You are on page 1of 44

[,

190;~
"2o S e n i ~ , s t r ~

: ~ U P P I . E M I ~ \ ' I ' ~ AI. VOIA',IIq IT. Slqlllt'; \


liE].

\['Ct\'l)

N.

('I\IFVF()

The General Statistical Problem in Physics and the


Theory ot Probability.
1). Bolr_~r (*) a n d W. S~Ht~'rZER
l.a,.,ld,

d~, tte I"ilo.~'o,ii,t.

('i~,t,i,.~

~, l, eh'a.~ - I , i c e r . ~ i d c t d e

dc ,~,'~io t ' a , i o

(rivt~xut~) il 1,5 Marz~ I!i5.5)

('4~xrt.:x'r.. 1. lntroductiov_.
2. Chance and Statistical Law. - 3. The
~'on~-epi of Probabilily.
4. The Irregular Collectives of von Mises. 5. The Modern Axiomati(. School. 6. On the General Statistical Problem
ax It 3,ri,~es in Statistical Mechanics.
7. A Further Example of the G-eneral Statistical Problem. The Deduction of a Probabilistie Law from
a I)eterminate lmw. - 8. The General Applicability of the Calculus of
Probability. - 9. A Criterion for Randomness.
19. Further Illustration
in Term~ of Random Pha,aes. -- 11. ('onelusion.

I.

Introduction.

Basi~. s t a t i s t i e a l p r o b l e m s in physics, (e.g., those eonneet.ed with s t a t i s t i c a l


m e c h a n i c s a n d the q u a n t u m t h e o r y ) h a v e , as in o t h e r fields, so g e n e r a l l y b e e n
~'iven a m a t h e m a t i u a l f o r m u l a t i o n in t e r m s of the t h e o r y of p r o b a b i l i t y , t h a i
i h e i m p r e s s i o n is ereat(,d t h a t p e r h a p s no o t h e r k i n d of f o r m u l a t i o n of such
p r o b l e m s is e v e n possible.

I n the p r e s e n l p a p e r , h o w e v e r , we wish to r~ise

th(, q u e s t i o n of w h e t h e r t h e t h e o r y of p r o b a b i l i t y r e a l l y r e p r e s e n t s t h e m o s t
a d e q u a t e possible m a t h e m a t i c a l i n s t r u m e n t in all of those e~ses to which it
is c u s t o m a r i l y applied. I n d e e d , as we shall see in this p a p e r , t h e r e is a wide
ran~'e of s~atisiieal a p p l i c a t i o n s in physics, in which a more reliable a n d a clearer
t r e a t m e n t can he o b t a i n e d b y using' m e t h o d s o t h e r t h a n those of the t h e o r y

t*) N~w ml h,ave of abs(qwe at Tr


lsr.el.

]laifa ] n s t i l . l e of Tectmolo~y, Haifa,

T H E G E N E R A L S T A T I S T I C A L P R O B L E M I N P H Y S I C S A N D T H E T H E O R Y OF P R O B A B I L I T Y

1005

of probability. I n the study of such applications, one sees t h a t w h a t is relevant is not so much the theory of probability, b u t r a t h e r w h a t we cull (~the
general statistical problem ~>, of which the problem of probability is only a
special case.
The general statistical problem m a y be stated thus: To find those relationships which hold in a statistical aggregate independently of a wide range
of variations of the detailed properties of the individual members of the aggregate, the size of this range of variation depending on the physical conditions
and the context of interest.
I t is evident t h a t if relationships of the t y p e described above hold, then
w h a t e v e r m a y be the character of these relationships, we shall be able to make
statistical predictions concerning the aggregate without having to t r e a t the
problem of just what the individual members are doing. I n those cases to which
the t h e o r y of probability applies, such statistical predictions can be made with
the aid of the well k n o w n probability calculus; b u t as we shall show in this
paper, there are a n u m b e r of r a t h e r simple problems, in which it is highly
advantageous to go outside the scheme of expressing the statistical relationship
in terms of nothing b u t the concept of probability. These examples will also
provide further insight into the meaning of the concept of probability; for
t h e y enable us to see this concept inside of the broader setting of statistical
laws in general.
After formulating the general statistical problem with the aid of certain
examples drawn from the field of statistical mechanics, we shall focus our
a t t e n t i o n on a few cases to which the theory of probability applies. Here,
we shall discuss the problem of how probabilities are related to determinate
laws, and even to a more general class of laws lying between the determinate
and the probabilistic types. Such a process of relating probabilistic laws to
other types of laws will also enable us to see more clearly w h a t is the n a t u r e
of randomness. Moreover, as we shall see, studies of this kind lead us into
new kinds of problems t h a t cannot even be f o r m u l a t e d in terms of the t h e o r y
of probability itself, such as t h a t of estimating in specific cases the error and
d o m a i n of applicability of various aspects of the t h e o r y of probability.
The general plan followed in this paper will be first to review some of the
basic quantitative ideas on the theory of probability t h a t have already been
current, t h e n to go on to present the (, general statistical problem ~, and t h e n
to specialize to the problem of probability. W e wish to m a k e it clear a t the
outset, however, t h a t an appreciable n u m b e r of points appearing in this p a p e r
will be a repetition, in our own terms of course, of ideas t h a t have already
been published elsewhere. This repetition is, however, essential for the clarity
of presentation of the material, because our principal conclusion is really t h a t
it is just the combination of these various ideas, which h a v e appeared mainly
in a scattered form until now, combined with our own results, which indicates
66 - Supplemento al Nuovo Ui~'~tenlo.

](ill6

It.

I'>I)IIM

a l i ( l W. 5('Ili:'I'ZLP

the i)ossil)ility of a more ~'eneraI p o i n t of view t o w a r d s t a t i s l i e a l i)rot)lems l l m n


has t h u s far ~'enerally b e e n considered.

2.

- Chance

and

Statistical

Law.

T h e most i m p o r t a n t an,1 c h a r a c t e r i s t i c apl)lieatioll of the t h e o r y of p r o t . ~ b i l i t y is t() a s t a t i s t i c a l ag~'regate of ,)bjeets, e v e n t s , or p h e n o m e n a s u b j e e t


in some way 1(, the a e l i o n of (, ('han(.e ,). To k n o w w h a t p r o b a b i l i t y m e a n s , it
is therefore necessary 1o h a v e a deliniti(m of (Qlallee. V~(, shall give here such
a d e l i n i t i o n (*).
In order to b r i n ~ out what is m e a n t l)y c h a n c e , we m a y e o u s i d e r a t y p i c a l
('hance e v e n t : mamely, a n a u t o m o b i l e a c c i d e n t . Now, il is e v i d e n t t h a t j u s t
wllel'e, WhOll a n d how a l ) a r l i e u l a r a e e i d e n t oe(,urs d e p e m l s on all ellorlnol.lS
mmfl)er of more or less e q u a l l y i m l ) o r l a n I factors, a slight ehal~ge of a n y one
of which (,(mid g r e a t l y ('llallKe tile c h a v a e l e r of the a c c i d e n t , or e v e n a v o i d
it alto~'ether (-). I l e n c e , r e l a l i v e to a c o n t e x t in whieh ~ e eonsider, for e x a m p l e ,
lhe l)reeauti(ms t h a t c~;n be t a k e n 1)y a p a r t i c u l a r m o t o r i s l , each a e e i d e n t has
a n aspeel t h a i is /or:,iton,~,. l~y this, we m e a n t h a t w h a t h a p p e n s is co~tinftc~t
on i n ( t e p e n d e n l f a e l o r s o('('urrinlz o u t s i d e the e o n l e x t i n (luestion, which h a v e
no essenlial r e l a t i o n s h i p to the ( ' h a r a e t e r i s t i e I r a i t s t h a t define j u s t what sort
of a person this m o t o r i s t is a n d how he will b e h a v e in a ~'iven s i t u a t i o n . For
this reason, we say 1hal r e l a t i v e Io such a c o n t e s t , a 1)artieular collision is n o t
all i n e v i t a b l e ( l e v e l o p m e n l , bul r a t h e r t h a t it is a n a c c i d e n t a n d eomes a b o u t
1)y ehan('e from which it also follo~vs t h a t ,'itkb~ tki,~ eont('.rt, t h e q u e s t i o n of
just where, when, a l n l how sueh a collision will t a k e place, as well as t h a t of
w h e t h e r it ~ill t a k e 1)laee or nol, is Unlwe(tietable.
Let us no~ eonsider a series of s i m i l a r a c c i d e n t s . Here, we n o t e t h a t t h e r e
is a (,omplieated an(l unpredietM)le v a r i a t i o n or f l u c t u a t i o n i n t h e precise d e t a i l s
of the v a r i o u s a c c i d e n t s (e.~'., precisely when a n d where t h e y t a k e 1)lace, precisely w h a t is destroye(i, etc.). T h e origin of this v a r i a t i o n is easily m l d e r s t o o d .
since as we have seen, the llUlllerolls causes Oil which the a(,eident d e p e n d s
are e s s e n t i a l l y i n d e p e m l e n t of each o t h e r a n d of a n y t h i n g t h a t a p p e a r s i n t h e
c o n t e x t of the a c t i o n s t a k e n 1)y t h e p a r t i e u l a r m o t o r i s t miller discussion. Since

(*) .\fler x~t, had lildsll4.d lhi.~ work. Prof. 31Al;lo 5cltOxl;Hr
called our attenti~m
lo lhe facl lhal ~oll c of lhe ideas ou chance and probabilily appearing in this paper
had already I>e(ql ~ngge~led by I'OI'RNIV]"(Erpositio]t ilO 1r ~]ts
de8 ('lt(o~(;(t6" et den pr,hal~ilih;.% Paris, I s43).
(-) E.o.: one of the molorists could have slarled lea seconds earlier or later, or
slowed down lo avoid a cat I]iat happened Io eros,- lh(, road, or he could have o'iwqt
lh(' wheel a slightly different tin'n, etc..

T H E G E N E R A L S T A T I S T I C A L P R O B L E M IN P H Y S I C S AND T H E T H E O R Y OF P R O B A B I L I T Y

I007

each of these essentially i n d e p e n d e n t sets of causes will in general act a t different and hot s y s t e m a t i c a l l y related times a n d places, a n d in different a n d
not s y s t e m a t i c a l l y related w a y s in all other respects, the net result will be
j u s t the complicated fluctuations in the details of different accidents which
we h a v e described above, fluctuations which will be unpredictable in the c o n t e x t
u n d e r discussion (although t h e y are in principle predictable in a sufficiently
b r o a d context).
As the n u m b e r of accidents under consideration becomes larger a n d larger,
however, new properties begin to a p p e a r ; for one finds t h a t individual variations t e n d to cancel out a n d regular statistical trends begin to show t h e m selves. Thus, the t o t a l n u m b e r of accidents in a p a r t i c u l a r region generally
does not change m u c h f r o m y e a r to y e a r ; a n d the changes t h a t do t a k e place
often show a regular s y s t e m a t i c trend. Moreover, this t r e n d can be altered
in a s y s t e m a t i c w a y b y the a l t e r a t i o n of specific factors on which accidents
depend. Thus, when laws punishing careless driving are enforced, or when
regular inspection of mechanical p a r t s is required, the m e a n accident r a t e in
a given region always undergoes a downward trend. I n the case of an indiv i d u a l m o t o r i s t taki~g a p a r t i c u l a r trip, no v e r y definite predictions c a n i~
general be m a d e concerning the effects of such measures, since there are still
a n enormous n u m b e r of causes of accidents t h a t h a v e not y e t been eliminated;
y e t statistically, as we h a v e seen, variations in a p a r t i c u l a r cause produce a
regular a n d predictable t r e n d in the effect.
The b e h a v i o r described a b o v e is f o u n d in a v e r y wide r a n g e of fields including social, economic, medical a n d scientific statistics, a n d a whole host of
o t h e r applications. I n all these fields, the following are the m o s t i m p o r t a n t
characteristics:
1) E a c h e v e n t or result depends contingently on a v e r y large n u m b e r
of essentially i n d e p e n d e n t causal factors lying outside the c o n t e x t u n d e r discussion.
2) These causal factors all h a v e c o m p a r a b l e orders of i m p o r t a n c e (*~)
{although t h e y will not in general be e x a c t l y equal in their effects).
3) These causal factors v a r y sufficiently so t h a t in a series of similar
events~ the events of interest fluctuate in a w a y the details of which c a n n o t
be predicted f r o m a n y d a t a t h a t can be t a k e n solely in the n a r r o w e r c o n t e x t

(*) If the effects of a few of the factors are many orders of magnitude more important than the effects of all the others, combined, then these factors would be regarded as operating systematically. The event in question would still be contingent
on these causal factors, but it would be inappropriate to describe their effects as
being those of chance.

1008

D. BOIt:~! a n d ~V.SCItI'TZER

u n d e r d i s c u s s i o n . It celt be p r e d i c t e d , ho~vever, t h a t in a l o n g e l t o u g h s e r i e s
of e v e n t s , e v e r y p o s s i b i l i t y will be b r o u g h t a b o u t in t h e s e f l u c t u a t i o n s . M o r e o v e r , t h e m o r e s e n s i t i v e l y d e p e n d e n t t h e e v e n t of i n t e r e s t is on t h e e x t e r n a l
c a u s a l f a c t o r s t h e m o r e f a v o r a b l e a r e t h e c o n d i t i o n s for t h e c r e a t i o n of e h a n e e
f l u c t u a t i o n s of a p p r e c i a b l e i m p o r t a n c e .
4) i n a n a g g r e g a t e c o n t a i n i n g m a n y m e m b e r s , t h e c h a n c e f l u c t u a t i o n s
t e n d to cancel o u t , so t h a t t h e a v e r a g e or l o n g - r u n p r o p e r t i e s c a n b e p r e d i c t e d
a p p r o x i m a t e l y on t h e b a s i s of statistical l~~cs. S u c h l a w s c a n b e a p p l i e d w i t h o u t
t h e n e e d to ~'o ~o t h e b r o a d e r c o n t e x t in w h i c h one w o u l d t a k e i n t o a c c o u n t
the a d d i t i o n a l c a u s a l f a c t o r s t h a t .a'overn t h e d e t a i l s of t h e f l u c t u a t i o n s of t h e
i n d i v i d u a l m e m b e r s of t h e afz~'regate.

3. -

The Concept

of P r o b a b i l i t y

l t i s t o r i c a l l y , t h e c o n c e p t of p r o b a b i l i t y was first g'iven a p r e c i s e f o r m in


c o n n e c t i o n w i t h g'ambling" ~:'ames. .\ ~'ood e x a m p l e is f u r n i s h e d b y t h e ~zame
of dice. Now, if we follow t h e r e s u l t s of e a c h i n d i v i d u a l t h r o w s of t h e d i c e ,
we d i s c o v e r t h a t t h e r e s u l t s t h l c t u a t e f r o m one t h r o w to t h e n e x t , i n t h e s a m e
c h a r a c t e r i s t i c w a y t h a t is d e m o n s t r a t e d b y a u-ide r a n g e of c h a n c e p h e n o m e n a
of t h e t y p e d i s c u s s e d in See. t. N e v e r t h e l e s s , g a m b l e r s h a v e d e v e l o p e d t h e
c u s t o m of b e t t i n g on a I d v e n c o m b i n a t i o n a n d of g i v i n g c e r t a i n o d d s t h a t
d e p e n d on t h e c o m b i n a t i o n in q u e s t i o n . E x p e r i e n c e h a s s h o w n t h a t corresp o n d i n g to each p o s s i b l e c o m b i n a t i o n , t h e r e seem to e x i s t a set of <,f a i r o d d s ~>,
such t h a t if t h e s e o d d s a r e offered, t h e n in t h e lon~z r u n t h e g a m b l e r will n e i t h e r
win n o r lose s y s t e m a t i e a l l y .
N o w t h e p r o b l e m s t h a t was a t t a c k e d b y t h e e a r l i e s t m a t h e m a t i c i a n s (*)
who t u r n e d t h e i r a t t e n t i o n to t h i s q u e s t i o n w a s to find a t h e o r e t i c a l w a y of
c a l c u l a t i n g w h a t t h e s e ~,f a i r o d d s ,~ s h o u l d be. I n t h e case of t h r o w s of a die,
for e x a m p l e , t h i s p r o b l e m was s o l v e d b y supposin~z t h a t a l l six f a c e s of e a c h
die a r e ~, e q u a l l y l i k e l y >> in e a c h tl~row. T h u s , t h e p r o b a b i l i t y t h a t a z i v e n
d i e will c o m e o u t a five is 1/6, an.d since t h e dice a r e <,i n d e p e n d e n t ,>, t h e p r o b a b i l i t y t h a t b o t h will c o m e o u t fives is I / 6 " z l / 6 - - . 1/36. H e n c e , t h e ~, f a i r
o d d s ,> a r e 36 to 1 in t h i s ease.
A l t h o u g h t h e m e t h o d of s o l u t i o n of t h e p r o b l e m i n d i c a t e d a b o v e c e r t a i n l y
w o r k e d in c o n n e c t i o n w i t h fz'ames of c h a n e e , it i n v o l v e d t h e i n t r o d u c t i o n of
t h e r a t h e r v a g u e n o t i o n of e q u a l <,likelihood >> or<< e , t n i p r o b a b i l i t y ,> of t h e
v a r i o u s p o s s i b l e r e s u l t s of a t h r o w . This n o t i o n c o n t a i n s a m i x t u r e of t w o

