1995 Book ComplexAnalysisAndSpecialTopic PDF
1995 Book ComplexAnalysisAndSpecialTopic PDF
1995 Book ComplexAnalysisAndSpecialTopic PDF
With 29 Illustrations
, Springer
Carlos A. Berenstein Roger Gay
Mathematics Department and Centre de Recherche en Mathematiques
Institute for Systems Research Universite de Bordeaux I
University of Maryland 351, Cours de la Liberation
College Park, MD 20742 33504 Talence (cedex)
USA France
All rights reserved. This work may not be translated or copied in whole or in part without the
written permission of the publisher (Springer-Verlag New York, Inc., 175 Fifth Avenue, New
York, NY 10010, USA), except for brief excerpts in connection with reviews or scholarly analy-
sis. Use in connection with any form of information storage and retrieval, electronic adaptation,
computer software, or by similar or dissimilar methodology now known or hereafter developed
is forbidden.
The use of general descriptive names, trade names, trademarks, etc., in this publication, even
if the former are not especially identified, is not to be taken as a sign that such names, as
understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely
by anyone.
987654321
ISBN-13:978-1-4613-8447-2
To our dear friends,
Mischa Cotlar and Jean-Henri Davancens
Yaqui me despido yo
Que he relatao a mi modo,
Males que conocen todos
Pero que naides conto.
Martin Fierro
Jose Hernandez
Preface vii
CHAPTER 1
Boundary Values of Holomorphic Functions and
Analytic Functionals 1
1.1. The Hardy Spaces in the Disk 2
1.2. Hyperfunctions 35
1.3. Analytic Functionals and Entire Functions of Exponential Type 51
1.4. Vade Mecum of Functional Analysis 77
1.5. Convolution of Analytic Functionals 85
1.6. Analytic Functionals on the Unit Circle 94
CHAPTER 2
Interpolation and the Algebras Ap 109
2.1. The Algebras Ap 109
2.2. Interpolation with Growth Conditions 118
2.3. Ideal Theory in Ap 136
2.4. Dense Ideals in AP(O) 160
2.5. Local Ideals and Conductor Ideals in Ap 166
2.6. The Algebra Ap of Entire Functions of Order at Most p 170
CHAPTER 3
Exponential Polynomials 198
3.1. The Ring of Exponential Polynomials 198
3.2. Distributions of Zeros of an Exponential Polynomial 217
CHAPTER 4
Integral Valued Entire Functions 260
4.1. The G-Transform 260
4.2. Integral Valued Entire Functions 278
ix
x Contents
CHAPTER 5
Summation Methods 299
5.1. Borel and Mittag-Leffler Summation Methods 299
5.2. The Lindelof Indicator Function 316
5.3. The Fourier-Borel Transform of Order p of Analytic Functionals 326
5.4. Analytic Functionals with Noncompact Carrier 333
CHAPTER 6
Harmonic Analysis 353
6.1. Convolution Equations in IR 354
6.2. Convolution Equations in C 385
6.3. The Equation J(z + 1) - J(z) = g(z) 405
6.4. Differential Operators of Infinite Order 419
6.5. Deconvolution 458
References 471
Notation 479
Index 481
CHAPTER 1
J
Cgo(]a, b[) the limit
lim
8--*0+
I(x + ie)cp(x) dx
exists. This limit is denoted (f(x + iO), cp(x)}, and f(x + iO) E 1)'(]a, b[), the
Schwartz space of distributions in the interval la, b[. From [BG, Proposition
3.6.12] we have that f has a boundary value in the sense of distributions if
and only if f is of slow growth. Moreover, f = 0 if and only if its boundary
value f(x + iO) = 0 as a distribution. Furthermore, we have also shown in [BG,
Theorem 3.6.19] that every distribution T E 1)'(]a, b[) can be obtained as
T = f(x + iO) - f(x - iO),
where I is a function holomorphic in the domain Q = la, b[ + ilR* and of slow
growth in Q. (The meaning of I (x - iO) and of slow growth in the lower half-
plane being entirely similar to the previously defined for the upper half-plane).
If f has a continuous extension F to the semiclosed rectangle Q', Q' =
{z E C: a < Rez < b, 0 ~ Imz < c}, then the function F(x) coincides with
the distribution f(x + iO) on la, b[. In fact, since F is continuous in Q' and cp
is compactly supported in ]a, b[, it is clear that
lim
8--*0+
J
f(x + ie)cp(x) dx = J F(x)cp(x) dx.
2 1. Boundary Values of Holomorphic Functions and Analytic Functionals
lim I(x
8--+0+
+ is) = I(x + iO)
exists not only a.e. on ]a, b[, but with some kind of uniformity then I(x + iO)
is still the boundary value of I in the sense of distributions. One instance of
this statement occurs when the functions x t-+ I(x + is) and I(x + iO) belong
to Ll([a + 8, b - 8], dx) and
b-~
This is the spirit that animates the first section of this chapter on H P spaces.
The functions in these spaces are holomorphic in the upper half-plane and
I(x + is) E U(JR, dx) for every s > O. It turns out to be easier to consider the
case of holomorphic functions in the unit disk, which is what we do in Section 1.
In the second section of this chapter we take an altogether different approach
to the problem of existence of boundary values of holomorphic functions.
Namely, we introduce a concept so general that every function has a limit!
These limits are the hyperfunctions. That such a general concept would lead to
interesting mathematics is really amazing, but in fact the school of Mikio Sato,
Masaki Kashiwara, and Takahiro Kawai have made it one of the major tools of
contemporary analysis.
The following sections of the chapter will introduce analytic functionals, i.e.,
the linear continuous functionals on the space Jf'(Q). We shall see that there are
close relations between analytic functionals, hyperfunctions, and the H P spaces
considered in § 1.1.
This section is a short introduction to the Hardy spaces ~P(D), HP(D) in the
unit disk D = B(O, 1). We refer to the excellent monographs [Du], [Ga], [Hof],
[Kol] for a thorough treatment of this classical subject (see also [Rud2]). We
refer also to [Rud2] for the background material in measure theory, and to [BG,
Chapter 4] for the properties of harmonic and subharmonic functions we use.
We include a proof of the celebrated Corona Theorem of Carleson, which
allows us to show the importance of solving the Cauchy-Riemann equation
with growth conditions.
It is convenient throughout this section to denote by T the unit circle and
da = de 12K, the normalized measure on T. Given a function I defined in D
and a number r, 0 ::: r < 1, we introduce a function Ir defined on T by
Ir(e i () := I(re i ().
1.1. The Hardy Spaces in the Disk 3
If I is a harmonic function and 1 ::; p < 00, the function I/IP is subharmonic
and therefore the function
P(/L)(z) := 2~ 1 P(zj9)d/L(0).
1.1.1. Proposition. Let 1 ::; P < 00 and I E LP(1'), then lim,-> 1- Pr(f) =I in
LP(1').
Proof. Let 8 > O. There is g E C(1') such that III - gllp < 8. (It is here that we
use p < 00.) Hence
IIP,(f) - Ilip ::; IIP,(f) - Pr(g)lI p + IIPr(g) - gllp + III - gllp·
Since Pr(f) - P,(g) = P,(qJ') * (f - g)(ei'P') and IIPrll l = 1, we have
IIPr(f) - Pr(g)lI p ::; IIPrl1t III - gllp ::; 8,
whence
4 I. Boundary Values of Holomorphic Functions and Analytic Functionals
On the other hand, P (g) is the solution of the Dirichlet problem in D with
boundary value gin 11', hence limr-+I- IIPr (g) - glloo = O. It follows that
o
1.1.2. Proposition. Let 1 < p :::: 00 and I E ~P(D). There is a unique f* E
LP(lI') such that I = P(f*).
Proof. If I E ~P(D) then the family (/r)O<r<1 is a weakly bounded, hence,
weakly compact set in LP(lI'), where this space is considered as the dual of the
separable Banach space U (11'), 1/ p + 1/q = 1. Therefore, there is a sequence
rj ~ 1 and f* E LP (11') such that Irj ~ f* weakly. For every j, z ~ I (rj z)
is a harmonic function in Izi < l/rj, hence for Izl < 1, we have
I(rjz) = _1
JrT P(z, e
i8 )/(rjei8 )d().
2rr
Since () ~ P(z, ei8 ) E U(lI'), we can take the limit and obtain
1.1.3. Remarks. (1) This proof cannot be adapted to the case p = 1 since
L 1(11') is not the dual of any separable Banach space. Furthermore, there is
a good reason for this. If 1= P(f*), f* E L 1 (1I'), then Ir ~ f* in L 1 (1I'),
hence pointwise a.e., but the Poisson kernel P(re i8 ) = Pr «() is in ~I(D) and
Pr ~ 0 a.e. In fact, the Dirac mass 8 at () = 0 is the only measure /J- such
that P = P(/J-). (From the reader's knowledge up to this moment one can only
say that Irn ~ 1* for a sequence rn ~ 1-, but the rest of the reasoning holds
anyway.)
(2) As we have pointed out during the proof, if 1 < P < 00 and I E ~P(D),
then Ir = Pr(/*) ~ f* in LP(lI'). In the case p = 00, we proved that Ir ~
f* a.e. and in LP (11') for any p < 00.
For ei8 E 11' and 0 < 8 :::: 2rr, let us denote by 18(ei8 ) := {e it : It - ()I < 8} the
open arc centered at ei8 andof length 28.
1.1.4. Lemma. Let (lj) l~h:.M be a finite lamily olopen arcs in 11'. One can find
a sublamily (ljk)l~k~N such that: .
(i) the Ijk are pairwise disjoint;
(ii) if Ij = 18j (~j), then
UI j £ U h8jk (~ik);
l~j~M l~k~N
1.1. The Hardy Spaces in the Disk 5
and
Proof. We can assume that M 2: 2 and the indices are such that OJ 2: OJ+I. Let
= =
h 1. If Ijl intersects every I j we let N 1. If not, let h be the smallest
among the j such that Ih n Ij =
13. Let us assume we have chosen h, ... , jk
such that for 1 :::::: I :::::: k - 1 the arc Ijl does not intersect Ij,+I' but it intersects
I j for any j[ < j < j[+I. If Ijk intersects every Ij with j > k we set N = k. If
not, we let jk+1 be the smallest index j > A such that Ijk n I j = 13. Clearly this
process stops for some index N :::::: M.
If j < jN there is a k such that jk :::::: j < jk+1 and, since OJ :::::: Ojk and I j n
Ijk =j:. 13, we have Ij ~ h8h (l;jk). If j 2: jN, then OJ :::::: OjN and Ij ~ I 38jN (l;jN) since
I j n IjN =j:. 13. This proves (i) and (ii).
Finally,
=3 L a (hik (l;jk») = 3a ( U IA ) . 0
l:5.k:5.N l:5.k:5.N
1.1.5. Definition. The maximal function M(f.L) of a Radon measure J.L on 'll' is
the function M(J.L): '][' -+ [0, +00] given by
i(J
M(J.L )(ei(J) -_ sup
1J.L1(/8(e ))
.(J'
Od:5.lT a(h(e ' ))
where IJ.LI is the positive measure of total variation of J.L.
1.1.7. Proposition. For any Radon measure J.L on 'll' we have that for any t > 0
Finally, taking the supremum over all the possible compact subsets K, we obtain
a ·0
({ e': M(tt)(e'·O})
)> t ~ - -II.
3 IItt o
t
1.1.8. Proposition. Let tt be a Radon measure on '][', lettts be its singular part
with respect to a, and let E:= {ei(} E '][': M(tt)(e iO ) < Do}. Then Ittsl(E) = o.
lim -
8
8-->0
I/,/,(e;8)
.
If(e"") - feel·0 )1 dfP = 0, aa.e.
1.1. The Hardy Spaces in the Disk 7
In particular,
I(e''II ) = lim -I
8->020
1 '
la(eiB)
I(e"") dcp, a a.e.
1.1.12. Proposition. Let I ELI (1[') and let eill be a Lebesgue point 01 I, then
I/(eill)1 S M(f)(e ill ).
Proof. We have I(e ill ) = limo-.o (1 la (/8» I/a l(ei'P)da(cp). Moreover, for
8 > 0, there is Os > 0 such that
1
_
1_
a(/8)
r l(ei'P)dCPI S M(f)(e iO ) +8
J1a
for all 0 < 0 < Os. The conclusion is therefore clear. o
8 1. Boundary Values of Holomorphic Functions and Analytic Functionals
1.1.13. Proposition. Let J-Ls be a Radon measure on 1r. singular with respect to
u. Then. the limit
Proof. We can assume J-Ls ~ O. Fix t > 0 and 8 > O. There is a compact set K
and a decomposition J-Ls = J-Ll + J-L2. where J-Ll = J-LsIK. u(K) = O. and
1IJ-L211 ~ 8. Let
1J (ifJ)
J-Ls e =
r J-Ls (18 (e
~~~p u(h(eifJ »
ifJ
».
Outside K we have 1JJ-Ll = 0 and 1JJ-Ls = 1JJ-L2. with the obvious meaning for
the notation. Therefore
K U {e ifJ : 1JJ-Ls(eifJ ) > t} = K U {e ifJ : 1JJ-L2(eifJ ) > t}
~ K U {e ifJ : M(J-L2)(e ifJ ) > t}.
Since u(K U {e ifJ : M(J-L2)(e ifJ ) > t}) ~ 38/t. and 8> 0 is arbitrary. it follows
that for every t > 0 the set {e ifJ : 1JJ-Ls(e ifJ ) > t} is contained in a Borel set of zero
u-measure. Therefore. the same holds for {e ifJ : 1JJ-Ls(eifJ ) > O} and we obtain
that 1JJ-Ls = O. u a.e. This implies that the proposition is true. D
1.1.14. Corollary. Let J-L be a Radon measure on 1r and let J-L f du + J-Ls be =
its Lebesgue decomposition in absolutely continuous and singular parts. Then its
derivative
1.1.15. Definition. For eifJ E 1r and a > 1 a Stolz region of opening a and
vertex eifJ is the region
ra(e ifJ ) := {z E B: Iz - eifJl < a(1-lzi}}.
1.1.16. Proposition. For any a > 1. any Radon measure J-L. and any eifJ E 1r one
has
Proof. Since M(J-L) = M(IJ-Li} and IP(J-L)I ~ P(IJ-Li}. it is enough to prove the
inequality when J-L ~ O. We can assume eifJ = 1 and M(J-L)(1) = 1. since. if
M(J-L) = 00. the inequality holds and if M(J-L)(l) = 0 then J-L = O.
1.1. The Hardy Spaces in the Disk 9
Let ra(1) := ra, Z Era, and r = Izi. Consider the value t = a(1- r) and
Vo := {{ E T: 11 - ~I < t}.
For k E N*, let
Vk := RET: 2k - 1t :::: 11 - ~I < 2k t}.
These sets are only interesting when 1 :::: k :::: ko, where 2ko - 1t :::: 2 < 2ko t.
Since M(JL)(1) =
1, we have
8
JL(I&(1» :::: 0"(18(1» = -,
n
hence
Iv P(z,eqJ)dJL:::: ---::::
and
-1 t 1
i
-a <a.
2n Vo nl-r n
For k ~ 1, eiqJ E Vb Z Era, we have
11 - eiqJl :::: 2ale iqJ - zi.
In fact,
Hence
As a consequence
+ L 2k = 17a:::: 20a.
100
P(JL)(z):::: a o
k~l
Proof. Let 8 > 0 and 80 > 0, such that for 0 < 8 < 80 ,
Let f..Lo = f..LII~o(ei9) and let f..Ll = f..L - f..Lo. P(f..Ll) is hannonic in C\ SUPPf..Ll
and f..Ll is zero on I~(ei9), hence
a - lim P(f..Ll)(Z) = O.
z~ei8
one has
a -lim sup P(f..L)(Z) = a -lim sup P(f..Lo)(Z) ::: 20a8.
z~ei9 z-+e i6
Given that 8 > 0 was arbitrary we obtain that the limit of the statement of the
proposition vanishes. 0
By the previous proposition, the limit of P(f..L)(z) along the Stolz region ra(e i9 )
is zero. Hence,
o
Proof. We write f..L = I du + f..Lso I ELI (1I'), and f..Ls singular with respect to
u. Hence If..Lsl is also singular with respect to u and 'D1f..Lsl =
0, u a.e. On the
other hand, the fact that IP (f..Ls) I ::: P (I f..Ls I) together with Proposition 1.1.17
imply that
a - lim P(f..Ls)(z) = 0, ua.e. o
z-+e i8
{ f dJL = lim_
IF r-> I IF
{(-2 io(21f
1
n
Pr(cp - e)f(ei'P)dCP) dJL(e iO )
1.1.22. Corollary. For every h E ~ I (D) there is a unique Radon measure /L such
that h = P (/L) and there is h * ELI elI') such that
a - lim h(z)
z-+e i8
= h*(ei8 ), (1 a.e.
Proof. It is enough to write /L = h* d(1 + /Ls, /Ls singular with respect to d(1
and apply Corollary 1.1.19. D
1.1.23. Definitions. For 0 < p ::: 00 we define the Hardy spaces HP(D) by
It is clear that the sup can be replaced by lim in the definitions. The spaces
HP(D), 1 ::: p ::: 00, are also Banach spaces (see Proposition 1.1.24 below).
For I ELI (,ll'), real valued, one can prove the following inequality that gener-
alizes the inequality that relates the geometric and arithmetic means. Namely,
1.1.24. Proposition. For 1 ::: p ::: 00, the spaces ~P(D) and HP(D) are Banach
spaces with the norms Ilfllp = limr..... l- Ilfrllp.
Proof. For 1 < p ::: 00, we have already shown that P: LP('Jl') -+ ~P(D) is an
isomorphism of normed spaces and the same is true for p = 1 with P: C' ('Jl') -+
~l(D). This shows the spaces ~P(D) are Banach spaces when 1 ::: p ::: 00.
Finally, the space HP(D) is closed in ~P(D) since the convergence in ~P(D)
implies the uniform convergence on the compact subsets of D by the Harnack
inequalities (see [BG, Proposition 4.3.9]). D
(d) The functions f{J E LP('Jl'), which are of the form f* for some f E HP(D),
1 < P ::: 00, are characterized by the conditions
1 o
27r .
f*(e,o)e- . dO
mo - = lim
27r r ..... l- 0
27r . . dO
f(re,o)e- Ino -
27r
= lim rn. r f(t;) dt;
r..... l- 2m JI~I=r t;n+l
=0
when n < O. This shows that (a) holds for f*. On the other hand, for r fixed,
P(f*)(z) =L
"7
(_1 r
27r Jo
27r
f*(eit)e-int dt) rlnleinO
ne",
= _1 r
27r Jo
27r
f*(e it ) dt.
1 - r'tz
=~ r f*(t;) dt;.
2m JI~I=l ~ - Z
14 1. Boundary Values of Ho1omorphic Functions and Analytic Functionals
In the second line we have used that we already know that (d) holds for f*.
Since P(f*)(z) = f(z), this proves (c). Moreover, this argument shows that if
rp E U(1l') has cn(rp) = 0 for all n < 0, then
P(rp)(z) = -1-.1
2m
rp({) d{,
1~1=1 { - z
i.e., P(rp) E JIf(D) n ~P(D). D
log If(O)1 ::: _1 f2:rr log If(re i8 )1 dO ::: _1 f2:rr log 1f*(ei8 )1 dO.
2rr lo 2rr lo
Moreover, if f ;;fo 0, then f* =f. 0, a a.e.
Proof. We can assume that IIflloo::: 1. Otherwise dividing f by IIflloo
will not change the above inequalities. We know that the function r 1-+
(1/2rr) J;:rr log If(re i8 )1 dO is increasing and that limr--+l-1og If(re i8 )1 =
log 1f*(ei8 )1, a a.e. Moreover, log If(e i8 )1 ::: O. Hence we can apply Fatou's
lemma
f2:rr liminf f2:rr log If(rei8)1 dO ::: _1 f2:rr log 1f*(ei8 )1 dO.
lo r->1- lo 2rr lo
This implies the required inequalities.
The only difficulty in proving that a({e i8 : f*(e i8 ) = O}) = 0 occurs when
f(O) = O. Let m be the order of z = 0 as zero of f. The function g(z) =
f(z)/zm also belongs to Hoo(D), Ig*1 = 1f*1, everywhere and log Ig(O)1 > -00.
Therefore
_1 f2:rr log 1f*(ei8 )1 dO > -00,
2rr lo
which implies f*(e i8 ) =f. 0, a a.e. This concludes the proof. D
The main building blocks of functions in HP(D) are the Blaschke products.
defines a function in Hoo(D) whose zeros are exactly the GYn and has a zero at
z = 0 of multiplicity k.
Proof. The general term of the series
1.1. The Hardy Spaces in the Disk 15
satisfies
an - z ) 1= (1-lanl) I(an +_Ianlz) I::: 1 + Izl (1 -Ianl).
11 - (~
an 1 - anz (1 - anz)an 1 - Izl
Hence, the infinite product converges uniformly in every compact subset of D
to a function B E .Tt'(D). Moreover, IB(z)1 ::: 1 since every factor has absolute
value 1 on 1I'. Therefore II B II 00 ::: 1. 0
This result shows that the condition En> I (1 - Ian I) < 00, often called
Blaschke condition, is sufficient for the existence of a function I E Hoo(D)
vanishing exactly at the an. In fact, we shall show below that this condition is
necessary even for functions in N(D) to vanish on the sequence (an)n~1 without
vanishing identically.
1.1.28. Proposition. Let IE N(D), I :f. 0, and let (an)n~1 be the sequence 01
zeros 01 I in D\{O}, counted with multiplicity. Then the sequence (an)n~1 satisfies
the Blaschke condition
~)1 - Ian I) < 00.
n~1
Proof. We can assume I has infinitely many zeros and also that 1(0) :f. o.
Otherwise, the function I (z) / zm is still in the Nevanlinna class and satisfies
this additional restriction.
Let v(r) be the counting function of the zeros of I in B(O, r). Fix k E N* and
choose 0 < r < 1 such that k < v(r). By Jensen's formula (see [BG, 4.4.29])
we have
1/(0)1 II _r ::: 1/(0)1 II r
I~n~k Ian I I~n~v(r) Ian I
Hence,
1.1.29. Lemma. Let 0::: Un < 1lor every n, then TIn>1 (1 - un) > 0 if and only
if En~1 Un < 00. -
Proof. It is well known that En>1 Un < 00 implies that the infinite product is
positive. To prove the converse, iet Pk = (1 - UI)··· (1 - ud, then PI ~ P2 ~
16 1. Boundary Values of Holomorphic Functions and Analytic Functionals
... ::: p = TIn> 1 (1 - un) > O. As it is very easy to see (see [BG, Proposition
4.6.12]), we have p ~ Pk ~ exp( -u 1 - U2 - ... - ud. Therefore Ln::o: 1 Un < 00.
D
lim -
1 12Jr log IB(relll)1
. dO = O.
r~l- 21l' 0
Proof. Let
and
lim - 1
r~l- 21l'
12Jr log IB(re,II)1
0
. dO = 0
and
-1
21l'
12Jr 10g(lB*(e ill ) IdO = O.
0
1.1.31. Proposition. Let lEN (D), I ;;jE 0, and let B be the Blaschke product
corresponding to the zeros 01 I. Let g = 1/ B. Then g has no zeros in D and,
besides, it satisfies:
(i) g E N(D) and IIgllo 11/110; and=
(ii) if IE HP(D) lor some 0 < p :::: 00, then g E HP(D) and Ilglip = II/lIp.
Proof. It is clear that III :::: Igl, hence 11/110:::: IIgllo. Moreover,
1
log+ Igl :::: log+ III + log+ WI = log+ III -log+ IBI·
Hence
1
log IIgllo:::: lim -2
121f log+ I/(re,9. )1 d()
r-+I- rr 0
- lim -
1 121f log IB(re ,9. )1 d() = log 11/110.
r-+I- 2rr 0
and define gN = I/BN. Since IBNI = 1 on '][' we have IIgNIl p = II/lIp. When
N -+ 00 then IgNI)" Igl. Therefore,
Ilgrllp = N-+oo
lim IIgN,rllp:::: II/lIp.
If I E H 2 (D), then
lim II/rll~
r-+I-
= r-+I-
lim "la
~
n I2r2n = II/II~ < 00.
n~O
Hence
L la l n 2 = II/II~.
n~O
18 1. Boundary Values of Holomorphic Functions and Analytic Functionals
o
1.1.33. Proposition. Let I E HI(D), then
a - lim I(z) = l*(e iO )
z~e,8
This implies that Ir ~ 1* in L I('R') and II f* III = II I III. Si~ce L I ('R') can be
considered as a closed subspace of C' ('R') and Ir ~ f.,/, in C' ('R'), it follows that
f.,/,s = O. Hence 1= P(f*) and, also, the Fourier coefficients cn(f*) = 0 for
n < O. As was shown in Proposition 1.1.25 this implies that the Cauchy integral
representation of I in terms of 1* is valid. 0
A Radon measure f.,/, on 'R' is called analytic if cn(f.,/,) = I.e- int df.,/, = 0 for
all n < O.
S(z) = exp (_
Jo
r 21f
e~1 + z df-Ls(t))
e'l - z
is an inner function. It is referred to as a singular inner function.
Proof. We have
log lSI = 2;r P(-f-Ls).
Since f-Ls :::: 0, it follows that lSI.:::: 1. Moreover, f-Ls singular with respect to a
implies
log IS*I = (log ISI)* = 0, aa.e.
Therefore S is an inner function. o
The simplest example of a singular inner function is
S(z) = exp (_ 1 + z)
l-z
corresponding to f-Ls = 8 = Dirac mass at () = O.
1.1.37. Proposition. The family of inner functions coincides with the products of
the form cBS, where c E C, lei = 1, B is a Blaschke product, and S is a singular
inner function.
Proof. It is clear that the product of two inner functions is another inner function,
hence all products cBS are inner functions. Conversely, given an inner function
f, let B be the Blaschke product of its zeros and g fiB. We have that g =
has no zeros, Ilglloo = IIflloo = 1, and Ig*1 = 1, aa.e. Therefore, -loglgl is a
positive harmonic function and has a representation
-log Igl(z) = Re 1o
21f eil + z
-,-
e'l - z
df-L(t)
20 1. Boundary Values of Holomorphic Functions and Analytic Functionals
for some measure IL :::: O. Hence IL = (1JIL) d(1 + ILs, and 1JIL = -(log Igl)* =
-log Ig*1 = O. It follows that IL is a positive singular measure. Let
log Ig(z)1 = Reh(z)
for some he JIf(D). Then
211' eil +z
h(z) = - 1o -'-,e'-z
- djJ,s(t),
eh is a singular inner function and g = ce h for some c e C, lei = 1. D
_1 r 211' I/r(e i6 W dB
27'( Jo
::::: _1 r21l' (_1 r 211' Pr(B _ t)q;P(e il ) dt) dB
27'( Jo 27'( Jo
Jor
21f
log 1/(0)1 ::: _1 log 1/*(ei9 )1 de
21r
becomes an equality if and only if I is a constant.
(v) Thelactorization I = cBSF is unique.
Proof. Consider first the case IE HI(D). Let B be the Blaschke product of
the zeros of I, g = fiB. We know that g E HI(D), 1/*1 = Ig*l, 0" a.e., and g
has no zeros. Therefore, it is enough to prove the above items when B = 1.
