Annals of Mathematics

Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Annals of Mathematics

Transformations of Weiner Integrals Under Translations


Author(s): R. H. Cameron and W. T. Martin
Source: The Annals of Mathematics, Second Series, Vol. 45, No. 2 (Apr., 1944), pp. 386-396
Published by: Annals of Mathematics
Stable URL: http://www.jstor.org/stable/1969276 .
Accessed: 29/04/2011 03:24

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .
http://www.jstor.org/action/showPublisher?publisherCode=annals. .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

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 The Annals of
Mathematics.

http://www.jstor.org
ANNEA OF MATHEMATICS
Vol. 45, No. 2, April, 1944

TRANSFORMATIONS OF WIENER INTEGRALS UNDER TRANSLATIONS


BY R. H. CAMERON AND WV. T. MARTIN

(Received September 20, 1943)

Introduction
In his paper on Generalized Harmonic Analysis (in which referencesto his
earlierwork are given) N. Wiener [I] definesan average or integralover the
space C of all functionsx(t) continuousin 0 ? t ? 1 and vanishingat t = 0.
The integralproved to be very useful in the theorydeveloped there. More
recently,the presentauthors[II] have shownthat the solutionof a fairlygeneral
non-linearequation can be expressedin termsof certainWienerintegrals. In
view of these ratherdistinctproblemsin whichthe Wienerintegralshave been
useful,it seems worthwhileto develop furthercertain aspects of the Wiener
integral. In the presentpaper we show how the integraltransformsunder a
translation,and we considerspecial cases of translationswhichseem to lead to
ratherinterestingresults. Our first(restricted)transformation theoremis as
follows:
THEOREM 1. Let F[y] be a functionaldefinedand WienersummableoverC,
boundedset,and let F[y] be
let F[y] be boundedin y(.) for y(-) in any uniformly
continuousin thesense thatif Iy(n)(t) is any sequenceof functionsof C which
converge uniformlyin 0 < t ? 1 to y(t) then
(1.1) lim F[y(n) - F[y].

xo(t) of boundedvariation
Let xo(t) be a givenfunctionof C witha firstderivative
in 0 ? t ? 1. Then underthetranslation
(1.2) y(t) = x(t) + xo(t)
the Wienerintegralundergoesthetransformation
W -f z[x't) I'dI t
J = I + -f 0x()d
0 d
(1.3) F[y]d y
et
F[x xo]e . x

We will recall a basic propertyof the Wienerintegral. Let n be a positive


integerand let H(,1q, ***, n) be a functionofn numericalvariables,j }, -* , qn -
Consider a functionalH[y(t1),* , y(tn)] for t , -, ttn fixed. If H is Wiener
summableand if 0 < t1 < t ?. <* _ <n 1, then by the theory,the Wiener
integralof H is given by

y(tg)] d

(1.4) -
n[wnt(t2
(t. 1. - t ..)]12 f;d1 ... d'qn{HQ,
( 'iXn)

2
~exp [- ('i6
[ 1('132-
(71_ v)2 (1_1_)2] I1)

386
TRANSFORMATION OF WIENER INTEGRALS 387

The idea to be used in the proofof Theorem 1 consistsin approximatingto


the functionalF[y] by some functionalwhichdepends only upon the values of
ytt)at some set of n points. This will enable us to use the relation (1.4) and
thus to work with the known transformation formulasof Lebesque integrals.
Our second (and more general) theoremis obtained by approximatingthe
characteristicfunctionalof a quasi-intervalby continuous functionals,thus
obtainingthe measureof a translatedquasi-interval. From limitsof sums of
quasi-intervalswe findthe measure of generaltranslatedsets and fromit the
translationformulaforunrestrictedWienersummablefunctionals.
THEOREM 2. Let xo(t) be a givencontinuousfunctionvanishingat t = 0
and havinga derivative x'(t) of boundedvariation0 < t < 1; let r be a Wiener
measurablesubsetof C, and let TF be thetranslationof r by xo(t); i.e., let TF
consistofall functionsx(.) ofC forwhichx(t) -y(t) - xo(t)and y(.) e F. Then

(1.5)
(1 s5) [rl =
~~measw[F] J1~~~f[
efo[0)]d
[w(j) ] 2dt
f W -2f
Tr
efo
Ix
I(0dx( (0)
dwx.

forwhicheithermember
Moreoverif F[y] is anyfunctional of (1.6) (below)exists,
thentheothermemberalso existsand theequation(1.6) holds:

(1.6) fr F[y] dwy = e fo t)l dt f F[x + xo]e2f


T~~~~~~~~~r
o 0 dzt dx.

