Mathematics Department, Princeton University

Functions of Exponential Type

Author(s): E. M. Stein
Source: Annals of Mathematics, Vol. 65, No. 3 (May, 1957), pp. 582-592
Published by: Mathematics Department, Princeton University
Stable URL: [Link]
Accessed: 08-04-2025 23:43 UTC

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@[Link].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
[Link]

Mathematics Department, Princeton University is collaborating with JSTOR to digitize,

preserve and extend access to Annals of Mathematics

This content downloaded from [Link] on Tue, 08 Apr 2025 [Link] UTC
All use subject to [Link]
ANNALS OF MATHEMATICS
Vol. 65, No. 3, May, 1957
Printed in U.S.A.

FUNCTIONS OF EXPONENTIAL TYPE'

BY E. M. STEIN
(Received April 12, 1956)
(Revised October 8, 1956)

1. Introduction
This paper will be concerned with various uses of a device related to the opera-
tion of convolution.
In Part I we shall show that various classical inequalities for functions of
exponential type involving "sup" norm are a priori extensible to analogous L,
inequalities. The novelty of the result is that we may also dispose of the case
of equality for the L, norm, which case was unsettled even for Bernstein's in-
equality.
Part II treats Kolmogoroff's inequality which relates the bounds of a function
and its derivatives. This inequality is extended to L, , and an application is
given.
Part III contains an extension of a classical theorem of Paley and Wiener.
All functions whose Fourier transforms vanish outside a given compact sym-
metric and convex domain in n-space are characterized. Another characteriza-
tion, due to Plancherel and Polya [9], has previously been known, but the one
given in this paper is of a different nature.

PART I
2. Bernstein's inequality and its generalizations
We shall make use of the following notation: E, will denote the class of analytic
functions of exponential type ? a. B,, will denote the class of functions in E,
which are bounded on the real line. We shall denote by fl f 11 = (f1 I f(x) I' dx)1p,
if 1 ? p < oc, and ffll = essup-.O<x<o I f(x) ;Lp, 1 ? p ? o, the class of
measureable functions f, for which 11f IIp is finite. Finally C(- 0o, + 00) will
denote the Banach space of bounded continuous functions f, with norm 11 f 11.
We shall adopt the following convention: for an analytic function f(z), j(x) will
denote its restrictions to the real line.
The following result is now classical.
THEOREM OF BERNSTEIN. Suppose that f(z) E B. * Then || f' ?1. _ a || f 11
and, if for some xo, I f'(xo) I = a II f II l then f(z) = aetn + be'z.
We shall deal with linear operators T and S, each of which satisfies the follow-
ing conditions:
(1) T is defined on B0, for every a, and T commutes with translations.

1 The contents of this paper are taken from the author's Doctoral dissertation at the
University of Chicago. The author wishes to take this opportunity to express his apprecia-
tion to Professor A. Zygmund for the help and encouragement he received during the prepa-
ration of this dissertation.
582

This content downloaded from [Link] on Tue, 08 Apr 2025 [Link] UTC
All use subject to [Link]
 FUNCTIONS OF EXPONENTIAL TYPE 583
(2) There exists constants Aa , so that 1f T(f) II= ? Aa 11 f 1100, for f e B,.
The following is the main result of this part.
THEOREM 1. Let T and S be linear operators on B, , for every a, which satisfy
the above conditions. Suppose that for some constants bXa
(1) || T(f) I I $ a 11 S(f) Il ., if f EBe ; then 1 T(f) jI < XA fl S(f) LP, if
fEE0 n Lp, 1 < p < oo.
(2) Moreover, if I Tf(xo) I = X, 11 S(f) 11. , for some xo implies that f(z) =
aeiaz + be-iaz, then 11 T(f) IIp = X, 11 S(f) I Ip implies f = 0, if f EE0 n L,
1 ? p < 00.
REMARK 1. The theorem contains the statement that if f EE, n Lp , 1 < p < x,
then f e Ba. (This we prove in Lemma 3.)
REMARK 2. If we take for T = d/dx, and for S = I, (where I = identity
operator and X. = a) then we get an extension of Bernstein's inequality to Lp.
The statement of inequality is well-known, but the case of equality is new. For
T and S we may choose a variety of operators corresponding to classical in-
equalities involving sup norm. For references for these inequalities, for which
the above theorem is applicable, see [1] Chap. 4 and [2] Chap. 11.

