638 D. G.

DICKSON [August

Successive integrations by parts and application of the conditions

fm(l)=0 gives I(x) =f(x).
The second solution can be verified in a similar way.

References
1. R. G. Buschman, An inversion integral for a Legendre transformation, Amer.
Math. Monthly 69 (1962), 288-289.
2. -, An inversion integral, Proc. Amer. Math. Soc. 13 (1962), 675-677.
3. A. Erdélyi et al., Tables of integral transforms, Vol. 1, McGraw-Hill, New York,
1954.
4. --, Higher transcendental functions, Vol. 2, McGraw-Hill, New York, 1954.
5. Ta Li, A new class of integral transforms, Proc. Amer. Math. Soc. 11 (1960),
290-298.
6. E. T. Titchmarch, Introduction to the theory of Fourier integrals, 2nd ed.,
Oxford Univ. Press, Oxford, 1959.
M. A. College of Technology, Bhopal (M.P.) India

INFINITE ORDER DIFFERENTIAL EQUATIONS

D. G. DICKSON

1. Introduction. Let /(z) = ]C™-o AkZk converge for \z\ <R, where
0 <R ^ oo ; let E he the vector space of entire functions of exponential
type less than R; and let 2D= XX o AkDk, where D is the differential
operator. The purpose of this paper is to provide a brief derivation
of the results of Muggli [2, p. 154] regarding the general solutions in
E of the equations
(1) 2D<¿>
= 0, and
(2) Sid, = $.
It will be shown that 2Dis a surjective endomorphism of E, reduc-
ing the problem of solving (2) to that of solving (1). It is easy to
show that if f is a zero of / of order at least h+l and of modulus less
that R, then zheiz is a solution of (1). If B is the set of all such exponen-
tial monomials, then Muggli's result says that B is a basis for the
solutions of (1) and that each solution cb is representable as a sum of
exponential monomials with exponent coefficients in the conjugate
indicator diagram of <b. Each solution of (2) is then representable as
the sum of a contour integral and a solution of (1).
Results similar to these have been obtained by Sheffer [3, p. 255]
Received by the editors March 14, 1963.
i964] INFINITE ORDER DIFFERNTIALEQUATIONS 639

and Sikkema [4, p. 203]. In case/ itself is entire and of exponential

type, the solutions of (1) are entire mean periodic functions and have
been studied widely. The result concerning (1) may be viewed as a
generalization of the fact that any function of exponential type ff and
of period 27r can be expressed in the form Yc*einz< where the sum is
over integers » with | w| ^<r. The explicitness of the result given here
and its brief derivation are due to the use of a lemma which is per-
haps of interest in itself.
2. Preliminaries. Let C denote the disk |z| <R, and let P' denote
the complement in the plane of the set P. For <j>in E, P(<p) will de-
note the conjugate indicator diagram of <p,and L<p will denote the
Borel transform [l, p. 73] of <p.Then P(<p) EC, and L<p is analytic in
P'(<t>). Also iltEC, then £>2e*<=/(/)«*',where 3l>,= £t% Akdk/dzk.
The fact that 3Dmaps E into E with P(£xj>)EP(<t>) follows easily
from the Pólya representation of <p [l, p. 74]. For if y is a simple
closed (rectifiable and positively oriented) curve in C containing P(<p)
in its interior, then

£>0(z) = (2«)-1 I emf(w)L<p(w)dw,

Jy
the change in order of integration and summation being justified by
the uniform convergence of this integrand on y. £>cbis obviously en-
tire, and since y may be chosen to be arbitrarily close to the boundary
of P(<p), it easily follows that the indicator function of D<£is less than
or equal to that of <pand so P(£xb) EP(4>)-
33 also maps E onto E; in fact if ypEE, then there is a ^0 in E such
that SD'/'o= \A and P(fto) =P(^). For suppose that 7 is a simple closed
curve in C about Pty) such that there are no zeros of / on 7 or in the
region common to the interior of 7 and P'(?p). Let

1 r e™hp(w)
(3) ^o(z) = —- I — dw.
2m J y f(w)

Then \poEE and since \}/ois independent of the choice of 7 for 7 in

the zero free region mentioned, P(\p0)EP(4')- But easily 3ûi/'o=^',
and since P(f)=P(®to)EP(4>o), P(W=P(f).
3. Representation of solutions.
Definition. If <j>EEand tEC, then