(*) The earliest mathematical discussions of probability seem to have been g~ven
b y PASCAL, FER~,IAT, HUYGHENS, BERNOULLI a n d I,APLACIL

TH]E G E N E R A L S T A T I S T I C A L P R O B L E M IX P H Y S I C S AND T H E T H E O R Y OF P R O B A B I L I T Y

1009

v e r y different interpretations of probability, which we m a y call respectively


the (~subjective ~ a n d the <~objective ~>. F o r the sake of clarity of presentation,
we shall give here, however, the pure forms of these interpretations, which
developed only some t i m e a f t e r the concept of p r o b a b i l i t y first b e g a n to "be
used systematically.
I n the subjective i n t e r p r e t a t i o n of probability, it is supposed t h a t probabilities represent, in some sense, incomplete knowledge or i n f o r m a t i o n concerning the events, objects, or conditions under discussion. Thus, in the g a m e
of dice, we h a v e no w a y of knowing with c e r t a i n t y before the dice are t h r o w n
w h a t the results of each individual throw will be (since these results are de'
t e r m i n e d b y the initial positions a n d velocities of the various p a r t s of each
die which are not accessible to us in practice). Hence, if the dice are sym~
m e t r i c a l l y constructed, we h a v e no knowledge favoring the suggestion t h a t
we will obtain a n y one side instead of another, a n d we therefore assign equal
probabilities to each side. I n this point of view, then, p r o b a b i l i t y is regarded
as something t h a t measures or reflects our degree of knowledge or ignorance.
I t is therefore essentially a description of a state of mind, a n d as a result, a n
essentially subjective category, since it would cease to be necessary, or even
to h a v e a n y meaning, if we could obtain precise knowledge concerning the
initial motions of the dice in each throw.
I t is clear, however, t h a t the i n t e r p r e t a t i o n of p r o b a b i l i t y as nothing more
t h a n a description of our m e n t a l reflexes under conditions of incomplete knowledge is not a d e q u a t e in t y p i c a l statistical problems, such as those a r i s i n g in
applications in science a n d in other fields. F o r such an i n t e r p r e t a t i o n gives
us no notion a t all of why in statistical aggregates cbntaining m a n y m e m b e r s ,
the a c t u a l relative ]requency is generally close to numerical value of the prob a b i l i t y (*). The mere f a c t t h a t we do not k n o w a n y reasons concerning either
the s t r u c t u r e of the die or its initial motion t h a t would f a v o r one face
of a die over a n o t h e r evidently does not b y itself necessarily i m p l y approxi m a t e l y equal relative frequencies for each face. Thus, a m o n g the things t h a t
we do not k n o w a b o u t the initial motions of the die, there could conceivable
exist hidden factors tending to f a v o r one result over another, even if we k n o w
t h a t the die is well-balanced in its construction. Such tendencies could evide n t l y exist w h e t h e r we k n e w of t h e m or not; a n d indeed it is just the nonexistence of these tendencies t h a t is necessary for the characteristic regular

(*) The importance of the relative frequency in connection with probability was
first recognized by ELLIS (On the Foundations o] the Theory o] Probabilities, in Trans.
o] Ca~b. Phil. Soc., 8, 1-6 (1844)), (paper read in 1842), and then by CouR~-o'r ( ~ p o sition de la tbdorie des chances et des p~vbabiltd s, Paris, 1843). However, neither of
these authors fully eliminated the subjective point of view, VEx~ (~l'hc Logic o/Chance,
London and New York, 1866) advocated systematically a pure frequency interpretation.

lo10

1). B o i i ; i

alld

w . SCllliTZI.;P,

b e h a v i o r of t h e r e l a t i v e f r e q u e n o y in t h e long runt~'. T h u s , we h a v e ~o ,lo


h e r e with a n objective p r o p e r t y of t h e die a n d of t h e p r o c e s s b y w h i c h i t is
thro~vn (*). which h a s no e s s e n t i a l relationshil~ to t h e d e g r e e of k n o w l e d g e
or i g n o r a n c e of t h i s process t h a t m a y e x i s t in o u r m i n d s . [ n d e e d , e v e n if w e
s o m e h m v were a b l e to o b t a i n perfect k n o w l e d g e of t h e i n i t i a l c o n d i t i o n s in
each ease, t h i s would n o t c h a n g e t h e f a c t t h a t in a t y p i e a l long series of t h r o w s .
t h e p r o b a b i l i t i e s w o u l d still p r o v i d e a g o o d a p p r o x i m a t i o n to t h e r e l a t i v e freq u e n c i e s predicted ,m t h e b a s i s of such p e r f e e t knowledg'e, ( p r o v i d e d of c o u r s e
I h a l t h e die is ~vell-tmla~wed a n d t h v . w n b v a p r o c e s s t h a t l e a v e s r o o m for a
sufficient deg'ree of , ' h a n e e l l u e t n a t i o n in t h e i n i t i a l c o n d i t i o n s ) . T h u s , t h e
p r o b a b i l i t i e s a r e o b j e v l i v e p r o p e r t i e s of l h e ,tie a n d of t h e p r o c e s s b y w h i c h
it is t h r o w n , which m a n i f e s t s itself a p p r o x i m a t e l y as a r e l a t i v e f r e q u e n c y in
a l,mg series ,ff thro~vs, l l m v e v e r , t h e r e l a t i v e f r e q u e n c y does n o t a p p r o a o h
uniformly, h u t l t t t e t u a t e s w i t h i n b o t t n d s t h a t lettd to (.onle closer a n d closer
to ea4.}l o t b o r as tht' ll/llltber of terl)ls illoroas{.s.
I t is clear, t h a t t h e o b j e c t i v e i n t e r p r e t a t i o n of p r o b a b i l i t y d e s m ' i b e d a b o w ,
is t h e one t h a t is a p p r o t n q a t e in st..qtistieM pr,~blems such as t h o s e t r e a t e d in
seiem'e a n d in o t h e r fields. IIf ,'om'se. t h e w o r d a p r o b a b i l i t y ~>, as c o m m o n l y
u.se,l, can also be g'iwm a subjective, meaning" in lertn8 of o u r j u d g ' e m e n t s rev ' a r d i n g t h e l i k e l i h o o d t h a t i n f e r e n c e s b a s e d on i n c o m p l e t e in~ormq.tion a r e
vali, I. N e v e r t h e l e s s . as we h a v e seen. t h e p r e d i e t ~ b l e reg'ular t r e n d s in t h e
r e l a t i v e f r e q u e n c i e s a r e p r e s e n t in s t a t i s t i t . a l ag'o'reg'ates of o b j e c t s or e v e n t s
d i s t r i l m l e d b y c h a n c e processes, w h e t h e r we m a k e inferen<.es a b o u t t h e m or
n o t : so l h a t t h e i n t r o d m . l i , , n of our d e g r e e of knowledg'e or i~'noranee is q u i t e
i r r e l e v a n t to t h e pt'ol}lem lllldev discussion.
W i t h reg'ar,l l o t h e role of v h a m ' e in p r o v i d i n g t h e g e n e r a l c o n d i t i o n s or
lm,'kgl'oUnd in x~hivh t h e v,m,'ept of p r o b a b i l i t y e a n e o r r e e t l y b e a p p l i e d , t h i s
follows inlnie<tiately f r o m t h e disvussion of l h e origiu of st:ttisti(.al l a w g i v e n
in See. 2. Thus. in t h e ,'ase ,,f t h e t h r o w s of a ,tie. we h a v e all (if tit(: (.hara(.teristie fa,'t(,rs tendin<.,' lo f a v o r t h e <.reati(m (,f (.han('e f l u e t u a t i o n s . F i r s t of all,
t h e r e is t h e r e l a t i v e l y g'real sensitivit.v of t h e die to t h e p r e c i s e (.ha r a ( ' l e r of
the i n i t i a l m o t i o n s with ~lii(.h il is t h r o w n . If t h e die is l h r o w n front a n
a p p r e , . i a M e heig'ht, so t h a t it h a s l i m e to t u r n one m' m o r e t i m e s b e f o r e i t
fails, t h e n a s n l a l l e h a n g e of its i n i t i a l c o n d i t i o n s c o u l d ehang'e t h e tinal r e s u l t
v o m p l e t e l y f r o m a n y ,me face to a n y other. M o r e o v e r , t h e h u m a . n b o d y is
e v i d e n t l y a v e r y eoml,tex s y s t e m , c<mlaining a g'real m a n y fluetuatino" elem e n t s ,ff COmlmrable o r d e r s of niag'nitude. T h u s , it ix quit, e n a t u r a l t h a t t h e s e
t l u e t u a t i o n s a r e stlcli t h a t in t h e hm~ r u n . a n y t a e e of t h e die c a n a p p e a r .

l*) I]oth ~',~CJ:XOT and \h,:N~ recognized t;hi~ fact explicitly. Indeed, the exposilimt in this se(.tiou follows closely atone" the ~eneral lines laid dowtt by {'OV,Rxo'r.

q ' n E G ] ~ N E R A L S T A T I S T I C A L P R O B L E M I N P H Y S I C S A N D T H E T H E O R Y OF P R O B A B I L I T Y

1011

And considering the effective independence of the behavior of these elements


relative to the physical state of the die, it is hardly surprising t h a t in the long
r u n and in the average, these fluctuations favor no particular face, and lead
t o a p p r o x i m a t e l y equal relative frequencies for each. We see t h e n in detMl
how tile p r o p e r t y of probability is a reflection of various aspects of the entire
process b y which u distribution of chance events comes into being. Thus,
t h e probability is a p r o p e r t y of this process as a whole, which c~n be given
meaning even prior to the realization of any particular sequence as the
p r o p e r t y of the process as a whole from which the predictable trends in the
relative frequencies come.
Finally, a word of caution must be given with regard to the use of the
theory of probability. F o r it must be r e m e m b e r e d t h a t the t h e o r y is valid
only to the e x t e n t t h a t the distrib i l i o n of objects or events has characteristics
similar to those t h a t would be b r o u g h t about b y the operation of a set of
genuine chance fluctuations. Such fluctuations depend, for their characteristic
properties, however, on the independence of the external a n d contiagent causes
relative to the events under discussion. Now in reality, because everything
in the universe is interconnected with everything else to some degree, perfect
independence cannot exist. Nevertheless, in m a n y cases, such as t h a t of throws
of a die, the degree of intereonneetion is so slight, t h a t the concept of pure
chance provides a good approximation. Where the independence is incomplete,
however, this concept will a p p l y only to a limited degree of approximation;
and then, m a n y features of the calculus of probability (e.g. the multiplication
of the probabilities of two events to give the probability t h a t t h e y shall b o t h
occur) will also apply only approximately. I n such cases, we do not have
pure chance, so t h a t a more precise representation of the statistical laws would
in general h a v e to go outside the framework of the t h e o r y of probability (*).

4. -

The Irregular Collectives

of y o n M i s e s .

The objective interpretation of the t h e o r y of probability was given a new


and more precisely defined m a t h e m a t i c a l formulation b y v o ~ MISES (1). Since
this interpretation raised the interesting problem of randomness, and moreover h a d an i m p o r t a n t influence on modern thinking on the subj'ect, in other
respects, it will be worth our while to discuss and analyze this point of view
briefly.
The basic idea of vo~" MISEs was to introduce the notion of the irregular
(*) We shall discuss this problem in more detail in Section 9.
(1) See, for example, R. vo~" M~SES: Wahrscheinlichkeit, Statistik ~nd Wahrheit,
dritte Aufl. (~'Vien, 1951).

1()]')

1). B()II'r

alld

vc. S C t l i ~ T Z E R

colh,ctic~, (<)r s t a t i s t i c a l e n s e m b l e ) ~s a m o d e l for a hmR' series of c h a n c e ew, n t s ,


or for q ( l i s t r i b u t i o n (.o||tainin;" m a n y objeets with p r o p e r t i e s t h a t are s u b j e c t
1o c h a n c e v a r i a t i o n s .

This collective was R'iven the followin,~ t h r e e e s s e n t i a l

(.ha ra eterist ies :


1) It was assmne(t to h a v e a n m l t i m i t e d n u m b e r of m e m b e r s .
2)

In the eolle(.tive, the r e l a t i v e frequen('y of e v e r y t y p e of o b j e c t o r

e v e n t was a s s m n e d to al)proaeh a definite n u m b e r , which was


the l)robability of t h a t tyl)e of ot)jeet or e v e n t (i.e., the

ide~ttified with

o,ly m e a n i n g ' of pro-

1)abilily was a s s m n e d to be in t e r m s of this r e l a l i v e f r e q u e n c y ) .


3) The d i s t r i b u t i o n of ol)jeets or e v e n t s was a s s u m e d to h a v e a c e r t a i n
prol)erty of

(, i ' a l l d O l l l l l e S s

~) e l '

<~ lawlessness

,.

To u n d e r s t a n d the suR'~eslions of v e x .~ltsEs with re~'ard to t h e c o n c e p t


of r a n d o m n e s s , it is c o n v e n i e n t to (.onsider the 1)rol)lem of stotistical .~,~pling.
N()~, it is well k n o w n t h a t in typi(.al (tistril)utions of (.hanee e v e n t s , one c a n
d r a w p a r t i a l s a m p l e s b y v a r i o u s m e t h o d s , an(1 t h a t the r e l a t i v e f r e q u e n c i e s
of v ' u ' i o u s (,lasses of e v e n t s in the s a m p l e s l)rovides a g'ood a p p | . o x i m a t i o n to
lhose in the s e q u e n c e as a whole. The sugg'estion of vo.x MiSES is efl'eetively
l h e n t h a t lhe 1)ossibility of d r a w i n g s a m p l e s of lifts k i n d be used as a dellh i l t o n of r a n d o m n e s s . More precisely, his sugg'estions a m o u n t e d to the idea
t h a t in a r a n d o m d i s l r i b u t i o n , sul)sequen(.es (lrawn o u t of the collective in
s u i t a b l e ways s h o u l d h a v e the s a m e limiting" r e l a t i v e frequen(,ies for each t y p e
of e v e n t as are l)resent in the m a i n s e q u e n c e (*).

1,2vi(lently, i n a v e r y regula,"

seque||ee, this would not be so. T h u s , if i n s l e a d of throwing" a die. we placed


it on t h e t a b l e in the s e q u e n c e 1 . 2 . 3 , 4, 5, 6, 1.2, 3, 4, 5, 6,..., etc., the limiting'
r e l a t i v e f r e q u e n ( ' y for each possible result ~vouhl c l e a r l y be 1/6.

Nevertheless,

if one s a m p l e d e v e r y sixth t h r o w , one would a l w a y s g'et the s a m e r e s u l t , so


l h a t the r e l a t i v e f r e q u e n c y of t h a i result would be u n i t y , a n d

nol o n e - s i x t h ,

as it is in the s e q u e n c e as a whole. T h u s , re~'ular s e q u e n c e s of this t y p e fail


lo possess e e r l a i n i m p o r t a n t p r o p e r t i e s wilh r e z a r d to sampling" t h a t e x p e r i e n e e

(*) Actually, ~(~x- MlsJ,:~ developed his delinition ori~imdly il~ cmmeetiou with lhe
proldcm of wim~illg a -ambling g'ame by t,eltin~ on ~ubsequences of plays ehoseli front
the main sequence l)y some predetermined kind of rule or ,, systeIn ,,. The fact tha,t in
real /ames, such efforis to beat the ~ame, have always failed is then taken to
demonstrate ~he randomness of the sequence of plays. It is evident, however, that
the possibility of Iindina" a , winninR" system ,, in a g'ambling game depends on whether
or nol ,~amples drawn from the entire sequence of plays will have the same distribution
of relati~ e frequencie,, a~ are present in the sequence as a whole. Thus, the fonnulatio~t
of the definition of randomness in terms el the possibility of drawing representative
salli|)|(~y; t ' O l ! l a i l l s lilt* ffamblinR' R ' a l l i e a ~ a speeial e a s e .