3. Proof of Theorem 1
LEMMA 1. Let fn(z) e Ea , and I f (x) j < 1. Then there exists a function f(z) e Ea
and a subsequence {fin (x) I of {f,(x) I so that fi, (x) -* f(x) uniformly on every finite
interval.
PROOF. This lemma is known. In virtue of Theorem 6.2.4 of [2], the family
f (z) is a normal family. That f(z) E E. follows from the fact that | fn(z) I < ealzl.
LEMMA 2. If f(z) E Ba , or f(z) e Ea n Lp, 1 < p < oo, then f(x) = ft f(x - y)
/'a(y) dy, where 4,,(y) = 1/ir[(cos ay - cos 2ay)Aay2)].
PROOF. We make use of the fact (easily verified), that the Fourier Transform
of 4ti,(x) equals 1 on the interval [-a, +a].
If we assume f(x) eEa n Lp, for some p, 1 _ p < 2, then we may use a well-
known theorem of Paley and Wiener, that the Fourier Transform of f(x) vanishes
outside the interval [- a, + a]; therefore the lemma follows in this case.
Now assume f(x) E Ea n Lp, 2 _ p _ oo . Let fa(x) = f(x)[(sin 5x)/lx], 8 > 0.
Then clearly fa(x) E L2, and fa(x) EEa, E if a' > a + &. Moreover Ifa(x) I 5
f(x) I and fa(x) -* f(x) everywhere, as 8 -* 0.
By what has been previously said, we have:
r+0
fa(x) = ffa(x - Y)Oa, (y) dy. Now let 6 -* 0; we get:
00

r+00

f(x) = f(x - y)a' (y) dy, for a' > a. Hence, finally
00

+00

00

2 A convolution device device similar to the one used above was also used by Kober in [51.

This content downloaded from [Link] on Tue, 08 Apr 2025 [Link] UTC
All use subject to [Link]
584 E. M. STEIN
LEMMA 3. If f(x) E, n Lp , 1 _ p < oo, then f(x) EBa , and in particular
!If I l K-of'"p 1Vfip I
PROOF. BY Lemma 2, f(x) = Jfh" f(x - y)4a(y) dy. Using Holder's inequality,
ll f(X) fl00 ' 11 IPl, f P . However, 11 41, fl _, < supx I 4t' (x) Il/p p11 <a _ Ko l'P.
Therefore, II f 1100 _ K *"P 11 f lIP .3
LEMMA 4. 11 T(f) IIP _A A 11 f IIp, iff EEa n Lp, 1 < p < QK. (A,, is the bound
which holds in the case of the L., norm.)
PROOF. Iff E n Lp, then by Lemma 3, f E B., and T(f) is defined. Let g(x) E
Lp,i /p ' + '/p' = 1, and 11 g(x) lIP = 1. Let Fn(x) = f+Jf(x + y)g(y) dy.
Then clearly I Fn(x) I _ I f IfIP and Fn(x) E Ea . Moreover, (TFn)(X) = fn (Tf)
(x + y) g(y) dy, since T commutes with translations.
By the assumptions on T, (TFn)(x) 1100 < Aw 11 Fn(X) 110 < A, 1lf(x) IlI.
Hence, I TFn(O) ? A,, l f llP, which means that I Jn (Tf)(x)g(x) dx _
A, fl f IlIP. Letting n oo, we see that I f+t (Tf)(x)g(x) dx I _ A, 11 f IP, with
g(x) arbitrary except l I IIP = 1. Hence I Tf II _ A,, f II P .
LEMMA 5. | T(f) IIP _ X, || S(f) IIP, if f Ea n Lp, 1 < p < co.
PROOF. The proof of this lemma is exactly parallel with the proof of the pre-
vious lemma.
Again let Fn(X) = ftnf(x + y)g(y) dy where we assume only that fl 9(y) II, = 1.
Then by assumption on T and S, I TFn(O) I _ X0 11 S(Fn(x)) II.. This implies
that I f+n(Tf)(y)g(y) dy I _ X. SUPz f Jn (Sf)(x + y)g(y) dy I _X, 11 S(f) IIP,
by Holder's inequality. Therefore, I f+n (Tf)(y)g(y) dy I _ X, 11 S(f) Ip. Again
letting n -oo we get, I f+O (Tf)(y)g(y) dy I < X, I S(f) IIP, for all g for which
11 g lIP = 1. Therefore 11 T(f) Ip P X" 11 S(f) lIP
LEMMA 6. Let U be a bounded linear transformation on C- oo, + oo) which
commutes with translations. Then, Uff](x) = f+tf(x + y) dK(y) where K(y) is
of total bounded variation. (This fact is well-known [4].)
LEMMA 7. Let Gn(x) E B,, I Gn(x) I < 1, and Gn(x) -* 0 uniformly on every
finite interval. Then T(Gn)(x) -O 0 uniformly on every finite interval.
PROOF. Let U(f) = T(H,(f)), f E C(- oo, + oo) where H,(f)(x) = ft+f(x - y)
/4,,(y) dy. Then U(f) clearly satisfies the conditions of Lemma 6. Therefore
U(f)(x) = f+tf(x + y) dKa(y), where K,(y) is of total bounded variation. How-
ever, by Lemma 2, H,(f) = f, if f - B, . Therefore T(f) = f+f(x + y) dKa(y)
if f e B . Now I Gn(x) , 1Gn, (x) E Ba and Gn(x) -O 0 uniformly or every finite
interval; hence T(Gn) (x) = fJ Gn(x + y) dK,(y) must converge to zero uni-
formly on every finite interval, and the lemma is proved.
Since we have already proved Part (1) of the theorem we shall now assume that
fl T(f) P = Xe 11 S(f) l P, Wheref e E, n Lp for some p, 1 _ p < oo . Let us now
fix g(x) = [I T(f) IP-' sign (Tf)]/[l1 I T(f) IP 11p ] if 1 < p < 0o, and g(x) = sign
(Tf(x)) if p = 1.
We notice that l!g 9p = 1. Let us set Fn(x) = f+nf(X + y)g(y) dy and F(x) =
f+tf(x + y)g(y) dy. Then clearly Fn(x) 6 E, , I Fn (X) I 11 f IP, I F(X) I
II f IIP, and Fn(X) -* F(x) uniformly on every finite interval. Therefore by Lemma