2. Proofof Theorem 1
Let n be a positiveinteger,denote by tj the point

(2.1) 1i -
n
and forany functiony(t) ofC definea polygonalizedformofy(t) by the relations

(2.2) Yn(t) = y(tj) + y(ti41)- -y(t,) (t - I) for t t < ti+1


4+1 4. j =0, 1, - *, n - 1.
The subscriptn on a functiony(t),x(t), xo(t)etc.,willdenotethat thispolygon-
alizing processhas been carriedout on the function.
We formour functionalF[yn] forthe functionsof (2.2). Since yn(t)depends
only upon the values of y(t) at the n points t1, ***, tn, it followsthat there
existsa function

(2.3) H(71 , * 'n)

of n real variables ql, .**, 7 such that

(2.4) F[Yn] = H~ly(tj), ys, (t,)].

Next let M be any positive number. We definetwo subsets CM and C v,


of C. The set Cm consistsof all functionsy(-) of C forwhich I y(l) I ? M for
388 R. H. CAMERON AND W. T. MARTIN

all t in 0 ? t _ 1, and Cm,n consists of all functionsy(.) of C for


which I y&j/n)1 5 M forj = 1, ***, n. We note that
CMn CM,2, , lim CM.2n = CM
(2.5)
CM C CM, if M < M', lim Ck = C.
MW--ao

We willdenoteby TC m and TCM,,n the images of Cm and Cm,,underthe trans-


lation (1.2). For example, TCM consistsof all functionsx(.) of C for which
x(t) = y(t) - xo(t) and y(.) eCM.
Now by (1.4) and (2.4) we have

J F[y.] d. y
(2.6) cMu, ii

=7%
M
... H'(771 * * n)e
7
tj -j-2 qj-tl dili... d-q,,
where
,Y" V-n /2Rl(2 .l
W . . (tn-t_)

We next observethat if y(t) is the image of x(t) under the translation(1.2)


and if x,(t), xo,,(t)and y,(t) are the polygonalizedfunctionscorrespondingto
x(t), xo(t)and y(t) respectively,
then

(2.7) y,(t) = z,(t) + xo,M(t).


If we write

(2.8) ilj = y(t,),


I , = x(t,), aj = xo(t,), j = 0, 1, .* , n,
thenunder (2.7) the t's undergothe translation

(2.9) ?1j = tj+ aj,j


On applyingthe transformation
(2.9) to the n-foldLebesgue integralin (2.5)
we findthat

JCF[yM dy = y. dAi.i. __dn [H(Q1+ a *, ,t + a)


(2.10)
-e j (i-'.-. e i-i 'q-j-l *e - l

The nextstepis to interpretthisn-foldLebesgueintegralas a Wienerintegral


it thiswayand using(2.8) and thefactthat
(see (1.4)). Interpreting
_r.j =_ _
j,
_
zet z.
_ _
' ,
TRANSFORMATION OF WIENER INTEGRALS 389

we find
W [zg(ti)-zg(t1-l) 12 .W

Jcy F[y.] do y= e i-l - J F[xn + xo,.]


(2.11) TCM.S

*e s-2 d.x.
This gives us a transformation formulaover C M,, and hence over CA,2Mfor
the polygonalizedfunctionsunderthe translation(2.7). To obtain our desired
transformation formula(1.3) we let n approach infinity,(over the sequence
1, 2, 22, 23 ... ), and thenMl. We willuse thefollowingtwolemmas.
LEMMA 1. If yn(t) is thepolygonalized formofy(t) as in (2.2) then

(2.12) lim f F[y.]d- y= f Fly] d.y.

LEMMA 2.

lim E [xo(t) xo(t,-1)1+ 2 [xo(t) -xo(4Q-i) [x(tj) -(t'-')]


(2.13) - -t-

= IL1 [x4(t)j2 dt + 2 f X(t)


dx(t),
theconvergencebeingboundedin x(.) for all x(.) in any uniformlyboundedset.
REMARK: Since n now varies,it must be mentionedthat the points tj = j/n
vary with n. For the sake of simplicityin writing,we do not add another
index n to the tj.
Lemmas 1 and 2 whenapplied to (2.11) willyieldthe desiredtheoremat once.
3. Proofof Lemma 1
Let y(t) be a givenfunctionof C and let e be a positivenumber. Then there
existsan integerno = no(e)such that
E ~~~~~~~1
(3.1) jy(t') - y(t") j < forall t'- t"j -.