Since I has no zeros on D we can assume 1(0) = 1. Hence log II I is
harmonic and
Jor
21f
0= log 1/(0)1 = _1 log I/(re i9 )1 de for 0 < r < 1.
21r
Writing log III = log+ III -log-III we obtain
_1 r
21r Jo
21f
log-l/(rei9)1 de = _1
21r
r
Jo
21f
log+ I/(re i9 )1 de ::: 11/110 < 00.
By Fatou's lemma the two functions log-II*I and log+ 11*1 belong to LI('ll'),
and log 11*1 E L I (1'). This shows that F is well defined, and FE HI(D). More-
over, IF*I = 11*1 =1= 0,0" a.e., since log 11*1 E L I (1'). If we could prove that
I/(z)1 ::: IF(z) I for every zED,
it will follow that cp = 1/ F is an inner function without zeros, and hence that cp
is a multiple of a singular inner function. Since log IFI = P(log 11*1), we need
to show that
log III::: P(log 1/*1).
For 0 < r < 1 and zED we have
log Ilr(z)1 = P(log+ Ilrl)(z) - P(log-llrl)(z).
On the other hand, if u, v E R+
Ilog+ U -log+ vi ::: lu - vi,
and limr-.I- Ir = 1* in L t (1'). Hence
lim P(log+ Ilrl)
r ..... l-
= P(log+ 1/*1).
Fatou's lemma now implies that
P(log- 1/*1) ::: lim inf P(log-llrl) =
r ..... l-
P(log+ 1/*1) -log III.
22 1. Boundary Values of Holomorphic Functions and Analytic Functionals
Hence
One can explicitly recover IE HP(D) from the data I*IE, where a(E) > 0
(see [Pat], [AC], [Ai]).
Having finished presenting the fundamental properties of the functions in
HP(D), we shall now tum to the consideration of the maximal ideals in the
algebra HOO(D).
The space HOO(D) is not only a Banach space, but a commutative Banach
algebra with a unit element. In fact, the required extra properties, II/glioo ~
1l/IIooIIglioo and 1· I = I, are evident. In this section we denote by M the
family of all multiplicative functionals cp: HOO(D) -+ C, that is, for II, h E
HOO(D), AI, 1...2 E C, we have
CP(Adl + A2h) = AICP(fI) + A2CP(h)'
cp(flh) = cp(fl)cp(h)·
Obvious examples of elements in M are the evaluation functionals CPI;' ~ E D,
closed and cp is continuous. This argument works in any unitary Banach algebra.
We will take the continuity of the elements cp E M for granted.
There are other elements in M besides the cp!;. One way to see this is as
follows. For a point eiO E '][' we define
which implies Icp(f)1 :::: 11111 for every I E Hoo(D). In other words, IIcpli :::: 1.
On the other hand, cp(1) = 1, which implies IIcpli ~ 1.
The set M is weakly closed: let 1jr E MW (weak-closure of M), then for
any I, g E Hoo(D) and any 8 > 0 there is cp E M such that
11jr(f) - cp(f)1 < 8, 11jr(g) - cp(g) I < 8, and 11jr(fg) - cp(fg) I < 8,
for cp E M.
The definition of the weak topology makes it evident that j is continuous on
M. Moreover, the map I ~ j is injective and is an isometry of Hoo(D) onto
a closed subalgebra of C(M):
Therefore, ker rp :2 ker rpi. and, since ker rp and ker rpi. are maximal proper ideals,
kerrp = kerrpi., that is, rp = rpi.. This shows that Jr-'(D) = 1::., hence I::. is open
in M, and Jrll::. is a continuous bijection to D. Moreover, the usual topology
of D can also be interpreted as being the weak topology defined by func-
tions in Hoo(D). Namely. let Zo E D and 0 < B < I - Izol. Then B(zo. e) =
{z E D: I/zo(z)I < e}, where !zo(z) = z - Zo E Hoo(D). Since zED corres-
ponds to rpz E I::. we have that B(zo, e) = Jr({rp E 1::.: lizo (rp) I < eD. This shows
that Jr II::. is a homeomorphism onto D. 0
The set C := M\I::. is called the corona (Le., crown). It is mapped onto T.
For every a E T, the set Ma = Jr- 1({aD is called the fiber over a. Clearly C
is compact and Ma is also compact.
1
h(z) := exp ( --2
:7r
h
1l'nB(a.s)
e +z
it
-.--log
ell -
. dt
1f*(e'I)1
z
).
1.1.46. Corollary. Let I E HOO(D) and a E T. The set /(M a ) consists 01 all
the limit points 01 the sequences (f(An»n~l lor every sequence (An)n~l in D
which converges to a, i.e., the cluster set 01 I at the point a.
A question that remained open for a very long time was the corona problem:
Is ll. dense in M? In other words, does the corona C have empty interior? The
26 1. Boundary Values of Holomorphic Functions and Analytic Functionals
affirmative answer was first given by L. Carleson in 1962. His original proof
was very hard. Since then several simpler proofs based on the idea of Tom
Wolff to use the a-operator have appeared. We follow here [Fi], the reader is
referred to [Ni3], especially its appendices, for sharper bounds in the construc-
tions of Theorem 1.1.48 as well as for constructive interpolation procedures. For
multiply-connected plane domains one can also consider the corona problem,
see [Fi], [Jo].
We need first to translate the corona problem into another form.
Proof. Let CPo E M\~. By definition of the weak: topology there exist
°
ft, ... , fn E Hoo(D) and a > such that CPO(fI) = ... = CPo(fn) = and °
{cpEM; Icp(jj)l<a, l~j~n}n~=0.
Therefore, for any ~ E D we have Icp~ (jj) I ::: a for some j. As a consequence,
n n
L: Icp~(jj)1 = L: Ijj(~)1 ::: 8
j=1 j=1
for every ~ ED. On the other hand, all jj E ker CPo, hence 1 is not in the ideal
generated by the jj. In other words, there would be no gj E Hoo(D) such that
ftgl + ... + fngn = 1
if ~ =1= M.
Conversely, assume there are fl,"" fn E HOO(D), a > 0, such that Iftl +
... + Ifn I ::: a and 1 does not belong to the ideal generated by the jj. In this
case, this ideal is proper and there is CPo E M such that CPo(jj) = 0, j = 1, ... ,n.
The neighborhood {cp EM: Icp(jj)1 < a/n, 1 ~ j ~ n} of CPo is clearly disjoint
from ~. If not, there would be ~ E D with Icp~(jj)1 < a/n, 1 ~ j ~ n, i.e.,
a
Ift(~)1 + ... + Ifn(~)1 < n- = a.
n
D
h j ..= n
J] 1 S j S n.
E Ifkl2
k=1
These functions are Coo and satisfy !1hl + ... + fnhn = 1 in B(O, p). The idea
is to modify them to make them holomorphic while preserving the desired
identity. Let Wj,k be Coo functions in B(O, p) satisfying the equations
aWj,k ahk
az = hj az , I S j, k S n.
Let
gj := hj + L (Wj,k - Wk,j)!k. 1 S j S n.
I::;k::;n
One has
n n n
Ljjgj = Ljjhj + L jjfk(Wj,k - Wk,j) = Ljjhj = 1,
j=l j=l j=l
and, moreover,
Since every w in this set is of the fonn Wo + <p, <p E .JIt'(B(O, p», with ow%z =
u, Wo E COO(B(O, p», we need to estimate
rwol de = JDr!l.(wof)
J1l'
log ~dm - 2n(wof)(O)
Izl
JD
r!l.(wof)
= log ~dm.
Izl
We have
a , au
!l.(wof) = 4 oz (uf) = 4uf + 4f oz'
Therefore, in order to obtain the desired bounds on II Wj.k 1100, we need to
estimate
r
JD
Igf'1 2 10g ~dm ::::: 2n IIgll~lIf11~.
Izi
Namely
gl' = (gf)' - g'l
and
1.1. The Hardy Spaces in the Disk 29
To see this, use Proposition 1.1.32 to write I = glg2, gj E H 2(D), and IIgjll~ =
11/111. Apply the inequality of Cauchy-Schwarz and (ii) to obtain the final
estimate.
(iv) If IE Hoo(D) and F E HI(D), then
rIF'I'llog~dm:::=:
JD Izl
2rrllFlllllflloo.
We factorize F = glg2, gj E H2(D), and IIgj II~ = 11F111 as done earlier. Then
r IF'I'llog~dm:::=:
JD Izl
r Ig;gd'llog~dm+
JD Izl
r IgIg;I'llog~dm.
JD Izl
The first term can be estimated by
8h k - 2 - '"' -
8z =f{JI£-f{J It ~ Id£·
I ~k:::n
Therefore
As a consequence,
Ig(~)1 ~ c L 1.fJ(~)I,
I~j~n
(~ ED)
then gEl.
The proof follows the same general lines of Theorem 1.1.48.
There are several natural ways of defining the spaces HP(Q), for Q a planar
domain [Fi]. Since we are interested in considering boundary values on JR of
holomorphic functions in the upper half-plane, the domain to consider is Q =
{z E C: Imz > OJ. In this case, we would like to define HP(Q) so that for
IE HP(Q) the boundary value f*(x) = I(x + iO) belongs to U(JR, dx). It
turns out that the correct definition is the following. For 0 < p < 00, we say
1.1. The Hardy Spaces in the Disk 31
IIfll p := sup
O<y<oo
(100 -00
I/(x + iy)iP dx
) lip
< 00.
The theory of these spaces in the upper half-plane is entirely similar to that of
HP(D) (see [Hot], [Du], [Fi], [KoID. This would seem entirely obvious using
the Moebius transformation
f/J:~t--+-I-
.~ +1
~ -1
of D onto Q. In fact, for p "# 00, this map does not provide an isomorphism of
HP(Q) onto HP(D), as one would expect that
where the right-hand side denotes the space of functions of the form
~ t--+- (1 _ ~)2Ip g(~), g E HP(D) (~E D).
Moreover,
(f/J-l)* HP(D) = (z + i)2lp HP(Q).
(See [Hof, pp. 129-130] for a detailed discussion.)
As a final remark on the choice of the definition of the H P spaces of the
upper half-plane, let us point out that the classical Paley-Wiener theorem (see
[PW], [DuD states that H2(Q) is isomorphic to the space L2(] - 00,0[, dt) via
1:
the Fourier transform. Namely, if F E L2(] - 00, O[dt), its Fourier transform F,
and
1 - 2 2
21T 11F1I2 = IIFIIU(]-oo,O[,dt)·
EXERCISES 1.1.
sup -
1 121f log+ I/(re i8 )ldB < 00
O<r<1 2n 0
is strictly stronger than the condition (*) in the proof of Proposition 1.1.39.
2. Show that the function 1/(1 - z) E HP(D) for any p < 1, and that it does not
belong to HI(D). Conclude that HP(D) :I Hq(D) if q :I p.
*5. Assume IE HI(D) is such that the set of all products of the form pi, p a
polynomial, is dense in HI (D). Show that I is an outer function.
*(c) For harmonic we say that U E ~P(D), 0 < p < 1, if sUPO<r<1 lIurll p < 00. Let
U
U U is real valued), show that its harmonic conjugate v E ~P(D) for every
E ~~(D) (i.e.,
p < 1. (Hint: Use that the boundary values of u can be written as a difference of two
positive Radon measures on T.)
8. Let IE HI(D) and assume f* coincides a a.e. with a function of bounded variation
g. Show that
21f
1o e- in8 dg(B) = 0, n < O.
9. Let f E HI (D).
(a) Show that if 11f E HI(D), then f is an outer function.
(b) Show that if Re f > 0, then f is an outer function.
(c) Show that if f is an inner function, then 1 + f is an outer function.
10. Verify the following two statements used in the proof of Proposition 1.1.45:
(a) Let B be a Blaschke product with zeros (an)n?:1 in D. Denote by A the set
of accumulation points of the set {an: n ~ I}, A ~ T, and fJn = 1/an • Show that B E
Jt'(C\(A U (fJn: n ~ I}).
(b) Let S be a singUlar inner function defined by a measure 11-. ~ 0 in T. Let K =
supp 11-•• Show that S E Jt'(C\K).
11. (a) Let f E Hoo(D) and IIflloo:5 1. Show that f = cB, where Icl = 1 and B is
the Blaschke product, if and only if
lim
r ... l- 10{21r log If(re
i8 )ld8 = O.
*(b) (Frostman's Theorem). Let f be an inner function neither constant nor a finite
Blaschke product. Show that the functions
f(z) - a
go(z) := 1 - iif(z)
are inner functions for any a E D. Moreover, with the exception of a E E ~ D, with
capacity C(E) = 0, every go is an infinite Blaschke product (up to a multiplicative
constant of absolute value 1).
(Hint: If C(E) > 0, we can assume E that is a compact subset of D, let v E 'P(E) be
the equilibrium measure, and consider
(iii) Let 1/1 E L 00 ([0, 2n]) be such that its Fourier coefficients en 2: 0 whenever n 2: o.
Let f E H'(D), fez) = Ln~oanzn. Show that
(Hint: Factorize f = gh, g, h E H 2 (D), Ilfll, = IIgll~ = Ilhll~, g(z) = Ln>oXnzn, h(z) =
Ln>o Yn zn , and use (i).) -
(tv) (Hardy's Inequality). Let f E H' (D), fez) = Ln~o anz n. Show that
*13. Recall that if f E A(D), the space of functions continuous in D and holomorphic
in D, then {.;- E-'lI': f(';-) = o} is a closed set of zero a-measure. This exercise shows that
any closed subset E of'lI' with aCe) = 0 is the set of zeros of a function f in A(D).
(i) Show that there is a function u: 'lI'\E -+ lR of class e', u 2: 1, u E L'('lI'), and
u(z) -+ 00 as z tends to any point in E.
(ii) Let
rp(z) := -
1 12rr eif + z .
-.--u(e'f ) dt, zED.
2n 0 elf-z
Show that rp E .1f(D) and has a continuous extension to D\E. Show that Rerp(z) -+ +00
if z E D\E tends to a point in E. Show that Re rp(z) 2: 1 in D\E.
(iii) Show that fez) := l/rp(z) is holomorphic in D, has a continuous extension to D,
vanishing on E, and Re fez) > 0 on D\E.
*14. (i) Let f E ek(D) n .1f(D), k 2: 1. Assume fez) # 0 in D. Let E be the set of
zeros of f. Write 'lI'\E as a disjoint union of open intervals of endpoints e ion and eibn ,
bn - an = In > 0, n 2: 1. Let M be a constant such that If(e iO ) - f(e i 9') I :::: MIO - rpl.
1 E lbn 10g[M(O -
Show that
21T
-00 < log If(eio)1 dO :::: an)]dO.
o n~l an
Hence E is the set of common zeros in D of f, f', ... , f(k). (See [Chol] for references
to these last two problems and generalizations.)
1.2. Hyperfunctions 35
15. Let (Zn)n~i be a sequence of distinct points in the upper half-plane 0, Zn i= i for
all n ::: 1. Assume that
~ Imzn
~--..,.<oo.
n~i 1 + IZnl 2
Show that for m ::: 0
b(z)= ( -z -i) m II -
Iz; + 11
---
Z - Zn
Z+i n~i z~ +I Z - zn
is a function in H""(O), Ib(x)1 = 1 a.e. on JR. (These are the Blaschke products in 0).
16. Let rp(z) be a Moebius map in 0, rp(z) = (az + b)/(ez + d), a, b, e, d E JR,
ad - be = 1. Show that the map
1
(Tf)(z) = --dl(rp(z»
ez+
(z E 0)
is an isomorphism of H2(0) onto itself such that IITllb = II/lb (Le., it is a unitary
map). Is there a corresponding statement for HP(O), 0 < p < oo? What about H2(D)?
17. Let 1 E HP(O) for some p, 0 < p ~ 00. Assume there is a rational function g
such that I(x + iO) = g(x - iO) a.e. on some open nonempty segment of the real axis.
What can you say about the location and multiplicity of the poles of g? What can you
say about the number of zeros of g? Does the answer depend on p? What happens when
o is replaced by D?
18. (a) Let G E .tf(D) n Li(D, dm), g E C(11'), be related to G by the fact that for
every {3 E C""(C) one has
rGa~dZ/\dz=l
JD az aD
g{3dz.
1.2. Hyperfunctions
Let Q be an open nonempty subset of the real line R We are going to intro-
duce a complex vector space 8(Q) containing as a subspace the space V' (Q) of
distributions in Q. This space 8(Q) will be called the space of hyperfunctions in
Q. As we have shown in [BG, §3.6], every distribution in Q can be obtained as
the boundary values of a pair of functions holomorphic above and below the real
axis, respectively. These holomorphic functions have to satisfy precise growth
conditions. What we are going to do is to generalize the concept of boundary
values to any pair of holomorphic functions, without any growth conditions
whatsoever. In a certain sense, the space 8(Q) is the largest conceivable space
36 1. Boundary Values of Holomorphic Functions and Analytic Functionals
where the usual operations of the differential calculus make sense. Some refer-
ences for this section and further developments of the theory of hyperfunctions
are [Schap], [Mor 3], [Kan].
In the rest of this section, Q will always denote a nonempty open subset of
JR, JR is identified as usual to the real axis of C. We denote C# := C\JR and
- #
Q=C UQ.
n is an open subset of C such that Q = n
n R Let us denote by U(Q) the
family of all open sets V in C such that V n JR = Q. Note that Q is necessarily
closed in every V E U(Q). Finally, if VI, V2 are open subsets of C such that
VI ~ V2, we denote by R V,,v2 the restriction map ~(V2) ~ ~(VI)'
By the preceding proposition for any V E U(Q) and any g E ~(V\Q), the
image under iy-I of the class of g in ~(V\Q)/ RV\Sl,v(~(V)) is a hyperfunction
in Q called the hyperfunction associated to g. Moreover, every hyperfunction
in Q is associated to a unique g E ~(V\Q). We say g represents this hyper-
function.
1.2. Hyperfunctions 37
and
h=g2+ h2 0nV2\Q.
The condition is necessary. If there is h E ~(n) such that II = 12 + h on C#,
we obtain
1.2.4. Examples.
(1) Let C+ ={z E C: Imz > OJ, C- =
{z E C: Imz < OJ, we denote by 1~
and 1; the hyperfunctions in Q associated, respectively, to the characteristic
functions Xc+ and Xc' One has
1~ = -1;.
In fact, Xc+ + Xc- = Xc' represents 1~ plus 1;, but, by Proposition 1.2.3, the
class of Xc' is zero.
(2) If T E B(Q) is represented by f E ~(V\Q), we denote by T the hyper-
function in Q which is associated to z 1-+ - f(z) E ~(V\Q), V = {z: Z E V}.
We say that T is the conjugate of T. It is clear that T = T, and that for any
scalars AI, A2 and hyperfunctions TJ, T2 we have
Al TI + A2T2 = ~I TI + ~d2'
=
One says that T is a real hyperfunction if T T. The real part of T is
Re T := !(T + T) and the imaginary part is 1m T: Oj2i)(T - T). It is easy
to see that Re T and 1m T are real hyperfunctions.
A hyperfunction is real if and only if there is f E ~(C#) representing it such
that f(z) = - f(z) for every z E C#.
(3) For every T E B(Q) represented by f E ~(V\Q), V E U(Q), we can
decompose T as
T = T+ + T-,
where T+ is associated to f+ = Xc+nv . f, and T- is associated to f- =
Xc-nv . f·
38 1. Boundary Values of Holomorphic Functions and Analytic Functionals
1~. h = -I n·h
by Example (1). It is also clear that if hI, h2 E Jt'(V), then (hi + h 2)T =
h I T+h 2T.
For the following proposition we need to recall that for any complex-valued
real analytic function qJ in Q there is an open set V E U(Q) and a holomorphic
function ell E Jf(V) such that elllQ = qJ. In fact, to say that qJ is real analytic
means that for every Xo E Q one can represent qJ by its Taylor series in an open
intervallxo containing Xo, lxo 5; Q,
=L -;--.
00 1 d j .
qJ(x) qJ(xo)(x - XO)l.
j=O J! dXl
That is, this series has a positive radius of convergence bigger or equal to r(xo),
where (xo - r(xo) , Xo + r(xo}} = lxo' The series
~ qJ(j) (xo) .
L....J ., (z - XO)l
j=O J.
clearly represents a holomorphic function in 8(xo, r(xo)} whose restriction to
lxo coincides with qJ. This way we can define a holomorphic function ell in
1.2. Hyperfunctions 39
One can verify without difficulty that T(p) is independent of the representative
taken. The maps DP: 13(Q) -+ 13(Q) are linear and they leave A(Q) invariant.
Furthermore, for cp E A(Q), DPcp is the usual derivative.
1.2.10. Remarks.
(1) One has that for cp E A(Q) and T E B(Q),
One can easily verify that TIQ I does not depend on the choice of V2 , f, or
VI. It is also clear that Rn"n 2 is a linear map: B(Q 2) -+ B(Qd.
We set B(¢) = 0 and R</>,n = 0 for every Q.
1.2.12. Proposition. Let Qj, Q 2 be two non empty open subsets ofR Then:
(1) R n "n 2 is surjective.
(2) 1~21QI = 1~, and In21Q I = In,·
(3) The map R n "n 2 commutes with conjugation, with taking real or imaginary
parts, with multiplication by real analytic functions, and with differentiation
operators. The map Rn"n2 preserves the class of holomorphic hyper/unc-
tions.
(4) If Q I ~ Q 2 ~ Q3, then
1.2. Hyperfunctions 41
Proof. We only show that Ro 1 ,o2 is surjective leaving the rest of the proof as
an exercise. Every T E B(QI) has a representative f E Jr(C#), to which one
can associate S E B(Q2). Clearly Ro h o2(S) = T. 0
The family B(Q), for Q open subset of JR, and the maps Ro 1 ,o2 determine a
presheaf on JR, this presheaf BJR. is called the (pre)sheaf of hyperfunctions.
1.2.14. Lemma. Let (Qj)jEJ bean open covering ofa nonemptyopen set Q ~ JR.
Let Tj E B(Qj), j E J, be a collection of hypeifunctions such that for every i, j
with Q; n Q j :f:. 0 we have
T;IQ; n Qj = TjIQ; n Qj .
Then there is a unique T E B(Q) such that TIQj = Tj for every j E J.
Proof. For every j let /j E Jr(C#) be a function representing Tj. Recall that
Qj = C# U Qj . For every i, j such that Q; n Qj :f:. 0 we have that gjj f; - =
/j admits an analytic continuation to a function, still denoted gij, which is
holomorphic in C# U (Q; n Qj) = OJ n OJ. If OJ n OJ n Ok :f:. 0, it is clear that
gij + gjk + gkj = 0 in C#, hence in Q; n Q j n Qk, since all three functions are
holomorphic in that intersection. By the Mittag-Leffler Theorem [BG, Theorem
1.2.23] there are gj E Jr(Qj), j E J, such that
This concludes the proof of the proposition, since we already know that the
restriction operators are surjective, which is the last point we need to verify that
BJR. is a flabby sheaf. 0
In the same way that we can talk about the support of a distribution, we can
define the support of a hyperfunction.
42 1. Boundary Values of Holomorphic Functions and Analytic Functionals
1.2.15. Proposition. Let T E B(Q) and let (Qi)iEI be a family of open subsets
ofQ such that TIQ i = 0, then TI(Ui Qi) = O.
Proof. Let Q' = Ui Qi, T' = TIQ'. Then (Qi)iEI is an open covering of Q' and
T'lQ i = 0, by the previous proposition T' = O. 0
For F a closed subset of Q, one denotes by BF(Q) the family of all hyper-
functions T in Q with supp T £; F. The previous proposition shows that this is
a well-defined concept.
The following properties of the support of a hyperfunction are easy to verify:
(1) supp( E A/Fj) £; U supp 1} (1) E B(Q), Aj E C, 1::: j ::: n).
l~j~n l~j~n
1.2.17. Proposition. Let Q be a nonempty open subset of~ and let (Fj)j"~l be
a locally finite sequence of nonempty relatively closed subsets of Q such that
Uj:'::l Fj = Q. Let T be a hyperfunction in Q. Then there is a sequence (1})j~1
of elements of B(Q) such that supp 1} S; Fj for every j and, such that for every
Q' open, Q' cc Q, there is an integer n = n(Q') so that
TIQ' = L (1}IQ').
l~j~n
I = II + gl in C#.
Let TI be the hyperfunction in Q associated to It IC#. Then supp Tl £; Fl.
To define the 1} we proceed by induction. Let k E N*, assume we have
already found a locally finite family of relatively closed subsets Fi,j of Q,
1 ::: i ::: k, j 2: i, and functions Ii E df(Vi), gi E df(V/), with V; = Q\Fi,i,
V;' = Q\(Uj~i+1 Fi,j), such that
I = 11 + ... + !; + gj in C#.
1.2. Hyperfunctions 43
TIO' = L (1j10'). o
I~j~n
1.2. IS. Proposition. Let Q be a nonempty open subset ofR and let (Fj)j"~1 be a
locally finite covering of Q by nonempty relatively closed sets. Let 1j E B(Q) be
such that supp 1j ~ Fj for every j. Then, there exists a hyper/unction T E B(Q)
such that for every relatively compact open subset Q' of Q there is an integer
n = n (Q') satisfying
TIQ' = (1jIQ') L
and
1jIQ' = 0 for every j > n.
Proof. We can write Q = Uj:::1 lj, where the lj are pairwise disjoint open inter-
vals in lR. Construct an increasing exhaustion (Kn)n:::1 of Q by compact sets as
follows:
( l)I~J~n
{x E lj: d(x,lj) ::: ~}) n [-n, nJ.
Let
(iii) For I :::: 3, the function LnEE/ fn belongs to ~«V\Q) U KI_ 1).
Applying Runge's Theorem [BG, Theorem 3.1.1] to V and the compact
subset K[ = KI-2 + i[ -I, I], we find a function hi E ~(V) such that
(2: In) -
Let now
gl := hi E ~«V\Q) U KI_I).
nEE/
It is easy to see that the series
g- L fk = L gl - L hi + L it.
no~/EEnl
The first term is holomorphic in (V \ Q) U Kn" the second in V, and the third
in (V\Q) U Q'. From this it follows that
TIQ' = L (TkI Q')·
I~k~no
1.2. Hyperfunctions 45
Furthennore, by the choice of no, Tk 10' = 0 for any k > no. This concludes the
proof of Proposition 1.2.18. 0
In [BG, Lemma §3.6.18] it has been shown that any T E e'(lR) (i.e., a distri-
bution T with compact support) can be obtained from the boundary values of
its Cauchy transfonn T, T = l/rrz T. Namely, *
1
= - 2i [T(x + iO) -
A A
T T(x - iO)].
We are now going to obtain similar results for hyperfunctions with compact
support.
1.2.21. Proposition. Let 0 be a nonempty open subset ofR and let T E 8(0)
have compact support. The function T defined in C# by
All
T(z) = -
rr
T(x)
--dx,
nZ-X
z E C#,
T(z) =~
rr
r T(t)t dt = _~rr inr Tz(t)dt = ~rr iyr tf(t)
in z - - z
dt
makes sense and it does not depend on the choice of the cycle y in Q\ supp T
of index 1 with respect to every point in supp T and index 0 with respect to z. It
is easy to verify that T is holomorphic in C\ supp T and has the value 0 at 00.