T H E G E N E R A L S T A T I S T I C A L P R O B L E M I N P H Y S I C S AND T I I E T H E O R Y OF P R O B A B I L I T Y

]013

has shown to be essential for a r a n d o m series of events; a n d it is for this re~son


t h a t v o ~ MISES proposes the identification of the p r o p e r t y of r a n d o m n e s s
with irregularity, or (( lawlessness )~.
The infinite collective of vo M~s~s is essentially a simplified model for
typical finite b u t long sequences of chance events. I n other words, it is supposed t h a t if one replaces the finite collective b y the infinite model, t h e n conclusions d r a w n b y using the model will generally a p p l y a p p r o x i m a t e l y to the
real finite collective. The infinite collective is evidently a n abstraction, which
cannot exist in reality. B u t it is a s s u m e d b y yon MISES to be an a b s t r a c t i o n
which enables h i m to represent in a simplified or (( economical )) f o r m t h e
essential relationships t h a t exist in a sufficiently large n u m b e r of chance events.
I n this regard he compares his idealized a b s t r a c t collectives to the i n t r o d u c t i o n
of the concept of points in g e o m e t r y , which he likewise regarded as a b s t r a c t i o n s
p e r m i t t i n g a simplified or economical r e p r e s e n t a t i o n of the essential aspects
of the geometrical relationships of real objects t h a t are v e r y small.
The f o r m u l a t i o n of p r o b a b i l i t y given b y v o ~ MrSES can be criticized on
several grounds. F i r s t of all, one can raise the question as to w h e t h e r it i*
a d e q u a t e to identify p r o b a b i l i t y completely as v o ~ MIsEs does, with a l i m i t i n g
relative fequency in the idealized collective. F o r f r o m the f a c t t h a t the p r o bability m~kes possible approxim-~te predictions of f u t u r e relative frequencies,
one would be led to conclude t h a t there should exist some p r o p e r t y corresponding to the probability, even be/ore the re~l collective has come into being.
F o r example, let us consider a g a m e of c~rds in which the rel_~tive f r e q u e n c y
with which a given card will be o b t a i n e d in a long series of plays is d e t e r m i n e d
a p p r o x i m a t e l y in t e r m s of a suitable probability. The question t h e n i m m e diately arises (( F r o m w h a t properties existing before the cards are d r a w n doe*
the a p p r o x i m a t e p r e d i c t a b i h t y of the relative frequency come? ,7. Surely, it
c a n n o t come f r o m the fact t h a t the plays of the ca.rds t h a t are potentially capable
of being m a d e belong to an ideahzcd b u t non-existent collective, which at
best m a y be reahzed a p p r o x i m a t e l y only in the future. Hence, the probabilities otlght no~ to be identified with nothing else but, relative frequencies in
the idealized ensemble, if only because such an identification implies t h a t the
a p p r o x i m a t e predictability of relative frequencies in real ensembles t h a t a r e
going to be p r o d u c e d in the future comes out of nothing at all existing in the
present. Such an implication is evidently unsatisfactory, since it does not
t a k e into account the f a c t t h a t definite conditions in the present are a c t u a l l y
needed to lead to predictable relative frequencies in the future. Thus, in the
ease of the card plays, the probabilities will give the right answers only if the
cards are similar in structure a n d if t h e y are a d e q u a t e l y shuffled b y some
chance process. Hence, it is evident t h a t the a p p r o x i m a t e predictability of
the relative frequencies comes out of something else existing a t present. One
is therefore led to conclude t h a t the eorrcept of p r o b a b i l i t y needs a b r o a d e r

]0]4

D, IIOI[M ~tlld v(. S('IIi:TZEI~

b a s e t h a n c a n be o b t a i n e d by i d e n t i f y i n K it with nothin~2' else b u t a rel'atiw*


f r e q u e n c y in a n i d e a l i z e d e n s e m b l e .
In l h i s r e g a r d , t h e 1)oint of view of vo:< M/SES is a s t e p b a c k w a r d s r e l a t i v e
to t h a t of C o u r n o t , w h o h a d a l r e a d y m a d e it ('lear t h a t e v e n b e f o r e a n ag'gre'_,'ate of o b j e c t s or e v e n t s is r e M i z e d , t h e p r o b a b i l i t y refleets v a r i o u s a s p e c t s
of t h e o b j e c t s a n d t h e p r o c e s s e s f r o m which t h e d i s t r i b u t i o n will e v e n t u a l l y
t'Ollle.

T h e s e c o n d i m p o r t a n t c r i t i c i s m of t h e w m Mises t h e o r y of p r o b a b i l i t y is
c o n n e c t e d with t h e p r o p o s e d c r i t e r i o n for r a n d o m n e s s . T h e e s s e n t i a l p r o b h ' m
here is t o define lnoFe p r e c i s e l y w h a t is m e a n t b y <~i r r e g ' u l a r i t y ,> or <<lawlesshess ,>. T h u s . \ o N MtSFS a l r e a d y r e a l i z e d t h a t he cmfid n o t r e q u i r e t h a t t h e
r e l a t i v e f r e q u e n c y in ccery ims.~ibb, s u b s e q u e n e e b e t h e s a m e as i n t h e m a i n
s e q u e n c e , since one could, for e x a m p l e in t h e c~se of t h r o w s of a coin, a l w a y s
choose t h a t s u b s e q u e n e e in whieh t h e r e s u l t s c a m e o u t h e a d s , so t h a i t h e
r e l a t i v e frequen<'y of h e a d s w o u h l be u n i t y , a n d n o t o n e - h a l f , as i t is for t h e
sequen<.e as a whole, l i e d e a l t w i t h l h i s p r o b l e m 1)3- requiring', t h a t t h e sul>s e q u e n c e be ('hosen b y a metho(1 of sampling" t h a t is defined belore all t h e
results of a ,,,'iven sequen<.e a r e a v a i l a b l e : or m o r e 1)reeisely, t h a t t h e s u b s~*quenee be chosen by a m e t h o d t h a t d e p e n d s o n l y on t h e posit.ion of a g i v e n
e w , n t in t h e s e q u e n c e a n d on t h e l>roperties of t h e e v e n t s b e f o r e it. E v e n so,
h o w e v e r , t h e c r i t e r i o n is v e r y v a / a e , a n d e a n n o l be a p p l i e d in a n y l n ' a e t i e a l
p r o b l e m (e./. one e o u l d h a r d l y <.ounl t h e r e l a t i v e f r e q u e n c i e s in a l l p o s s i b h '
s u h s e q u e n c e s fittino/ i n t o t h e a b o v e c r i l e r i o n ) . M o r e o v e r , it is b y no m e a n s
vleav t h a i such a c r i t e r i o n is s e l f - c o n s i s t e n t , since it h a r d l y s e e m s l i k e l y t h a i
,',','r?l m e t h o d of sampling' fallin~z w i t h i n its scope will give t h e s a m e r e l a t i v e
f r e q | | e n e i e s as in t h e m a i n s e q u e n c e .
T h e lwoblem of d e f i n i n / w h a t one m e a n s b y <<irrela'ularity >> or <<lawlesshess ,> is a p a r t i c u l a r l y t h o r n y one. since in t h e v e r y a c t of d e f i n i n g it, one is
i n e v i t a b l y as<.ril>ing to it s o m e k i n d of reg'ularity a n d law. I n d e e d , v a r i o u s
m e m b e r s (*) of t h e school of \(>x 31[SES s e e m to h a v e Iziven u p t h e a t t e m p t
to d e l i n e r a n d o n m e s s as lqwlessness. T a k i n / as a s t a r t i n / p o i n t a s e q u e n c e
of p l a y s in a g'amblin,g" / a m e . t h e y p r o p o s e to define r a n d o m n e s s in t e r m s of
a specitie a n d well-defined l w o p e r t y of such a s e q u e n e e ; viz, t h a t if a p l a y e r
s y s t e m a t i c a l l y follows a n y one of a c o u n t a b l e i n f i n i t y of rules of p l a y w h i e h
/ h e y detine, he <.an n e v e r <, h e a l ,, t h e g'ame. T h e s e efforts to d e l i n e r a n d o m n e s s
a r e in t h e i r t u r n , h o w e v e r , s u b j e e t to i m p o r t a n t c r i t i c i s m s (e). Aside f|'o|n
sharing' in t h e defect of t h e approa<.h of vo.\ 3[tSES t h a t it is n o t <,lear h o w
t h e y c o u l d be a p p l i e d in p r a c t i c e to a wide rang'e of p r o b l e m s , t h o s e c r i t e r i a

t:::) .~. \VAIA). A .

H.

(.~OPEL,,NI~, e{3C..

/ e) J. \'tL1.E: F3tude Critiqzu, tte [a Notion de ('ollectif (Paris, 193.9).

T H E G E N L R A L S T A T I S T I C A L P R O B L E M LN P H Y S I C S A N D T H E T H E O R Y OF P R O B A B I L I T Y

1015

for r a n d o m n e s s involve a simple transference to all possible statistical problems


of a m e t h o d t h a t m a y p e r h a p s be a p p r o p r i a t e in the special p r o b l e m of g a m b ling g a m e s (*). Finally, there has not y e t b e e n given a d e m o n s t r a t i o n t h a t
these criteria c a n consistently be applied in all of t h e domains where the t h e o r y
of p r o b a b i l i t y has been used.

5. -

The

Modern

Axiomatic

School.

The m o d e r n a x i o m a t i c school of p r o b a b i l i t y developed b y KOL)rOGO~OFF (3)


a n d others is based on essentially the same general a s s u m p t i o n s concerning
chance a n d statistical law t h a t we h a v e discussed in Secs. 2 a n d 3. I t s essential
new characteristic is its stress o~ t a k i n g the laws of the t h e o r y of p r o b a b i l i t y
as m a t h e m a t i c a l abstractions, which c~n be studied s e p a r a t e l y f r o m the question of the p r o b l e m of their i n t e r p r e t a t i o n , a n d of their applicability to real
problems.
I n this a p p r o a c h , the probabilities are a s s u m e d without f u r t h e r analysis
as (<primitive)> m a t h e m a t i c a l concepts, while the relationships b e t w e e n t h e m
are t h e n t a k e n as axioms. The purpose of the m a t h e m a t i c a l theory of probability is t h e n r e g a r d e d as being the deduction of the logical consequences of
these a x i o m s in progressively g r e a t e r detail. This procedure is c o m p a r e d with
t h a t used in g e o m e t r y , where one likewise introduces certain a b s t r a c t entities
(e.g., points a n d lines) as p r i m i t i v e concepts not subject to f u r t h e r analysis,
a n d where the function of the t h e o r y is likewise conceived of ~s the deduction
in ever greater detail of the logical consequences following f r o m a certain
a s s u m e d basic set of axioms, b y which it is a s s u m e d t h a t the p r i m i t i v e concepts are related.
The principal a d v a n t a g e s of the a x i o m a t i c point of view is t h e n t h a t
w i t h o u t identifying p r o b a b i l i t y p r e m a t u r e l y a n d excessively n a r r o w l y with
a n y specific p r o p e r t y , such as relative f r e q u e n c y either in a real or in a n idealized ensemble, it enables one to o b t a i n a clear a n d precise f o r m u l a t i o n of the
purely m a t h e m a t i c a l p a r t of the theory.
Certainly if one simply considers such a process of a b s t r a c t i o n in itself,
no criticism w h a t e v e r could possibly be m a d e a g a i n s t c~rrying it out within
the d o m a i n of pure m a t h e m a t i c s , provided t h a t it is clearly understood t h a t

(*) As we shall see in Section 9, there are strong reasons to suppose that the criteria
for randomness may have to possess a nmeh wider range of adjustability to the specific
characteristics of the various statistical problems to which they are applied, than is
possessed by the criteria that we arc discussing here.
(a) A. KOLMOGOROFF:Grundbegri]je der Wahvsc]~einlichl~eitsrechnang, in Erg. d. Math.,
2, m 3.

lO16

i).

B~IIlM itli(|

W. ~CII[~'I'ZEII

perhaps under other conditions alld ill. connection witll new kinds of statistieaI
problems, new kinds of m a t h e m a t i e a l abstractions m a y be needed (Just as
the abstractions of Euclidean ~'eometry m a y have to be replaced b y the more
~'eneral ones of non-Euclidean ~'eometry, to deal with astronomical d o m a i n s
of space), llox~ever tile l)osition taken by m a n y members of the modern
axiomatic school seems to imply that the calculus of probability already represents the most ~'eneral possible 1)urely m a t h e m a t i c a l means of treating'
statistical problems. Such a point of view tends to confine tile horizo,| of
our thinking about statistical problems to nothing' move t h a n the calculation
of one kind of probability in terms of another kind. On the other hand, as
we shall see in the later sections of this paper there exists a wide variety of
aetnal eases in which arise nlu(,h more ~'eneval statistical problems t h a t can
a.dvanlag'eously be treated by methods going' beyond those of the theory of
probability and that in m a n y cases must be treated by such methods. Thus.
it is our opinion t h a t while tile era'rent theory of probability m a y consistently
be re~'a|,ded as a self-eontained logical diseipli|w, it m a y also be studied both
m a t h e m a t i c a l l y and in more ~'eneral ways, within tile context of a broader
point of view. in which the meaning" of ils basic concepts becomes clearer, as
one comes to see their setti||~' in the ~'eneral domain of statistical problems,
as well as tile processes by which tile laws of probability <,ome to be valid,
and the factors limiting" the domains of validity of these laws.

6.

- On t h e

General

Statistical

Problem

as It A r i s e s in S t a t i s t i c a l

Mechanics.

Thus far, we have discussed tile t r e a t m e n t of statistical problems in terms


of the coneept of probability. We shall now ~ee that in the domain of statistical meehanies, a much more ~'eneral notion of what constitutes a statistical
problem is not only advanta~'eous, but in fa(,t, in m a n y problems, absolutely
necessary.
In order to help bring' out our point, we first review a few elementary ideas
of statistical mechanics, Now, as is well known, the basle object of statistieM
mechanics is to prediet various macroseopi(' properties of a system composed
of a very larg'e llllIllbel' of moleeules (of the order of 1.02a), on the basis of suilable hypotheses ~'(meernin~" the motions of its constituent molecules. To
describe the maeroscopie properties, we m a y divide the six dimensional space
having" (x. p) as coordinates (which was called :z space 1)3" EmtEXFEST) (~) into
cells that are so small t h a t x and p do not chan~'e appreeiably within them.
Now if we consider a systenl composed of m a n y molecules, the macroscopic

(a) P. and T. EHRr:NFI.:S'r: Etmyklop. der Math. Wi.s's., 4, 4 (Leipzig, 1907-19141.

T I l E G E N E R A L S T A T I S T I C A L P R O B L E M I N P H Y S I C S A N D TH)] T H E O R Y OF P R O B A B I L I T Y

i017

properties of this s y s t e m can be specified b y giving the n u m b e r of molecules N~,


in each cell. Such a specification involves giving a series of n u m b e r s (n~, n2,...
n~,..., n,) -~ [n~]. I n typical problems, the t o t a l n u m b e r of molecules in t h e
s y s t e m is so g r e a t tha~ the [~r] ~re all lairly large, e v e n w h e n the cells are
chosen small relative to macroscopic orders of m a g n i t u d e .
On the other hand, the basic laws of mechanics work in t e r m s of a microscopic speci]ication of the s t a t e of the system. This is done b y specifying a
point in the 6n dimensional p h a s e space of the s y s t e m as a whole (called the
F space b y EHP~ENFEST).
An i m p o r t a n t question t h a t we now consider is t h a t of comt3uting the
votmne of F space corresponding to a certain set of the [n~]. As is well known,
this p r o b l e m is of crucial i m p o r t a n c e in calculating the statistical properties
of the s y s t e m (~). This volume is
n~
(6.1)

Q [n~]

nl ! n~ ------T.
! ... n~ A ~ '

where A~2 is the Volume in phase space of the cell corresponding to a n y one
a.rr~ngement leading to the s a m e [nr].
W e now consider, as is usually done in stutisticul mechanics, a region of
phase space lying b e t w e e n two hypersurfaces of c o n s t a n t energy, E, a n d E + A E ,
where A E is quite smull c o m p a r e d with E. As is well known, a n isoluted
s y s t e m moves along a surface of c o n s t a n t energy, so t h a t if we s t u d y t h e properties of such a shell of thickness A E a n d let A E a p p r o a c h zero, we shall
o b t a i n a n a c c u r a t e r e p r e s e n t a t i o n of the properties of the h y p e r s u r f a c e on
which such a s y s t e m moves.
The n e x t p r o b l e m is to o b t a i n a n a p p r o x i m a t i o n to t h e expression (6.1),
which is valid when the n, ~re large. This is conveniently done in t e r m s of
the expression for In/2In,]. W e use Stirling's a p p r o x i m a t i o n for the n,, obtuining
(6.2)

In Q[nr] ~ n in n + In A~2 -- ~ n i In ni .
i

~Now~ Q[nr] has a v e r y steep m a x i m u m . Thus, as we shall see presently


a good a p p r o x i m a t i o n to this function can be o b t a i n e d b y e v a l u a t i n g the set
of values of [n~] for which ~[n~] is a m a x i m u m , a n d t h e n b y e x p a n d i n g
In Q[n~] as a series of powers of ~a~ ~ n~--n~0, where n~o is the v a l u e of n~
for which Q[n~] is a m a x i m u m .

(~) This problem has been treated in various ways by many authors. See, for
example, P. and T. E r m ~ S T :
Encyelopedie des Sciences Mathdmatiques, 4, 2, Sup~pl6ment II.

I{tlS

I).

ll(fll'~l

~l)ld

W.

