Annals of Mathematics The Logic of Quantum Mechanics Author(s): Garrett Birkhoff and John Von Neumann Source: Annals of Mathematics, Second Series, Vol. 37, No. 4 (Oct., 1936), pp. 823-843 Published by: Annals of Mathematics Stable URL: http://www.jstor.org/stable/1968621 . Accessed: 23/12/2013 13:10 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. . Annals of Mathematics is collaborating with JSTOR to digitize, preserve and extend access to Annals of Mathematics. http://www.jstor.org This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions ANNALS OF MATHEMATICS Vol. 37, No. 4, October,1936 THE LOGIC OF QUANTUM MECHANICS By GARRETT BIRKHOFF AND JOHN VON NEUMANN (Received April4, 1936) 1. Introduction. One of the aspects of quantumtheorywhichhas attracted the most generalattention,is the noveltyof the logical notions which it presupposes. It assertsthat even a completemathematicaldescriptionof a physical system' does not in generalenable one to predictwithcertaintythe result of an experimenton I, and that in particularone can never predictwith certaintyboth the positionand the momentumof (E (Heisenberg'sUncertainty Principle). It furtherassertsthat mostpairs of observationsare incompatible, and cannot be made on ' simultaneously(Principleof Non-commutativity of Observations). The object of the presentpaper is to discoverwhat logicalstructureone may hope to findin physicaltheorieswhich,like quantum mechanics,do not conformto classical logic. Our main conclusion,based on admittedlyheuristic arguments,is that one can reasonablyexpect to finda calculus of propositions whichis formallyindistinguishable fromthe calculus of linear subspaces with respectto setproducts,linearsums,and orthogonal complements-andresembles the usual calculusof propositionswithrespectto and, or,and not. In orderto avoid beingcommittedto quantumtheoryin its presentform,we have first(in ??2-6) stated the heuristicargumentswhichsuggestthat such a calculus is the properone in quantum mechanics,and then (in ??7-14) reconstructedthiscalculusfromthe axiomaticstandpoint. In bothpartsan attempt has been made to clarifythe discussionby continualcomparisonwithclassical mechanicsand its propositionalcalculi. The paper ends with a fewtentative conclusionswhichmay be drawnfromthe materialjust summarized. I. PHYSICAL BACKGROUND 2. Observationson physicalsystems. The conceptof a physicallyobserv- able "physical system" is present in all branches of physics, and we shall assume it. It is clear that an "observation" of a physicalsystem( can be described generallyas a writingdown of the readingsfromvarious' compatiblemeasurements. Thus ifthe measurements are denotedby the symbolsAl, *, An) then . 1 If one prefers, one mayregarda set ofcompatiblemeasurements as a singlecomposite "measurement"-andalso admitnon-numerical withsubsereadings-withoutinterfering quentarguments. Amongconspicuousobservablesin quantumtheoryare position,momentum, energy, and (non-numerical) symmetry. 823 This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions 824 GARRETT BIRKHOFF AND JOHN VON NEUMANN an observationof S amountsto specifyingnumbersxi, *. *,x, corresponding to the different ilkIt followsthat the mostgeneralformof a predictionconcerning( is that the point(xi, ***, x,) determined by actuallymeasuring Al, ** , ;z, will lie in a subset S of (xi, ***, xn)-space. Hence ifwe call the (xi, **, xn)-spacesassociatedwith(, its "observation-spaces,"we may call the subsetsof the observation-spacesassociated with any physicalsystemA, the "experimentalpropositions"concerning 5. . 3. Phase-spaces. There is one conceptwhichquantum theoryshares alike withclassicalmechanicsand classicalelectrodynamics. This is the conceptofa mathematical"phase-space." Accordingto this concept,any physicalsystem25 is at each instantlyhypotheticallyassociated with a "point" p in a fixedphase-space 2; this point is supposedto representmathematicallythe "state" of 5,and the "state" of e is supposedto be ascertainableby "maximal"2observations. Furthermore, the pointpoassociatedwithe at a timeto,togetherwitha prescribedmathematical"law of propagation,"fixthe point pi associated with e at any later time t; this assumptionevidentlyembodiesthe principleof mathematicalcausation.3 Thus in classical mechanics,each point of I. correspondsto a choice of n positionand n conjugatemomentumcoordinates-and the law of propagation may be Newton's inverse-squarelaw of attraction. Hence in this case 2 is a iegion of ordinary2n-dimensional space. In electrodynamics, the points of 2 can only be specifiedaftercertainfunction8-suchas the electromagnetic and electrostatic potential-are known;hence2 is a function-space ofinfinitely many dimensions. Similarly,in quantumtheorythepointsofZ correspondto so-called "wave-functions," and hence Z is again a function-space-usually4 assumed to be Hilbertspace. In electrodynamics, the law of propagationis containedin Maxwell's equations,and in quantum theory,in equations due to Schrodinger. In any case, the law of propagationmay be imaginedas inducinga steady fluidmotionin the phase-space. It has proved to be a fruitfulobservationthat in many importantcases of classical dynamics,this flow conservesvolumes. It may be noted that in quantummechanics,the flowconservesdistances(i.e., the equations are "unitary"). 2 L. Pauling and E. B. Wilson, "An introductionto quantum mechanics," McGraw-Hill, 1935,p. 422. Dirac, "Quantummechanics,"Oxford,1930,?4. 