46 1. Boundary Values of Holomorphic Functions and Analytic Functionals
Let us show that flc# represents -2iT. It is enough to show that the function
-(1/2i)T - f has an analytic continuation across suppT. Then -(l/2i)T - f
would be entire and represent the zero hyperfunction. Let y be a cycle in
Q\ supp T with index 1 with respect to every point of supp T. Let Int(y) denote
the union of Int(Yj) for the components of y. We claim that for any z E Int(y) n
r
C# one has
-~T(z) - fez) = __1_. f(t) dt.
21 211: 1 JY t - z
If the claim were true, then -(l/2i)T - f would have an analytic continuation
to the Int(y). To verify the claim, let Yl be another cycle in Q\ supp T with
Yl ~ Int(y), and such that the index of Yl with respect to supp T is 1 and with
respect to z is zero. By Cauchy's formula we have
fez) = _1_. r f(t) dt = ~ r
f(t) dt __1_. r
f(t) dt
2m JY - Y1 t - Z 211:1 Jy t - z 2m JY1 t - z
= _1_.
211: 1
r f(t) dt _
JY t - Z
~T(z),
21
exist in the sense of distributions in V' (Q) and one can define its boundary
value b(f) = b+(f) - b-(f). Conversely, any element T E V'(Q) is of the
form T = b(f) for some holomorphic function of slow growth in Q (i.e., of
slow growth in some open set W as above) (see [BG, Theorem 3.6.19]).
These theorems can be translated in the language hyperfunctions into the
following proposition:
1.2.22. Proposition. For every nonempty open subset Q ~ JR, there is a canon-
ical injection of the space V'(Q) of distributions in Q into B(Q). The image
of V' (Q) under this injection is the subspace of B(Q) of classes of holomor-
phic functions of slow growth at the points of Q. This injection commutes with
restrictions.
1.2. Hyperfunctions 47
Proof. Choose a partition of the unity (an)n~1 in Q, an E 1)(Q). Let T E 1)' (Q).
Then for every n ::: 1, the function
and
if j > n.
Note that in view of the second condition, we might as well have written SIQ' =
Lj~I(SjIQ').
We want to see that S is independent of the choice of partition of unity. Let
(f3n)n~1be another Coo partition of unity in Q. The same procedure allows us to
construct a hyperfunction S in Q. To prove that S = S, it is enough to show that
for every Q' open and relatively compact subset of Q, we have SIQ' = SIQ'.
We know there is n I E N* such that
SIQ' = L (SjIQ'), SIQ = L (SjIQ'),
I~j~nl I~j~nl
and SjlQ' = SjlQ' = 0 for all j > nl. The Sj are the hyperfunctions in Q asso-
ciated to (i /2)(f3j Tf. We can also assume that Q' n supp aj = Q' n supp f3j = 0
if j > ni.
The distribution LI~j~nl (aj T - f3j T) has compact support contained in
Q\Q'. Hence, its Cauchy transform LI~j~nl «aj Tf - (f3j Tn is holomorphic
in C# U Q'. This proves that SIQ' = SIQ'. Therefore, the map T f-+ S is a well-
defined map from 1)' (Q) into 8(Q). The linearity now follows immediately
from the above construction.
To prove the injectivity it is enough to show that supp T 5; supp S.
In fact, if Q' is a relatively compact open subset of Q\ supp T, there is a Coo
partition of unity (an)n~1 in Q such that al == I in a neighborhood of Q' and
al ;: 0 in a neighborhood of supp T. The first condition on al implies that SIQ'
is the hyperfunction associated to the holomorphic function
~/ Tt , al(t)) , Z E C#.
27r1 \ t - Z
But this function is entire (and, in fact, identically zero) by the second condition
on al. Hence S IQ' = O. It follows that supp T 5; supp S.
Finally, we want to show that the map T f-+ S from distributions to hyper-
functions commutes with restrictions. Let Q' 5; Q be two open subsets of IR
48 I. Boundary Values of Holomorphic Functions and Analytic Functionals
The only thing left to prove in Proposition 1.2.22 is the fact that the image of
V' (Q) can be identified to the classes of holomorphic functions of slow growth.
This is the statement of the following Proposition 1.2.23:
1.2.24. Examples.
(1) Given a E R, the function -1/[2rri(z - a)], holomorphic in C\{a}, repre-
sents a hyperfunction in B(R) denoted 8a , supp8a =
{a}. This hyperfunction
coincides with the image of the Dirac measure at a in B(lR.) under the map
of Proposition 1.2.22. For n ::: 1, 8~n) is represented by the function obtained
differentiating this function n times, i.e., (-l)n+'n!/[2rri(z - a)n+'].
(2) Let T E B(Q), then for every n E N* and a E JR., there is an S E B(Q) such
=
that (x - a)n S T. In fact, if a ¢ Q, the solution S is unique and determined
by multiplying T by the real analytic function (x - a)-n, or, what amounts to
the same thing, if g E oIf(C*) represents T, then S is associated to the function
g(z)/(z - a)n E oIf(C#). If a E Q, let S, be the hyperfunction in Q associated
to g(z)/(z - a)n. This hyperfunction satisfies (x - a)ns, = T, but it is not the
1.2. Hyperfunctions 49
in B(O).
(3) Let us recall that Arg z is the principal value of the argument of the
complex number z in C\(] - 00, 0]), its values lie in ] - Jr, Jr[.
The function
-1 i
F(z) = -2
Jrl
. (Log Izl + i Arg(-z)) = - Log(-z),
2Jr
determines a hyperfunction Y E B(lR), which coincides with the usual Heavi-
side function, that is, the function X[O,oo[' In fact, note that Y must correspond
to a distribution since F has slow growth. Furthermore, as Log Iz I is locally
integrable on the real axis and continuous except for z = 0, and Arg z has a
jump of 2Jr across the negative real axis, one obtains immediately that
b(F) = Y = X[O,oo['
50 1. Boundary Values of Ho10morphic Functions and Analytic Functionals
EXERCISES 1.2.
+ .- 2..-1 b«
x".- -z)"),
Ism:7ra
where one is using the principal branch of the power function.
(a) Prove that supp(x~) = [0, oc[ and the hyperfunction coincides with the function
x" on its support.
To study this hyperfunction for a E Z, consider the auxiliary function
x~ I = 8(n)(x).
r(a + 1) ,,=-n-I
What is the value of the right-hand side in the last formula for a = n?
3. Is the function exp(1/z) of slow growth? What is the support of the corresponding
hyperfunction?
4. Solve the differential equation
x 2 DT+T =0
in the space B(lR).
5. (a) Show that the space of hyperfunctions with support at the point 0 E lR is isomor-
phic to the space Jf(C) of entire functions. (Include the zero hyperfunction in this space.
Hint: Identify T E B(lR), supp T S; {OJ, with F(1/z), F E Jf(C).)
(b) Prove that if F(z) = En:;:o anz n is an entire function, then the function
f(z) := ~ an zn
n:;:O n!
1.3. Analytic Functionals and Entire Functions of Exponential Type 51
is also an entire function that satisfies the additional property that for every e >
is a positive constant Ce such that
° there
(z E C).
Prove also the converse. The function I is said to be of infraexponential type or expo-
nential type zero (see [BG, Chapter 4], [Lev]).
(c) Let 0 0 be the space of functions holomorphic in a neighborhood of zero. Show
that it can be identified to ne>O Je(B(O, e»,
the intersection of the spaces of hoi omorphic
functions in the disk B(O, e) of center zero and radius e. Let L: 0 0 -+ C be a linear
functional which is continuous when restricted to every Fn!chet space Je(B(O, e».
Prove
that there is a unique sequence (an)n~O of complex numbers such that if cP E 0 0 then
I(z) := LanZ n
n~O
= Le-n
00
I(x) cosn 2x
n=O
is Coo in ffi. but its Taylor series diverges (i.e., it has radius of convergence zero) at
every point except x = 0. Construct now g E £(ffi.) such that its Taylor series diverges
at every point.
The family of all these analytic functionals is the topological dual .if' (Q)
of the space .if(Q). Since .if(Q) is a closed subspace of the Frechet spaces
C(Q) (of continuous functions in Q) and £(Q) (of Coo functions in Q). Every
analytic functional can be represented, in an essentially nonunique way, either
by a measure with compact support in Q or by a distribution with compact
support in Q. For instance, let a E Q, then the analytic functional
! t-+ !(a),
(T,h) -1-.1
= 27r1 y,
h(z)dz,
where Yr = 8B(0, r). The circle is holomorphically convex but T also has
carriers Yrl' Yr2' rl =f r2, which are not ordered by inclusion.
We will be able to prove the existence of a smallest carrier when Q is holo-
morphically convex in C, i.e., when Jt?(C) is dense in Jt?(Q). Let us recall from
[H03], [BG, Example 3.1.4] that we have the following characterization:
Proof of Proposition 1.3.3. It is clear that (1) => (3) => (4). Let K' = K02 n Q1.
It follows from (4) that K' and K" are two disjoint compact sets whose union
is a holomorphically convex subset of Q2, namely Ko 2 • For any f E Jt?(Q 1)
we can consider the function F = f on K' and 1 on K" and, by the theorem
of Runge [BG, Theorem 3.1.1], this function can be uniformly approximated
in K' U K" by functions in Jt?(Q2). In particular, Jt?(Q 2) is dense in Jt?(Qd.
Therefore (4) => (1). If we choose f =0, the fact that K 5; K' and the values
in K" of a holomorphic function in Q2 can be estimated by its values in K',
implies that K" = 0. Hence (2) also holds. Since (2) => (3) is clear, we have
the equiValence of the first four conditions.
54 1. Boundary Values of Holomorphic Functions and Analytic Functionals
aw n2 :2 w:2 L.
-
Since (2) implies that aWn2 -
= aWn! ~ Q!, we have L ~ Q!. Therefore L = 0,
as we wanted to show. D
1.3.5. Theorem. Let Q be a Runge open set and let T E Jt'" (Q). Then the follow-
ing properties hold:
(1) Let K be a carrier ofT and let w be an arbitrary, holomorphically convex,
compact neighborhood of K. Moreover, let /-Lw be a measure with support in w
representing T and let
/-L",(Z) .- -
A ._ 1
Jr
1
d/-L",(~)
w
.
z-~
The function fl", is holomorphic in W C and zero at infinity. When w runs over all
the compact holomorphically convex neighborhoods of holomorphically convex
carriers ofT in Q, thefamity flw defines, by analytic continuation, a holomorphic
function
T(z) := -
A 1( 1)
Jr
T/;,--
z-~
in the open set U(T), which is the union of the complement of these carriers.
(2) The set K(T) := C\U(T), the intersection of all the holomorphically
convex carriers of T, is the smallest holomorphically convex carrier of T in Q.
(3) Let U be an open neighborhood of K (T) and let 8 be a cycle in U\K (T)
with integral coefficients, 8 E Z! (U\K (T), Il), homologous in U(T) to the cycle
1.3. Analytic Functionals and Entire Functions of Exponential Type 55
We clearly have
inf Ifz(~)1 2: sup Ifz(~)I.
~eU, ~eV,
Let W := Wg, then w\llJ £ Q\K, and we can cover w\w with a finite number of
open sets of the form UZI ' ••• , Uz., with Zl, ... , Zn E w\w. Let V := nl<_1_
'<n VZ}"
it is a compact neighborhood of K contained in Q.
We show that V =
V by verifying successively that V is disjoint from Q\w,
w\w, and w\ V.
First, V £ w implies V £ W. Second, if z E w\w, there is a j such that
z E UZj ' Hence,
56 1. Boundary Values of Holomorphic Functions and Analytic Functionals
holds. By the Hahn-Banach theorem one can find a measure /-tw of compact
w
support in such that
(T, f) = J
f d/-tw, f E ~(w).
J
f-+
~
rr
\7's' _1_) -_
1
Z-~
/-tw z -
A ( ) - ~
rr
d/-tw(~) .
z-~
By Morera's theorem, flw is holomorphic in we and has the value zero at infinity.
Let K\, K2 be two holomorphically convex compact carriers of T, and let WI,
W2 be holomorphically convex compact neighborhoods of K I , K 2 , respectively.
Let /-tW\' /-twz be measures representing T obtained as above. For z E wi n w~,
we have
flw\ (z) = flwz (z).
Namely, since z ¢ (WI U (2), to evaluate (Ts' 1/(z - ~»), it is enough to choose
an approximating sequence to (z - 0- 1 on the compact subset supp(/-twJ U
supp(/-twz) of WI U W2.
From this it follows that the collection {flw} determines a holomorphic func-
tion f in U(T), which vanishes at infinity.
To show that T is carried by K(T) = C\U(T), we are going to prove first
part (4) of the statement of the theorem. Using that T = Tt we shall then
conclude that T is carried by K(T). Statement (3) will also be true since it is
a consequence of (4).
1.3. Analytic Functionals and Entire Functions of Exponential Type 57
[ .7(z)h(z) dz = [ .7(z)h(z) dz
Ja Ja 1
for every h E .1f(C), since 8 and 81 are homologous in U (T). Since .1f(C) is
dense in .1f(Q), this identity also holds in .1f(Q). This shows that T:r is carried
by K.
Let T = T:r, to see that the function T coincides with .7, we observe that both
functions are holomorphic in C\K, zero at infinity, and that for Izl sufficiently
large we have
T:r(z) = 2. / T/;,
7r \
_1_)
z- {
= _~
27rl
[.7R)d{ = .7(z),
Ja ~ -z
by Cauchy's formula. (Note that 8 turns clockwise around z when Izl is large,
in other words the index Inda(z) = -I, and also .7(00) = 0.) Since C\K is
connected, T = .7 everywhere in C\K. This ends the proof of (4).
To finish the proof of the theorem we still have to show that if T E .1f'(Q)
is given and if .:J = T, then T = T:r. We have shown in the proof of (4) that
T = T:r.
For any S E .1f'(Q), the Laurent development at infinity of S is of the form
S(z)
A
= -7r I( S/;, - ,
Z -
I) = -1
.,
L(S/;, ~ )
7r n;::O
n
Z
1
n+I'
We are now going to consider the case which occurs most often in practice:
Q is an open convex subset of C. We shall see that for any T E Je"'(Q), there is
a smallest compact convex carrier in Q. It is clear that when Q is convex, there
are convex carriers for T. Namely, for any carrier K of Tin Q, the convex hull
cv(K) is a convex compact subset of Q, which still carries T.
If T E Je'" (Q) is carried by the compact convex subset K of Q, the function
T is holomorphic in K C and we have by the previous Theorem 1.3.4 that, for
1
every h E Je"(Q)
1
(T, h) = --:- T(z)h(z) dz,
A
21 y
C(T)(z) = / T/;,
\
_1_)
z-~
(z f. K(T».
C is a linear isomorphism by Theorem 1.3.5 and Remark 1.3.7. Note that with
this notation we have
(T, h) =~
21f1
1 .y
C(T)(z)h(z) dz.
1.3.10. Proposition. Let Q be a convex open set and let T E ;e' (Q) have as
convex support the set K, then for every e > 0 there is a constant Ce > 0
such that
(z E C),
Then, there is a unique analytic functional T E ;e' (Q) such that 'J(T) = f.
Proof. If f(z) = Ln~o anz n, let
"" n!an
B(w) := ~ w n+ 1 •
n~O
(It is enough to choose R > 0 so that K S; B(O, R). If Ro is the smallest radius
such that K S; B(O, Ro), we can choose e > 0 with Ro + e < R, Ce = C, and
use HK(z) :::: HiJ(O,Ro)(Z) = Rolzl.)
60 1. Boundary Values of Holomorphic Functions and Analytic Functionals
p>O
()
Ian I :::: inf max If(z)l/ pn
Izl=p p>O
~~
:::: C inf(e Rp p-n) = C--.
nn
Moreover, Stirling's formula states that
n! = nn e-nJ27l'n(l + sen)),
sen) ~ 0 as n ~ 00. Hence
This implies that the series defining B converges whenever Iwi> R as we had
claimed.
Besides the purely formal way we have used to introduce the function B,
we observe that B coincides, in the half-plane Re w > R, with the Laplace
transform of the function f. In fact,
1 o
00
L
n~O
lanltne-t(Rew) dt = Ln!lanl(Rew)-n+l
n~O
< 00,
Figure 1.1
°
Since z E Pu , there is a > such that Re zu ::: HK(u) + a. In fact, the same
inequality is valid in a half-plane strictly contained in Pu • Hence
HK(t) + 81tl- Rezt = S(HK(U) +8 - Rezu) ::: S(8 - a).
CPu(z):= r !(t)e- zt dt
lLu
converges uniformly in the half-planes {z E C: Rezu > Hdu) + a}, for every
a > 0. Therefore CPu is a holomorphic function in Pu.
Let us now consider two distinct values UJ, U2, Iud = IU21 = 1, < °
Arg(u2lul) < Jr.
For p > 0, let YP be the loop in C formed by the portions of LUI and LU2
inside B(O, p), together with r p , the secant to the smallest of the two arcs
determined by the two rays in aB(O, p) (see Figure 1.2). We claim that
lim
p-+oo
r !(t)e-
lrp
tz dt = 0,
Figure 1.2
= sup(A(Re~ul) + (1 - A)(Re~u2»
l;eK
~ AHK(Ul) + (1 - A)HK(U2),
which proves the convexity of HK. On the other hand, Z E PU1 n Pup hence for
some a > 0 we have
ReZUl = (zlul) ~ HK(ud + a,
Rezu2 = (zlu2) ~ HK(U2) + a,
and therefore,
Retz = p(zlAul + (1- A)U2) = p[A(zlul) + (1 - A)(zlu2)]
Let us compute J(T)(z) for this analytic functional T. We set h(~) = eZ~ in
the definition of T and we see that we can replace y by the circle of center °
and radius p > R because h is an entire function. We find
= 2:n!an2Jri
1 d~ = =
I e~Z
n+J 2: anzn f(z).
n;::O Y ~ n;::O
The third identity is valid by the uniform convergence of the series repre-
senting B on y.
Finally, the equation J(TJ) = J(T2) for h T2 E Je'(Q) implies that
(TJ, ~n) = (T2, ~n) for every n ::: 0. Namely, for z E C fixed, the series e~z =
2:n;::O(zn~n In!) converges in Je'(Q), hence for T E Je'(Q) we have
J(T)(z) = 2:
n;:: O
(T, ~n) zn.
n!
This shows that J(T) determines (T, ~n) for all n ::: 0. The density of Je(C) in
Je(Q) implies that TJ = Tz. This concludes the proof of the theorem. 0
1.3.12. Definitions.
(1) The function B introduced in the proof of Theorem 1.3.11 is called the
Borel transform of the function ! and the Cauchy transform of the analytic
functional T.
(2) The representation of ! = J(T) in the form
f(z) = 21Jrl.1 y
B(t)e tz dt
convex subset K of Q such that for every e > 0 there is a constant Ce > 0 so
that f satisfies the estimate
T ~~(T).
f ~B(f).
T ~C(T) =Td.
The diagram below commutes
=
If u eirp , 0::: ({J::: 21l', HK(u) is the maximum value in K of Re(zeirp) =
=
(z le- irp ) x cos ({J - Y sin ({J, which is the projection of z on the ray Lu directed
by u= e- irp . We have already used this fact above.
It is an immediate consequence of the above definition that HK = Hcv(K),
where, a usual, cv(K) is the convex hull of K. For that reason we will assume
K is a convex compact set in what follows:
1.3. Analytic Functionals and Entire Functions of Exponential Type 65
1.3.14. Proposition.
(1) If K\ and K2 are two nonempty convex compact sets, then
hf () · log If(rz)1
z = 1Imsup
r .... oo r
(z E C*).
1.3.16. Proposition.
(1) Let II, h be two entire functions of exponential type, then
hfIh :::: hf) + hh' hj)h :::: hj) + hj2'
hf)+h :::: max(hfI , hfJ, hj)+h :::: max (hj) , hj).
(2) If fez) = e(a+ib)z, a, b E~, then
h
E
1
-log If(rw)1 :::: -12 1
-log If(r(w + u))1 dm(u),
r 7rp B(O,p) r
one obtains
. 1
hf(w) = hmsup -log If(rw)1
r-->oo r
:::: - 12
7r P
h .
B(O,p)
1
hmsup-Ioglf(r(w+u))ldm(u)
r-->oo r
= ~
7rp
r
} B(O,p)
hf(w + u)dm(u)
The above steps are justified applying twice the following variant of Fatou's
lemma:
1.3. Analytic Functionals and Entire Functions of Exponential Type 67
limsup
r ..... oo Jx
rgrdlL ~ Jxr(limsupgr)dlL.
r ..... oo
Proof of Lemma 1.3.1S. It is a consequence of the usual Fatou's lemma [Rud2]
once one observes that if r E [n, n + 1[, then
sup fs ~ sup fs ~ sup fs.
s~n+1 s~r s~n
hence
inf
r>O Jx s~r
r (sup fs) = inf r (sup fs) dlL.
n~1 Jx s::::r
o
Now we use that for each WE C we have hi(w) = limsupj ..... oohf(w + ~j)
for a sequence ~j ~ 0, which depends on w. Then, from the inequality preceding
Lemma 1.3.18, we have
~~ ~ limsuphj(w+~j+u)dm(u)
rrp JB(O,P) j ..... oo
~~ ~ hj(w + u) dm(u).
rrp iB(o,p)
The last inequality is due to the upper semicontinuity of hi- It follows that hi
is subharmonic in the extended sense. Now we shall try to prove that hf "¥= -00
and that it is a continuous function in C*. It will follow that h f = hi in C*,
hence hf will be subharmonic in C*. Since both functions are homogeneous it
is enough to consider hf restricted to the unit circle.
1.3.19. Lemma. Let 91, 92 be two real numbers such that 0 < (h - 91 < rr.
Assume hf(eUh ) ~ hi and hf(e ilh ) ~ h2 and let
g(9) := hi sin(82 ~ 9) + h2 sin(8 - ( 1) •
s1O(82 - 8d
Then
Since g is the only function of the form a cos 8 - b sin 8 which takes the
values hj at 8j, this lemma says that there is a unique exponential fo(z) =
exp«a + ib)z) such that hfo(ei8j) = hj and hf(ei8 ) ~ hfo(e i8 ) for 8 1 ~ 8 ~ 8 1•
Proof of Lemma 1.3.19. Let ~ > 0 and g&(8) = a& cos 8 - ba sin 8 be the unique
sinusoidal of the form a cos 8 - b sin 9 taking the values hj + ~ at 8j • Let
fa(z) = f(z)e-(ol+ibl)z.
68 1. Boundary Values of Holomorphic Functions and Analytic Functionals
Then Ih(z)1 = If(z)le- g6 (1!)l z l, z = Izle i9 , is bounded in the rays L" L 2, where
L j = {se i9j : s :::: OJ. Since 0 < (h - (h < 1(, we conclude from the Phragmen-
LindelOfprinciple (see [BG, §4.9.37]) that 1181 is bounded by a positive constant
M in the sector (h ~ e ~ e2. Therefore, If(z)1 ~ M exp(g8(e)lzl) and hf(e i9 ) ~
g8(e). Passing to the limit 8 -+ 0 we obtain the lemma. 0
Let us remark that if hf (e i91 ) = -00 and hf (e ilh ) = -00, the above lemma
implies that h f (e i9 ) = -00 fore\ ~ e ~ e2. It is enough to take h\ = h2 = -M,
for some large M > 0, and let M -+ 00.
1.3.20. Lemma. Let e\ ~ e2 ~ e3 be such that 0 < e3 - e\ < 1(. Let g (e) =
a cose - bsine be such that h f (e i91 ) ~ g(e\) and hf(e ilh ):::: g(e2), then
hf (e i93 ) :::: g(e3 )·
Proof of Lemma 1.3.20. Let us assume there is 8 > 0 such that hf (e i93 ) ~
g(e3 ) - 8. Define the auxiliary sinusoidal
sinCe - e\)
g8(e) = gee) - 8 . (e
sm 3 - \
e ).
Then g8 (e\) = g(e\), g8 (e2) < g(e2), and g8 (e3) = g(e3) - 8. From the hypo-
theses and Lemma 1.3.19 we conclude that
'£I
hf(e' 2) ~ g8(e2) < g(e2),
and
1.3. Analytic Functionals and Entire Functions of Exponential Type 69
is absolutely and uniformly convergent for Re(zeilJ ) ::: hf(eilJ ) + e, e > 0, where
L = {se ilJ : s ::: OJ. Since the convex support of T is the smallest compact convex
region on whose complement B is holomorphic, it follows that HK(e ilJ ) :::
hf(eilJ ).
On the other hand, let e > 0 and use the P6lya representation
f(z) = 21.
1l"1
rB(t)e
Jy
tz dt,
with y a rectifiable Jordan curve such that K S; int Y S; V (K, e). Then
We also have h(e iBz ) = Re G(e iB2 ) and h(z) < Re G(z), when z E aB(e iB2 , p),
01 ::: Arg z < O2 • This contradicts the mean-value property for the subharmonic
function h - Re G. Therefore, we can conclude that
We claim that this inequality shows that h is convex. First, we observe that
Lemma 1.3.20 and its consequences can now be applied to h in place of hf'
hence h(e iB ) > -00 everywhere.
Now let g be a sinusoidal chosen so that g(OI) = h(e iB1 ) and g(03) = h(e iB3 ).
Let Zl, Z3 be any two points with arg Zl = OJ, arg Z3 = 03 and z = AZI +
(1 - A)Z3, 0 < A < 1. Then 0 = arg z E [0 1 ,03] and
h(z) = Izlh(e iB ) ::: Izlg(O) = Izl(a cosO - b sinO)
1.3. Analytic Functionals and Entire Functions of Exponential Type 71
Let us coIisider a few special cases of compact convex carriers K. The first
special case occurs when K = {OJ.
(? = n
OEQ
JIf(Q),
then
{T,({J} = Lan{T,~n).
n~O
72 1. Boundary Values of Holomorphic Functions and Analytic Functionals
so that
Hence
q;(n) (0)
(T,q;) = Ln! i n - , - = Linq;(n)(o).
n~O n. n~O
where
i(D)q;(~) := L inDnq;(~)
n~O
Consider now the case where K = [a, b], -00 < a < b < 00, is an interval of
the real axis. Any analytic functional T carried by K will have as a Fourier-Borel
transform i = ~(T) an entire function of exponential type such that hf = hKI'
for some KI = [aI, bd, a :5 al :5 b l :5 b. That is,
'(}
hf(e' ) = max{al cos8, b l cos8}.
Moreover, the function -(1/2i)T is holomorphic in C\K and represents a
hyperfunction S with compact support, supp S 5; K. Proposition 1.2.21 indi-
cates that S should be identified to the analytic functional T, in fact, we have
already done so in Proposition 1.2.22, when the analytic functional T is induced
by a distribution with compact support in K. Note that supp T in the sense
of hyperfunctions coincides exactly with the smallest holomorphically convex
carrier K (T) of T. This follows from Theorem 1.3.5 and the observation that
any compact subset of IR is holomorphically convex.
