have here).
T o evaluate $(P), it is convenient to change to the coordinate representation.
As was noted in 0 2, any complete orthonormal set pn(x) can be used, and it can be
shown that (compare Eqn (2.48) )
(x
IP(P) I x> = c
. . . . . .(6.8)
From Eqn (6.8) and the orthonormality of the
where
operates on pm(x) only.
set y n it follows that
(x IP(0) x> =
m Pm(X) exp ( -PG) Pm(X),
cm p,*(x>
P,(X)
= S(X
- x).
. . . . . .(6.9)
We must thus find a solution of Eqn (6.3) which satisfies the boundary condition
(6.9). We shall assume the magnetic field to be along the z axis so that we can
write Eqn (6.3), using Eqns (6.6) and (6.7), in the form
If 2 = 0, Eqn (6.10) reduces to the heat conduction equation with the solution
(for instance, Carslaw and Jaeger 1947)
P2.=o
v3I2exp [ - VV(X
- x . x - x)],
I,
m/2nh2P.
. . . . . .(6.11)
If # # 0, one finds a more complicated expression (see Sondheimer and Wilson

(1951) for a derivation of this equation)
(x
IP(P) I x> = f(P>exp [ - (ie 2 / 2 & c )(.y -Y.)
-g(P) ((2- X l 2 + (Y-YYI

. . . . . .(6.12)
- VI,(z - 2 ) 2 ] ,
where f ( P ) and g(P) are given by the expressions

e 2
g = -coth4A c
e7i2P
2mc
f=
e 2
e7i2PP
__
cosech ___
2hc
2mc
. . . . . .(6.13)
*
The partition function per unit volume is given by

2 = v-l/(x\p(p)lx)dx,
. . . . . .(6.14)
and from Eqns (6.12), (6.13) and (6.14) we get

2 = ~ ~ ~ [ y / s i n h yy] ,= PpBA?,
where p B = eR/2mc is the Bohr magneton.

expression for x, if we use Eqn (6.5),
x=-
. . . . . .(6.15)
Eqn (6.15) leads to the following
PB(COth y - y - l ) ,
. . . . . .(6.16)
D.ter Haar
336
which for weak fields reduces to the Landau expression (1930)
. . . . . .(6.17)
= -13 P P B 2 *
If we are dealing with conduction electrons, we must replace the Hamiltonian

of Eqn (6.6) by
. . . . . .(6.18)
A = A+ P,
where
I? is
the periodic lattice potential. It is no longer possible to evaluate

exactly, but one uses an expression in powers of #, the term in A?2
giving the field-independent part of the susceptibility (Hebborn and Sondheimer
1960).
T o conclude this section we consider briefly the related question of the Meissner
effect, that is, the exclusion of a steady magnetic field from a superconductor.
We consider the external magnetic field as a perturbation, and write for the
Hamiltonian
A = ri,+A,,
. . . . . .(6.19)
(xl$(P)lx)
where H , is the kinetic energy and H , given by the equation (we are now considering
a system which is not necessarily uniform and Z? is once again the Hamiltonian of
the total system)
HI = - c-I
(A(x). j(x)) d 3 x ,
. . . . . .(6.20)
where j is the total current density given by the expression :

j
e2
e%
($* V$ - $V$*)-. -** *A*

mc
2mi
. . . . . .(6.21)
T h e components J,(x) of the average current density are given by the equations
J,W
<&m.
. . . . . .(6.22)
As this average is evaluated using the equilibrium density matrix (4.2) we get to a
first approximation for J, a linear expression in the components A,, of the vector
potential
r
(6.23)
From Eqns (6.20) and (6.22) and the expansion
. . . . . .(6.24)
exp ( -Pa> exp ( - PAO)[ I - PQll
it follows that the K,,, are proportional to the correlation functions C$JX - x)
given by the equation
. . . . . .(6.25)
&V(X - x) = ( ~ v o 4 ~ p ( x ) D o ,
where the average is taken over the unperturbed equilibrium ensemble, that is,
the ensemble with $(P) = exp (-PI?,). We see that the $ p v are essentially an
example of the correlation functions (3.14).
The behaviour of the Fourier transforms of the Kiiydetermines the diamagnetic
properties of the system. From general symmetry considerations it follows that
the pv-dependence of the Fourier transforms K J k ) can be expressed as follow-s
K,VW = (A2 8, -A,,
AV)
K(k).
. . . . . .(6.26)

If K(k2) has no singularity at k
however, K ( P ) has a pole at k = 0,
337
0 one finds ordinary diamagnetism.
K(k2) = olk-2+ K,(k2))
If,
. . . . . .(6.27)
where K,(k2) behaves regularly at k = 0, the term K,(k2)leads to ordinary diamagnetism, but the first term leads to an extra term, J,, in J which can be shown
to satisfy the London equation
. . . . . .(6.28)
[V A J,] = d3,
from which the Meissner effect follows in the usual way. We see from this that
the Meissner effect is related to a singularity in the Fourier transform of the kernel
K J x - x), that is, it is related to long range correlations in the system.
9 7.
N O N - E Q U I L I B R I UPM
R O C E S S E; S T R A N S P OTRHTE O R Y
In the preceding sections we have mainly been concerned with systems in
equilibrium. T h e question arises, however, how equilibrium is reached, and what
happens if we are dealing with a system which is permanently in non-equilibrium,
but which has reached a steady state ? Such situations are described by transport
equations. T h e basic, exact equation of motion is Eqn (2.37), and this is the
equation from which the equations discussed in the present section are derived.
We shall briefly discuss the derivation of a transport equation starting from Eqn
(2.37), but the main discussion in the present section will be of the evaluation of
transport coefficients without the intermediary of a transport equation, a subject
which has been developed very extensively in recent years and which was started
by the fundamental papers of Kubo (1956, 1957 a) and Nakano (1956, 1957, 1959,
1960 a, b, c). We may refer also to papers by Callen and Welton (1951), Nakajima
(1956), Kohn and Luttinger (1957), Argyres (1958, 1960 a, b), Edwards (1958 b),
Greenwood (1958), Lax (1958), Luttinger (1958), Luttinger and Kohn (1958),
Mattis and Bardeen (1958), Adams and Holstein (1959), Argyres and Roth (1959),
Bench-Bruevich (1959 e), Chester and Thellung (1959), Klinger (1959 a, b, c, d, e),
Konstantinov and Perel (1959, 1960), Kubo, Hasegawa and Hashitsuma (1959),
McLennan (1959), Montroll and Ward (1959), Zigenlaub (1959), Green (1960),
Kanazawa and Watabe (1960), Yokota (1960) and Zubarev (1960). Among the
papers dealing with the more general problem of irreversible processes and transport equations we refer to those by Born and Green (1946, 1947 a, b, 1948), Green
(1947), Mori and Ono (1952), Ono (1953, 1954, 1958), Ross and Kirkwood (1954),
Van Hove (1955), Kummel (1955), van Kampen (1956), Mori (1956, 1958, 1959),
Gurzhi (1957), Husimi, Kitano and Nishiyama (1958), Klimontovich and Temko
(1958), Prigogine and Balescu (1959 a, b), Prigogine and Ono (1959), Helfand (1960),
Helfand and Rice (1960), Klimontovich and Silin (1960), Nishikawa (1960),
von Roos (1960), Saitb (1960), and Zelazny (1960). We also refer to the Proceedings
of the 1959 Varenna Summer School (Varenna 1959) and the 1956 Brussels Conference (Prigogine 1958). Finally, we refer to the discussion in 9 where we shall
consider relaxation processes.
I n classical statistical mechanics the counterpart of the equation of motion for
the density matrix
iti; = [Q,p]. . . . . .(7.1)
22
D.tev Haar
338
is the Liouville equation. Both in Eqn (7.1) and in the Liouville equation we are
dealing with the total N-body system. T h e ordinary transport equation, on the
other hand, involves, if possible, only quantities referring to a single particle. I n
classical theory the transport equation is obtained by integrating over the coordinates
of all but one of the particles. As the Hamiltonian occurring in the Liouville
equation will in general contain terms referring to two particles (two-body interaction terms), the transport equation for the single-particle distribution function
will contain the two-body distribution function. One can either make some simplifying assumptions about this two-body distribution function, expressing it in terms
of the single-particle distribution function, or one can set up a transport equation
for the two-body distribution function by integrating the Liouville equation over
the coordinates of N- 2 particles only. T h e resultant equation will, however,
contain the three-body distribution function, and we have to continue the process.
We may refer here to a similar situation which we met in 3 3 with respect to the
Green functions.
I n quantum mechanics a similar situation arises, but there is an additional
complication in that the density matrix i; occurring in Eqn (7.1) is not the immediate
counterpart of the N-body distribution function f(xl, , .,,xAy; pl, . , , pN). T h e
latter is a function of N coordinates xi and N momenta pi, while the density matrix
in coordinate representation (xl, .... xN i; I xl, .... xN)is a function of N coordinates
xi and N coordinates xi. I n working with a transport equation, and especially
in evaluating transport coefficients from a solution of the transport equation, it is
often more convenient to deal with a distribution function such as the classical
one. One can actually construct from the density matrix such a distribution
function, the so-called Wigner distribution functions (Wigner 1932)
~
(XI,
....x g ; pl, .... p,y) = (za)-3~J<Xi- &%TiI i; I xi+ $&Ti) exp
From Eqn (2.7) for the average value of a quantity A^,
(A^)
Tri;A,
- i I:(Tj.pj)jd3s7.
j=1
AV
. . . . ..(7.2)
. . . . . .(7.3)
Eqn (7.2), and the well-known properties of Fourier integrals, one can prove that
for any function A(xi, p i ) of the coordinates and momenta
(A)
pi)fr(xi; pi)d3-1Xd3-Vp,
. . . . . .(7.4)
so that t h e f have, indeed, the properties of distribution functions. Moreover,

in the limit as R - t O (classical limit) the f go over into the classical distribution
functions.
From Eqns (7.1) and (7.2) it follows that f satisfies the equation of motion
P
- i Z p ( x i , pi) =
( 2 ~ ) -(H(qi
~ ~ + &Rki,yi - ~ R T -H(qi
~ )
- @ k i , y i ++hi))
x f ( y i , qi) exp ~ C , [ (. qj
T ~- pi) + (k, .y j - xi)] d3 T d31Vk d32rq d3x y,
. . . . . .(7.5)
where H(xi, pi) is the Hamiltonian of the system. As E--+ 0, the equation tends to
the Liouville equation.
339
By integrating f over the arguments of all particles, but one (or two, ...) we
obtain a one-particle (two-body, ...) distribution function, and by similar integrations
of Eqn (7.5) we obtain the transport equations for the quantum mechanical (Wigner)
distribution functions. T h e situation is now completely similar to the one in
classical statistics.
I n view of the fact that the transport equation for the single-particle distribution
function contains either the two-body distribution function or involves simplifying
assumptions, many authors have discussed the possibility of obtaining transport
coefficients without solving a transport equation. I n all these papers one assumesas is usually done in a discussion of the transport equation-that the perturbing
field, density gradient, temperature gradient, ... are so small that one can restrict
the discussion to an approximation which is linear in the perturbation. If this is
the case, one can show that the transport coefficients, that is, typical non-equilibrium
quantities, can be evaluated using an equilibrium density matrix. I n fact, such
properties as electrical conductivity are directly related to the response of a system
to an external force, and Callen and Welton (1951) have shown that this response
in turn is related through Nyquists fluctuation-dissipation theorem to time
correlation functions, that is, to Green functions (see 5 3). Of course, in as far as
one uses the equation of motion (7.1) for the density matrix to obtain the expressions
for the transport coefficients, one has used a transport equation, but only one for
an N-body system, not the one for the one-particle distribution function. One
can show that the expressions obtained for the transport coefficients are the same as
those obtained from the usual transport equation in the appropriate limits of
sufficiently weak perturbations.
T o derive an expression for the electrical conductivity we first of all consider
the general problem of the response of a system to a time-dependent, oscillating
perturbation which is switched on adiabatically. That is, we assume the total
Hamiltonian to be of the form
A =fl0+l3(t),
. . . . . .(7.6)
where
A(-co)= 0, A(t)= edeiWtP ( E + o , ~ > o ) ,
. . . . . .(7.7)
with P a time-independent operator and w the frequency of the perturbation.
assume the density matrix at t = --CO to be the equilibrium density matrix
p( - C O )
..... .(7.8)
po = 2-1exp ( - PAo),
where 2 is the partition function : 2 = T r [ exp ( - P f z O ) ] .

at a later time we write
A
$(t)= i;O+AP,
We
For the density matrix
. . . . ..(7.9)
and from Eqns (7.1), (7.6) and (7.9) we get, if we neglect second-order terms,
A
i7&
[Ao,
Ap]-
+ [I?,pol-.
......(7.10)
Introducing the Heisenberg operators
6.
exp (;Aotiti) fi exp ( - iAotih),
we have
ih@
Po]- exp (- ig0t/R),
= exp (!Ao
t / h )[A,
..... .(7.11)
......(7.12)
D.ter Haar
340
with the solution
A
exp [ig0(~
- t)/h] [A'(T),
80]-exp [ - Z'Z?~(T - t)/h] d ~ . . . . . .(7.13)
From Eqns (7.3), (7.9) and (7.13) we get for the average value
( A ( t ) )= (A^),+
([A^(t),P(T)]-)o eiW7+E7d7)
(%)-I
(A),
. . . . . .(7.14)
where ( ,.)o indicates an average over the equilibrium ensemble j30.