3For theexistenceofmathematical causation,cf.also p. 65 ofHeisenberg's"The physical principles of the quantum theory,"Chicago, 1929. " Cf. J. von Neumann, "MathematischeGrundlagender Quanten-mechanik,"Berlin, 1931. p. 18. This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions LOGIC OE QUANTUM MECHANICS 825 4. Propositions as subsetsof phase-space. Now beforea phase-spacecan become imbued with reality,its elementsand subsets must be correlatedin some way with "experimentalpropositions" (which are subsets of different observation-spaces). Moreover,thismustbe so done that set-theoretical inclusion (whichis theanalogueoflogicalimplication)is preserved. There is an obviousway to do thisin dynamicalsystemsofthe classicaltype.' One can measure position and its firsttime-derivativevelocity-and hence momentum-explicitly,and so establisha one-one correspondencewhich preservesinclusionbetweensubsetsof phase-spaceand subsetsof a suitable observation-space. In the cases of the kinetictheoryof gases and of electromagnetic waves no such simple procedureis possible, but it was imagined for a long time that "demons" of small enoughsize could by tracingthe motionof each particle,or by a dynamometer and infinitesimal point-charges and magnets,measurequantitiescorresponding to everycoordinateofthe phase-spaceinvolved. In quantumtheorynot even thisis imagined,and the possibilityofpredicting in generalthereadingsfrommeasurements on a physicalsysteme froma knowledge ofits "state" is denied;onlystatisticalpredictionsare alwayspossible. This has been interpretedas a renunciationof the doctrineof pre-determination; a thoughtful analysisshowsthat anotherand moresubtleidea is involved. The centralidea is thatphysicalquantitiesare related, but are not all computable froma numberofindependent basicquantities(such as positionand velocity).' We shall showin ?12 that thissituationhas an exact algebraicanalogue in the calculusofpropositions. 5. Propositionalcalculi in classical dynamics. Thus we see that an uncritical acceptance of the ideas of classical dynamics (particularlyas they involven-bodyproblems)leads one to identifyeach subset of phase-spacewith an experimentalproposition(the propositionthat the systemconsideredhas position and momentumcoordinatessatisfyingcertain conditions) and conversely. This is easily seen to be unrealistic;forexample,how absurd it would be to call an "experimentalproposition,"the assertionthat the angular momentum (in radiansper second) of the eartharoundthe sun was at a particularinstanta rationalnumber! Actually,at least in statistics,it seemsbest to assume that it is the Lebesguemeasurablesubsetsof a phase-spacewhichcorrespondto experimentalproposi0.7 iftheirdifference has Lebesgue-measure tions,twosubsetsbeingidentified, ' Like systemsidealizingthesolarsystemorprojectilemotion. ' A similarsituationariseswhenone triesto correlatepolarizationsin different planesof electromagnetic waves. 7 Cf. J. von Neumann,"Operatorenmethodenin der klassischen Mechanik," Annals of Math. 33 (1932),595-8. The difference of two sets Si, St is the set (St + S2) - St-2 of thosepoints,whichbelongto one ofthem,but notto both. This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions 826 GARRETT BIRKHOFF AND JOHN VON NEUMANN But in eithercase, theset-theoretical sumand productofany two subsets,and to experimental the complementofany one subsetofphase-spacecorresponding propositions, has the same property. That is, by definition8 The experimental propositionsconcerningany systemin classical mechanics, correspondto a "field" of subsetsof its phase-space. More precisely:To the "quotient"of such a fieldby an ideal in it. At any rate theyforma "Boolean Algebra."9 In the axiomaticdiscussionof propositionalcalculi whichfollows,it will be shownthat this is inevitablewhen one is dealing with exclusivelycompatible measurements, and also that it is logicallyimmaterialwhichparticularfieldof setsis used. 6. A propositionalcalculus for quantum mechanics. The question of the connectionin quantum mechanicsbetweensubsets of observation-spaces(or "experimentalpropositions")and subsets of the phase-spaceof a systema, has not been touched. The presentsectionwill be devoted to definingsuch a connection,provingsome factsabout it, and obtainingfromit heuristicallyby introducinga plausible postulate, a propositionalcalculus for quantum mechanics. Accordingly, let us observethatifal, , an are any compatibleobservations on a quantum-mechanical systemS withphase-spaceX, then'0thereexistsa set ofmutuallyorthogonalclosedlinearsubspaces Uj of Z (whichcorrespondto the familiesof properfunctionssatisfyingaif = Xi.if,-... anf = Xi.,f) such that every point(orfunction)f e Z can be uniquelywrittenin theform . celf + f = C2f2 + cafS + *[fi e( Hence ifwe state the DEFINITION: By the "mathematicalrepresentative"of a subset S of any observation-space(determinedby compatibleobservationsai, - - , a") for a quantum-mechanical system 5,willbe meanttheset ofall pointsf ofthephasespace of (5, which are linearlydeterminedby properfunctionsfk satisfying alfk = Xlfk, * * -* afk = Xnfk, where(X1,***, Xn) e S. Then it followsimmediately:(1) that the mathematicalrepresentativeof any experimentalpropositionis a closed linear subspace of Hilbert space (2) since all operatorsof quantum mechanics are Hermitian,that the mathematical representative of the negative"of any experimentalpropositionis the orthogonal F. Hausdorff, "Mengenlehre," Berlin,1927,p. 78. H. Stone,"BooleanAlgebrasand theirapplicationtotopology," Proc. Nat. Acad. 20 (1934),p. 197. 10Cf.von Neumann,op. cit., pp. 121,90,or Dirac, op. cit., 17. We disregardcomplications due to the possibilityof a continuousspectrum. They are inessentialin the present case. 11By the "negative" of an experimentalproposition(or subset S of an observationofS in to the sef-complement space) is meantthe experimental propositioncorresponding the same observation-space. 8 9M. This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions LOGIC OF QUANTUMMECHANICS 827 of the mathematicalrepresentativeof the propositionitself(3) the complement threeconditionson two experimentalpropositionsP and Q concerning following a giventypeofphysicalsystemare equivalent: (3a) The mathematicalrepresentativeof P is a subset of the mathematical ofQ. representative (3b) P impliesQ-that is, wheneverone can predictP withcertainty,one can predictQ with certainty. (3c) For any statisticalensembleof systems,the probabilityof P is at most theprobabilityofQ. The equivalenceof (3a)-(3c) leads one to regardthe aggregateof the mathepropositionsconcerningany physical oftheexperimental maticalrepresentatives thepropositionalcalculusforS. mathematically system(B, as representing We nowintroducethe representatives productof any twomathematical POSTULATE: The set-theoretical is itselfthe system, a quantum-mechanical concerning propositions ofexperimental proposition. ofan experimental representative mathematical REMARKS: This postulate would clearly be implied by the not unnatural operatorsin Hilbert space (phaseconjecturethat all Hermitian-symmetric space) correspondto observables;12 it would even be impliedby the conjecture that thoseoperatorswhichcorrespondto observablescoincidewiththe HermiM.13 elementsof a suitableoperator-ring