Chapter 6

Foundations of Probability Theory and Statistical Mechanics

EDWIN T,. JAYNES

Department of Physics, Washington en i \ -ersity St. Louis, Missou ri

1. What Makes Theories Gro w P

Scientific theories are invented and cared fur bypeople : and so have the properties of any other human institution - vigorous gnw,;thwhcl1 all the factors are right ; stagnation, decac.encc, and even retrograde progress when they are not. And the factors that determine which it will be are s ldom the ones (such as the state of experimental or mathematical techniques) that one might at first expect. Among factors that have ~ ,>·m xl, historically, to he more important art: practical considerations, acciden ts of birth or personali t y uf i ndiv idua l people: and above all, the gi,~ncral philosophical dim ate in which the scientist lives, which determines whet her efforts in a certain direction wi II be a pproved or deprecated by the scientific C0I1)!11unity as a whole.

Huwev«r n.ucl; the" pure') scientist may clcplc\re it, the t<h't remaintha t rni litary or engineering applica tions of science ha '·C. over and U\TI" again, provi led the impetus with ut which a field would have remained stagnant. \\"e know, for example, that .-\HCHD1EIlE:';' w» k: in mcchunirs was at till' fun-frun t of efforts to defend. ~~,T"11·lbt' :lg;lin;;t tl n- l{nrnal1.,";': and that l.\X.\IFOIU)·:--; experiments \\'11i(1I l,~d cvcnt uu llv t o the tir-t 1.1\\ of thermodvnu mics were performed in the cour.«: of btlring c.mnon. The dcvdopll1vntof microvva.vc theory and tccl.uiqucs during \\"orld \\"ar l l, and the present high Icvel of activity in plasma phvsi 'S are n1uH' reecn t ('X,;lIH pi, 'S of til is kind qf in t -ract ion ; .md it 1-; ( 11·;t r t h.u t IH' pa ..... l decade of unprecedented ad \'<m~'( S in sol idst.a tc plly ..... i1 '"' i." 111 It ~·ntlrcly unrelated to commercial a pplica t ions, part icularly in electronics.

Another f;l .. tOT" more important historically hut probably not today, is simplv a m.it.ter of chance. Often, the clc\ dnpu:'I·nt of a field of knowk<lgl' has been dcpendcn t 011 neither mattr-rs of l~,)gic nor practical applications. The peculiar \'i::~i\ .\1. or blindnc-.s, of ind! vidual persons. can

be decisive for the Idilirdkm a field takes, and th'!.! views of one man COm persist for centurieswhether r.~ght or \.vrong. It scorns incredible to. ~JS today that the views of Aristotle and Ptolemy could have dominated tho-ught in mechanics audastronomy for arnillenlumjun til GAr. n.no and others pointed out that \1iC are all surrounded daily by 'adual evidence to the: contrary ;U.:rl.{~ equally incredible that, alU~lou.gl~ thermometers (or ra ther, thermoscr pes) were made by CAtn .. E,Q before f GOO; it rcqul red another i 60 years before the distiucticnbetwcentern perature and heat was clearly recogniz d, by JQSEl~H BtAcr. (Even here, however, the practical applicaticns were neverout of sigh[ ~ for CALIlEO"S dUB'1-lllo:scapes were im medi; tely used by his u)Heab ues in til medical school at Padua f'm:- ,diagnosing fever; and JO'~n~PH .BLACI~~S prize pupil was named _Jt\:'[ES \\,F_\IT). In an age averse toany specutntiou, F'R,F.SNI!L was nevertheless able, thr-ougll pure speculation about elastic \·rb~atlsms,. 'to fh:l(l tln correct rnathcmatical relations g vcrning the prnpagatlunj reflection, and refraction '0[" pt)~ari"zr!"¢t~i,gh;t a half-~,e.ntury before .1\'1AA\~iT~U:S. electromagnetic theory: while at the same time the blindness of a Icw others dckl.y(~d recognition Q[ Hre Iirst law of the:rnmdynamlcs for fOlrtjt y.eu]"s.

01' far _grc'at'er imporeance than these. however, is the general ahilosopbical elimat that determines. th • n officinl" views and. standards of -value of the scientific. .commuuity, and the degree .of pressure toward cOl:ifo:rrflity with those iews that the community exerts. on those with a. tendency to O:f[-gi]l.ality. The reaHty and effectiven ,·S,._; of this blctor are no less greatbeaus.c. by it$, very nature, indlvklual cases are more difficult to, document; its cHccts a in HIe Iarge " arc .,' aSliy seen 0..' ~oHows.

If you make a ligt. of what you regard as the major advances in ph ysical tit eory throughout the history of science. look up the date of eacl "and plot 0. histogram shm':~ng their distribution by .dcca,ck:s, you will be strneki mmediatelv bv the faC' that advances in UW(If'l d(} not

J .J ..F

take place ilildcpe,ndefll,uy .;:-wd r;:;~ml~.)m~y; they b,l'~[~ n :sh:9nU h~nd·c-nc,Y

t appear in small elise clush.'[s; ~rac('(I, ahou sixt y to SeYll'nty year:::; apart. \rhJ..~ \'i.'C are oh~er\zing hN'4' is the result of .m inkr'i;.':=.'lin~ .~od.~l phcnomenon ; thi:\ prrss,urc tnw;u"fl (·,~nfonnity \\,"lth certain IiHki~lUy proclaimed ~·iew:s.,a.:n:d aW~'I!r (n)wl. fn . .!c ::ii)e~uL';l!tiun, is subject t'l'~1 large periodic fluctuation. The last' h:n:..' c· rvclcs can be Iollowed very ,en:, iilly, am] the· pr-t,::s:s.U!'"l: maxima andminlma (all be dated rather l)red~('~:Y'

At the point (J,r the cycl ' where tlrepressure is. least! condil iU:t1S1~U'e ideal Jor the creation of new theories. At these times, no one f~~c]5 'very sure jinst where t.b truth lies, and so free speculation isenconraged. New ideas of ;)lJlj" kind are welcomed, fl.l1d judged as. ~\U thcoriesO<~Jsll~ t to be ju(tg(·d; on ,g"rollnds of their lOSl!{,"~ll consistency and agre'CI'l'II'N1-t with 'ex peri rn en t. 0 f course, \Vt .. are on ly lnm}:~ 11; and. so we alsoha ve a. st rons-

preference for theories which have a beautiful simplicity of concept. However, as stressed 1,:,- many thinkers from Occ.vxi to EI?>:'STEIN, this instinct seldom leads us away from the truth, rind usually leads us toward it.

Eventually. one 0.1 these theories proves to be so much Ill-ore successfun than its competitors that, in a remarkably short time the pressure starts rising, all effective opposition ceases, and only one voice is hear-d. A well-known human frnil ty - overeagernes-. of the fresh convert - rides rough-shod over all lingering doubts, and the successful theory hardens into an unassailable official dogma, whose absolute, universal, and final validity is proclaimed independently of the factual evidence that led to it. \Ve have then reached the peak of the pressure cycle; a High. Pri -sthood arises whose members believe very sincerely that they are, at last, in possession of Absolute Truth, and this gives them the right and duty to combat errors of opinion with ;111 the forces at their command. Exactly the same attitude was responsible, in still earlier times, for the Spanish Inquisition and the burning of witches

At times of a pressure maximum, all free exercise of the imagination is frowned upon, and if one persists, severely punished. New ideas are judged, nut on grounds of logic or fact, but on grounds of ideological conformity with the official dogma. To openly advocate ideas which do not conform is to be branded a crackpot and to place one's professional career in jeopardy; and very few have the courage to do this. Those who are students at such a time are taught only one view; and they miss out on the give and take, the argument and rational counter-argument, which is an essential ingredient in scientific progress. A tragic result is that many fine talents are wasted, through the misfortune of being born at the wrong time,

This high-pressure phase starts to break up when new facts art' discovered, 'which clearly contradict the officialdogma. As soon as one such tact is known, then we arc no longer sure just what the range of validity of the official theory is; and we usually 1!(\ \'C, t'lh)llgb dues bv then $0 that additional disconcerting facts can be found without ditIiculty. TIle voice of the High Priests fades, and soon \\"0 have agdin reached a pressure minimum, in which nobocl _}r feels very sure where the truth lies and new suggestions are again given a fair bearillg, S0 that creation of n. w theories is again so .iallv possible.

Let us trace a fewcycles of this pressure fluctuation (see Fig, O· The pressure minimum that occurred at the end of the eighteenth century is now known as the" Age of Reason ".

During a fairly short period many important advances in physical theory were made by such persons as LAPL-\CE, L:\GRA;\GE; LA \·OISIER, and FOUHIER. Then a pressure maximum occurred in the first half of the

nineteenth century, which is well d scribed in some thermodynamics textbooks, particularly that of EpSTEIX [1]. This period of hostility toward free speculation seems to have been brought about, in part, by the collapse of SCHElLH:C'S iVatlirphilosophie, and its chief effect was to del a) recognition of the first law of thermodj namics for sev eral decades, .\:::; already noted. FRES:\EL was one of thcvery few physicists who escaped this influence sufficiently to make important ad anCC3 in theory.

Another pre, . UH' minimum was r a .h d during the third quarter of the nineteenth century, when a new spurt of advances took place in a period of only filtc('nycars (1855-1070), in the hands of JL\X\VELL, K£L\T~. HEinz, HEUIHOLTZ, CLAUSI'·. BOLTz:..rA~i\J and several

\

\.

