You are on page 1of 14

A THEOREM OF STONE AND VON NEUMANN

By

GEORGE W. MACKEY

1. Introduction.

Stone in [15] has stated and indicated a proof of the following

THEOREM. Let C be a Hilbert space and let P1, P2, Q,, P QI Q,. 1, 2, ...) denote the members of a set of 2n self adjoint operators in C. For each real number and each j n let U exp (iaPi) and V exp 1, 2, --1. Suppose that this system of operators is irreducible in the (iaQi) where i

(n

sense that there is no proper closed subspace of C which is carried into itself by all of the U and all of the Vi Suppose finally that the following permutation relations are satisfied. For all j, k, a and r, U,U U,U, and V,V, V,V, For all V,U, For all j, a and r, U,V, exp (-ir) j, k, and r with j k, V$ U$. Then there exists a one-to-one linear norm preserving transformation S of onto the Hilbert space of all complex valued functions of summable square on Euclidean n space of such a nature that SPS-lf (x x2 xn) iOf (x, x2 xn)/Ox and SQS-If(xl x x) xf(x, x x,).

UV

In [8] von Neumann has given a proof of a sharpened form of this result. He omits the hypothesis of irreducibility and proves that is the direct sum of at most countably many invariant subspaces for each of which the above conclusion is valid. Rellich in [12] has recently given a proof of a variant of Stones theorem in which the assumed permutation relations involve the P and the Q directly. In the present paper we prove two related theorems. One of these (Theorem 1) is about generM separable locally compact Abelian groups and reduces to a theorem equivMent to the Stone-yon Neumann result when the group is taken to be the additive group of Euclidean n space. It is possible to state Theorem 1 in such a form that with a slight change in the hypotheses it has meaning for locally compact groups which are not necessarily Abelian. While Theorem 1 is not a special case of the resulting Theorem 2 it is an easy consequence of Theorem 2 and well-known theorem on the unitary representations of locally compact Abelian groups. The argument used in the proof of Theorem 2 is closely related to and was in part suggested by the argument given by Gelfand and Neumark [4] in analyzing the irreducible unitary representations of the non-commutative two parameter Lie group. In fact Theorem 2 may be used to obtain the irreducible unitary representations of the members of a class of groups which includes that discussed by Gelfand and Neumark. This and related mtters we expect to discuss in another paper. The statements of Theorems 1 and 2 will be found in 2, and their proofs in 6 and 7. The other sections contain preliminary materiM of various kinds.
Received November 2, 1948.

313

314

GEORGE W. MACKEY

The material in 4 is closely related to the theory of direct integrals of Hilbert spaces which has been developed by von Neumann [9]. We are indebted to Professor von Neumann for having given us the opportunity of looking at this paper in manuscript and for having called our attention to its applicability to a problem similar to the one discussed in this paper.
2. Statements of the principal theorems. It is easy to see that the permutation relations in the Stone-yon Neumann theorem are equivalent to the following. For every set of 2n real numbers n, r r2, 2, rn, UI,,U. U:.V,,V. V, exp (-i(#,r, #2r V,U,U." #,r,))V,,V,. U, U, U$, and the mappings x,U,, U2, and *a 2 -* V, r regarded from as the mappings r r2, rl V:. additive group Gn of Euclidean n space to the group of unitary operators in .are homomorphisms. Moreover G, is isomorphic to its character group G in such a manner that the character function (, r) for t G and 1, r, G: is given by the expression (, r) r exp (i(,r r, r2, .r2 nr)). The fact that U, and V, are generated by self adjoint operators in the manner indicated implies (and is implied by) the fact that U,f.g and V,f.g are continuous functions of, and r respectively for each choice of f and g in Thus the hypotheses of the Stone-von Neumann theorem may be expressed entirely in terms of the notions of the theory of locally compact Abelian groups. (The reader is referred to [11] and [16] for expositions of this theory.) The same is true of the conclusions. Because of the well-known manner in which the self adjoint operator P and the one parameter subgroup exp (iP) determine one another it is easy to see that the conclusion in question is equivalent to one which differs from it only by making ssertions about the operators SU, S and SV, S instead of about the operators SPAS-1 and SQiS-Z. These assertions are of course to the effect that SU, S takes f (x x x) into f (x rl x r. x) into exp (i(rx r2x x ) and SV, S takes ](xl x x). It is now clear how the Stone-yon Neumann rnx))f(x x. theorem may be regarded as a special case of a theorem about locally compact Abelian groups. If we make the trivial but convenient substitution --, -1 and define a unitary representation of a locally compact group G in a Hilbert space as a homomorphism U ( U,) of G into the group of all unitary transformations of into itself which has either and hence both of the equivalent (see [6; footnote 12]) properties (a) U,f.g is a continuous function of for each f and g in and (b) U,f is a continuous function of for each f in we may state this theorem az follows.