tian-symmetric Now the closed linearsum Q%+ Q2of any two closed linear subspaces Q. of Hilbert space, is the orthogonalcomplementof the set-producta'- 2' of the henceifone adds theabove postulateto the Q. ofthe Qua; orthogonalcomplements then one can deducethat of theory, usual postulates quantum complement and closedlinearsumofany two,and theorthogonal The set-product an mathematically of any one closedlinearsubspaceof Hilbertspace representing systemS, itself experimentalpropoition concerninga quantum-mechanical concerning (S. proposition an experimental represents This defines the calculus of experimental propositions concerning A5, as a calculus with three operations and a relation of implication,which closely resemblesthesystemsdefinedin ?5. We shallnowturnto theanalysisand comstandpoint. parisonofall threecalculifroman axiomatic-algebraic II. ALGEBRAIC ANALYSIS 7. Implicationas partial ordering. It was suggested above that in any physicaltheoryinvolvinga phase-space,the experimentalpropositionsconcern12I.e., that given such an operatora, one "could" findan observableforwhichthe properstateswerethe properfunctionsofa. 1I F. J. Murrayand J. v. Neumann,"On rings of operators," AnnalsofMath., 37 (1936), p. 120. It is shownon p. 141,loc. cit. (Definition4.2.1 and Lemma4.2.1), that the closed linearsets of a ringM-that is those,the "projectionoperators"of whichbelongto Mcoincidewiththeclosedlinearsetswhichare invariantundera certaingroupofrotationsof Hilbert space. And the latter propertyis obviouslyconservedwhen a set-theoretical is formed. intersection This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions 828 GARRETT BIRKHOFF AND JOHN VON NEUMANN ing a systeme correspondto a familyof subsetsof its phase-space2, in such a way that "x implies y" (x and y being any two experimentalpropositions) means that the subset of 2 corresponding to x is containedset-theoretically in the subsetcorresponding to y. This hypothesisclearlyis importantin proportion as relationshipsof implicationexist between experimentalpropositions corresponding to subsetsofdifferent observation-spaces. The presentsectionwillbe devotedto corroborating thishypothesisby identifyingthe algebraic-axiomatic propertiesof logicalimplicationwiththoseof setinclusion. It is customaryto admitas relationsof "implication,"onlyrelationssatisfy- ing Sl: x impliesx. S2: If x impliesy and y impliesz, thenx impliesz. S3: If x impliesy and y impliesx, thenx and y are logicallyequivalent. In fact,S3 need not be stated as a postulateat all, but can be regardedas a definition oflogicalequivalence. Pursuingthislineofthought,one can interpret as a "physicalquality,"theset ofall experimental propositionslogicallyequivalentto a givenexperimental proposition.'4 Now ifone regardstheset S. ofpropositions implyinga givenpropositionx as a "mathematicalrepresentative"of x, thenby S3 the correspondence between the x and the S. is one-one,and x impliesy if and onlyif S. C S,. While conversely,if L is any systemof subsetsX of a fixedclass r, thenthereis an isomorphismwhichcarriesinclusioninto logical implicationbetweenL and the systemL* ofpropositions "x is a pointofX," X e L. Thus we see that the propertiesof logical implicationare indistinguishable fromthoseofset-inclusion, and thattherefore it is algebraically reasonableto try to correlatephysicalqualitieswithsubsetsofphase-space. A systemsatisfying S1-S3, and in whichthe relation"x impliesy" is written x C y,is usually'6called a "partiallyorderedsystem,"and thusour firstpostulate concerningpropositionalcalculi is that thephysicalqualitiesattributable to anyphysicalsystem forma partiallyordered system. It doesnotseemexcessiveto requirethatin additionany suchcalculuscontain two special propositions:the proposition0 that the systemconsideredexists, and theproposition? thatit does notexist. Clearly S4: ? C x C 0 forany x. ? is,froma logicalstandpoint,the"identicallyfalse"or "absurd" proposition; LIis the "identicallytrue" or "self-evident"proposition. 8. Lattices. In any calculus of propositions,it is natural to imagine that thereis a weakestpropositionimplying,and a strongestpropositionimpliedby, 14Thus in ?6, closed linearsubspacesof Hilbertspace correspond one-manyto experimentalpropositions, but one-oneto physicalqualitiesin thissense. 15 F. Hausdorff, "Grundzage derMengenlehre,"Leipzig, 1914,Chap. VI, 11. This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions LOGIC OF QUANTUM MECHANICS 829 a givenpair ofpropositions. In fact,investigationsofpartiallyorderedsystems fromdifferent anglesall indicatethat the firstpropertywhichtheyare likelyto possess,is the existenceofgreatestlowerboundsand least upperboundsto subwe state sets of theirelements. Accordingly, DEFINITION: A partiallyorderedsystemL will be called a "lattice" if and onlyifto any pair x and y ofits elementstherecorrespond S5: A "meet" or "greatestlowerbound" x n y such that (5a) x x ny Cy, (5c)z Cx andz Cy implyzcx n y. n y C x, (5b) S6: A "join" or "least upperbound"x nflysatisfying (6a) x U y D x, (6b) x U y D y, (6c) w D x andw D y implyw D x U y. The relationbetween meets and joins and abstract inclusioncan be summarized as follows,'6 (8.1) In any latticeL, thefollowingformalidentitiesare true, n a = a and a U a = a. n b = b n a and a U b = b U a. afn (b n c) = (afn b) n c and a U (b U c) = (a U b) U c. a U (a n b) = a n (a U b) = a. Moreover,the relationsa D b, a n b = b, and a U b = a are equivalent-each Li: L2: L3: L4: a a impliesbothofthe others. (8.2) Conversely,in any set of elementssatisfyingL2-L4 (L1 is redundant), a n b = b and a U b = a are equivalent. And if one definesthemto mean a D b,thenone revealsL as a lattice. Clearly L1-L4 are well-knownformalpropertiesof and and or in ordinary logic. This gives an algebraicreasonforadmittingas a postulate(ifnecessary) the statementthat a givencalculusofpropositionsis a lattice. There are other reasons17 whichimpelone to admitas a postulatethestrongerstatementthat the set-productof any two subsets of a phase-spacewhich correspondto physical qualities, itselfrepresentsa physicalquality-this is, of course,the Postulate of ?6. It is worthremarkingthat in classical mechanics,one can easily definethe meet or join of any two experimentalpropositionsas an experimental proposition-simplyby havingindependentobserversread offthe measurementswhich eitherpropositioninvolves,and combiningthe resultslogically. This is truein quantum mechanicsonly exceptionally-onlywhen all the measurementsinvolved commute(are compatible);in general,one can only expressthe join or 16The finalresultwas foundindependently by 0. Ore, "The foundationsof abstract algebra. I.," AnnalsofMath. 