In other words, let K be the compact subset of IR and let the space Jf"(K) of
functions holomorphic in a neighborhood of K be considered with the projective
limit topology induced by the identity
Jf"(K) = n
K<;O
Jf"(Q),
where Q runs over all open sets (or all Runge open sets) containing K. Then
any element T E Jf"'(K) is an analytic functional which is carried by K and
it can be represented, in the sense of hyperfunctions, as boundary values of a
1.3. Analytic Functionals and Entire Functions of Exponential Type 73
The reader should also observe that when K ~ JR, JIf(K) coincides with the
space A(K) of real analytic functions in a real neighborhood of K and we
recall from the previous section that the hyperfunctions with support in K are a
module over A(K). This is clear, since if T E JIf'(K), cp E JIf(K), cpT makes
sense in JIf' (K).
Later in this chapter we will discuss a similar situation for analytic functionals
on the unit circle.
It is clear that we need some precision about the properties of projective limits,
inductive limits, and their duals. In particular, to be able to study the topological
properties of Exp(Q), JIf' (Q), JIf~(QC), etc. For this reason the following section
will be an overview of the functional analysis tools we use in the remainder of
the book.
EXERCISES 1.3.
1. Let a E n and let n be an open set in Co Show that the distribution T = 2::=1 c"8~"),
8~"} = (a" /az")8 a can be represented by a measure when considered as an analytic func-
tional in n. Give examples of non trivial measures that represent the zero analytic
functional.
2. Show that if T is an analytic functional so that there is A ::: 0 with the property that
J'(T)(z) = o (exp(A I Rezl + 8Izl), then T is carried by a compact subset of JR. Find
a function F holomorphic in C\([-A,Al+i{O}) such that T =b(F) in the sense of
hyperfunctions.
3. Verify that C = B . J' as stated after Definition 1.3.12.
4. Find the analytic functional T whose Fourier- Borel transform is I (z) = (sinh z) / z.
How about S such that J'(S)(z) = I(JZ).
5. Let T be an analytic functional with smallest holomorphically convex carrier K
and let n be a Runge open set, KEn, I E JIf(n). Define a function g by
g(z) := (T~, I(z + ~»).
Show that g is a holomorphic function in a neighborhood of the origin. Let n' =
{z E C: z + K S; n}, show that n' is open and g E JIf(n').
6. Let K be a compact convex subset of C, k ::I 0. Show that T E JIf'(K) satisfies
the following property:
3C > 0 such that VIE JIf(C): I(T, 1)1 ::s CII/IIL2(K)'
if and only if there is h E L2(K) so that
7. Let P(D) = 2:o~Oj~" ajDj be a differential operator with constant coefficients and
its symbol P'(~) = 2:o~j~" (-l)j aj~j, a nonzero polynomial of degree n. Show that if
74 1. Boundary Values of Holomorphic Functions and Analytic Functionals
Use this to show that the kernel of the mUltiplication map ,;t?'(C) -+ ,;t?'(C) given by
T 1-+ zT is the vector space c80• c E C.
9. Let T E ,;t?'(C). Show that the following four conditions are equivalent:
(i) T can be represented by a distribution S E V' (C) with support in a finite number of
points at • ...• an E C of the form S = Lk,l Ak,18~? (the sum is necessarily finite).
(ii) j is a rational function. zero at infinity. and with poles contained in the set
{at •...• an}
(iii) ~(T) is an exponential polynomial of the form
~(T)(~) = L Pk(~)eakl;.
l:5k:sn
Pk polynomials.
(iv) There is a nonzero polynomial P E C[~] such that the product P . T = O.
10. In this exercise. for P E C[z] we shall denote by P* the polynomial P*(z) :=
P(Z). For two nonzero polynomials p. Q we shall study the bijectivity of the operator
Q(D)P: ,;t?(C) -+ ,;t?(C) given by Q(D)P(f) := Q(D)(Pf). Pf being the ordinary
multiplication of functions. 1m P denotes the image of the multiplication operator.
(a) Show that Q(D)P is injective if and only if 1m P n ker Q(D) = (OJ.
(b) Show that Q(D)P is surjective if and only if ,;t?(C) = 1m P ker Q(D). +
(Hint: Recall Q(D) is surjective.)
=
(c) Show that the condition deg P deg Q is necessary for the operator Q(D)P to be
bijective. Is the converse true?
(d) Consider the Fock space A 2 (C) of all functions
(This part of the exercise requires the use of duality and some properties that appear in
the next section.) The bijectivity of P*(D)P in Jt'(C) is called the Fischer problem, it
can be shown to be true. The corresponding problem in two or more variables has not
yet been solved (see [NS), [MerS), [MerY)).
*11. (a) Let T E Jt"(C) be carried by B(O, R), and define Tn E Jt"(C) by
Show that Tn -+ T weakly in Jt"(B(O, R». That is, for every I E Jt'(B(O, R», one has
lim (Tn, f)
n-+oo
= (T, f).
In what follows, K = K(T) and S is an analytic functional carried by L ~ aB(O, R).
(b) Show that if K ~ B(O, R)\L, then ST is a well-defined analytic functional carried
by K.
Define <II: Jt"(B(O, R» -+ Jt"(C) as follows. Let Tn be defined as in part (a),
then Dom(<II) := (T E Jt"(B(O, R): limn-+oo STn exists weakly in Jt"(C)} and <II(T) :=
limn-+oo ST, if T E Dom(<II).
(c) Show that if K n L = 0, then T E Dom(<II) and <II(T) = ST.
(d) Show that Dom(<II) is dense in Jt"(B(O, R». (Note that Jt"(B(O, R» ~ Dom(<II)
and Jt'(B(O, R» = (Jt"(B(O, R»)'. It follows that if h E Jt'(B(O, R) is orthogonal to
Jt"(B(O, R», then hIB(O, R) = 0, hence h == 0.)
(e) Define the transpose <III: Jt'(C) -+ Jt'(B(O, R», by Dom(<III) := (h E Jt'(C): T ~
<II(T)(h) is continuous on Dom(<II)}, and <II1(h)(T) = <II(T)(h). Show <III is well defined.
(f) If Dom(<III) =/: {OJ, show that for any h E Dom(<III)\{O} one has
Conclude that S = <III (h)/ h in a neighborhood of B(O, R). Show that S is a rational
function vanishing at infinity.
(g) Conversely, show that if S is a rational function vanishing at 00, then
Dom( <III) =/: {O}.
(h) Show that Dom(<III) =/: (OJ, if and only if ~(S) is an exponential polynomial with
frequencies in L, i.e.,
°
l~j~m
12. Let I be holomorphic in the first quadrant, Q = {z: Rez > and Imz > OJ,
°
continuous in the closure and of exponential type, I/(z)1 ::::: Boe Bdzl for ZEn. Assume
further that for some constants c ~ 0, M ~ 0, A ~ one has
I/(iy)1 ::::: Aecy (for y ~ 0),
I/(x)1 ::::: M (for x ~ 0).
Show that
I/(x + iy)1 ::::: max(A, M)e cy ,
for all x ~ 0, y ~ 0. (Consider the auxiliary function g(z) := I(z)e icz • and use the
Phragmen - Lindelof principle.)
76 1. Boundary Values of Holomorphic Functions and Analytic Functionals
13. Let I be holomorphic in H = {z: Imz > OJ, continuous in ii, of exponential type
in ii, and such that: (i) hf < 00; and (ii) I/(x)1 :::: M for all x E JR. Show that
I/(x + iy)1 :::: M exp(hf(i)y),
for 0 < y < 00. (Start by showing that for every e > 0 there is Ae > 0 such that I/(z)1 ::::
Ae exp«hf(i) + e)y). For fJ > 0, Xo » 1, and A > 0, consider
+ A)]/(z + xo) exp(i(hf(i) + e)z),
g(z) := [z/(z
where fJ > 0 and Xo » 1 are chosen so that I/(x)1 :::: K + fJ for x 2: Xo. First, prove that
limy-++ oo I(xo + iy) exp(i(hf(i) + e)(iy)) = 0 and show that for A conveniently chosen
one has Ig(iy)1 :::: K + fJ. Hence, I/(z + xo)1 :::: [Iz + AI/lzl](K + fJ)e(hj(i)+e)y, for x 2: 0,
Y 2: 0.)
15. Use the previous exercise to show that in the case K = 0, then
lim I(x
x-+-+oo
+ iy) = 0
uniformly for y in any compact subset of ]0,00[. (Use Montel's theorem.)
that hS J = S. Prove that S2 := T - 8SJ/8i has compact support contained in the set of
zeros of h and, moreover, r(S2) = R.
(i) Show that for the analytic functional R from part (h), its Cauchy transform R is a
rational function. What can you say about R?
1.4.4. Definitions.
(1) A Hausdorff LCTVS E is said to be a Schwartz space if for every convex
balanced neighborhood U of 0, there is another convex balanced neighbor-
hood V of 0 such that V ~ U and the map Epv --+ EPu is compact.
78 1. Boundary Values of Holomorphic Functions and Analytic Functionals
One can prove that any FS or DFS space is reflexive, and that the strong dual
of a DFS space is an FS space.
In particular, a DFS space is an C'J space.
1.4.5. Proposition.
(1) If E is a Schwartz space and F is a closed subspace of E, then F and E/ F
are Schwartz spaces.
(2) A product of Schwartz spaces is a Schwartz space.
(3) A countable inductive limit of Schwartz spaces is a Schwartz space.
1.4.8. Proposition.
(1) Let Un: En+! ~ En be a sequence of compact linear maps between Frechet
spaces. The projective limit of the sequence (En)n~! is an FS space.
(2) Let Un: En ~ En+l be a sequence of compact linear maps between Frechet
spaces. The inductive limit of the sequence (En)n~l is a DFS space.
The following two propositions constitute the duality principle that underlies
the relation between Harmonic Analysis and Complex Analysis:
1.4.12. Proposition. Let E and F be two FS spaces or two DFS spaces. In order
that a continuous linear map u: E ..-.+ F be an isomorphism of E onto 1m u, it is
necessary and sufficient that tu: F' ..-.+ E' be surjective.
The map tu is surjective if and only if u is injective and has a closed image.
To verify that 1m u is closed it is enough to show it is sequentially closed.
It is clear that the following result must be well-known but we have been
unable to find an explicit proof for it. The proof below was kindly provided by
Marcel Grange (Univ. Bordeaux I).
If the claim were not true, there would be no ~ I and a sequence (Ym)m~l
of points in the unit ball of Eno such that vno(Ymlm) ¢ P. This means that
vno(Ymlm) E F - x. Hence, there are fm E F such that fm = x + vno(Ymlm).
Let nl ~ no and x E En, be such that vnl (x) = x. If we let Xm = X +
Vnl-lo'" oVno(Ymlm) E En" then Xm ~ x
in En, and vnl (x m) = fm E F. Since
x ¢ F1, we have a contradiction, therefore P is bornivorous.
The definition of bornivorous ensures that V;;l(p) is a neighborhood of 0 in
En for any n ~ 1. (In fact, it contains the ball of radius IrAn.) This means that
P is a neighborhood of 0 in the inductive limit E. Since x E F, we must have
F n (x + P) =j:. 13. Let Y E F n (x + P), then y - x E F - x, and also Y - x E
P. This is a clear contradiction with the definition of P. It follows that F = Fl,
which is what we wanted to prove all along. D
(
SUP
k>l
Ihk(W)I) ( sup - I
z.~eKn 21r
1
8
Idtl
It - zllt - ~I
)
weKn+ l
Exp(n) = U EXPK(n).
K£;O
(T, h) = 21.
TO
r
iy
C(T)(z)h(z) dz.
the restriction maps from Je(V(K, lin» into Je(V(K, 1/(n + 1))), which are
compact. It is easy to verify that, when K is compact and convex, the FS space
Je'(K) is isomorphic, via the Fourier-Borel transform, to the Frechet space
Exp(K) = {f E Exp(C): Vs > 0, 3Cg > 0 such that
If(z)1 :::: CgeHK(z)Hlzl for all z E q,
with the norms
so that
= (ioj F/L(~).
More generally, if P is a polynomial then
if I~I ~ R.
1 g(;)ei~t
Fourier transform we can choose
00
({J(t) := _1 d;,
2rr -00
so that
Im(t)1 A'
< _e(B-t)'1 l°O+ i '1 Id~ I
..,.. - 2rr -oo+i'1 (1 + I~ 1)2 .
Let us remark that we have also shown in this proof that when f satisfies (*)
then f = :FfJ" supp(fJ,) £ [-B, B].
We can introduce in the space PW the topology of the inductive limit of the
Banach, spaces PWn , n E N*,
On the other hand, the space £ (JR) is an FS space so that it is reflexive and £1 (JR)
is DFS. As a consequence of Proposition 1.4.12 we now obtain the following
result:
1.5.1. Definition. Let Q be a convex open set and let T be an analytic functional
whose convex support is the compact convex set K. For any I holomorphic in
the convex open set Q + K, the function T * I is defined by
(T * f)(z) = r
iV(K,S)
I(z + l;) df.L(~),
which is defined in ~Ls' o
1.5.3. Proposition. The map T*: ;f(Q + K) --+ ;f(Q) given by I 1-+ T *I is
linear and continuous.
Proof. If L is a compact convex subset of Q and e > 0 is chosen sufficiently
small so that L S; Q-s, we represent T by a measure f.L of compact support in
V(K, e), then
sup IT
zEL
* l(z)1 ::: ( sup
wEL+V(K,e)
I/(W)I) (riV(K,e)
dlf.Ll) . 0
°
We recall from § 1.3 that an analytic functional T =1= is a differential operator
of infinite order if and only if K(T) = {OJ. (More precisely, T can be identified
to such an operator.) In this case
K(T * S) = K(S)
for any analytic functional S.
An important property of the convolution is the following:
in other words, S = T * R. o
It is not generally true that if Q is a convex open set, K = K(T) =1= 0, then
T*: Jr(Q + K) ~ Jr(Q) is surjective. It depends on more subtle properties of
the function ~(T).
1.5.9. Definitions.
(1) A measurable subset E ~ [0, oo[ is said to be of relative zero measure
if
. ml(E
11m n [0, rD
=0,
' ..... 00 r
where m I denotes the Lebesgue measure on the real line ~.
(2) Let <p be a function, defined on ]0, 00[, for which there is a set E of
relative zero measure such that the limiting value of <p(r), when r ~ 00 while
r f/ E, exists. We denote this value by lim;..... 00 <p(r). (Note that this value is
independent of the exceptional set E.)
88 1. Boundary Values of Holomorphic Functions and Analytic Functionals
I r~oo r
and, furthermore, these limits are uniform in e.
We refer the reader to [Lev, Chapter 3] for the properties functions of
completely regular growth.
L pj(z)eUjZ ,
l~j~m
We claim that the entire functions T * (zj eaz ) vanish identically when 0 ::; j ::;
m - 1. In fact,
and the coefficients of the last sum vanish identically when 0 ::; j ::; m - 1.
This shows that all the exponential polynomials appearing in the statement of
this proposition lie in the kernel of T*. Let us denote
V := (f E Je(Q + K): T *f = O},
the kernel of T* as a map from Je(Q + K) into Je(Q), and
N := span{zjeaz : ~(T)(a) = 0,0::; j ::; m - 1, m = multiplicity of a}.
Hence N = V.
To show the surjectivity of T* onto Je(Q), we have to show that
I(T*): Je'(Q) ~ Je'(Q) is injective and has a closed image. The injectivity
*
is part of Corollary 1.5.6, since I(T*)(f.L) = T f.L. Furthermore, the preceding
=
computations show that Im(I(T*» Nl., which proves that Im(I(T*» is closed
in Je'(Q + K). 0
T *f = Laj/(j).
j?:O
Here we have P = J(T), T has convex support {O}, and h p == O. It follows
from [Lev, Chapter 3] that P has completely regular growth.
We shall see in Chapter 3 that all the exponential polynomials have completely
regular growth, hence Proposition 1.5.12 applies to any T that is a linear combi-
nation of the derivatives of Dirac point masses.
Let us remark that the last part of the proof of Proposition 1.5.12 can be
easily modified to obtain the following result.
EXERCISES 1.5.
1. Prove in detail Corollary 1.5.6. In particular, discuss the meaning of the identity
t(S*)(T) = t(T*)(S), when a priori t(S*), t(T*) depend on an open convex set S'2 and
the convex supports K and L of Sand T, respectively.
2. Prove Corollary 1.5.10.
3. Show that if P(z) is a nonzero polynomial, cx E C, T E Je'(C). then T *
(P(z)e'XZ) == 0 implies that ~(T)(~) vanishes at ~ = cx. with multiplicity ~ deg P.
More generally, show that if 1 is an exponential polynomial with frequencies CXj.
I(z) = Ll::::j::::n Pj(z)e ajt • and T * 1 = O. then v(~(T), CXj) ~ deg Pj , 1 .:::: j .:::: n, where
v(~(T), CXj) is the order of CXj as zero of ~(T). Prove Proposition 1.5.13.
4. Prove that if S'2 is a convex open subset of C and P(D) is a nonzero infinite
order differential operator with constant coefficients, then P(D): Je(S'2) --+ Je(S'2) is an
open map.
*5. (Theorem of Cramer-P6Iya.) Let 1 be a function holomorphic in an open set
S'2 that contains the half-plane {z: Rez > a}. Let us assume that in that half-plane the
function 1 is represented by a Dirichlet series
I(z) = Lane-Ant
n;::O
1.5. Convolution of Analytic Functionals 91
cp(z) := L ang(An)e-).'z
n2:0
6. (a) Let n = B(O, .,fi), and let a(z) := sinz/z, fJ(z) := sinhz/z. Show that a, fJ E
Exp(n), a, fJ have not common zeros, and afJ ¢ Exp(n).
(b) For n E N*, let r n be the contour in Figure 1.3 below.
Show that the function In defined by
r eZw
In(z):= Jr. a(w)fJ(w) dw
is entire.
(c) Show that the sequence (fnk~_l converges in Jf(n) to the function I given by
r eZw
I(z):= Jr a(w)fJ(w) dw,
Figure 1.3
92 1. Boundary Values of Holomorphic Functions and Analytic Functionals
Figure 1.4
(d) Let T and S be the analytic functionals such that :F(T) = a and :F(S)(~) =
a(~)/(~ - n). Show that S E .}t"(O) and
(S, f) =
r/ eZW
Jr \ S" a(w){J(w)
) 2ni
dw = (J(n)·
(T * fn)(~) = Jr.
r e~w
(J(w) dw,
and deduce from this formula that for I~ I < ~ one has
Conclude that T * f = O.
(f) Prove that if R is the analytic functional such that :F(R) = {J, then R * f = O.
7. Let v E C, Re v > - ~, and let Tv be the analytic functional defined by the measure
Xv, of support [-1, 1] ~ JR,
for -1::::;t::::;1.
1.5. Convolution of Analytic Functionals 93
Compute explicitly F(Tv ) and show these functions are of completely regular growth,
by applying directly the definition given in the text. (Hint: Look up the definition of the
Bessel function in a table, for example, in [GR].)
8. Let T = 8_ 1 + L i + 81 + 8i , compute F(T) and show directly it is a function of
completely regular growth.
9. Show that if E ~ [0, oo[ has finite Lebesgue measure then it has zero relative
measure.
10. Prove that a function f e Exp(C) is an exponential polynomial if and only if its
Borel transform B(f) is a rational function. Show that the coefficients of this exponential
polynomial are constant if and only if the poles of B(f) are simple. What do these
characterizations mean in terms of the analytic functional T and of its Cauchy transform
f, where f = F(T).
= =
11. Let g F(T) and h F(S), T e .1'('(B(O, .» and S e .1'('(B(O, u». Let B(g),
B(h), and B(f) be the Borel transforms of g, h, and f := gh. Show that, for any E > 0,
(b) We assume from now on that B(h) is holomorphic for Iwl > R > 0, and that it
has at least one singularity on Iwl = R. We assume also that RI is such that B(h) is
meromorphic for Iwl > RI and there is a singularity on Iwl = R I . Up to a change of
variables, we can assume this singularity occurs for w = R I •
Let us enumerate the poles f3j in a decreasing order for the lexicographic order. (That
is, Z -< ZI is equivalent to either Rez < Rezl or, if Rez = Rezt. then Imz < ImZI')
Show that RI -< RI + (131 - f3i) and that RI + (131 - f3i) is exterior to B(O, R I ) for
i = 2, ... ,n. Conclude that
(i = 2, ... , m)
is meromorphic in a neighborhood of R + 131.
(c) Show that
as Inl -+ 00.
The dual V' (11') of the space V(lI') is the space of distributions on the circle
11'. Recall that Je'(1I'), the sPi1ce of holomorphic functions on 11', is the inductive
limit
Je'(1I') = lim Je'(Q),
~
over all open subsets Q of C containing 11'. This space is considered with the
corresponding inductive limit topology and is an C'J space. One can verify
that Je'(1I') £ V(lI') and is dense. The elements T of Je"(1I') will be called
analytic functionals on 11' or hyperfunctions on 11'. (An older terminology
is hyperdistributions on 11'.) As was the case for the hyperfunctions on JR, this
generalization of the distributions on 11' will allow us to attribpte to an arbi-
trary holomorphic function in D a boundary value, something that could only
have been done for functions of moderate growth if we would have wanted the
boundary values to lie in V'(lI'). As an application we will give the characteriza-
tion of all the eigenfunctions of the hyperbolic Laplacian (or Laplace-Beltrami
operator) in D as integrals of powers of the Poisson kernel multiplied by
hyperfunctions on 11'. There are just not enough distributions on 11' to solve
this problem.
We start by characterizing the elements of Je'(1I').
-
unique and given by
1
1
fl(z) = -1. -few) ~cnz n,
- d w = '"' Izl :::: 1,
27r1 n"':O
= -. 1
Yl W - Z
-
h(z) 1 -few)
-dw ~cnz n,
= '"' Izl ::: 1,
27r1 ~ W - Z n<O
where YI, Y2 are two positively oriented circles centered at the origin of radii PI,
P2 such that r2 < P2 < 1 < PI < rl·
Proof. (1) The equivalence of (i) and (ii), and the fonnulas for ]1 and h
follow immediately from the properties of the Laurent development of f. The
uniqueness of ]1 and h can be seen as follows: If ]1 + = gl + g2 on 'JI', h
with ]1, g1 E Jff(D), h, g2 E Jffo(C\D), then the function g1 = g2 - ]2 h-
on 'JI' is the restriction of an entire function which vanishes at 00. Hence it is
identically zero by Liouville's theorem.
(2) The equivalence of (i) and (iii) follows from Hadamard's fonnula for the
radii of convergence of the two series
and
Therefore
neZ n"':O
96 1. Boundary Values of Holomorphic Functions and Analytic Functionals
(2k)!
:::: C (1 _ (2)2k+l .
Hence
( 1If(k)112)l
/ k = ((J(2k)!)l/k) =
k! 0 k! 0(1),
""
IIglloo :::: L..J Icn(g)1 = Ico(g)1 + "" Icn(g')1
L..J -In-I-
neZ neZ'
:::: 2~ 11"
o
2
g(O)dOI + L nI2
( ) 1/2 (
neZ'
L:
neZ
Icn(g')1 2
) 1/2
We can now show that the topology of .;E'('ll') can be characterized in different
ways.
1.6. Analytic Functionals on the Unit Circle 97
(iv) The topology of the inductive limit on the index M > 0 of the Banach spaces
(v) The topology of the inductive limit on the index L > 0 of the Banach spaces
on Izl = 1.
Hence, the Fourier series of P is given by
Now let f.J. E C be a fixed value, and keep re icp E D also fixed. Since
P(re icp , b) > 0 for bE 1I', the function w ~ wI-' which takes the value 1 at
w = 1 is well defined on the range of P. By Remark 1.6.6, we have that
b ~ (P(re icp , b))1-' is also in Jf(1I'). We denote pl-'(z, b) := (P(z, b))l-'.
The computation of the explicit Fourier series development of pI-' can be
found in [Er]:
pl-'(re icp , eiB ) = Lan(r)e-incpeinB,
neZ
with
r(f.J. + In!)
a (r)
n
= (1 - r2)l-'rlnl F(ll
r(f.J.)r(lnl + 1) ,.",.,
+ Inl', 1 + Inl'" r 2 )
II.
where
. .z _ r(c) ' " r(a + n)r(b + n) zn
F(a, b, c, ) - r(b)r(a) L...Jo r(c + n) n!
n~
to an analytic functional, i.e., V'('ll') s;;; Jf'('ll'). For instance, if h E LI('ll'), then
1.6.9. Definition. For T E Jf' ('ll'), the Fantappie transform of T is the pair
(ii, T2) defined by
TI(z) := _1_)
(Tb' b-z (Izl < 1),
= -. 11- -
27l'1 Y2
T2(z)f,(z)dz + -. 11--
27l'1 YI
T,(z)h(z)dz,
= LCn(f)c-<n+l)(T),
nel.
which proves the first part of statement (3).
If Izl < 1 is fixed, we have 1/(b - z) = L.n>O znb- n- 1, the series is conver-
gent in Jfo(C\D) and hence in Jf(T). Similarly~if Izl > 1, we have 1/(z - b) =
L.n>O z-nbn- 1 , this series is convergent in Jf(D), hence also in Jf(T).
Therefore, if Izl < 1 we have
Tl(Z) = (T, b ~ z) = L
n~O
zn(T, b-n- 1 ) =L
n~O
cn(T)zn
T2(Z) = / T,
\
_1_) =
z-b
Lz-n(T,b n- 1 ) = Lc-n(T)z-n.
n>O n>O
This shows the proposed expansions for T" T2 are correct and that (T" T2)
belongs to Jf(D) x Jfo(C\D). Moreover, these series have a radius of
convergence ~ 1, hence for any r E [0, 1[ one has
f f-+ -1.
27l'1
1 !I-
Y2
S2 (z) (z) dz + -.
1
27l'1
1 -
YI
Sl (z)h(z) dz
1.6. Analytic Functionals on the Unit Circle 101
- = ( 1) = 11
T j (z) Tb, -b- -2' dw
Sj (w)-- = Sj (z) if Izl < 1,
- = ( z-b1) = -.11
- Z 7r1 YI W - Z
T2(z) Tb, - - dw
S2(W)-- = S2(Z) if Izl > 1,
2m Y2 z-w
these computations are justified using the uniqueness of the Fantappie pair
associated to b t-+ 1/(b - z), Izl < 1 and Izl > 1, respectively, and the Cauchy
representation formula. Hence, ij = Sj, j = 1, 2, which proves both the surjec-
tivity in (1) as well as the relation in (3). The surjectivity in (2) follows from
the relations between the two transforms. 0
1.6.11. Definition. Let T E Je' ('JI') and (Tj , T2) be its Fantappie transform. The
Fantappie decomposition of T is the pair (T\, T2) of elements in Je' ('JI') defined
by the fact that the Fantappie transform of T\ is (T\, 0) and that of T2 is (0, T2).
We say that T\ is the boundary value of the holomorphic function T\ E Je(D)
and that T2 is the boundary value of the holomorphic function T2 E Jeo(C\D).
1.6.12. Definition. Let T E Je' ('JI') and let (T\, T2 ) be its Fantappie transform.
°
For every < r < 1, the function z t-+ T\ (rz) belongs to Je(D) and the function
z t-+ T2(z/r) belongs to Jeo(C\D), and there is a function Tr E Je('JI') whose
Fantappie transform is precisely (T\(rz), T2(z/r)). The functions Tr are called
the r-approximations of T.