V.V....

.,

--

- + +

THEOREM 1. Let G be an arbitrary separable locally compact A belian grou Let G be its character group. Let U ( ---, U,) be an arbitrary representation of G in the Hilbert space 5C. Let V (r V,) be an arbitrary representation of G in Suppose that U and V are so related that U,V, (r, r)V,U, for all r , G and all r , G Then 5C may be represented as a direct sum of at most countably

THEOREM OF STONE AND YON NEUMANN

315

many closed subspaces each of which is carried into itself by each U and each V, and for each of which there is a one-to-one linear norm preserving map S onto the Hilbert space of all complex valued functions on G which are square summable unth respect to Haar measure of such a nature that SUS-f (x) f (xr) and SVTS-1f (x)

(x, )f().
Theorem 1 takes on a somewhat different aspect if we look upon V as a mapping from a class of functions on G to the bounded linear operators on rather than as a representation of G n Since the members of G are linearly independent as functions on G, there is a unique extension of V, to the ring of functions generated by the characters, which is a linear ring homomorphism. If h is defined by the equation h(x) cl(x, rl) c2(x, r2) cn(x, rn), where the c are complex numbers and the ri members of G we have Vh clVT, c2V,. c,V,.. An easy computation leads to the equation UVh V,U, where h, is the function such that h,(x) h(xa). Thus the permutation relations may be put into a form independent of the theory of characters. Applying the generalization to locally compact Abelian groups of Stones well-known theorem on the unitary representations of the real line it is possible to extend V still further so as to be defined on the ring of all bounded complex valued Borel functions defined on G. The resulting map has the expected algebraic properties and in addition a certain continuity property described in detail below. Thus in order to prove Theorem 1 it is sufficient to prove a variation of it in which the representation of the character group is replaced by a certain kind of representation of the ring of bounded Borel functions on G. In this form however the theorem may be stated for non-Abelian groups and it is in this form that we shall prove it. In other words Theorem 1 is a consequence of the generalized Stone representation theorem and the following theorem about general separable locally compact groups.

n,

THEOREM 2. Let G be a separable locally compact group. Le U ( U,) be a representation of G in the Hilbert space C. Let V (h V) be a linear homomorphism into the ring of all bounded linear operators on 5C of the ring of all bounded complex valued Borel functions defined on G which has the following additional properties: () If h(x) is the complex conjugate of h2(x) for all x in G then V, is the adjoint of V. (b) If h h2 are the characteristic functions of disjoint Borel subsets of G and h(x) then V, converges h(x) h.(x) Vh. to V in the weak (and hence also in the strong) operator topology. Suppose that U and V are so related that UV V,U, for all r G and all bounded Borel C h. as a direct sum of at most countably many be Then may represented functions closed subspaces each of which is carried into itself by each U, and each V and for each of which there is a one-to-one linear norm preserving map S onto the Hilbert space of all complex valued functions on G which are square summable with respect to right invariant Haar measure of such a nature that SU, S-f(x) f(xa) and

...,

SV,S-f (x)

h(x)f (z).

316

GEORGE W. MACKEY

We conclude this section with some remarks about the meaning of Theorem 2. Let 2(G) denote the Hilbert space of functions on G square summable with
respect to right invariant Haar measure. Each a in G defines a unitary operator U in 2(G) such that Uf(x) f(x) and each bounded Borel function h on G defines a bounded operator V such that V,f(x) h(x)f(x). Since the mapping -* U is one-to-one we may regard the mapping a --* U, of Theorem 2 as a mapping from the U to the U,. Moreover it is not difficult to show directly from Lemma 3.3 without invoking the rest of the proof of Theorem 2 that 0 for almost all x then Vh 0. Thus the mapping h --* Vh whenever h(x) may be regarded as a mapping from the V to the V. Consider the (algebraic) ring generated by the U and the V ;that is the set of all operators in 2(G) of the form U,V, U,V, U,,,V,,, One may ask whether or not the mappings U, --* U, and V V may be blended and extended to a homomorphism of this ring into ((), the ring of bounded linear An easy calculation shows that the permutation relation operators on U, Vh V,,U, is precisely the condition that this be the case. In other words Theorem 2 may be regarded as stating that any adjoint preserving linear homomorphism of the ring generated by the U and the V into (c) is a direct sum of at most countably many replicas of the identity representation provided that the given homomorphism satisfies a certain (unfortunately rather complicated and unnatural) continuity condition. In case G is finite it is readily verified that the ring generated by the U and the V coincides with ((2(G)) and Theorem 2 becomes the well-known theorem of algebra which says that a representation of a full matrix algebra over the complex numbers is a direct sum of replicas of the identical representation. To summarize: the Stone-yon Neumann theorem is a consequence of Theorem 1 and Theorem 1 is a consequence of the generalized Stone representation theorem and a theorem (Theorem 2) which may be regarded as an infinite dimensional generalization of a classical result on the representations of full matrix algebras. It naturally occurs to one to wonder whether the separate continuity conditions on U and on V may be combined into one natural continuity condition on the ring homomorphism which they generate and whether the domain of this homomorphism may be replaced by a larger one which is closed in some natural topology. These are questions which we have not yet investigated but hope to return to at some future time. In this connection we refer the reader to a paper of Segal [14], to the unpublished manuscript of which we have had access. Here he discusses a ring of operators on 2 (G) which is like ours in that it is generated by two classes of operators one involving multiplication by functions defined on G and one defined by means of group translation. It differs in that it is closed in the uniform topology, in that a smaller class of fllnctions is used, and in that group translation proper is replaced by convolution with functions defined on G. It turns out that an analogue of Theorem 2 is not in general true for this ring.