36 (1935),406-37,and by H. MacNeille in his Harvard DoctoralThesis, 1935. 17 The firstreason is that this implies no restrictionon the abstract nature of a lattice- anylatticecan be realizedas a systemofits ownsubsets,in sucha waythata n b is thesetproductofa and b. The secondreasonis thatifoneregardsa subsetS ofthephase-spaceof ofobserving(S in S, thenit is naturalto assume to thecertainty a systeme as corresponding that the combinedcertaintyof observingS in S and T is the certaintyof observing ; in So T = SfnT,-and assumesquantumtheory. This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions 830 GARRETT BIRKHOFF AND JOHN VON NEUMANN meet of two given experimentalpropositionsas a class of logicallyequivalent experimentalpropositions-i.e.,as a physicalquality."8 9. Complementedlattices. Besides the (binary) operationsof meet- and join-formation, thereis a third(unary) operationwhichmay be definedin partiallyorderedsystems. This is theoperationofcomplementation. In thecase oflatticesisomorphicwith"fields"ofsets,complementation corresponds to passage to the set-complement.In the case of closed linear subspaces ofHilbertspace (or of Cartesiann-space),it correspondsto passage to the orthogonalcomplement. In either case, denotingthe "complement" of an elementa by a', one has theformalidentities, L71: (a')' = a. L72: a n a' = ? and a U a' L73: a C b impliesa' D b'. = : By definition, L71 and L73 amount to assertingthat complementationis a ofperiodtwo. It is an immediatecorollaryofthisand the "dual automorphism" dualitybetweenthe definitions (in termsofinclusion)ofmeetand join, that L74: (a n b)' = a' U b' and (a U b)' = a', b' and anothercorollarythat the secondhalfofL72 is redundant. [Proof:by L71 and the firsthalfof L74, (a U a') = (a" U a') = (a' n a)' = ?', whileunder inversionof inclusion? evidentlybecomes0.] This permitsone to deduceL72 fromtheevenweakerassumptionthat a C a' impliesa - . Proof:forany x, (x fnxI)I = (' u x") = x' u x Dxf n x'. Hence if one admitsas a postulatethe assertionthat passagefroman experimentalproposition a toitscomplement a' is a dual automorphism ofperiodtwo,and a impliesa' is absurd,one has in effectadmittedL71-L74. This postulateis independentlysuggested(and L71 proved) by the fact the "complement"of the propositionthat the readingsxi, ***, x, froma seriesof compatibleobservationsyA1,** , /Anliein a subsetS of (xi, **- , xn)-space,is by thepropositionthatthereadingslie in theset-complement ofS. definition 10. The distributiveidentity. Up to now, we have only discussed formal featuresoflogicalstructurewhichseemto be commonto classicaldynamicsand thequantumtheory. We nowturnto the centraldifference betweenthem-the distributive identity ofthepropositionalcalculus: L6: a U (b n c) = (a U b) n (a U c) anda fn (b U c) = (a Rb) U (afn c) whichis a law in classical,but notin quantummechanics. 18 The following If a, b pointshouldbe mentionedin orderto avoid misunderstanding: are twophysicalqualities,thena U b,an b and a' (cf.below) are physicalqualitiestoo (and so are 0 and 0+). But a C b is not a physicalquality; it is a relationbetweenphysical qualities. This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions LOGIC 831 MECHANICS OF QUANTUM From an axiomaticviewpoint,each half of L6 impliesthe other.'9 Further, eitherhalfof L6, taken withL72, impliesL71 and L73, and to assume L6 and of a Boolean algebra.20 L72 amountsto assumingtheusual definition From a deeper mathematicalviewpoint,L6 is the characteristicpropertyof set-combination. More precisely,every "field" of sets is isomorphicwith a Boolean algebra, and conversely.21 This throwsnew lighton the well-known fact that the propositionalcalculi of classical mechanicsare Boolean algebras. ofthe It is interesting thatL6 is also a logicalconsequenceofthecompatibility observablesoccurringin a, b, and c. That is, if observationsare made by independentobservers,and combinedaccordingto the usual rules of logic,one can proveLi-L4, L6, and L71-74. These facts suggestthat the distributivelaw may break down in quantum mechanics. That it doesbreak down is shownby the factthat if a denotesthe experimentalobservationof a wave-packet4I'on one side of a plane in ordinary on the otherside,and b the obserthe observationof4fr space, a' correspondingly about theplane,then(as one can readilycheck): vationof6 in a statesymmetric f b n (a U a') = bn = b> ? = (b n a) = (bna) = (b n a') U (bna') REMARK: In connectionwith this, it is a salient fact that the generalized law oflogic: distributive in L6*: II n i \m E aid) j 1 = E jgi) (f[ i 1 \ aij(i)) breaksdownin the quotientalgebraof the fieldof Lebesgue measurablesets by theideal ofsets ofLebesguemeasure0, whichis so fundamentalin statisticsand oftheergodicprinciple.s theformulation 11. The modular identity. Although closed linear subspaces of Hilbert space and Cartesian n-space need not satisfyL6 relativeto set-productsand closedlinearsums,the formalpropertiesof theseoperationsare not confinedto L1-L4 and L71-L73. to satisfythe In particular,set-productsand straightlinearsumsare known22 so-called"modularidentity." 19 1931,vol. 2, p. 110. R. Dedekind,"Werke,"Braunschweig, " G. Birkhoff,"On the combination of subalgebras," Proc. Camb. Phil. Soc. 29 (1933), withrespectto inclusion 441-64,??23-4. Also, in any lattice satisfyingL6, isomorphism this need not be trueif L6 is not impliesisomorphism withrespectto complementation; assumed,as thlelatticeoflinearsubspacesthroughthe originof Cartesiann-spaceshows. 21 M. H. Stone, "Boolean algebras and theirapplication to topology," Proc. Nat. Acad. 20 (1934),197-202. 22 A detailedexplanationwill be omitted,forbrevity;one could refer to workof G. D. Birkhoff, J. von Neumann,and A. Tarski. 