Proof. This proposition follows from the fact that Tl (rz) converges to Tl (z)
in .1'f(D) and T2 (z/r) converges to T2 (z) in .1'fo(C\D), plus the topological
isomorphism between .1'f'(1l') and .1'f(D) x .1'fo(C\D) given by the Fantappie
transform. 0
We are going to apply the theory of analytic functionals on 11' to the study
of the Laplace-Beltrami operator of the hyperbolic disk D. There are other
applications, for instance, to harmonic analysis and to signal processing. The
objective is to show that all the eigenfunctions in D of the operator
~ = (1- (x 2+ i»2~,
~ = 82 / 8x 2 + 82 / 8y2, corresponding to the eigenvalue A E C, are the functions
F,Az) = (Tb, PIL(Z, b»),
!
where 4f-L(f-L - 1) = A, Ref-L ~ and T E .1'f'(1l') is arbitrary.
The operator ~ is the Laplace-Beltrami operator in D for the hyperbolic
metric. Recall the hyperbolic metric has arc length
d s2 =
dx 2 + dy2
----:----=--::--::-
(1 - (x 2 + y2»2
and its isometries are precisely the Moebius transformations. The equation
~f - AI = 0
is equivalent to the elliptic equation (r2 = x 2 + y2) :
A
~f - (1 _ r2)2 f = 0,
1.6. Analytic Functionals on the Unit Circle 103
I
f(W,Ao+h)-f(W,AO) _ af(w A)I = laf(W A +fJh)- af(w A)I
h aA ,0 aA' 0 aA' 0
°
for some < fJ < 1. The last expression tends to zero as h -+ 0, uniformly
for W E U by (ii). Hence, the difference quotient (f(w, AO + h) - few, Ao))1 h
tends to (aflaA)(W, AO) in the topology of Je(11'). This proves the lemma. 0
One uses afterward the previous lemma to guarantee that we can differentiate
twice under the bracket ( , ), with respect to x = Re z and y = 1m z. Let Zo =
xo + iyo E D fixed, ro = (x5 + Y5)1/2 < 1. Fix r, p, ro < r < p < 1. We allow
z to vary only in the disk Izl < r. Set U = {w E C: p ~ Iwi ~ lip}. Apply
the lemma to f = pM, f = apM I ax, f = apM I ay, successively with A = x or
A = y. The proposition follows. 0
1.6.17. Proposition. Let A E C, J1. E C, such that 4J1. (J1. - 1) = A, Re J1. :::: ~,and
let n E Z.ll IE C 2(D)\{0} is such that lor any () E lit zED,
l(ei(Jz) = ein(J I(z),
the lollowing two properties are equivalent:
(i) ~I = Af·
(ii) I is proportional to the lunction ifJn defined by
d 2I dI ( 2 Ar2)
r2 dr 2 + r dr = n + (1 _ r2)2 I
for f = fer), 0 ::::: r < 1, i.e., f restricted to the real positive axis. Hence fer)
is a real analytic function of r, I (r) = L:k>O akrk, and replacing into the differ-
ential equation we obtain two systems of equations for the coefficients ak:
n2ao = 0,
(4 - n2)a2 = Aao,
(I)
(II)
One sees immediately that if k does not have the parity of n, then ak = O. Also,
if k < Inl, then ak = O. The ak are uniquely determined once alnl is given. If
1.6. Analytic Functionals on the Unit Circle 105
one changes alnl by a factor, they all change by the same factor. Since CPn was
a solution, it must be that f is proportional to CPn. D
We are finally ready to prove the main theorem, due to Helgason [He]. Let us
point out that the proof in [He] of the preceding lemma is more closely tied to
the underlying semisimple Lie group SU(1, 1), hence intrinsically more general
than ours.
1.6.18. Theorem. Let D be the disk {Izl < 1}, 'Jl' = aD. Let
Li = (1_lzI2)2~,
be the Poisson kernel. Let A, f.J. E C such that 4f.J.(f.J. - 1) = A, Re f.J. ~ ~. The
eigenfunctions of Li with eigenvalue A are exactly the functions
r
fn(z) = _1_ 27r f(eiOz)e-inO de.
21r Jo
It is clear that fn(eiOz) = e fn(z). Furthermore, differentiating under the inte-
inO
gral sign one finds Lifn = Afn. (This is simply a consequence that Li is invariant
under rotations.) Therefore, there is a constant bn such that
fn(z) = bnCPn(z, f.J.).
Taking () = 0 in the above Fourier expansion we obtain that for every zED
and
L Ibnl ICPn(z, f.J.)1 < 00.
nE'L
since the fPn(z, /L) are the Fourier coefficients of PIL(Z, b) (see Proposi-
tion 1.6.10(3». Hence, we have reduced the problem to show that for every
r E [0, l[ we have
To show this we need to use the asymptotic behavior of the functions IfPn (z, /L) I
as Izl ~ 1. In the formula for IfPnl the term that multiplies r 1nl is
We want to see that for fixed r, 0 < r < 1, its presence or not does not
affect the convergence of the series. First, we simplify this expression using the
transformation law [Er]:
-1 ~
- + L.J (n + 1) ... (n + k/ + Pm+l
Ak k
k=O
where
IPm+ll ~ am Iylm+l (n + 1)2(Re IL)-m-3,
am a constant independent of y and n. Chose first m so large that 2(Re J.L) -
m - 3 < 0, hence Pm+l ~ 0 as n ~ 00. This shows that for y fixed (i.e., for
r < 1 fixed)
lim F(/L, 1 - /L, n
n..... oo
+ 1; y) = 1.
Therefore, for n ::: no, we have
The other factor that varies with n in IfPn I is the quotient of r -factors. By
Stirling's formula
(as n ~ 00).
1.6. Analytic Functionals on the Unit Circle 107
Since Re f..L ::: t, we have for some positive constant A and for n ::: n 1 ::: no that
r(f..L+ n ) I A
1
r(f..L)n! ::: (n + 1)1/2'
Thus, for some constant B > °and , fixed in [0, 1[, we have
,Inl
L
Inl~nl
Ibnl ICPn(z, f..L)1 ::: B L
Inl~nl
Ibn I (Inl + 1)1/2'
Since the left-hand side is convergent, we obtain that the power series
for any , E [0, 1[. This concludes the proof of Helgason's theorem. 0
EXERCISES 1.6.
1. Prove in detail that V' (T) can be characterized as the space of Fourier series of
slow growth.
2. Verify that the topology of Jt'(T) is, in fact, C:!='. Show that it is a DFS space.
3. (a) Verify that, for fixed zED, the Fantappie decomposition of I: b ~ 1/(b - z),
is (0, h), with h(w) = 1/(w - z).
(b) Verify that if Z E C\D, then the Fantappie decomposition of the function I: b ~
1/(b - z) is (h. 0) with i,. (w) = 1/(w - z). Is there any contradiction between the two
answers? .
4. Show that the Poisson kernel b ~ P(re i8 , b) is the restriction to T of the function
(1- r2) z
z ~ - - - . -----,:::-
Z - re i8 1 - zre- i8
and verify that the Fantappie decomposition of P given in Example 1.6.7 is correct.
5. Let h E Jt'(T)\{O}. Characterize the solutions R E Jt'/(T) of the equation hR = O.
(Hint: See Exercise 1.3.16.)
6. A weight w in Z is a function w: Z .... [1, oo[ such that
w(m + n) :::: w(n)w(m) for any m,n E Z.
Let us denote Jt'... (T) = (f E C(T): III... := Enez Icn(f)lw(n) < oo}.
(a) Show that Jt'... (T) is a Banach algebra for the ordinary product of functions.
108 1. Boundary Values of Holomorphic Functions and Analytic Functionals
(b) Define
(T. f) = Lcn(f)C_<n+l)(T).
neZ
(T, f) = r-->I-
lim -21 . rf(n[C(T)(~/r) - C(T)(rmd~.
10 JT
7. Let G E L1(D,dm) and let g E C(1f) be such that for every 13 E C""(C) one has
ID G(Bf3/B'i) d'i /\ dz = IT f3g dz. Show that G E Jr(D) n C(jj) and GI1f = g. (Hint:
Consider the Fantappie transform (G 1• G 2 ) of g and. choosing conveniently 13. show that
G 2 = O. Use the fact that the Cauchy transform of g inside D coincides with the Poisson
transform to conclude that G 2 = G and it is continuous to the boundary). Compare with
Exercise 1.1.18.
CHAPTER 2
In the first chapter, we have seen how the Leitmotiv of the boundary values
of holomorphic functions lead us naturally to introduce several transforms, in
particular, the Fourier-Borel and Fourier transforms, and found out that many
questions can be posed in equivalent terms in the algebras of entire functions
with growth conditions, Exp(Q) and F(£'(R)), specially problems relating to
convolution equations. In the case of distributions, this relation will be come
more evident in Chapter 6. The aim of this chapter is to study a more general
class of algebras, the Hormander algebras, Ap(Q). We shall see that the ideal
theory of these algebras is intimately related to the study of interpolation vari-
eties. In the previous volume [BG, Chapter 3], we have shown that to be the
case for the algebras of holomorphic functions Jt"(Q), and we found out that
one could study interpolation questions with the help of the inhomogeneous
Cauchy-Riemann equation. The same will be the case here. This time, though,
we shall be obliged to consider the problem of solving the Cauchy-Riemann
equation with growth constraints.
The algebras Ap were used by Hormander in [H03], [H04], who proved they
were the correct context in which to study the Cauchy-Riemann equations in
en. Throughout this chapter we use this work, as well as that of [KTl], [KT2],
[Gul], [SkI], [Sk2], [BTl], and [BT3].
e
2.1.1. Definition. Let Q be an open subset of and let p be a subharmonic
function in Q. Let us denote by Ap(Q) the space of all functions f E Jt"(Q) for
which there exist constants A :::: 0, B :::: 0, such that for every z E Q one has
the estimate
If(z) I ::: Aexp(Bp(z)).
109
110 2. Interpolation and the Algebras A p
It is clear that Ap(Q) and Wp(Q) are algebras under the ordinary product of
functions.
Since
j'(z) = _1 1
2rri I~-zl=i
f(l;) dl;,
(l; - z)2
then
Therefore f E Wp(C).
2.1. The Algebras Ap 111
Recall that A(I/I, z, r) denotes the area average of III over the disk B(z, r).
Then
This is a Banach space for the norm II· liB. Choosing a sequence (Bn)n~J,
0< B n, strictly increasing to 00, we can introduce an C'J topology on Ap(Q)
as the inductive limit
n->oo
2.1.2. Proposition (A Priori Inequality for ojoi). Let n be an open subset ofe
and P a subharmonicfunction in n. For every f E V(n) we have the inequality
: : inrIfI202p~ePldm.
oz OZ
If now P is a C 2 function which is subharmonic in n, set PI (z) = p(z) +
210g(1 + IzI2). Then
o2PI o2p 2 2
--
OZ oi
= --
oz oi
+
(1 + Iz12)2
> ----;:--::-
- (1 + Iz12)2
and
eP1(Z) = eP(Z) (1 + IZI2)2.
Hence, the previous inequality becomes,
In :~
1 12 (1 + IzI2)2e P(z) dm(z) ::: 2 In Ifl2e Pdm
in Qp = {z E Q: d(z, QC) > p}, but for 0 < p sufficiently small we also have
supp f ~ Q p • Therefore,
Similarly, the dual H{ of HI can be identified to L 2(Q, e-p(z) /[(1 + Iz12)2] dm)
1
(by the same pairing)
(g,f)j := fgdm.
T(f) = - !~.
This operator has a well-defined adjoint IT (since dom(T) is dense), IT: H{ --+
H~. Recall that the domain of IT is
2 12M
IIlulil l :s zlllvilio = 2'
If f" ... , fm E Ap(Q) we set IfI 2 := If1l2 + ... + Ifml 2 and for c > 0,
C > 0, we introduce the open sets
S(lfl, c, C) := {z E Q: If(z)1 < ce-Cp(z)}
°
(1) A E Ap(C);
(2) strictly positive constants c', C', A', B', < c' < c, C' > C;
(3) holomorphic functions a" ... , am in S(lfl, c', C'); and
such that for every z E S(lfl, c', C'):
(a) A(Z) = );.(z) + L aj (z)jj (z);
'~j~j
and
(b)
Remark. If (zkh~, is the family of common zeros to all the jj, j = 1, ... , m,
and if their multiplicities (as a common zero) are denoted by mb one concludes
from (a) thatA (j) (zd = );. (j) (zd for °: :
j ::: mk - I and every k ::::: 1. Therefore,
given a sequence (ak,j h,j of complex numbers the problem of finding A E Ap (C)
such that
).
., k::::: 1,
can be reduced to the same problem for a function);. in S(lfl, c, C). This
explains the reason of the terminology "semilocal" applied to Theorem 2.1.4.
Proof of Theorem 2.1.4. Let us first show that there are
A, > 0, and B, > 0, such that, for z E S(lfl,
< c, < c, C, > C,
c"
C,), the distance from z to
°
as(lfl, c, C) is at least exp(-A,p(z) - B,).1t is enough to find those constants
for those z E C such that d(z, as(lfl, c, C» ::: 1 (if this set is nonempty).
Let z' E as(lfl, c, C) be such that Iz - z'l = dist(z, as(lfl, c, C» ::: 1. For
t E [0, 1], let ((J(t) = If(z + t(z' - z))l2. There are A2 > 0, B2 > such that °
1({J'(t) I ::: A21z - z'l exp(B2P(Z». (Here we use not only that jj E Ap(C) but
that p is a weight.) Since we know exactly the value of If I on as(lfl, c, C),
we have
c 2 exp(-2Cp(z'» = If(z')1 2 = ({J(I) ::: ((J(O) + tE[O,']
max 1({J'(t) I
when Z E S(lfl, eJ, Cd. Since p(z') :::: Cop(z) + Do, if C I ~ 2CCo and ef <
~e2e-2CDo then we get
e2
Iz - z'l ~ 2A2 e- 2CDo exp( -(2CCo + B2)p(Z»,
2.1.5. Lemma. There is a function X of class COO in C, 0 :::: X :::: 1, such that:
(a) supp X ~ S(lfl, e, C);
(b) X == 1 on S(lfl, ej, C 1);
(c) there are two positive constants A3, B3 such that
(z E C).
Proof of Lemma 2.1.5. The fact that for z E S(lfl, el, Cd we have
the inequality d(z, oS(lfl, e, C» ~ exp(-AIP(z) - B 1), allows us to apply
Whitney's partition of the unity lemma (see [BG, §1.3.5]). 0
Vj
. i 0
.= -lfl 2 oi (X A),
- j = 1, ... ,m.
These functions are also Coo and there are constants A4, B4 > 0 such that
IVj(z)1 :::: A 4 e B4P (z).
Let us choose now 0<82 < 81, C2 > CI, E > 0, F > 0, such that if z E
S(I!I, 82, C2), then dist(z, SO!I, 8" Cln > exp( -Ep(z) - F). This is possible
by the reasoning at the very beginning of the proof. We also know we can find
K > 0, M > 0, such that for every j:
1
::: - exp«KCo + 2E)p(z) + K Do + 2F)
7r
x ~ IUj(~)12e-Kp(n dm(~)
} B(z,exp(-Ep(z)-F»
M
::: _eKDo+2F exp«KCo + 2E)p(z».
7r
This proves all the estimates required in the theorem. o
EXERCISES 2.1.
1. The purpose of this exercise is to give two slightly different proofs that if BulBi = v
in the sense of distributions and v E Coo(n), then u E Coo(n), as stated in Theorem 2.1.3.
(a) [BG, Theorem 3.2.1] states that BIBi: Coo(n) --+ Coo(n) is surjective, and [BG,
Proposition 3.6.5] states that if WE V'(n) and BwlBi = 0, then W E .;f(n). Use these
two statements to conclude that the function u E HI found in the proof of Theorem 2.1.3
belongs to Coo(Q) if v E Coo(Q).
(b) (Direct proof.) Let E = 1/7rz and recall that (BIBi)E = 8 in the sense of distri-
butions. Let now u E V'(n), v E Coo(n), and BulBi = v. Let n' be an arbitrary open
subset of n, n' cc n, and choose rp E V(n) so that rp == I on a neighborhood of n'.
(i) Show that (BIBi)(rpu) = v + w, W E £'(n), supp(w) ~ (n')c.
(ii) Show that if 1{! E V(n), 1{! == 1 on a neighborhood of supp(rp), then
B
Bi (rpu) = 1{!v + WI, WI E £'(n),
2. The aim of this exercise is to show that Theorem 2.1.3 can be used to prove that
alai:
°
COO(Q) ~ COO(Q) is surjective for any open set Q in C.
(i) Let (Xn)n~l be a strictly increasing sequence of numbers Xn > such that
limn .... ooxn = 00, and (Vn)n~l' another sequence with Vn ~ 0.
Show that there exists a strictly increasing convex function rp such that rp(xn) ~ Vn
for every n ~ 1.
(ii) Let (Kn)n~l be an exhaustion of a proper open set Q ~ C, by compact
subsets, (J.Ln)n~1 a sequence of real numbers, J.Ln > 0, and let d n = d(Kn, QC). Show
there is a convex, strictly increasing function rp: [0, oo[ ~ [O,oo[ such that p(z) :=
rp( -log d(z, QC» satisfies the inequalities
p(z) ~ 10gJ.Ln,
=
for all z E 8(0, rn)\B(O, rn-l), n ~ 1. (Define ro -1.)
(iv) Let f E £(Q), show that there is a subhannonic function p in Q such that
:i:
[aG, §3.2.1], that
COO(Q) ~ COO(Q) is surjective.
3. Let P be a subhannonic function in an open set Q "I C satisfying (a) log (1 + Iz12) =
°
O(p(z», z E Q, and (b) there exist constants C lo C2 , C3, C4 > ~uch that if z E Q, ~ E
C, and Iz - ~I ::::: exp(-C1P(z) - C2 ), then ~ E Q and p(~) ::::: C3P(Z) + C4 • Show that
Wp(Q) n .;t"'(Q) = Ap(Q).
j E J}. The Weierstrass theorem [BG, §3.3.1] asserts that for every multiplicity
variety V, there is 1 E Jt'(iC) such that V = V (f).
Given a multiplicity variety V, we also defined the space A(V) of holo-
morphic functions on V, i.e., doubly indexed sequences of complex numbers
(ak,/h,/,O :::: Z < mk. The reason for this notation is that there is a natural restric-
tion map p: Jt'(iC) ~ A(V) given by
g(I)(Zk))
p(g)= ( - - .
Z! k,/
Given a sequence a E A(V) we say that the entire function g interpolates it,
if p(g) = a. The interpolation theorem [BG, §3.4.1] states that p is always
surjective. The proof uses in an essential manner that variety V satisfies V =
V(f) for some 1 E Jt'(iC)\{0}. In·this section we discuss the corresponding
interpolation problem for the space Ap(iC). We shall assume throughout that the
multiplicity variety V satisfies the nontriviality condition
V 5; V(f)
for some 1 E Ap(iC)\{O}.
We follow [BT 1] and we refer the reader to the references therein for other
approaches to the interpolation problem.
We start first by defining a space of holomorphic functions with growth
conditions on a multiplicity variety. This space must be a good candidate to
be the image of Ap(iC) by the restriction map p: Jt'(iC) ~ A(V). It turns out
that there are several natural candidates. The first one is given in the following
definition:
2.2.2. Proposition. For any 1 E Ap(iC) there are constants A ::: 0, B ::: 0, such
that
/(1) (z) I
~ 1 -Z-!- :::: A exp(Bp(z)) (z E iC).
Proof. Assume that I/(t)1 :::: A1eB1P(t) for every t E C. Let z E C be fixed. For
any t E B(z, 2) there is ~ E B(z, 1) such that It - ~I :::: 1, hence
p(t) :::: Cop(~) + Do :::: C5P(z) + (Co + 1)Do.
Taking these inequalities into account, the Cauchy inequalities
1(I)(z) I 1
I-Z-!- :::: 21 It~~21/(t)1
120 2. IntetpOlation and the Algebras A p
The same argument as that in Proposition 2.2.2 leads to the existence of A > 0,
B > 0, such that
f(l)(z)
~ I -I-!-
Irl::: Aexp(Bp(z,r)), for any z E C and r > O.
II I(ak,l) II I := sup L
k2:1 O~l<mk
laul.
gjO(Zk)
-I-!- j~&,ak,l'
Therefore,
and
p(q;d = Aa k •
glQ = L gj(/jIQ).
l:"Oj:"Om
2.2.8. Definition. Let /J, ... , 1m E Ap(C). We say they are jointly invertible
if
/(/J, ... , 1m) = Iloc(iI,···, 1m).
2.2.9. Lemma. Assume that Ii, ... , 1m are jointly invertible. Given A > 0, there
are B > 0 and C > 0 such that ifh E 11oc(fI, ... , 1m) and
h = L gj/j
I~j~m
and
Igj(z)1 :::: Cexp(Bp(z)), lor 1:::: j :::: m.
Proof. We give a proof similar to that of Lemma 2.2.6. Consider
S = {h E l(fl,"" 1m): Ih(z)1 :::: exp(Ap(z))},
and
as k -+ 00.
If V = V(ft, ... , 1m), (Zj, mj) E V, then hnk vanishes at Zj with multiplicity
~ mj. It follows the same is true for h. Hence,
As pointed out above, the ideal I (flo . .. ,1m) = 11oc(f" ... ,1m) is a closed
subset of Ap(C).
For n E N*, denote
Sn := {(g" ... , gm) E (Ap(c))m: Igj(z)1 :::: ne np (z)(1 :::: j :::: m)} n s.
We claim that Sn is a closed subset of S. In fact, let (h/)/?!I ~ Sn and h E S
be such that d(h/, h) -+ 0 as 1-+ 00. Each h/ = 8(gl,/, ... , gm,/), Igj,/(z) I ::::
nenp(z). We can therefore extract subsequences (gj,hh?!l and functions gj E
Jt"(C) such that
124 2. Interpolation and the Algebras A p
Hence
for all Z E C,
and
L gj,/k/j ~ L gj/j in Jt"'(C).
I~j~m I~j~m
and
8(gl, ... , gm) = Ah.
Letting B = no and C = nolA we conclude the proof of the lemma. 0
Remark. Note that if hE I\oc(/I, ... , fm) and h satisfies Ihl .::: De Ap , then we
can conclude from Lemma 2.2.9 that h = L:I~j~m gj/j and Igjl .::: DCe BP •
2.2.10. Theorem. Let /I, ... , fm E Ap(C) and V:= VUh ... , fm) be the
multiplicity variety of their common zeros. Assume there are e > 0, C > 0, such
that for every (Zb mk) E V one has
IJ.'mk\Zk) I
(*) L J , :::eexp(-Cp(Zk».
I~j~m mk·
Then V is an interpolation variety for Ap(C).
Conversely, if V is an interpolation variety for Ap(C) and if the functions
/I, ... ,fm are jointly invertible, then there are e > 0, C > 0, such that the
inequality (*) holds for every (Zb mk) E V.
Proof. Let us start proving that in case I\oc(/I, ... , fm) = /(/1, ... , fm), the
inequality (*) is a necessary condition for V to be an interpolation variety.
°
If V is an interpolation variety, from Lemma 2.2.6 we conclude there are
constants CI, C2 > and functions hk E Ap(C) such that
Ihk(z)1 .::: CI exp(C2P(Z»
and
The functions
2.2. Interpolation with Growth Conditions 125
vanish at every Zj with multiplicity ~ mj. Hence ({Jk E Iloc(flo ... , /m). More-
over, for some A > 0 we have
+ IZkl)(1 + Izl)lhk(z)1 ::: CI(1 + IZkl) exp(Ap(z» ,
l({Jk(Z) 1 ::: (1
where we have used that log(1 + Izl) = O(p(z». Hence, we can apply
the
previous lemma and the corresponding remark to conclude that there are
constants B, C > 0, and entire functions gl,b ... ,gm,k such that
and
(Z E C),
with C2 = CCI.
The choice of hk ensures that the Taylor expansion of ({Jk about Zk is of
the form
using once more that log(l + IZkl) = O(p(Zk». This inequality is clearly equiv-
alent to (*).
We shall now prove that the inequalities (*) imply that V is an interpolation
variety for Ap(C). We start by reminding the reader that a corollary of the
Minimum Modulus Theorem is the following:
2.2.11. Lemma. Let 1/1 be a/unction holomorphic in B(O, 2eR), assume that
max 11/I(~)1::: M
1~1~2eR
and
11/1(0)1 ~ 8 > 0,
then there is a value r, Rj4 ::: r ::: Rj2 such that
Proof of Lemma 2.2.11. Let us recall the Minimum Modulus Theorem as stated
in ([BG, Theorem 4.5.14] or [Lev, Chapter I, Theorem 11]): If f is a hole-
morphic function in a neighborhood of B(O, 2eR), /(0) = I, and 1/ E ]0, 3ej2[,
126 2. Interpolation and the Algebras A p
then there is a finite family of exceptional closed disks, the sum of whose radii
does not exceed 4TJR, such that for z E B(O, R) and outside the exceptional
disks, the following inequality holds
In other words,
min log 11/I(z) I > 9 log 0 - Slog M(1/I, 2eR),
Izl=r
Note that the same argument shows that one can take r E ]3R/4, R[.
We also need the following lemma:
Let us now return to the proof of Theorem 2.2.10. Let (Zk, mk) E V and define
dk := inf{1, inflzi - zkl}·
i#
If dk < 1, choose some Zi such that dk = IZi - zkl. Since we are assuming that
(*) holds, there must be an index j, 1 :s j :s m, such that
l/mi)(Zi)1 8
J ~-exp(-Cp(Zi»'
mil m
We shall apply now Lemma 2.2.12 to the function g(~) := /j(Zi + ~). Then
max Ig(~)1 :s max max 1/j(Zi
11;1:::1 I:::j:::m 11;1:::2e
+ ~)I :s M := Mo exp(NoP(zi»
for some constants Mo, No > 0 that depend only on the bounds for the generators
/j. We have now lal = dk and q = mk in the statement of the lemma, hence
m 8
- exp( -Cp(Zi».
Mdk k ~
m
Since P(Zi) :s Cop(zd + Do, we obtain
(tt) d';k ~ 81 exp(-AIP(Zk»
for some Al > 0,0 < 81 :s 1. Clearly the same inequality holds if dk = 1.
In order to apply the semilocal interpolation Theorem 2.1.4 we need to show
that we can chOose 80 > 0 and Ao > 0 such that, for every k ~ 1, the connected
component Uk of S(lfl, 80, Ao) that contains Zk satisfies Uk ~ B(zk, dk/2).
Let us choose j so that
(#)
Translating this inequality into an inequality for jj, we have that there are
constants e3 > 0 and A3 > 0 so that
Ijj(Zk + dk~)1 ::: e34-mk exp(-A 3P(zd),
I/rk)(Zk)1
mk.,
= _l_!