3. Some measure theoretic lemmas.

Slight variants of our first lemma are

A THEOREM OF STONE AND VON NEUMANN

317

proved in various places in the literature. (See [2], [3], [17].) Since in none of these places is it stated exactly as we need it, and since the proof is fairly short, we include it here for completeness. In this section and throughout the paper will be used to denote the characteristic function of the set E. the symbol

LEMMA 3.1. Let M and M. be measure spaces with measures and respectively. Suppose that each is the sum of countably many measurable sets of finite measure and that there is a countably generated Borel field of measurable subsets of M such that every measurable subset of M differs from some member of by a set of measure zero. Let f be a complex valued function defined on M X M. which is measurable and essentially bounded as a function on M for each fixed point in M. Suppose that f f (x, y)(y) d(y) is measurable on M for each measurable subset E of M. which is of finite measure. Then there exists a function f measurable on M X M such that for all x in M f(x, y) f(x, y) for almost all y in M Proof. We shall call a subset of M1 M. bounded if it is a subset of a set of the form E1 X E where E and E. are measurable subsets of finite measure of MI and M respectively. Let ff be the family of all bounded measurable subsets F of M X M such that f f(x, y)(x, y) d(y) is measurable on M It is readily verified that ff has the following properties: (a) If E and E. are measurable subsets of finite measure of M and M. respectively then E X E is a member of if; (b) If F where the F; are mutually disjoint F k_) F. k_) members of and F is bounded then F if; (c) If F lim; F (j 1, 2, ...) F_ where the F; are members of ff such that F then F ft. It follows from [13; 83, Lemma 9.4] that ff coincides with the family of all bounded measurable subsets of M1 X M.. For each F ff let ,(F) f (f f(x, y)(x, y) d2(y)) dt (x). It is clear that is completely additive and is absolutely continuous with respect to the measure X t2. It follows from the Radon-Nikodym theorem that there exists a measurable complex valued function g defined on M X M2 such that (F) f g(x, y) d (x, y) for all F fi;. Applying the Fubini theorem we are able to conclude that for all F 5:,

f (f f(x,

Setting F E E where E M and E M. we find that for each measurable E. M. we have f(x, y).(y) d(y) f g(x, y).(y) d.(y) for almost all x in M Applying once again the lemma from [13] referred to above it is easy to see that the set of measure zero on which this equality need not hold may be chosen so as to be the same for all E and hence for all measurable subsets of M.. It follows at once that for almost all x, f(x, y) g(x, y) for almost all y. Changing g on a set of tl X measure zero we obtain a function f with the desired properties.

__ _ __
y)y(x,

y)d(y))d(x)= f (f g(x, y)(x, y)d2(y))d(x).

LEMMA 3.2. Let G be

measure in G such that the measure

be a Borel a separable locally compact group. Let and every right set is finite of every compact

318

GEORGE W. MACKEY

translate of every set of measure zero is again of measure zero. Let f be a complex valued function defined and X measurable on G X G. Let g be defined on G X G X G by the equation g(x, y, z) f(xy, z). Then g is X a X measurable

onGGG.