\

.2

1850

19S0'

1900

Fig. 1. ~OTl'lC recent Iluctua tious in social pressure in science

others. During this short period tit .rmodynamics, el ictroruagnctic rheorv, and kinetic theory were developed nearly to their present Iorm ; hut the very su Tess of these effort' led to another of the inevitable pressure maxima, which we recognize as being in full Ilower in til . period 100;-11)00. One of the tragedies {at l'ast from tile standpoint of physics) caused by thi: was the virtual loss of til talents of P I:\C!\I(I~.\Vhik hi' . ntributions to physical theory arc onsidcrable, still they are hardly commensurate with what we know of his enormous abili tics. This was recocniz d and explained by I:.., T. BELt [2j in these words: «He had the misfortune to be in his prime just \\"1H'11 physics had reached nne of its recurrent periods of senilitv." The orfirial dogma at that tillll' was that all til' i':1cb of physic.; are to he explained in terms of .:\('\.\'tullian l11t' .hanics: particu larly t ha t of pan ic Iv:-i in t ·'r;~d i ng thmu.gl\ rcn tral flll·'·~'S. Herculean effurts were made to explain away ~L\,\WL~LI.':-; electronWf,.l1 tic theory bv more ancl more complicated mechanical models of rhc ether ~- cO()rts which remind us \' ·ry much of the earlier ~ingleminded insistence thatall the facts of astronomy must be cxpl.iu ed by adding mort: and more Ptolemaic epicycles.

An intcre ting man i lcstat ion toward the end of this period was the rise of the school of "Encr~cticsJ'. championed by :\I.\CH and OSTWALD, whi .h r pre's nts an early attempt ,i th p sitivist philosophy to limit the scope of sci '11 -e, This sehoul held that, to usc modern t f1llinolugy. the atom was not an "observable", and that physical theories should not, th refer , make usc of the concept. The dcmis of this school was

brought 'O-b.QtJt t~pidly by PG'(~J'<lI;;t~i quanritative rneusurements on fill;) Brownianmotion, which verified E.l'~'STEI~·S predi~;tio:n~ and provided run experimental value for A.VOGADRO'S number ..

Tl l t "G J ~I .,\ ,f"PI '1 b' .] '1 ~: t. . -

llfl: .~S _ '. r .. =p O<~CB /11.g: e 0 ~Llet)ty' .' r~Ug ~J "~~ pu nt l~lI'r t ~ ~e cnsu.tiY.f;J) r ~tB~

sure n'1h~hHI,UI"~~ Iasted f];1()\ma.bont 19:10 to19JOI 'and produced our present general realifivity and (l~,a[!lturntheode:::L~~gain, the spectacular :suuct£.~ -of: Hae l"a tter - literally tbousands 01' quanti tat i vel yccrrect predictions which could not b~~!natchcd by ~m,r co].-;npetin:g th~~~~ri" - brolilgllt ·about the. inevitable pressure rise, and for twenty- n." "}' years .( [9) 5 - '19:(0). theoretlcal phy:-;.ic:~ \"!;.';;:LS/ p;,wal.y.lcd bJ' {)ofrc of thi(~ mcstintensc .and pro-

. -

~m"l.g-edhigl)-priCS.:St'l!J.t'~ periods yet recorded, During this pedod the of-

ficial dogma has been that an of physic'S i:§ now to be explained by ])lft~~tribing inlh~1 ;lud Imal ~t:),tc vectors in fl Hilbert ~';lp!~!..'e" and ,coluputillg transition matrix el 'f:J.1ents bf:itt:'ween them. ;A ny artempt to find ~1 J;l~:()t~· detailed (k"cl':iption than this stood in conflict \\""lth the onida~ ].d.eOoillogy,

·afld\,\.'~s quickly suppIlress&d. w[thou:t ~lny aU~~rnpt to:' e~hhJbit alogic:l! inconsistency or a Gon:Wct xvi.th experirnent ; this time, ~"R fewindividuul cases can lie CkH .. .umented [;3].

Ther.¢ nrc now !11..a;tl.:r signs ·that the I~H');;$snre'htl::' st~tl'tl~d do!;rn a_g~tin ; several of Hl.e supposedly universal principles of quantum theQry have been ceufronted witbnew facts, or new investigation»: which make tI:i unsure ot their q-X1U;t r2l!1lg~"O:f varidit.r. In. particular I nne of the Iundamental 'tasks of any theory ]s" to prescribe the CliillS", uf phy';)ical stares allowed by Salute" Iu )b.XWEU/S electrorrl'~lgi;!.ctic theorv, for exarnpk-, -any mathematieal soiutlon of ThL<'tXWELL'§' equations is ]~d~ to represent apossible physical state. which couM in rdllt~pk he Pi!!'OdU!f~d fn the labQrah"ry, [nqu8cllftH;ntll£:O""ry. we were 't:;lug.ht for many yearf., thnJ the 'Cl-a$~ or possible physical states is in jrt .corre.spoIltlenre \\']tll'baJ~H~on:j of the Schrodiuaer-enuation that are either svmmetric or arrtlsvrnntcrrio

~ '~1 ~ ~

uncl~\r ·:per.IHUit'Qitlg.ris 'of iderttica.l pa:rtid't',~. Ol;fr,(()[l!"fldlenc(.l in th~~ uni-

versal va ]:idi tv of this rule ha-s,re[cuth.·, heen sl.KI.k~.m in t\'·'O l'e~I,'~{'ll:<:.-

. , ~

In the ffrst place, study of "parastari . stics" has shown that much more

g;'('"u(":ral types of syl:nm-e.t:ry in corl.fi,g_ura:ti.IHl space OHl .abo be de-scl'~be:d by file ltlotdlirf-:fY ·0'£ ,quantized \1j,'Tt,~efunitjnn$> .;:tnd, these ne\O\" possibili til's a':n~.:. nvt rnj.~tl out byexperimental f'yidencE'; .. SC-Cf..'!UCUy. the superposition principle [wliich mo.)' bCf f{~g;lrded as a COD.i€;((i1,ence ofthe -above-rncnHori.cd rule, al.thoqgh H is usually 'cr!lhideI.~ed itfi· a sti]] more ge-Tlt'ra.l "'~r.ll'.,e} hold~ that, i.f.~:'l~Uld .~~~ are: anytw« possible physicat-states, thl~H auv linear combinarion 1ijJ ~t.11 ¥'.t + ~l~¥'::! is al_st"i" a pfD:ssible. physical state.

- - - ."

But with the appearance o-f superselection rules, we are no Ionger sure.

what the r~m.·ge of validity of the' superposition principtei« ..

The di::;to'l;ery of parity nOiHC(H1:Ser1i;:ahQH was al great psychologi~hI

] k "'!k' 'I ~ -~ ~ , ~ . . ' f 'I .,

$1!~Ot7 r ; @l prlnnp·.ie '''''ld_Cll ~Hu.)l 1J:'~~(!ntaugl1!:t toa getH2'ratlOn '~' PllY::HClst;:;

82

EOWIN T, JAYNES.:

as a universally valid physical law, so firmly established t-hat it could be used to rule out a priori certain theoretical possibilities, such as \VEYL'S twocomponent relativistic wave equation, was found not to be universally valid after all; and again \'vTC are unsure as to itsexact range of validity, and v~rEYL'S equation has been resurrected.

Several. quantum mechanics textbooks assure us that the phcnomenan of spontaneous emission places a fundamental irreducible minimum value on the width of spectral lines. Such statements arc now confronted with the laser, which - in instruments now commercially available, and as simple to operate as a sixty-watt light bulb - produce spectral lines over a million times narrower than the supposedly fundamental limit! Thus, all around the edges of quantum theory \ve sec the familiar kind of crumbling whick historically, has always signalled the. inci picnt breakdown of the theory i tselL

I hasten to add that, of course, none of these developments affects the basic. "hard core" of quantum theory in any way; they show only that certain gratuitous additions to quantum theory (which had, how ever, become very closely associated with the basic theory) "were unsound in the sense that they were not of universal validity, But it is inevitable that, faced these developments, more and morc physicists will ask themselves how many other principles arc destined to crumble a little at the edges, so that they can again be considered valid objects for inquiry; and not articles of faith to be asserted dogmatically for the purpose of discouraging inquiry.

In particular, the unccrtaint y principle has stood for a gencwtioll, barring the way to more detailed descriptions of nature; and yet, with the lesson of parity still fresh in our minds, how can anyone he quite so sure of its universal validity when we note that. to this day, it has never been subjected to even one direct experime-ntal test?

Today, elemcnt.uv particle t heorists are bu-ilv questioning and rve xarnining ail the (oundat ion- of quantum fidel theory, in a WJS that \\'o111d have been l"('ganJl'J. as utter heresy ten years ago; and some have suggested that perhaps the whole apparatus of fields and Hamiltonians ought to be simply abandoned in favor of more abstract appreaches, It would be quite inconsist cnt with the present mood of theoretical physics it \<,:0 failed to question and re-examine ali of tlIl' supposedly sacred principles of quantum theory.

For all these reasons. I think we arc going to see a rapid decrease in pressure in the immediate future, and another period of great theoretical ad varices will again 1Jc social! y possible in perhaps ten years. And I think 'we can predict with confidence that some of the dues which will lead to the next found of ad ··'J.Tlce::i arc to be found in the many suggestions

'already made by dissenters from the Copenhagen theory ~ suggestions which have, tbus fat', been met ,only by sneers a:nd ~U.a.dks .. which no attem pt to studs their real potcntiaut ies.

