Sunteți pe pagina 1din 6

Mathematical Proceedings of 

the Cambridge Philosophical 
Society
http://journals.cambridge.org/PSP

Additional services for Mathematical 
Proceedings of the Cambridge 
Philosophical Society:

Email alerts: Click here
Subscriptions: Click here
Commercial reprints: Click here
Terms of use : Click here

The basis of statistical quantum 
mechanics
P. A. M. Dirac

Mathematical Proceedings of the Cambridge Philosophical Society / Volume 25 / 
Issue 01 / January 1929, pp 62 ­ 66
DOI: 10.1017/S0305004100018570, Published online: 24 October 2008

Link to this article: http://journals.cambridge.org/
abstract_S0305004100018570

How to cite this article:
P. A. M. Dirac (1929). The basis of statistical quantum mechanics. 
Mathematical Proceedings of the Cambridge Philosophical Society, 25, 
pp 62­66 doi:10.1017/S0305004100018570

Request Permissions : Click here

Downloaded from http://journals.cambridge.org/PSP, IP address: 147.226.7.162 on 22 Apr 2013
62 Dr Dirac, The basis of statistical quantum mechanics

The Basis of Statistical Quantum Mechanics. By P. A. M.


DXBAC, Ph.D., St John's College.
[Received 17 October, read 29 October 1928.]
Tn classical mechanics the state of a dynamical system at any
particular time can be described by the values of a set of co-
ordinates and their conjugate momenta, thus, if the system has
n degrees of freedom, by In numbers. In quantum mechanics, on
the other hand, we have to describe a state of the system by a
wave function involving a set of coordinates, thus by a function
of n variables. The quantum description is, therefore, much more
complicated than the classical one. Let U3 consider, however, an
ensemble of systems in Gibbs' sense, i.e. not a large number of
actual systems which could, perhaps, interact with one another,
but a large number of hypothetical systems which are introduced
to describe one actual system of which our knowledge is only of a
statistical nature*. The basis of the quantum treatment of such
an ensemble has been given by Neumann f. The description
obtained by Neumann of an ensemble on the quantum theory is
no more complicated than the corresponding classical description.
Thus the quantum theory, which appears to such a disadvan-
tage on the score of complication when applied to individual
systems, recovers its own when applied to an ensemble. It is the
object of the present note to examine this question more closely
and to show how complete the analogy is between the quantum
and classical treatments of an ensemble.
Classical Theory.
We shall 'first repeat briefly the well-known basis of classical
statistical mechanics, in order that it may be ready to hand for
comparison with the quantum theory. The state of a system of
n degrees of freedom can be described by its coordinates and
momenta qr, pr (r = 1,2 ... n) and can so be represented by a point
in 2n-dimensional phase space. This point will move according to
Hamilton's equations

^=wr[qr'Hl Pr
=-dq-r=[Pr>U] (1)>
where the brackets [ ] denote a Poisson bracket expression. An
ensemble of systems can be represented by a dust of points in this
* The so-called " statistics" of Einstein-Bose or Fermi applies only to an
assembly of actual systems which could interact with each other, and has no
meaning for a Gibbs' ensemble.
+ J. v. Neumann, Gutt. Nachr. 1927, p. 245. Principally section in.
Dr Dime, The basis of statistical quantum mechanics B3
space, which dust will be describa"ble by a density function p (pr, qr),
giving the number of points per unit 2w-dimensional volume in
the neighbourhood of the place (pr, qr). This density function
must satisfy the hydrodynamical equation of continuity, which is

In this equation we can substitute for qr, pr their values given by


(1), which gives

dt~ ~r\dqr\Pdpr) dprVdqr

r
\dqrdpr dprdqr
(2).
This is the equation of motion that governs the ensemble.
If p is considered as a function of the coordinates and momenta
of a single system, its equation of motion is

of which equations (1) are particular cases. This equation is the


same as (2) except for the sign. The rate of change of p following
the motion of the dust, usually denoted by Dp/Dt, is given by the
sum of its rates of change according to (2) and (3), and is therefore
zero. This constitutes Liouville's theorem.
Quantum Theory.
We shall now consider how to describe an ensemble in quantum
mechanics. We can no longer represent each of the systems by a
point in phase space, since we are not allowed to attach numerical
values to the coordinates and momenta simultaneously. Instead
we must describe each system by a wave function. We may
suppose any wave function -v/r to be expanded in terms of a set of
normalised orthogonal >}rm's, thus

and the coefficients cm will then define y(r completely. Each system
can thus be described by a set of numbers cm. If we suppose for
definiteness that the number of ylrm's is finite, equal to N say, then
the cm's may be regarded as the coordinates of a point in
iV-dimensional space, so that each system may be represented by
a point in this space. The ensemble can thus be represented by
a dust of points in the i\r-dimensional space. In general the cm's
64 Dr Dirac, The basis of statistical quantum mechanics
are complex numbers. When this is the case, to get a representa-
tion in real space it would be necessary to split up each cm into
its real and imaginary parts, so that we should obtain a 2 JV-dimen-
sional space.
It may easily be proved that the rate of change following the
motion of the dust in the i^-dimensional space, or, in the general
complex case, in the 2iV-dimensional space, is zero. One might
think that one had here the analogue of Liouville's theorem. The
number of dimensions of the space with which we are here dealing
is, however, usually much greater than in the classical Liouville's
theorem, since N is usually much greater than n, and is, in fact,
infinite in most practical cases. Our quantum statistical mechanics
would thus be much more complicated than the classical one if we
allowed it to stop here. We can, however, effect a great simplifica-
tion in it and obtain a much closer analogy with the classical
theory.
Let us inquire into what observable facts we can calculate
about the ensemble. Quantum mechanics does- not allow us to
calculate the value of a variable for a system in a given state, but
only its average value when the observation of the variable is
repeated a large number of times upon systems in the same state.
If we do not know .precisely which state a system is in but only
that it can be described by a certain Gibbs' ensemble, then the
only observable thing which we can calculate about it is the average
value of a variable for the whole ensemble. If x is any variable, it
can be represented by a Hermitian matrix xmn whose rows and
columns refer to the yjrm's in (4). The average value of x for
a system in the state given by the ^r of equation (4) is then