Proof. Let T be the function from G X G X G to G G defined by the equation T(x, y, z) xy, z. We must show that T-I(F) is measurable whenever F is measurable. Since T is continuous and every a X a X a measurable set differs from a Borel set by a set of a X a X a measure zero, it is enough to show that X X (T-I(F)) 0 whenever F is a Borel set and X a(F) 0. Supposing F to be such a set, for each z G let F. be the set of all u G for which u, z F. Then a(F,) 0 for almost all z. Now x, y, z T-(F) if and only if x Fy (the right translate of F, by y-). Since F.y has measure zero whenever F. does, it follows that a(Fy -) 0 for a a almost all pairs y, z. Thus if is the characteristic function of T-(F) then for almost all pairs y, z we have (x, y, z) 0 for a almost all x. Since T-I(F), being a Borel set, is measurable, it follows from the Fubini theorem that T-I(F) is of X a X measure zero.

IEMMA 3.3. Let G be a separable locally compact group. Let (x be a non-identically zero Borel measure in G such that G is the union of countably many measurable sets of finite (x measure and such that every right translate of every set of (x measure zero is again of measure zero. Then is equivalent to right invariant Haar measure in G in the sense that each is absolutely continuous with respect to the other.

Proof. Since any measure with the countability property described in the statement of the theorem is equivalent to a finite measure, there is no loss in generality in supposing that a(G) is finite. Let E be any Borel set in G. Then E is a Borel set; and for all x and y, the set E of all x, y G X G with xy tE,(x, y) tE(xy-). Applying the Fubini theorem to the measurable function we find that

The left member, because of the right invariance of is equal to (E)a(G). Moreover since (xy -1) -l(yx -1) -1(y) we see that the right member is equal to ] a(E-x) d(x). Thus for each Borel set E in G we have (E)a(G) f a(E-x) d(x). The truth of the lemma is an obvious consequence of this 0. equation and the fact that (E) 0 if and only if t(E -1) We are indebted to L. H. Loomis for suggesting this proof of Lemma 3.3. Our earlier proof (of a weaker result) was much longer and more complicated.
4. A ring isomorphism. We begin with some notational conventions. If M is any measure space with measure a we shall denote by 2(M, a) the class of square summable complex valued functions defined on M and by (M, a) the

THEOREM OF STONE AND YON NEUMANN

319

class of essentially bounded complex valued measurable functions defined on M. We shall denote by (M, a) the Hilbert space associated with .2(M, ). If f denotes an element of 2(M, ) we shall denote by f- the corresponding member of C(M, a). If C is any Hilbert space we shall denote by (C) the ring of all bounded linear operators on C. If g t )(M, a) then Tg will be used to denote the member of (B((M, a)) such that To(f-) (gf) for all f .2(M a). Now let M be any measure space with measure such that C(M, a) is separable and M is the sum of countably many sets of finite measure. Let o be the direct sum of finite or countably many replicas of (M, a). For each g t 9(M, let T be the member of ((Co) such that T(f?, f;, ...) T(f-), To(f;), t Co. We define (R as the ring of all T t (B(o) such that for all f? f;, TT T,T for all g t (M, a). For definiteness we shall suppose from now on that Co is the direct sum of countably many replicas of C.(M, ). Everything that we say, however, will be true, with obvious modifications, when there are only a finite number of direct summands. Let Cl be the Hilbert space of all sequences of complex numbers cl, c2, such thatll,,c2, =-,Ic;I <. AfunctionAfromMto will be said to be measurable if A (x)sx .s is measurable as a function of x for each K for almost all x and some K > 0, pair s, s2 of elements of Cl. If ll A (x) II then A will be said to be bounded and the least K which will do will be denoted by A II. We shall denote by ff the set of all bounded measurable functions from M to (B(C). If A and A. are in then the functions A3 A4 and A5 defined respectively for all x t M by the equations A(x) A.(x.), A(x) A(x) A(x)A(x) and A(x) (A,(x))* are readily seen to be members of We shall denote them by Ax A2, AA2 and A* respectively. In this section we shall introduce and study a certain natural mapping A A- of ff onto (. Given For each j 1, 2, let e be the element of Cx whose k-th term is A t ff we define A t ((C(M, a)) as follows. First for each j and each k we define ai t (M, ) by the equation a(x) 1, 2, A(x)e.e. This done we define A on a dense subspace of Co by setting A-(f?, f;, f:, 0, 0, ..) where for each j h(x) _,= a(x)f(x). Then 1, 2, h? h; ]1 h- h; II 7-1 f =1 a(x)f(x) 12 da(x) f (_.= a(x)f(x) 12) da(z). But for almost all z, I] A(x) I I A II. Hence for almost all x, _,2-, a(x)f(x)12 <_ II A -1 f(z)12. It follows that h:, II I[ A I f _,.l f(x)12 da(z) I A I1 [] f;, f;, .-., f:, 0, 0, I[ 2 In other words A- is bounded and II A- [I [[ A I1 on the dense subspace of o on which it is defined. We extend A- to all of Co by continuit);.