If we apply this to the case of the electrical conductivity, A^ is replaced by one
of the components j, of the current and g' is of the form
. . . . . .(7.15)
A' = - e&( E . xj) eiWt
where we have now dropped the adiabatic factor est and where E is the amplitude
of the electrical field strength. We shall apply Eqn (7.14) to the case A^ = j p ,
rewriting it slightly, using the fact that (j,),, = 0, and using Eqn (7.11) for jiLH.
T h e result is
(i,)
(Z'/h)Tr/-aexp [iflo(,
- t)/h][exj( E . xj), fjO]- exp [ - ig0(T
- t)/h]eiwr$, dT
(i/h)Trle[e&(E, xj), pÔ]-
. . . . . .(7.16)
eiw(t-7)dT.
JhH(7)
T h e complex conductivity tensor, which is defined by the relation
(j,)
. . . . . .(7.17)
= oPvEyeiwt,
follows from Eqn (7.16) and is equal to

opUy
= (i/h)TrIome-i~7[e&xjv,& - f , " ( T )
dT.
. . . . . .(7.18)
We shall use the identity

P
[fi,exp(--/3g0)]- = -i?iexp(-/3g0)/
exp(Xl?o)~exp(-hZ?o)dh,
. . . . . .(7.19)
where Eqn (2.38) has been used. This equation follows easily, as was observed
by Kubo (1957 a) if one uses a representation in which
is diagonal. As fjo is
given by Eqn (7.8))we find from Eqn (7.19)
no
[e& xj,,
801-
(ih/Z)exp ( -/3go)/
exp (Qn)JL,
exp ( - hA0) dX,
. . . . . .(7.20)
or, using Eqns (7.18), (7.20) and (7.11))

ojlv
- z-'jad~
e-iWT/o'{jvH(- iEh)j,"(T))n dh,
n
. . . . . .(7.21)
341
and we see once again the importance of the correlation between currents (compare
Eqn (6.27) of the previous section).
Equation (7.21) has been used by Montroll and Ward (1959) and Konstantinov
and Perel (1960) as the basis of diagram techniques to evaluate transport
coefficients.
3 8.
POLARIZATION
SC
, A T T E R I NAGN D A N G U L A R
CORRELATIOK
EXPERIMENTS
We mentioned in the introduction that often one can use an operational approach
to the density matrix. This is especially the case when one discusses polarization
or scattering experiments, as one is in that case interested in only a few of the
many parameters which specify the system and one can use a density matrix which
refers to only those degrees of freedom which are studied experimentally. T h e
simplest example is the polarization of a beam of electrons which we shall discuss
in some detail, A related problem is that of the polarization of light which we
shall also consider here, without going into a very detailed discussion. Our final
discussion will be of scattering and angular correlation experiments. For a
consideration of such experiments elaborate matrix techniques have been developed,
some of which are based upon the density matrix, but we must refer to the literature
for a discussion of such techniques (see, for instance, Tolhoek and de Groot
1951 b, c, Cox and Tolhoek 1953, Tolhoek and Cox 1953, Hartogh, Tolhoek and
de Groot 1954, de Groot and Tolhoek 1955, Huby 1958). We can only touch
upon some aspects of density matrix techniques ; for more details of those
techniques as applied to polarization, scattering, and angular correlation experiments
in nuclear physics we may refer, for instance, to the papers by Tolhoek and de Groot
(1951 a), Lipps and Tolhoek (1954 a, b), Kotani (1955), Tolhoek (1956), Fano
(1957), Hagedorn (1958) and Zaidi (1959). We may also mention some unpublished
lecture notes based upon lectures delivered by L. Rosenfeld to Nordita in
Copenhagen which have been of great use in writing this section. T h e use of
density matrices in discussing the polarization of electromagnetic radiation is
based upon the fundamental ideas of Stokes (1852). Recently Fano (1949, 1957)
and Wolf and Roman (see, for instance, Wolf 1954, 1959 a, b, 1960, Roman 1959,
Parrent and Roman 1960) have discussed this problem in great detail, but we can
refer here only to those papers which are relevant to our discussion of density
matrix techniques. I n the present section we shall consider the polarization
of particles with spin 8 ; the case of larger spin values will be mentioned only
briefly.
Let us consider a beam of particles with spin 4,for instance, a beam of electrons.
If we are only interested in the polarization or spin-orientation properties of this
beam, we have a system of particles with two degrees of freedom, as long as we
neglect negative energy states as we shall do here. The system should thus be
describable by a 2 by 2 density matrix, and we need only 3 independent parameters
to determine fully the density matrix (compare the discussion in 4 2). The physical
situation is completely defined, if we know the polarization vector P, that is, the
average value of the spin vector in the system,
P = (8),
. . . . . .(8.1)
D.ter Haar
342
where 8 is the vector the components of which are the Pauli matrices,
We note that these matrices satisfy the following relations

TrG, = TrG,
TrG,G,
TrGuG, = TrG,G,
. . . . . .(8.3a)
TrG, = 0;
0, TrGx2= TrGU2= TrG2 = 2.
. . . . . .(8.3c)
As P has three components we can use these as the three independent parameters
to determine the density matrix i;. As i; is a 2 by 2 matrix, we can express it in
terms of the unit matrix 'i and the Pauli matrices,
. . . . . .(8.4)
i; = a . l + ( a . 8 ) .
From the normalization condition (2.9) and Eqn ( 8 . 3 ~we
) get
Tri;= 1 =2a,
. . . . . .(8.5)
a = 12 ,
or,
while Eqn (8.1) leads with the aid of Eqns (8.3a) and ( 8 . 3 ~to
)
P = (8)
= T r i;e = T r [a8
+ 8 ( a ,S)]
2a.
. . . . . .(8.6)
Combining Eqns (8.4) to (8.6) we get
......
i; = &[l+(P.S)] = 1
This equation shows that, indeed, p^ is determined, once P is known.

Let us now consider the case where this beam passes through a magnetic field,
and ask what will happen to the polarization of the beam. We could use Eqn (2.37)
for the rate of change of the density matrix and as i; contains P obtain in that way
the equation of motion for P. T h e drawback of this procedure is that one cannot
apply it with the same ease to the case of particles with spin greater than &. We
shall instead use Eqn (2.38) for the rate of change of average values. From this
equation and Eqn (8.11) we get
aP - a(8)
at
at
i
- ( [ e ,ASIA
......
where
is the Hamiltonian referring to the spin coordinate (the spin Hamiltonian)
which is given by the equation
A, = -(@.*)
-gyE(e.x),
. . . . ..(8.9)
where X is the magnetic field, p the magnetic moment of the electron (@= eE8/2mc
with e, electronic charge ; m, electronic mass ; c, velocity of light), and y the
magnetogyric ratio (y = e/mc). If we write Eqns (8.3b) in the symbolical form
SA^]
23,
. . . . . .(8.10)
343
we find from Eqns (8.8) and (8.9)

apjat = ;iy([a, (6.x)]-)
&iy( [%A (6A s)])

aP/at = -?/[*A PI,
=
or,
= -y [ x A
(e)],
. . . . . *(8.11)
which is just the classical equation of motion for the polarization vector. One
could prove Eqn (8.11) starting from the Schrodinger equation, but the proof is
cumbersome. T h e ease with which we could prove Eqn (8.11) is an example of
the advantages of the density matrix. Eqn (8.11) itself is a consequence of
the generalized Ehrenfest theorem (Ehrenfest 1927, Kramers 1957, 3 30) which
states that any quantum-mechanical average will obey the corresponding classical
equation of motion.
We can easily generalize the discussion leading to Eqn (8.11) to the case of
larger spin values. Let j ( > 4)be the largest possible value of the angular momentum of the particles. T h e density matrix will now be a 2j+ 1 by 2j-t 1 matrix,
and apart from the components of the polarization vector P we need other quantities
to determine i; completely, T h e vector P can be called the dipole polarization
vector, and the other quantities which can be used to determine p^ are the quadrupole polarization tensor ( 5 components ; its components and the 3 components
of P are the 8 parameters which are sufficient to determine i; if j = I), the octupole
polarization tensor (7 components ; it comes into play if j > 1)) ..., in general the
2z-polarization tensor (with 21+ 1 components ; for a given j , all multipoles with
1 6 2 j will be involved). W-e do not have the space here to go into a detailed
discussion of the determination of the density matrix for this general case and refer
to Fanos review article (1957) where further references can be found. T h e
polarization vector P is now defined by the equation
P=
(I)j%)
. . . . . .(8.12)
where J is the angular momentum. It satisfies the commutation relation which