SCII{:TZItl

To o b t a i n tire m a x i m u m of In _(2[;~]. s u b j e e t to t h e e o n d i t i o n followin~z~fronl

the

i s o l a t i o n of t h e

system,

that

the

total

ener~'y is a

constant.

number

of

particles is a

constant

(~,,E,--E

= 0) a n d

(~ ,, -- ~

()) we use the well k n o \ v n m e t h o d of La~'ran~'e's p a r a m e t e r s , i.e..

~,, ~-)'

(6.3)

that

the

h, ~-2 - - :.( 2

total

,,, - - ,, ) - - : , ( 2 , , , = ,

- - ~:).

T h e (.on(liti(m for" a mc~xim|||n is t h e n


F

(6.1)

2,~,

""

1)I'

((;.5)

,;, -

;;,,

,1 ex 1) 1-- [,/',', I 9

As is well k n o \ v n from eal~.u!ations carried out in s t a t i s t i c a l m e e h a n i e s u comes


out e q u a l t~) 1 1:7", where 1,. is B o l t z n m n n ' s c o n s t a n t , a n d T tile t e m p e r a t u r e .
~ h i h , ,1 is a n(n'|nalizing" o o n s t a n t ohosen to yield ~ i ~ , - ,

=-0.

For t h e ease of mole(.ules with ne~'li~ibh, inte|.a0tion~, to which we shall


rest~'i('t the diseussior~, the ener~'y is j u s t tire k i n e t i c energy.
tim! we o b t a i n

(i~.i;)

.,

~1 e x p l - -

.~v2,'2/,T],

E, = 89

,o

This is t h e w e l l - k n m ~ n .~lax~vell distril)t~tion of velocities.


W e see t h e n t h a t the .~laxwell d i s t r i b u t i o n uorrespollds to the l a r g e s t l~)~sihle VOJL/llle Of [" apace. \Vi, shall now be i n t e r e s t e d , h o w e v e r , i n t h e v o l u m e
of 1" space (.orreslrondin=" to s m a l l d e v i a t i o n s f r o m the M a x w e l l distributiot~.
To ~'ah'ulate these v o l u m e s , we e x p a n d eq. (6.3) up ~o se~:ond p o ~ e r s in
atr,

--

~,

--

~,,, ol)tainin~"

(t;..)

(|;.s)

_re'= -gl exp [ - ~ ~ ,,.(,)/,)q.


t

x~here r~/, ~ 6~,/~,o is the fr::(.tional d e v i a t i o n froll? a Maxwell distribut.iu~.


W e see t h e n t h a t d e v i a t i o n s f r o m a 5 l a x w e l l i a n d i s t r i b u t i o n h a v e a g a u s s i a n
de~lsity ill p h a s e space. I n d e e d . b e c a u s e ~,o is larva'e, t h e v o l u m e corresponding"
to a p p r e c i a b l e f r a c t i o n a l d e v i a t i o n s is, i n g e n e r a l , v e r y small, h i o t h e r w o r d s .

T H E GENJERAL S T A T I S T I C A L P R O B L E M I N

P H Y S I C S A N D T H E T H E O R Y OE P R O B A B I L I T Y

1019

the m a x i m u m of f2[n,] is a v e r y steep one (*). Indeed, it is so steep t h a t one


m a y express the result more pictorially b y saying t h a t a surface of c o n s t a n t
energy corresponds to a Maxwellian ocean in which there are ~., few v e r y t i n y
non-Maxwellian islands (+). Now~ the usual t r e a t m e n t in statistical mechanics
has been to assume t h a t e v e r y point on the shell of p h a s e space, in which tile
energy has a value b e t w e e n E a n d E ~ - A E is equally probable. Such an ass u m p t i o n would evidently lead to the correct conclusion t h a t the s y s t e m
practically always has a distribution v e r y close to the Maxwellian; for the v e r y
small volume of p h a s e space corresponding to the non-Maxwelllian distribution
would t h e n practically never be entered b y the p h a s e point.
The question of the justification of the hypothesis of equiprobability on
a shell of constant energy is discussed in t e x t s on sta.tistical mechanics (s).~
This question is, however not p r i m a r i l y w h a t interests us in this paper. Hence,
we shall here content ourselves with the s t a t e m e n t t h a t such a justification
usually involves a t t e m p t s to p r o v e either a quasi-ergodic t h e o r e m or metricindecomposability of the orbits in p h a s e space, a t t e m p t s which thus far h a v e
not been successful except for a few of the simplest t y p e s of systems (-). The
point t h a t interests us here, however, is t h a t the a t t e m p t to p r o v e equi-probability m a y h a v e little or no relevance to the p r o b l e m of d e m o n s t r a t i n g the
Maxwellian distribution. F o r because of the f a c t t h a t the non-Maxwellian
distributions correspond to such a m i n u t e v o l u m e of p h a s e space, it is clear
t h a t a l m o s t a n y continuous p r o b a b i l i t y distribution, uniform or otherwise,
would lead to a distribution v e r y close to a M a x w e l h a n distribution. B u t
more i m p o r t a n t even t h a n this is the f a c t t h a t a v e r y wide range of determinate laws governing an individual system would still give a n a p p r o x i m a t e l y
Maxwellian distribution, for practically all of the time, as long as t h e y did
not i m p l y an unusual t e n d e n c y for t h e orbit to r e m a i n n e a r the special nonMaxwellian regions. And since as we shall see in Sec. 7, we h a v e a v e r y large
n u m b e r of particles in unstable motion, it is quite plausible to expect t h a t
in systems of the k i n d with which we work in statistical mecharrics, t h e a c t u a l
orbit in phase space will t e n d to w~nder sufficiently t h r o u g h o u t the accessible
regions of p h a s e space, so t h a t it will not long r e m a i n in the non-Maxwellian
regions.

(*) A much more elegant and precise method is available for deducing this result.
See, for example, the proof of the central hmit theorem in A. I. KuI~Cm~: Statistical
Mechanics {New York, 1949). The method that we have chosen here is, however, based
on essentially the same ideas, and has the advantage of bringing out more clearly what
is important for our purposes.
(+) This fact has long been weli'known. See, for example, reference (b).
(6) See for example, A. I. K n i f e , I N : Statistical Mechanics (I~ew York, 1949).
(-) We shall discuss a few such cases in Section 9.

\Ve c o n c l u d e t h e n t h a t t h e effort to b a s e s t a t i s t i c a l m e e h a n i e s on equip r o b a b i l i t y in a g'iven region in p h a s e s p a c e is a n u n n e c e s s a r i l y n a r r o w w a y


of a t t a c k i n g t h e p r o b l e m . [n s o m e eases (if t h e r e is t i m e for t h e o r b i t to fill
such a re~ion to a g o o d d e g r e e of a p l ~ r o x i t n a t i o n ) , it c o u l d be b a s e d on e q u i p r o b a l ) i l i t y (*). B u t in most p r o b l e m s , t h e r e a l b a s i s s e e m s to b e t h e insensit i v i t y of l a r g e - s c a l e averatz'es in s y s t e m s c o n t a i n i n g m a n y c o m p o n e n t s to t h e
p r e c i s e d e t a i l s of t h e m o t i o n s of t h e s e c o m p o n e n t s . B e c n u s e of t h i s i n s e n s i t i v i l y , one m a y h a v e p r a c t i c a l l y c o n s t a n t a n d d e t e r m i n a t e s t a t i s t i c a l a v e r a g e s ,
w i t h o u t h a v i n g e q u i d i s t r i b u t i o n , a n d i n d e e d w i t h o u t h a v i n g a n y k i n d of p r o b a b i l i t y d i s t r i b u t i o n a t all (:). T h u s , t h e p r o l d e r n of f i n d i n g a.n a d e q u a t e
b a s i s for s t a t i s t i c a l m e c h a n i c s is g r e a t l y s i m p l i f i e d , b e c a u s e i t is e v i d e n t l y
m u c h e a s i e r to p r o w * t h a t t h e p h a s e p o i n t does n o t r e m a i n lone2 in t h e t i n y
n , m - M a x w e l l i a n regions t h a n it is to p r o v e t h a t it c o v e r s t h e whole s p a c e in
a q u a s i - e r g ' o d i e or m e t r i c a l l y i n d e e o m p o s a b l e way.
On t h e b a s i s of t h e a b o v e r e s u l t s , we a r e t h e n led to c o n s i d e r a n e w a n d
h r o a d e r p o i n t of view on s t a t i s t i c a l p r o b l e m s , which p e r m i t s us to go b e y o n d
t h e m e r e e a h . u l a t i o n of one k i n d of p r o b a b i l i t y in t e r m s of a n o t h e r k i n d .
l u s t , c a d , we m a y c o n s i d e r w h a l has b e e n c a l l e d in See. 1 t h e <,g e n e r a l s t a t i s t i c a l p r o b l e m ,~. This is to find which p r o p e r t i e s of a s t a t i s t i e a l a g g r e g a t e a r e
a p p r o x i m a t e l y i n d e p e n d e n t of t h e v a r i a t i o n s t h a t a.etually occur in t h e d e t a i l e d
b e h a v i o r of t h e i n d i v i d u a l s y s t e m s c o m p o s i n g the a g g r e g a t e u n d e r t h e cond i t i o n s t h a t define t h e p r o l ) l e m . T h e p r o o f of t h e q u a s i - e r g o d i e t h e o r e m a n d
t h e o r e m s ii~ m e t r i c i n d e e o m p o s a b i l i t y w o u l d t h e n b e s p e c i a l eases of s o l u t i o n s
of this p r o b l e m , b e c a u s e w h e r e t h e s e t h e o r e m s a r e v a l i d , t h e l o n g - r u n p r o p e r t i e s of t h e o r b i t in p h a s e s p a e e a r e e s s e n t i a l l y i n d e p e n d e n t of i n i t i a l cond i t i o n s . T h u s , p r e d i c t i o n s of t h e m e a n b e h a v i o r b e c o m e p o s s i b l e w i t h o u t a
d e t a i l e d s p e c i t i c a t i o n of t h e s t a t e of a n i n d i v i d u a l s y s t e m .
W e s h a l l now f u r t h e r e x p l a i n s o m e of t h e c o n s e q u e n c e s of t h e b r o a d e r a~td
m o r e ~a'enerat a t t a c k on s t a t i s t i c a l p r o b l e m s t h a t w e h a v e d e s c r i b e d a b o v e .
I n t h i s c o n n e c t i o n , we h a v e t h u s f a r c o n s i d e r e d t w o e x t r e m e s , one of w h i c h
is to d e d u c e t h e s t a t i s t i c a l p r o p e r t i e s f r o m p r o b a b i l i t i e s , a n d t h e o t h e r of which
is t o d e d u c e t h e m f r o m p r e c i s e d e t e r m i n a t e laws g o v e r n i n ~ t h e i n d i v i d u a l

(*) In tile problem t h a t ue are considering here, there is not enough time for this
to h~ppen. Thus, as ~ consequence of a theorem proved by PoI~ca~s (.Iota Matt~.,
13, 07 (1890)), in a typical mechanical system, eont.aining something like l02a molecules, it would take a fantastically long time, of the order of trillions of trillions of
years for the orbit to come reasonably near to every point of phase space.
(+) This point seems to have been for a long time known to a number of writers
in the field (see, for example, _~. 1. Kmxc~Hx: Matt~e~natical Fo,t:ndations o/ Statistical
.IIeclmnics, ~ee. 13). Itowever, as far as we can tell its full significance with regard to
the lack of neeessit.y of using a prohabilistic treatment of stat.istieal mechanics does
not seem to have been realized.

TII]~5 G E N E R A L S T A T I S T I C A L P R O B L E M I N P H Y S I C S A N D T H E T H E O R Y

OF PROBABILITY

I021

systems. We may, however, consider the possibility of deducing the statistical properties from intermediate types of laws, which are neither completely
probabilistic nor completely deterministic. The need for such intermedi~.te
~ypes of laws can be seen b y considering the fact t h a t no system can be described completely in terms of a n y finite n u m b e r of properties. First of all,
a n y system is always interacting with other systems outside of it. In m a n y
cases, these interactions seem to be small. B u t because the detailed motions
of individual systems are unstable, t h e y can in the long run, be affected a
g r e a t deal even b y small disturbances arising outside. Secondly, laws operating at a particular level are generally subject to disturbances arising a t
other levels, not included in the laws in question. F o r example, the laws of
t h e macroscopic level are subject to fluctuations of the t y p e of Browni~n motion,
arising f r o m the motions a t the atomic level. B u t the laws of classical mechanics operating at the atomic level are subject to further fluctuations arising
~rom the quantum-mechanical properties of matter. And because of the instability of motion, these fluctuations can have i m p o r t a n t effects on the detailed
behavior of the orbits even a t a level where classical mechanics would otherwise be a good approximation. Thus, one comes to the conclusion t h a t in a n y
real problem, in which one necessarily leaves out of consideration an unlimited
n u m b e r of effects coming from outside from the system and from other levels,
a precisely determinate and complete causal prediction c a n n o t be obtained.
I n general, therefore, we expect to obtain f r o m our laws predictions only of
the m e a n behavior of a system, and of the range and general c h a r a c t e r of the
fluctuations away from this mean, arising from the infinity of factors t h a t
have not been t a k e n into account. Because of the insensitivity of large scale
averages in statistical mechanics, however, a v e r y wide range of laws of the
t y p e described above is evidently possible, which lead to a p p r o x i m a t e l y the
same large scale statistical properties (such as the Maxwellian distribution).
On the basis of considerations of the t y p e discussed above, one is led to
consider the question of whether one ought not to seek a more general sta ~
tistical formalism (*) and not always to t r y to f r a m e all statistical problems
in terms of the culculus of probabilities. Of course, the t h e o r y of probability
has demonstrated its applicability in a wide range of fields, starting with
gambling games, and going on to the fields of medical, social, industrial, economic, and scientific statistics. Nevertheless, as we have already seen, in our
discussion of the foundations of statistical mechanics {certainly a s t u d y t h a t
falls within the scope of statistical problems), the theory of probability already

(*) This problem would be a


problem that we are raising here.
than the currently used calculus of
fo us by Prof. M:xRIO SCIL6~t~ERC
67 - Supp~e~nento al ]Vuovo G i r ~ n t o

mathematical reflection of the general statistical


The notion that new statistical formalisms, other
probabilities, might be possible was first sugges%ed
(Private communication.).

seems t() have little relevance. To be s u r e , it is possible (o justify statisliea|


mechanics on the basis of the h.vpotlmsis of equiprobability in phase space,
hut in most problems, this hypothesis (~an be shown not to have much to do
with the real problem (*). W h a t is significant in the real problem is t h a t large
scale averages are insensitive to most of the details of the molecular motions,
so t h a t for a wide range of possible motions, a p p r o x i m a t e l y the same macroseopic averages are obtained. The fact t h a t one m a y within a certain context
of problems, obtain correct results using the irrelevant hypothesis of equiprobability, is then a consequence of the close relationship between the equidistribution of probability t h a t would arise in an extremely tong period of
time and the properties of the motion t h a t lead to the statistical taws that
arc valid over periods of time of practical interest. ()ne m a y compare the
probabilities t h a t would apply in an extremely long period of time to a sort
nf shadow, the behavior of which can more or less faithfully represent some
of the statistical properties t h a t are relevant in shorter periods of time. [)ur
point of view is then t h a t it could be profitable to study the essentials of the
problem more directly r'~ther t h a n to study nothing" more t h a n the probahilistie <(shadow ,>.
As an example of how statistical problems can l>e treated in ~ a y s olher
t h a n those based on the theory of probability, let us consider the problem of
finding the meaning of the e n t r o p y on the basis of the molecular properties
of the system. Now, it has been c u s t o m a r y to defin( ~ entropy in terms of prohability: and for a b r o a d range of purposes, this definition has served to makc
possible the deduction of the t h e r m o d y n a m i c relations and the calculation of
some of the t h e r m o d y n a m i c properties of various systems. |gut from the work
of P. HERTZ and LEVI-CIVWA (7) one sees t h a t entropy lnay be associated
with mechanical properties of the system rather t h a n just purely probahility.
Thus it seems quite likely t h a t new models for the entropv eou|d be u o r k e d
out. We do not wish to imply t h a t either the probabilistie or the mechanical
model necessarily exhausts all t h a t there m a y be to the (:oneept of entropy.
Indeed, still other models m a y perhaps be proposed later which give us even
deeper insight into the meaning of entropy. But the mere fact t h a t models
different from the probabilistic one can lead to similar results in a broad domain
is a sign t h a t it would noI be wise to identify entropy completely with the
current probabilistie model, just because this model has led to correct results

(*) Thu.-. t)ecausc of the (~xlremely great length of lhe Poincard cycles, there is
11o ~illle for a system to come reasonably ('lose to every point on a surface of constant
onelg 3 ill phase space.
(7) p. ttERTZ: lieber-(;(r,~, in ]iepertori.um d. 1)t~ysil,. Bd. 1, 2 (Leipzig, 1916);
T. 1,FW-('IW'rA: Drei I'orle.sttttfte~ iiber adiabali.~('l~e l~raria~de,, in tbh. d. 31al],. .~'e.m,i~ar.~ Ilamb.~fl, 6. 323 (192~).