-<

- .
...,

<-

<-

-<

IEMMA 4.1. For each A A- (. Proof. Let g be any member of 9r(M, ). An obvious calculation shows that TA-(f-[ f; f: O, O, ...) f: O, O, ...) A-T(f- f; for each member of Co of the form indicated. It follows from continuity that

...,

TA

ATe.

320

GEORGE W. MACKEY

LEMMA 4.2. If A

ff then

A"

if and only if A (x)


A

0 for almost all x.

Proof.

Obvious.
5 there exists

LEMMA 4.3. For each B

such that A"

B and][ A I]

and each f- C(M, a) let E(f-) be the mem1, 2, is component ber of o whose k-th f-.and whose j-th component is zero whenever k k. Let Bk (j, 1, 2, ..-) be the member of (C(M, a)) which takes fj of into the j-th component BEd(f-). It follows at once from the fact that TB BT that ToBi B, To for all j, k 1, 2, and all g 91Z(M, a). Hence for each j and k there exists bk 9(M, a) such that Tbk Bk Thus the fact that ll B(f: f; f: O, O, ...) ]1 <_ I] B I]II f: f; "", f: O, O, may be written in the form

Proof. For each k

...,

This inequality continues to hold if we replace each f by f where E is an arbitrary measurable set. It follows that we may drop the integral signs and conclude that
/=I
k=l

2) da(x)

b(x) f (x)

<- II B II

k=l

f (x)

12

for almost all x. This is true no matter what element f7, f;, f:, O, O, we choose from Co. In particular we may select a measurable set E of finite measure and n complex constants c, c2, c, and set f(x) c(x). But then the inequality becomes

for almost all x in E. Since M is a sum of countably many sets of finite measure it follows that the inequality holds for almost all x in M. For each such x we where for eachj c,, O, O, ...) defineA(x)(c, c. d d. 1, 2, d __,. b(x)c, Then A(x) as a linear operator on the dense subspace of C on which it is defined is bounded with bound not more than I1 B 11- Thus it may be extended uniquely to be a member of (3Ct) with bound not more than II B [I. We define A (x) for all x by letting it be zero whenever it is not already defined. Then for each j and k 1, 2, we have 1, 2, Thus for x. almost all is A(x)e,.e b,(x) A(x)e,.e measurable as a function

THEOREM OF STONE AND YON NEUMANN

Clearly of x. It follows easily that A is measurable and hence is a member of is It B. of a an B obvious calculation A consequence that AII II II II. Since in general I] A-- II ]i A ]1 we conclude finally that ]1 A ]1 I] B I! thus completing the proof of the lemma. The proofs of the next two lemmas are easy and will be left to the reader.

-<

321

-<

LEMMA 4.4. If A ff then (A-)* (A*)-LEMMA 4.5. If A Y and B Y then (A :t: B)-- ALEMMA 4.6. If A ff and B ff then (AB)" A~B ~.

+/-

B~.

be defined as in the proof of Lemma 4.3. Proof. Let Ek for k 1, 2, Let A(x)e.ek a(x) and B(x)e.e b(x). Then for f (M, a) we have where for j A~((blf) (bk2f)-, ...) h-l, h., 1, 2,--A--B~Ef am(x)b,(x)f(x) and the convergence is in the mean. On the h(x) other hand (AB)~E(f -) where ci(x) (clf)-, (cf)-, A(x)B(x)e.ei But B(x)e bki(x)e -]- b(x)e. W for k, j 1, 2, Hence A(x)B(x)e, where the convergence is pointwise. b(x)a(x) + b.(x)a.(x) + e Comparing these expressions we see that h(x) c(x)f(x) for almost all x It follows that A"B and (AB)-- coincide. and all j, k 1, 2, In order to state the final lemma of this section we need two further definitions. Let co be a one-to-one measure preserving map of M onto itself. Then for each A ff the function A defined by the equation A(x) A(co(x)) is clearly a member of ft. Moreover he member X of (B(Co) defined by the equation X(f-( f; ...) (h h; ...), where hi(x) f(oo(x)) for eachj 1, 2,

_,,-_

-,

....

....

is clearly unitary. The proof of the lemma which we can now state is routine and will be left to the reader.