can be expressed by a formal equation similar to Eqn (8.10))
A
. . . . . .(8.13)
[J A J] = i i i .
T h e spin Hamiltonian is given by the equation
. . . . . .(8.14)
A, = - Y ( J . X ) ,
and from Eqns (8.12)) (8.13)) (8.14) and (2.38) we find that Eqn (8.11) holds also
for the general case.
Let us now consider the case of the Polarization of electromagnetic radiation.
I n this case there are also two degrees of freedom, and we can thus expect a formal
similarity between this case and the case of the electron beam. I n both cases the
wave functions describing the systems in the ensemble can be written in the form
$k = c ~ $ I + c , + ~ ,
1,2, ...,
. , . , . .(8.15)
where and $2 are two orthogonal wave functions. I n the case of electrons G1may
describe either the case where the electron spin is in the +x-direction, or the case
where the electron spin is parallel to the electron momentum. T h e function g2
D.ter Haar
344
will then describe the case where the electron spin is in the -2-direction, or the
case where it is antiparallel to the electron momentum. I n the case of photons $1
could describe either a plane polarized wave or a right-hand circularly polarized
wave, and y!J2 would then describe a plane polarized wave with its plane of polarization perpendicular to the plane of polarization of the wave described by $1, or a
left-hand circularly polarized wave.
Each of the y!Jk corresponds to a totally polarized beam. If we were dealing
with a pure state all $k would be the same, say, equal to i,b0 :
$o
= C1(O)
C2(O)
. , , . . .(8.16)
$b2.
I n Q 2 we mentioned that one can find a filter such that it would correspond exactly
to a given pure case ; for linearly polarized light a Nicol prism would be such a
filter. We shall call such an instrument here a detector although strictly speaking
a filter is, of course, not used to detect (compare the discussion in Q 10) and we shall
describe it by a density matrix Fdet. T o find the form of this density matrix we
first of all note that it should correspond to a pure state with wave function
p e t = Cldet $1
+ c 2 d e t qJ2.
. . . . . . (8.17)
T h e response of this detector to a state with wave function (8.16) will be given by
the overlap of $o and $det, and the probability W for a response will be given by
the expression
If we are not dealing with a pure state, we must introduce for the beam the density
matrix i;, and we shall require that W is now given by the expectation value of Gdet,
W = (?deb)
. . . . . .(8.19)
Tr @et.
One can easily check that one obtains for W expression (8.18) for the case of a
pure state corresponding to all $k being equal to i/ro if we take for p d e t the density
matrix which would correspond to a wave function $idet :
I Cldet 12
CldetB C2det
Cldet C2det*
I C2det 12
. . . . (8.20)
If the detector is a filter corresponding exactly to the state $o, $det should be equal
to i/ro, and W reaches its maximum value 1. If now i; also corresponded to the pure
state $o ( p = Bo) we find also from Eqn (8.19) that W = (Bdet) = T r Bo2 = T r p0 = 1,
where we have used the condition (2.26) for a pure state.
We have here assumed a complete response of the detector for the pure state
i/rdet and we shall therefore have a zero response for the pure state with the wave
function which is orthogonal to $det. We do not have space to enter here into a
discussion of the case where the detector is not ideal, but responds less than completely to the most favourable polarization and still responds to the least favourable
(opposite) polarization. We refer to Fanos review (1957) for a discussion of this
point.
We want to express j;det in slightly different form. We mentioned earlier that
a given wave function such as (8.17) corresponds to a well-defined polarization.
345
Let Pdet be the corresponding polarization vector. By arguments similar to those

leading to Eqn (8.7) we can then write for i;det
:
pdet
= *[
F + (PdCt . S)] .
. . . . . .(8.21)
Using Eqns (8.21), (8.19) and (8.7) we get immediately an expression for the
response of the detector to a beam with a polarization P :
W = Tr
= T r $[?
+ (Pdet,S)] [I + (P .S)]
= g[l+(Pdet*P)],
. . . . . .(8.22)
where we have used Eqns (8.3). Eqn (8.22) leads to W = 1, if PdetllP (note
that we have P defined in such a way that P2< 1) and to W = 0, if Pdet/I - P. I n
general, we can write
. . . . . .(8.23)
* ( l + P n ) , P, = I P ( c o s 0 ,
w=
where 0 is the angle between P and Pdet.

T h e vector P has an immediate physical meaning in the case of electrons. I n
the case of photons, however, we must specify in more detail what is meant by P
and also what is meant by ' opposite ' polarization, a term used a moment ago.
So far we have only used the formal similarity between electron spin and photon
polarization. I n the case of photons P is a vector in a symbolical space (compare
the isotopic spin vector in nuclear physics) which is introduced in order that we
can continue to use the same formalism. T h e simplest way to introduce P for
photons is to consider the complex vector potential A exp [i(k. x)] (k is the wave
vector of the photon) as the wave function $. Instead of Eqn (8.16) we have now
Aexp[i(k.x)]
where
a,A,exp[i(k.x)]+a2A,exp[i(k.x)],
(A,. k) = (A2.k) = (A,.A2) = 0,
A12 = A,'
A2,
. . . . . .(8.24)
. . . . . .(8.25)
which means that we have chosen #, and #2 to describe two plane polarized waves
with mutually perpendicular planes of polarization. T h e coefficients a, and a2
completely describe the polarization of the photon. For this pure case, the
density matrix is constructed by analogy with expression (8.20),
(8.26)
and P is defined by analogy with Eqn (8.7) :
P2 = U, ~ 2 +
* U,* a2, P, = i ( a , ~ 2 *- U,* 4. . . . . . . (8.27)
P, = I U, l2 T o emphasize the fact that now P is defined in a ' polarization ' space rather than
in physical space we have denoted its 3 components by P,, P, and P,, rather than
by P,, P, and P, as was done in the case of electrons. For a discussion of the
relation between P and the so-called Stokes parameters (1852) or the coherence
matrix used by Wolf and Roman we refer to the papers quoted at the beginning of
this section.
We shall use the formalism developed here to discuss briefly the process where
a positron and an electron in a singlet state annihilate one another with the simultaneous emission of two photons (compare the discussion of this process by Pryce
and Ward (1947), Snyder, Pasternack and Hornbostel (1948) and Bleuler and
,I
D.ter Haar
346
ter Haar (1948)). One can show that the two photons (A and B) which are
emitted can be described by a joint density matrix PA, (compare the discussion at
the end of 3 2) which is given by the equation
PAB
4-'(
= *['Ai,
. . . . . .(8.28)
where the subscripts A and B refer to the two photons, and the vectors Q are
, and
are two by two matrices
symbolical vectors whose components G ~ G,
which have the same form as the Pauli matrices (8.2) and which therefore satisfy
equations similar to Eqns (8.3). We do not wish to give a proof of Eqn (8.28) but
accepting its validity we want to investigate its physical meaning. T o do this
we assume that there are two detectors described by their density matrices
and PBdetwhich can measure the polarization vectors PAdetand PBdet. This means
that (compare Eqn (8.21) and the definition of c3, and QB)
Pgdet = &[i,+(PAdet.Qa)],
pBdet
= 3[IB+(PBdet.QB)].
. . . . . .(8.29)
If we wish t o know the polarization of photon A we must evaluate the expression
W,
Tr,,
PABT,
pBdet =
. . . . . . (8.30)
which corresponds to zero polarization, This we could also have seen by using
Eqn (2.51) to obtain the density matrix pAk,
. , . . . .(8.31)
= hlA,
p-4 = TrB $AB
and #2 means that there is just as much polarization in one

which in terms of
direction as in the opposite direction. As all our expressions are symmetric in
A and B we find, of course, that the photon B is also unpolarized.
Let us now consider whether there is any correlation between the two polarizations. T o do this we evaluate
W,,
TrAB/;,,fjAdetp",det
$[l-(PAdet.
PBdet)],
. . . . . .(8.32)
which expresses the fact that the two photons have opposite polarization : W,, is
a maximum when P,det 11 - P B d e t and vanishes when PAdet
11 PBdet.
T o conclude this section we shall briefly discuss scattering and correlation
experiments in the density matrix frame-work. Let us first of all consider the case
where both the detector and the initial state of the system are pure states with wave
functions +deb and +(O). T h e scattering is described by the Schrodinger equation
itis;(t)
with the solution
. . . . . .(8.33)
= Zl#(t)
+(t)= [exp ( - iBt/h)l %(O).
. . . . . .(8.34)
T h e probability of detection is given by the equation

(8.35)
or, expressed in the coefficients ci(t) and
terms of some complete orthonormal set
W ( t )= [
xi
tidet of
tidet
ci*(t)
l2
an expansion of $(t) and
= Tr
pdetp(t),
+det
. . . . . .(8.36)
in
347
where we have used Eqn (2.6) for the definition of the elements of the density
matrices. I n a mixed state, Eqn (8.36) will still hold and i;(t) will depend on F(0)
through the solution of the equation of motion (2.37),
i;(t) =
[exp ( - iHt/ti)]$(o) exp (iHt/E).
. . . . . .(8.37)
Let us now consider how we can obtain transition probabilities. I n that case
we are interested in detectors which are sensitive to systems within a small energy
range of final states dE,, and for the transition probability per unit time P, which is
W(t)/t,we get from perturbation theory for the pure state case
(8.38)
where we have assumed that we can use first-order perturbation theory, where
H is the perturbation Hamiltonian, and where we have assumed that the detector
is not sensitive to unscattered particles.
If we are dealing with a mixed state we must generalize Eqn (8.38) and the
result is easily seen to be
= (2n-/ti) dE,Tr
[phdetHi;Ht],
. . . . . .(8.39)
where A?is the Hermitian conjugate of A.

Equation (8.39) is a general equation. T h e Hamiltonian A is determined by
the scattering process considered, i; by the initial conditions, and Pdet by the final
conditions. If, for instance, we are interested in the angular correlation between
a photon of a well-defined plane of polarization and an electron with a well-defined
spin direction which are produced in a Compton scattering process, Pdet will be
the direct product of a factor of the form (8.21) for the electron, a factor of the
form (8.29) for the photon, and a factor depending on the electron momentum
and the photon wave vector. T h e actual evaluation of expression (8.39) may still
be complicated (see, for instance, the papers by Lipps and Tolhoek (1954 a, b) on
Compton scattering) but a great deal has been gained in not having to consider all
degrees of freedom which are irrelevant to the final cross section in which we are
interested.
9 9. R E S O N A N CAEN D R E L A X A T I O PNH E N O M E N A
We saw in the preceding section that the density matrix is particularly suitable
for discussing systems with a small number of degrees of freedom. It has, indeed,
been very useful in discussions of various resonance phenomena involving a
limited number of energy levels ; recently, for instance, the theory of masers
and maser-like devices has been discussed in density matrix terms (see, for instance,
Anderson 1957, Clogston 1958, Suhl 1958). I n the present section we shall
restrict our discussion to the consideration of absorption of microwave radiation
in a gas following Karplus and Schwingers treatment (1948) and to a consideration
of relaxation effects in nuclear magnetic induction following the classic paper by
Wangsness and Bloch (1953). Among the steadily growing literature on the
application of density matrix techniques to resonance and relaxation phenomena
we may refer to papers by Karplus (1948), Bloch (1956, 1957), van Kranendonk
D.ter Haar
348
and Bloom (1956), Mori (1956), Seiden (1956 a, b), Wangsness (1956 a, b), Fano
(1957), Kubo (1957 b), Redfield (1957, 1959), Abragam and Proctor (1958),
Hubbard (1958), Kaplan (1958), Lamb and Sanders (1960), Lamb and Wilcox (1958),
Schumacher (1958), Solomon (1958), Tomita (1958 a, b), Di Giacomo (1959),
Kotera and Toda (1959), Skrotskii and Kokin (1959), Wannier (1959), Yatsiv
(1959), Halbach (1960), Nakano (1960 d), Sher and Primakoff (1960), Wilcox and
Lamb (1960), Opechowski (1953) and Yvon (1960 a).
T h e first problem we shall consider is the evaluation of the absorption coefficient
a: for radiation of angular frequency W.
This absorption coefficient is proportional
to the ratio of average dipole moment (6) to the field strength F ( t ) . T h e field
may be an electric or a magnetic field, and the total Hamiltonian of a particle in
the gas will be of the form
A(t)= Ao-(P.F(t))
= Ao+P C O S W ~ ,
. . . . . .(9.1)
where is the dipole moment operator, where gois the Hamiltonian of an isolated
particle if no field is present, and where we have used the fact that
F ( t ) = F COS w t .
. . . . .(9.2)
The state of the particles in the gas will be described by a density matrix i; which
satisfies the usual equation of motion (2.37) with the Hamiltonian given by Eqn (9.1).
In Eqn (9.1) interactions between particles are not taken into account and we must
therefore be careful to choose suitable boundary conditions for i;. We shall
assume that collisions take place randomly with a mean time of flight T , and that
they are so strong that immediately after the collision i; is given by the equilibrium
expression (see Eqn (4.2) ),
$ ( t o ) = i;o(to) =
exp [ - p ~ ( ~ o ~ l / ~ ~ { ~ ~ P *~* - . ~. a(9.3)

~ ( ~ o ~ l } ,
where to is the time when the last collision took place.