Withn 2 noand t, 5 t ? tar we have by (2.2) and (3.1)

2 V(t) y (t)l| j|t+1)


Y(4) It - t I + j
y(t) - y(t)j

(3.2) ~ ~ Sn..1 =(
1n- 2n 2
Hence
(3.3) Urmy.(t) = yi(t) uniformlyin 0 - ? 1,
a-,.
390 R. H. CAMERON AND W. T. MARTIN

and by the continuityof F (see (1.1)) we have


(3.4) lim F[y] = F[y].

Also, since by hypothesisF is bounded over Cm the convergencein (3.4) is


bounded fory(.) in CM. Thus (2.12) hOldsforeach fixedpositivenumberM.
This yields Lemma 1.
4. Proofof Lemma 2
For conveniencein writingwe define
(4.1) 4t[x(.); t] = xo(t) + 2x(t)
and

(4.2) Pjx ] E1 Xo(tj)- tj-


xX(tH..G1)
- t)] - 0x(); -ill.
j t4.
Then by the mean-valuetheorem
n
(4.3) Pn[x]= Ex0(t*) {[x(); tJ - 0W(); ti 11I
j-=1

forsome t* in tji- < t* < tj . By the definitionof the Stieltjes integral,since


x0(t) is of bounded variation and 4/[x(.), t] is continuousin t, this expression
Pn[x] approaches the integral

(4.4) .l dot4 ,F[x(*); t]

as n or. Thus, we have proved the convergence in (2.13). We have yet to


show that the convergence is bounded in x(.) for all x(s) in any uniformly
bounded set. This is quite easy. Let B be any uniformly bounded set of C.
This means that there is a constant K = KB such that

(4.5) I X(t) I < K for x(-) e B.

We apply Abel's transformation to (4.3); it yields


n-1
(4.6) Pn(X) = Xo(tn)O (Ix); i - EI {xo(t*+) - X?(t*') 4}x( ); tiI

Now 46[x(*); t] xo(t) + 2x(t) is bounded in. t and x(*) for 0 < t _ 1, x(.) e B,
and also xo(t) is of bounded variation in 0 < t < 1. Hence the Pf[x] are bounded
in n and x(.) e B. This yields Lemma 2.
On applying Lemmas 1 and 2 and the convergence property (1.1) of F to equa-
tion (2.11) we obtain
W W f|1 [z'(0)2d9-2f1z(9) dz(9)
(4.7) 1CMF[y]dwy= J F[x + xo]e ? ? d.x.

On letting M approach infinity we obtain the desired result (1.3).


TRANSFORMATION OF WIENER INTEGRALS 391

5. Proofof Theorem2
We begin our proofby eastblishing(1.5) and we firstdo this in the special
case in whichr = I, whereI is the quasi-interval
(51)~~~~~~~' n
(5.1) {< Y~,
y(t
j) ' ~ ~< ti <0 ... < <tn
(It is permissiblefor any t to be -c and any {, to be + co.) Now let
e > 0 and let jbe a continuous"trapezoidal" functionwhichequals zero
outside the interval e < v < t'i + e, equals unity inside the interval
to -

,t' <
',< and is linear on the remainingintervals. (If h, takes on the im-
propervalue - oo,so does -e, e etc.)
Let x'(y) be the characteristicfunctionalof I, (i.e., xl(y) = 1 if y e I and 0
otherwise),and let
n
(5.2) XiIe(Y) =llIe(Y(tD).
i-i
)
It is easy to see that F(y) = xl,,(y) satisfiesthe hypothesesof Theorem 1, for
boundednessand Wiener summabilityare obvious, and continuityin the uni-
formtopology followsfromthe continuityof the individual Hence j

(1.3) holds for XI,e and we have


I fw[-1' o( t )j]2d t W _ x; ( t )dz ( t)
e
(5.3) XJ,e (y) d. y = )]d Xi,e[x + xo~e fZ(~d~)d.. x.

But it is also obvious that for each y in C, limb-,oxr,[y] = Xi[yJ,for


if y e I, x',[y] = 1 forall e, while if y 4 I, at least one of the inequalities (5.1)
is not satisfied,and forsufficient
small e the corresponding (pj,(y(tj)) willbe zero.
Thus lettinge -> 0 and applying the principleof monotonicconvergenceto
(5.3), we have

] XJ~x+ xole_,
0
Xi(y) d.y =
ef 250 d. x,

whichis equivalentto (1.5) with r = I.