3 The lemma is known; see [2], 6.7.1.

This content downloaded from [Link] on Tue, 08 Apr 2025 [Link] UTC
All use subject to [Link]
 FUNCTIONS OF EXPONENTIAL TYPE 585
1, F(x) E, . If we now let Gn(x) = Fn(x) - F(x), then Lemma 7 tells us that
T(Gn(x)) -* 0 uniformly on every finite set. In particular, (TGn))(0) -O 0.
However, T(FL(x)) = f2n (Tf)(x + y)g(y) dy; therefore TF(O) =
f+x (Tf)(y)g(y) dy = 11 T(f) 11 , by the construction of g. Therefore, TF(O) =
f 11 S(f) flu, by assumption.
Similarly (SF)(x) = f+OO S(f)(x + y)g(y) dy. Therefore, X, I SF(x)
X,, I S(f) plp = (TF)(O).
Owing to the fact that TF(O) I < ?X sup. I (SF) (x) , we obtain that
T(F)(O) = c supx I SF(x) . By the assumption of the theorem we must have
F(x) = ae"X + be-zx.
We have therefore established that:
I+00
(3.1) aet + be-t' - J f(x + y)g(y) dy. 00

Let us dispose first of the case where both a and b vanish. In this case, F(x) 0 ,.
and by the above, 1f T(f) 1II = 11 Sf IIP = 0. Therefore, trivially, (Tf)(O) =
X sup., p Sf(x) I and by the assumptions of the theorem f(x) = a'et'X + b'e-i.
This contradicts the fact that f e Lp, unless f-O.
Now we may assume that not both a and b in (3.1) vanish, and we shall show
that this leads to a contradiction. We break up consideration into two cases.
Case 1, 1 < p < oo. It is well known that the convolution of a function in Lp
with another in Lp, must vanish at infinity, if 1 < p < oo. This leads to a con-
tradiction in (3.1).
Case 2, p = 1. By what has been said above

T(aezo'x ? be-so) = L0(Tf) (x + y) g(y) dy.