L.EMM 4.7. If A Y then (A)- XA-XWe conclude this section with the remark that the ring isomorphism we have discussed here is a special case of a more general one. Let M and M. with measures a and a. satisfy the conditions demanded above for M and a. Then there exists a natural mapping A --. A of the bounded mea,urable functions A from M to 6((M, a.)) onto the ring of all members of (B(C(M X M, al X a)) which commute with all members of the form To where g I)E(M X M a X a) and depends on the first argument alone. The special case we
have just discussed is that in which M is countable and every point has measure one. The general case may be reduced to this one by making use of the obvious fact that C(M X M. a X a) admits a one-to-one linear norm preserving map onto C(M X M, a X aa) which carries the ring 6t associated with one onto that associated with the other.
5. Abelian rings of projections. As has been shown by Nakano [7] the theory of multiplicity of spectra for normal operators in a separable Hilbert

322

GEORGE W. MACKEY

space can be reduced to the theory of so-called Abelian rings of projections. Here we shall recall briefly certain facts about the latter theory which we shall need in the next two sections. The statements which we shall make are either explicitly proved in [7] or are readily deducible from results which are proved there. An Abelian ring of projections is a set (P of projections in a separable Hilbert space 3C such that (a) if P1, P. then PP. P2P (P, P -t- P2 PP. ( and PP and P 0 for i j then P1P2 (, and (b) if P P. ( where convergence may be taken in either the weak or the P1 P -[strong topology. We shall deal only with rings which contain the identity I. The property of being uniformly n dimensional (n 1, 2, ...) is defined for Abelian rings of projections in such a way that the following statements are true. A ( is uniformly one dimensional if and only if whenever (V is an Abelian (e. Let C be a Hilbert space and ring of projections such that ( ( then let (V be a uniformly one dimensional Abelian ring of projections in C. Let 3C be the direct sum of n (n 1, 2, ...) replicas of . and let (P be the set of ---, Pfl Pf. all projections in C of the form f f. where P Then ( is uniformly n dimensional. Conversely every uniformly n dimensional ( is unitary equivalent to one constructed in this manner. A ( is uniformly one dimensional if and only if there exists f such that the closed linear span of the Pf with P (P coincides with C. For such an f, Pf 0 if and only if 0. If (P is arbitrary then is uniquely a direct sum of closed subspaces P some of which may be zero dimensional, such that for each 3C(R), 3Cl, C, P ( and each j and such that 0 rela1, 2, Pf C for each f tivized to C is uniformly j dimensional. We shall refer to the canonical decomposition of C with respect to (.

6. Proof of Theorem 2. Each of the three following paragraphs contains one of the three principal steps in the proof. In the first we show that we need only consider the case in which a certain family of projections is uniformly n dimensional. In the second we show that we may suppose that C is the direct sum of n replicas of C(G, #) in such a manner tha the Vg are what they should be. In the third we show that the representation may be so altered that the U, also fall into line. Let denote a right invariant Haar measure in G. For each Borel set E let P be the projection V where is the characteristic function of E. Let E be Borel subsets of G. Then Ps,Ps, Ps,ns. and if E E then E P, P.. If the E; are disjoint then P,.... P,_, In particular it is clear that the set ( of all Ps is an Abelian ring of proG we have UPU (t. In fact jections. Moreover for each E and each U,PsUUV,U- V,,, (F Eo"-) by the assumed permutation relation. Since the canonical decomposition of with respect to ( is unique it follows that each C; is carried into itself by each U. Obviously each C; is carried into itself the hypotheses of Theorem 2 hold by each V,. Thus for each j 1, 2,

...

THEOREM OF STONE AND YON NEUMANN

323

with replaced by C;. It follows that we may assume without loss of generality For definitethat (e itself is uniformly n dimensional for some n o, 1, 2, ness we shall assume from now on that n --o. The same argument will hold with at most minor modifications for finite n. Since (P is uniformly o dimensional we may suppose that C is the direct sum of countably many replicas of a Hilbert space and that there exists a uniformly one dimensional Abelian ring (P of projections in C such that for each Borel G there exists a member P of (P such that for all fl, f2, in C we set E Since (P is uniformly one dimenhave PB(f f2 ") P(f l), P(f.), such that the set of all P(fo) has the whole sional there exists an element fo of for its closed linear span. We define a finite Borel measure a in G by setting o(E) Ps(fo)"fo Then it is readily verified that there exists a unique linear norm preserving map $1 of (G, a) on 3C such that whenever E1 E2, E are disjoint Borel subsets of G and c, c., cn are complex numbers we have

...