2. Statistical Mechanica

At this point I se'e that you are looklng about anxiously and wonderIn£) if you are in the r,ight 1'60.J1\; for ·t1lq, announced title of this talk was, "Foundations of Probability Theory and Statistical )ledlr,anksJ,. What has all this to do w"~th statistical mechanics t' IV-eU, I wanted to say a Iew things fir t about genera,] properties. 'Cl,f physical theories because st.,ltis,ti(:;,J!.l mechanics is. in several respects, an exceptional case. Statisnc~d methods exist independently of physical theories, and so statistical mechanics is. subject to additional outside interactions from -other fields. The field of probability and s tatistics is also su bject taperlodic" rluctuations, but they arc not in phase with the Iluctuaticns t:aklng pbcc in physics {they are tight now ata deep pressure minimum] ; ana S9' the history of statistical mechanics is: more complicated,

In particular, statistical ntech .. mics missed out on, the latest pressure minimum in physics, because H colncided witn a pressure maximum .in statistics tthe transition to quantum statistics took place quietly and uneventfully without. any real change ill the basic formalis m of GW:6S, and without any extension of the range of .applicability of the theory, There was no advance in understanding, as witnessed QY the fact that .debates about irreversibilit Y" continue to this da~·I' repeating exactly the SaUlc. arguments and connter-argumcntsthat were used in the time ef Bot1'z~r,\ xx: and the newest and oldest textbooks you can find hardly di (fer at 3 U in their presen ta tion of fundamen tab. In short, sta tist ical mechanics has suffered a; perio-d of stagnation and decadence that makes .it unique in t he recent history ,0,( science ..

~-\. 11";.\'\" era of active work in statistical mechanics started, however.

Alilmt 1/35, 10 phase witlra re".:arutl~n in statistical thought but not at 'first directlv influenced bv it This was caused .. in qatt,b\ practlcal

....... .£: ..

n~eds·; an. understanding of irreversible precesses becamc Increasingty

necessary' in chemical and mechanical ·e~ugiIlieer:i ng as one demanded more. efficient Industrial processing plants, stronger and more reliable materials, and. bigger and better bombs. There .i~ always a movement of sl'ien tific talent into areaswhere gene-rOllS .f:inan .. .ial support is there for "the takine __ Another cause \YfIJ,S the appearance of a 'few people \\;ho were genuinely interested En the field for its 0\\'1] sake ; and p.e-Thaps it helped to. rl.,.>fle'c:t that, since it bad been virtually abandoned for decades, 'one might be able to work ill this' field fref2: of the kind of pressure n.lJted above, which was p.a.ra.1yz·ing rrrative thougllt in other areas of ph~-..:;_ic,:,.

84

j~Jl\n:s T. J ,\ Y:-; liS:

Regardless of the reasons for this renewed activrty, we have now made considerable progress in theoretical treatment of irrcvcr: iblc processes; at [east in t he sense of successful calculation of a number of particular cases. It is all opportune time to ask whether this has been accompanied by any better understanding, and whether tile fOlllldation:-:. of the subject can novv Ill! put into some kind of order, in contrast to tl.c chaos that has persisted for almost a century. I hope to show now that the answer to both of these questions i:; yes; and that recent dcvel )pmerits teach us an important lesson about scientific methodology in generaL

let me state the lesson Iirst, and then illustrate it by examples from statistical mechanics. It is simply this: }"Olt cannot base a general mnibenutiical thea)')' on im/,rccisc(I' d('filled conccbts. YOll can make some. progress tlta! ;:C(iY,' l.u! .'iOO},ICl' or lut.:: tile theor v is bound to dissolve in ambiguities icliicli PJ't"'oli )'01./ from extending it further, Failure to rccognize this fact has .mot her unfort unatc consequence which is, in a practical sense, even more disastrous : L ulcss the COl1ceptllalproblems 0/ a field lunie been dC'arty resolved, you Crill uoi say which matheut«! ical problems are the relevant ones uorlh <"orhug 01.1; ({wi your efforts are more than, likely to be 't('astcd. [ believe that, in this century, thousands of man-years of our fi nest mnt hema t ical talent have been lost through failure to understand this simple principle of methodology; and this remark applies with t'qua'J f( rcc to physics and to statistics.

2.1. BOLTZ:'-L\\ ~ 's Collision Equation

Let us consider some case histories, BOLTZ:\L\~::-: sought to dcscri be the approach to cqui I ihr: um in a gas in terms of the distribu tion I (x, p, f). In his first work, this [unction was ddincd as giving the actual number ,~f particles in variou-. cells of phase space; thus if Ii. denotes the set of PI .irits comprici ng a 1\';'; ion (,f si x-cl illh"n:.;i()l1a 1 pha:-:c space, the 11 umber n f vHti.:k;:; in R I::; to 1)1.' ('()mpHtul Ir.uu

r '( It {3

u J,' .!.I",,,, t) 1 ' .r I p ..

J,

( 1 )

_-1.fter some phvsir al .ll'.::,um -nt s which need nut concern us here, BOLTZ:\L\):)'" concluded tha t tilt: ti me evolution of the gaj should be described by his famous "r olli-iou ~'1uati()n",

~t ~ } [l~ ~ I ':- F"J, of j -0 f (-3]>-' I 2' o U I' =l l') a (.2)

(it ___. 1/1 ex c· P. I

c£ ,., :t

where F:.r, is the cc-componcnt of external [orce acting on a particle; and the right-hand sick represents the effects of collisions in redistributing

particles in phase space, in a way familiar to physicists . As a consequence of this equation, it is easily shown that the quantity

can only decrease (in this equation we in b .. ?gra te over all the accessible phase space); and so BOLTZ\L\);~ sought to identify the quantity

Su:'-'=: ~ kHl)

(4)

with the entropy, making the second la\\- of thermodynamics a conseq ucrice of the dynamical laws, as expressed by (.?), As we know, this was challenged by' ZEID1ELO and Loscrormr who produced two counterexamples, based on time-reversal and on the POI:\CAHE recurrence theorem, showing that Eq. (2) could not possibly be an exact expression of the dynamical equations of motion, aiu! thereby placing the range 0/ validity 0/ Boltznian ns theory in doubt"

At this point! confusion entered the subject; and it has never left it, F'oT 130LTz2lf:\i\~ then retreated from his original position, and said that he did hot intend that I (x, p, t) should represent necessarily the exact number of particles in various regions [indeed. it is clear that the only funr t ion l which has exactly the propcrtv of Eq. (1) is a sum of delta-funct ions : it». p,.t) = Lj b (x-- Xi) 6 (P- PJ, where xdt), Pdt) are the position and momentum of the i-th particle". It represents only the probable number of particles; or perhaps the .Eyec_ggf number of part iclcs ; or perhaps it givesthe p_-robab£tidy that a. given particle is to be found in various regions. The decrease in HjJ 1S then not something \v11ic11 must - happen e-c:'ery time; but only what will nios! praGi/My happen: or perhaps what will. happen on the avernge, etc.

Unfort unatelv. neither BOLTZ:'l!:\::\~ nor anvbod v else has ever become

-' ~ .

more ex li.i t than this about just wha t BOLTDL\:\\,,'S I; and therefore

BOLTZ.\L\;':;': 's H -thcorcm, means. \ Vhcn our concepts are not precisely defined, they are bound to end up meaning different things to diffcren t people, thus creating rooorn for l'ndk;:;s and fruitless debate. of exactly the type that has been going on ever since. Furthermore, when we debate about imprecise concepts, we can never be sure whether we are arguing about a question of fact; or only a quesr ion about the meaning of words, From BOLTZMA::\;';'S clay to this, the debate has never been able to rise above this level.

If you think nl)-' characterization of the situation has 1 een too laconic, and unfair to many honest seekers after the truth, I invite yon to examine a recent review article on transport theory [4]. On page 27-1, the author states that" The Boltzmann distribution function - is the (probable) number of particles in the positiona] range d3 x and th.:

86

EDWIN T_ ] AY:--"ES:

velocity range c{Lu "On page 274 this is altered to: "The quantity I, the Boltzmann distribution function - is, roughly speaking! the average number of particles in a cell in the x- () space (the ,u-spacc). f refers to a single system. A more precise definition of I can be obtained through the use of the master function P." Consulting this master function; we find that neither the definition of P, nor ·its connection with I, is ever given. This, furthermore, is not a par ticularl y bad example; it is typical of what one finds in discussions of BOLTZl\.IANN'S theory. ..- ..

Let us note some of the difficulties that face the practical physicist because of this state of utter confusion with regard to basic concepts. Suppose we try to assess the validity of BOLT2::IL\N:;;'S equation (2) for some particular problcm ; or we try to extend it to higher powers in the density, where higher order 'collisions will become important in addition to the binary ones that are taken into account, in some sense, in (.3). If we agree that I represents an average number of particles, we must still

~speC~.fY w hat this a veragc is ~o be taken _over. I.' sit an average o.ver. the .particlcs, an average over time for a SIngle system, an average over

man)' copies of the single system, or an average over some probability distribution? Different answers to this question are going to carry different implications about the range of validity of (2), and about the

correct way of extending it to more general situations. Even without answering it at all, however, we can still see the kind of difficulties that are going to face us. For if I (x, p, t) is an average OH;f something. then the left-hand side of (2) is also an average over this same something. So also, therefore, is the right-hand side if the equation is correct. But on the right-hand side we see the product of two j's; the product of two a verages,

If you meditate about this fora moment, I think you will find it hard to avoid concluding that, if j is an average, then the right-band side ought to conta in the aH'r:q.;e of a product, not the product of the averages. These quanti ties arc surciv different; btl t we cannot say how diifercn t until \VC say what we arc averaging over. Until this a;n!J:igltity in the deliniiion .0/ BoLtzmanJl's I is cleared up, ire cannot assess the range 0/ ,'(didit_y 0/ Eq (2), ond,t'e cannot say hot» it should be extended ta inorc general problems, Because: of imprecise concepts, the theory reaches an impasse at the stage where it has barely scratched. the surface of any real treatment of irreversible processes l

2.2_ Method of GIBBS

For our second case history', we turn to the work of GIBBS. This was done some thirty years after the aforementioned work of BOLTZ~L-\X:-;, and the difficulties noted above, plus man.y others for which we do

Foundations of Probability Theory and Statistical Mechanics 87

not have time here, were surely dear to GlIms, who was extremely careful in matters of logic" detail, and definir ions.