T H E G E N E R A L S T A T I S T I C A L P R O B L E M I N P H Y S I C S A N D T H E T H E O R Y OF P R O B A B I L I T Y

1023

in a certain domain. Rather, the way should be left open for new statistical
concepts, and associated m a t h e m a t i c a l methods, which m a y shed new light
on the meaning of entropy.
Another i m p o r t a n t application in which the concept of probability m a y
have only a r a t h e r limited validity is the s t u d y of non-eqnilibrium problems
in statistical mechanics. If we are interested in how a system approaches
statistical eqlfilibrium, then a certain difficulty of principle arises in connection
with the application of the concept of probability. F o r the statistical properties of a system, (e.g., pressure, temperature, velocity, etc.), are changing
with the time. The usual m e t h o d of defining probability as the average over
an unlimited period of time t h e n will not work. Of course, if the statistical
averages are changing slowly, one m a y define the probability a p p r o x i m a t e l y
in terms of the average over a period of time during which these properties
do not change very much (s). F o r rapid changes, however, even this m e t h o d
does not work. The other possible m e t h o d of defining probability is through
a statistical ensemble. B u t it seems unsatisfactory to have to t r e a t w h a t is
a f t e r all the approach of a single system to equilibrium in terms of an average
over a non-existent ensemble. Suck a t r e a t m e n t m a y give correct results in
certain contexts, b u t again it is r a t h e r like ~studying the motion of a thing in
terms of the motion of its shadow. Perhaps new statistical methods and
concepts would here m a k e possible a deeper anderst~.nding of the problem of
the approach to equilibrium in such systems.
E v e n if we restrict ourselves to statistical laws t h a t are expressible in terms
of the laws of probability, our point of view shows t h a t one can still t r e a t a
broader range of problems t h a n just the calculation of one kind of probability
in terms of another. F o r within the setting in which statistical law is regarded
as more general t h a n probability, it has meaning to d e m o n s t r a t e the validity
of the laws of probability in certain contexts, a n d to calculate the numerical
values of the probabilities, on the basis of laws t h a t are more general t h a n
the laws of probability. Thus, as we h a v e seen, we can calculate a probability
f r o m a d e t e r m i n a t e law, or f r o m a n intermediate t y p e of law, which is p a r t i a l l y
determinate. Vice-versa, a partially determinate law can be obtained from a
probabilistic law, when the n u m b e r of elements N, in the distribution is large.
F o r example, according to BernouUi's theorem, one c~n for practical purposes
predict t h a t the error in a certain average will decrease as h r-89 As N approaches infinity, the partially d e t e r m i n a t e law will become more a n d more n e a r l y
a completely determinate law. Thus, we see how the adoption of a b r o a d e r
point of view toward statistical law permits us to a t t a c k the problem of the
interconnection of various kinds and levels of law.
(8) See, for example, M. SCH/kNB~G: Nuovo CimenIo, 10, 419 (1953) where a discussion is given of the problem of taking averages in a non equilibrium process.

1024

7. -

i).

A Further

Example

of a P r o b a b i l i s t i e

I'OIIM alld

of t h e

w.

General

,~CHI?'I'ZI.iR

Statistical

Law from a Determinate

Problem

- The

Deduction

Law.

In tile previous section, we have seen in terms of all e


drawn from
statistical mechanics, in which one deals with a system containing" a very large
number of particles, how statistical relationships can be treated without the
need for their being expressed in terms of the theory of probability. I n this
section, and in the next. we shall go to the opposite extreme and consider a
system deseribed by a sing'le variable only, satisfying a simple determinate
law, a law in which, however, the motion has the essential characteristic of
extreme instability (*). We shall then see that this law leads to a long run
behavior that is independent of initial conditions, a l o u g run behavior t h a t
is just what is predicted by the theory of probability. We shall also see t h a t ,
as happened in the many-particle systems treated in statistical mechanics,
essentially the same kind of long-ram behavior will follow from a wide range
of laws t h a t are between the determinate and probabilistie types. Thus, the
justification of the laws of probability in a given application is seen to be a
part of the general statistical problenl.
In order to motivate the example that we are going" to choose, let us consider the path of a molecule in a gas, which is contained in a box. We suppose t h a t this molecule ean be represented approximately as a hard elastic
sphere. Now, it is well k n o w n that in its collisions with other molecules, this
particular molecule will follow a very complicated path, in wieh the velocity
undergoes a wide range of rapid fluctuations, both in magnitude and in direction. In the long run, however, the probability that this partieulur molecule will occupy ans" given small region of space is the same as t h a t it occupy
any other such region h a v i n g the same volume. Likewise, the probability t h a t
the direction of motion lies in any speoific range of angles becomes uniform
in the long run.
Let us now analyze h o u such a uniform probability comes about. We
first note t h a t the path of a n y given molecule possesses an extreme instability.
Thus, suppose the initial position of the i-th molecule to be x ~ and its initial
velocity v,~ We now consider a collision of this moleenle with a n y other one,
say moleeule j. Focussing our attention on the p a t h of molecule i, we see
that, the elmnge of angle of motion ,dO,, is evidently very sensitively dependent
on the collision parameter, h, (shown in Fix. 1) and therefore on the initial
position and velocity of this particle ('~).
(*) POI~(:AR~has already pointed out tile importance of instability in this problem.
\Ve shall discuss P~lXCAm::',~work and co,npare his result~ with ours ~t the end of this
,~eetion.
(+) It also depends on those of particle j, but in this discussion, we are interested
only in following what happened io particle i.

TI"IE GENERAL STATISTICAL :PROBL]~31~IIN PHYSICS AND THE THEORY OF PROBABILITY 1025

In fact~ if its initial position is changed by an a m o u n t of the order of the


size of an atom, its angle of motion after collision will change by an a m o u n t
of the order of 2m
Since the change of position suffered by the particle
between
the times t~ ~nd .~ is (as tong as no collisions
~',A01
occur between t, and t2)
__2 .........
scottermg perhcle
scofter~ngparticle

Fig. I.

Ax~ =

v~

-- h),

it is clear t h a t in a series of collisions, there is a continual multiplication of the instability, so t h a t after


a sufficiently large number of collisions, a change of initial position of less
t h a n the size of an a t o m can carry the particle to a very different region of
space and to a very different direction of motion,
9
as shown in Fig. 2.
I t is already clear intuitively t h a t under these
/
conditions a very irregular motion will develop
/
which does not favor a n y particular region of
space. Thus the equiprobability of all regions of
...... 20
space becomes plausible on the basis of the kind
of motion t h a t exists.
;2Z2ZJZZ:::-_~D
:
07
I t would evidently be desirable, however, to
d . o'
prove this equiprobability in a more rigorous
way. Unfortunately, in the problem described
"'~)
above, the mathematical difficulties a t t e n d a n t on
such an exact proof have not yet been resolved.
Fig. 2.
We shall therefore propose here a simplified
example, which permits an exact treatment, and which still retains the essential property of instability of motion, t h a t is evidently at the root of the
development of an equiprobability in the distribution
~ C
of colliding molecules.
~. Oo,
In this model, we consider a point, P , , in a two
"',
dimensional space. The coordinates, x , , y , , repre7~. . . . .
sent the location of the n-th particle during the
n-th collision, while 0~ represents its angle of motion
zf,koo___after this collision, as shown in the figure below.
v.
We now assume t h a t the particle, makes a new eoll.'ig. 3.
lision after moving a definite distance, L, (In reality,
this distance L fluctuates from collision to collision;
but by regarding this distance as a constant, we are further simplifying the
problem.)
We further assume t h a t in the process of collision, the angle 0~ is mul-

1026

1,. t~.OHMand w. 8':H(h:ZER

tiplied b y a large irrteger, K. or t h a t

0,~+1- KO,,.

(7.1)

We see t h a t this hypothesis embodies the essential element of instability~


since a small change of angle ~0, produces a much larder change of angle,
AO~+~ -- KAO,~. A change AO,, = 2~/K produces a change ~0,,~ - 2~. But,
whenever the an~'le changes b y 2~ it comes back to its original value. Thus,
we should write 0,, :, ~ K O , - [KO,,] where [KO,,I is the larta'est multiple of 2~

iu KO,,.
We see also that our simplified model gives a schematic representation of
tile significant aspects of tire actual process of collision. First we note t h a t
*he angle 0,_~ changes by 2~r when tile initial position of tlre particle just before
it, strikes the (~ 1)-th obstacle changes b y LdO
2~L/K. Now if K islarge
it is clear that a small change of position will lead to a change of angle of 2a.
Thus we e m b o d y in our model something sinfilar to tire extreme sensitivity
of the angle of collision to the collision parameter, t h a t is so i m p o r t a n t in
real collision processes for bringing a b o u t a uniform distribution. In fact,
one could regard this model as more or less equivalent to having relative to
each collision point a uniformly distributed set of obstacles on a circle of
radius L. each of diameter 2.~L/K. Of course, in a n y given collision, tile locations of the effective obstaeles will depend ml where the previous collision
took place: wherea,~ in the real process it does nol. In this sense, our model
is a further simplification: but in the present work, the essential aspe('l of instability of motion is the principal feature of the real process t h a t we are interested in retaining in the model.
Let us now consider a sequence of collisions, starting with 0 :-: 0o. We
have

[ O,

liO,, - [ KOo l,

[ o . . . . . K"-o. -

(7.2)

tK%t

t 0 , := h ' v 0 , , - - t K % l

We then write our sequenee associated with N collisions as

,,,( x , O o / = {oo, o, .....

o,,}.

Note that each 0o determines an entire sequence.


We now divide 0 into elements of size A0, of small, but fixed dimensions.
And in each sequence ,~(N, 00) we call ask how ma-ny ,'alues of 0,, fall inside
the i-th element (located near 0- :0i). This q u a n t i t y we define as ~o(N, 0o, 0~)A0,.

T H E ~4ENERAL STATISTICAL P R O B L E M IN P:[IYSICS AND THE T H E O R Y OF P R O B A B I L I T Y

102~

Now, in order to d e m o n s t r a t e the d e v e l o p m e n t of a uniform distribution,


wha.t we m u s t do is to show t h a t for a r b i t r a r y 00 (with the possible exception
of a set of m e a s u r e zero) Q(N, 00, 0~) a p p r o a c h e s a constant, independent of
00 a n d 0~ when N a p p r o a c h e s infinity.
There exist well k n o w n t h e o r e m s in the t h e o r y of n u m b e r s , which p e r m i t
us to p r o v e t h a t in our case (as well as in a m u c h wider class of similar problems), p(N, 00, 01) does indeed a p p r o a c h a c o n s t a n t (9). I n the present paper,
however, we shall d e m o n s t r a t e our results in a n o t h e r w a y which has the
a d v a n t a g e of bringing out more clearly the essential features of the processes
leading to e q u i p r o b a b i l i t y t h a n is possible with the more a b s t r a c t methods,
such as those used b y WEYL.
L e t us first consider a sequence in which the particle arrives in the region
A0, a f t e r the n - t h collision. We now a s k : W h a t would h a v e been the values
of 0 before the n-th collision? One possible set of values could h a v e been a
region of width AO,_j -~ AO~/K n e a r 0 : _ 1 = O~/K. B u t since the physically
significant v a l u e s of 0 r u n b e t w e e n 0 a n d 2~ the particle could also h a v e come
f r o m a n y one of K regions of similar width located n e a r (O~/K)~,2gM/K
where M is a n integer running f r o m 1 to K - - ] . Thus, the possible regions
are u n i f o r m l y distributed around the circle. (See Fig. 4).
B u t each one of these regions z~0.n-- 1 c o u l d h a v e come
f r o m a set of similar uniformly spaced regions of width
AO,-2 : AO,:_~/K = AO,~/K:. If we continue this analysis
b a c k to 00, we discover t h a t the values of 0o which can lead
to a given AO,~ are situated inside K" uniformly distributed
regions, each of width AOo::AO,/K".
Let us now consider a n y given sequence, S(N, 00) where
0o lies in the r a n g e AOo, 01 in AO~, etc. F o r the sake of
Fig. ~.
convenience, we choose all of our ranges AOo, ..., AO~vequal
to each other a n d equal to 2z~m/K where m is an integer which is m u c h
smaller t h a n K, b u t large c o m p a r e d with unity. Thus, the unit circle is
divided into K/m regions, each of w i d t h 2zm/K.
We now ask (( W h a t is the r a n g e of 00 t h a t can lead to a n y given sequence,
in which the particle goes t h r o u g h the regions AO~, AO~, ..., AOv? ~>.This quantity, we call the <~m e a s u r e )> of our sequence.
Now, as we h a v e seen, there will be K uniformly distributed regions
of width (~Oo~--AO~/K f r o m which the particle started, if it arrived in
the region, AOj. The t o t a l measure of these regious is K A O ~ / K : AO~.
B u t now, if the particle goes t h r o u g h the region A02, there are likewise K
uniformly distributed regions of the angle 0 each h a v i n g w i d t h AO2/K, f r o m

(~) See, for example, H. ~'~,YL: Math. A~n., 77, 313 (1916).

|()~8

I). B l l l | ' q

;llld

V~. S('III~'I'ZE|{

which t h e p a r t i e l e could h a v e s t a r t e d its second step. B u t of these; o n l y m


will be i n t h e r e / i o n of width lOj, in which the l>arlAele was a c t u a l l y l o c a t e d
.just before this second step.

Henee, t h e m e a s u r e of t h o s e t r a j e c t o r i e s passinK

t h r o u g h l>oth 10+ a n d J0., will be m ~O~/K--

100-(d00/2,~). S i m i l a r l y , the mea-

sure of those t r a j e e t o r i e s p a s s i n g t h r o u g h 10~, J0e a n d ,J03, will be J00" (/100/2~) '2


a n d t h a t of those passing" t h r o u / h 10p . . . . . 10,. will be JOo.(/lOo/2X) `v. T h u s .
only a v e r y s m a l l f r a e t i o n of all possible i n i t i a l v a l u e s of 0o e a n lead 6o a seq u e n c e of ano~les passinR' throuo'h a / i v e n set, lO;,..., 40~,. The p r o b l e m is
v e r y s i m i l a r to t h a t of having" to ~z'et a b a l l to ~,o t h r o u g h a series of i r r e g u l a r l y
spaced holes, if t h e b a l l is deflected as it passes t h r o u g h each hole b y a n a m o u n t
t h a t is v e r y s e n s i t i v e to the d i s t a n e e f r o m the center. To get t h e b a l l t h r o u g h
all of the holes, a f a n t a s t i c a l l y R'ood a i m would be needed.
~)nr n e x t step is to lind the m e a s u r e of those v a l u e s of 0o t h a t

can lead

to a ~'iven d i s t r i b u t i o n , N,
9(N, 0o. 0,) I0,.
Now, as we h a v e seen, a n y gi~',>+ s e q u e n c e has a m e a s u r e of /10.( ,10/2x)".
W e n m s t no~ c o u n t all s e q u e n e e s t h a t ean lead to a ~iven o(N, 0o, 0,) a n d
therefore l o a

~'iven set of N,.

This mm~ber is (*)

-V'. NI'.

X>!

... N , ~

Thus. lhe m e a s u r e of ihose v a l u e s of 0,, leadin~" to a ,/iven dist|'il>ution of t h e


:V, is

(7.3)

.11=

.V~

'..\.,.
='

,- , ( = 1 0 2 7 ) u

l0.

['sim,/ ,~tirlin~'s a p l ) r o x i m a t i , m , we ~et

(7.4)

In ( J l ' 10) ~ .VIn N - - "~ N, In N, -- N In (30/2~) s .

We o b t a i n the resutl t h a i t h e d i s t r i b u t i o n havin,_," the hi~z'hest m e a s m ' e is


(7.5)

.V

c o n s t a n t - - : \ ~ = N 10 2.~.

(*) llere, we make use ~f lhe h , ' t thai if any ~iven ~equence. /(0 o, =Y) is po.~.~ible
lh#ll ~|lly other .-e~luence involvin~ an exchange of any rwo of the magles is also possible. This follows from the lar~e value of the inte~er, E, which has as a consequence
thai ,_,'iven any ~ alues ,}f 0,,. ~e ,'an always obtain any desired ~ alue of 0,~ ~1 simply bs:
a small change of q, thai doe,~ not take il out of the region :10,~ of width. 2.~m/K. :

T H E G E N E R A L S T A T I S T I C A L PROBL]dM IN P f t ' i - S I C S A N D T H E T H E O R Y OF P R O B A B I L I T Y

109~}

B y defining 6N~ = N~ -- N, we obtMn


(7.6)

In M ~ In Mo -- ~ (~N~)~/27e = In Mo -- ~2~(6N~)2/2NAO.
/

Writing for the fractional deviations, 6]~ = 6 N J N - - 2 z 6 N J N A O


(7.7)

we get

M ~ M~ e x p [ - - ~ 89
i

We see t h e n t h a t the measure of those distributions in which 6]~ is a p p r e ciable becomes negligible as N - + c~. Thus, fixing a given AO and a given
6]io, however small, we can always choose N so large t h a t the measure of
those sequences in which ~]~ > ~]~0 is less t h a n a n y specified number.
W h a t is the meaning of this result? We first remind ourselves t h a t all of
the possible sets of initial angles, t h a t can lead to a given distribution,
N ~ = O(0o, 0~, N) arrange themselves into small elements of width, AO/K ~.
Thus as N - + c~, the width of each element approaches zero v e r y r a p i d l y .
This behavior corresponds to the e x t r e m e instability of motion, since a change
of A O o - - A O / K ~ would change 0~ b y AO and therefore alter the distributio n.
Thus, to get any given t y p e of sequence, as N - + ~ we would need a
perfect fit for the b o u n d a r y conditions. B u t as we have seen, the overwhelming
majority of sequences correspond to distributions t h a t are v e r y nearly uniform.
Only a comparatively small n u m b e r of regions of width flO/K ~ will lead t o
a n y t h i n g different (such special regions will occur, for example, where 0o is
rational). Although the n u m b e r of such special regions approaches infinity
as N--> c~, this approach is much less rapid t h a n is the decrease in width
of each region, so t h a t the total measure of such special regions ~pproaehes
zero,

aS

2V--+ c~.