+00

Therefore, since T commutes with translations,

r+0
a'ei ? b'e = L7- (Tf)(x + y)g(y) dy.
00

Since g(y) = sign (Tf(y)), therefore I a' + b' = jj Tf Ili . By periodicity, how-
everI f+ T(f)(y)g[y + (2ir/o)] dy = jj Tf Ii. Therefore since Ig(y) = 1, we
must have: g[y + (2rr/o)] = g(y), i.e. g is periodic with period 2i/o L Let g(y)
Z. ane^?1' be the Fourier expansion of g(y), and let o-(y) denote the (C, 1)
means of the expansion of g.
Then clearly,

limno f|+f(x + Y)cfn(y) dy = f+tOf(x + y)g(y) dy = aei?X + be-C~

However, since f c F.,, and f c L1 the theorem of Paley-Wiener implies that
f+0 e k?8f(y) dy = 0 if I k I > 1. We are therefore led to
r+0
ae io + be- "Y = ao f(x) dx.
00

This again is a contraction, and the proof of Theorem 1 is complete.

This content downloaded from [Link] on Tue, 08 Apr 2025 [Link] UTC
All use subject to [Link]
586 E. M. STEIN
PART II

4. Kolmogoroff's inequality
Kolmogoroff has given the following theorem:
THEOREM OF KOLMOGOROFF [6]. Let f(x), f'(x) ... f( )(x) be continuous and
bounded on the interval - o < x < + Xo . Then:

(4.1) || f(k)() | < Ck,n f(x) flfl fl f (x) Ik

where 0 < k < n, and Ck,n = (Kn.k)n/(Knf)n-k, and where

Ki = 4/rZ E O (_)k/ (2k + 1)i?+

for even i, while Ki = 4/7rZ=o 1/(2k + 1)t+? for odd i. Moreover the inequalities
are best possible, since there exist piecewise smooth functions for which equality is
attained.
Our aim is to prove the following theorem, which is the principal result of
this section.
THEOREM 2: Let f(x) , f1(x) ***f fl) (x) be continuous, assuming also that f (n-) (x)
is locally absolutely continuous, and assume further that f(x), f(l) (x) ... f(n) (x) all
belong to Lp (- o, + so) for some p, 1 _ p < oo. Then:
||Ok) (X) II n <1 n-k |f~) 11 f (n x) 11P,

where Ck ,, are the same as in Kolmogoroff's theorem.

PROOF OF THEOREM 2.
LEMMA 8. Let g(x) be locally absolutely continuous, and suppose that both g(x)
and g'(x) e Lp(-x, + ?o); 1 _ p < oo. Then [g(x + h) - g(x)]/h -g'(x) in
the Lp mean.
The lemma is well-known.

Now let h(x) = sign (f(k)(x)) f (k)(x) p-

if I < p < , h(x) = sign(f(k)(x)), if p 1. Then obviously fl h(x) I1p -1, and

j fk)(x)h(x) dx = f|k) H1).

Let

(4.2) F(x) = ff(x + y)h(y) dy.

Then repeated application of Lemma 8 to (4.2) above yields that F(r) (x)
exists, is bounded and continuous, and that:

(4.3) F(r)(x) = f f(r)(x + y)h(y) dy, where 0? r ? n.

00

Now F(k)(O) = I+of(k)(y)h(y) dy = I f k) II, But applying the theorem of Kol-

mogoroff to F(x) yields,

IF(k)(0) ? Cki n II F(x) IIk I|I F (n(x) 11k0C

This content downloaded from [Link] on Tue, 08 Apr 2025 [Link] UTC
All use subject to [Link]
 FUNCTIONS OF EXPONENTIAL TYPE 587
However,

F(x) 110 < 1f(x) jip , and 11 F"n)(x) 10 f_ Ifl)(X) flip

by applying Holder's inequality to (4.3) above. Therefore,
Fk) (0) I n Ci || f n-k. JJ f(n) kii

The last implies that:

11 f k)11 n - kn| |p-k 11 f(") fIlk

and the inequality of Theorem 2 is proved.

5. Application
In Theorem 2 we assumed not only that f(x) e Lp and f(n) (x) e Lp, but also
that all intermediate derivatives of f(x) belong to L,. We shall now prove a
theorem which complements Theorem 2, and which shows, in effect, that if
f(x) e Lp, and f n) (Xx)ce Lp, then all f(k)(X) (e Le, 0 < k < n. The idea of the
proof is to approximate f(x) by functions all of whose derivatives are in Lp .
Then we apply Theorem 2 to show that the approximation is uniform in the
intermediate derivatives.
THEOREM 3. Let f(x), f(')(x) ... f (n'(x) be continuous on (-00, + 00) and
f(n-1) (x) be locally absolutely continuous. If f(x) and f(n) (x) eLp, 1 ? p < cc,
then f(k)(x) ELp, 0 < k <n.
REMARK. The above theorem has been proved in the case p = 1 by Bochner
and Chandrashekaran, [3] p. 29, by the use of the Fourier transform. This theo-
rem is of the same kind as that introduced by Littlewood [8], which relates the
boundedness of a function and its derivatives at infinity. For related results,
see [7].
PROOF. Construct a function M(y) with the following properties:
(a) M(y) = 0, if I y l _ 1
(b) M(y) ? 0
(c) ft M(y)dy= 1
(d) M(y) is indefinitely differentiable.
Let Me(y) = 1/eM(y/E), E > 0.
LEMMA 9. If ME(Y) is as defined above, and g(y) is locally integrable, then
r+0
Me(y)g(x + y) dy -> g(x) a.e. as e 0.
co