equation of motion for i; will be (2.37),
iz; = [A,PI-,
Between collisions the
. . . . . .(9.4)
with A given by Eqn (9.1). In view of this splitting up of the changes in i; into
a smooth part governed by Eqn (9.4) and a discontinuous part which is governed
by the boundary condition (9.3) it is useful to emphasize this fact and to write i; as
a function of two arguments i;(t,to), where the second argument indicates when
the last collision took place. T h e function $(t,to) thus satisfies Eqn (9.4) with
the boundary condition
. . . . . .(9.5)
i ; ( t o , t o ) = i;o(to).
T h e appearance of this second argument implies, however, that when we evaluate
the average dipole moment we must take not only the usual average using the
density matrix but also an average over all possible values of to. I t turns out to be
convenient to take this average first and t o introduce a density matrix
defined
bv the equation
FBv(t) = ~ o ~ i ; ( t , t - a ) e x p ( - B / r ) d i j ~ ,
. . . . . .(9.6)
349
where we have introduced the Poisson distribution exp( - 8/r) dO/r for the probawe get
bility that the last collision took place between t - 6 and t - 6 + do. Using
for the average dipole moment
(is> = TriiPa,.
. . . . . .(9.7)
T o evaluate (p) we need to know the equation of motion for Pav. As (B) vanishes
A
when there is no field, it is convenient to introduce the difference Ap between
PIY(t) and the density matrix Bo(t) for instantaneous thermal equilibrium. I n
A
terms of Ap, Eqn (9.7) becomes
(B>
T r is&,
AP
= ?h(t)-po(t).
. . . . . .(9.8)
T h e equation of motion for Ap can be found from Eqns (9.4) and (9.6),
A
ihAp = - iE3, + iEFAv
, A
- ih?, + [ H ,Apl- - A P / T ,
. . . . . .(9.9)
where we have evaluated the integral involving (aja8) by integrating by parts, used
Eqn (9.6), and the fact that [A,fiO]- = 0. We must emphasize that this does not
imply that bo = 0, as A is an explicit function of the time.
We shall now assume that the field is weak so that we may neglect second-order
A
A
terms in P, Ap, or their product. This means that we can replace [ A , A p ] - in
A
Apl-. Using a representation in which gois diagonal, denoting

Eqn (9.9) by [Ao,
the eigenvalues of goby E, and introducing the notation
. . . . ..(9.10)
Emmn = Em - En,
we get from Eqn (9.9)
[3+
iwmn +
7 APmn = - 3 (Po)mn*
..... .(9.11)
If the field is weak, one can evaluate (po)mn using the Bloch equation (4.6)
~$(P>/aP= ( g o +Q)$(P),
where $(P) = exp ( -PI?) and A eP cos w t . If we put
P(P) = $o(P) J(PL $o(P> = =P [ -PQol,
we get for B(P) the equation
aSjap = $o-lI?$ 0 B,
. . . . . . (9.12)
. . . . . .(9.13)
. . . . . .(9.14)
which is equivalent to the integral equation

(9.15)
where we have used the fact that $(O) = j O ( O ) = 1, or, J(0) = 1. As I? is assumed
to be small we can approximate Eqn (9.15) by
J(P)N 1+ /p$o(P)-lA$o(P)
dP.
0
. . . . . . (9.16)
D.ter Haar
350
Using Eqns (9.13) and the representation in which gois diagonal we get for (po),nrL
(PO),,
where
pm(0)
= pm(0)
a,, + (p,
- p,)
(Vmn/kwmm)
COS ut,
...... (9.17)
is given by the equation

Pm(O) = ~ X (-PEm)/[Xnex~
P
(-PEn)J,
. . . . . .(9.18)
and the V,, are the matrix elements of the operator P in the same representation.
T h e pm@) are the diagonal elements of the equilibrium density matrix. In Eqn
(9.17) Tve have used the fact that up to terms linear in the field
T r exp ( - pa) = T r exp ( - Pg0).
From Eqns (9.8), (9.11), (9.17) and (9.18) we can find (p> and thus the absorption
coefficient. If w lies near a particular resonance frequency, wo, say (kwo = E, - E,),
and if the other w, are such that w,,-wo
is always much greater than l / ~we
,
find for the absorption coefficient per particle
which shows that l / will

~ be the resonance width (we have assumed E, and E, to
be degenerate).
T h e derivation of Eqn (9.19) is straightforward, but tedious.
T h e other topic we want to discuss here is the derivation of the Bloch nuclear
induction equation (Bloch 1946),
dP = y [ p A x ] - - - if!,
-
jPy k(e-4)
--
dt
7 2
72
. . . . . .(9.20)
71
where P ( t ) is the nuclear polarization due to an external magnetic field X ( t ) ,

which consists of a strong, constant field *WOin the x-direction and a relatively
weak field XI,
& = k X o+ X1(
t ).
. . . . . .(9.21)
T h e vectors i, j, k are unit vectors in the x-, y - , and x-directions, y is again the
magnetogyric ratio, and Po is the equilibrium polarization in the field Xo. T h e
quantities T~ and T~ are the longitudinal and transverse relaxation times. If there
P, and Py would tend to zero with a time constant T ~ and
,
P,
were no field Xl,
would tend to Po with a characteristic time T ~ .
Actually, we shall use a simpler model than the one discussed by Wangsness
and Bloch, and as a result we shall find T~ = T* ( = T ) so that we do not need to
specify the direction of the constant field, and instead of Eqn (9.20) we shall find
dP
P- Po
PA\]--,
. . . . . .(9.22)
dt
T
where Po is the equilibrium polarization in the field X,,,while we write for 2
instead of Eqn (9.21) simply
X = XO+ZI(;(t).
. . . . . .(9.23)
Our first simplification consists in assuming that we are dealing with particles of
spin 3 ; this entails that a great many troublesome terms which occur in the general
case vanish, and that P is given by Eqn (S.l),
P = (6).
. . . . . .(9.24)
351
T h e system we are considering consists of the spins which are embedded in a

lattice-the surroundings or bath. We shall assume the spins to be non-interacting-except through the bath of which they are part. I n that case we can just
consider a system consisting of a single spin in interaction with a bath, and obtain
the final results for our macroscopic system by taking suitable statistical averages.
T h e Hamiltonian of this system is of the following form
......(9.25)
A = HIy1-k-??B + g j n t ,
is the magnetic energy of the spins in the magnetic field,
where
-??>I
= AM,+Hll,, HhI0= +yZ(e.y%),
$yB(G.Yi",),
..... .(9.26)
is the Hamiltonian of the bath, and

the interaction Hamiltonian describing
the interaction between the spin and the bath. We shall assume that this interaction takes place through random magnetic fields Hrond(t)
produced by the motion
of charges in the bath (see Redfield 1957) so that
= - hti(* Yiorrand),
. . . . . .(9.27)
where the fact that Xrrand

is random in both direction and magnitude is expressed
by the relations
..... .(9.28)
..... .(9.29)
Eqn (9.29) defines the time-correlation of the random field. We shall later on
make the usual assumption that Eqns (9.28) and (9.29) still hold if we replace the
upper limit of the integral by a ' sufficiently large ' time. Assumption (9.27) is
responsible for the fact that T~ = T~ so that Eqn (9.20) reduces to Eqn (9.22).
It is convenient to use a representation in which HIyI,and -??, are diagonal.
The density matrix will then have matrix elements (mb I PI m'b') where m ( = 2 3)
is the magnetic quantum number, and b a quantum number describing the state
of the bath. We shall assume that the bath is at all times in thermal equilibrium.
If that is the case we can write
(mb
where
I P^ I " b ' )
'
P ( b )<mI Psp I m ' )
Pap is a reduced spin density matrix,

Psp = Tr,P,
. . . . . .(9.30)
..... .(9.31)
Tr, indicating the trace over the bath degrees of freedom, and P(b) are the
normalized Boltzmann factors [E, P(b) = 11.
We shall now consider gnIo
and H B to be large compared with gbIl
and Hint,and
treat the latter as perturbations. It is further convenient to work in the interaction
representation. Operators in the interaction representation will be indicated by
an asterisk,
fi* = 8-1J-1 lfjJB,
(9.32)
......
where
A = exp ( - ifTllotiti),
. . . . . .(9.33)
B = exp ( - iZTB t / ~ ) .
. . . . ..(9.34)
352
D.ter Haar
We note for future reference that A^ and commute, and that if fi depends on B
only, and not on the bath degrees of freedom, Eqn (9.32) reduces to
. . . . . .(9.35)
fi* = A-lfiA.
From the equation of motion for p
itiappt
we get for
[ABIo
+ ElllI+A, +ETint,p1-
. . . . . .(9.36)
[A,,* +Aint*,p^*]-.
. . . . . .(9.37)
e* the equation of motion

itiap*/at
From Eqn (9.37) we can find ai;,,*/at and then ai;,,/at, using Eqn (9.35), and so
finally %'/at. We must draw attention to the fact that as pSpis independent of the
bath degrees of freedom, relation (9.31) also holds between $* and psp*.
As the eigenvalues of
A^
are ecimwt,where
. . . . . .(9.38)
= rlYi"01,
we get from Eqn (9.35) with
fi~p,,
. . . . . .(9.39)
T o find a$,,*jat we evaluate the change in pSp* during a ' small ' time interval At.
A
A
This increment ApspS is obtained from Eqn (9.37) by first evaluating Ap* and then
taking the trace over the bath degrees of freedom. From Eqn (9.37) we find
A
Ap" = A, p"
where
A
iRA, p*
= /oAt[f?lI,*(t),
i;*(O)]- dt
+ A2 p",
. . . . ..(9.40)
[Qint*(t), p^*(O)]- dt
. . . . . .(9.41)
At
0
is the first-order change and where e p * is the second-order change in p^*. T o