23 G. Birkhoff, op. cit., ?28. The proofis easy. One firstnotesthat sincea C (a U b) nc ifaCc,andbncC(aUb) (bnfc) C (aU b) nc. Thenonenotes ncinanycase,aU This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions 832 GARRETT BIRKHOFF L5: If a C c, thena U (b AND JOHN VON NEUMANN n c) = (a U b) n c. linearsubspacesof Therefore(sincethe linearsum ofany two finite-dimensional and consequentlyclosed) set-products Hilbertspace is itselffinite-dimensional and closedlinear sums of the finitedimensionalsubspaces of any topological linearspace suchas Cartesiann-spaceor Hilbertspace satisfyL5, too. One can interpretL5 directlyin various ways. First, it is evidentlya restrictedassociativelaw on mixedjoins and meets. It can equally well be regarded as a weakened distributivelaw, since if a C c, then a U (b n c) = (a U c). And it is self-dual: (a n c) u (bfnc) and (a U b) n c= (a U b) replacingC, n, U by D, u, n merelyreplacesa, b,c, by c, b,a. Also, speaking graphically,the assumptionthat a lattice L is "modular" (i.e., satisfiesL5) is equivalent to24saying that L contains no sublattice isomorphicwiththelatticegraphedin fig.1: n FIG. 1 Thus in Hilbert space, one can finda counterexampleto L5 of this type. Denote by ti, t2 t3 ..- a basis of orthonormalvectorsof the space, and by a, b, and c respectivelythe closed linear subspaces generatedby the vectors (Q2n + 1O-"tl + 10-2n2n+i), by the vectors 6,2n and by a and the vector ti. Then a, b,and c generatethe latticeof Fig. 1. Finally,themodularidentitycan be provedto be a consequenceof theassumpd(a), with the properties tion that thereexistsa numericaldimension-function D1: If a > b, thend(a) > d(b). D2: d(a) + d(b) = d(an b) + d(a U b). to latticesin whichthereis a This theoremhas a converseunderthe restriction ? < a, < a2 < ... < a. < 0 of finiteupper bound to the lengthn of chains25 elements. Since conditionsD1-D2 partiallydescribe the formal propertiesof probability,the presenceof conditionL5 is closely related to the existenceof an that any vector in (a U b) nc can be written t = a + , [a e a, , e b, t e c]. But = t- a is in c (since t e c and a e a Cc); hence t = a +B e aU (bfnc), and aU (bf c) D (aUb) nc, completing the proof. 24 R. Dedekind, "Werke," vol. 2, p. 255. 25 The statements of this paragraph are corollaries of Theorem 10.2 of G. Birkhoff, op. cit. This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions LOGIC OF QUANTUM MECHANICS 833 "a priorithermo-dynamic weight of states." But it would be desirable to propertiesofquantumphysics. interpretL5 by simplerphenomenological 12. Relation to abstract projectivegeometries. We shall next investigate how the assumptionof postulatesassertingthat the physicalqualities attributable to any quantum-mechanicalsysteme are a lattice satisfyingL5 and L71-L73 characterizesthe resultingpropositionalcalculus. This question is evidentlypurelyalgebraic. We believe that the best way to findthis out is to introducean assumption limitingthe lengthof chains of elements(assumptionof finitedimensions)of thelattice,admittingfranklythattheassumptionis purelyheuristic. L5 and L72 is the that any latticeoffinitedimensionssatisfying It is known26 directproductof a finitenumberof abstractprojectivegeometries(in the sense ofVeblenand Young), and a finiteBoolean algebra,and conversely. It is a corollarythat a latticesatisfyingL5 and L71-L73 possesses REMARK: independentbasic elementsofwhichany elementis a union,ifand onlyifit is a Boolean algebra. Again, such a lattice is a singleprojectivegeometryif and only if it is irreducible-that is, if and onlyif it contains no "neutral" elements.2Yx;S ?, 0 such thata = (a n x) u (a n x') forall a. In actual quantummechanicssuch whichcommuteswithall projecan elementwould have a projection-operator, ofobservables,and so withall operatorsofobservablesin general. tion-operators in quantummechanics.28 This wouldviolate the requirementof "irreducibility" calculusofquantummechanicshas the Hence we concludethat the propositional as an abstract geometry. samestructure projective Moreover,this conclusionhas been obtained purelyby analyzinginternal propertiesofthe calculus,in a way whichinvolvesHilbertspace onlyindirectly. 13. Abstractprojectivegeometriesand skew-fields. We shall now try to get a freshpicture of the propositionalcalculus of quantum mechanics,by recallingthe well-knowntwo-waycorrespondencebetweenabstract projective geometriesand (not necessarilycommutative)fields. and conNamely,let F be any suchfield,and considerthefollowingdefinitions structions:n elementsxi, ... , Xnof F, notall = 0, forma right-ratio [xi:. : Xn]r X : being called "equal," if and only two right-ratios [xi: ... Xn], and [.: ... .nl]r ifa z e F with t, = xiz, j = 1, *.. , n, exists. Similarly,n elementsyi, *. , Yn two left-ratios[yr: ... : Ynh of F, not all = 0, forma left-ratio[yi: . .. :Y and [1: ... 71,,] being called "equal," if and only if a z in F with ?li = zyi, i = 1, ***, n, exists. 2r G. Birkhoff "Combinatorial relations in projective geometries," Annals of Math. 36 (1935),743-8. 27 0. Ore,op. cit.,p. 419. 28 Usingthe terminology offootnote,'and ofloc. cit. there: The ringMM' should conthan0, 1, or: the ringM mustbe a "factor." Cf. loc. tain no otherprojection-operators cit.'3, p. 120. This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions 834 GARRETT BIRKHOFF AND JOHN VON NEUMANN Now definean n - 1-dimensionalprojectivegeometryPn,-(F) as follows: The "points" of Pn,-(F) are all right-ratios[x:...:xn]r. The "linear subspaces" of Pmi-(F) are those sets of points,which are definedby systemsof equations ak1X1 + + akVXI, = 0, k = 1,*,m. (m = 1, 2, *, theaki are fixed,but arbitraryelementsofF). The proof,that this is an abstractprojectivegeometry,amountssimplyto restatingthe basic propertiesoflineardependence.29 The same considerationsshow,that the (n - 2-dimensional)hyperplanesin torn= 1,notall ai = 0. Put aii = yi,thenwe have Pm-1(F) correspond (*) y1xl + * + y.xn = O. not all y = 0. This proves,that the (n - 2-dimensional)hyperplanesin Pm._(F) are in a oneto-onecorrespondence withthe left-ratios [y.:... :Y.] So we can identifythemwith the left-ratios, as points are already identical with the right-ratios, and (*) becomes the definitionof "incidence" (point C hyperplane). Reciprocally,any abstractn - 1-dimensionalprojectivegeometryQn-,with n = 4, 5, *-- belongsin this way to some (not necessarilycommutativefield withPn..1(F(Qn..1))Y10 F(Qn1), and Qn-,is isomorphic in to involutory 14. Relation of abstractcomplementarity anti-isomorphisms skew-fields. We have seen that the familyof irreduciblelatticessatisfyingL5 and L72 is preciselythefamilyofprojectivegeometries, providedwe excludethe