S((c,rhs, -}- c@s,

-+- cnrhs.)-)

clP,(fo) -I- c2P.(fo) -{-

Now since Ps,

U,-,PsU-d it follows that a(Ea) 0 if and only if a(E) 0. Thus we may apply Lemma 3.3 and conclude that there exists an everywhere positive measurable function s such that for each Borel set E, a(E) f 4,s(x)s(x) d(x). For each f- z C(G, a) let S2(f-) be the element of (G, ) defined by the function fs Then $2 is a linear isometric map of C(G, a) on (G, ). Thus S.S-; is a linear isometric map of on (G, ). Moreover it is readily verified that S2S-[PSS; is for each Borel set E G the operator taking f- z C(G, ) into (fb)-. Thus finally, extending V to measurable functions in the obvious fashion, we see that we may suppose that comes to us as a direct sum of countably many replicas of C(G, ) in such a way that for each g Z(G, ) and each we have V(f-;, f-, ...) (gf)-, (gf.)-, f-, f;, It is now clear that in order to complete the proof of Theorem 2 we have only to find a unitary transformation S of C(G, ) onto itself such that for each g ilrg(G, ) we have S-VoS Vo and for each G we have S-U,S W, where W is defined by the equation W,(f; f; ...) g( g; and g(x) Let Q, UW so that U, Q,W for fi(xa) for all x, G and j 1, 2, all , G. The fact that --* U, is a representation of G tells us that for all Q,W,,Q.W. which (since --* W, is also a repreand a in G, Q,W,. sentation) may be written

-,

c,,Pn(fo).

....

....

(,)

Q,, ,.

Q,,

Moreover the fact that the U, and the V, satisfy the same permutation relation that the W, and the Vo do tells us that Q,Vo VoQ, for all g e r(G, ). At this point we make use of the theory developed in 4 with G, and respectively taking the place of the M, a, and C0 of that section. By Lemma 4.3 each Q is of the form A for some A, e ft. Moreover since A(A)* (A,A*,)" 1 it follows that A,(x) is unitary for almost all x. Thus by suitable choice of A, we

324

GEORGE W. MACKEY

may suppose that A,(x) is unitary for all a and x. Let us denote A(x) by the symbol A(a, x). Then it follows from (*) above and the various lemmas of 4 that for each al and a2 in G

A (ala. x)

A (o"1, x)A (o., xol)

for almost all x in G. For each and x in G and each pair of positive integers j and/c, let ai,(o, x) A(a, x)e.ek Then a;k(a, x) is bounded and measurable in x for each fixed a, j and k. Now if f:, f, and g:, g;, are members of C then Q,(f[ f ...). (g g; ...) is a measurable function of a. This follows readily from the fact that we get a measurable function if we replace Q, by either U, or W 1. Let us choose a measurable set E of finite measure and /c set f: ;: or 0 according as n j or not and g: ;: or 0 according as n or not. Then

Q,(f, f;, ...).(g;, g;, ...)

f a,((r, x)E(x) dtt(x).

It follows from Lemma 3.1 that A may be chosen so that a; is a measurable function on G X G for each pair of integers j, /c. Thus A is measurable as a function from G X G to 3Cl Hence for almost all x, A (, x) is measurable as Lemma 3.2 (applied to the a.) tells us that A(alz. x) and a function of A (., xal) are measurable as functions of the three variables x, 1, It follows G G, and hence that for readily that (**) holds for almost all triples in G Thus there exists Xo in G such almost all x it holds for almost all pairs 1 that A (, Xo) is measurable as a function of and

..

A (r xo)A (0"2, XoO) Let B be the member of ff such that B(x) holds for almost all pairs iT iT A (xx, xo) for all x G. It follows then from (***) that B(xy) A (XIxy, Xo) A(xx, xo)A(y, x) for almost all pairs xlx, y. Hence for almost all x, y we have B(xy) B(x)A (y, x). Thus for almost all G it is true that B(xo-) B(x)A (o-, x) for almost all. x. Operating on both sides with and applying Lemma 4.7
A (a(r., Xo)
xa we find that for almost all r, to the measure preserving transformation x WB"W-21 B"Q B"UW-21. Thus for almost all W., B"UB "-. Hence for all W B"UB "-. But both sides are weakly continuous in Finally B"Vo VB- for all V and since each B(x) is unitary B- is unitary. B and the proof of the theorem is complete. Thus we may take S

7. Proof of Theorem 1. We apply the Ambrose-Godement-Neumark (see [1], [5] and [10]) generalization of Stones theorem on the representation of one V,. It tells us that there is parameter unitary groups to the representation a mapping E P of the Borel subsets of G into the set of projections in 3C such
are disjoint Borel sets then PiP that whenever E1 E2 and such that for all f and g in and PE,,... P, -t P, -t

,,

0 for j

(Vf.g)