simplify Eqn (9.41) we assume that At though ' small ' is ' sufficiently large ', so
that we can use Eqn (9.28). T h e second integral then vanishes and as
is
independent of the bath degree of freedom we get from Eqn (9.41)
;RAl
pSp*N [I?3Il*,
$sp*]l-
. . . . . .(9.42)
At.
There is as yet no term depending on the bath, and we must go to the second
A
order. In Aep* we can, however, neglect all terms involving glr,so that we are
left with
A
A ZP* = -R-2/oA'dtr/~[f&,,*(t.),[I?int*(t"), i;*(O)]-]-
dt".
. . . . . .(9.43)
From this equation and Eqn (9.29) we get, provided At satisfies again certain
conditions as to smallness and largeness,
iRA, pSp*N
X,,'[Gi",
[ej*,p^,,*]-]-
At,
. . . . . .(9.44)
'
353
where the hij are determined by the hij of Eqn (9.29). Substituting Eqns (9.42)
A
and (9.44) into Eqn (9.39) and putting a$,,*/at = Apsp*/At, expression (9.42)
combines with the first term on the right-hand side of Eqn (9.39) to give
(ml [l?31,ijsp]-~m)
and this term leads, as we saw in the preceding section, to
the first term on the right-hand side of Eqn (9.22). T h e second term is more
complicated. T h e general procedure is as follows : one uses Eqn (9.35) to go
back from the starred to the unstarred operators and expands the double
commutator in Eqn (9.44). After that one multiplies the result by a and evaluates
the trace over the spin variables using the relations (8.3) between the components
of a. T h e result is equal to a constant times [(a) - a ] , where the constant can be
identified with - T - ~ and is determined by the X i j , showing once again the general
importance of correlation functions in statistical physics, and where a can be shown
to be equal to Po. We refer to Wangsness and Blochs paper for details of this
evaluation.
$ 1 0 . T H ET H E O ROY
F MEASUREMENT
We shall finish this article with a brief discussion of the theory of measurement.
This is of less practical importance than some of the applications with which we
have been concerned in the preceding sections, I t is, however, of theoretical
importance and the present author felt that a survey article such as the present v7ould
be incomplete without this concluding section. A related topic is quantummechanical ergodic theory, and we shall have occasion briefly to mention that topic
as well. T h e classic papers on the theory of measurement in quantum mechanics
are those by von Neumann, summarized in his monograph (1932) and the beautiful
expos6 by London and Bauer (1939). Among recent contributions to the subject
we may mention papers by Ludwig (1953, 1954), Baker (1958), Band (1958),
Green (1958), Daneri and Loinger (unpublished) and Wakita (1960), while
we refer to papers by Sakai (1937), Tolman (1938), ter Haar (1954, 1955),
van Kampen (1954, 1956), Van Hove (1957, 1959), Bocchieri and Loinger (1959)
and Caldirola (1959) for a discussion of quantum-mechanical ergodic theory.
A measuring process consists essentially of two separate parts-a fact which was
pointed out by Margenau (1950) and which is not always sufficiently stressed.
T h e first part is the preparation for the measurement which consists in the ideal
case in preparing a pure case. This can be done-as we have discussed in earlier
sections-by letting the system pass through a suitable filter. After that the system
must still be observed to be in the particular pure state which has been prepared.
T o fix the ideas let us consider a beam of spin 1 particles, a case discussed by London
and Bauer which we shall use to illustrate the ideas of the present section. If we
wish to find out the polarization of such a beam, we can perform a Stern-Gerlach
experiment and let the beam pass through a region in space in which there is a
magnetic field, This field will split the original beam into three beams, each of
which corresponds to a well-defined value of the spin component in the direction
of the magnetic field. This splitting into three beams itself is not a measurement,
but is the preparation for the measurement which will consist in having three
counters, say, at well-defined positions in space, such that the clicking of one of
them corresponds t o one of the three possibilities for the spin component. Once
23
3 54
D.ter Haar
the actual measurement, or observation, has taken place the pure case has been
destroyed through the interaction between detector (counter) and system.
Let us consider the process of the preparation for the measurement in somewhat
more detail. T h e actual measurement we shall assume to be done by a counter.
There are a number of problems connected with the actual action of detectors (see,
for instance, Green 1958) but we shall assume that the actual detection is straightforward. As far as the preparation process is concerned, we shall assume that the
system is in a pure state before this process begins. This restriction is not necessary
and can be removed but it enables us to discuss more clearly various consequences
of the measuring process. If the system is in a pure state, it can be described by a
wave function t,!Jo(x)where x stands collectively for all coordinates describing the
system. Let fl be the physical quantity the value of which we wish to measure,
and let uk andf, be, respectively, its eigenfunctions and eigenvalues. It is convenient to expand t,!Jo in terms of the uk,
Let P be the operator corresponding to the apparatus preparing the system for a
measurement, and let v,(y) and p , be, respectively, its eigenfunctions and eigenvalues, where y stands collectively for all coordinates describing the apparatus.
I n order that our apparatus should be suitable for our purpose we must assume that
the eigenvalues p , can be assigned to different channels such that each channel
corresponds uniquely to a well-defined eigenvalue f,. Before the measuring
process takes place, the apparatus should be in a well-defined state, c0 say,
corresponding to a ' neutral ' situation. T h e combined wave function of system +
apparatus will thus before the measuring process be of the form
Once the measuring process starts, the apparatus and the system are coupled and
the wave function will be of the general form
.(10.3)
I n view of the fact that our apparatus was chosen such that each p,, can be assigned
to a definitef, we can simplify expression (10.3) and write
YcoupIed
C, b k U,(.)
vd~),
. . . . . .(10.4)
indicating that each zi, will uniquely lead to a well-defined state vls, or, that if the
system were in the state U, it would definitely lead to the state zk of the apparatus.
As, moreover, in the case of an ideal measuring set-up-such as, for instance, the
Stern-Gerlach experiment described earlier-the apparatus should not change the
probability amplitudes, we must require that
b,
ak,
. . . . . .(10.5)
and the wave function after the preparation of the measurement will be of the form
Theoyy and Applications of the Density Matvix
355
We are once again in a familiar situation. We are interested only in part of the
degrees of freedom of the combined system, and each part is no longer described
by a wave function but by a density matrix corresponding to a mixture. T h e
quantity I a, 12 is now both the probability that the system was originally (and still
is) in the state uk (see Eqns (10.2) and (10.6)) and also the probability that the
apparatus is in the state zk leading to an observation from which we conclude to the
eigenvalue p , and thus to the eigenvalue flc.
We note that the preparation for the measurement, that is, the coupling of the
system to a filter, changes the situation from one corresponding to a pure state to
one corresponding t o a mixture (which, by the way, increases the entropy as it
increases the lack of detailed knowledge). After the uncoupling the detectors
which are uniquely assigned to the different channels will determine the probabilities
for the different eigenstates of the apparatus to be realized and thus will determine
the probability distribution corresponding to the physical quantity p . We have
no space here to go into related problems such as the fact that the experiment must
be repeated in order that we can find the probability distribution.
Let us consider again the Stern-Gerlach experiment on a beam of spin 1
particles. If there were no magnetic field, the beam would pass on undeflected.
T h e same would be true, if the spin of the particles were in the direction of the
magnetic field, while spin components + R or - E along the direction of the magnetic
field would produce deflections of the beam in opposite directions. Before the
coupling we have (compare Eqns (10.1) and (10.2) )
$0 =
\k'before =
a-, U-,
+ a, U0 + a, U,,
%[a-, U-, + a0 U 0 + a,
%I,
. . . . . .(10.7)
. . . . . . (10.8)
where U-,, u0 and U, are the eigenfunctions of the spin component in the direction
of the magnetic field with eigenvalues - R , 0 and 7i respectively, while uo indicates
that the beam would pass undeflected through if there were no spin-field interaction.
Eqn (10.8) corresponds to Eqn (10.3) with
. ..
b/im = a k am07
* *
.(10.9)
and to the pure state density matrices
.....
(10.10)
After the coupling and uncoupling we find (see Eqn (10.6))

corresponding to
Tuncoupled
= a-, U-1 U-,+
a0 U 0 U0
+ a, U1 V l
bkm = a, Bkm,
and
(~suat)kl=
I a/i l2 Ski,
. . . . . .(10.11)
. . . . *(10.12)
*
(Papp)mn = am I'amn,
. . . . . .(10.13)
as should be the case.

It is of interest to verify that the transition from (10.8) to (10.11) occurs through
a unitary transformation, as should be the case, since it should be the evolution of
the combined wave function under the influence of a Hamiltonian I? given by the
expression
. . . . . .(10.14)
I? = Qsyst + Q a p p + Qinint,
D.ter Haar
3 56
c,pp
where Hsvstand
are the unperturbed Hamiltonians of the system and of the
apparatus, while Hintis an interaction term describing the coupling between system
and apparatus. One can easily verify that
. . . . . .(10.15)
= Z , n Skm,ln ( h n h e f o r e ,
(b/A"onpled
where Skm.lnis the matrix