With regard to the theory of probability, the essential meaning of the


above result is t h a t the long run average distribution of angles is practically
independent of the initial angle, 0o. I t is true t h a t there is a special set of
measure zero t h a t could give rise to non-uniform long-run distribution, b u t
one can show t h a t such special values of 0o play no role in real problems. To
see why this is so, we note t h a t in order t h a t one of these special non-uniform
distributions should come about, it would be necessary t h a t the value of 0,
be defined exactly (i.e., with no error whatever). We know, however, t h a t it
is impossible in reality to have a physical system in which the value of a
continuous variable is defined as exactly equal to any given number. F o r every
system is always interacting with other systems in the environment which also
undergo irregular fluctuations in the motions of their elements (such as a t o m s
and molecules). Thus, even if a particle accidentally did h~ve its initial value
of 0o exactly equal to one of the special set of measure zero described above~
it would be sure to obtain a slight disturbance from its environment; and
because of the instability of motion, even the slightest disturbance would c a r r y

/o30

D.

BOI[M

alld

w.

S('II[YTZER

it far from this special state and lead eventually to a unifornl distribution.
S o w , it is true t h a t such a disturbance might accidentally throw the system
into a n o t h e r of the special angles, 00. of the set of measure zero, not leading
to a uniform distribution. This would require however a very special exact
coordination between the external disturbance and the motion of the system.
I t is evident t h a t for practically all possible motions, this kind of eoordinat,ion
would not exist. Thus, it is consistent with the t h e o r y to suppose t h a t the
motions are such l h a t there are no such coordinations. Moreover, this supposition is also consistent with all of our practical experience. For example,
if we consider a cone standing on its point, this position is one of unstable equilibriunl. It would be conceivable that the fluctuations of the surrounding air
and of the table on which it is supported would be just, such as to maintain
it in its position of unstable equilibrium for all tinle but evidently this never
happens in reality (*).
We conclude then that the t r e a t m e n t of the sequence given here is a special
ease of the general statistical problem, ill the sense t h a t certain long run or
average properties can be predicted without taking into account the precise
factors needed for a prediction of the details of the behavior of the system.
For as we have sPell, once we exclude tile special set of O0 of lneasure zero
described above, then the mean distributiou of angles is independent of 00,
and is therefore independent of the b o u n d a r y eonditious t h a t determine the
precise motion. We e<)llelude then t h a t the average value of any int.egrable
x

function of O, ](0) - - lira(1 I X ) ~./(0,,) will, indel)endently of initial conditions,


Y-+e~

be equal to

(7.s)

" ~

l(0)

--

~ ;t/

tf(~176 9
o

The above is evidently equivalent to a statement of the equidistribution


over angles. Thus, as far as computing averages of any funetion of 0 are,
(.oneerned, we obtain the same result as that whieh follo~vs fronl using the
theory of probability, with the probability distribution, P(O) set equal lo a
constant.
We shall now compare our results with those of POINCARg: (~o), who treated
a similar problem, but by different methods. PO~NCAt:~ ~'ealized clearly the

(*) it i.~ interesting to note that ('OURXOTill his ~J'l)ositiot~ de la Thdorie des (~i~atu'e.~
et des P.robabilit~ already realized tlle importance of this point, and in fact used the
example of the cone standing on its point as all argument in favor of the the~is that
in chance phenomena, no special properties depending on the infinitely precise ~peei~ication of any particular quantity could be of physical importan(.e.
(lo) H. POIXCARs ('ale,l de.~" t'roh,,hilitd.~ (Paris, 1912).

T I I E G E N E R A L S T A T I S T I C A L P R O B L E M I N P H Y S I C S A N D T H E T H E O R Y OF P R O B A B I L I T Y

1031

i m p o r t a n c e of i n s t a b i l i t y in leading to a n equidistribution. Indeed, he considered t h e case of a p a c k e t of asteroids, each in neighboring, b u t slightly different circular orbits. Because the p h a s e angle of a n y p a r t i c u l a r asteroid relative to t h a t of a n o t h e r is u n s t a b l e {i.e., it increases indefinitely with the time),
one expects t h a t in t i m e the asteroids will h a v e a uniform a n d more or less
r a n d o m distribution of p h a s e angles (which is in f a c t found in a s t r o n o m i c a l
o b s e r v a t i o n s (*)).
I n order to t r e a t this p r o b l e m , POI~CAR~ considered a n a r b i t r a r y continuous
initial p r o b a b i l i t y distribution (e.g., one h a v i n g a v e r y n a r r o w b u t continuous
p e a k n e a r a certain point). H e t h e n showed t h a t i n d e p e n d e n t l y of the precise
initial f o r m of such a distribution, the p r o b a b i l i t y would e v e n t u a l l y b e c o m e
a c o n s t a n t t h a t is i n d e p e n d e n t of the angle. F o r discontinuous initial distributions (e.g., a delta function) this would not in general occur. B u t here
POINCA]~ used a n a r g u m e n t similar to the one b y which we h a v e excluded
sets of m e a s u r e zero as not r e p r e s e n t i n g real physical possibilities, l i e did
this b y pointing out t h a t in all real problems, there is a b a c k g r o u n d of fluct u a t i o n s coming f r o m influences outside the scope of w h a t is t r e a t e d in the
p r o b l e m (e.g., the motions of other planets, a n d even the molecular motions)
so t h a t no discontinuous p r o b a b i l i t y function (such as a delta function) can
represent a n y real physical situation.
POINCAR]~S conclusion was thus essentially the same as ours. B u t the
weak point in his a r g u m e n t is t h a t his m a t h e m a t i c a l methods do not correspond to the real context of his physical ideas. Indeed, f r o m a strictly
logical point of view, the a s s u m p t i o n of the initial n a r r o w l y p e a k e d a n d continuous initial distribution is not f u n d a m e n t a l l y different f r o m m a k i n g assumptions of the u l t i m a t e u n i f o r m p r o b a b i l i t y distribution to begin with. I t
would thus b e concluded t h a t while P o I ~ C A ~ has demi)nstr~ted a n interesting
a n d useful relationship b e t w e e n two p r o b a b i l i t y distributions t ~ k e n a t different times, he has still r e m a i n e d within the basic f r a m e - w o r k of lust calculating one p r o b a b i l i t y in t e r m s of another.
On the other hand~ the significance of our result is t h a t the d e v e l o p m e n t
of a n eqnidistribution .can be u n d e r s t o o d w i t h o u t a n y reference to the concept
of p r o b a b i l i t y a t all. This we h a v e done b y recasting WEYL'S results on the
d e v e l o p m e n t of a n equidistribution in a f o r m in which the physical significance of the sets of measure zero could easily be seen, so t h a t we can unders t a n d w h y t h e y m u s t be excluded (+).

(*) The rings of Saturn, m~de of sm~ll-meteor-tike particles, present ~ similar


problem.
{+) In section 10, we shall see how an example similar to the one discussed by
POI~'CAR~ can be treated, without the use of probability.

1032

D. B()IIM

ltIld

w.

S('ttl}TZEtR

F i n a l l y , ~e shall consider the possibility of p r o v i n g e q u i - d i s t r i b u t i o n in


terms of l a ~ s i n t e r m e d i a t e b e t w e e n d e t e r m i n a t e a n d probabilistic t y p e s , dis~'ussed in the previous section, in connection with statistical m e c h a n i c a l p r o b lems. It is i m p o r t a n t to consider such problems, b e c a u s e t h e n o t i o n of a single
nieehanieal variable, following a d e t e r m i n a t e law, defined with u n l i m i t e d pre('ision. is evidently a n a b s t r a c t i o n . I n reality, every such m e c h a n i c a l varif+ble
is subject to flu('tuatin~' p e r t u r b a t i o n s of the t y p e t h a t we h a v e m e n t i o n e d
in the l)revious section a n d in the a b o v e p~ragr~l)h, h i o t h e r words, it is in
reality coupled, even if only slightly, to a n infinity of o t h e r p a r a m e t e r s . The
e x t r e m e instability of m o t i o n could c o n c e i v a b l y m a k e such coupling significant
over a long time. I t is i m p o r t a n t therefore to show t h a t t h e r e exists a wide
range of cases in which the e q u i d i s t r i h u t i o n would n o t be significantly affected
b y su('h pertu|'bations.
In order lo do this, we introdu(.e the n o t i o n of the ,~eq~e~e~' space. The
c o o r d i n a t e s of a point in this space are specified b y the values. (~o, qG ..... c~.,,)-[q~,,] of the v a r i o u s an~,'les in a n y specified sequence t h a t we wish to consider.
Moreover, on('e we h a v e chosen of.. their a n y point in the subspaee ( ~ ... ~ x ) - - [-~c,,:cfo1, which starts at 9~--g,, ~.or|'esponds to a p a r t i c u l a r d e t e r m i l m t e rela tionshi 1) defining e v e r y l erm in the sequence in t e r m s of q%. (The relationship
rf,, = K,,T. ) is evidently a spe(.ial (,ase). l~y considering a range, d ~ ... d~_,.,
of points (.orresponding to a given q:., we are effectively considering a r a n g e
of possible relationships to d e t e r m i n e the sequem,e.
Now, as we (lid in our discussion of statistical m e c h a n i c s in Se('. 5, we can
define the ravage' o/ seq~,)~c~, ,~pace eorrespondin~ to a given distribution of the
angles. 0 (instead of the r a n g e of pt~o,~c ,~paee considered in See. 5). \Ve need
not go t h r o u g h the m a t h e m a t i c a l details here, as t h e y are v e r y similar to
w h a t u as done in the statisti(.al n~eehanieal p r o b l e m , b u t we shall m e r e l y quote
the result here.
An equidistributi~m of t h e angles, 0, corresponds to t h e m a x i m u m possible
v o l u m e of sequence space (*), distributions n e a r the u n i f o r m one c o r r e s p o n d i n g
lo a v o l u m e of sequen(,e spa('e t h a t decreases a c c o r d i n g to a G a u s s i a n curve.
W h e n X is large, this decrease is so r a p i d t h a t a h n o s t all of sequence space
c o r r e s p o n d s to a u n i f o r m distribution.
The a b o v e result has a n u m b e r of interesting consequences. F i r s t of all,
we see t h a t the special law (~c, - - K"9~0) t h a t we are discussing is n o t essential
to an e q u i d i s t r i b u t i o n , so t h a t there are a wide range of d e t e r m i n a t e l a w s

(*) In tile st,atisli('al m e c h a n i c a l problem, t h e Maxwellian distribution, instead of


the uniform one. wa,~ o b t a i n e d as t h e one corresponding to m a x i m u m v o l u m e in phase
~paee, be~.ause of t h e r e q u i r e m e n t s of c o n s e r v a t i o n of energy. No analogou~ requirement a,ppears in the present problem.

~ I I E GEN~ERAL S T A T I S T I C A L P R O B L E M I N P H Y S I C S A N D T H E T H E O R Y OF P R O B A B I L I T Y

1033

t h a t could lead to the same statistical results. Indeed, even ~ law intermediate
between determinate and probabilistie could h a v e the same statistical effects,
provided t h a t as h r increases without limit, it is quite rare t h a t the point in
sequence space finds itself inside the minute volume corresponding to distributions deviating appreciably from a uniform one.
The above result must, however, be interpreted with care. I t does not b y
any means signify, for example, t h a t all distributions must be uniform. Laws
are evidently quite possible t h a t would lead to non-uniform distributions, or
even to distributions t h a t approached no p a r t i c u l a r limiting behavior at all.
All t h a t we h a v e shown is t h a t a wide range of laws exist t h a t could le~d to
uniform distribution, so t h a t the appearance of such a distribution need
not be sensitively dependent on having a certa.in special kind of law determining
the sequence. Moreover, we can also u n d e r s t a n d why a wide range of perturbations of the system need not disturb the uniform character of the distribution significantly.

8. - The General Applicability of the Calculus of Probability.

Thus far, we h a v e considered only the development of equidistribution in


t h e mean behavior of the angle, 0n. The calculus of probability deals, however,
with properties t h a t are more general t h a n just the mean behavior of a simple
event. I t deals also with the m e a n b e h a v i o r of various combinations of events.
Thus, in the case of throws of a die, the t h e o r y of probability permits us to
estimate the relative frequency with which we shall obtain two sixes in a row,
b y multiplying the probabilities for each result separately, according to the
well known rules t h a t apply for combinations of ~(independent ~>events. Moreover, the t h e o r y of probability also applies to problems in which the probability distribution is not uniform.
I n this section, we shall show in terms of the same sequence t r e a t e d in
See. 7 (0, ~ K"Oo) how a determinate law can lead to the validity of the rules
for combining probabilities. I n the course of the t r e a t m e n t of the above
problem, we shall also show how all of our results can be generalized to cases
in which the probability distribution is not uniform.
The first problem t h a t we shall consider is t h a t of estimating the mean
n u m b e r of times in which a given angle Oi is followed b y a n o t h e r specified
value, 0j. More complex combinations involving three, four or more angles
m a y evidently also be of interest.
To t r e a t this problem, it is convenient to introduce as a n i a t e r m e d i a t e
concept the notion of conditional probability. Thus, in our problem, a typical
conditional probability is p~lidO~dO~ the probability t h a t if the angle after
given collision lies between 0~ and 0~q- d0, the angle after the n e x t collision

1034

I).

will lie between 0 , : a n d 0 , pound event i~i.x is then


(8.1)

B~)tl'd

dO,.

dP,,

;llLd

~,.

t:('III:'I'ZFt~

[! is evident t h a t the lwobal)ility of the c o m -

p , "l~,,, dOqdO, .

.\ special case of considerat)le interest arises when p + , is i n d e p e n d e n t of i~.


In o t h e r words, the pvot>ability <listribution f<>r the second angle does ,lot
d e p e n d on the first ang'le. I t is easy to see t h a t in this ease pq+ - - p~. Thus"
the l>rol>ability of the c o m p o u n d event be<.omes
('q.:.')

<l P i]'

]},' P'_,"dOdlO' -

a n d we o b t a i n lhe well-known p r o d u c t rule. which applies for <~i n d e p e n d e n t ,>


events. This rule implies t h a t t h e r e i s no correlation b e t w e e n suceessive a.ngles.
For if there w e r e correlations the 1)robability of obtainin~ 0,, would d e p e n d on
which ang'le e a m e before.
W e m a y also be interested in the conditional l)robal)ility if the angle after
a xiven collision lies b e t w e e n O, and O, . (t0, the a n / l e a f t e r :q a d d i t i o n a l collisions will be b e t w e e n O, a n d 0,: -dO,:. This we d e n o t e 1)y

p;~,.(10,, d0,~.
Moreover. we m a y l)e interested in c o m b i n a t i o n s involvinR' m o r e events.
Thus. we wish to define p r o b a b i l i t y t h a t if a f t e r a ~iven collision t h e an~'le
is b e t w e e n O, a n d O+ dO, a n d a f t e r x~ a d d i t i o n a l collisions b e t w e e n O.q and
OqC-dO,: the ano'le a f t e r still a n o t h e r y.., collisions will be b e t w e e n Oq aitd
Oq dOq. I t is e v i d e n t t h a t a r b i t r a r i l y hio/h order conditional probabilities
can be defined in this wax-. The p r o b a b i l i t y of a coml)omld e v e n t (i,i2 ... it-)
is the|~
(s.3)

d P' .....~"-' = =

'~

)" i'"'~'-'dO,,

dO,x

_\ case of .areal interest is one ill which t)q, is i n d e p e n d e n t of i ,, P q s q indep e n d e n t of i t . i.,, etc.. It is readi]ly verified t h a t for this ease

(8.4)

t),,,.

P,

t I tff,~

P,~ .....

P,,,, .... x

= t),x

and

(8.5)

<1t',,

...,+

p, 1),_~ ... p,,. dO, ... d0,~..

This is the case of e(mJlilete a b s e n c e of correlation up to order K, or of


complete <~statistical i n d e p e n d e n c e ~,.