This lemma is well-known.

Now with our given f(x) construct fe(x) as follows:
+00

(5.1) ft(x) = f Me(y)f(x + y) dy.

We note first that:
+00

(5.2) fer) (x) = J Me(y)f )(x + y) dy, 0 < r < n.

0O

This content downloaded from [Link] on Tue, 08 Apr 2025 [Link] UTC
All use subject to [Link]
588 E. M. STEIN
The above follows by differentiating (5.1) under the integral sign, which is per-
inissible since M,(y) has finite support. Integrating (5.2) by parts yields:

(f r) (X) ( = ( M 0J ME(y)f(X ? y) dy.

Next, we note that by (5.3) each f(r) (x) must be in L, . This follows from the
fact that f(x) is in Lp, and that M~ (x) is in LI. We may therefore apply The-
orem 2 to fe (x), resulting in:

(5.4) fk) f p 1 Ckn f| fl ! |n-k 11 gf(n) Ilkt

By (5.2) it follows that 11fn) I1P-I! flf) %1 and 11Kf IIp < 11 f 11, ,since 11 M 1 = 1.
Therefore (5.4) becomes:

11 k 1 - C 11 f I! n-k 11 f (n) Ilk

Filially, by Lemma 9, fk (X) f(k) (x) a.e.; therefore applying Fatou's lemnma
proves that f (k) (x) E Lp, concluding the proof of the theorem.

PART III

6. Generalization of a theorem of Paley and Wiener

Our aim will be to generalize the following well-known result of Paley and
Wiener:
If f(x) E L2(- co + co), then the necessary and sufficient condition that the
Fourier Transform of f(x) vanish outside the interval [-a, + a] is that f(x) be of
exponential type a.
We shall denote by En the euclidean n-space, and by x = (x1, x2, ... Xn) a
generic point in it. E' will denote another euclidean n-space, and the generic
point in it will be denoted by y = (y1, yn). Also define the inner product
X Y = X1Y1 + X2Y2 * + XnYn . For any function f(x) EL2(En), the Fourier
Transform of f(x), given by

1/(27r)nl2 eX f(x) dx,

En

will be abbreviated by F.T. (f).

Let Q be an arbitrary compact, convex and symmetric4 domain in E' . We shall
denote by Q* its conjugate domain in En; i.e. Q* = {x EEn I I xy I < 1 if y E Q}.
Then as is well-known, Q is the conjugate domain to Q*; i.e. to every y 4 Q, there
is an x E Q*, so that x-y > 1.
We shall now define means of f(x) taken with respect to the domain Q*. Let
A/(x) denote the characteristic function of Q*, and let AIt(x) = F-ip(x/t). Let
f(x) be any locally integrable function. Define

(6.1) Ut(f)(x) = A t(z)f(x + z) dz. pn

(6.1) defines the means of f(x) taken with respect to the domain Q*.

4 Symmetric means that if y e Q, then -y e Q.

This content downloaded from [Link] on Tue, 08 Apr 2025 [Link] UTC
All use subject to [Link]
 FUNCTIONS OF EXPONENTIAL TYPE 589

We are now in a position to state our theorem.