All important advances have their precursors, the full significance of which is realized only later'; and the innovations of GIBBS were not entirely new. For example, considerations of the full phase space (r-space) appear already in the works of i\L-\XWELL and BOLTZ.'IL\X~; and Grnns canonical ensemble is dearly only a small step removed from the distribution laws of l\L-\XWELC and HOLTZMAN:\,. However, GIBBS applied these ideas in a way which was unprecedented; so much so that his work was almost totally rejected ten years later in the famous Ehrenfest review article [D], which has had a dominating influence on thought in statist.cal mechanics for fifty years. In this article, the methods of Gums are attackcclrcpeatedly, and the physical superiority of BOLTZ:\L\,;\:X'S approach is proclaimed over and over again. Fat example, GIiiBS' canonical and grand canonical ensembles are dismissed as mere "analytic!! tricks ", which do not solve the problem; but only enable GInES to evade what theauthors consider to be real problems of the subject!

Since then, of course, the mathematical superiority' of GIBBS' methods for calculating equilibrium thermodynamic properties has become firmly established; and so statistical mechanics has become a queer hybrid, in which the practical calculations are always based on the methods of GIBBS; while in the pedagogy virtually all one's attention is given to repeating the arguments. of BOLTZMANN.

This hybrid nature -- the attempt to gr;-aft together two quite incompatible philosophies - is nowhere more clearly shown than in the fact that the "official" commentary on GIBBS' work [6J devotes a major amount of space to discussion of ergodic theories. Now, it is a curious fact that if you study GIBBS' work, you will not find the word "ergodic;' or the concept of ergodicity, at any point. Recalling that ergodic theorems, or hypotheses, had been actively discussed by ot her writers for over thirty years. and recalling GIIm:;' extremely meticulous attention to detail, r think the only possible conclusion we can draw is that GW13S simply did not consider ergodiciiy as relevant to the foundations o] the subject, Of course" he was fa.r too polite a man tu sa)' so opcnlv ; and so he made the point simply hy developin.; his theory wit h-iu t making an): use of it. Unfortunately, this tactic was too subtle to be appreciated by most readers; and the few who did notice it took it to be a defect in Gruns' presentation, in need of correction by others.

This situation has had very unfortunate consequences, in that the , v ork of GIBBS has been persistently misunderstood; and in particular, the full power and generality of the methods he in traduced have not yet been recognized in any existing textbook. However, it is not a q nest ion of placing blame on anyone; for we can understand and sympathize

EDWli'< T. J:\Y;:-';ES:

with the position of everyone involved. I think that a historical study will convince YllU, as it has convinced me, that all of this is the more or Jess inevitable result of the fact that GIBBS did not live long enough to complete his work. The principle he had discovered was so completely new, and the method of thinking so completely different from what had gone before, that it was not possible to explain it fully, or to explore its consequences for irreversible phenomena, in th time that was gran ted to him.

GIBI3S wa-. in rapidly failing health at the time he wrote his work On statistical mechanics. and he lapsed into his final illness vcry soon after the manuscript 'was sent to the publisher. In studying his book, it is dear that it \\';).S never really finished ; and we can locate very accurately the place where time and energy rail out on him. Tile first eleven chapters are written in his familiar style - extremely meticulous attention to detail, while unfolding a carefully thought out logical development. At Chapter ":12, entitled, "On: the Motion of Systems and Ensembles of Systems Through Long Periods of Time", we see an. abrupt change of sty lc; the treatment becomes sketchy, and an ounts to littl more than a random .ollection of observations, trying to state in words what he had not yet been able to reduce to equations. On pages 1..lJ -144 he tries to explain the methodology which led him to his canonical and grand canonical ensembles, as well as the ensemble canonical in the angular momenta which was presented in Chapter 4 but not applied to any problem [7J. However, lie devotes only two sentences to this; and the principle he states is what we would recognize today as the principle of maximum en trapy! To the best of my knowledge, this passage has never been noted or quoted by any other author (it is rather well hidden amung discussions of other topics); and I discovered it myself only hv accident, three years after I had written some papers [Sj advccating this principle us a general foundation for statistical mcchan.r s. This dis\'(l\Try convinced n e that there was much more to the history of this subject than one (inds in auv textbook, and induced me to study it from the origiual ::;OlHC(,S; some of the resulting con-

clusions arc being presented ill this talk. .

GIIms' discussion of inc rcrsibihty in this chapter does not advance beyond pointing to a qua lit afivc analogy with the stirring of colored ink in water; "and this forms the basis for another of the EHHEXFEST'S criticisms of his work. I think that. had GIBBS been granted afew more years of \ ig:orou~ health) this would have been replaced by a simple and rigorous demonstration of the second law based OIl other ideas. For it turns out that all the clues necessary to point the way to this, and all the rna thernatical material needed tor the proof, were already present in the first eleven chapters of his book; it requires only a little more

:ph:j."5kal reaseniagto s,e.e, tl m t [nlrodlH:tiOll! of (:C".)~u':;e-gr.8! ilMd. d.i;:;\ tri btl t ioas does-not advance our undcrsta nding of irreversibil i ty and the serornrdhlw, for the $in:1ple reason that 'thfl ]a.Her ,:are cxpC'.riiucnt.dlj - observed macroseop$'c properties: and 'Hl{~ Hn~-gI'dlined and coarse-graiusddistributions lead. to just the same PHH:illictions lor i1]] ~1~tH'i'Q;;i,r(}p~C quantities. Thus., the '.diUer,en'c€b~h\'een the finc""gr.ained~l.l1d l'oarse-grai:ned H-functi(')n~ has ftlot,hi[)g to do with the' experimentally u'bs~r\'~'d entropy: it dt'lPfUds o~~y 011 t1]C par ticular war in\1,'hic4h we 'Ch(;1o~~~ t~~ CtKU-se-gmin',

On the other band I' th~ variational ,maXiHl:mnentloPY) property noted hy Gr BBS (:i:QCco; leadus ir!!lU1C'ditltel~f to ,(tFr~~lof, not O~1]Y ·(j.e the se,('),:nd law, but of all extension of tb-e seeond Iaw ro IH}nC'qll.ilibr~'li~m··~ta.t,!.!;;. I have recently p(:rirl,t,ed thiscut (ft] and 5U~)p'Ht!(1 the very simple proof. which I think is just the argufilcIlt Glf)nS'W(}~lkl Iiave giy!";'·n H he: har] been i'l:h~.e to complete hr:is work, Hu\\"e\~p'i thisis not the main point [ \vish to di;:,ct:].$.s tOIlil~ht • so let us turn back to other topirs.

Ih Uleft-'n~c of the EHRE:-:F.q,;::;·'r's position, it .ha;-; 10 be adrnitredtturt, UlfQ\lgll !'!If) fa.ult of his O\\'U, GmBd d]d fail [I) present any dear description of the motivation behind his work, £ ht::~rl;\'C thut it was virtually. impossible to rrud,eJ[staH.il. what Gn!-m:~ n:-1("thod~ nmeunred tc!.~ md.lhen> lore hca; grent: was ih(fi~·r ge.n,erali:(y ami ·rang(J ael 'i:_'dh,diiy. lInt];] the appearancc of Si1lA::-;Xti;S'S '!:l"qrk '0'11. Informntion 'Tbeory~ in. our own time [10]. '[, inally .. uutil r~c.:erl}tlr the situation :~Il~jrfib\\bi]hy theerv itself, which '\.~a;~ in ah:igh-pressuPJ phase cO'mpk.tely 9wnil1;_tt~d by H}t) fre'quen~y fhei()fj·,. which only sneers and aU(I,;I2k,'-) O~.t l'be the-ories of LA.N .. \CE and JEFFR£YSj has made it lmposslble even to discuss, niuch less pub! ish, the \de\Clj·pofnt. and approach ,\,\{hkh I believe ]\~as now s6h"f'·Qi these problems .

NO\i(, in order to iliend a. little more substance to ~11:{r~U remarks ]t~ 's examine' SOUle eq1]a:tlkn:ig, the net result of Gum~:\Ym-1L Considering <J; r1o~c~d system (i.e., no particles enter nr leave), tbe HH~r'!.ti(,.J:]~Hnl~C pro.p,('nic-s are tobccnlculated fnllu11 th,~ Hai·uillonirulll1{rJi' P'£) as {uIJ0\~J;,., fir~:t.\\:(~ define fhr. partitio,.1 /mlr.:l:£(m

where V!(·e Integrate ever al! ~he a,cccs~ib~~ pi:ta;;;.e ;-.p:Lt'l'J and the dt'r~ndefLGe rpn tlH.:' volurne fl aris,e,::,;Q·QCal.l:St the range oiinteS"ratlon over [hie' '['0- ordinates f{i depends Ion l.f. If we succeedIn e .... ·p,]u~tting this runction, thfju. "BtU thermodynamic- properties are kn61i,fl1:; for the ,energy II][lct['i)H (wh~dll determines the thermal p-mperti,es) 'is gi\'{"n by

U· a 'l z

= - -._ . O~J

,,:on 0,

V,pl

in which we interpret ,{3 as (kTtl, where k is BOLTZMANN'S constant and T the KELVIX temperature; and the equation of state is

p= ;- 'a?v·1og Z.