TIlE

G]~NF~I~AL S T A T I S T I C A L

PROBLEM

IN PHYSICS

AND

TH~

THEORY

OF PROBABILITY

~03~

We shall now show t h a t with our sequence, 0, = K'~Oo, the conclusions


drawn from using probabilities of compound events are valid. To do this,
we must deal with two questions. First, do quantities t h a t would correspond
to the probabilities of compound events exist {i.e., do the associated relative
frequencies approach limits independent of 0o)? Secondly, if so w h a t are the
values of these quantities?
In order to s t u d y this problem, we generalize the s t a t e m e n t of equidistribution given in eq. (7.8). First of all, eq. (7.8) applies only for uniform
probabilities. One could however define a more general probability, P(0) such
that
2y~

(8.6)

l(O) =

2Yg

(0) t(O) dO
0

(0) dO = .,Vlira
.-..-).co

1(0.)

which would hold even if tile distribution were not uniform. I n order to prove
t h a t a probability, P(0), really exists for a given t y p e of sequence of angles 0~,
it would be necessary to show t h a t it is possible to find a function such t h a t
eq. (8.6) holds independently of 0o, except possibly for a set of measure zero.
In Sec. 7 we have already done this for the sequence, 0" ---- K'Oo, in which
case it turns out t h a t P(O) is a constant. Evidently, however, a more general
problem can be a t t a c k e d in a similar way. I t is not difficult to find a simple
determinate sequence leading to a non-mfiform P(0). I n this paper, however, we restrict ourselves to merely pointing out t h a t the m a t h e m a t i c a l framework is already broad enough to p e r m i t t h e t r e a t m e n t of problems involving
non-uniform probabilities.
The probabilities of compound events can be t r e a t e d similarly. To do
this, we consider a n y integrable function of two variables ](0~, 02). L e t us
recall t h a t in our sequence b o t h 0~ a n d 0.+~, are determined in terms of 00.
Thus, once 0~ is given, 0, is certainly determined. In general, such a determination would lead to correlations (e.g., 0,+~ 1 could be equal to 0 . - ~ ) .
W e shall see, however, t h a t in our problem, the relation between 0,+~, and 0.
fluctuate sufficiently as n changes so t h a t no such correlations exist.
If a probability P~,(0~02) of the compound event exists, t h e n we m u s t have

(8.7) lira

/(0~, 0,+6,) =

Assuming t h a t a suitable function, P~,(0102) exists, t h e n evidently P~,(0~02)


can be interpreted as the probability of obtaining an angle between 01 a n d
01
in the n-th step followed b y an angle between 02 a n d 02~-d0~ in the
(~ ~- ~l)-th step.

10;{(j

D.

BOIIM

alld

w.

SCtlI;TZEtl

P e r ~he s p e c i a l ease in whi(,h P(Ot02)-- P(O~)P(O,.,) t h e r e is t h e n no ( : o f


r e l a t i o n b e t w e e n 0, a n d 0. . . . , while a still f u r t h e r s i m p l i f i c a t i o n is o b t a i n e d
w h e n P(O~) a n d P(0._,) a r e b o t h eonst.~nt.
To a p p l y eq. (8.7) to o u r s e q u e n c e , we s h a l l r e s t r i c t o u r s e l v e s to l(0~, 0,+~,)
t h a t c a n b e a p p r o x i m a t e d a d e q u a t e l y in t e r m s of a finite b u t a r b i t r a r i l y l a r g e
n u m b e r of t e r m s in a F o u r i e r series
[

(~.8)

.v

x,,~/(o,,,
o,,
--

,,/-- 2

n !..,

',,,,, "~p

[iifo,,

: ,,,o ..... /3.

As we shall see l a t e r such a f u n c t i o n is ~'eneral enon~'h for m o s t a p p l i c a t i o n s .


W c t h e n h a v e (using" 0 ..... = K~'O,)

(s.o)

~ I(0,,, 0 ..... /

exp

[iO,,(~

,,~ So',)]

W e t h e n w r i t e 0,,
h'"O, a n d ( / - ) ~ t K " ) = C,,, ( p r o v i d e d t h a t C,,,,~-0).
replacino,. 0, b y 0,,,, :- ~'~,,,0~ we ~et t h e s u m in eq. (8.9)

2"',,,v

Tlten

1 ~ e x p [iK"O,,,,]

Since. t h e d i s t r i b u t i o n of ang'les, 0 , ~ - K"O,,,< is u n i f o r m , each t e r m in t h e


a b o v e d o u b l e series will v a n i s h ( e x c e p t for a set of 0,,,, of m e a s u r e zero).
T h i s e o n e l u s i o n , h o w e v e r , does n o t a p p l y for C z , , , - - 0 a.nd since K was
a s s u m e d to b e an i n t e g e r in o u r o r i ~ i n a l p r o o f of t h e equidis~ribu(sion, Lerms
with r
can a l w a y s a p p e a r . W e n o t e h o w e v e r t h a t W v Y i . h a s s h o w n
t l m t t h e a b o v e s u m is zero e x e e p t for a set of m e a s u r e zero e v e n w h e n K is
n o t r a t i o n a l . To s i m p l i f y t h e p r o b l e m we s h a l l t h e r e f o r e s u p p o s e f r o m n o w
on t h a t K is i r r a t i o n a l , so t h a t C,,,~ is n e v e r zero, unless 1-- m 0. W e t h e r e f o r e
c o n c l u d e t h a t ( e x e e p t f o r a set of 00 of m e a s u r e zero)

(~.io)

hm

N ](O,,, O~ ,,)

ao,,.

1/111 if we a s s u m e a prolmbili~y, P(0IO.,)= c o n s t a n t , t h e n in eq. (~4.7) we


]|ax'e
(8.1.1)

j)

(01 0.,) dO, dO._. ~- a .....

Tim,s, we h a v e p r o v e d t h a t i n d e p e n d e i i t l y of 0o, t h e a v e r a - e s of a r b i t r a r y
funot.ions, ](010.,_) t h a t a r e e x p a n s i b l e in a finite n u m b e r of t e r m s in a t~'oul'ier

' ~ H E G ] ~ N E R A L S T A T I S T I C A L P R O B L E M I ~ ~ P H Y S I C S A N D TH]d T H E O R Y O F P R O B A B I L I T Y

1037

series, are the s a m e as those o b t a i n e d f r o m a p r o b a b i l i t y P(O~02) = constant.


Hence, we h a v e also p r o v e d t h a t there is no correlation b e t w e e n O~
and 0,~+~,.
I~ is evident t h a t these ideas can be applied to the case of the probabilities
of K-fold c o m b i n a t i o n s of events; a n d a simple extension of the a b o v e results
shows t h a t
(8.12)

P ..... ~ _ (0~ ... 0~) = c o n s t a n t ,

so t h a t the K-fold correlations vanish, where K is a r b i t r a r i l y large.


A t this point, it should b e n o t e d t h a t t h e set of values of 0o for which t h e
distribution of the simple events 0(0), is not uniform, will not in gel~eral coincide with the sets 0~m for which the various distributions of compound events
fail to be uniform. Thus, if 00 is one of the values for which 0(0) is n o t uniform, t h e n as we h a v e seen, Ozm = C~mOo= (l~,mK~')Oo will be a value for
which the distribution of twofold events corresponding to a s e p a r a t i o n of al,
fails to be uniform. Hence, the m o r e kinds of distributions t h a t we t a k e into
a c c o u n t a n d the more t e r m s t h a t we t a k e in the Fourier series, the more values
of 0o there will be for which some distribution fails to be uniform. Of course,
as long as the n u m b e r of distributions a n d the n u m b e r of t e r m s in the Fourier
series is finite, we can e v e n t u a l l y o b t a i n a n u m b e r of elements of the sequence
as a whole, N, so large t h a t the t o t a l measure of all these speciM angles is a s
small as we please. ~ e v e r t h e l e s s , the more different kinds of distributions
a n d the m o r e t e r m s t h a t are used in the various Fourier series, the larger N
will h a v e to be in order to m a k e t h e m e a s u r e of this special set less t h a n a n y
n u m b e r , s, which represents the precision t h a t is significant in the conditions
of the p r o b l e m of interest.
I n order to bring out more clearly w h a t we m e a n b y the notion of significant precision, let us first note t h a t in each p r o b l e m there will in general
exist some r a n g e of angles, AO, such t h a t no i m p o r t a n t properties within the
c o n t e x t of interest will depend on a definition of the angle with a precision
greater t h a n AO. We t h e n introduce t h e so called characteristic function,
Z of the set of angles of interest, which is zero outside the set a n d equal to
u n i t y inside the set. T h e m e a n value of Z is t h e n equal to t h e p r o b a b i l i t y
t h a t the s y s t e m will be within the r a n g e AO. Now, to a p p r o x i m a t e a function
such a~ Z, we need a Fourier series h a v i n g a n u m b e r of t e r m s of the order
of 1/AO. Thus, we see how the n u m b e r of t e r m s in the various series is determined.
I n order to show w h a t determines the n u m b e r of distributions of comp o u n d events t h a t needs to be t a k e n , we first point out t h a t in a n y paxtieular
problem, only a finite n u m b e r of combinations of events will in general be
i m p o r t a n t . F o r exaniple, in a n effort to (~b e a t )) a g a m e of coin throws, a
68

SappZemento al N u o v o Ci'me~do.

1038

D. B()tI?,I a l l d

w.

~CIII-TZEIr

player might choose a <<system ,7 of plays, such t h a t if he obtains a head, and


after :q throws, another head, he will then bet on heads for the play comin~~ throws later still. The vanishing of three fold correlations is enough to guarantee t h a t this systems will not ~,beat ,~ the ,,,,'aIne. More general systems
could be invented such that he would bet only when the results appeared in
some other specified but predetermined w a y involving a n y desired n u m b e r
of throws, and as lon}z as the corresponding combination of events has a uniform distrilmtion, he cannot win systematically by a n y such choice. A different example which is more conlplex, b u t which illustrates the same principle
comes from statistical mechanics. Thus, the total potential energy of the
molecules of a .._,'as is E -- 12 ~
V ( x , - x~) where V ( x , - x~) is the potential
energy of interaction ()f a pair of molecules. The mean potential energy would
then evidently depend on the two-fold correlations in the position distribution
of the molecules.
From the above example, we see that the combinations of events t h a t are
i m p o r t a n t may depend p a r t l y on what in of interest to us, and partly on the
physical conditions of the l)roblem.
When the distribution of a r b i t r a r y combinations of compound events is
uniform so that the distribution possesses the property of statistical independence, (as is true in our sequence of angles) then one can, by the well knowa
techniques of the t h e o r y of probability, deduce a n u m b e r of i m p o r t a n t results.
For example, let us divide the basic circle into two equal parts, denoting by A,
a case in which the angle, is in the first part, and B. a case in which it is in the
second part. This division provides us with a model of a pro(.ess leading to
results analogous to those in a game of coin throws. Evidently, in the long
run, the angles will be uniformly distributed over A and B (except for a set
of 0,, of measure zero). Moreover, a simple calculation shows that various
combinations of events (e.g. combinations in which the )~-th event is separated from the next b y :~ steps) also have a uniforln distribution. Thus we
reproduce another p r o p e r t y of the game of coin throws. (For example, the
probability t h a t a head will be followed :~ steps later by a hea(l is the same
as t h a t it will be followed b y a tail.)
I t is clear that we can now calculate the probabilities of c o m p o u n d events
in the same way as is done with the theory of probability. Thus the probability of a combination of )~ events in which lhe angle is found n, times
m A and ~, times in B (with ~ = *L, =- ~ ) is just I1"~ (!~"n!/~'t~!)~[2, the
well-known binomial distribution. If n, n, and ~% are large, one m a y obtain
the gaussian approximation IV(n, b ] ) ~ exp [--(~/2)(b]) 2] where 5 ] - : &~/n and
dn is the deviation of ~ / ~ from its probability (which in {/.
This result means t h a t if we followecl the t r a j e c t o r y for a n y 00 (except for
a set of measure zero), we would in the long run obtain a gaussian distribution

T I t E G~;N]~RAL S T A T I S T I C A L P R O B L E M I N P14YSICS A N D T t I E T H E O R Y O F l a R O B A B I L I T Y

1039

in 5n/n. Thus, we see how the gaussian distribution which is so t y p i c a l of


the theory of probability, can come out of a d e t e r m i n a t e sequence (*).
A f u r t h e r step is to deduce BernoullPs t h e o r e m . F o r as n becomes l a r g e r
a n d larger it is evident t h a t one will e v e n t u a l l y o b t a i n ~ negligible n u m b e r
of eases in which 5/ is g r e a t e r t h a n some specified n u m b e r , e. This me~ns
t h a t for a typical t r a j e c t o r y , p r a c t i c a l l y all choice of a sample of n t e r m s will
lead to nA/n v e r y close to 89
Finally, let us note t h a t these results can be generalized, b y showing t h a t
t h e y also follow f r o m t y p e s of l~ws i n t e r m e d i a t e between d e t e r m i n a t e a n d probabilistic. This can be done b y m e t h o d s analogous to those a p p e a r i n g in
Secs. 6 a n d 7, b u t we shall not c a r r y out such a d e m o n s t r a t i o n in detail here.

9. -

A Criterion

for Randomness.

I t can be seen t h a t in the previous section we h a v e justified the calculus


of probabilities, as a meaJns of t r e a t i n g the statistical properties developed in l
the long r u n b y our sequence, 0~ - - K~Oo. I n doing this, we h a v e also g i v e n
the elements with which one can o b t a i n a criterion for randomness.
I t is our conclusion t h a t the p r o p e r t y of statistical independence (i.e., absence of correlations) furnishes a n a t u r a l a n d useful criterion for randomness.
A t this point, we m u s t emphasize~ however~ t h a t statistical independence
with regard to all possible combinations of events will occur only in r a t h e r
special problems (such a s p e r h a p s in the g a m e s of coin throws a n d dice). More
generally, we m a y expect approximate a n d partial statistical independence.
The degree of r a n d o m n e s s m a y t h e n be a p p r o x i m a t e , in the sense t h a t the
K-fold correlations v a n i s h only in some degree of a p p r o x i m a t i o n , a n d not
exactly. I t m a y be pai'tial in t h a t some of the K-fold correlations v a n i s h ,
b u t not all of them. As we h a v e pointed out in Sec. 8, the i m p o r t a n t p r o b l e m
is then, w h e t h e r the kinds of correlations t h a t h a v e physicM significance in
a given problen b a n d those t h t t are of interest to us in c e r t a i n applications,
will be small enough to be neglected. I f this is so, we m a y say t h a t the distribution is <(r a n d o m in the c o n t e x t u n d e r discussion )).
As a justification for our proposed criterion for r a n d o m n e s s , in t e r m s of
absence of correlations, we first point out t h a t this is effectively t h e p r a c t i c a l
criterion most ]requently used in applications in physics, (e.g., s t a t i s t i c a l
mechanics) (+) as well as in other applications of statistics (e.g., p r o b l e m s
(*) The gaussian approximation has already been obtained from determinate sequences in other ways. See, for example, M. KATZ and H. STEIN~IAIJS:Studia Mathewatictt, 6, 59 (1936).
(+) In Section 10, we shall give an example of how this definition of randomness
is used in statistical mechanics, in connection with the development of (( random phases ;.

of distributions of various kinds of members in a certain population). Secondly,


we point out t h a t it contains what is really essential in the definition given b y
yon MISES (*). For if a sequence has the property of statistical independence
of its members, then as we have pointed in See. 8, the gaussian law follows,
so t h a t most samples will tend to give good representations to the relative
frequency distribution in the sequence as a whole.
The criterion t h a t we have given for randomness avoids the problems
appearing in the definition given by vo~ M~SES. First of all, we do not define
randomness in terms of irregularity or lawlessness and thus, we do not become
involved in the very probably unsolvable problem of defining w h a t irregularity
and lawlessness mean. Instead, we re~'ard randomness as a speeia.1 manifestation of law, a manifestation which can appear however only in the context
of a statistical aggregate and one of whose characteristic aspects is statistical
independeuce. In this regard, we adopt a position similar to t h a t of WALl> (+)
,and various other members (*) of the school of y o n M~sEs, who a d o p t as a
criterion for randomness a certain class of laws suggested b y the impossibility
of , b e a t i n g ~, a gambling game. Ore" eriterion differs, however, from theirs,
in two aspects: (A) It is much easier to apply in practice in a wide range of
statistical problems, and indeed corresponds to just the practical definition
t h a t is most frequently used. (B) The criteria for randonmess are not chosen
completely a-priori or by the simple transfer to all possible problems of the
criteria that m a y Ire significant in a given problem (e.g., gambling games).
Instead, the criteria are determined either by the a c t u a l physical situation
or by the statistical problems t h a t we wish to treat, beeause the kinds of
correlations that are signifieaut m a y depend on the context in which the theory
is applied.
W i t h regard to tile question of irregularity or law-lessness, it <~I)pears to
us tilat this is not ,an essential feature of the problem. For example, in the case
of throws of a coin, regular orders, such as t h a t of ten heads in a row will
certainly be quite rare. But a n y other speciJied order, however irregular it
m a y seem to us, will be equally rare. Thus, the rarity of <,regularity ~ is not
so much a property of the sequenee itself as of our own tendency to pick out
a few special orders a.nd <.all them <<regular ~>, while the overwhelming m u j o r i t y
of orders are relegated to the very broad category of <,irregular >>. IIence,
a n y distribution that fluctuates over a wide range of possible orders will inevitably seem irregular to us most of the time.
Before ending this section, we shall make a fe~ remarks on the relationship between randomness and chance. Now, as we have seen, randomness

(*) ,See ~ect.ion 4.


t+) See Seeti(>n 4.