THEOREM 4. Let f(x) E L2(E.), f(x) bounded; let F(y) = F.T. (f). Then a neces-
sary and sufficient condition that F vanishes outside a compact, convex, and sym-
metric domain Q, is that Ut(f) (x) is for each fixed x an analytic function of expo-
nential type ? 1 in t. U,(f) is defined by (6.1) and represents the means of f(x)
taken with respect to the domain Q* conjugate to Q.
REMARK. It is not difficult to verify that the above theorem actually con-
tains the one dimensional form of the Paley-Wiener theorem quoted above. We
omit the proof of this fact.
When the convex domain Q is taken to be the sphere of radius a with center
origin, the theorem may evidently be put in the following form:
COROLLARY. Let f(x) E L2(En), and bounded. Then a necessary and sufficient con-
dition that the Fourier transform of (f)x vanish outside the sphere of radius a and
center origin, is that

Vt(f)(x) = nt | f(x + z) dz
be for each x, a function of exponential type < a in t. Here, wC denotes the volume of
the n-dimensional unit sphere.5

7. Proof of Theorem 4

Let us recall that #6(x) is the characteristic function of the domain Q*, and
that #t(x) = tUnJ(x/t).
LEMMA 10. Let 4(y, t) = F.T. (*t(x)). Then 4(y, t) = Jm=O [02.(y)e~j/[(2m)!].
Also | 02m(Y) I < K if y e Q, where K is independent of m and y. Moreover if y # Q.
then

limmo, |402m(Y) |rn > 6(y) > 1,

where 5(y) is uniformly greater than 1 with y uniformly away from U.

PROOF. '1(y, t) = 1/(27r) 12 e-z'yi't(x) dx = 1/(2Xr) '2fa* etx-" dx.
Hence, for each fixed y, since we integrate over a bounded domain in x, 4(y, t)
is analytic in t. Therefore,

4)(y, t) = .,o [c(y)tM"/m!.

Since the domain 9* is symmetric, it is evident that 4(y, t) is even. Therefore

(7.1) 4p(y) t) = Eao=o [52m,(Y)t m]/(2m)!

However, by (7.1)

+2, (Y)= {=a2m"4(y) t)I/atymn I = (-i)2,/(2 7r)"n2 (X.y)2' dx.

If y e Q1, then I x y | ? 1 for every x e Q*, therefore clearly 1402m(y) I _ K. Now

5 The factor con is introduced only for the sake of convenience.

This content downloaded from [Link] on Tue, 08 Apr 2025 [Link] UTC
All use subject to [Link]
590 E. M. STEIN
if y 4 Q, there exists an x EQ* so that I ye x > 1. By continuity, there exists a
neighborhood of x in W*, where J y x _ 6 (y) > 1. The fact that

qS2m(y) = (-i)2/(,X)n/2 L. (.y)2m dx,

then proves the lemma.

The necessity is now easy. For assume that F(y) = F.T. (f(x)) vanishes out-
side the domain Q2. Now

Ut(f)(x) = f(x)*/t(x) = ,t (z)f(x + z)dz =fb (y, t)F(y)e-i'x dy

=E, el /(2m)! [ 0f2ci(y)F(y)eGiYX dy] = Z [t2m/(2m)!] cm.

However, by Lemma 10, J 02m(Y) J K, if y e Q2, therefore

J l 2.(y)F(y)eivx dy < K', i.e. cm < K'.

Therefore the series Emjo [Cmt2m]/(2m) ! is of exponential type ? 1. The neces-

sity of the conditions of the theorem are therefore proved. As for sufficiency we
shall temporarily make the assumption that f(x) is indefinitely differentiable and
all its derivatives are bounded and in L2(En). We shall lift this restriction of
smoothness at the end of the proof.
LEMMA 11. Let f(x) satisfy the above smoothness condition, and suppose that
F(y) = F.T. (f). Then
(7.2) (27r) - F.T. { [( 2m Ut(f))/at2m] t=o} = k2m(y)F(Y).
PROOF.

(2 n) F.T. [Ut(f)] = (27r)'-F.T. (4*t'f) - F.T. (Vlt)'F.T. (f)

= (ZEan=o [4,2m(y)2n]/(2m) !) F(y).

Therefore,

(27r) . F.T. [Ut(f)] = (Zm=O [02m(y)t2m]/ (2m)!) F(y).