(7)

Now; isn't this a beautifully simple and neat prescription? For the first time in what has always been a rather messy subject, one had a glimpse of the kind of formal elegance that we have in mechanics, where a single equation (H.A)IlLTON'S principle) summarizes everything that needs to be said. Of an the founders of statistical mechanics, only GIBBS gives us this formal simplicity, generality, and as it turned out, a techniq ue for practical calculation which the labors of another sixty years have not been able to improve on .. The transition to quantum statistics took place so quietly and uneventfully because it consisted simply in the replacement of the integral in (5) by the corresponding discrete sum; and nothing else in the Iormalisru was altered.

In the history of science, whenever J. field has reached such a stage, in which thousands of separate details can be summarized by, and deduced from, a single formal rule. - then an extremely important synthesis has been accomplished. Furthermore,by understanding the basis of this rule it has always been possible to extend its application far beyond the original set of facts for which it was designed. And yet, this did not happen in the case of GIBBS' formal rule. Witl: only a few exceptions, writers on statistical mechanics since GlOBS. have tried to snatch a way this formal elegance by grafting GIBllS' method onto the substrate of BOLTZ:ilIANX'S ideas; for which GIBBS himself had no need. However, a few, including TOL)[AX and S~HRODI:-':-GEh.,. liavo seen GlOBS' work in a. different light - as something that can stand by itself without having to lean on unproved ergodic hypotheses. intricate. but arbitrarilv defined cells in phase space, Z-:::tars; and the like. Thus, while a detailed stu lv will show that there are as many different opinions as to the reason for GIBBS' rules as there are writers on the subject. a more coarsc-graincd view shows that these writers are split into two basic; camps; those who hold that the ultimate; justification of GIBBS' rules must be found in ergodic theorems; and those who hold that a pri nciple [or assigning a priori probabilities \\'i11 provide a sufficient justification. Basically, the confusion that :still exists in this field arises from the fact that, while the mathematical content of Gums' formalism can be set forth in a Icw lines, as we have just seen, the C-OtICr:PtU((J o .. isis underlying it has never ben agreed upon.

Now, while GIEBs'fom1alism has a great gCll~rality ~ in particular. it holds equally well for gas and condensed phases, while BOL TDL\XX'S results apply only to dilute gases ~ it nevertholcss fails to give llS many

things that BOLTZMANN'S c c collision equation" does yield, however imperfectly. For DOL TZMANN'S equation can be applied to irreversible processes; and it gives definite theoretical expressions for transport coefficients (viscosity, diffusion, heat conductivity), while GIBBS' rules refer only to thermal equilibrium, and one has not seen how to extend them beyond that domain. Furthermore, in spite of all my carping about the imprecision of BOLTZ)'·lA~~'S equation, the fact remains that it has been very successful in giving good numerical values for these transport coefficients; and it docs so even for faidy dense gases, where we really have no right to expect such success. So, my adulation of Gibbs must be carried to the point of rejecting BOLTl~r..\X:\:·s work; it appears that we need both approaches!

AU right. I have now posed the problem as it appeared to me Q number of years ago. Can't we learn how to combine the best features of both approaches, into a new theory that retains the unity and formal simplicitv of GIBBS' work with the ability to describe irrever .ible processes (hopefully, a better ability) of BOLTZ~1:\~~'S work? This queslion must ha ve occurred to almost every physicist who has made a serious study of statistical mechanics, for the past sixty years. And yet, it has seemed to many a hopelessly difficult task; or even an impossible one. For example, at the 1956 International Congress on Theoretical Physics, L. VA)," HO\"E [11J remarked, "In contrast to the case of thermodynamical equilibrium, no general set of equations is known to describe the behavior of many-particle' systems whenever their state is different from the equilibr iurn state and, in view of the unlimited diversity of possible noneq uilibriurn situations, the existence of such a set of equations seems rat her doubtful.".

Now. while I hesitate to say so at a symposium devoted to Philosophy (Ii Science, the injection of philosophical considerations into science h~15 usually proved fruitless, in the sense that i t does not, of itself, ie,id tc any advances in the science. But there is One extremely important exception to this; and 'it is in exactly the situation 110\\- before us. At r;le stage in development of a theory where we already' have a formalism successful in one domain, and we are trying to extend it to a wider one, some kind of philosophy about what the formalism" means" is absolutely essential to provide us with a sense of direction. And it need TIl);: even be a "true" philosophy - whatever that may mean - for its rcu] justification will not lie in whether it is "true", but in whether it docs point the way to a successful extension of the theory.

In the construction of theories. a philosophy playssomewhat the same role as scaffolding does in the construction of buildings; you need it desperately at a certain phase of the operation, but when the construction is completed you can remove if it you wish; and the structure

92

ED\VI:i '1'.. J,\ Y)';ES:

will still stand of i b O\\"Il accord. This [maIO;":y l~ im perfect, however, because in the case of theories, the scaffolding is rarely ugl}', and many will wish to retain it as an integral part of the final structure. At the ( pposite extreme to th is conservative at.ti tude stands the radical positi vist, who in his zeal to remove every trace of scaffolding, abo tears down part of the building. Almost always, the wisest course will lie somewhere between these extremes.

The point which [ am trying to make, In til is rather cryptic \\'uy, is just the one which we have already noted in tile attempt to evalua te and extend BOLTZllL\'~:\ 's collision equation. Differen t philosophies of what that equation means carry different implivations as to its range of validity, and the correct \vay of extending it . And we arc now at just the same impasse with regard to Gmus' equations; because their conceptual basis has Hot been pr~'c"isd_'l defined, the theory d issoloes iii ambiguities which have prevented us, for sixty vvars, from extending ito to

, , /0

new domains.

2,3, Conceptual Problems of the Ensemble

The fact that two different camps exist, with diametrically opposed views as to the justification of GIRl'S' methods, is simply the reflection of two diametrical.y opposed philosophies all ut the real meaning of the GIB (IS ensemble; and this in turn arises from two eli ff 'fen t philosophies about the meaning of ailY probability dist ribution. Thus: the foundations of probability theory i tsclf arc involved in th ~ problem of ext riding GIBBS' methods.

Statistical mechanics has always bl'l'Il troubled with questions concerning the relation between the ensemble and the individual system, even apart horn possible extensions to noncquilibrium Cas.C3. In the theory, we calculuu numbers to com pare wi t h ~:\pl'ril1l('nt by taking ensemble averages; tlJat is what we are dliin~ ill l~qs. (()) :lIHI (.7). _\]1(1 yet, nul' ex peri mcnts to check these prcdictious ,H" nut performed on enscmblcs ; they arc pr;rfonned on the one indii-idna! system that exists in the laboratory. Nevertheless. we find that the predictions are verified aCCU[Zlt .ly : a rather astonishing result, but one without which w > would have Ettie interest in ensembles. For if it were necc:-;:-;:try to repeat a thermodynamic measurement 1,000 times and average tile results before any regularities (laws of thermodynamics) began to appear, both thermodynamics and statistical mechanics would be virtually useless to us; and they would not apP('ar in our physics curriculum, Thus, it appears that a major problem is to explain 'why GIBBs' rules work in practice: and not only why they work so well, but \\'l1Y they work at all!

\Ve can make this dilemma appc~r still worse by noting that the relation between the.; ensemble and the individual system is usually

described by ;:luppnsing: that the individual Sy::lEI:!;Hl can be reg;xlrd,ud as h.aving bi[Hen drawn {"·at lraJld~jnl jj. bdm. the ensemble. I personally have never been ab~.¢ to comprehend what " at random means; [or 1 ask myselt: \Vhat b· the rriterien, whcllt Ls the test, b~' \\.'hich~\·e could decide ,\'hcthrr U was or W;,;~~ not r~aUy ~.~ r:8i.l1dmn~·?' Dn~.}.:.~ it make sense to ask \yhpther it was .c:.i(rtt.a.;~" ta'udp~, or appr,lJJ.imdld_'j,'r~tndurl1 ? - 1il,.nd n12ithcl' the literature nor my Introspection gi\~e me Cl.nYfilb,w0L Ho'wtlvet, ~\'en \\'iUwut lLl'"l1d.erstam.iing thi~ipoin't, the real diWculty hi obvious;' Ier the saina individual s},sf{'l'n Uta} ·$urelYj anti with.equal justice, be regarded a:-:: ,ln~\ri.ng been 4;[;;:l'\\:nj',u:t.raI1ld,o'm I, from nn~" nne of an infinite number of (ll/i.cren{ e.lls~mbh.~·~ But the: measured properHe$ ulan indi\']"(]lJIul ~ystem depend on the shrlt. ojthe..systenl ; and. unt 0]1 which ensemble you or I ril.",g:inl itt as 'kt YltlA been It d:r:a wn ft~;n~'~. H O\~.' ~ then Is 1 t ·~lo~:-:.ihk tha.tl ensemble U\·'"erag~'s;· 'f'oinddi." with expt·rillU:n:w] vulues]