270. IZ-Zkl=e (Z ) m +1
- Zk k
jj(z) dz
'
so that
In other words,
and i == 0 on S(I/I, eo, AO)\(Uk:::1 Uk). Since (ak.!h,1 E Ap(V), there are
constants A, B > 0 such that
L lak.!l:s: AeBp(Zk) (k ::: 1).
O:::;i<mk
Hence
Using condition (ii) from Definition 2.2.13, we can replace this inequality by
Ig(z)1 ~ E exp(F(p(z)),
for all z E [ja, for some E, F > 0 independent of the component Ua. Therefore,
this inequality holds in SOfl, B, A). On the other hand, for z ¢ S(lfl, B, A),
CeDp(z) C
Ig(z)1 ~ If(z)1 ~ -; exp«A + D)p(z)).
It is now clear that g E Ap(iC) and I (f) = I'oc(f). D
2.2.15. Proposition. Let p be a radial weight that satisfies the doubling condi-
tion p(2z) = O(p(z)), then every f E Ap(iC)\{O} is slowly decreasing.
Proof. It is enough to show that there exist a point Zo and two constants B, A > 0
such that for all r » 1 one can find r" r2 satisfying
r
2 ~ r, < r < r2 ~ 2r,
inf{lf(z)l: Iz - zol = r, or Iz - zol = r2} ~ Bexp(-Ap(r)).
Since, in this case, we can find a strictly increasing sequence of radii Rn ,
Rn ~ 00, Rn+l ~ 4Rn, such that If(zo + ~)I ~ exp(-Ap(Rn)) if I~I = Rn.
130 2. Interpolation and the Algebras Ap
The doubling condition imposed on the weight p implies that if z and z'
satisfy Rn ::::: Izo + zl ::::: Rn+t. Rn ::::: Izo + z'l ::::: Rn+ l , then p(z) ::::: C~ + C;p(z')
for some convenient positive constants C~, C;. Thus, the components of the set
S(I!I, e, A) satisfy the requirements of Definition 2.2.13.
In order to conclude the proof of the proposition, we choose Izol ::::: 1 such
that !(zo) =f. O. We can now twice apply Lemma 2.2.11 to the function 1/r(l;) =
!(zo + l;) to obtain rl, r2. D
Ip(z)1 = O(lzn
for some K > 0, while the inequality p(z) ::::: Cop(l;) + Do whenever Iz -l; I ::::: 1,
only implies the growth condition
Ip(z)1 ::::: O(e A1z1 )
for some A > O. In fact, for a weight like p(z) = e 1zl one does not automatically
obtain that every! E Ap(C)\{O} is slowly decreasing, but only that! is slowly
decreasing in Aq(C) with a weight q slightly bigger than p (see [Heyl]).
The advantage for the interpolation problems of knowing that a function
! E Ap(C) is invertible is that then V = V(f) is interpolating if and only if
there are e > 0 and C > 0 such that
1!(mk)(Zk)1 > -Cp(Zk)
ee
mk! -
holds for all (Zb mk) E V. So the interpolation problem has a clear-cut answer
in this case. Whence, the interest of Proposition 2.2.15.
It is natural to ask whether condition (*) of Theorem 2.2.10 always holds
when V = V(f) is interpolating, even if ! is no invertible in Ap(C). The
answer is negative as shown by the following example of Squires ([Sq2], see
also [Hey2]). Consider the weight p(z) = 11m zl + 10g(1 + Iz12) and the corre-
sponding space Ap(C), which coincides with the space F(£'(ffi.)) of Fourier
transforms of distributions of compact support. If ep E 1)(ffi.), then there is a
constant A > 0 such that for every n :::: 0
Cn
IFep(z) I < (z E C),
- (1 + Izl)n exp(Allmzl)
for some Cn > O. Clearly Fep E Ap(C), but it cannot be slowly decreasing. (It
turns out that in this space, slowly decreasing and invertible are two equivalent
conditions, hence no such function is invertible.) Since the derivatives of Fep
also satisfy similar estimates, condition (*) of Theorem 2.2.10 cannot hold for
Fep. Nevertheless, Squires has shown that one can find ep E 1)(ffi.) such that
sin]l' z
Fep(z) = 00 ( z2) .
fI 1--
n=1 4n
2.2. Interpolation with Growth Conditions 131
Hence V(.rq» ~ V(sinnz) (all the zeros are simple). On the other hand,
V(sinnz) = {k E Z, mk == I} is interpolating for Ap(C), namely
°
slowly decreasing is equivalent to the following:
There is A > such that for every x E ~ the function I satisfies
max{l/(x + t)l: t E ~, It I :::: A 10g(2 + Ix!)} ~ (A + Ixl)-A.
We shall show in the next chapter that every exponential polynomial
N
I(z) = LCj(z)eiajZ,
j=]
satisfies this type of lower bound condition, and hence is slowly decreasing.
More generally, the ideas of this chapter are also well adapted to the Beurling
weights,
p(z) = Ilmzl +w(lz!),
1 D
00 wet)
--2
1+t
dt < 00.
These functions appear when we consider spaces of Coo functions where one
introduces conditions on the growth of the derivatives (see [Bj], [BD3], [Chou)).
°
To conclude this section, let us rephrase the interpolating procedure used
in Theorem 2.2.11. We used condition (*) to find 8, A > such that for any
a E Ap(V), there is a holomorphic function). on S(I/I, 8, A) satisfying the
correct growth conditions and p().) = a. It is clear from the proof that the map
a 1-+ ). is a linear continuous map into Ap(S(I/I, 8, A)). It is therefore natural
to ask whether the procedure a 1-+ )... of Theorem 2.2.11 can be also made linear
and continuous, in other words, whether there exists a right inverse E to p. Or
what amounts to the same thing, does it exist as an extension operator,
po E = id,
where E is linear and continuous, under the assumption that V is an interpolation
variety? One way to do this is to define)... using an interpolation series, typically
132 2. Interpolation and the Algebras A p
for a convenient sequence of integers i-Lk ::: 0 (see [Ge], [BrK] , [Lev]). It turns
out that this is not possible for every weight p. In a seminal paper, B.A. Taylor
[Ta2] considered the problems of finding a right inverse for p and for the
Cauchy-Riemann a-operator in the correct spaces. He found that the situation
is very different for p(z) = Izla, a > 0, and p(z) = I Imzl + 10g(1 + IzI2). In
a
the first case, there is a right inverse for and, consequently, linear continuous
extension operators E: Ap(V) -+ Ap(C) exist when V is interpolating. In the
second case, the extension operator exists if and only if V satisfies the additional
condition of lying in a logarithmic strip about the real axis, i.e., there is A > 0
for all Zk E V.
EXERCISES 2.2.
Throughout these exercises we denote by V = (Zk. mkk~.I a multiplicity variety, which
we assume is infinite and satisfies V S; V(f), for some IE Ap(IC), unless explicitly
stated otherwise.
1. (a) Show that the failure ofthe condition IZk I -+ 00 in the definition of a multiplicity
variety is equivalent to saying there is no entire function 1'# 0 such that V = V(f).
Prove that in such a case the map p: Jt'(1C) -+ A(V) is injective but cannot be surjective.
(b) Assume now that IZkl -+ 00, but that there is no IE Ap(IC)\{O} such that V S;
V(f). Then:
(i) show that p: Ap(lC) -+ A(V) is injective;
(ii) let ej E A(V) be defined by eL = ~j,k~I,O' prove that no ej E p(Ap(IC».
2. (a) Show that if V is an interpolation variety for Ap(1C) and W S; V, then W is
also an interpolation variety.
(b) Prove that W = N (with multiplicity 1) is an interpolation variety for the space
Exp(lC).
*(c) Can you find a single function IE Exp(lC) such that V(f) = N? Can you find
two functions II, fz E Exp(1C) such that V (fl, fz) = N? (Hint: Recall from [BG, Exer-
cise 2.5.20] that if {Zdk~1 denotes the zeros of I E Exp(IC), repeated according to their
multiplicities, then the sums LO:::lzkl<r Ijzk are bounded, independently of r > 0.)
2.2. Interpolation with Growth Conditions 133
3. Let p(z) = IzI P, P > 0, compute explicitly the function p(z, r) defined just before
Definition 2.2.3.
4. Show that if the multiplicities mk :::; M < 00 for all k, then Ap.",,(V) = Ap(V).
*5. For p(z) = Izl, construct a multiplicity variety V such that Ap.",,(V) =J: Ap(V).
(Hint: Use a variety for which mk ::::: IZkl.)
*6. Let p(z) = IzIP. Use Exercise 2.2.3 to show that a sequence (ak,/) E Ap,,,,,(V) if
and only if for some A > 0
exp(Alzkn
(*) lak,II :::; A (1 + IZk!)1 (0:::; I < mko all k).
Show also that in this case, (*) is equivalent to the existence of B > 0 such that
Bexp(BlzkI P)
lak,II :::; (l!)I/p .
*7. Let p(z) = IImzl + log(l + IzI2), then (ak,/) E Ap,,,,,(V) if and only if for
some A > 0
la I< AI+I exp(Ap(Zk» .
k,1 - I!
Thus Ap,,,,,(V) = Ap(V) if and only if
mk =0 ( P(Zk) ) as k --+ 00.
logp(zd
What restriction on the multiplicities does this condition imply when Zk is on the real
axis? (Hint: Compute p(z, r) for this weight function.)
8. Simplify the proofs of Lemmas 2.2.6 and 2.2.9 using the open mapping theorem,
i,e., Proposition 1.5.1, instead of the Baire category argument given in the text.
9. Let V be an interpolation variety for Ap(iC),
d k := min{l, min Iz)· - zkl}.
j#
Show that there are positive constants 8, A, B, C > 0 such that for all k
and
mk :::; Bp(zd + C.
(Hint: Use Lemma 2.2.6 and reconsider the proof of Theorem 2.2.10 without assuming
that V is of the form V(fl, ... , 1m), with /J, ... , 1m jointly invertible.)
10. (a) Use the previous exercise to show that a necessary condition for a sequence
(Zk k:: I of distinct points converging to
00 to be an interpolation variety for the space
Exp(iC) is the existence of a constant A > 0 such that
inf{lz - zkleA(lzkl+IZjl)} > O.
j-Ik )
134 2. Interpolation and the Algebras A p
(b) Let (zkh~1 be a sequence of distinct real numbers, IZkl --+ 00. A necessary condition
for it to be an interpolation variety for the space :F(£'(R» is the existence of N > 0
such that
11. Let V = V (fl, ... ,fm), f]' ... , fm jointly invertible in Ap (<C). Show that V is
an interpolation variety if and only if there are e, A > 0 such that:
(i) each Zk E V belongs to a component of S(lfl, e, A) of diameter at most 1; and
(ii) no two distinct points of V lie in the same component of S(lfl, e, A).
12. Let p(z) = Izl, V = {k E Z, mk = I; ±(k + e(k», mk = I}, with e(k) E R, 0 <
le(k)1 < !.
(a) Show that there exists f E Ap(<C) such that V =
V(f). Find a condition on 8(k)
so that there cannot exist /) > 0, such that
Yk E Z.
*13. Find an even entire function f of exponential type such that it vanishes at all the
points x + iy E Z + iZ of the form Ix - nJI :s nj, Iyl :s nj, for an increasing sequence of
integers n j, such that nj --+ 00 sufficiently fast. Show that V (f) cannot be an interpolation
variety for Exp(<C), even though the conditions on db mk from Exercise 2.2.9 are satisfied.
(Hint: Check whether the value of 1f'(nJ)1 is too small. This depends only on the behavior
of the finite product IT(l - (Z/Zk)2), where Zk lies in the block of zeros surrounding nJ.>
Find necessary and sufficient conditions for V (f) to be an interpolating variety in Ap (<C)
in terms of the behavior of rp for large n. Similarly, for p(z) =
IImzl + log(l + IzI2).
(Hint: Use the elementary fact that I sin;rrxi 2: Cd(x, Z) for some C > 0.)
15. Let T E £'(R), f(s) := (Tx, e-ix~) be the Fourier transform:FT of T. Let P(D) =
L:J=o aj Dj and be a linear differential operator with constant coefficients.
(a) Show that if there is S E £'(R) such that P(D)S = T, then f(s)/P(is) is an
entire function.
=
(b) Let q(z) amz m + ... + ao be a polynomial of degree m, h a holomorphic func-
tion in 8(0, 1). Show that
(Hint: Use that when Izl = 1, Iq(z)1 = Ir(z)l, r(z) = am + am-lZ + ... + aoz m.)
s
(c) Assume that g(s) := f(s)/ P(iO is an entire function. For fixed consider q(z) :=
P(i(z + s» to show that
lamllg(s)1 :s max If(s + z)l·
Izl~l
2.2. Interpolation with Growth Conditions 135
Conclude that there is S E e'(lR) such that :FS = g, T = P(D)S, and cv(supp S) =
cv(supp T). (This proves that any polynomial is invertible in Ap(C), for p(z) =
IImzl +
10g(1 + IzI2). In fact, it proves the same result holds for any weight p.)
(d) Show that for some constant C > 0, IP(z)1 ~ Cd(z, v(p))m. Use this to prove
that P is slowly decreasing in any Ap(C).
16. Prove that I sinzl ~ Cd(z, nZ)ellrnzl for some C > O. Use this to show that sinz
is slowly decreasing in any Ap(C) such that p(z) ~ IImzl + 10g(1 + IzI2).
17. For B > 0 let us define
Yk,1 := Yk,I(B) := inf{r- I exp(Bp(Zb r)): r > OJ,
Show that a = (ak,l) E Ap,oo(V) implies that for some A > 0, B > 0,
*18. Let fI, ... , 1m be jointly invertible and V = V(fl, ... , 1m) a weak interpolation
variety. Follow the notation of the previous exercise. Show that for every B > 0 there
are constants e, C > 0 such that
(Hint: Consider functions hk like those in Lemma 2.2.6 with condition (iv) replaced
by (iv)' hi
mk -I)(Zk) = (mk - I)! nCB). Follow the proof of the necessity part in
Theorem 2.2.10.)
for Iz - zkl < ~Rb where 1/, C > 0 are suitable constants. Follow the rest of the proof
of Theorem 2.2.10.)
136 2. Interpolation and the Algebras A p
*20. Let p(z) = IzIP, P > 0, and follow the notation of the preceding exercise.
(a) Show that mk = O(P(Zk)), Rk = O(p(zd), and condition (ii) always holds for this
particular weight.
t
(b) Show that condition (iii) is equivalent to
22. Show that if for some A, B > 0, lakl ~ Ae B1kl for all k E Z, then the function
~ sin1l"z (Z)lkl
I(z) := L.." ak(-I)k -
kE"L 1I"(z - k) k
(where (z/ k)lkl == 1 for k = 0), is an entire function of exponential type such that I(k) =
ak for every k E Z.
23. Let V = {Zdk2:I' IZkl -* 00, mk == 1, and let A(V) be the set of all sequences
{adk2:I' The object of this problem is to prove there are no linear continuous exten-
sion operators E: A(V) -* Jr(C), poE = id ([RudlJ). The topology on V is that of
convergence for each index k.
(a) Show that the topology we described is the compact open topology on the space
of all functions V -* <C.
(b) Let ek E A(V), ek(Zj) = 8jk . Assume E exists, set gk := E(ek). Show there is
a E C such that gk(a) # 0 for all k.
(c) Let k edgk(a) E A(V). Show !k -* 0 in A(V).
(d) Use the homogeneity of E to verify Elk(a) = 1 for all k. Conclude that E cannot
be continuous.
where
(i) TI is a continuous linear operator and we denote its adjoint Tt: H2 -+ HI;
and
(ii) T2 is a closed linear operator with dense domain, denoted dom(T2).
2.3.1. Proposition. G 2 = T, (ker T2) if and only if there is a constant C > 0 such
that
II Tt(X2) + T2*(X3)1i, ~ CIIX2112
for every X2 E G 2 and every X3 E dom(T2*).
In this case,jor every X2 E G 2 there is x, E kef T2 such that T, (x,) = X2 and
IIx,lI, ::::: (1/C)lI x2112.
Proof. Let G, = kerT2, it is a closed subspace of H,. Denote still by T, the
operator T,IG,: G I -+ H2 • The following lemma shows that T,(G I ) = G2 if
and only if there is C > 0 such that:
To prove the converse, assume the existence of a constant C > 0 such that
for every y E £2, we have lIyll2 :s CIIT*(y)llt. Clearly T* is injective, hence
for every y E £2 the following map is a well-defined linear map on Im(T*):
u: T*(z) f-+ (zlyh,
which verifies
for all z E £2. Therefore, T(x) = y. This concludes the proof of the lemma. 0
Let us now return to the proof of the proposition. G I is also a Hilbert space,
hence G I is isomorphic (including the identity of norms) to its dual G~. On the
other hand, G~ ~ Ht/ Gt, also with equality of norms. Since G I = ker T2 and,
by a previous observation (see [HorD,
HI = ker T2 EB Im(T2*),
we have
and use II· III to denote both the norm in L2(Q, ((JI) as well as that in HI.
Consider
TI(hl,···,h n):= L gjhj,
l:::oj:::on
which is clearly continuous HI ~ H2. Define an operator T2: HI ~ H3 by:
(a) h = (hi, ... , hn) E dom(T2) if and only if hj E dom(T) for every j;
(b) T2(h) = (ohJ/oz, ... , ohn/oz).
In this case, the space ker T2 is exactly (L2(Q, ({Jd n Jf'(Q»n. The operator TI
sends ker T2 into a closed subspace G2 of L 2(Q, ({J2),
G2 := L2(Q, ((J2) n Jf'(Q).
We are in a situation like that of Proposition 2.3.1, hence, in order that every
holomorphic function I in L2(Q, ({J2) could be written in the form
1= L hjgj ,
l:::oj:::on
with hj E L2(Q, ((JI) n Jf'(Q) , it is enough to prove the estimate in Proposi-
tion 2.3.1.
,We will assume henceforth that Q is an open bounded set with Coo regular
~~undary and that ({JI, ({Ji are defined and of class C 2 in a neighborhood of
Q. We let ({J := ({J2 - ({JI. For any h = (hi, ... , hn), denote, as usual, Ihl =
(L l:::oj:::on Ihj 12) I /2. Moreover, we will assume that
Igl > 0 in a neighborhood of n,
i.e., the functions gj have no common zeros in n.
Let us compute Tt(u) for u E L2(Q, ((J2). Tt(u) is defined by the formula
<Tl(h)luh = (hITt(u)h. On the other hand, for h = (hi,"" h n) E HI, we
have
140 2. Interpolation and the Algebras A p
We conclude that
Tt(u) = (glue-"', ... , gnue-"').
Let v = (VI,"" vn) E dom(Tn. Since T2*(v) = (T*(VI),"" T*(v n», the
inequality
= 2Re L (T(gjue-"'lvj»1
l::sj::sn
= 2Re L
l::sj::sn n
1aza-(g·e-"')v·e
J
- -"'I dm
J
with IvI2
= Ll~j~n IVjI2. For that purpose, we note that IIT2*(v)lIr =
Ll~j~n IIT*(vj)lIr, hence it is enough to obtain this estimate for the case of
a single function (i.e., n = 1).
Let c(n) be the family of all Coo functions on n (i.e., in a neighborhood of
n). For f E c(n) n dom(T*) we claim that
a
T* (f) = -e'l'l - (f e-'1'1).
iJz
In fact, we must have
(T(h)lfh = (hlT*(f)h
for every h E D(Q). Now,
(T(h)l!)l = 1. --=-
ah
n az
fe-'l'l dm
and, from
1. a
we conclude that
(T(h)l!)l = - 1. a -
h--=:(fe-'I'I)dml
n az
+ -
--=:(hfe-'I'I)dm
n az
since the other term vanishes, as one sees by applying the Stokes formula.
142 2. Interpolation and the Algebras Ap
(T(h)l/h = -
Jnr h [e'P[~(fe-'P[)]
az
e-'P[ dm = - (hle'Pl~(fe-'Pl))
az 1
,
whence
as claimed.
We shall admit a technical result whose proof can be found in an even more
general setting in [Ro3, Lemma 4.1.3].
2.3.4. Lemma.
(i) The space c(Q) is dense in dom(T), endowed with the graph norm
(lIfIIf + lIaf/azllf)1/2.
(ii) The space c(Q) n dom(T*) is dense in dom(T*), endowed with the graph
norm (1I/1If + IIT* fIIf)1/2.
r
Jan
gje-'P[ dz =0
2.3. Ideal Theory in Ap 143
for every g E e(Q). Let s be the arc length parameter in an, the function dz/ds
is Coo on an, hence we can find a function g E e(Q) such that g = fdz/ds
on an. We conclude that f E dom(T*) if and only if f = 0 on an. 0
We can now show that the estimate we needed for IIT2*(v)lIi holds.
IIT*(f)lIi ~
a2cp~ dm.
(lfI 2e-q>\
az azin
Proof. From Lemma 2.3.4 we conclude that it is enough to prove the inequality
for f E e(Q) n dom(T*).
Let us recall that we have shown in the proof of Proposition 2.1.2 that for
any u E C 2 one has
I
{ eq>\ au 12 dm = {Iu 12eq>\ a2cp~ dm + { I
eq>\ au + acpI 12 dm
in az inaz az in az az
~ (lul2eq>\ a2cpI_dm.
in az az
When f E e(Q) n dom(T*) we have f = 0 on an and T*(f) =
-eq>\ (a/az) (e-q>\ f), so we can apply the previous inequality to u = -eq>\ f and
obtain
(i)
(ii)
Hence,
Therefore,
- ag
2
L g.-
az J
j
l::,j::,n
L g- ag.az- 2] .
J
j
l::,j::,n
which is valid for any choice aj, b j E C, 1 :::; j :::; n. This concludes the proof
~~ 0
2.3. Ideal Theory in Ap 145
2.3.8. Lemma. Letaj, bj , Vj E C, I ::s j ::s n, lal 2 = EI:;:j:;:n lajl2, and similarly
IvI2 = EI:;:j:;:n IVjI2. Then
2
On the other hand, the Lagrange identity used in Lemma 2.3.7 yields
2
a2f{J1
Ivi 2- ex L a Ivi 2-a-f{J .
2
az -
az - -Igl 2 .
I:;:J:;:n
erp-(g·e-rp)v·
az J J
>
- az az
Since IgI 2e- 2rp -rp, = erp-2rp, = e-'P2, with f{J2 := f{J + f{JJ, we have obtained the
following result:
146 2. Interpolation and the Algebras A p
2.3.9. Proposition. Let Q be an open bounded subset in the complex plane with
Coo regular boundary. Let 1/1 be any C 2 function in the neighborhood ofn which
is subharmonic in Q. Assume that gj, 1 ~ j ~ n, are holomorphic functions in a
neighborhood ofn such that Igl 2 = 2:1~j~n Igjl2 > 0 on Q. Fix any a ~ 1 and
let f{JI = 1/1 + a log(igI2) and f{J2 = 1/1 + (a + 1) log(igI2). Then,for any u E G2
and v E dom(TD one has the inequality
f = L hjgj
I~j~n
(resp. f = 2:j~1 hjgj , the series being convergent in Je(Q)). Moreover, the
following inequality holds:
We can appeal now to Proposition 2.3.1 to obtain the theorem in this case.
Let us now eliminate the condition that n is finite. Let (gj)j~1 be a sequence
of functions which are holomorphic in the same neighborhood of and 0 < n
2:j~1 Igj 12 converges uniformly on n.
Let f be a holomorphic function satis-
fying condition (*) with respect to this sequence. There is an integer no » 1
such that for n ~ no we have 2:1~j~n Igj 12 > 0 on n
and condition (*) holds
2.3. Ideal Theory in Ap 147
when we replace the infinite sequence by gl, ... , gn. Hence, there is a sequence
(hj,n)j?l such that hj,n = 0 for j > n:
(i) 1= Lgjhj,n;
l:::;j
(We have used that Lj?l Ih},n(~)12 is a subharmonic function, being a finite sum
of subharmonic functions.) Therefore, we can find a sequence (nk)k?lo nk ~ 00,
such that for every j, h},nk converges to a function hj in Jt"(Q). Given K cc Q
and N an arbitrary integer,
with Ihl 2 = L}?l Ihj l2. In the same way as above, one obtains that Lj?l Ihj l2
converges uniformly over compact subsets of Q.
148 2. Interpolation and the Algebras Ap
(i) f=~gjhj,n;
j~1
This function is Coo and proper. Moreover, by Sard's lemma the set of critical
values of p is of zero measure. Therefore, there is an increasing sequence of
2.3. Ideal Theory in Ap 149
(ii)
~ _a_
a -1 Jo
r1/12Igl-2a-2e-Vt dm.
For m fixed and n > m, the sequence (hj,n) is bounded in Om. It follows
that we can find an increasing sequence nk ~ 00 such that for every j :::: 1 the
sequence hj,nk has a limit hj in Jt"(0) when k ~ 00. It is easy to see now that
the sequence (hj)j~l has the required properties.
To conclude the proof we only have to eliminate the supplementary hypothesis
that Igl > 0 on O. Let Z be the discrete set of common zeros in 0 of all the gj,
and apply the theorem to Of = O\Z. Since 1/1 and Igl are locally bounded above,
the functions h j are locally square integrable in 0, hence their possible singu-
larities at the points of Z are removable, as shown by the following elementary
lemma:
2.3.11. Lemma. Let 1 be holomorphic in 0 < Izl < r, ~Z\<r 1/1 2dm < 00, then
the singularity 011 at z = 0 is removable.
Proof of Lemma 2.3.11. Let 1 (z) = EkeZ akzk be the Laurent development of
fin 0 < Izl < r. For any 0 < e < r we have
1e:::lzl:::r
r
I/(z)1 2 dm=21l'la_11 2 10g-+1l'
e
L lad r
k#-l
2k +2 -
k +1
e2k+2
.
rIhI2e-C4'"
Jo.
dm ::: _a_
a-I
r
Jo.
IfI 2e- C3 1{! dm,
Therefore, there are (hj)j~l holomorphic in 0 such that f = L:j~l gjhj and
rIhI2IgrZae-c~I{!
Jo.
dm ::: _a_
a-I
r
Jo.
IfI2e-C31{!.
As another application of Theorem 2.3.10 we shall see what are some of its
consequences for the algebra Ap(O), 0 an open set in C. We assume the weight
p satisfies the conditions (i)' and (ii)' of §2.1. That is, there are positive constants
K 1, K 2 , K 3 , K4 such that z E 0, ~ E C, and Iz - ~I ::: exp(-K1P(z) - K 2 )
implies that ~ EO and p(~) ::: K 3 P(z) + K4, and 10g(1 + Iz12) = O(p(z»
(though this last condition is not essential for what follows).
2.3. Ideal Theory in Ap 151
We remind the reader that / E Ap(Q) if and only if / E ;f(Q) and there is
°
C :::: such that
h'/'2e-CP dm < 00.
From the second assumption above, it follows that Ap(Q) contains the restric-
tions to Q of the polynomials. More generally, consider a family P of subhar-
monic functions such that if PI, P2 E P, then PI + P2 E P and max (PI , P2) E
P. We associate to P the algebra Ap(Q) of those holomorphic functions in Q
such that 1/1 :s Ce P for some PEP (i.e., Ap(Q) = UPEP Ap(Q». It follows
that Ap(Q) contains the restrictions of the polynomials and that / E Ap(Q) if
°
and only if there is PEP and C > such that
That is, one can define the growth conditions in Ap(Q) either by L oo estimates
or by L 2 estimates.