Differentiate 2m times with respect to t, and set t = 0 (the passage underneath

the F.T. is easily justified because of the smoothness assumed for f). The lemma
is therefore proved.
Now let g(x) be an arbitrary function in L2(En), subject only to the condition
that g(x) = 0 outside some bounded set, and 11 9(X) 112 = 1, and set:

(7.3) H(t) = L Ut(f)(x)g(x) dx. En

Then by the assumptions of the theorem, U,(f)(x) is bounded for each x and
real t, and is of exponential type < 1 in t, for each x. Using Lemma 1 (Section 3),

This content downloaded from [Link] on Tue, 08 Apr 2025 [Link] UTC
All use subject to [Link]
 FUNCTIONS OF EXPONENTIAL TYPE 591
it is an easy matter to verify that H(z) is analytic and of exponential type < 1.
Moreover, by Schwarz's inequality:

1 H(t) I < I"' f 112 for real t.

Here, V is the volume of the domain Q*.
Using Bernstein's inequality (theorem stated in Section 2) repeatedly, we get:
H(2m)(0) 2 < f 2
Hence by (7.3)

J [ (&a( ) Ut~f))/(8t )t=o-g(x) dx j < VI f 2

However, because of the arbitrariness of g(x), we must have:

fl[(&2m'U(f))/&t2m]t.0 2 2 if2
11 [( Ut())/8t ]t-0112 _< VI 11 f 112

Now Plancherel's theorem, and Lemma 11, implies

(7.4) 11 (P)2.(y)F(yJ) 112 -< (27r)'V- I! f ~12.

Now let mn -- oo in (7.4); and recall that if y o Q then 2m(Y) "m -_ 6(y) > I
(Lemma 10). Therefore F(y) = 0 a.e. if y 0 Q. The theorem is therefore proved
under the above restriction-that f(x) is indefinitely differentiable with all its
derivatives bounded and in L2(En).
To conclude the proof, let f(x) be bounded, and in L2(En) and assume Utf(x)
is for each x analytic and of exponential type < 1 in t. Let so11(x) be an "approxi-
mation to the identity" in the usual fashion. That is, each pnn(x) is indefinitely
differentiable with compact support, and f*, -o f in L2 norm, as n > oo. (f*pn de-
notes convolution of f with en). Then clearly each frl(x) = f(x)*(p,,(x) is indefi-
nitely differentiable and its derivatives are bounded and in L2(En). Moreover,

Ut(fn)(X) = Ut(f)(X)* on(X).

An easy application of Lemma 1 (together with the boundedness of f(x)),

insure that Ut(fn)(x) is for each x analytic and of exponential type <1 in t.
Hence by the case we have proved above, F.T. (f,,) = 0 if y o Q.
Since f, -> f in L2, then F.T. (f) = 0 if y o Q, and the proof of the theorem is
complete.

8. Concluding remarks
The requirement in Theorem 4 that f(x) is bounded is needed so that Ut(f)(x)
is uniformly of exponential type < 1 as a function in t. The theorem could be
evidently reformulated so as not to involve the boundedness of f(x) explicitly.
The device of convolution, which in the above takes the form of (7.3), could
be used in other connections in deriving certain n-dimensional theorems from
one dimensional ones. We hope to return to this elsewhere.

MASSACHUSETTS INSTITUTE OF TECHNOLOGY

This content downloaded from [Link] on Tue, 08 Apr 2025 [Link] UTC
All use subject to [Link]
592 E. M. STEIN
REFERENCES

1. AKHIEZER, 14., Vorlesungen uber approximationstheorie, Akademieverlag, Berlin, 1953.

2. BOAS, R. P., Entire functions, Academic Press, New York, 1954.
3. BOCHNER, S. and CHANDRASEKHARAN, K., Fourier transforms, Princeton University
Press, Princeton, 1949.
4. KITAGAWA, T., On the theory of linear translatable functional equations and Cauchy's
series, Jap. J. Math., vol. 13 (1937), pp. 233-332.
5. KOBER, H., On the approximation to integrable functions by integral functions, Trans.
Amer. Math. Soc., vol. 54 (1943), pp. 70-82; Approximation of continuous func-
tions by integral functions of finite order, ibid., vol. 61 (1947), pp. 293-306.
6. KOLMOGOROFF, A. N., On inequalities between upper bounds of the successive derivatives
of an arbitrary function on an infinite interval, Amer. Math. Soc. Translation 4,
1949.
7. HALPERIN, I. and PITT, H. R., Integral inequalities connected with differential operators,
Duke Math. J., vol. 4 (1938), pp. 613-625.
8. LITTLEWOOD, J. E., The converse of Abel's theorem on power series, Proc. London Math.
Soc., vol. 9 (1910), pp. 434- 448.
9. PLANCHEREL, M. and POLYA, G., Fonctions entieres et integrales de Fourier mutiples,
Comment. Math. Helv., vol. 9 (1937), pp. 224-248.

This content downloaded from [Link] on Tue, 08 Apr 2025 [Link] UTC
All use subject to [Link]