Annals of Mathematics

On One-Parameter Unitary Groups in Hilbert Space

Author(s): M. H. Stone
Source: Annals of Mathematics, Second Series, Vol. 33, No. 3 (Jul., 1932), pp. 643-648
Published by: Annals of Mathematics
Stable URL: http://www.jstor.org/stable/1968538 .
Accessed: 23/07/2013 13:14

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 support@jstor.org.

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 128.148.252.35 on Tue, 23 Jul 2013 13:14:11 PM

All use subject to JSTOR Terms and Conditions
 ON ONE-PARAMETER UNITARY GROUPS
IN HILBERT SPACE.'
BY M. H. STONE.

In this note, I give the detailed proof of a result(TheoremB below)

whichI announcedin theProceedingsoftheNationalAcademyofSciences,
16 (1930), pp. 173-4 and whichI originallyintendedto publish in my
forthcoming book, "Linear Transformations in Hilbert Space and Their
Applications to Analysis" (ColloquiumPublicationsof theAmericanMathe-
maticalSociety). The chapter on group theorywas completedin May,
1930, but was laterexcludedfromthe book. Recent applicationsof this
theoremmake furtherdelay in publicationhighlyundesirable. J. v. Neu-
mann,forexample,has foundit necessaryto generalizethe statementof
the theoremand has given a new proof,whichappears in this numberof
the Annals.2 My own methodseems to me to be a "correct" one from
the pointof view of the theoryof groupsand theirlinearrepresentations;
and in any case, as I had observedin my originalmanuscript, it yields
withoutany modification the generalizedformof the theoremrequiredby
v. Neumann(TheoremC below). The numbereddefinitions, theorems,and
lemmasto whichI referin these pages, will be foundin my book.
THEOREM A. If H is a seif-adjointtransformation in abstractHilbert
space,D, thenU(T;) eiTH < < + ct, is a family of unitary trans-
-

formations withthegroupproperty
(a) U(O+i?) U(@) U(i)
and thecontinuity
property
(Ai) U~r) U(i) wvhen a -> v (Def. 2.3).

The existenceof U(r) ei= H is establishedinTh. 6.1, its unitarycharacter

followsfromTh. 6.6. The groupproperty (a) is a consequenceofTh. 6.1,
(6), (7). The continuity property(,8) followsfromthe relation

I(UW() - U(T))f12 = Wea - eiTXI2diE(I)fI2,

whereE(s) is the resolutionof the identitycorresponding

to H, as we
have pointedout in Th. 6.1 (5) and Th. 6.2.
1ReceivedMarch19, 1932.
2 Vide supra,pp. 567-573.
643

This content downloaded from 128.148.252.35 on Tue, 23 Jul 2013 13:14:11 PM

All use subject to JSTOR Terms and Conditions
644 M. H. STONE.

THEOREM B. If U(r), - o <r < + x, is a family of unitary trans-

formationswith the groupproperty(a) and thecontinuity property(ai), then
thereexists a unique self-adjoinettransformationH such that U(T) eiTH.
of
property(ai) impliesthe continuity
It is evidentthat the continuity
f and q in D, by virtueof
(U(r)f, g) as a functionof r for arbitrary
Def. 2.3 and Th. 1.3. Hence we may generalizeTheoremB as follows:
THEOREM C. If U(r), - x <r <+ x, is a family of unitary trans-
formationswith the group property(a) and the measurabilityproperty
(y) (U(T)f, g) is a measurablefunction of T for arbitraryf and 4 in lS,
H such that U(r) =-ei rr.
thenthereexistsa unique self-adjointtransformation
TheoremA shows,therefore, that thegroupproperty(a) and themeasurability
property(y) togetherimply the continuityproperty(O).
Theorem
We now proveTheoremC. First, by the Fourier-Plancherel
formulas
(Th. 3.10), the reciprocal

(la) P; 1) 21 J r e-iTd

(lb) )1l J ip(i; l)etirdT

are valid when el(1) f 0, and lead to the relation

X+0
(2) (l-m) f? (T; 1)tP(-;r M) d = t/;1)-t(1f; m)

when the unitarycharacterof the Fourier-Plancherel transformation is

coupled with equation(1 a) and the equationobtainedfromn(1 a) by the
substitutionof a - r for T. We find

V('; 1) 0, r>0, P(l) >0; t~;)ie ilr, 0,J()O

=(T; i
ie-ilrT, z>0, y(l)<0; Pt(T;) 0, T<0, PM(<0;

since (I b) is obviouslyvalid with this choice of At. We findalso that

(4) My(T;1) = (-T; 1).