As an immediate corollary Theorem 2.3.10 we have the following theorem
due to Hormander [H03]:
2.3.15. Corollary. Let gl, ... , gn E Ap(Q), then the/ollowing three conditions
are equivalent:
°
(1) I(gl,oo.,gn) = Ap(Q).
(2) There is pEP and A > such that
1Q
e-P
---dm < 00.
igi 2a +2
2.3.16. Remarks. (1) Hormander considered the case P = {cp: c > o}, P a
fixed subharmonic function.
(2) If there is a sequence (gj)j,,:1 of holomorphic functions in Q such that
Ale-PI :s Igl :s A2eP2
for some constants AI, A2 > 0, and PI, P2 E P, then for every / E Ap(Q),
there is a sequence (hj)j":l ~ ;f(Q) such that:
(i) / = Lj":l gjhj in Q; and
(ii) Ih I :s A 3eP3 for some P3 E P, A3 > 0,
which is Hormander's result extended to the case of infinitely many generators.
Let us now give a situation where the need to consider an infinite sequence
arises naturally.
Let P be a subharmonic function in an open set Q. Assume that for every
j E N*, the set Qj := {z E Q: p(z) < j} is relatively compact in Q. We recall
152 2. Interpolation and the Algebras Ap
that Ap(O) carries an inductive limit topology, which is not metrizable. Never-
theless, the spaces Ap,B(O) are Banach spaces and thus, from Proposition 1.4.13,
the closure of any subspace F of Ap(O) coincides with its sequential closure.
Therefore, an ideal I in Ap(O) is dense if and only if it is sequentially dense
in Ap(O), i.e" there is a sequence (/kh~1 in I that converges to 1 uniformly
over every compact subset of 0 and besides, there are two constants C 1, C2 ::: 0
such that
(Z EO, k::: 1).
In this case, for every j ::: 1 there is kj such that
then
T-2
e- 2P (Z) ~ ~e-2j ~ e-2jlfk/Z)12 = Igj(z)1 2 ~ Ig(z)1 2 •
This shows that the existence of four positive constants, A, B, A', and B' so
that
Afe-B'p ~ Igl ::::: Ae BP .
Conversely, Remark 2.3.16(2) shows that if we can find a sequence gj E
I, and positive constants A, B, A', and B' such that the previous inequality
holds, then I is sequentially dense. Therefore, we have proved the following
proposition:
fn(z) := -
1 12lr e-Itze-
.. Int dt =
1 - e- 2lriz
.
21l' 0 21l'i(z+n)
2.3. Ideal Theory in Ap 153
Since for each Z E C, the fn (z) are the Fourier coefficients of the function
t r-+
L'" Ifn(z)1 2 = -
1 1
e- itz , we can apply Parseval's formula and obtain
2Jl'
le- itz l2dt = -
1 e4Jl'(Imz) - 1
.
nEtu
27r 0 47r Imz
It is easy to verify the inequalities of Proposition 2.3.17 and conclude that the
ideal generated by the fn, nEZ, is dense in F(C"(JR».
Let us consider a finite family gl,"" gn E Ap(Q). Denote h =
I(gl,"" gn), the ideal generated by them in Ap(Q) and let Z(/I) be the set
of their common zeros in Q. A function f E .;r; (the radical of II) if there are
k E N*, hj E Ap(Q), such that
fk = L gjhj .
We conclude that
Let now 12 be the family of f E Ap(Q) for which there are A > 0, pEP
such that
If I :s: Algle P •
It is easy to see that h is an ideal in Ap(Q) and that II ~ h
If f E 11, then there are A > 0, PEP, such that
If I :s: Algl 3e P ,
which implies, via Theorem 2.3.10, that fEll. Therefore, we have the following
statement:
1= L hj}j.
I::;j::;n
and hence
h = L
I::;j::;n
hjfih = L
I::;j::;n
hjgj Eft.
for u E G2, V E dom(Tn, ({JI = t/I + log(lgI2), ({J2 = ({JI + log(lgI 2) = t/I +
210g(lgI 2). Let us replace t/I by t/I + 2 log (1 + IzI2). Then, for this new weight,
the meaning of G2 changes but, since
for u E G2 , V E dom(Tn, f{JI = 1/1 + log(lgI 2) + 2 log (1 + IzI 2), f{J2 = f{JI +
log(lgI2). We apply this observation to prove the following proposition:
Let p:= 1/1 + 210g(lgI 2) + 210g(1 + IzI 2), then there are functions hj E Lloc(Q)
such that:
1L
2
l(wlv)I1 2 = w/!Jje-'P'dm
n I~j~n
:s (
in
rIw12(1 + IzI2)2e-'PI dm) ( inr (1 +Ivl2IZ12)2 e-'PI dm) .
Therefore, if we let f{J2 = p, to conform with the notation of Proposition 2.3.9,
we have that for u E G2 and v E dom(Tn
Then:
(i) 0 is well defined because if Tt(u) + T2*(v) = 0 then
r Ivl2
in (1 + Iz12)2
e-'P'dm =0
'
hence v = 0; and
156 2. Interpolation and the Algebras A p
In other words,
ah·
(a) a{ = Wj, 1 5: j 5: n.
1 O\Z
Ifl2
- 4 (1
Igi
+ ~(log Igl»e- 1ft dm < 00,
there are hI, ... , h n E K(n) (resp. a sequence (hj)j~1 in K(O» such that
f = L gjhj
I::;:j::;:n
and
f If I: e-1/I 2 2dm 5: 2 f If I: (1 + ~(log Igl»e-1/I dm.
Jo Igi (1 + Izl ) JO\Z Igi
Proof. One starts by assuming that 0 is a bounded open set with regular
boundary of class Coo, 1/1' is C 2 and subharmonic in a neighborhood of g,
the n functions gj are holomorphic in a fixed neighborhood of g, and Igl > o.
Afterward, one eliminates these restrictions one by one as done in the proof of
Theorem 2.3.10. We shall just indicate how to prove the first step.
One can write f = LI::;:j::;:n gjhj with
h'·
j.=
i.i2 •
f Ig1
2.3. Ideal Theory in Ap 157
Wi = 1~4 L gj (gj ~~ - gi ~; ) .
l~J :9
Iwi 2 = L -1/128
Igl
L g'] (a gi aaz-
g 'az- - g i
]
gj )
l~i~n l~j~n
Hence,
1W,2
1 -e-'/Idm<
n Igl 2
11/12
- 1 1gag
- n Igl 4 l~i,j~n Igl 4 ] az
L - ag-j 12 e-'/Idm.
' -i - g i
az
Therefore,
From Proposition 2.3.21 we see there are functions h'j, 1 :::: j :::: n, such that:
ah"
(1) af = Wj;
(2) 1n ( Ll~J~n
gjh'j) ue-f{J2 dm =0 (u E G 2 ); and
for u E G2.
Hence Tl (h) =1- Tl (h") is also square integrable in n, Tl (h) - 1E G2, and
(T1(h) - Iluh = 0
158 2. Interpolation and the Algebras Ap
1.
We have
Ihl2 e-1/I
-
2
n Igl (1 + Iz12)2
dm=
1.
Ih' - h"12 e-1/I
n Igl2 (1 + Iz12)2
dm
r
Ih'I2 e-1/I Ih"12 e-1/I r
~ 2 in Igl2 (1 + Iz12)2 dm + 2 in IiI2 (1 + Iz12)2 dm.
But Ih'I2 = IfI 2/lgI 2, hence
1. 2
-
n Igl (1 + Iz12)2
1.
Ihl2 e-1/I dm <2 -Ifl2 e-1/I dm+2 -e-1/Idm
- n Igl (1 + Iz12)2
4
Iwl2
n Igl 2
1.
~ 2
Ifl2
1.
- 4 (1 + ~(log Igl)e-1/I dm.
n Igl
As pointed out earlier, the remainder of the proof can be carried out as in the
case of Theorem 2.3.10. 0
2.3.23. Corollary. If gl, ... , gn E Ap(Q) and there is apE P such that
r
in\z
~(log Igl)e- P dm < 00,
then
EXERCISES 2.3.
1. Let gl,"" gn, fl,"" fn, and h be holomophic functions in a disk B satisfying
gj = iJh (1 :5 j :5 n) and for some A > 0,
Show that the functions iJ cannot have a common zero in B (see the proof of Proposition
2.3.20).
2. (The objective of this exercise is to prove a very simple case of a theorem of
Bombieri [R03, Theorem 4.4.4].) Let p be a subharmonic function in an open set Q.
(a) Let U E Jf"(Q) satisfy Io
lul 2e- p dm < 00. Show that if ~o E Q is such that e- P
is not integrable on any neighborhood of ~o, then u(~o) = O.
(b) Let B(zo, r) ~ Q, r > 0, be such that e- P E LI(B(zo, r), dm). Show that there is
U E Jf"(Q), such that u(zo) = I and
Conclude that there is a nonzero U E Jf'(Q) such that u satisfies the condition (*) and
vanishes at every point ~ E Q such that e- P ¢ L I (N~) for any neighborhood N~ of ~.
*3. Let !/f be a subhannonic function in B = B(O, 1) such that !/f(0) = 0 and !/fez) < 1
for every z E B. Let t-t = t!,.!/f.
(a) Show there is a Radon measure du on aB(O, 1), du(O ::::: Id~ I such that
!/fez) = ~ r log
2n J B
I1z-- z~~_I dt-t(~) + r
JaB
1 -lzl:dU(~).
Iz - ~ I
(Hint: Use the Riesz decomposition theorem for B(O, r) and let r -* 1-. [BG, §4.4.24].)
(b) Show that
rlog~dt-t(O+ r (ld~l-du)=2n.
JB I~I JaB
l
Conclude that
ldU(O' ::::: 4n.
aB
11 -lzl:dU(~)1
(c) Show that
I~
2n aB Iz - ~I
::::: 6 if Izl:::::~.
1 r
a = a(R) := -2 dt-t(n
n JB(O.R)
Show that
0::::: a::::: (-!ogR)-1 < 2.
(e) Fix R E 14, e- 1/ 2 [. Show there is a constant C > 0 such that for z E B(O, 1/2) one
has
11 R<I~I<1
Iz-~I
log - - - dt-t(~) < C
11-z~1
I
-
and
exp (-4 r
JB(O.R)
log Iz - ~~ dt-t(~») <
11 - z~1 -
r
JB(O.R)
( Iz - ~~ )
11 - z~1
-a dt-t(~)
2na-
< C.
*4. Let rp be a subhannonic function in C and Hrp the set of all entire functions u such
that
for some N ~ O.
160 2. Interpolation and the Algebras A p
(a) Let U E Jfr(C) be such that lul 2e-'I' E Lloc(C), Fix R > 0 and let X E 1)(C), X == 1
on B(O, R + 2). Let v := XU, f := 8v/8z. For t > 0 define
(b) Show there exists Ut E Coo such that 8u,j8z = f and satisfies
(i) lim J; IUt 12 exp( -qlt) dm = 0;
1-+00 C (1 + Iz12)2
(ii) tl!.~ ~zl<R+l lutl 2 dm 0; and=
=
(iii) lim Ut 0 uniformly in Izi < R.
t-->oo
r
(c) Show that
IUtl 2 -'I'
lc (1 + IzI2)2+(t/2) e dm < 00;
!
and
Iv-utl2 -'I'
(1 + IzI2)2+(t/2) e dm < 00.
(d) Conclude that the closure of Hrp in Jfr(C) contains the set of all U E Jfr(C) such
that lul 2e-rp E Lloc(C),
2.4.2. Remarks.
(1) It is clear that we can assume that the sequences (Uj)j?J. (Kj)j?J.
and (Sj)j?l are increasing. Under these conditions we shall denote Sj =
(F!. ... , FN), where (Nj)j?l is an increasing sequence of integers.
(2) A set S is exhaustive if and only if the ideal I (S) generated by S in Ap (Q)
is exhaustive. Let us show, for instance, that if I (S) is exhaustive, then one can
find subsets Sj 5; S, verifying (4) for the same sequence (Uj)j?l associated to
I (S). In fact, there is a finite, nonempty set <5j = {It. ... , INj} 5; I (S) such
that
Let us write !k = ~i ak,i Fk,i, for some finite collections Fk,i E S, ak,i E Ap(Q),
both collections depending on fk. Let Sj be the (finite) collection of all Ft.i E S
appearing in one of the above representations of Ik E <5j. Then
J~N' Ij,(') I' " IfNi (~Ia", (,) I') (~IF'A') I')
for some Aj , Bj > O. This gives a choice of Sj and constants C;,j' q,j for the
set S.
(3) If S is exhaustive, its elements cannot have a common zero Zo E Q. Other-
(L
wise,
(j :::: 1, z E Q).
(z E Q, j :::: 1).
2.4.4. Lemma.
inr10<tJj o.
Vk dm) =
lim (lim sup
J-+OO k-+oo oz 12 e-
Proof of Lemma 2.4.4. Choose k(j) :::: 1 such that Uk :::: 0 on supp(qJj) for all
k :::: k(j). Let K; = supp(qJj) n (Q\Kj ), then for k :::: k(j) we have
Since M j = sUPzEn\K d (z, QC)2 /[ (1 + Iz 12)2] :::: 1/j2 and Uk :::: 0 on sUPP qJj, we
c51
j
have
1 oi
n 1
-OqJj 12 e-vkdm < -
- P
dm
n (1 + Iz12)2
.
Let Ej := In e-
IOqlj/ozI 2 Vk (j) dm, with the strictly increasing sequence k(j)
chosen as in the proof of the preceding lemma, so that limj-H'O Ej = 0 and
Vk(j) ~ 0 on the support of qlj.
From Theorem 2.4.3 it follows that there are functions Vrj such that 0Vrj / 0Z =
Oqlj/oz and
r
in IVrjl
2
(1
e-Vk(j)
+ Iz12)2 dm :::: 'iEj.
I
The functions h j := qlj - Vrj are clearly holomorphic in Q. We claim that all
hj E I and that the sequence (hj)j~1 converges to 1 in Ap(Q). Let us prove first
the last item. We need to show first that there exist constants C > 0, B > 0,
such that, for every k ~ I and every z E Q,
e-Vk(j)
e-Cp(z) < B---"7
- (l + IzI2)2·
In fact, letting fJ = log B, this inequality is equivalent to
-Cp(z) :::: fJ - Vk(Z) - 2log(1 + IzI 2),
i.e.,
Vk :::: fJ + Cp(z) - 2log(1 + Iz 2 ).
1
But, Vk = 4(Uk + log(1 + IzI 2 )/d(z, QC)) and there are positive constants AI,
B I , A 2 , B2 such that
log(l + Iz12) :::: Al + BIP(z),
1
d(z, QC) :::: A2 exp(B2P(Z)),
r 11 - qlj
in
2e- Cp dm::::
1 r
in\K j
e- cp dm,
.lim
J-+OO inr 11 - qlj 12e- Cp dm = O.
Therefore,
164 2. Interpolation and the Algebras Ap
lr:.[ Ih'1 2
J 3 e-CjP dm < 00,
D C::;~Nj IFk1 2)
then there are functions hj,l' .•• , hj,Nj E Ap(Q) such that
hj = L hj,kFk,
l::;k::;Nj
and, moreover,
L Ih j ,kl 2 2
[1::;k::;Nj -CjP d < 2 [ Ih j I -CjP d
lr:. 2 e m lr:. 3 e m.
D C!:~Nj IFk1 2) - D C!:~Nj IFk1 2)
The existence of the constants Cj > 0 follows from property (4) of Definition
2.4.1 and a reasoning entirely similar to those leading to the estimates for 1/tj. D
2.4.5. Corollary. Let I be an ideal in Ap(Q), which has no common zeros, and
for which there exist finitely many F 1 , ••• , FN E I and e > 0, C > 0, such that
every connected component of the set
We have
D = {z E Q: u(z) < OJ.
2.4. Dense Ideals in Ap(Q) 165
Define
u(z) if z ¢ K)·,
u,(z)·- {
) .- max(O, u(z)) if z E Kj ,
and Sj being equal to the set {FI. ... , FN} to which we have adjoined a finite
subset 6 j of I with no common zeros in Kj • It is easy to verify that (Uj)j~1.
(K j )j~i, (Sj )j~1. satisfy the condition of Definition 2.4.1. Hence I is exhaustive
and the corollary has been proved. 0
2.4.6. Remark. The proof of Lemma 2.4.4 also shows that if u 1 ::: U2 ::: •••
satisfy the conditions (1), (2), and (4) of Definition 2.4.1, and if
u(z) := ,lim Uj(z),
)-+00
3. Let p(z) := log(1 + Iz12) and Q = {z: Rez > O} as in the previous exercise. Also
let 8n(z) = (n/(z + n»n, n ~ 1.
(a) Show that 8n E Ap(Q) and limn_co 8n = 1 in Ap(Q).
166 2. Interpolation and the Algebras Ap
(b) Find explicitly the function rpo E Ap(Q) such that rpo = limHooe~.
(c) Prove that the ideal I = rpoAp(Q) is different from Ap(Q).
(d) Is rpo invertible in Ap(Q)?
2.5.1. Definition. The local ideal hJC associated to an ideal I in Ap(Q) is the
collection of all I E Ap(Q) such that for every Z E Q there is a neighborhood
U of z and there exist functions /J, ... , In E I, gl, ... ,gn E .te(U), verifying
I = L jjgj in U.
l:'Oj:<on
It is clear that if V = (Zk. mkh:O:1 denotes the multiplicity variety of the ideal
I, and I(V) (resp. leV)) is the ideal in Ap(Q) (resp. in .te(Q», corresponding
to V, i.e.,
= {f E Ap(Q): I vanishes at every Zk with multiplicity ~ md,
I(V)
then hoc = I (V). Moreover, if i: Ap(Q) ~ ;f(Q) is the canonical injection, 1
is the closed ideal generated by I in .te(Q) (i.e., by iU)), then 1 = leV) and
Iloc = i-I(J). Hence, Iloc is closed in Ap(Q). Furthermore, the identities
2.5.2. Definition. Given an ideal 1 in Ap(Q) the conductor ideal (or quotient
ideal) of I is the ideal 1* defined by
[* := {f E Ap(Q): Ig E I for every g E Iloc},
1* 5; (i)*.
Proof. We already know that (i)* is closed and that I loc = (i)loc' Hence for any
f E [* and g E (i)loc, then f gEl 5; i. Therefore, f E (i)*. It follows that
1* 5; (i)*. 0
This concept will play an important role in the study of convolution equations
later on. We see that for I to localize it suffices to prove that [* is dense in
Ap(Q), but this requires at least that l* has no common zeros. In fact, we can
show that this is always the case.
2.S.S. Proposition. For every ideal I of Ap(Q), the ideal [* has no common
zeros.
Proof. We can assume that I =f. {OJ (Why?). Let us argue by contradiction and
assume there is Zo E Z(I*). We claim that if g E [*, then the function G(z) :=
g (z) / (z - zo) belongs to [*. To continue the proof we need the following very
simple lemma:
1
IF(z) I ::: max IF(s)1 = - max If(s)l,
I~-zol=r r I~-zol=r
168 2. Interpolation and the Algebras Ap
We return to the proof of the proposition. Since V(l*) ~ V(l), the multi-
plicity m of [ at Zo verifies that m 2: 1. Let h E [ be such that the multiplicity
of h at Zo is exactly m. Then we can write
h(z) = (z - zo)m H(z),
2.5.10. Corollary. Let [ be an ideal in Ap(Q) for which there are functions
F l , •.. , FN E [ and constants 8 > 0, C > 0, such that every connected compo-
nent of the set D = {z E Q: (Ll::J:<:N IFj(z)l)I/2 < 8 exp( -Cp(z))} is relatively
compact in Q. Then i = hoc.
Proof. The preceding Proposition 2.5.8 shows that [* has no common zeros
and, since F l , ... , FN E [*, we can apply Corollary 2.4.5. We conclude that [*
is sequentially dense in Ap(Q). Hence i = hoc by Proposition 2.5.6. D
In general, it is not true that i = [loco We show this with an example due to
Gurevich ([Gu2l, see also [KT2] and Exercise 2.4.2).
2.5. Local Ideals and Conductor Ideals in Ap 169
2.5.11. Example. Let Q = {z E C: Rez > O}. For z = re ill , -rr/2 < e < rr/2,
let
p(z) := exp( -ar" cos e + r P) + r Y ,
with 0 < {3 < a, 0 :s y < 1, a > O.
Since {3 < a, the quantity -ar" cos e + r P --+ -00 whenever r --+ 00 for e
fixed in] - rr /2, rr /2[. Therefore, every function f E Ap(Q) is of order y along
every ray starting at the origin, but the type of this function could increase to 00
as this ray approaches the vertical axis. We shall show that the principal ideal
[ = e- z Ap(Q) is not localizable in Ap(Q). It is clear that [ =I- Ap(Q) since
eZ .;. Ap(Q).
Since [ has no common zeros, ['oc = Ap(Q). We want to show that i =I-
Ap(Q). Assume that the opposite is true. We would then have a sequence (gnk'::i
in Ap(Q) such that:
(i) gn(z)e- Z --+ 1, uniformly on every compact subset of Q; and
(ii) Ign (z )e- ZI :s C, e C2P (z) for all n ~ I and some C" C2 > o.
Consider the open set Q, := {z E Q: Rez > rY}. On the set Q, we have
-ar" cose + r P < -ar Y+"-' + r P•
Assume that
{3<y+a-1,
then there is a constant Co > 0 such that
p(z) :s Co + r Y for all z E Q,.
Therefore, we have
Ign(z)e-ZI :s C, exp(COC2 + C2 r Y )
and
since Re z = r Y on aQ,.
In order to apply the Phragmen-Lindel6f principle consider the auxiliary
function
f(z) := exp(-C~zY), C' _ C2 + 1
2 - cos(rry /2)
We have
Ifgnl :s C; on aQ,
and the order of fgn is at most y < 1 in Q,. The Phragmen-Linde16fprinciple
asserts that
The preceding example raises the natural question of whether there are weights
p such that every ideal in Ap(C) localizes. In the next section we prove this for
p(z) = IzIP,O < P < 00. The very important case p(z) = IImzl +log(1 + Iz12)
is left to Chapter 6.
We denote here Ap the algebra Ap(C), p(z) = IzIP, 0 < P < 00. Note that all
the functions of this algebra are either of order < p or order p and finite type.
(In the literature sometimes a function of order p and infinite type is also said
to be of order at most p.) For this algebra we want to give more precision
to the results of this chapter about interpolation and ideals. Many reasonings
extend to the Case where p is a radial weight satisfying the doubling condition
p(2z) = O(p(z». We start with a result of Lindelof which we take from [Lin],
and we also suggest [Lev], [Bo] for a complete study of entire functions and their
zeros. Part of this section is based on the doctoral thesis of Squires [Sql], [Sq2].
Let f be a nonconstant entire function and recall that n (r) denotes the number
of zeros of fin B(O, r), counted with multiplicities. We assume, for simplicity,
that f (0) = 1. Let then denote, as usual,
M(r) = max
Izl=r
If(z)1 and N(r) = 1o
r n(t)
-dt.
t
Hence Jensen's fonnula (see [BG, Chapter 4]) is simply
2.6.1. Lemma. II I has finite order p > 0 and finite type -r, then
. n(r)
hmsup - :::: ep-r.
r ..... oo rP
In particular, n(r) = O(rP).
Proof. In fact, for e > 0 given there is r(e) > 0 such that
log M(r) :::: (-r + e)r P, r ~ r(e),
hence
N(r)
--;:;;- :::: -r + e, r ~ r(e).
2.6. The Algebra Ap of Entire Functions of Order at Most p 171
Now let f3 > 1 and r 2: r(e), then nCr) being increasing implies that
n(r)logf3:s 1 r
f3r net)
-dt:S
tot
1
f3r net)
- d t = N(f3r):s f3P(r +e)r P.
Therefore
nCr) :s (r
rP
+ e) inf
f3>1
(L) =
log f3
ep(r + e), r 2: r(e).
If the order p of the function f is an integer, one cannot expect such a simple
result. The reason is that either the term e Q can dominate or the infinite product
P can have infinite order while n(r) = O(r P). These possibilities occur in the
examples e Z and II r(z). In the latter case, n(r) = O(r) but the function has
infinite order. There is a classical result of Lindelof that deals with this case.
Note that in case 2(i) we are also assuming that the series is convergent. For
the proof we need some preliminary results.
2.6.4. Lemma. Let f be a nonconstant entire function such that f (0) = 1. Let
a > 0 and let 27rrl(a, r) denote the linear measure of the set
A(a, r) := {z E aB(O, r): If(z)1 ::: M(r)-O"},
then for r » 1 one has
1
l(a, r) :s 1 + a .
Proof. Let a(a, r) := {e E [0, 27r[: re jlJ E A(a, r)}, then its measure is l(a, r)
and
_1 1
27r a(O".r)
log If(reilJ)1 de :s -al(a,r)logM(r),
0= log If(O)1 :s 2~ 1 2
1£ log If(reilJ)1 de :s (1 - (a + l)l(a, r)) log M(r).
As a consequence, as soon as log M (r) > 0, we have
1 - (a + l)l(a, r) ::: O. o
2.6. The Algebra Ap of Entire Functions of Order at Most p 173
1o
00 net)
----:;T d t < 00.
tP
Therefore,
nCr) < p
rP -
1
r
00
net) dt
t P+ 1
= 0(1) as r -+ 00.
< Krp-I
-
r
Jo
net) dt
tP
+ KrP 1r
00
net) dt.
t P+ 1
Let C > 0 be such that net) :::: Ct P, t > O. Given e > 0, let R > 0 be such that
n(t)/t P :::: e, then
1 o
r net)
-
tP
dt :::: C lR 0
dt +e 1
R
r
dt = (C - e)R + er.
Hence,
log lP(z)1 :::: (C - e)K Rr p- I + Ker P + o(l)r P = o(r P),
as we wanted to prove. o
2.6.6. Lemma. Let n (t) be the counting function of entire function f of integral
order p, f(O) = 1, then for Izi = r we have
L 109! E (~ , p - 1) !+ n>n(r)
I:,,:n:,,:n(r)
L 109! E ( :n ,p) !
< C (r p- 1
- P
r
Jo
net) dt
tP
+ rP+11OO
r
net) dt) .
t P+2
Proof. For fixed r > 0 consider two auxiliary sequences: the first, (un)n, is
finite, Un = Zn for 1 :::: n :::: nCr); the second, (vn)n, is the cofinal part of (zn)n,
for n > nCr). Let f.L" Vr be their respective counting functions. Then
net) if 0:::: t :::: r,
f.Lr(t) = { nCr) if t > r.
174 2. Interpolation and the Algebras A p
if 0 ::: t ::: r,
v,(t) =
{ 0n(t) - n(r) if t > r.
Then the first tenn we want to estimate corresponds to the canonical product of
the sequence (un)n, the second to the sequence (vn)n.
Let p = I, then from [BG, §2.6.1O] (or [Lev, p. 12]) we have
l'
n ,
n(t)
= -dt+n(r),
o t
"L.J
n>n(,)
I ( z ,I ) I ::: 4
log E Zn {1°Or
,
v,(t)
'(2dt +2r2 1T
,
00
v,(t) dt }
= 4 { 21' 2 1 n~t)
00
dt - n(r) } .