We next establish (1.5) in generalby progressivelyenlargingthe scope of r.
Since bothsides of (1.5) are completelyadditivefunctionsof r, it followsat once
that (1.5) holdsforany enumerableset ofquasi-intervalswhichhave onlybound-
ary pointsin common. But any measurableset P can be expressedas a set ro
of measurezero plus the limitof a decreasingsequence of sets IF D P2 D .*-
in whicheach Pk is the sum of an enumerableset of quasi-intervalswith only
boundarypoints in common. Applying(1.5) to each rk and taking the limit
as k -a o, we obtain (1.5) for r by monotonicconvergence. Thus (1.5) holds
in general.
Coming now to a considerationof (1.6), we note that
(154 E x2dt-2) I0(t)d(t)
(5.4) E(x)e-- ??
392 R. H. CAMERON AND W. T. MARTIN

is a Wienermeasurablefunctionalof x(-), and hence that the measurabilityof


F(y) on r impliesthe measurabilityofE(x)F[x + xo]on TP. We need therefore
only considerthe absolute value of the functionalso far as integrabilityis
concerned.
Let us firstconsiderthe case in whichthe functionalF is bounded and non-
negative. Let 0 < F(y) < M on r, whereM is an integer,and let us consider,
for& fixedpositiveintegern, thesets rk suchthaty e r1' if y e r and if
k
(5.5) -* F(y) < - k = 1,2, Mn.

Then we have

meaE6 (y) d. y -
(5.6) n (rk () ) fr: n meas, (rk),

and from(1.5), (5.4), and (5.5) we have

measw(rP) = | E[x] d,,x ? J F[x + xoJE[x]


dx
(5.7) n? f rk

k| E[x] d,,x - meats,(rk).


n Trk n
But (5.6) and (5.7) implythat

fr|F[y] 4y -J F[x + xo]E(x) 4 x | meas, (rP)


, Or, n
and we obtain on adding for k = 1, 2, *.,Mn
I
F[y] d4y - F[x + xoJE(x)dx ? measw(r).
fr for n
Lettingn -- A, we obtainthe equalityoftheseintegrals,and (1.6) is established
for non-negativebounded functionals.
Next take a non-negativefunctionalF whichis not bounded,and let FM(y)
nun (M, F(y)). Then we have (1.6) holdingforFM, and hence forF, by mono-
tonic convergence. Finally, if F is any real functional,the theoremholds for
I F I . Then if the integralsexist for I F I, they exist for the positive part of
F and the negativepart of F, and (1.6) holds forboth and hence forF itself.
This concludesthe proofof Theorem2.

6. An applicationofthe translationtheorem
By applyingthe translationformulasof Theorems1 and 2 to various special
cases of functional FMY]we are able to obtain formulasfor the evaluation of
various Wiener integrals. To illustratethis, we shall give a new proof of a
special case of a theoremof Paley and Wiener[III].
TRANSFORMATION OF WIENER INTEGRALS 393

LEMMA 3. Let a(t) be real and of boundedvariationon 0 _ t < 1, and let a


be so normalizedthat
(6.1) a(1) = 0.
Let
(6.2) A = i/f[a(t)]2dt
functiondefinedon -
and letG(u) be a (real or complex)measurable X0 < u < X .
Thena necessaryand sufficient conditionthat

G [f x(t) dac(t)]
be a Wienermeasurablefunctionof x(.) overC is that
e-u G(Au)
be ofclass L1 on - Xo < u < Xo. Moreover,
if thisconditionis satisfied,

(6.3) fW G[fx(t) da(t)]dwx= - ? e"G(Au) du.

Beforegivingthe proofof this lemma,we will derivea corollaryto it.


COROLLARY. Let ,B(t)be a realfunctionofclass L1 on 0 < t < 1, let

(6.4) A= 1i/f| [| ,) d{] dt

and let G(u) be a (complex)measurablefunctionon - 00 < u < mo. Then a


necessaryand sufficient
conditionthat

G [| (t)x(t) dt]

be WienersummableoverC is thateu2 G(Au) be of class L1 on - 00 < u < 00.


we will thenhave
Moreover,

(6.5) f G [f 3(t)x(t) dt]dwx = A L eu G(Au) du.

The corollaryfollowsfromLemma 3 by putting

(6.6) a(t) = -f () d{.