n
\ k, m
-1,-1
-1, 0
- 1, 1
0,-1
0, 0
0, 1
1,-1
1, 0
1, 1
- 1,-1
-l,o
-1,l
1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0,-1
O,o
0,l
1,-1
l,o
1,l
0
0
0
1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
1
0
0
Each of the three beams produced by the magnetic field corresponds to a welldefined value of the spin component, and thus to a well-defined pure case, but the
total of the three beams corresponds to a mixture, produced by the magnetic field
from the original pure case.
C O N C L U D I NRGE M A R K AS N D A C K K O W L E D G M E N T S
I n this paper I have given a bird's eye view of a few of the many aspects of
density matrix theory. Lack of space has made it necessary to give only a sketch
of many of the topics which were discussed. I hope to discuss this subject in
more detail elsewhere.
I should like to express my gratitude to the great number of people who have
kindly sent me preprints of relevant papers by themselves and their collaborators,
and especially to Drs. Argyres, Brush, Golden, Kanazawe, Lowdin, McWeeny,
Rosenfeld, Siegert, Thouless, Wolf and Yvon.
I should also like to express my thanks to Dr. W. E. Parry for helpful comments.
REFERENCES
T h e number in brackets refer to those sections in this paper where the paper in question is
referred to.
ABRAGAM,
A., and PROCTOR,
W. G., 1958, Phys. Rev., 109, 1441 (9).
ABRIKOSOV,
A. A., GOR'KOV,
L. P., and DZYALOSHIKSKI~,
I. E., 1959, J . Exp. Theor. Phys.,
U.S.S.R., 36, 900 (Souiet Phys.,JETP, 9, 636) (I, 3).
ADAMS,
E. N., and HOLSTEIN,
T. D., 1959,J. Phys. Chem. Solids, 10, 254 (7).
ALFRED,
L. C. R., 1960, Techn. Rep., No. 7 , Project NR-051-362, Brandeis University, U.S.A.
(5).
P. W., 19j7,J. Appl. Phys., 28, 1049 (9).
AKDERSON,
ARGYRES,
P. N., 1958, Phys. Rev., 109, 1115 (7) ; 1960 a, Phys. Rev., 117, 31 j (7) ; 1960 b,
Westinghouse Sci. Paper 6-94760-2-P16 (7).
357
ARGYRES,
P. N., and ROTH,L. M., 1959,J. Phys. Chem. Solids, 12, 89 (7).
AYRES,R. U., 1958, Phys. Rev., 111, 1453 (5).
BAKER,G. A., 1958, Phys. Rev., 109, 2198 (IO).
BALESCU,
R., 1959, Physica, 25, 324 (7).
BALIAN,R., BLOCH,C., and DE DOMINICIS,
C., 1960, C.R. Acad. Sci., Paris, 250, 2850 (4).
BALIAN,R., and DE DOMINICIS,
C., 1960 a, Nucl. Phys., 16,502 (3,4) ; 1960 b, C.R. Acad. Sci.,
Paris, 250, 3285 (4) ; 1960 c, C.R. Acad. Sci., Paris, 250, 4111 (4).
BAND,W., 1958, Amer. J . Phys., 26, 440 (IO).
BELYAEV,
S. T., 1958 a, b,J. Exp. Theor. Phys., U.S.S.R., 34, 417, 433 (Soviet Phys.,JETP,
7, 289, 299) (3).
BETHE,H. A., 1956, Phys. Rev., 103, 1353 (4).
BLATT,J. M., 1956, Nuovo Cim., 4, 430 (5).
BLATT,J. M., and MATSCBARA,
T., 1958, Progr. Theor. Phys.Japan, 20, 553 (4, 5).
BLEULER,
E., and TER HAAR,
D., 1948, Science, 108, I O ( 8 ) .
BLOCH,C., and DE DOMINICIS,
C., 1958, Nucl. Phys., 7, 459 (4) ; 1959 a, b, Nucl. Phys., 10,
181, 509 (4).
BLOCH,F., 1932,Z. Phys., 74, 295 (4) ; 1946, Phys. Rev., 70, 460 (9) ; 1956, Phys. Rev., 102,
104 (9) ; 1957, Phys. Rev., 105, 1206 (9).
BOCCHIERI,
P., and LOINGER,
A., 1959, Phys. Rev., 114, 948 (IO).
BOGOLYUBOV,
N. N., and SHIRKOV,
D. V., 1959, Introduction to the Theory of Quantized Fields
(New York : Interscience) (3, 4).
BOGOLYUBOV,
N. N., and TYABLIKOV,
S . V., 1959, Dokl. Akad. Nauk, S.S.S.R., 126, 53
(Soviet Phys., Doklady, 4, 604) (3, 6).
BONCH-BRUEVICH,
V. L., 195j , J . Exp. Theor. Phys., U.S.S.R., 28, 1 2 1 (Societ Phys., JETP,
1, 169) (3) ; 1956 a,J. Exp. Theor. Phys., U.S.S.R., 31, 5 2 2 (Soviet Phys.,JETP, 4,457)
( I , 3) ; 1956 b,J. Exp. Theor. Phys., U.S.S.R., 30, 342 (Soviet Phys.,JETP, 3, 278) ( 3 ) ;
1956 c, J . Exp. Theor. Phys., U.S.S.R., 31, 254 (Soviet Phys., JETP, 4, 196) (3, 6) ;
1957, Fiz. Met. Metallov., 4, 546 (Phys. Met. Metallogr., 4, No. 3 , 133) (3) ; 1958 a, b,
Fiz. Met. Metallov., 6, 590, 769 (Phys. Met. Metallogr., 6, WO.4, 13 and No. j , I) (3) ;
1959 a, Dokl. Akad. Nauk S.S.S.R., 124, 1233 (Soviet Phys., Doklady, 4, 172) ( 3 ) ;
1959 b, Dokl. Akad. Nauk S.S.S.R., 126, 539 (Soviet Phys., Doklady, 4, 596) (3) ;
1959 c, Dokl. Akad. Nauk S.S.S.R., 129, 529 (Soviet Phys., Doklady, 4, 1275) (3) ;
1959 d, Fiz. Met. Metallov., 7, 174 (Phys. Met. Metallogr., 7, No. 2, 13) (3) ; 1959 e,
J . Exp. Theor. Phys., U.S.S.R., 36, 924 (Soviet Phys.,JETP, 9, 653) (7).
BONCH-BRUEVICH,
V. L., and GLASKO,
V. B., 1959, Dokl. Akad. Nauk S.S.S.R., 124, 1015
(Soviet Phys., Doklady, 4, 147) (3).
BONCH-BRUEVICH,
V. L., and KOGAN,SH. M., 1960, Ann. Phys., New York, 9, 125 (3).
BONCH-BRUEVICH,
V. L., and MIRONOV,
A. G., 1960, Fiz. Tverd. Tel., 2, 489 (Soviet Phys.,
Solid St., 2, 454) (3).
BOPP, F., 1959, 2. Phys., 156, 348 ( I , 5).
BORN,M., and GREEN,H. S., 1946, Proc. Roy. Soc. A, 188, I O ( 7 ) ; 1947 a, Proc. Roy. Soc. A,
190, 455 (7) ; 1947 b, Proc. Roy. Soc. A, 191, 168 (7) ; 1948, Proc. Roy. Soc. A, 192,
166 (7).
BROUT,R., 1959, Phys. Rea., 115, 824 (4, 6).
BRUSH,S. G., 1961, Rev. Mod. Phys., 33, 79 (4).
BUTLER,S. T., and FRIEDMAN,
M. H., 1955, Phys. Rev., 98, 287 (4).
CALDIROLA, P., 1959, lvuovo Cim., 14, 260 (IO).
CALLEN,
H. B., and WELTON,T. A., 1951, Phys. Rev., 83, 34 (7).
CARSLAW,
H. S., and JAEGER,
J. C., 1947, Conduction of Heat in Solids (Oxford : University
Press), p. 217 (6).
CHENCHUN-SYAX,
1959, Dokl. A k a d . Nauk S.S.S.R., 125, 1238 (Soviet Phys., Doklady,
4, 413) (3).
CHESTER,
G. V., 1954, Phys. Rev., 93, 606 (4).
CHESTER,
G. V., and THELLUNG,
A., 1959, Proc. Phys. Soc., 73, 745 ( I , 7).
CHIRGWIN,
B. H., 1957, Phys. Rev., 107, 1013 ( 5 ) .
CLOGSTOP.;,
A. M., 1958,J. Phys. Chem. Solids, 4, 271 (9).
D.ter Haar
3 58
CORSON,
E. M., 1951, Perturbation Methods in the Quantum ililechanics of n-electron Systems
(London : Blackie) (5).
Cox, J. A. l L , and TOLHOEK,
H. A., 1953, Physica, 19, 673 ( 8 ) .
DI GIACOMO,
A., 1959, Nzioao Cim., 14, 1082 (9).
DIRAC,
P. A. M., 1929, Proc. Camb. Phil. Soc., 25, 62 (I, j ) ; 1930 a, b, Proc. Camb. Phil. Soc.,
26, 361, 376 ( I , j ) ; 1931, Proc. Camb. Phil. Soc., 27, 240 ( I , j ) ; 1935, Principles of
Quantum Mechanics (Oxford : University Press), 2nd edition, 0 37 (I).
DZYUB,
1. P., 1960, Dokl. Akad. Nauk S.S.S.R., 130, 1241 (Soviet Phys., Doklady, 5, 125) (3).
EDWARDS,
S.F., 1958 a, Phil. Mag., 3, 119 (6) ; 1958 b, Phil. Mag., 3, 1020( I , 7).
EHRENFEST,
P., 1927, 2. Phys., 45, 45 j (8).
EHRENREICH,
H., and COHEN,M. H., 1959, Phys. Rev., 115, 786 (5, 6).
ENZ,C. P., 1960 a, b, Helv. Phys. Acta, 33, 89, 11j (6).
EZAWA,
H., TOMOZAWA,
J., and UMEZAWA,
H., 1957, iliuovo Cim., 5, 810 (3).
FALK,
D. S., 1 9 j 9 a , b, Phys. Rev., 115, 1069, I074 (3).
FASO,U,, 1949, J . Opt. Soc. Amer., 39, 859 ( 8 ) ; 1957, Rev. Mod. Phys., 29, 74 ( I , 2 , 8, 9).
FEYNMAN,
R. P., 1948, Rev. Mod. Phys., 20, 367 (4) ; 1953, Phys. Rev., 91, 1291 (4).
FRADKIN,
E. S., r g j g a , b, J . Exp. Theor. Phys., U.S.S.R., 36, 951, 1286 (Soviet Phys.,
JETP, 9, 672, 912) ( I , 3 ) ; 1 9 j 9 c , Nucl. Phys., 12, 465 ( I , 3, 5 ) ; 1959d, e, Dokl.
Akad. Nauk S.S.S.R., 125, 66, 311 (Soviet Phys., Doklady, 4, 327, 347) (3).
FRIEDMAN,
H. M., and BUTLER,
S.T., 1955, Phys. Rev., 98, 294 (4).
FUJITA,
S., 1959, Phys. Rev., 115, 1335 (3, 4).
FUJITA,S., and HIROTA,
R., 1960, Phys. Rev., 118, 6 (3, 4).
GALITSKI:,V. M., 1958,J. Exp. Theor. Phys., U.S.S.R., 34, 1 5 1 (Soviet Phys.,JETP, 7, 104)
(3).
GALITSKI?,
V. M., and MIGDAL,
A. B., 1958, J . Exp. Theor. Phys., U.S.S.R., 34, 139 (Soviet
Phys., JETP, 7, 96) (3).
GAUDIN,
M., 1960 a, Nucl. Phys., 15, 89 (4) ; 1960 b, Nucl. Phys., 20, 513 (4).
GLASSGOLD,
A. E., HECKROTTE,
W., and WATSON,
K. M., 1959, Phys. Rev., 115, 1374 (4).
GOLDBERGER,
M. L., and ADAMS,E. N., 1952,J. Chem. Phys., 20, 240 (4).
GOLDEN,
S., 1957 a, Phys. Rev., 105, 604 ( 5 ) ; 1957 b, Phys. Rev., 107, 1283 ( 5 ) ; 1960 a,
Rev. Mod. Phys., 32, 322 ( I , j ) ; 1960 b, Techn. Rep. No. 6, Project NR-051-362,
Brandeis University, U.S.A. (5).
GOLDSTONE,
J., 1957, Proc. Roy. Soc. A, 239, 367 (4).
GOLDSTONE,
J., and GOTTFRIED,
K., 1959, Nuovo Cim., 13, 849 (5, 6).
GOMBAS,
P., 1956, Handbuch der Physik, 36, 109 (Berlin : Springer) ( 5 ) .
GORKOV,
L. P., 1958,J. Exp. Theor. Phys., U.S.S.R., 34, 735 (Soviet Phys.,JETP, 7, 5 0 5 ) (3).
GREEN,H. S., 1947, Proc. Roy. Soc. A, 189, 103 ( 7 ) ; 1958, Nuoco Cim., 9, 880 (I, IO).
GREEN,M. S., 1960, Phys. Rev., 119, 829 (7).
GREENWOOD,
D. A., 1958, Proc. Phys. Soc., 71, 585 (I, 7).
DE GROOT,
S.R., and TOLHOEK,
H. A., 195j, Beta- and Gamma-Ray Spectroscopy (Amsterdam :
North-Holland), p. 613 (8).
GURZHI,
R. N., 1957,J. Ex@.Theor. Phys., U.S.S.R., 33, 451 (Soviet Phys.,JETP, 6, 352) (7).
TER HAAR,
D., 1954, Elements of Statistical Mechanics (New York : Rinehart) ( I , 2, 4, 6, IO) ;
1955, Rev. Mod. Phys., 27, 289 ( I , 4, I O ) ; 1958, Introduction to the Physics of ManyBody Systems (New York : Interscience) ( 5 ) ; 1960, Physica, 26, 1041 ( I , 5 ) .
HAGEDORN,
R., 1958, CERN Report No. 58-7 (2, 8).
HALBACH,
K., 1960, Phys. Rev., 119, 1230 (9).
HALL,G. G., 1960, Proc. Phys. Soc., 75, 575 (5).
HAMILTON,
J., 1959, The Theory of Elementary Particles (Oxford : University Press) (3, 4, 5).
HARTOGH,
C. D., TOLHOEK,
H. A., and DE GROOT,S.R., 1954, Physica, 20, 1310 ( 8 ) .
HEBBORN,
J. E., and SONDHEIMER,
E. H., 1960,J. Phys. Chem. Solids, 13, 105 (6).
HELFAND,
E., 1960, Phys. Rev., 119, I (7).
HELFAND,
E., and RICE,S.A., 1960,J. Chem. Phys., 32, 1642 (7).
HUBBARD,
J., 1959, Phys. Rev. Letters, 3, 77 (4, 5 ) .
HUBBARD,
P. S., 1958, Phys. Rev., 109, 1153 (9).
HUBY,R., 1958, Proc. Phys. Soc., 72, 97 ( 8 ) .
359
HUGENHOLTZ,
N. M., 1957, Physica, 23, 481 (4).
HUGENHOLTZ,
N. NI., and PINES,D., 1959, Phys. Rev., 116, 489 (3).
HUND,F., 1956, Handbuch der Physik, 36, I (Berlin : Springer) (5).
HUSIMI,
I<., 1940, PYOC.
Phys. Math. Soc., Japan, 22, 264 ( I , 2 , 4, 5).
HUSIMI,K., KITAKO,
Y., and NISHIYAMA,
T., 1958, Fortschr. Phys., 6, I (4, j, 7).
ICHIMURA,
H., 1956, Progr. Theor. Phys.,Japan, 15, I j I (6).
VAK KAMPEN,
N. G., 1954, Physica, 20, 603 ( I O ) ; 1956, Fortschr. Phys., 4, 405 (7, IO).
KANAZAWA,
H., and ~TATSUDAIRA,
N., 1960, Prog. Theor. Phys.,Japan, 23, 433 (6).
S., and FUJIT.4, E., 1960, Prog. Theor. Phys.,Japan, 23, 426 (6).
KAKAZAWA,
H., MISAWA,
KAKAZAWA,
H., and WATABE,
M., 1959, Progr. Theor. Phys.,Japan, 22, 466 (3) ; 1960, Progr.
Theor. Phys.,Japan, 23, 408 (6, 7).
KAPLAK,
J. I., 1958,J. Chem. Phys., 29, 462 (9).
KARPLUS,
R., 1948, Phys. Rev., 73, 1027 (9).
KARPLUS,
R., and SCHWINGER,
J., 1948, Phys. Rea., 73, 1020 (9).
KASCHLUHN,
F., 195j a, Ann. Phys., Lpz., 16, 257 (4) ; 195j b, Ann. Phys., Lpz., 16, 304 ( 5 ) .
KASUYA,
T., 1959,J. Phys. Soc.,Japan, 14, 410 (7).
KEMBLE,
E. C., 1937, Fundamental Principles of Quantum Mechanics (New York : NIcGrawHill) (2).
KINOSHITA,
T., and NAMBU,
Y., 1954, Phys. Rev., 94, 598 (3, 5).
KITTEL,C., 1958, Elementary Statistical Physics (New York : Wiley) (I).
KLEIN,A., and PRANGE,
R., 1958, Phys. Rev., 112, 994 (3, 4).
KLEIN,A., and ZEMACH,
C., 1957, Phys. Rev., 108, 126 (3).
KLIMONTOVICH,
Yu. L., and SILIK,
V. P., 1960, Usp. Fiz. Nauk, 70, 247 (Soviet Phys., Uspekhi,
3, 84) (7).
KLIhIONTOVICH, Yu. L., and TEMKO,
S.v., 1958, J . Exp. Theor. Phys., U.S.S.R., 35, 1141
(Soviet Phys., J E T P , 8, 799) (7).
KLINGER,31. I., 1959 a, Fiz. Toerd. Tel., 1, 674 (Soviet Phys., Solid St., 1, 613) (3, 7) ;
1959 b, c, d, Fiz. Tverd. Tel., 1, 861, 1226, 1385 (Soviet Phys., Solid St., 1, 782, 1122,