(x, ) d(P.f.g). Here the symbol d(P,f.g) means as usual that (x, r) is to be integrated as a function of x with respect to the set function E (Pf.g). For each bounded Borel function h defined on G let Vh be the unique bounded linear operator on C such that (Vhf.g) f h(x) d(Pf.g) for all f and g in C. It follows from Theorem 2 of the paper of Ambrose referred to above that the mapping h V satisfies all of the conditions laid down in the hypotheses of Theorem 2 of the present paper except possibly for the permutation relations. However evaluating the expression U, VTU-lf.g in two ways first using the fact that UVU V(Ulf) (, r) V, and then the fact that UV,UIf .g (Ulg) we are able to conclude that

A THEOREM OF STONE AND VON NEUMANN

325

-,

f (xa, r) d(P,f

g)

f (x, ) d(U,P,U;If

g)

U,PU for all a, G and all for all f, g and r. It follows easily that PB,-, Borel sets E. This combined with the definition of Vh leads us at once to the result that U, Vh Vh, U, for all G and all bounded Borel functions h. Thus all of the hypotheses of Theorem 2 are satisfied. Since the conclusions of Theorem 2 imply those of Theorem 1 the proof of Theorem 1 is complete.
Added in Proof. L. H. Loomis has recently shown that the hypothesis of separability in Theorems 1 and 2 is superfluous. He does this by giving a proof of Theorem 2 which avoids the measure theoretic preliminaries of the proof presented here as well as use of the spectral multiplicity theory. In the meantime the author has obtained a further generalization of Theorem 2. This generalization gives an analysis of systems U, V where V is a homomorphism of the ring of bounded Borel functions on the homogeneous space defined by a closed subgroup Go of G and U is as before. It reduces to Theorem 2 when Go is the identity. At this writing it seems doubtful that the method of Loomis will apply to this generalization.
REFERENCES

WRRE AMBROSE, Spectral resolution of groups of unitary operators, this Journal, vol. 11(1944), pp. 589-595. 2. S. BOCUNEa hND J. vos NEUMANN, On compact solutions of operational-differential equations, I, Annals of Mathematics, vol. 36(1935), pp. 255-291. 3. J. L. Doo, One-parameter families of transformations, this Journal, vol. 4(1938), pp. 7521.

774.
4. I. GELFAND AND M. NEUMARK, Unitary representations of the group of linear transformations of the straight line, Comptes Rendus (Doklady) de lAcadmie des Sciences de IURSS (N. S.), vol. 55(1947), pp. 567-570. 5. ROGER GODEMENT, Sur une gnralisation dun thorme de Stone, Comptes Rendus de lAcadmie des Sciences, Paris, vol. 218(1944), pp. 901-903. 6. ROGER GODEMENT, Les fonctions de type positif et la thorie des groupes, Transactions of the American Mathematical Society, vol. 63(1948), pp. 1-84. 7. HDEGOR5 NhKANO, Unitdrinvariante hypermaximale normale Operatoren, Annals of Mathematics, vol. 42(1941), pp. 657-664.

326

GEORGE W. MACKEY

8. J. v. NEUMANN, Die Eindeutigkeit der SchrSdingerschen Operatoren, Mathematische Annalen vol. 104(1931), pp. 570-578. 9. J. voN NEUM/NN, On rings of operators, Reduction theory, To appear in Annals of Mathematics. 10. M. NEUM.RK, Positive definite operator functions on a commutative group, Bulletin de lAcadmie des Sciences de IURSS [Izvestia Akademie Nauk SSSR], voh 7(1943), pp. 237-244. 1.1. L. PONTRJA(IN, Topological Groups, Princeton, 1939. 12. F. RELLICH, Der Eindeutigkeitssatz fir die LSsungen der quantenmechanischen ?ertauschungsrelationen, Nachrichten der Akademie der Wissenschaften in GSttingen. Mathematisch-Physikalische Klasse. Mathematisch-Physikalisch-Chemische Abteilung, (1946), pp. 107-115. 13. S. Sxs, Theory of the Integral, New York, 1937. 14. I. E. SEXL, A class of operator algebras which are determined by groups, Submitted to this Journal. 15. M. H. STONE, Linear transformations in Hilbert space. III. Operational methods and group theory, Proceedings of the National Academy of Sciences of the United States of America, voh 16(1930), pp. 172-175. 16. A. WEL, Lintgration dans les groupes topologiques et ses applications, Paris, 1938. 17. KSsxu YOSIDA AND SHIZUO KAKUTANI, Operator-theoretical treatment of Markoffs process and mean ergodic theorem, Annals of Mathematics, vol. 42(1941), pp. 188-228.

IIARVARD UNIVERSITY.

You might also like