The hV;i) different philosophical Gl.mr~ trv tq extricate themselves from this dl1~'.'mil\~~ 'iil biro entirely different W~\~·:» TiN H ~rgodk" camp, ()f COl)fS('. b;cnnlpd:il'tlli of those whobelieve thut ;~l prohability distribution describes au Qbjlerli.7i.'(·lr rt:alphY$]C-<lL1 s"itllt,tkm; that it stands for an ~~~s,erti[Hl about expcri mentallymeasurable !rr3qutIUJes ~ that ] t is there", ltifiC dUl('t' correct Of' i.nftJrn;~ctc; and th~U rhi~. ('an, h"ll)ir'i]'{("i·p}~,c,. hi! decided by ~l~rformi.ng'! random experiments". T~~cy no~~,e tbat wlmtwe rneasur« in any lex~rinh;J.1t 1:;5 ne\,"'\es5ari1y II time rt\'c-ruge OY~T'<'\ time that i~ ]Oll~ en ~hc at0l11l'C $cah.e of thing's;. a.n¢l.. SO the sueressof GWllS~ methods will be ~K'~·o.unl,t;d fen: H we ~.a1!1J. pro\:e., from the lu~(rn5copk; equations of .m{)trn·U!l,. th.;d the ihm·(f-t·t·rage for an individual ·S.:v;.;lci.n is. €:{Ltlal to' the ,(1"HSIJ:})lbh' ,(~;'~'}'(tge: O''C'c-r the particular ensembles g~.\·t:Ii by .GlEBS,.

This\l'iC',,_,'tH)~d·"t lla$ much to, r,e,~m"Ji)mend it. .I f! tIw fin:;t place, pllYildst;;.,;; have a na tur a. I h'!Hlc!1Jcy tobelieve '[hat, .. -iuce th.(· observed 'pl<opt~rhl::"~ of matter co, iu the tH'~'t:" arc .simotv the r'I',ultant ofits properties "in

,,:.~ .l .- L~

rhC':-.rn.;lU. ,. rn uH.illlh·l] !lIii-~Hl!y 'times 0\ ·t~r. it DU~'ht [0 he pn.~~ibl~ t 0 ~~btai.J1

the !~l;t('r)(.l,~~·opi("bl'ha r vior by $trJct l~gJ".:ai d~~d1!H,·tinil fronl the micro..on:q)i.c laws (J,W r;hyi=i,('~; and the ,< el'godk' 'J approarh f{l\~c~ promise uf b~;ing uble t·p ~Jo this, S&(\E.H~dlr,wh]le the '!Wrt;i:.;arr theorems h~ixt.'! uo~ hcen established ri~c:itO,u.slyand universallv, thu wnrk cLone' [(m this prOh~l'U] t hus fa r 11" S ulade ] t h ~,ghl Y pl a U,,": i bk t hat, m n a syste m intera·cting with ru. large' he.fitt bath .. the /re(j_1t"cJHi('::' \~·i th which variousmicro~.;,copl(::. (on dit 16n s :;1 r(~' real ized 'in the long ru n are indeed gi vcn e;Ol'T~C t 1 Y h\t~he Grsns canonical ensemble. Thishas been rendered so extrentelv

.,. - -. .-_. .. ..- . . - -

p]tm~~b]cthHt 1 think .~iIiO reasonable person cau :set~ol.~":.dJ/ dQubt tho.t it ~~ true, ~~Uhohlgh we rnnnot rule aut "the possibility of oconsioual "p~dhb" ]OgTh~:llr~ ;except~on.~, 'Thus' "the "ergodic 1~ school qK thQught has, it]. rnyopin ~.on r very nearly succeeded in its aim of e~t.d:blishi,ng equality of time a\"f"fagr's ~l.nd. -enS(:"ln1)lr a"i:eragI25/0"f tJlNJ Jmrticrdar (',a'se! oj ,GU,.f)s' f{iU()ll'I;~'al

fl;fl<semU!J: arid in the [QH'Q\\'ing- I ant simply gOeing hl grunt, for the sakeo.f tb:c. argumeht~ t]lat thI~·p[tJgr.anl has s'utc(~£d~>;tl.ent.iT~ly.

;N:" evertheless ~ the" ergodic" scl mol () f eh oug h t sti U fa,ce£ a 'fundaUl e [it •• tal difficulty; and OJ1J:,e thatwas first pointed 0m~ by BOiLTZ:Mi'i.?o<~ himself, and she~se.d III the EHl{][.NFE:~nr review attide. Curicusty, there exists to this day a group ofwodi(~I"5 in Europe 'l,:dlO refuse to recognize the seriousness of this dTU:U:t:ultj" j u[}d den:.r that ~ t ]lUlTI.Hdates their app(oa:c:h. Tbe'dHfh::.t;l[ty is that, even if [one Imd ~,lJcCf;cc1echn proving·thesfl ergo di~:: t be()r.rems ri gt)i[il~usl r a m1 d un i ve rsa n :f!t he resu It wou~d ha ve been .established only [or time averages (rae!' iu/iuilf.! times; 'wlJ!e:rt'as the experirnentswhich verify GU3l)S' rules ~nC;;:fS:lJ.U'c ~in:~I(:' ~.lYCr;"lt?c.S only 0\-(:'1" fh:llte times .. Thus, a further mathematical demeastratiunwould in tln~event he necessary, to show th;iBl!t the${~ Iinlte ti.~:11Ie a 'i/Cr;ji~p~S, have sulnci.ent~:y ~,pp;lroximatcGl their]j mit s for in'Hnite times.

Now we.can .giv~ simpleand g,e'r1er~lli counter-examples proving tbp,t suchan :1idditi(inalq,enlO~gtr~tioTI CWUlOt be g:i.ven; and indeed that any macroscopic system, gliven a~im€ millions of fimes the age of the nniVc0rs:e, 'still could not '~~amp]d:f'[non~ Utan an inflndc8im'~1 fraction of ~U themkrosc.opk states which ba\~c: high pt:0Qabmt)' En the 'Qj;m).nk~l emsemble ; and thus an.'-:, asaertion about the Iretj~u'llc£eS\"ith \'\;':h idl ~if~ ferent 1'llic[q8GO~.lic:;}tn.tt::'::l;O!rc. realized irran individual ~y$tenl1 •. i$C'O[D.~ p1etdr devoid of QperatiQl1aJ Ille~l1ing.

The easiest way 01 seeing' th~3 is j ust .,[00 note that. H n macroscopic sysift:en1 could sampJe ~n microscopic states in the; tiruc in. which raeasurements are. made, so thn t UrL;C mea:s:t1n;stl thnc tl veragcs "\'~::mmd he equal tQ ensemble av-era;gc:i;~ then the measured valueswould necessarily ~l\yay·s; be: the equilibrl:~fm values: \\T' ~, .. ould not Cy(?U l~n(}\!\' abl)~t irreversible processes: The l.ut: #l~d ~~:.~ r~p: Hh·&:.}..(~r~ .. ,flit fdie oj ast frr'!'t,tys.ibl .. ~ P)'Ql.'t!!3$ {.~lr,eq:d)' P"O'i.;t/i tkrt~ the. timtJ. F~·.q1'rin .. 'd jQr If nrfJr.fSf}rMJh't ;S{,f.1H,Pl";ug 'OJ ·m~~C:f()$lC!rtr;.s 1YH!'Sl Lenusch. l(mr;~~r thnn ,the U~N'; ret]ltired to 1~l,6rk,e. our nteasnrements. Thus, any rn;lq)(rrt.~d proof that time a\'er~'ugcs over tho fTi::dk 'timesinvelvodin ac'~~~d measurerncn t6 ~liroe ~rl~lal "t[ i, r:an~mk~tl ,lmSCi.n]~]e 9:VeHlgc-s would .. , fal' r rom jus dtying statistical luechJ nics, stand in clear conJli'<:t with the very expcrjmental facts about ]rm\"trsil]~lmty that ,\:\'C are lryi.ng toaccount for' h~' e~tenQ:]ng. G1B1JS,' methods!

the: fhintr w"hh:h h~s to be explained [s., not that ,cu$cm'q,]e averages

. .

are eq U9.1 to time ~tvC'l"a,gej. ; but the nU.1ch st ro:ng'ct's t,:,:d omen t thad!:

ensemble a\'e,r;tgC$ are eq'ltl!31 ~o~:~q]{!rhfJ!enta[ va~uc'S, The most that ergodic. tne'C)::rerns ceukl possibly establish is that ens(!mble av-erages are equal to time8i..vcTa~cs over infinite Hme, and so the l< ergodic. I' approach :t;~nno,t even justify '(.l::qJtlfl'iibriunl stttt['s.ti·cal :m~cha.nks\vifhoutcon tradict- 1 ng lex perimen t 011 fac:~s. OLl 'V~():Ui l~· • such ana pproac~~' ·c;.~:Ul not.be .. extended to irreverslble ,tyf0("tiSSt:lt \ ... ·here" in order fOT ('Tt.;'elHlb~I.:" thenry to be of

any use" the ensemble averages must still be equal to experiment ... d values: but the very phenomena tOI beexplained consist of th Iaet that these are ~,l:ot equal to time averages.