1269) (7) ; 1959 e , Fiz. Tverd. Tel., 2nd collection of papers, p. 136 (7).
KOBELEV,
L. YA., 1958 a, Izv. Vyssh. Uch. Zaved., No. I , 68 (3, j) ; 1958 b, Izv. Vyssh. Uch.
Zaved., No. 2 , 103 (3) ; 1958 c, d, Fix. Met. Metallov., 6, 3 5 4 , 7 5 0 (Phys.Met. Metallogr.,
6, No. 2, 156 and KO.4, 165) (3) ; 1958 e, Fiz. Met. Metallov., 6, 943 (Phys. i2let.
Metallogr., 6, No. 5, 172) (3, 5 ) ; 1959, Izv. Vyssh. Uch. Zaved., NO. 2, 60 (3).
KOGAK,Sw. M., 1959, Dokl. Akad. Nauk, S.S.S.R., 126, 546 (Soviet Phys., Doklady, 4, 604)
(3).
KOHK,W., and LUTTINGER,
J. M., 1957, Phys. Rev., 108, 590 ( I , 7).
KONSTANTINOV,
0. V., and PEREL',V. I., 19j9,J. Exp. Theor. Phys., U.S.S.R., 37, 786 (Soviet
Phys., J E T P , 10, 560) (7) ; 1960, J . Exp. Theor. Phys., U.S.S.R., 39, 197 (Soviet
Phys.,JETP, 12, 142) (7).
KOPPE,H., 1951, Ann. Phys., Lpz., 9, 423 (3, 4 ) ; 1957, 2. Phys., 148, 135 (5, 6).
B., 1958,Z. Phys., 151, 613 (6).
KOPPE,H., and M~HLSCHLEGEL,
KOTANI,T., 195j, Prog. Theor. Phys., Japan, 14, 379 (4, 8).
KOTERA,T., and KODA,M., 1959, J . Phys. Soc., Japan, 14, 1475 (9).
KRAICHNAN,
R. H., 1958 a, b, Phys. Rev., 112, 1054, 1056 ( 3 ) .
KRAMERS,
H. A., 1957, Quantum Mechanics (Amsterdam : North-Holland) (5, 8).
VAK KRANENDONK,
J., and BLOOM,M., 1956, Physica, 22, 545 (9).
KRIVOGLAZ,
M. A., and PEKAR,
S.I., 1957 a, b, Izv. Akad. Nauk, S.S.S.R., Ser. Fiz., 21, 3 ,
16 (6).
KUBO,R., 1952, J . Chem. Phys., 20, 770 (4) ; 1956, Canad. J . Phys., 34, 1274 (7) ; 1957 a,
J . Phys. Soc.,Japan, 12, 570 ( I , 3, 7) ; 1957 b, iVuovo Cim. Suppl., 6, 1063 (9).
H., and HASHITSUME,
N., 1959, J . Phys. Soc., Japan, 14, 56 (7).
KUBO,R., HASEGAWA,
K ~ M M E H.,
L , 1955, Z . Phys., 143, 219 (7).
LAMB,
W. E., and SANDERS,
T. NI., 1960, Phys. Rev., 119, 1901 (9).
LAhIB, W. E., and WILCOX,
L., 1958,J. Phys. Radium, 19, 801 (9).
LANDAU,
L. D., 1930,z. Phys., 64, 629 ( 6 ) ; 1958,J. Exp. Theor. Phys., 34, 262 (Soaiet Phys.,
J E T P , 7, 182) (3).
360
D.ter Haar
LANDAU,
L. D., and LIFSHITZ,E. M., 1958 a, Quantum Mechanics (London: Pergamon)
(I) ; 1958 b, Statistical Physics (London : Pergamon) (I).
LAX,M., 1958, Phys. Rev., 109, 1921 ( I , 7).
LEE,T. D., and YANG,C. N., 1959, Phys. Rev., 113, 1165 (4).
VAN LEEUWEN,
J. M. J., GROENEVELD,
J., and DE BOER,J., 1959, Physica, 25, 792 (4).
LEVINE,
H. B., 1960, Phys. Fluids, 3, 225 (4).
LIPPS, F. W., and TOLHOEK,
H. A., 1954 a, b, Physica, 20, 85, 395 (8).
LONDON,F., and BAUER,E., 1939, La The'orie de l'observation en Michanique Quantique
(Paris : Hermann) ( I , 2, IO).
LOWDIN,P.-O., 1954, Proc. 10th Solvay Congress (5) ; 1955 a, b, c, Phys. Rev., 97, 1474,
1490, 1509 ( 5 ) ; 1956, Advanc. Phys., 5, I (5) ; 1959, Advanc. Chem. Phys., 2, 207 (5) ;
1960, Rev. Mod. Phys., 32, 328 ( I , 5).
LUDWIG,
G., 1953, 2. Phys., 135, 483 ( I O ) ; 1954, Grundlagen der Quantenmechanik (Berlin :
Springer) ( I , IO).
LUNDQUIST,
S.O., 1959, A r k . Fys., 16, NO. 30 (4).
LUTTINGER,
J. M., 1958, Phys. Rev., 112, 739 (7).
LUTTINGER,
J. M., and KOHN,W., 1958, Phys. Rev., 109, 892 (7).
MCCONNELL,
H. M., 1958,J. Chenz. Phys., 28, 1188 (5).
MACKE,W., 1955 a, Phys. Rev., 100, 992 (5) ; 1955 b, Ann. Phys., Lpz., 17, I (5).
MCLENNAN,
J. A., 1959, Phys. Rev., 115, 1405(7).
MCWEENY,
R., 1955, Proc. Roy. Soc. A, 232, 114 (5) ; 1956 a, Proc. Roy. Soc. A, 235, 355 (5) ;
1956 b, Proc. Roy. Soc. A, 237, 496 (5) ; 1957, Proc. Roy. Soc. A, 241, 239 (5) ; 1959 a,
Proc. Roy. Soc. A, 253, 242 (5) ; 1959 b, Phys. Rev., 114, 1528 (5) ; 1960, Rev. Mod.
P ~ Y s .32,
, 335 ( 1 , 5 ) .
MCWEENY,
R., and M~zurio,Y., 1961, Proc. Roy. Soc. A, 259, 554 (5).
MCWEENY,
R., and OHNO,K. A., 1960, Proc. Roy. Soc. A, 255, 367 (5).
MARCH,
N. H., 1957, Advanc. Phys., 6, I (5).
MARCH,Iv. H., and YOUNG,
W. H., 1958, Proc. Phys. Soc., 72, 182 (5).
MARGENAU,
H., 1950, Nature of Physical Reality (New York : McGraw-Hill) ( I O ) .
MARTIN,
P. C., and SCHWINGER,
J., 1959, Phys. Rev., 115, I342 ( I , 3, 5).
MATSUBARA,
T., 1955 a, Prog. Theor. Phys., Japan, 13, 628 (6) ; 1955 b, Prog. Theor. Phys.,
Japan, 14, 351 ( 1 , 3, 4, 5, 6).
MATTIS,
D. C., and BARDEEN,
J., 1958, Phys. Rev., 111, 412 (7).
MAYER,
J. E., 1955, Phys. Rev., 100, 1579 (5, 6).
MAZER,
P., and OPPENHEIM,
I., 1957, Physica, 23, 216 (4).
MIGDAL,
A. B., 1957,J. Exp. Theor. Phys., U.S.S.R., 32, 399 (Soviet Phys.,JETP, 5 , 333) (3).
MIZUNO,
Y . , and ISUYAMA,
T., 1957, Prog. Theor. Phys.,Japan, 18, 33 (5, 6).
MONTROLL,
E. W., and WARD,J. C., 1958, Phys. Fluids, 1, 55 (3, 4 ) ; 1959, Physica, 25, 423
( 1 , 3, 4, 7).
MORI,H., 1956,J. Phys. Soc.,Japan, 11, 1029 (7, 9) ; 1958, Phys. Rev., 112, 1829 (7) ; 1959,
Phys. Rev., 115, 298 (7).
MORI,H., and ONO, S., 1952, Progr. Theor. Phys., Japan, 8, 327 (7).
MORI,H., and Ross, J., 1958, Phys. Rev., 109, 1877 (7).
MORSE,P. M., and FESHBACH,
H,, 1953, Methods of Theoretical Physics (London : McGrawHill) (3).
NAKAJIMA,
S., 1955, Advanc. Phys., 4, 363 (4, 6) ; 1956, Proc. Phys. Soc. A, 69, 441 (6, 7) ;
1959, see Varenna 1959, p. 275 ( 7 ) .
NAKANO,
H., 1956, Progr. Theor. Phys.,Japan, 15, 77 ( I , 7) ; 1957, Progr. Theor. Phys.,Japan,
17, 145 (7) ; 1959, Progr. Theor. Phys.,Japan, 22, 453 ( 7 ) ; 1960 a, b, c, Progr. Theor.
Phys.,Japan, 23, 180, 182, 526 (7) ; 1960 d, Progr. Theor. Phys.,Japan, 23, 527 (9).
NAQVI,M. A., 1959, Nucl. Phys., 10, 256 ( I , 5).
VOX NEUMANX,
J., 1927 a, b, c, Nachr. Ges. Wiss. Gottingen, I , 245, 273 ( I ) ; 1932, Mathematische Grundlagen der Quantenmechanik (Berlin : Springer) (Mathematical Foundations
of Quantum Mechanics, Princeton University Press, 1955) ( I , IO).
NISHIKAWA,
K., 1960, J . Phys. Soc., Japan, 15, 78 (7).
NITSOVICH,
M. V., 1959, Fiz. Met. Metallov., 7, 641 (Phys. Met. Metallogr., 7,No. 5 , I ) (6).
361
ONO,S., 1953,Proc. Kyoto and Tokyo Conf. Theor. Phys., 449 (7); 1954,Prog. Theor. Phys.,
Japan, 12, 113 (7); 1958,see Prigogine 1958,p. 229 (6,7).
OPECHOWSKI,
W.,
1953,Rev. Mod. Phys., 25, 264 (9).
OPPENHEIM,
I., and MAZUR,P., 1957,Physica, 23, 197 (4).
OPPENHEIM,
I.,and Ross, J., 1957,Phys. Rev., 107,28 (4).
OSAKA,
Y., 1959,Prog. Theor. Phys.,Japan, 22,437 (6).
PARRENT,
G.B., and ROMAN,P., 1960,Nuovo Cim., 15,370 (8).
PAVLIKOVSKI!,
A., and SHCHURUVNA,
V., 1958,Dokl. Akad. Nauk, S.S.S.R., 118, 61 (Soviet
Phys., Doklady, 3, 71)(4); 1959,Dokl. Akad. Nauk, S.S.S.R., 124, 69 (Soviet Phys.,
Doklady, 4, 95) (4).
PEACOCK,
T. E., and MCWEEKY,
R., 1959,Proc. Phys. Soc., 74, 385 ( 5 ) .
PENROSE,
O., 1960,Proc. Roy. Soc. A, 256, 106(5).
PITAEVSKI~,
L.P., 1959,J.
Exp. Theor. Phys., U.S.S.R.,36,1168(SovietPhys.,JETP,9,830)(3).
PRANGE,
R., and KLEIN,A., 1958,Phys. Rev., 112,1008(3).
PRIGOGINE,
I. (editor), 1958,Transport Processes in Statistical Mechanics (New York : Interscience) ( I , 7).
PRIGOGINE,
I., and BALESCU,
R., 1959a, b, Physica, 25, 281, 302 (7).
PRIGOGINE,
I.,and ONO,S., 1959,Physica, 25, 171(7).
PRYCE,
M. H.L., and WARD,J. C., 1947,Nature, Lond., 160,435 (8).
REDFIELD,
A. G.,
1957,I.B.M.J. Res. Developm., 1, 19(9); 1959,Phys. Rev., 116, 315 (9).
RIESENFELD,
W. B., and WATSON,K. M., 1956,Phys. Rev., 104, 492 (4).
ROMAN,
P., 1959,Nuovo Cim., 13,974 (8).
VON Roos, O., 1960,J.
Math. Phys., 1, 107(7).
ROSS,J., and KIRKWOOD,
J. G., 1954,J . Chem. Phys., 22, I094 (7).
S A I T N.,
~ , 1960,Phys. Rev., 117, 1163(7).
SAKAI,1937,Proc. Math. Phys. Soc.,Japan, 19, 172 (IO).
SALAM,
A., 1953,Prog. Theor. Phys.,Japan, 9,550 (3, 6).
SALPETER,
E.E.,1958,Ann. Phys., IYew York, 5 , 183(4).
S A R O L ~L.,
A , and KOPPE,H., 1958,Z.Phys., 151,385 ( 5 ) .
SAWICKI,
J., 1960,Prog. Theor. Phys.,Japan, 24, 213 (3).
SCHAFROTH,
M. R., 1951,
Helv. Phys. Acta, 24, 645 (6).
SCHAFROTH,
M. R., BUTLER,S.T., and BLATT,J. M., 1957,Helv. Phys. Acta, 30, 93 (4,6).
SCHRODINGER,
E.,I 948,Statistical Thermodynamics (Cambridge : University Press) ( I ) .
SCHURIACHER, R. T., 1958,
Phys. Rev., 112,837 (9).
SEIDEN,
J., 1956a, b,C.R. Acad. Sci., Paris, 242,454,763 (9).
SHER,A., and PRIMAKOFF,
H., 1960,Phys. Rev., 119, 178(9).
SHULL,H.,
1959,J.
Chem. Phys., 30, I405 ( 5 ) .
SIEGERT,
A. J. F., 1960,Physica, 26, S30 (4).
SIEGERT,
A. J. F., and TERAMOTO,
E., 1958,Phys. Rev., 110, 1232 (4).
SKROTSKI!,
G.V.,and KOKIN,A. A., 1959,J.Exp. Theor. Phys., U.S.S.R., 36, 169 (Soviet
Phys.,JETP, 9, 116)(9).
SNYDER,
H. S., PASTERNACK,
S., and HORNBOSTEL,
J., 1948,Phys. Rev., 73,440 (8).
SOLOMON,
I., 1958,Phys. Rev., 110,61(9).
SONDHEIMER,
E.H.,
and WILSON,
A. H., 1951,
Proc. Roy. Soc. A, 210, 173(6).
STILLINGER,
F.H.,
and KIRKWOOD,
J. G., 1960,Phys. Rev., 118, 361 (4).
STOKES,
G.G.,
1852,Trans. Camb. Phil. Soc., 9, 399 (8).
SUHL,H.,
1958,J . Phys. Chem. Solids, 4, 278 (9).
TAKABAYASI,
T., 1954,Prog. Theor. Phys.,Japan, 11, 341 (4).
THOULESS,
D. J., 1957,Phys. Rev., 107, 1162(4); 1959,Phys. Rev., 116, 21 (4); 1961,
Nucl. Phys., in the press (5).
TODA,
M., 1958,J.
Phys. Soc.Japan, 13, 1266(7).
TOLHOEK,
H.A., 1956,Rea. Mod. Phys., 28, 277 (8).
TOLHOEK,
H. A., and Cox, J. A. M., 1953,Physica, 19, I O I (8).
TOLHOEK,
H. A., and DE GROOT,S.R., 1951a, b, c, Physica, 17, I, 17,81 (8).
TOLMAK,
R. C., 1938,The Principles of Statistical Mechanics (Oxford : University Press)
(1, 2,4,10).
362
D.ter Haav
TOMITA,
K., 1958 a, Prog. Theor. Phys.,Japan, 19, 541 (9) ; 1958 b, Prog. Theor. Phys.,Japan,
20, 743 (9).
TREDGOLD,
R. H., 1957, Phys. Rev., 105, 1421 (5, 6).
VALATIN,
J. G., 1958, Xuovo Cim., 7, 843 (6).
VAKHOVE,L., 1955, Physica, 21, 517 (3, 7) ; 1957, Physica, 23, 441 (3, IO) ; 1959, Physica, 25,
268 (IO).
V.\REKNA,
1959, Proceedings of the ' Enrico Fermi' International Summer School on Thermodynamics of Irreaersible Processes (Bologna : Zanichelli) (7).
VEDEXOV,
A. A., 1959,J. Exp. Theor. Phys., U.S.S.R., 36, 641 (Soziet Phys.,JETP, 9,446) (3).
VEDESOV,A. A., and LARKIN,
A. I., 1959, J . Exp. Theor. Phys., U.S.S.R., 36, 1133 (Societ
Phys., J E T P , 9, 806) (3).
WAKITA,
H., 1960, Prog. Theor. Phys.,Japan, 23, 32 (IO).
WAKGSNESS,
R. K., 1956 a, Phys. Rev., 101, I ( 9 ) ; 1956 b, Phys. Rev., 104, 857 (9).
WANGSNESS,
R. K., and BLOCH,F., 1953, Phys. Rev., 89, 728 (9).
WANNER,G. H., 1959, Elements of Solid State Theory (Cambridge : University Press), Ch. 7
(9).
WATSON,
K. M., 1956, Phys. Rev., 103, 489 ( 3 , 4).
VON WEIZSACKER,
C. F., 1935, 2. Phys., 96, 431 (5).
WICK,G. C., 1950, Phys. Rev., 80, 268 (4).
WIGNER,E., 1932, Phys. Rev., 40, 749 (7).
WILCOX,
L. R., and LAMB,
W. E., 1960, Phys. Rev., 119, 1915 (9).
WILSON,
A. H., 1953, Theory of Metals (Cambridge : b-niversity Press) (I).
WOLF,E., 1954, h'uovo Cim., 12, 884 (8) ; 1959 a, Nuoco Cim., 13, 1165 (8) ; 1959 b, Proc.
Phys. Soc., 74, 269 (8) ; 1960, Proc. Phys. Soc., 76, 424 (8).
YAGLOM,
A. YA., 1956, Teor. Veroyatn. Primenen., 1, 161 (4).
YAMAMOTO,
T., 1960, Phys. Rev., 119, 701 (6).
YATSIV,S., I9j9, Phys. Rev., 113, 1522 (9).
YOKOTA,
T., 1960, J . Phys. Soc., Japan, 15, 779 (7).
YOVNG,W. H., and MARCH,N. H., 1960, Proc. Roy. Soc. A, 256, 62 (5, 6).
YVOX,J., 1957, iVucl. Phys., 4, I (4) ; 1958, C.R. Acad. Sci., Paris, 246, 2850; 1960 a, J .
Phys. Radium, 21, 505 (4, 9 ) ; 1960 b,J. Phys. Radium, 21, 569 (4).
ZAIDI,M. H., 1959, Phys. Rev., 116, 241 (8).
ZELAZNY,
R., 1960, Phys. Rea., 117, I (7).
ZIGENLAUB,
R., 1959, Fix. Tverd. Tel., 1, 1053 (Soviet Phys., Solid State, 1, 964) (7).
ZUBAREV,
D. S . , 1954, Dokl. Akad. Nauk, S.S.S.R., 95, 757 (4, 6) ; 1960, Usp. Fiz. Nauk, 71,
71 (Soviet Phys., Uspekhi, 3, 3 2 0 ) ( I , 3, 7).