A more completediscussionof some of these relationsis givenin the

proofofTh. 10.9. Secondly,we definea boundedlineartransformation
Xi
withdomainQ by the equation
+0
1)J t(~ ) ( U (r) fj ) dr
(Xzf~ (1) +O.

The integralis to be interpretedas a Lebesgue integralin accordance

with the measurabilityproperty(y). By the unitarycharacterof U(r)

This content downloaded from 128.148.252.35 on Tue, 23 Jul 2013 13:14:11 PM

All use subject to JSTOR Terms and Conditions
 GROUPS IN HILBERT SPACE. 645

we have I(U(T)f, g) I _ I U(T)fI IgI = If I IgI this in-

By combining
equalitywith(3), we findthat the integralis absolutelyconvergent and
does not exceed If I IgI/IaJ(l) in absolutevalue. We can therefore
apply
a theoremof Fre'chet(Ths. 2.27-2.28) to establishthe existenceof the
desiredtransformation Xi. Thirdly,we obtainessentialpropertiesof Xi
correspondingto (2), (3), and (4). The firstis
(2 bis) ( -m)XIXm X, Xm, 0(l) t 0, aJ(m)t 0.
of Xi and the groupproperty(a), we have
From the definition

(XI Xmf,g) = (Xmf, Xl'%) =_ t(e; m)(U(e)f, XI*g) de

- /(e; m)(Xi U(e)f, g)de

- 00?(; (e;m)(U(r)U(e)f,g) drde

-JJ I)V(e;m)(UMe+,)f, g)drde.
tV(r;
This integralis an absolutelyconvergent Lebesgue integralby virtueof
the measurabilityproperty(y),3 the relations (3), and the inequality
J(U(e+?T)f,g) <Ifj IgI. Hence we can introduce newvariablesc e=+
and T = T and can integratein any convenientorder. If we perform the
integrationwith respect to the new variabler first,we obtainwiththe
aid of (2) the result
(i-m) (XiXmfg ) =J (p (or;i)-(or; m))(U(') f, g) d = ((Xi - Xm)fg),
whichevidentlyimplies(2 bis). The second propertyis
(3 bis) Xmf = 0 implies f = 0.
0 impliesXjf = 0 for all not-real1
f
From (2 bis) we see that Xmf
and henceimpliesthat lf(T; ) (U(T)f, g) dr vanishesforall not-realI
and all g in &. By makinguse of the explicitformof V given in (3),
we conclude'that(U(T)f, g) = 0 for almostall r and all g in ). Hence
3 It is to be notedthat if f (T) is a measurablefunctionofT, -() <r<+OO, then
F(e, T) f (Q+ T) is a measurablefunctionof (Q, T).
4We have to show that the equationJ e'? F(T) dr = 0, holdingforall I such that
0 (1)>0, implies F(r) = 0 almost everywhere, T >O, wheneverF(r) is bounded and
measurable. If we put I = p + i', v>0, and hold v fast,we can obtainthis resultby
a simpleapplicationof the Fourier-Plancherel
Theorem.

This content downloaded from 128.148.252.35 on Tue, 23 Jul 2013 13:14:11 PM

All use subject to JSTOR Terms and Conditions
646 M. H. STONE.

we see thatU(T)f 0 for almostall T and thatf- 0 by virtueof the

unitarycharacterof U(r). The thirdproperty is

(4 bis) XI - X, J() 4 0.
The group property(a) impliesthat U(O) I, U(-u) = U-1 (a) = Us (r).
Hence we obtainfrom(4) the relation

I, g) - (XigfJ)
(XY* fq(r; 1) (U(r)g, f)d

-JXiw(-,; 1,U*(r) f,y)dr

P+m
-f (T;i )(U(r)f, g) dT

(Xif, 9),
which holds for all f and g in S) and therefore implies(4 bis). In the
fourthplace, we observethat, in accordancewith Th. 4.19, the three
propertiesofXI whichwe havejust provedenableus to assertthe existence
of a unique self-adjointtransformationH such that Xi (H I)-' for
all not-real
1. Finally we must show that the transformationV(W)= eiTH
coincideswithU(0). Accordingto TheoremA, the familyV(T) satisfies
the hypothesesof TheoremsB and C. Hence we can introducethe trans-
formationY1 definedby the relation
+tch
(Y f7 g) J 00(T; l(V( )f, g) d.
By Th. 6.1, we have