two-dimensionalcase. But what about L71 and L73? In other words, for which P,1(F) can one definecomplementspossessingall the known formal propertiesof orthogonalcomplements? The presentsection will be spent in answeringthis question.3a ofB. L. Van derWaerden's"ModerneAlgebra,"Berlin,1931,Vol. 2. Cf. ??103-105 ... means of course n - 1 2 3, that is, that Q._, is necessarily a "DesargueNew York,1910, sian" geometry. (Cf. 0. Veblenand J.W. Young, "ProjectiveGeometry," in the classicalway. (Cf. Veblen Vol. 1, page 41). Then F = F(Q-,,) can be constructed betweenQni and the and Young, Vol. 1, pages 141-150). The proofof the isomorphism above,amountsto this: Introducing(not necessarilycommutative) Pn,,(F) as constructed theequationsofhypercoordinatesxi, **, x,,fromF in Q,,-,,and expressing homogeneous planes withtheirhelp. This can be done in the mannerwhichis familiarin projective geometry,althoughmost books considerthe commutative("Pascalian") case only. D. 7th edition,1930,pages 96-103,considersthe nonHilbert,"Grundlagender Geometrie," and n - 1 = 2, 3 only. geometry, case, but foraffine commutative Consideringthe lengthyalthoughelementarycharacterof the completeproof,we propose to publishit elsewhere. 29 30n = 4, 5; 30a R. Brauer, "A characterization of null systemsin projectivespace," Bull. Am. Math. Soc. 42 (1936),247-54,treats the analogous questionin the oppositecase that X ? 0 is postulated. This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions n x' LOGIC OF QUANTUM 835 MECHANICS First,we shall show that it is sufficient that F admit an involutory antisomorphism W: x = W(x), thatis: Q1. w(w(u)) = U, Q2. w(u + v) = w(u) + w(v), Q3. w(uv) = w(v) w(u), witha definite diagonalHermitian formw(xz1),yi + + * W(xn)7n^, where Q4. w(x1),ylxl+ *** + w(x.)-yfx.= 0 impliesxi = ** = xn = 0, the ey;beingfixedelementsofF, satisfying w(yi) = ey. Proof: Consider ennuples (not right- or left-ratios!)x: (xi, ,), ofF. Defineforthemthevector-operations ) ofelements Q1, I: ** xz: (XIz, X + t: (XI + *.* (z in F), , xnz) {i X *so X + + + n and an "innerproduct" (Q1x) = W(e1)-YiX W(fn)YnX. Then thefollowingformulasare corollariesof Q1-Q4. (x, 0) = w((Q,x)), (Q,xu) = (Q,x)u, (eu, x) = w(u)( , x), (Q, x' + x") = (Q, x') + (Q, x"), (Q' + V", x) = (i', x) + (W", x), (x, x) = w((x, x)) = [x] is # O ifx # 0 (that is, ifany xi #0 ). IP1 IP2 IP3 IP4 We can definex 1 t (in words:"x is orthogonalto i") to mean that (%,x) = 0. This is evidentlysymmetric in x, t,and dependson the right-ratios [xi:... :xn],, ... : a .1 b,betweenthe therelationof"polarity," [.. onlyso it establishes points b: [ti: ... : n]r ofPn.-(F). a: [X1: ... :Xn]7, ] : td] ofPn-i(F) constitutea linearsubspace The polarsto anypointb: ...: ofpointsofP,1(F), whichby Q4 does not containb itself,and yetwithb generates wholeprojectivespace P._,(F), sinceforany ennuplex: (xi, * X* , xn) x = X' + t.(]-1(QP, x) whereby Q4, [t] $ 0, and by IP (Q, x') = 0. This linearsubspaceis, therefore, an n-2-dimensional hyperplane. elementofP.-,(F)1 one can set up inductively Hence ifc is any k-dimensional k mutuallypolar pointsb(l), ***, b(k)in c. Then it is easy to showthat the set c' of pointspolar to everyb(l), ***, b(k)-or equivalentlyto everypoint in cconstitutean n-k-1-dimensional element,satisfyingc nc' = ? and c U c' = 0. Moreover,by symmetry(c')' D c, whence by dimensionalconsiderations c -* c' defines c" = c. Finally,c D d impliesc' C d', and so the correspondence ofPnpl(F) completing an involutory theproof. dual automorphism This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions 836 GARRETT BIRKHOFF AND JOHN VON NEUMANN In the Appendixit will be shownthat thisconditionis also necessary. Thus the above class of systemsis exactly the class of irreduciblelattices of finite dimensions> 3 satisfyingL5 and L71-L73. III. CONCLUSIONS 15. Mathematicalmodels forpropositionalcalculi. One conclusion which can be drawn fromthe preceding algebraic considerations,is that one can models fora propositionalcalculus in quantum meconstructmany different by knowncriteria. More precisely,one chanics,whichcannotbe differentiated satisfyingQ4 (such can take any fieldF havingan involutoryanti-isomorphism fieldsinclude the real, complex,and quaternionnumbersystems3l),introduce and thenconstruct suitablenotionsof lineardependenceand complementarity, n a modelPn(F), havingall ofthe propertiesofthe foreverydimension-number propositionalcalculus suggestedby quantum-mechanics. models P,(F) whose elements One can also constructinfinite-dimensional spaces. But consistofall closedlinearsubspacesofnormedinfinite-dimensional of laws" formal (which Hankel's principleof the "perseverance philosophically, of analysis technical and mathematically, L5)32 leads one to try to preserve continuous-dimensional a lead one to prefer in Hilbert space, spectraltheory modelPC(F), whichwillbe describedby one ofus in anotherpaper.33 PC(F) is veryanalogouswiththe modelfurnishedby the measurablesubsets ofphase-spacein classicaldynamics.34 16. The logical coherence of quantum mechanics. The above heuristic statementsin suggestin particularthat thephysicallysignificant considerations quantummechanicsactuallyconstitutea sortof projectivegeometry,whilethe a givensystemin classicaldynamics statementsconcerning physicallysignificant constitutea Boolean algebra. They suggesteven more stronglythat whereas in classical mechanics any propositionalcalculusinvolvingmorethan two propositionscan be decomposed into independentconstituents(direct sums in the sense of modern algebra), quantum theoryinvolvesirreduciblepropositionalcalculi of unbounded complexity. This indicatesthat quantum mechanicshas a greaterlogicalcoherence 31 In therealcase,w(x) = x; in thecomplexcase,w(x + iy) = x - iy; in thequaternionic case, w(u + ix + jy + kz) = u - ix - jy - kz; in all cases, the Xi are 1. Conversely, A. Kolmogoroff,"Zur Begriindungder projektivenGeometrie,"Annals of Math. 33 (1932), 175-6 has shownthat any projectivegeometrywhose k-dimensionalelementshave a locally relativeto whichthe latticeoperationsare continuous,must be over the compacttopology real,the complex,or the quaternionfield. in P,(F) onlyelementswhich by the artificeof considering 32 L5 can also be preserved whichare offinitedimensions. eitherare orhave complements 33 J. von Neumann,"Continuous Proc. Nat. Acad., 22 (1936),92-100and geometries," 101-109. These maybe a moresuitableframeforquantumtheory,thanHilbertspace. 