TH]~ G~:NERAL S T A T I S T I C A L PROBL]~M IN P H Y S I C S A N D T I t E T H E O R Y OF P R O B A B I L I T Y

10~1

is a statistical p r o p e r t y of a given system. I t m a y evidently come from the


effects of external disturbances t h a t already h a v e a 4egree of randomness in
them, and it m a y also arise from purely internal causes (as happened with our
sequence 0~ ~- K~Oo); b u t most generally it arises because of some combination
of these two kinds of causes. The assumption t h a t statistical theories must
always be restricted to the calculation of one kind of probability in terms of
another is then equivalent to neglecting the possibility t h a t a p a r t of the
randomness, can arise internally in a system, w i t h o u t the action of some
external causes t h a t already h a v e in t h e m some degree of randomness.
On the other hand, as we saw in Sec. 2, chance refers to a relationship in
which the events in a given context depend contingently on (( independent ))
fluctuating causal factors lying outside the context of interest. Thus, while
a r a n d o m distribution can develop of its own accord in a given context, chance
can b y its v e r y definition refer only to the effects of causes t h a t lie outside
the context under discussion.
There is, nevertheless, a close relationship between randomness and chance.
F o r in those cases of chance fluctuations t h a t are most typical (e.g., coin
throws), the external contingencies (e.g.~ the motions of the h a n d Chat throws
the coin) on which the results depend, have a range of fluctuation within which
the distribution is effectively random. Such cases, correspond to the category
of (~pure chance ~ discussed in See. 2. There m a y , however, exist a m o r e
general kind of dependence in fluctuating causal factors (*) t h a t are not entirely r a n d o m in their distribution. This dependence still leads to a kind of
chance, b u t it is not pure chance. I n such a ease, the calculus of probabilities
evidently would be a t best an approximation. Here, then, we have ~n additional example of the general statistical problem; for it is evidently of interest
in such cases to study the question of how good is the degree of a p p r o x i m a t i o n
provided b y the theory of probability, as well as perhaps to seek a more general
calculus t h a t might apply even when we do not h a v e pure chance.

10. -

Further

Illustration

in Terms

of Random

Phases.

In this section, we shall apply some of the ideas suggested in the previous
sections to a problem t h a t has widespread application in statistical mechanics;
namely t h a t of the development of r a n d o m phases in a system of oscillators
or rotators.
The simplest example is the problem of r a n d o m phases with a system of
harmonic oscillators. L e t x~ be the coordinate of a given oscillator, y~ the

(*) This possibility has also been discussed in Section 2.

1047

1~, BOHM a l l d -~v, ~CII[}TZEtr

corresponding canonically conjugate m o m e n t u m . The equations of motion of


such a set of oscillators (which arc, in g'eneral, coupled), ean be derived from
tile hamiltonian
(1t).1)

H:

~ ~

(aid/dl;

-'- b ~ / ' d L

- c~.r,,/. ) .

i.i

As is well known, there exists a linear iransfornmtion, to the normal coordinates (~), q~ and to the eonjuffate momenta, pC, such t h a t the hamiltonian
reduces to a sum of squares, eaoh equivalent to a separate harmonic oscillator.
not coupled to the others.

The equations of motion (0;


(10.3)

~, - -

; H I l l > ,. [,,

(B,/A,)q,

o,

;H,;q,)

~,, -

(lt,/A,)p,

then lead to
()

so that each oscillator lms angtlIar fre(lueney o~,- ( t L / A , ) ~.


can then conveniently be expressed in the oomplex form
(10.4)

q, -- ip, -

The soluiions

K , e x p [io>,ll,

where

K, is an a r b i t r a r y complex numl)er, which ('~n be written as

(10.5)

1(~ -: .J, ex I) 12i7,,, I.

(I0.6)

q, -- ip,

-~ ( J , ) i e x p [i,),l -~ q~.,J .

Evidently J, is proportional to the euerg'y of the j-th oscillator and q%,


is a constant of the motion eorresp(mdin~' to t h e initial vatue of l~he t)hase
an~'le of the )-th oscillator.
.ks a fun('tion of the time. the l)hase au~le is then ~iven h y

(10.7)

(f,

--- (fo,

(o,t

As we shall s h o w in this section, if the o, are ineommensm'al)te (as they


usually are in most applications) the phases, (f, will in time become r a n d o m
relative to each other.
The problem of r a n d o m phases arises in a g'reat m a n y i m p o r t a n t problems

(~i) H. (;. ('.OLt)STe.~: ('las,~'ie~al _]teeka~H~.s (I(.t52}, Chap. ]tJ.

"~I]E GENERAL STATISTICAL PROBLEM IN PHYSICS AND TIlE THEORY OF PROBABILITY

10~3

in statistical mechanics, e.g., t h e r m a l v i b r a t i o n of a solid, the t h e r m a l excit a t i o n of r o t a t i n g molecules, motions which can be solved in t e r m s of angle
~nd action variables a n d m o s t i m p o r t a n t of all, in the foundations of q u a n t u m
statistical mechanics (~). The a p p l i c a t i o n of the notion of r a n d o m phases
in the f o u n d a t i o n s of q u a n t u m statistical mechanics, we shall discuss in more
detail a t the end of this section.
I n order to illustrate in a simple w a y just how the phases b e c o m e r a n d o m (*),
let us begin b y considering the case of two variables. Now, in the p r o b l e m
t h a t we are t r e a t i n g here, we h a v e the new feature, not present in the t r e a t m e n t
given in sections 7 a n d 8, t h a t instead of a sequence of discrete values, we
h a v e to deal with a continuous function of the t i m e t. F o r this reason, we shall
replace the s u m over the N t e r m s of the sequence a p p e a r i n g in eq. (8.7) b y
an integral over the time, representing the m e a n value of an arbitra~T function,
] ( ~ ( t ) , f2(t)). (The use of the continuous t i m e a v e r a g e in this case corresponds
to the a c t u a l practice in the definition of physical quantities in statistical
mechanics (~3)). We t h e n obtain as ~ criterion for a uniform distribution t h a t
the following relationship m u s t hold i n d e p e n d e n t l y of initial conditions (except
possibly for a set of m e a s u r e zero):
ff
(10.8)

2~ 2:z

Tli~ ( 1 / T ! t '(~O 1 ($), ~2(1))d~ = I /1 '(~01, ~2 ) d~l d ~ 2 .


o
o o

The limits on ~ , ~2 are t a k e n f r o m zero to 2~, because as can be seen froul


(10.6}, the s y s t e m comes to the s a m e physical s t a t e when ~ or9~2 change b y 2z.
This means also t h a t all physically significant functions of q~l and 9~2 can be
expressed as periodic functions with period 2z~ or t h a t
(10.9)

J(gh, q~2) = ~ ]~J exp [i(n~q~x -F nj~2)],


i,i

n~ a n d n~ being integers.
Now, writing ~ = 9~o~ w~t, ~0~ = 9~o.2 w2t we obtain
(10,10)

](~01~2) = ~ ]iJ exp [i(ni~ool ~ n~9~o~)-F i(niC01 ~- ~/JO)2)t] 9


i,J

I n s e r t i n g (10.10) in (10.8), we note t h a t if ~ol a n d co2 are not e o m m e n s u r (~) See, for example, R. (3. TOLMA~,~: The Principles o] Statistival Mech,,~aic~, (Oxford,
19~0), Chap. IX.
(*) The development of random phases is already well known (see, for example,
reference (~)). However, we shall give a brief sketch of a proof here which is convenient
~or illustrating the concepts involved.
(la) See, for example, reference (~).

]044

I).

BOII'~I

;111(|

XV.

>('lti:'l'Zt-.l~

able e v e r y term on the left h a n d exeept [o. will i n t e g r a t e out to zero as t h e


time T a p p r o a c h e s infinity. On the right h a n d side, we see f r o m (10.9) t h a t
likewise only /0o will survive i n t e g r a t i o n o v e r qOl a n d ~2. T h u s we d e m o n s t r a t e
t h a t the criterion for e q u i d i s t r i b u t i o n is satisfied i n d e p e n d e n t l y of initial conditions (*) (note t h a t in this p r o b l e m , there is no e x c e p t i o n a t all, not even
of a set of m e a s u r e zero). We conclude then t h a t t h e r e exists a u n i f o r m p r o b a b i l i t y d e n s i t y t h a t the p h a s e point will ()('eul)Y a n y specified region of the
configuration space of ~ , ~ .
For a more z e n e r a l ease. in which there does not e x i s t a u n i f o r m distrib u t i o n , eq. (10.S) should be replaced b y
/

(lO. l l )

l..

2.'r 2,7

at

T >co
ti

i[

-_

ii

~1

where P(qG, ~ ) is lhe p r o b a b i l i t y density in the eonfiguratiol~ space of (/?1. (~:~


a n d where the e q u a t i o n n m s t hoht i n d e p e n d e n t l y of initial conditions (except
for a possible set of m e a s u r e zero), if the motion is Io be such as to p e r m i t the
(lelluition of a l)rol)ability.
To show in more detail how the c q u i d i s t r i b u t i o n eomes a b o u t ill t h e case
t h a t we h a v e t r e a t e d , let us e(msider the ?q, g2 plane as shown in Fig. 5, where
g~. %_, ~'o only over their l)hysieally significant
rang'e of 2n. (%nsider a n initial point, P, with
~F~-=
9COl a n d g ~ - ~c0e. T h e n with t h e passage of
0 Lv 0
time, the phase poinl will m o v e until it finally
reaches lhe point, t,l, where one of t h e variables.
gl or ~e is equal to 2n (in t h e d i a g r a m ; it is of,, t h a t
is equal to 2n). At this point, P m u s t decrease
0<
a b r u p t l y 1)y 2~ a n d j m n p to the point, r
Then
Fi~..~.
the p o i n t m o v e s until it reaches Q". t h e n j u m p s to
O'" moves to Q~", etc. The t h e o r e m of equidistrib u t t o n t h e n shows that in time the phase point will fill the a b o v e region equidensely. E v i d e n t l y this c a n h a p p e n only if oq a n d ~'~e are i n e o m m e n s u r a b l e .
for if t h e y are eommensural)le, the l)hase point will describe a elosed curve.
I n the Slmee of the p, a n d the q, we o b t a i n the well-known Lissajous figures.
which fill lhe spaee densely, b u t n o t in general equi-densely.
The g e n e r a l i z a t i o n to a n y u m n b e r of variables is quite s t r a i g h t - f o r w a r d :
a n d we need only state the result t h a t if all the <,~ are i n c o m m e n s u r a b l e , then

(*) Thi~ re~uli i, esuenrially a l)roof of the qua,~i-ergodie theorem for t;he specia, t

problem that we ar~ treating. {See ~eetion 6).

TH]~ G E N E R A L

S T A T I S T I C A L P R O B L E M 'iN P H Y S I C S A N D T H E T H E O R Y

OF PROBABILITY

1045

the distribution will h a v e a uniform p r o b a b i l i t y density in the configuration


space (~0i...~). Eqs. (10.8) a n d (10.11) t h e n r e p r e s e n t a s t a t e m e n t of the
f a c t t h a t the t i m e a v e r a g e b e h a v i o r of a single s y s t e m can be represented b y
a suituble a v e r a g e over angles of an ensemble of systems, t a k e n a t a single
i n s t a n t of time, t. Thus, the use of statistical ensembles (which are essentially
the s a m e as the collectives of v o ~ MzSES) to represent the statistical properties
of a single s y s t e m is justified. The u n i f o r m i t y of the distribution in the configuration space evidently implies t h a t P(~I, ~02 ... ~%) ~ P ( ~ ) P @ ~ ) ... P(~%) so
t h a t the phase angles of t h e different oscillators are uncorrelated, a n d therefore statistically independent. Physicists h a v e gradually come to refer to a
s y s t e m of variables t h a t b e h a v e s in this w a y as (( r a n d o m >). (Thus, we obtain
a justification of the definition of r a n d o m n e s s suggested in Sec. 9, a t least a~
f a r as applications in physics are concerned).
As we h a v e a l r e a d y pointed out, the p r o b l e m of r a n d o m phases has u wide
application in statistical mechanics. This p r o b l e m is p a r t i c u l a r l y i m p o r t a n t
in q u a n t u m theory, because it enters as an essential p a r t of the f o u n d a t i o n s
of q u a n t u m statistical mechanics. To see how this comes about, we note t h a .
as is well known, the q u a n t u m s t a t e of a s y s t e m is defined b y its w a v e functiont
~,(x~ ... x,.) where x~ ... x~ are the coordinates of the system. An a r b i t r a r y
w a v e function can be expressed in terms of the series of eigenfunctions y J , , ( x ~ . . . x , )
of the H a m i l t o n i a n operator, corresponding to eigenvalues, E ~ . T h u s ,

(lO.12)

yJ = ~ C~nv2E~ ( x l ... Xn) exp [-- i E ~ t l h ] .


n

(We are assuming the s y s t e m to be in a b o u n d e d region of space, so that, the


eigen-values are discrete, a n d we are assuming t h a t there is no degeneracy.
The case of degeneracy requires a more extensive t r e a t m e n t , which we shall
not give here.)
W e then write
(10.13)

C~n - a,~ exp [i(Po,~] ,

obtaining

(10.14)

~f = ~ anyJB~exp [i(p~],
~t

with
(10.15)

~f. ~ q~o. - - E . t / h .

C o m p a r i n g (10.14) a n d (10.15) with (10.6) a n d (10.7) we see t h a t f o r m a l l y

1046

1). BOtIM all(] w. S('H(:TZEP

speaking, the wave function acts like an infinite collection of tmrmonic oscillators, one for each eiffenfunction of the hamiltonian. This is not surprising,
since the Sehr6dinger equation is linear ill V, and as is well known a wave
field satisfying a linear partial differential equation can be expressed in a
series of terms each oseillating with its own characteristic freqnency.
Now we note t h a t in general, the angular frequencies (,),~ = E,~t/D of oscillation of the wave function in differeni s t a t i o n a r y slates, will not be ('oremensurable (*). I n the h)ng run. we shall therefore have a uniform distribution in the eonliguration spa(.e of the phases, and the l)hases will be random
relative to each other.
Finally. we remark t h a t the problem of r a n d o m phases, is formally of the
same lype as the one treated by POIXCAR~ in connection with the distribution
(ff the phase an~les of a swarm of asteroids in neighbourin~" orbits (see Section 8).

11.

Conclusion.

In this paper, we have shown that in a large numl)er of stalistieal problems


arising in physics, il is either ,~dvantageons or necessary to go outside the
frameworl< of concepts thai ~t)pea|' in t h e lheory of probability. Among such
lwoMems are the following:
(1) The calculation of the lar~'e-sc~le statistical and t h e r m o d y n a m i c properties of an aggregate (,4retaining" a large numbers of molecules.
(2) The t r e a t m e n t (ff non-equilibrium slales of such an aggrep~'ate.
(3) The demonstration t h a t a given kind of sequenee contains statisti('al
relationships t h a t permit its long-run behavior l<) he lreated by
means of the ('aleulus of pr(>l>~bilities.
(4) The estimate of error in the lheory of probabilily.
(5) The t r e a t m e n t of ai)proximate and i)artial rand(mmess.
(6) The t r e a t m e n t of ehaneo lluetuali(ms ori~'inating in coniingeneies
that are nol totally independent of what happens in the context
under diseussi()n.
An inspection of the above lisl would suggest thai the l~roblems of statistical physics in whi(,h the theory of probability (l(~es not al)l)ly are perlmps

(*) They at(' comiuensura|)le for th( ~ ol~e-(limen,~ional h;irm(,ni(' os~.illator, but. nor
Jn mosl other ~'enerM problenl~.

T H E [TcEN~3RAL S T A T I S T I C A L P R O B L E } [ IN P H Y S I C S A N D T I I E T H E O R Y

OF P R O B A B I L I T Y

104,7

even more i m p o r t a n t t h a n those in which it does apply. A systematic a t t a c k


in statistical problems going outside the scope of the t h e o r y of probability,
coupled with an effort to develop new formal systems t h a t are more general
t h a n t h a t of the calculus of probabilities, would therefore be likely to produce
fruitful results.
r

We wish to t h a n k Professor *~[AKIO SCH6~BEPG, Dr. MARIO BU~GE a n d


J. A. MEYEn for muny stimulating and instructive discussions, which played
a v e r y i m p o r t a n t role in the clarification of our ideas.

You might also like