(Yif, g) lim U p(T; I)(U eikT

d(E(A)f, g))dT
a i+X )00 j4ext (;) ) Ei),

lim 00 a;tTV(r )( d (i) (A) f,

a-*+oo -a0 e-O
i
lim ;? d (E (i) f, g)

=_1Je~
Xd-;E(E())f)g) ~

whereE(X) is the resolutionof the identityassociatedwithH, and the

+ sign or the - sign is taken accordingas el(1)<0 or el(1)> 0. Now
Ths. 5.7 and 6.1 show that YJe(H- 1-1 = Xi. Hence we have

This content downloaded from 128.148.252.35 on Tue, 23 Jul 2013 13:14:11 PM

All use subject to JSTOR Terms and Conditions
 GROUPS IN HILBERT SPACE. 647
P+ 0
fo P(r; 1) ((U(T)f, g)-(V(T)f, g)) dr 0
and can concludethatU(T) = V(W)for almostall r. If r is fixedand e
varies in the set specifiedby the identity
U(e) -= V(), thenthedifference
a T - e varies over the entire range - o <a< + oo with the possible
exceptionof a null set. Hence a assumesat least one value such that
U(a) = V(a). We see therefore that
u(r)- U(e + d) = U(Q) U(a) = V(W)V(a) =- T(Q + a) = V(r)
for arbitraryi and suitablychosen e and a, by virtue of the group
property(a). This completesthe proof.
and U(r) = eigH,
THEOREM D. If H is a self-aqjointtransformation

thentherelation-U(r) - f* is valid in S if and onlyif f is in

thedomainof H and f* = i Hf.
If f is in the domainof H, the integral J 2 d E ()fJ2 exists by
Th. 5.9. From Th. 6.1 (5) and Lemma6.1 (6) we see therefore
that
lim H-eirH- 2 1- eir_1 2
TAO~ ~~ TAO
ei -o 2T

l1 212 (t)1J1
_____ dIE

() - U() f f* =
iHf whenever
f is in the
Thisresultmeansthat
domainof H. On the otherhand, we can definea transformation
T by
the relationiTf = f*, valid whenever (1(7) U(0) f convergesin -
to an elementJf. Then T is an extensionof H withthe propertythat

(Tf g) 1 ( U(T)U(0) f, q)

lim(f, U( T)- U(O) -(fTg)

whenever f and g are elementsin its domain. Hence T is a symmetric

extensionofH. SinceH is self-adjoint, we musthave T = H in accord-
ance with Th. 2.13. For a modification of this theorem,we referto
v. Neumann'spaper in this numberof the Annals.
THEOREME. A boundedlinear transformation T is permutable(Def. 8.1)
withtheself-adjointtransformationH if and onlyif it is permutable
with
everytransformationof thefamilyeiTH, X <, < + m. -

This content downloaded from 128.148.252.35 on Tue, 23 Jul 2013 13:14:11 PM

All use subject to JSTOR Terms and Conditions
648 M. H. STONE.

From the integralrelationconnecting (H- 1)-1 witheITH,it is evident

that T is permutable witheverytransformation eYTH if and onlyif it is
permutable with(H - )-1 for all not-real1. As we proved in Th. 8.1,
T is permutable with (H- 1)-1 for all not-real1 (or, for a singlesuch1)
if and onlyif it is permutablewith H.
Because of lack of space, I do not go intocertainothergeneralizations
of the theoremsprovedhere. I pointout that the resultsobtainedlead
easily to a characterization of all r-parameterabelian unitarygroups
U( T. . . .. . e n, whereH,, ..*, H. are permutableself-
adjointtransformations (cf. Def. 8.2). A less obviousgeneralizationdeals
withthe familyof isometrictransformations U(T), 0 ? r< ?+o, where
U('?+r) = U(a) U(T) for a > 0 r > 0. I hope to state this problem
in preciseformand to studyits solutionon some otheroccasion.
YALE UNIVERSITY.

This content downloaded from 128.148.252.35 on Tue, 23 Jul 2013 13:14:11 PM

All use subject to JSTOR Terms and Conditions