'"rbe: above. line of ~'le:~150nwug convinced nW,ben years ago~ that iu:rtilcr adl vances .in the basic formulation of statistical mccha nics c.a.nRo~t be made wifhin the framework of the ,uel1',gmiic 'j viewpoint: and, rirghU} or wrongly, it seemed. equally dear to. me that the n~aUy Iurrdamental trouble wl: • Ich 'was ptevenJing further advances, both in statistical mechanics and in the field of statistics in general, was this dogmatic. single-minded ' insi tcnce on the Irequency the-orr of probability which had domina" d the Held ((':If SO many yc~u:s.. ec~t that time, vhtuaUr every writer on probability theory ·feU Impelled to insert 'an introductory paragraph or two, expressing his. denunciation and total rejection of th,(! so~can-',.' "subj ctivc " interpretaticn ot prnbabmty. as- advocated by L.,u.lr •. \~CEr D'E AiORG.'\ x, POI S C;i\lli. KEY N.E~. and JEFF·]~E~,,; ~11I!d this was done, invus lably, without a.ny att mpt to, understand the arguments and results which these people ~ particularlj L.,,_PL\C.E and JE.FFRhYS - had advau .ed, The situation Wj1S, psycholpgicaUy, exactly like the one which has dominatedAmerican Politics since.abou t '~()J.(ll; the Republlcans contitnuany analyze the statements of Democrats and issue. counter~u·&11Uncnts. whkh the Demorrats contemptuouslv Iismiss without any attempt to understand them oranswer til m,

O[~ ~ha otherhand, I had taken the trouble o read all of JEFFliEY:$' wvrk~ and much of L.;\PL.\O(S. on probability theory .and was unable to H~1td any of thee terrible things about which the" Irequentist ·J\~·rilters hudwamcd ~U.' On thl;l philosophical side I found their arguments t'OI be. l;lf fto~n h:!I"l'5pon~'J,hlc and useless, so eminently SOIU:nd and reasonable UW~lt I could not imagine ally sane person disputing them, On the mathcrnatjcal si le. I found that in problems of statistical estimatlon and hypothc:;i;:; 't~~ting •. anyproblem fi)J" \\;hich the: .. .fr1,;·(IUdltist ,. offered {my solution at all was also solved with ease by th methods of L.lPL;\.·t~ ·HHl Jr;prl:tE'I!"S,,; and. ~heir results were either the stun€: 0.1." dernonstrublv suporior te the ones found by the frcqucntists, Furthermore, the methods of ·L.\l"LAC~~. and. Jw:..nrm:rl.·s (whIch were, of course, based on, )3'J\YE.3.' theorem as 'the hmdamenltaJeool of .s.t:atistk.'l.) ";"ere apptie.di '!i.v~th ,eqt1a~ ease to many problems which, accordi ng fa the fre'[ltHmt:ist, did nor belong to the fseld I.,'f pr' "babHity theory at all ; and they still ) i it d

. -

per'fccdj' reasonable, and s.den'tHkall.y useful, results]

I dOfl~twaFiJJt £0 dwell ~t length on the situation in probability th~', lily. because time is running short and a rather large expcsition of this.with ~'t1n mathema tical details, is heing readied for publication else', ... -herc ... Bur I,ct me j,tlst rnerJ;ti~,n one ex..amp~,e of ·"vha t one finds H be takes the trou ble to gn beyond polemics and study the mathematical facts of the mar: '·r.

In problems of interval estimation of unknown parameters, the frequentist has rejected the method of LAPL.:\CE and ]EFfREYSt on grounds that I can only dr-scribe a ideologi al, and has advocated vigorously the method of confidence intcrva b. Now it is a matter of straightforward mathemati s to show that, whenever the fr qucntists "e:.timator" is not a sufficient statistic (in the terminology of FISHEH), thor is always a class of possible samples f r which the method of confidence int rvals leads to absurd (II' {langer-ous!y misleading results, In the sensu that it yields a wrong answer far more frequently {or, if one prefers, with far higher probnbilitv] than one would suppo."e from the stated .onlidc nee I vel. The cmdi(lc'ncc interval can. in some case:;. contradict what can be prove I 11 strict dcdu .tive reasoning from the observed saruplo .. One can even invent problems, which arc not Itt all unrealistic, in which the probabilit ' of this happening is greater than tl c stated confidence lev 1.

This is somcth i ng which. to the best of my knowledge, you cannot find mentioned in an r of the' "orthodox" statist ical literature; and I shudder to think of some of the possible consequences. if important decisions are hci ne made on the basis of CI nlidcnce interval analvscs,

" ~

The method of LAPLACE and jEFFRE\'S is demonstrably fr 'C from this

defect; it cannot contradict deductive reasoning and, in the case of the aforementi n tl "h:1d" class of samples. it automatically d teet" them and yields a wider interval, so that the probability of a correct decision remains cqu: I to th ' stated value. Once one is aware of such facts, the arguments advanced against the method of L.\l'L:\CE and JEFFREY" [1;'[1(1 in favor of confidence intervals (i.c .. that it is m xminglcss to speak of the probability that f) lies in a certain interval, because 0 is not a Ii random variable, , but only an unknown constant) appear very much like those of the 17th ccnt urv scholar who claimed his th ·oIogy had proved there could be no moons on j upitcr, and st adlastly refused, to look through G.\LILEO·~ t('k:: ·O})('.

Si nee the rl·;.~:-;I)ni nO' by which the !1 :frcq1.11~11 t ist " has rejected L;\PL.\CE'S nwthllib i;; so pat en t lv nn$o~1111.1. and since attempt: t{ x nd, or even justifv. Cums' methods in. terms of tl.e frequency theory of probabilitv havr met with 311 impasse, it would appear that we ought to explore the possibilities of applying L.\IJL\Cr::'S II subjective Ii th ory of probabilitv to thi~ probl .m. ~\[ any rate. to reject this procedure without botherins to explore its potcutialitics, is hardly what 'we mean by a "scientijic" a tl itucle l So, I undertook to thin k through statistical mechanics all. v r ag~lin. u~ing the cone .. pt of H subj:cctivc" probability ..

It became clear, \'cry quickly, that to do this makes all the unsolved roblems of the theory appear in a v 'r)' differ fit light; and possibiliti es for extension of Caws' methods are seen in entirely different directions. Once we clearly : nd explicitly free ourselves from the delusion that an

ensemble describes an "objectively real" physical situation, and recognize that it describes Dilly a certain tieto. 01 kU!)1plcdge" then it i$ clear that. in the case of irrev' rsible processes, the knowledge which we haveis of a. differentnature than inthe case ofequllibrium, \V·~·can then see the problem as one which cannot even he formulated in terms of the frequency theory of probability. It is simply this; JVha"t .probalJ£lity .(lssigmnc~tt to n« .. i.c.rosti.de$ ,Mrtec#y de:scriiJ(.s tlu.:.' st{l.te 0/ lm.(f;"1i'lelige. wh.ich 'lVE hece, iu pract.icf, abonta 'nemeqnilioriwn, state? Such a question just doesn't nrake sense in terms of the Ircquency theory; but, thanks to the work of GUHJS and SIr~.\}i~O}{, I believe that it makes extremely geed sense, and in fact has a, very gcneralnnd mathcmntically unambiguous solution in terms of subjective probabilities.

3. The GCilctal.lIvbxiiIl.um-Ent;:QPY f·orfll:.dism

If we accept SHA:-iXONJ$ intcrpretaticn (w11.1(h can be justified by other mathematical arguutents entirel y independent of the ones gh,rt:u by-Saaxxox) that the quantity

is an , . information measure." for any probability distribution »: i.e, that it measures Ole '.r amount of uncer-tainty" as to the truevalue of i, then an ancien t principle OE wisdom - 'that one ought to tl.CkHCl\.~~,cdgG: frankly the CuB extent of his ignoran.GC - tetls us that the d[istriU[lHon that maximizes H subject to constraints which 'represent wharcver information 'we have; provides the most honest descrlpticn of what we know" The probabi [i ty is, by this process" j, sp read out" as w idel y as possible, without contradlcting the available in fQrmatiQrI"

Hut recognition ,~)f this simple principle. suddenly makes all the maximum-minimum properties given by C ns ssin his Chapter XI -"livh8!t I believe to he the climax of Guass' work, and just the place where tim-c and cfwrg}' ran out on hirn - ·acqul.re 'a, much deeper meaningvIf we. spet'ify the expectation value (')f the energy, this princ:iplt:: uniquely determines GIn BJ:S J canonical ensemble. If we specify thc expcctntious o[ cnelrg,,.. and mole numbers, it uniquely determines GWB's" grand canonical ensemble [8J, If we specify the expectations 01 'energy and angular momentum, it uniquely determines Gums' rotational ensemble [7]1. Thus. all the: results of Gnrss on stutisticnl mechanics follow immediately Irom the principle of maximum entropy, and their derivation is' astonishingly short and simple compared to the arguments usually found in textbooks.

But the generaJizatioI1J of Gums' fon.ilnHi:iw 'to nonequilibrium probIems also follows immediately (although I have to confess tha t J spent

(8)

six years trying to do this by introducing n wand more complicated principles, before I finally saw how simple the problem was). For this principle in no way depends on the physical meaning of the quantities we specify; there is nothing unique about energy, mole numbers, or angular momentum. If we grant that it represents a valid method of reasoning at all, then we must also grant that it applies equally well to any physical quantity tuhatsoeucr. So, let us jump immediately. in view of the time, to the most sweeping generalization of GIBBS' formalism ..