T<t>(t)
= [a), f 4>(s)e^^ds\ .
L J 0 Jj-o
640 D. G. DICKSON [August

T<bis well defined; for if tEC, then eztEE and its convolution with
<phas as a type the maximum of | /| and the type of <6.Hence the con-
volution and its image under 2DZare in E. Further, T<p is analytic in
C. For suppose A is a closed circular neighborhood in C and tEN.
Then the convolution of <b and ezt is of a type <r<R, and its Borel
transform is (z —t)~1L(b(z). By choosing a circle ß about the origin
with radius between <rand R and which contains N in its interior, we
may write the convolution as a Pólya integral over ß and obtain

/f(w)L<b(w)
H(w, t)dw, where H(w, t) =-— •
ß 2iri(w — t)
Application of Morera's theorem yields the analyticity of T<p in the
interior of N.
Lemma. If <pEE and tECnP'(<p), then T<p(t)=f(t)L<b(t)-L£Hp(t).
Proof. Let <j>he of type a<R and e= (R —a)/3. Let a, ß, and y
he the circles \z\ =cr + e, \z\ =a + 2e, and \z—(a+R)/2\ =e/3, re-
spectively. The left member of the identity to be established is
analytic in C, while the right member is analytic in C(~\P'(<p) since
P(2Dci>)EP(4>)- Since both are uniform, it suffices to establish the
identity for / inside y. Considering only such /, write <p(t) as a Pólya
integral over a, and then operate on this with 2Dand then with L,
writing the latter as a Laplace transform from 0 to ». Then with
H(w, t) as in (4), LSxb(t)=—faH(w, t)dw, the change in order of
integration being justified by the observation that /0™exp[(w —t)s]ds
converges absolutely since (R(Z—w) >0 for w on a. Write faH(w, t)dw
as the difference of integrals over ß and y. As in (4), the integral
over ß is T<b(t) while the integral over y is f(t)Lcb(t).
Definition. Let {^jitex be the zeros of / in C with mk+i the order
of f*. Let ck he a circle in C about f* containing no other zeros in
or on itself. For each kEK and natural h, define the linear functional
£)*»on E by
(t - h)hT<p(t) J
h<t>
=—
£>kh<t>
= — II --.-dt.
2iriJCk /(/)
Using the lemma, it is easy to show that ©jfc>,(zpexp f3z) = h\5phStk
when f„ is a zero of/of order at most p + l and hence the elements in
B are linearly independent.
Theorem. Let <bEE; then £xb= 0 if and only if

(5) 4>(z)= E^E -T— zA>

k h=o h\
i964] INFINITE ORDER DIFFERENTIAL EQUATIONS 641

where k ranges over those k for which ¿j¿£P(</>).

Proof. The fact that such a sum satisfies the equation is obvious.
Suppose that 33$ = 0. Upon writing the sum in (5) with the coeffi-
cients in integral form, the mk may be replaced by infinity since the
terms so introduced are zero by Cauchy's theorem, and then the
sum over h may be replaced by an exponential function. The sum of
the integrals over the ck may be replaced by one integral over a closed
path 7 in C about P(<p) whose interior contains no zeros of / other
than those in P(4>). Using the lemma and the fact that 33$= 0, the
resulting integral is the Pólya representation of <j>.

Corollary. Let <pand 4>EE; then <£>4>=4>

if and only if

<Kz)= ^o(z) + Y ¿kZY-**,

k a=o A!

where k ranges over those kfor which ^kEP(4>~4'o) andipo is given by (3).
Other interesting results follow from the theorem and its corollary
by observing their implications when P(<b) or P(<p—\j/o) contains one
or no zeros of/, as is the case when one of the diagrams is a point. It
follows from the theorem that any solution of (1) in E is actually the
solution of a similar equation of finite order having a characteristic
function dividing / in the ring of functions analytic in C.

References
1. R. P. Boas, Entire functions, Academic Press, New York, 1954.
2. H. Muggli, Differentialgleichungen unendlich hoher Ordnung mit konstanten
Koeffizienten,Comment. Math. Helv. 11 (1938), 151-179.
3. I. M. Sheffer, Systems of linear differential equations of infinite order, with
constant coefficients,Ann. of Math. (2) 30 (1928), 250-264.
4. P. C. Sikkema, Differential operators and differential equations of infinite order
with constant coefficients, Noordhoff, Groningen, 1953.

University of Michigan