The average value of x for the ensemble is thus


U i g Zimn Cm Xmn C« = U 4 f f l n 00mn 2., Cm Cn,
where u is the number of systems in the ensemble and £„ denotes
a summation over these systems. Thus, if we know the value of
^sCmcn .(5)
we know enough about the distribution to be able to calculate the
average value of any variable x, so that (5) is an adequate
description of all observable properties of the ensemble.
In the case when the cTO's are all real so that an iV-dimensional
representation is possible, (5) has a simple geometrical interpreta-
tion. When n = rn, (5) becomes 2«cm2, which is the second moment
of the dust of representative points with respect to the hyper-
plane cm = 0, while when n + m we get a "product of inertia" with
respect to the two hyperplaues cm = 0, cn = 0. The quantities (5)
Dr Dirac, The basis of statistical quantum mechanics 65
are thus just the coefficients that enable one to obtain the moment
of inertia of the dust about an arbitrary axis through the origin.
It is only the moments of inertia of the dust that are of any
physical importance, the finer features of the distribution of
representative points being unobservable. When the cm's are
complex and we have to use the 2i\T-dimensional space, the
expression (5) still gives second moments of the dust, but they are
now of a less familiar kind.
Put u~1'E8cmcn = pnm --(6).
The numbers pnm are the elements of a Hermitian matrix p, and
this matrix describes the ensemble adequately for all physical
purposes. The average value of any variable x is given by
m = -D (xp) = D {px) (7),
where D denotes the diagonal sum. This form for the average
value of any variable for an ensemble has been obtained previously
by Neumann by a very simple argument. Neumann points out
that the average value of a variable a; is a linear function of that
variable and is thus a linear function of the matrix elements
xmn representing it. Hence this average value must be of the
form (7), where the pnm's are coefficients capable of adequately
describing the ensemble.
We shall now obtain an equation for the rate of change of the
pKm's. This will be the equation of motion that governs the
ensemble. The cn's vary according to the well-known equations
ihcn-2kHnkck,
-ihcm=2kckHkm,
where the matrix Hnk represents the Hamiltonian. Hence from (6)
p'nm = w1 !„ {ihcncm - cn(— ihcm)\
- cnckHkm]

This may be written


ihp = Hp — pH
or P = -[p,H] (8),
where the brackets [ ] denote the quantum analogue of a Poisson
bracket expression. This equation is formally the same as (2). It
thus appears that the matrix p denned by (6) is the analogue of
the density of the representative dust in phase space in the classical
theory.
VOL. XXV. PART I. c
66 Dr Dirac, The basis of statistical quantum mechanics
This matrix p may be transformed to a scheme in which the
coordinates qr are diagonal matrices. Its general ma.trix element
will then be afuuction of the variables qi, q2'... qn' and q^', q2"... qn",
which are In in number. Thus the quantum description of the
ensemble requires a function of the same number of variables as
the classical one. We could alternatively regard p as a function of
the non-commuting quantum variables qr, pr (since any matrix
expressible in the ^-scheme may be so regarded) and this makes
the analogy with the classical theory most perfect. If p were
an ordinary dynamical variable of a single system, instead of being
a description of an ensemble, it would vary with the time accord-
ing to the equation

analogous to the classical equation (3). If we define Dp/Dt to be


the sum of the rates of change of p in the two cases we get the
result Dp/Dt = 0. This appears to be the true analogue of
Liouville's theorem. We can no longer give any graphical mean-
ing to Dp/Dt, but for all analytical purposes the analogy is perfect.
To obtain a basis for statistical mechanics we must look for
a density function that remains constant even when arbitrary
perturbations are added to the Hamiltonian. We can then assume
that this density function gives the a priori distribution of the
ensemble. On the classical theory we have the well-known result
that a uniform density over the whole of phase space has this
property, i.e. the density
P=c (9),
where c is a number. From (8) we see that this makes p vanish
also on the quantum theory. Thus we can take (9) to be the
a priori density function also on the quantum theory. It has been
shown by Neumann* that this leads at once to the usually assumed
result that all the discrete states of a system have the same
a priori probability.
Note added in proof.—The analogy between the methods of
obtaining averages from the classical and from the quantum
density function should be pointed out. In the classical theory,
to obtain the average of a variable x for the ensemble, we multiply
x by p and integrate over the whole of phase space, i.e. over all
states. In the quantum theory we also multiply x by p and we
then take the diagonal sum, which means summing over all states.
* Loc. cit., p. 269.

S-ar putea să vă placă și