34 In quantummechanics,dimensions are uniquelydeterminedby but not complements the inclusionrelation;in classical mechanics,the reverseis true! This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions LOGIC OF QUANTUM MECHANICS 837 than classical mechanics-a conclusioncorroboratedby the impossibilityin quantitiesindependently. generalofmeasuringdifferent 17. Relation to pure logic. The models for propositionalcalculi which have been consideredin the precedingsections are also interestingfromthe standpointof pure logic. Their nature is determinedby quasi-physicaland fromtheintrospective and philosophicalconsideratechnicalreasoning,different to comtionswhichhave had to guidelogicianshitherto. Hence it is interesting pare the modificationswhichthey introduceinto Boolean algebra, with those whichlogicianson "intuitionist"and relatedgroundshave triedintroducing. seems to be that whereaslogicianshave usuallyassumed The main difference that propertiesL71-L73 of negationwere the ones least able to withstanda identitiesL6 as criticalanalysis,the studyof mechanicspointsto the distributive theweakestlinkin thealgebraoflogic. Cf. thelast two paragraphsof ?10. Our conclusionagrees perhapsmorewith those critiquesof logic,whichfind most objectionablethe assumptionthat a' U b = 0 impliesa C b (or dually, the assumptionthat a n b' = ? impliesb D a-the assumptionthat to deduce that an absurdityfromthe conjunctionof a and not b, justifiesone in inferring a impliesb).36 18. Suggested questions. The same heuristicreasoningsuggeststhe followquestions. ing as fruitful What experimentalmeaningcan one attach to the meetand join of two given experimentalpropositions? What simpleand plausiblephysicalmotivationis thereforconditionL5? APPENDIX 1. Considera projectivegeometryQ,,i as describedin ?13. F is a (not necessarilycommutative,but associative) field,n = 4, 5, ... , Q.-, = PnA.(F)the projective geometryof all right-ratios[xI: * :xnlI, which are the points of are representedby the left-ratios Qn-1. The (n - 2-dimensional)hyperplanes [yI: ... Yn1 incidenceof a point [xI:... xn], and of a hyperplane[Yi:... Yn1l beingdefinedby n (1) z yixi = =0 1YXi All linearsubspacesofQ.-, formthelatticeL, withtheelementsa, b,c, *ii Assumenowthatan operationa' withthepropertiesL71-L73 in ?9 exists: L71 (a')' = a =? L72 ana' and a U a' = O, L73 a C b impliesa' D b'. 35 law to show,thatassuming ouraxiomsL1-5and7, thedistributive It is notdifficult L6 is equivalentto thispostulate:a' U b = Uimpliesa C b. This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions 838 AND JOHN VON NEUMANN BIRKHOFF GARRETT They imply(cf. ?9) (afn b)' = a' U b' and (a U b)' = a' L74 f b'. in a, b,owingto L73 and L71. Observe,thattherelationa C b' is symmetric 2. If a: [x1:... : x,],is a point,thena' is an [yr:... yJh. So we maywrite: [x:l*** Xn1r,= (2) [lyi .. ** s]l ^ and definean operationwhichconnectsright-and left-ratios. We know from ?14, that a generalcharacterizationof a' (a any elementof L) is obtained,as soon as we derive an algebraic characterizationof the above [xi:*...* : x]r . We will now findsuch a characterizationof [xi: * :x"] , and show, that it justifiesthedescriptiongivenin ?14. in In orderto do this,we will have to make a ratherfreeuse of collineation3 which Qn-1. A collineation is, by definition,a coordinate-transformation, replaces[xi: ** :xn]4 by [xl: ... **n]. , (3) n xi= i -2 Xi aii forj = 1p -n. Here the wijare fixedelementsofF, and such,that (3) has an inverse. (4) n xi= j=1 Oiji, fori =1, ,n, theOijbeingfixedelementsofF, too. (3), (4) clearlymean lifk = 1 8k1 = (5) n s2iOtj j=1 0 ~ 11~ (ki ifk 5e 1 ik Sik J n1 (j Oik = i-1 jik= Considering(1) and (5) theyimplythe contravariantco6rdinate-transformation forhyperplanes:[yi: ... :yn],becomes[pi: ... :9,]j, where Yi= (6) yi= (7) n i 21 n j= 1 Yioii, forj = 1,- aii n, fori= 1,***,n. on the leftside of the variablesin (Observe,that the positionof the coefficients (4), (5), and on theirrightside in (6), (7), is essential!) 3. We willbringabout (8) [dil:Si n**R]' = [MI:... * in~l fori This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions = 1, * n, LOGIC 839 MECHANICS OF QUANTUM by choosinga suitablesystemof coordinates,that is, by applyingsuitable collineations. We proceedby induction:Assume that (8) holds fori = 1, m - 1(m = 1, * , n), then we shall finda collineation which makes (8) true for i= 1,.***,m. Denote the point [5il: ..*. :sn]r by pi, and the hyperplane[6il: ... : 65in by h*our assumptionon (8) is: p*' = h*fori = 1,.**, mr-1. Considernow a point a: [xl:* *:x,],, and the hyperplanea': [yi:... :y** ]i. Now a _ p*' = h* means (use (1)) x, = 0 and p* < a' means (use (8)) y; = 0. But thesetwo statements are equivalent. So we see: If i = 1, * , m - 1, then xi = 0 and y, = 0 are equivalent. As 8mS= 0 for Put Pm: [y*:.. . :y]. Consider now p*: [3mi:...:*mnnr. , m-1. Furthermore, i = 1, ***,m-1, so we have y =0 fori = 1, p f pea o, p* . o, so p*not _ p*'. By (1) thismeans y#P O. Formthecollineation(3), (4), (6), (7), with ... - 09si= col; 1sV im = @swi = YM for i = m + 1, *;,n, Y* all other0,s,ws = 0. One verifiesimmediately,that thiscollineationleaves the coordinatesof the n, invariant, and similarlythose of the Pi: [5i:l * in], i = 1, :5iin]i, i = 1, ***m - 1, whileit transformsthoseof p*: [aj:. into [8i: ...* 5*M]l So after this collineation (8) holds fori = 1, **, m. Thus we may assume, by induction over m = 1, ** , n, that (8) holds for all i = 1, ... , n. This we willdo. The above argumentnow shows,that fora: [xi: ... xi = 0 is equivalent to yi = 0, (9) 4. Put a: [xi: ...- :xlr, a': [yi: ... :yn]z, further xi = 1. Then (9) gives *y"]1 for i = 1, ... , n. : and b: [i: ... -n]rb': Assume first i1 = 1, X2 = 17117=3 ?I= = ** = we can normalize ti = 1, and Only, so tb = f2(77). Assume a': [yr:... :xn]r, n = 0. n = 0. yj i 1n] [771 i 5= O, so {a can depend on 712 = Then (9) gives 0, so we can normalize Yi = 1. = 0. Now a S b' means by (i) 1 + IX2 = 0, and b S a' means 1 + Y2f2(G) So if x2 # 0, we may put These two statements must, therefore, be equivalent. If x2 = 0, then = =-(f2(1i, and obtain Y2 = fX-'))-f' 2(f))-' of 0 by (9). Thus, x2 determines at any rate y2 (independently y2= Permuting the i = 2, * * *, n gives, therefore: Xa, Xn) : Y2 = (P2(X2). There exists for each i = 2, ... , n a function pi(x), such that y; = (ii) * * Or: (10) If a: [1: XI ... : Xn]r, then a': [1:V2(x2): ...