Theory and Application of Density Matrix

Uploaded by

Copyright:

Available Formats

Theory and Application of Density Matrix

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Theory and Application of Density Matrix

Uploaded by

Copyright:

Available Formats

Home

Theory and applications of the density matrix

Please note that terms and conditions apply.

THEORY AND APPLICATIONS OF THE

General properties of the density matrix

Theory and Applications of the Density Matyix

Theovy and Applications of the Density Matrix

the change in which is governed by a Hamiltonian. If we were interested in the

Theory and Applications of the Density Matyix

developments in the evaluation of the equilibrium density matrix of systems of

Theory and Applications of the Density Matrix

Theory and Applications of the Density Matrix

or, using Eqn (Ll),

Introducing the density matrix i; by its matrix elements in the p,-representation,

we can write Eqn (2.4) in the form

<A)= c,, n Amn Pnm = T r (?A^),

where a,, is the Kronecker delta function.

Theory and Applications of the Density Matrix

and using Eqn (2.15) we get from this equation

dqk = Enenk Snp*.

If we denote the transformed density matrix by a prime, we have instead of

X:k,m,n enk* Sng emk s m p *

or, in matrix notation,

which is the normal equation for the transformation of an operator.

(A) = T r ?A = T r S-I p^SS-lAS= T r S-1 BAS

T r SS-l p^A= T r p^A= (A),

where we have used the property of the trace

CZ (0) cn(o)* = Cm(O) cn(0)* =

From Eqn (2.6) we then get

Theory and Applications of the Density Matrix

Z? is the Hamiltonian of the system. I t is convenient

where we have used the definition of the trace in coordinate representation.

(XI i; I x> = Em,,

We then find for i; the following equation of motion.

amnih[$n*(x, t ) 4m(x, t ) + & * ( X I ,

- ?Lm(x,t ) $hn*(x,t ) dx(x

in accordance with Eqn (2.37).

x(x) dxx*(x)~ ( x =) x*(x) x(x)

that is, the condition (2.26) for a pure case is satisfied.

Theovy and Applications of the Density Matvix

where the density matrix

PA for the sub-system

and where A^ is an operator acting upon the sub-system A only,

Theory and Applications of the Density Matyix

T h e propagator Green function occurs in field theory when one discusses

where we have used Eqn (2.9) and the fact that

It is of interest for statistical applications of Green functions to consider also

where we have assumed two-body interactions, and where u ( x , x ) is the potential

- - d3 X a ( x , x) G+l$2,;;$,t = S(x - x ) 8(t - t).

Theory and Applications of the Density Matrix

which is most easily attained by letting p^ be a function of I?. We have mentioned

where k is Boltzmann's constant, and that v = pp, where p is the chemical or

? and the potential

We then have from Eqn (4.6), which is also satisfied by $(p),

This equation reduced to the equation of stationary state scattering

Theory and Applications of the Density Matrix

We shall consider a system of bosons which interact through two-body forces.

$o(P> = exp ( - PQO).

amnih[$n(x, t ) 4m(x, t ) + & ( X I ,

x(x) dxx(x)~ ( x =) x(x) x(x)

($* V$ - $V$*)-. -** A