\Ve have a number of physical quantities about which we have some experimental information. Let them be represented by the H isenberg operators F,_ (x, t), F2 (x, t), .. " F;J' (x. l). In general they will depend on the position x and. through the equations of motion, on the timet. For example. F;_. might be the particle d .nsity, l~ the density of kinetic 'energy. F; the If mass velocity" of the {lui I. l~ thc (y z)-component of the stress tensor, Fs the intensity of magnctizution, .... and so on; whatever information of this type is available, represents our definition of the nonequilibrium state,

Now we wish to construct a density matrix .(1 which incorporates an this information, When I say that a d msity matrix" contains" certain information, I mean by this imply that, if we apply the usual rule for prediction; i.e. calculate the expectation values

(F;. (x, t) > = Tr [e Fdx, l)]

(9)

we must be able to recover this information from the density matrix. Thus, the mathematical constraints on the problem are that the expectation values (9) must agree with the experimental information:

I" (x, t) = Tr [g ~ (x, t)L

(10)

where I" (x, t) represent the expcriuu-utal V~hH;S. and RI; is the spacetime region in which we have informat ion about l't; in g .ncral it may be different for different J;;,. SUbject t ) these constraints, we are to maximize the" information entropy"

5 i= - Tr (u 10"· (])

(1'1)

which is the appropriate generalization of (8), as found many years ago by VON NElJ:<..lA::-'::S:. The solution of this variational problem is::

1 {1)1. }

e = Z exp "5' j ([3 x tit Adx, t) ~ (x, t).

. Ik=l RJ.

( 12)

where the All (x; t) are a set of real functions to be determined presently (they arise mathematically as Lagrange multipliers in solving the

vtl!'f~a tional problem 'VoJ it h cons tr.tt~ n ts]; and for Tlorm aJ] ~~ t itiriUM~ par titi on function of Gums has been generalized to UY'CF:lld'lU()u lu.m;Jl.{),IIU1~,.'

M.,'hk:h is a generalization of Guu~S'c(ruation (6)~; V\ih~re &: denotes the fundi(ma~ deriwltiv·c.MathenHltLca[ analysis shows that (1.4) is just sufficient to determlue uniquely thci,ntegrn.ls in UU~ exp~m.ent ,of (t2); it does not necessarily d'etl~n'lli;i:lJt: the Junctions *~ (x~ ll, but it d.Lle:~ detennivlrlC the only property of U:lO~C, Iunctious which isneeded in the thnory; a \J,¢l-Y ,I, fTh t eres t i ng,ex. am p'l e o f ma t:h:"m,~lu cal ccouemy,

The cl.cn$j,lty nlah-i~ Th~avinglJecri fhus found, prediction !fJf any other quantity ]{Jf., t) in its S~);}o¢-timc dependence is then ioundby" applyhlg the usual rul e ;

In. Eqs, (~2) to 1(,15,) wehave the sencralization of Gun,~5' a]gorith:1ll to a:rbHr~lry nl;m,e'~lllnibrhun p.roblenls.Fr(:)m this poi.nt on, it is, simply a question of mathematlcs to apply the th!eJ(H."j.· to a.nyproblern YOU\'VlS]l.

Of course, it requJ.res: a great deal u.fri(Jntrivial mathemat ics to Cltlrry 011'[ these steps explicitly (or .4ny nontrivial problem! If GrBBS~ odgbna[ formalism was somewhat de .. c:epti:yc", in tba:tij~s formal sirnplinity conceals ~11 en~d11(ms. ,ruY'i,OUi:1t 0.€' Intticate detajl,thesa,.rne' is, true with a v.el1g~l:uce f9f this geher.aHzation. Nevertheless, it is still 'C)'DIy· Hm thernatics: and if it were important e.nol1:g)] to get. a: certain resu It c;me Goulid <JJway5 hh"lf;': a buihjil~g. hdl ().f niri~hl~;mut icians ui[)!d cemputers to glind it ·VI1!it ;U1ef<l.>a~~c U9 further quc~tiqn:'§ (}rpril~dple toworryabeut.

F,().f the past three years, my slu:d ents and lh,a.\ -e beenexp 10r'[ [lg these mathematical problems, and we have a la·rge mass (l·f results that wi11be reported ,~n d,ne coarse .. \Vwtho:u t g(}ing into furtherdetails, let me [ust say that all the, previously known results in theory of irreversible PTOC;CSS:es can be dedv~d. easily from :tr.h~" a.l,gorl thm. Dissipative eHc,c,t';) such as Vi.5C05,ity~ d~.ffusion" heat comluctivity are obtained by .Jb~ct qu adratures. llsill'g (fS) ,\"!itll no need. for the f orward Integratien and cd.9,rsc"'grillining operations characteristic of previous treatments, For statlc tr(1n~port codlidcnts 'we obtain formuJas essentially equivalent b)' those of KU[i,o; we.can exhibit certain ensembles fer whk:h Kuno"s resul ts, inriginaUy o-bta.u'lliud by pe,rtuFbatl()n fb:eorJ'~ are In fact exac t,

J •• -l~~ l J \,1.\~.t!·

Because we arc freed from the need for time-smoothing and other coarse-graining operations .. the theory is no longer restricted to the quasi-stationary, long-wavelength limit. It gives, with equal case, general formulas for such things as ultrasonic attenuation and for nonlinear effects, such as those due to extremely large temperature or concentration gradients, for which previously no unambiguous theory existed .. Because of these results, I feel quite confident that we are on the right track, and that this generalization will prove to be the final form of nonequilibrium statistical mechanics"

Let me dose with a couple of philosophical remarks, relating this development to things I mentioned earlier in this talk. In seeking to extend a theory to new domains, some kind of philosophy about what the theory tI means;' is absolutely essential. The philosophy which led me to this generalization was, as already indicated, my conviction that the «subjective" theory of probability has been subjected to grossly unfair attacks from I coplc who have never made the slightest attempt to examine its potcntialitics : and that if one docs take the trouble to rise abov e ideology and study the facts, he will find that It subjective" probability is not only perfectly sound philosophically; it is a far more powerful tool for sol ving practical problems than the frequency theory. ] am, moreover, not alone in thinking this, as those familiar with the rise of the II neo-Bayesian " school of thought in statistics are wen aware.

Nevertheless, that philosophy of mine was only scaffolding, which served the purpos . of telling me in what specific way the formalism of GIBBS was to be generalized. Once a philosophy has led to a definite, unambiguous mathematical formalisru by which practical calculations may be carried out, then the issue is no longer one of philosophy; but of fact. The formalism either will or will not prove adequate in practice; and it will be judged, quite properly, not by the philosophy which led to it, but by the results which its gi\'llS. If you do not like my philosophy, but you find that the formalism, ncvcrthel ... -ss, docs give useful results, then I am quite sure that you will be able to invent some other philosophy by which that formalism call be justified! And, perhaps, that other philosophy will lead [0 still Iurt hcr generalizations and extensions, to which my own philosophy makes rue blind. That is, after all, j ust the process by which all progress in theoretical physics has been made.

HEFEtU~:'\t'ES

[1] EpSTEI~. P. S,: Textbook of thcrmod ymuuics, p. 27-34. ~e\V York: John \Viley & Sons, [nco 1937.

(2J BELL, E. T.: ;:\I¢n of mathematics, p. S·l6. Kti\\" York: Dover Publ. Inc. 1937. [3] Sec, for cxarnplc : Niels Bohr uu.d the development of physics (W. PAULI, cd.), p. 17-2$, and footnote, p. 76. :l'\ew York: Pergamon Prc'5s195 5; Obscrvation and interpretation (S. KOR:-;£R, cd.), p. 4·1-45. Nc\ .. · York: Academic

Fouuuu tions of Probability Theory and Staiistica.l :\Iccha.nics 1 DI

Press, Inc. 19S7; \V. I-IEISE:.iOERC"Phy!)ics a.nd philosophy, p, ~ 28-'146, New y ork : Harper & Brothers, PlJbt 1958; N. It. I-I. ·SON. Am, J. Phys, 27" 1 {1959); Ql1lantaand reality (D. E~)GE. co.) p., 85-93. Larchmont (Ne'w YOd()I: AnL Research Council 1: 962.

[il] DRES'OEN" ~J.: Revs. ~Iod. Phys. 33" 265 (t9Gt).

[5] En REN:YEST", P. and 'T.: Encykl, !\f!.lth. Wiss .. 19112. English translation by 1\1. J. lMoRA'\lCS rx, The 'cone 'ptual Iou ndations of the sta tlstical approach in mechanics. Ithaca (N. Y.): COfl1.CU University Press 1959.

[6] GIBBS, J., ,y,: Collected works and commentary, vol. II (A. HAAS, ed.), p. 46 'I ~4 88. Yale UI;l it versi ty Press New E f aven (C onn.}: 1 936.

[7] A StH."CCS8rU~ application of Gnu~s' rotationally canonical ensemble 'to the theory o.fgyromagnctic effects has since been givcn: S. P, H,6I:>-IS and! E.. T. J~\YXES. Revs. Mod. Phys, 34 .. , '143 (1962).

[8] ]AY~ES. E, T.: by-so Rev .. l0i6. 620; 10,8.1'71 (1957).

[9'] - Chapter -4- of Statistical physics (19'62 Brandeis Lectures) I(I{, "\'\ . FORO', ed .. ).

New York: ,,V. A. Benjamin, Inc. 196.); Gibbs. vs Boltzmann entropies, Am. J, Phys, J3~ 391 (1.965).

[.10] SIL,\:-.rN(~Nj C. E., and \V .. 'VE~~VER: The mathcmaticat theory o,E communication.

Urbana (IlL): University of Illinois Press 194'~. [11] IIQvJ!:j L. VA~: Revs, nlod. Phys, 29. 200 (1 ~S 7).