By specializingthe functionG in (6.5) we obtain various results. For exam-


ple, if G(u) = u2n we find

| [A %(t)x(t)
dt( x -0U2n du
({;.7) = (2n + 1)
~~~~~~A;
n
394 R. H. CAMERON AND W. T. MARTIN

In the next section,we prove Lemma 3.


7. Proofof Lemma 3
We carrythroughthe proofof Lemma 3 in successivestages.
STEP 1. We establish (6.3) for

(7.1) G(u) = e
where X is an arbitraryreal number. Take

(7.2) xo0) = fa a )d

and apply the translationformula(1.3) of Theorem1 to the integral 1 d,,y

fldwy
under the translationy(t) = x(t) + xo(t). Thus
W fc W -f (t) ]2d t-2fI 00
(
(7.3) 1= J ?. Lee l' (t) d(t)
xoe,

so that
0
(7.4) - fcWe d) x,

or
A2f1[a (tfl2dt W la( x(

(7.5) f7
W (efda
- (') d

A directcalculationyields
ax
v- L eU2eXAu du =
1 2A2
(7.6) 4

and thus (6.3) holds forG(u) = e).uwithX real:

(7.7) L e fo d.x = d u
e A2e>Adu.

STEP2. We nextestablish(6.3) forG(u) = e~uwith X complex. (While the


Wiener integralis originallydefinedforreal functionals,we can easily extend
the definition to complexfunctionalsby breakinginto real and imaginaryparts).
If A is real, we have (6.3) forG(u) = eAUand eAu and hence forG(u) = cosh
Au; the existenceof the integralis guaranteedby our translationtheorem,as
applied in Step 1. Now in the domain I X I < A, the function eXUis dominated
by 2 cosh Au. If we considerthe leftmemberof (7.7) as a functionof X and
TRANSFORMATION OF WIENER INTEGRALS 395

integratearound a contour,we are able to interchangethe orderof integration


by the mixed Fubini theorem. But the integrande 0 is analytic in
X foreach x(.) in C, and its integralaround a closed contourin the X-plane is
zero. Hence, by Morera's theoremthe integral
W exfs(t)da(t)

is analyticin X. Obviously,the rightmemberof (7.7) is analytic in X. Since


both membersof (7.7) are analyticin X and since they are equal forX real, the
equality (7.7) must persistforcomplexvalues of X. This completesStep 2.
STEP 3. We establish(6.3) whenG(u) is a real absolutelycontinuousfunction
vanishingoutsidea finiteintervaland havingan essentiallyboundedderivative.
Let g(v) be the Fouriertransform

(7.&) g(v)
e-iuWG(u)
= 2- L du.

Then g(v) is an entirefunctionof v and is of class L1 on the real axis, and


00

(7.9) G(u) = Legtg(v) dv.


00

Thus

WG [ x da(t)] d f{jj e: g(v)dv}d. x

= g(v) {f eivfxz(t)da (t) x}dv

= L: g(v){v- jj et ei' dt} dv

= L f e-2{f eitAtg(v) dv} dt

= 1 L e- G(At) dt,

and (6.3) holds forthe functionG of Step 3. (The changes of integrationare


justifiedby Fubini's theorem,sinceg(v) is of class L1, and the complexexponen-
tials are of absolute value 1).
STEP 4. We establish(6.3) whenG(u) is a real bounded measurablefunction
vanishingoutsidea finiteinterval. This is done by the use ofa smoothingproc-
ess on G(u) and the principleof dominatedconvergence. We omit the details.
STEP 5. We next establish the generaltheoremfor real non-negativeG(u).
This is establishedin the customaryway by use of the principleof monotone
396 R. H. CAMERON AND W. T. MARTIN

convergence. The existenceof one side proves the existenceof the other,and
hence,we obtain the desirednecessaryand sufficientconditionof Wienersum-
mability.
STEP 6. The case of general complex-valuedfunctionsis established by
combiningthe fourparts, real positive,real negative,pure imaginarypositive
and pure imaginarynegative.
This establishesthe generallemma,includingsummabilityof the integral.
Lemma 3 and the Corollary furnishmeans of calculating various Wiener
integrals.
THE MASSACHUSETTS INSTITUTE OF TECHNOLOGY
CAMBRIDGE, MASSACHUSETTS, AND
SYRACUSEUNIVERSITY, SYRAcusE, NEW YOR

BIBLIOGRAPHY
I. N. WIENER, "Generalized Harmonic Analysis," Acta Mathematica, 55 (1930), pp.
117-258, esp. p. 214-234.
II. R. H. CAMERON AND W. T. MARTIN, "An Expression for theSolution of a Class of Non-
Linear Integral Equations," to appear in American Journal of Mathematics.
III. PALEY AND WIENER, "Fourier Transformsin the Comples Domain," American Mathe-
matical Society Colloquium, Vol. XIX, pp. 151 and 156.

You might also like