* :n(Xn)]l- This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions 840 GARRETT Applying this to a: [1:X2: and c < a' are equivalent,so (11) AND JOHN VON NEUMANN BIRKHOFF ...: X]r and c: [1: u1:... n :u shows: As a < c' n E Vi(ui)xi = -1 is equivalentto E jp(xi)vi =-1. i -2 i 2 that(9) becomes: Observe, (Pi(x)= Oifand onlyifx = 0. (12) 5. (11) withX3= Xn = = *.. is equivalentto IP2(X2)U2 = - 1. equationholds,and so bothdo. Choose X2, = U3 If X2 U2 in thisway, butleave x3, 0, u2 ... (11) becomes: n (13) E i-3 Nowputx5 = i-3 = =n*= = s.Un 0 is 0. shows:IP2(U2)X2 (-IP2(X2))-f, Xn, U3, n P3(U3)X3 + io4(u4)X4 = thatis (forX4,U4 5s , = (pi(ui) xi = 0 is equivalentto E .* 0 n = = *.. z- ... = - 1 thenthe second unarbitrary. Then (pi(xi)Ui = 0. Then(13) becomes: equivalentto IP3(X3)U3 + IP4(X4)U4 = 0, 0): (a) x3x7l = so4(U4)-1 '93(ug) (b) u3u-1 to (14) is equivalent 4(x4) -1 v(x8). = Let x4, X3 be given. Choose u3, U4so as to satisfy(b). Then (a) is true,too. Now (a) remainstrue,if we leave U3, U4unchanged,but changex, z4,without thatis, the changing x3x41. So (b) remainstoo trueundertheseconditions, value of P4(x4)-1V&(X3) does not change. In otherwords:(P4(X4)- sP3(x3)depends on x3x-1only. That is: (P4(X4)-1 V3(X3) = (P34(X3X-4). Put X3 = XZ,X4 = x, thenwe obtain: (15) (P3(xz) = IP4(X) {134(Z). Thiswas derivedforx, z $ 0, butit willholdforx or z = 0, too,ifwe define 134(0) = 0. (Use (12).) (15), withz = 1 givesIP3(x) = J04(X)a34,where a34 = 4,34(1) 0 0, owing to thei = 2, * , n gives,therefore: (12) forx 0 0. Permuting (16) Vi(x) = ros(x)ai,, (Fori =jput aii = 1.) Now (15) becomes (17) W((Z) = (P2(ZX) = whereaij 0 0. (P2(X)w(z) a4234'(z)a23. This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions LOGIC OF QUANTUM 841 MECHANICS Put x = 1 in (17), writex forz, and use (16) withj = 2: (Ps(x) = (18) #w(z)yi, (O = 6. Compare(17) forx = 1, z= Then whereA, yj F 0. IP2(1)l -yi = ai2). u;x=uz=v;andx=1,z=vu. w(vu) = w(u)w(v) (19) results(12) and (18) give w(u) = 0 if and only if u = 0. (20) Now writew(z), ,y for,w(z)t-', jBy. Then (18), (19), (20) remaintrue,(18) in so far,as we have i = 1 there. So (11) becomes is simplified n Ew(Ui) 'Yix = - (21) i =2 (21) is equivalentto n E i -2 W(Xi) jiUi =-1 u and all otherxi = ui = 0 give: w(u)Y2x =-1 is equivalentto If x # 0, u = -72 1 w(x)-', thenthesecondequationholds, But and so the firstone gives: x =-y1 w(u)-1 =-2 -y(w(--yj'w(x)-1))-l. (19), (20) implyw(1) = 1, w(w-') w(w)-', so the above relationbecomes: X2 = x, u2 w(x)'yu = = -1. - X = -n'(w(-8^ = = -2 -w(x71))- -y:'W(W(X)(-Y2)) x = 1,as w(w(1)) Put herein =-72 = W(1) = lW((-8-l(X)-l)-l) W(-2)W(W(X)). 1, SO -'2y results. Thus the above equation becomes (22) and w(-Y2) = 1W(-Y2) =-2 w(w(x)) = x, = 7-2 gives,ifwe permutethei = 2, Put ui =-'yr' n, w(-'yi) = -Fi. (23) (24) W(--Y2) in (21). n E . -2 Then considering(22) and (19) n xi = 1 is equivalentto 2: w(xi) = 1 i =2 * = n =O. Then(24) obtains.Put x2= x, x3 = YX4=1-X-yX6= gives w(x) + w(y) = 1- w(1 - xy). So w(x) + w(y) dependson x + y only. Replacingx, y by x + y,0 shows,that it is equal to w(x + y) + w(O) = w(x + y) (use 20). So we have: (25) w(x) + w(y) = w(x + y) This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions 842 GARRETT BIRKHOFF AND JOHN VON NEUMANN (25), (19) and (22) give together: ofF. antisomorphi8m w(x) is an involutory Observe,that (25) impliesw(-1) = -w(l) w('yi) = (26) =-1, and so (23) becomes 'vi. 7. Consider a: [x1:... : Xz]r, a': [yi:... : yn]. If xi 5 0, we may write a: [1:x2x': ... :xfx1],r,and so a': [1:w(x2X')-Y2: ... :w(nx71)'Y,],. But Yi = W(Xl)-Iw(xi)-y;, w(xi~xT)'vi= W(XT1)W(xi) and so we can write a': [w(x1):W(X2)*2: ... W(X:)-Ynl too. So we have fori = 1, ... yi = w(xi)'vi (27) n, wherethe -Yifori = 2, * , n are those from6., and 'vI = 1. And w(1) = 1, (27) with'vj so (26) holdsforall i = 1, * , n. So we have the representation 0. obeying(26), ifxi 5 Permutationofthei =1, *... , n shows,thata similarrelationholdsifx2 5 0: yi = w+(xi)iT+, (27+) 'vit W+(uS) = (26+) of F. (w+(x), -ytmay differfrom w+(x) being an involutoryantisomorphism now 1 have we = of w(x), Syi!) Instead -yv y+ = 1, but we willnotuse this. = a': Then [yr: 1. all Put .*:y,]z can be expressedby both formulae xi so w(1) = w+(1) = 1, (27) and (27+). As w(x)lw+(x) are both antisomorphism, ...**zl : :* obtains. Thus ['= and therefore[yi: ... :y]l = [-I: ... *: (-I+l)-1- (yi)-f i = 'i,'i+ = - ye+7ifor i = 1, ... , n. but as we are Assume now x2 5 0 only. Then (27+) gives yi = w+(xi)'y+i, dealingwithleftratios,we mayas wellput Yi = (+4)-1w+(xi)yt = ('1+) Yw+(x)'y+ty,. Put = 'v 9+ 0, thenwe have: ei yi (xi)-+ W+ (27++) Put now xi = X2 = 1, X8 = X, all otherxi = 0. Againa': [yi: ...* yn]lcan be expressedby both formulae(27) and (27++), again w(1) = w+(1). Therefore [y1 y2: y y4:... :yn1l = [l:-Y2:W(X)-Y3:0 :... ?]1 = [v1:'2:t+-W+(X)f'+Y:0: .. ?]1 obtains. This impliesw(x) = ft+w(x)t+ forall x, and so (27 ) coincideswith (27). This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions LOGIC OF QUANTUM MECHANICS 843 In otherwords: (27) holds forx2 z- 0 too. Permutingi = 2, ***, n (onlyi = 1 has an exceptionalr6lein (27)), we see: (27) holds if xi 5 0 fori = 2, * , n. For xlI5 0 (27) heldanyhow,and for somei = 1, ... , n we musthave xi 0O. Therefore: (27) holdsforall pointsa: [xi: *-:- x4],. 8. Consider now two points a: [xi: ... :x4] and b: [ti - n.. Put Xthenb ? a' means,considering(1) and (27) (cf. the end of 7.): a'i:[y.... : yn,1 n (28) E i-1 a ? a' can neverhold (a ifall xi = O. Thus, (29) u(xi)yi(i = 0. n a' = 0, a # 0), so (28) can only hold forxi = (i, n = 0 impliesxi = Ad w(x1),yix i-1 *-- = Xn = 0. Summingup the last resultof6., and formulae(26), (29) and (28), we obtain: Thereexistsan involuitory w(x) ofF (cf. (22), (25), (19)) and a antisomorphism n definite diagonalHermitianform E w(xi)yi i in F (cf. (26), (29)), suchthatfor a: [xl:*. (28) b: [i: :Xnl]7, .*. i-1 ] b < a' is defined bypolaritywithrespecttoit: n A w(xi)yi{i = 0. This is exactlythe resultof ?14,whichis thusjustified. THE SOCIETY OF FELLOWS, HARVARD UNIVERSITY. TEns INSTITUT FOR ADVANCED STUDY. This content downloaded from 131.252.130.248 on Mon, 23 Dec 2013 13:10:10 PM All use subject to JSTOR Terms and Conditions
© Copyright 2026 Paperzz