< C {r P- 1
- P
r
io
n(t) dt
tP
+ rP+1l°O n(t) dt}.
, tP+2
0
Proof of Theorem 2.6.3. Let us first prove that the conditions (i) and (ii) in
part (1) are enough to conclude that f is of finite type in the case when p = p.
We denote
D = O(rP- 1) + C
- P
(r P- 1 r
Jo
n(t) dt
tP
+ rP+1jOO n(t) dt)
r t P+2
.
We have n(t) :s At P for some A > O. It is clear that the two integrals together
are dominated by 2Cp Ar P, so that D = O(r P). Since S(r) is bounded, it follows
that:
log I/(z)1 = D + Re (zp {ao + ~s(r)}) = O(r P).
In other words, 1 has finite type dominated by 2CpA + S, where S is an upper
bound for lao + (1/p)S(r)l.
A similar argument can now be made for part (2), also when p = p. For £ > 0
given, we can find ro such that n(r) = £r P for all r ::: roo Then the integral terms
in the earlier estimate of D can be bound by
C r P-
P
1 1
0
ro n(t)
-
tP
dt + 2C
P
up.
Since the series Ln:::l z;;P is convergent to -aop, for r ::: rl we have
1
ao + - '"""' ~ z-P
n < £,
-
p l:::;n:::;n(r)
whence,
10gl/(z)1 = O(r P - 1) + (2C p + l)u P ,
so that 1 is of type zero.
Consider now the case where p = p - 1. Then the series Ln> I IZn r p < 00,
and the infinite product P(z) has type zero by Lemma 2.6.5. Wehave
Therefore, in part (1) we immediately obtain that log I/(z)1 = O(r P ), the type
of 1 is exactly laol in this case. In part (2), ao = 0 means that deg Q < p, then
log I/(z)1 = o(r P ).
This last argument concludes the proof of the sufficiency of the given condi-
tions. We now want to show these conditions are also necessary. In part (1),
we already know that 1 being of finite type implies n(r) = O(r P). The bound-
edness of S(r) is only a problem when p = p. Returning to the proof of the
176 2. Interpolation and the Algebras Ap
For r:::: 1 we can estimate log I/(z)1 :::: 10gM(r):::: Br P, but we need to esti-
mate I log I/(z)ll. We can apply Lemma 2.6.4, with a = 3, and obtain that the
measure of the set A(3, r) £ aB(O, r), where
log I/(z)1 :::: -3Br P
is 2Jl'rl(3, r), with l(3e, r) :::: ~. Therefore, in the complementary of the set
Ur>r
_0 A(3, r) we have
We want to conclude from this that S (r) remains bounded. If S (r) is not
bounded, then the expression between square brackets is not bounded. This
means that for some sequence of values rj -+ 00, we have for some ({Jj E [0, 2Jl' [
and Rj > °
Let s > °be given, then we have > °such that the term
ro o(r P) is bounded by
uP for r :::: ro, and we can also assume log M (r) :::: sr P for r :::: roo As above,
the measure of the set A(3, r) £ aB(O, r) where
I log I/(z)11 :::: 3sr P , Izl = r :::: ro,
2.6. The Algebra Ap of Entire Functions of Order at Most p 177
If not, there are rj --+ 00, Rj > 8e, ({Jj E [0, 21f[ such that for Z = rjei9 ,
Re {Zp lao +.!. L z~l} = Rjrf cos(O +((Jj) 2: ~Rjrf > 4erf
p l:5n:5n(r)
on a subset of aB(O, rj) of linear measure (21f/3)rj. But, any point of this set
belongs to the exceptional set A(3, rj). Using the expression (*) we obtain a
contradiction, so that the inequality (**) is true.
We need to show that (**) implies that the series Ln> 1 z;;P is convergent and
has the sum -aop. Consider a finite sum -
if r 2: roo Hence,
lim sup ao
N ...... oo
+.!. L -.;.1:::: ge.
P l:5n:5N Zn
2.6.7. Proposition. Let the canonical product P(z) = Iln>l E(z/zn, p) repre-
sent a/unction in the space A p , and PN(z) = Ill:::;n:::;N E(zlzn, p). Then
and let n(t) be the counting function of the whole sequence. First, consider the
case p ¢ N. Then we have p < p < p + 1 and some constant K = K(p) > 0,
OO}
< Kr P { f --dt +rJ--dt
r n(t) n(t)
- io tP+l tP+2
r
where n(r) ::: Ar P , since P e Ap. So that the sequence PN is a bounded sequence
in A P' therefore
lim PN
N-+oo
=P in Ap.
In the case p e N*, p = p, we recall that the function S(r) = I:l:on:on(r) z;;P
is bounded, say IS(r)l::: C. It follows that the functions ISN(r)1 =
II : 1:on9N(r) z;;PI are bounded independently of N, say by a constant C 1 > 0.
(Here we use n(r) = O(r P ) as in the proof of the necessity of conditions (i),
part (2) of Theorem 2.6.3.) As in Theorem 2.6.3 we rewrite
In the last chapter (Definition 1.3.24) we found that the space Expo(1C) of
entire functions of infraexponential type, i.e., order 1 and type 0, is isomorphic
to the space of infinite order differential operators. For that reason it is interesting
to consider the spaces of functions of minimal type.
2.6.8. Definition. Let Ap,o denote the Frechet space of entire functions of order
p, p > 0, and type zero. The. norms are given by
PN N.... ~P'
in Ap,o, The same statement if true if P E N* and the genus p =p - 1.
Proof. It is the same as that of the first part of the preceding proposition. Note
that in the case p E N*, p = p - 1, the products PN E Ap,o. 0
2.6.10. Remarks. (1) In the case p E N*, we have a difficulty when the genus
p = p. In this case, the canonical factor E«z/zn), p) does not belong to Ap,o.
Hence, in general, we have the same problem for the partial products. On the
other hand, let us consider the case of an even function I E Expo(lC), 1(0) = 1.
Then the zeros appear in pairs ±ak, and if we order them so that
then
L-=O,
1
n~! Zn
so that if I:n~! (1/lzn I) = 00, the canonical product has genus p = I and
do belong to Expo(lC), they converge uniformly over compact sets, and for every
mE N*, ("IN"m)N~! is a bounded sequence. This implies that
IN N.... ~I
in Expo(C).
Note that if we denote by iN(z) = IlbN(1 - (z/ak)2), then iN(z) E Expo(1C)
and iN ~ 1 in Expo(lC).
(2) Similarly to the last part of the previous remark we have that if P (z) =
Iln~! E(z/zn, p) is a canonical product of a function in A p, p > 0 (resp. Ap,o,
p ¢ N*), then
in Ap (resp. Ap,o).
180 2. Interpolation and the Algebras Ap
Proof. Let us assume first that I is a closed ideal in Expo(C) such that V(I) =
0. Let f E I\{O}. Multiplying f by 1 (I(z) = f(-z», if necessary, we can
assume f is an even function. The same reasoning as that of the previous
proposition, shows that if Zo is a zero of multiplicity m, then f(z)/[(z - zo)m], as
well as f(z)/[(z2 - z5)m], belong to I. Therefore we can assume that f(O) = 1,
that
2.6. The Algebra Ap of Entire Functions of Order at Most p 181
Recall that ft, 12 being jointly invertible means that if hEAp, V(h) :2 V,
then there are gl, g2 E Ap such that h = gdl + g2h.
Proof of Proposition 2.6.13. Let V = (Zk. mkh~I' then its counting function nv
is given by nv(r) = Llzklsr mk. The condition V ~ V(g) implies
nv(r) :s ng(r) = O(rP).
has the following properties. Let Z = (zkh~l' Z(II) = set of zeros of II (without
counting multiplicities), then:
(i) there are two disjoint open sets Sh S2, such that
182 2. Interpolation and the Algebras A p
Once iI has been found we conclude the proof as follows: Let rp be a Coo
function which is equal to 1 On S2, 0 On SJ, 0:::: rp :::: 1, and satisfies everywhere
l
satisfying the estimate
'u 12e-MiziP
----:--:- dm < 00
IC (1 + Iz12)2
for some M > 0, as well as the equation aujaz = (ljiI)(arpjaz). It is possible
to obtain a pointwise estimate for u from these two properties. Recall the
following lemma from [BG, Exercise 3.2.8].
2.6.14. Lemma. Let Q be an open set in C and let K be a compact subset ofQ,
there is a constant C > 0 such that for every function v of class C I in Q
v(a) = x(a)v(a) = - .
1
2m n az
1. aX v(z)
--=-(z)--dz /\ dz
z- a
1
+ -.
2m n
1. av
x(z)--=-(z)
az
dz/\dz
z- a
.
Since Iz - al :::: 8> 0 on supp(aX/az), we can estimate the first term by const.
In Ivl dm. Using that 1/(z - a) is locally integrable, we can estimate the second
integrand by sUPn lav/azl. D
We apply this lemma to n= B(O, 1), K = {OJ, v(z) = u(zo + z), for a fixed
C, and obtain
Zo E
On the other hand, for some constants A', A" > 0, we have
and
x
(1 lul2e-MlzlP d
.:........:.-----:-""7
B(zo,l) (1 + Iz12)2 m
) 1/2
::: A" eM'lzolP
with M' = M/2. It follows that lu(z)1 ::: C l e C2iziP and 12 E Ap.
Let hEAp be such that V(h);2 V. The function h(l - h) E Ap and
V(h(l - h» :2 V(fl). Therefore, there is gl E Ap such that h(1 - h) = gl/I'
Hence
(i) Re { 2: mk + 2: ~} ::: K;
IZkl:o=r zf lajl:o:r aj
(ii) 1m { 2:
IZkl:o=r
mk
zf
+ 2:
lajl:o=r
~}
aj
::: K; and
(iii) nv,(r) = O(r P ).
184 2. Interpolation and the Algebras Ap
The reason to separate the real and imaginary part of Lindelof's condition is
that for p odd we shall need to place the aj on the real and imaginary axes
respectively to satisfy those two conditions, while for p even we need to place
them On the rays e = 0, e = 7T I p, for the condition (i), and On the rays e = 7T 12p
and e = 37T 12p, for the condition (ii). We shall show in detail for the case of p
odd how to place zeros on the real axis to achieve the boundedness condition.
The other cases, being entirely analogous, are left to the reader.
The proof is a bit delicate; it takes six steps and follows the blueprint of the
proof of the Cartan-Boutroux lemma [Lev], [BG, §4.5.13].
For n ::: 1, let An = {z E C: 2n - 1 ::: Izl ::: 2n}, Bn = B(O, 2n), and denote the
points of Z n An by Vn = (bih::;j::;).n' repeated according to their multiplicities.
Therefore, for AD = n v< 1), we have
AD + Al + ... + An = nv(2n).
points of Vn •
We replace the disks Cj by concentric disks Ci of radii R(Cf) = R(Cj) +
30n. For z f/. U I <i <i Ci, the disk B (z, 2on) has distance at least On from every
Cj. Since there are nat most An disks Cj we have
L R(Cn::: 5H = 57]2n.
1::::; ::::in
can only appear as points from disks starting from points of Vn-J. Vn, or Vn+1•
The length is at most 2 x 57](2n- 1 + 2n + 2n+l) = 357]2n. This indicates that
eventually we shall need 7] < 1/70 to have enough room to maneuver.
2.6. The Algebra Ap of Entire Functions of Order at Most p 185
Third Step. Let us estimate the maximum value of IRe LI:;:j:9.n l/(bJ)PI.
Since IbJ I :::: 2n - l , we have
Re L n
An
< -(--1)-'
(b.)P -2 n - P
I:;:j:;:i.'n J
Fourth Step. We are now trying to find where to locate the new zeros. One
of the conditions is that the disk of radius 20n centered at a new zero does not
intersect any of the omitted circles or any of the other previously added disks
of radius 2on •
We shall show that the new zeros can be chosen as a subsequence of the
sequence «~j)l/p)jEZ' for a conveniently chosen constant 0 < ~ < 1. We shall
choose ~ so that dn, the smallest distance among the points in «~j)l/p)jEZ nAn,
is smaller than On, and this holds for every n E N. For that reason we need to
estimate dn • The distance between successive points is
On the other hand, since n v (r) .::: CorP, r :::: 1, we have An .::: Co2 np , and
YJ2n YJ2n YJ
0-->------,--
n- An - Co2np - Co2n(p-I)'
(iv) The distance between two of the points (aj)l:;:j:;:jn is at least 4on; and
(v)
We note that (v) implies (i) almost immediately. In fact, if r E [2 n , 2n +1[ and, by
abuse of language, S(r) denotes the sum in (i), then IS(2n) - S(r)1 .:::
C2np /2(n-l)p = C2P, while
Since d n < 8n, we can choose a positive integer f such that 48n < fdn <
88n. By choosing the (a'j)j as a subset of the sequence «~fj)l/p)jEZ n An, we
guarantee (iv). To simplify the notation, let I; = e~.
The idea about how to satisfy (v) consists in showing we can do it in the
worst possible case, e.g., Ll:;:j:;:An l/(b'j)P = -An/2(n-l)p. (The case An/2(n-l)p
is dealt with similarly, since An n «I;J)I/ p)jEZ is symmetric with respect to the
origin, and p is odd.) Let us start by proving an auxiliary lemma.
2.6.15. Lemma. = lan, bn[ 5; I = la, b[, 0 < a < b, 1:::: n :::: N,
Let In
Ll<n<N(bn - an) = L, 0 < a = inf{bn - an: 1 :::: n :::: N}, T = b P - (b - L)P,
and- W= (a + a)P - a P. Denote by (l;lk)I/ P, 1 :::: k :::: M, the points of the
sequence (l;j)l/p which belong to 1\ Ul<n<N In. Assume that no consecutive
pair of points in the sequence (l;j)l/p lies fn-a single In, then
L -I;1 ~ L --:-
1:;:k:;:M I; )
8k
1 log {b-
1 ~ "2
I; I;
I;j EJ'
P
[
(a P + ( I;
W + 1) T)/I; ] -I}
P
b
= 21;1 log { aP + (I;/W + l)T
}
.
m+1
1) ~ ~ log ( -n ) .
m
o
m:;:k:;:n
We apply the lemma to the case the omitted circles intersect the interval
[2 n- 1 , 2n] in a set of the largest possible measure. We have a = 2n- 1, b = 2n,
a~ 48n, T = 2np (1 - (1 - 351J)P) (from the second step),
and l; ::: 8onP2 n(p-l), hence l; I W + 1 ::: 2P + 1. Therefore, using all the avail-
able points (aJ:h::;k::;kn , we obtain
log({2P 1(1 + 1]')}
L
l::;k::;k n
1 1 I
(an)p::: 2l; og{
k
2P 1 '
I( + 1] )}::: 168 2n(p-lj ,
nP
with 1]' = (p2P + 1)2P(1 - (1 - 351])P). We need to obtain that this lower bound
exceeds AnI2(n-l)p = 1]2n IOn2(n-ljp, i.e., we need
log({2P 1(1 + 1]')}2n- 1 1]2n
~~--~~----- > --
16p2np - 2(n-ljp·
It is clear that, after dividing out the common factors, letting 1] -+ 0 makes the
left-hand side approach (1/32) log 2, while the right-hand side tends to zero.
Therefore, for a convenient choice of 1] > 0 we can achieve
L (n)p + Re L
1 1
(bn)p::: 0,
l::;k::;k~ ak l::;j::;An j
in the case the second term is negative, choosing all the aJ: > 0 (as we pointed
out, the situation is symmetrical and we only need to consider this case). By elim-
inating points aJ: we decrease this sum and, since each term (ak)-P ::: 2-(n-ljp,
we can stop just before the sum becomes negative. In this case we obtain
1 1
0< """' - -
- ~ (n)p
+ Re """' - - < 2-(n-l j p < 2-(n-l j
~ (bn)P -,
l::;j::;jn aj l::;j::;An j
Fifth Step. Let Wn := Ul::;k::;n Vk U {aj: 1 ::: k ::: n, 1 ::: i ::: in}, and
Pn(z):= IT (z - Wk).
WkEWn
We want to show that for z E A n- 1 U An and outside any of the omitted disks,
corresponding to k ::: n, and any of the disks B(aj, 2o k), 1 ::: k ::: n, 1 ::: i ::: in,
the polynomial Pn satisfies the inequality
points of V£. Now, if for some e, we have Iz - dl < (e/2)0k, let m = (e/2) or
(e + 1)/2, according to e being even or odd, then B(z, mOk) contains at least
the e points d, ... ,
yl, while 2m - 1 :::: e. Since (Sm/4) - 1 < 2m - 1, this is
a contradiction. Hence, the claim is correct.
Consider now the polynomial
Qk(Z) = II (z - Yl).
l:'Oi:'Oqk
As a consequence of the previous claim, outside the excluded disks it satisfies
as claimed.
Sixth Step. The estimation of Pn will help us to show that the entire function
II defined as the canonical Hadamard product with zeros in W = UI<k<oo Wk
satisfies the inequality -
log 1!l(z)1 > -H2 np
for every z E A n- I U An outside the omitted disks and all the B(aj, 20k), H is
a strictly positive constant independent of n.
The proof is essentially the same as that of the Minimum Modulus Theorem.
We can assume !I (0) = 1. We have W = {Wk: k ~ I}, IWkl :::: IWk+li,
1
We have
1 2n + 1 net) logM(2 n+ l )
Pn = n(2n):::: - - --dt:::: ,
log 2 2n t log 2
2.6. The Algebra Ap of Entire Functions of Order at Most p 189
YJ ) logM(2n + l )
log Ig(z)1 ~ ( log 8e log 2 '
outside the omitted disks and the B(aj, 28 k). If we choose YJ > 0 smaller if
necessary, we can assume that the sum of the radii of the omitted disks and the
disks B(aj,28k) which intersect An-I U An is less than 2n-2. Therefore, there
is an R, 3 x 2n - 2 < R < 2n such that the last inequality holds on the circle
Izl = R. Let us now choose () such that Ih(Re ili )I = M(h, R) = sUPlzl=R Ih(z)l.
Since M(R, f) = M(R) .::: M(2 n + I ), we have
We conclude that for z E An-I, outside the omitted circles and B(aj,28d
we have
for some CJ, C2, C3 > O. This shows that if s = e- c" the set SUI, s, C) has
the three required properties, namely, we let S2 = Uj.k B(aj, 28k) and SI its
complement.
This concludes the proof of Proposition 2.6.13. D
2.6.16. Corollary. Given V = (Zk. mkh"?1 ~ V(g), g E Ap. The necessary and
sufficient condition for V to be an interpolation variety for Ap is the existence
of fl' hEAp, and constants s, C > 0 such that V = V(fl, h) and
o~ j ~ mk - 1,
II(Yk,j)11 = SUp{IYk,jle-CP(zKl}.
k,j
It follows that p(Un/e) nD = D. Hence there are entire functions !t.j such that
l!t.j(z)1 ~ ~enP(Z),
e
2.6. The Algebra Ap of Entire Functions of Order at Most p 191
°
h,j(Ze) = {exP(Cp(Zk)) if 1= k and i = j,
j! otherwise.
The functions fk.j := exp(-Cp(Zk))h,j satisfy the conditions (i), (ii), and (iii)
of the lemma. 0
2.6.20. Lemma. If V = V(/!, ... , fn), with /!, ... , fn jointly invertible in
°
Ap(iC), and V = (Zb mkh~1 is a universal interpolation variety for Ap(iC), then
there is a constant Co > such that
L e-COP(Zk) < 00.
°
k~1
Let dj := inf{lzk - zjl: k =j:. j}, dj := min{1, dj /2}, and Bj = B(zj, dj ). These
disks are disjoint, hence
j?:.l.
On the other hand, there are Ao, Bo > °
such that for any
AOP(Zj) + Bo. Therefore, for some, 81,82 > 0,
Z E Bj , p(z) :::
Hence
o
°
j~1
with the functions obtained in Lemma 2.6.12 for a value C > to be chosen
below.
If this series were uniformly convergent over compact sets and F E Ap(C),
then we could differentiate it term by term and evaluate it at Zk. We would then
192 2. Interpolation and the Algebras A p
obtain
F(mk) (Zk) !k.mt- 1(Zk)
---- -1
mk! - (mk - I)! - .
Let us prove the convergence of the series and estimate F.
Since we are in the conditions of Theorem 2.2.10, it follows that the property
(*) given there holds. In the course of the proof that the property (*) just
mentioned implies that V is an interpolation variety, we proved that there are
constants E, F > 0 such that
for every k ~ 1.
Using that estimate and the fact that from the properties of p we have for some
L, K >0,
Izl ::: L exp(Kp(z)) ,
we obtain
IF(z)1 ::: 2A exp«B + K)p(z)) Lexp(-(C - E - K)P(Zk)).
k~l
2.6.22. Proposition. Let V = (Zko mkh~l = V(/J, ... , fn), where /1. ... , fn
are jointly invertible in Ap(C). Assume V is an interpolation variety for Ap(C)
(or equivalently, a universal interpolation variety), then there are constants
C, D > 0 such that for every k ~ 1 we have
(1) io
rk n(Zk, t, V)
t
dt ::::: Cp(zd + D,
Cp(Zk) + D
(2) mk< .
- logrk
Proof. Let F be the function obtained in Proposition 2.6.18. For every k ~ 1
we have
Since F E Ap(C) and p(z) ::::: K1P(Zk) + K2 for Z E B(Zko rk), we obtain
1::::: Aexp(Bp(zd)/r;'k
for some A > 0, B > O. Taking logarithms we obtain (2).
Let us now apply Jensen's formula to the function f(z) = F(z)/(z - zd mk •
We have for f(O) = 1, hence
io
rk n(z t F)
ko ,
t
dt = -1 1211' log IF(Zk + rke,(i)
21l'o
. IdB - mk log rk.
In the case of the algebras A p , one can obtain now clear geometric condi-
tions to decide when a multiplicity variety is an interpolation variety. Note that
these are the conditions that exclude the examples of the type occurring in
Exercises 2.2.12 and 2.2.13.
(1')
(2')
Proof. From Proposition 2.6.13 we know that V = V(/J, h), /J, h jointly
invertible in Ap. The proof of the necessity is a verbatim copy of the proof of
Proposition 2.6.22.
To prove the sufficiency of the conditions (1'), (2') when V = V(g), we need
to show that for some 8 > 0, A > 0, one has
°
We can apply the minimum modulus theorem ([BG, §4.5.14], [Lev, p. 21]) to
obtain constants 81 > 0, Al > (independent of k) such that for Z E B(O, 21zkl)
and outside a finite collection of exceptional disks, Bk.l, ... ,Bk •jk the sum of
whose radii does not exceed ~IZkl, one has
Ig(z)1 ~ 8Ie-AlizkIP.
It follows that there must exist a value r, !IZkl < r < IZkl, such that
aB(Zb r) n Bk,j = 0, 1 ::: j ::: jk.
1 1
Apply now Jensen's formula, [BG, §4.4.30] to g in the disk B(Zb r). One obtains
g(mkl(Zk) r n(V, Zb t) 1 2". .
log I 1+ mk logr + dt = - log Ig(Zk + re,e)1 dO.
mk! 0 t 2rr 0
From this identity, (1'), (2'), and the choice of r, it is clear that
g(mkl (Zk)
log I I ~ -AlzklP - B
mk!
for some A, B > 0. o
2.6.24. Remark. [Sq2]. For the weight p(z) = I Imzl + 10g(1 + IZI)' and a
multiplicity variety of the form V = V(h), given by h slowly decreasing for
Ap(C), one has that V is a (universal) interpolation variety if and only if
1 o
P(Zkl n(V, Zk. t)
- - - d t : : : Cp(zd
t
+D
and
Cp(Zk) D
mk < - " - - - -
+
- log P(Zk)
for some C, D > 0, independent of k.
sufficient conditions do not coincide, though they are close to each other. This
is a phenomenon characteristic of functions of infinite order of growth [BL].
In [GR] similar conditions for spaces of functions with prescribed indicator of
growth have been found.
Interpolation in the space of functions of infraexponential type has very
interesting applications to the theory of Dirichlet series, as we shall see in
Chapter 6. A classical theorem of P6lya - Levinson [Levs, Theorem XXXI] states
the following:
Let V = {zdn:::! be a sequence of nonzero complex numbers such that
Rez n > 0 , · -n = 0,
11m ll'm 1 1m Zn 1 = 0
Zn
n-HXl n--->oo IZn 1
and there exists c > 0 such that
Note that (t) implies that points Zn are distinct. One can prove that the main
point (t) of the statement is exactly the condition that V is an interpolation
variety for functions of infraexponential type. In [Vi2] it has been proved that the
conditions on Re Zn > 0 and 11m Zn I/lzn 1 -+ 0 are not necessary. In fact, given
any sequence V = {znln:::!, n(r) = o(r), and {znln:::! satisfies V, then there is a
function 1 of infraexponential type such that
(i) V ~ V(f);
(ii) all zeros of f are simple; and
(iii) V (f) is an interpolating variety in the space of functions of infraexponen-
tial type. (The condition (tt) also holds when dist(z, V(f» 2: c/8.) We
provide a proof of this result in Chapter 6.
It is possible also to find necessary and sufficient geometric conditions for a
variety to be interpolating in space of type zero for any radial weight satisfying
the doubling condition [BLV].
EXERCISES 2.6.
1. Let 1 be an entire function, 1(0) = 1. For r > 0, let ai, ... , an denote the zeros
of 1 in B(O, r) repeated according to multiplicity.
(a) Show that
-1
21l'
1°
23r
e-'Pu
'n
log I/(re'U)1
'n
dB
involving Sp(r) = E1otl::o, a;p. Use this expression to show that Sp remains bounded as
r --+ 00, whenever lEAp. (Hint: Divide out the zeros of I to obtain a nonvanishing
holomorphic function g in B(O, r). All this exercise requires is to compute the Fourier
coefficients of the harmonic function Re(log g) (see also [BG, Exercise 4.6.9]).
2. Let E p , Eoo be respectively the spaces of entire functions of order at most p
(resp. finite order). For 0 < p < 00, n E N* consider the Banach spaces Ep.n of all entire
functions such that
II/lIn = sup(l/(z)1 exp( -lzIP+I/n» < 00
(a) Show that Ep = nn>1 Ep.n, Eoo = Un>1 Fn. Ep is an FS space and is a DFS space.
(Hint: Use Proposition 1.4.8.) -
(b) Show that every closed ideal in Ep (resp. Eoo) is principal. (Hint: Find g such that
V(g) = V(I).)
3. For 0 < p < 00, n E N*, let Ap,n be the Banach space of all entire functions
such that
II/lIn = sup (I/(Z)I exp ( -~IZIP)) < 00.
(a) Assume lEAp, show that e- Q 1/ PR E Ap, and that for every g E Ap one has
. e- QI .
lim - - g = g III Ap.
R~oo PR
(b) Let I be a closed proper ideal in Ap, V = V(l), I E I\{O}, g E I(V). Write PR
as a product PR =
P~P; such that yep;) =
V n B(O, R). Show that fI P~ E I. Use this
to prove that gEl.
(c) Prove parts (a) and (b) in the case Ap,o, p fj. I'll, and also when p E N* but p =
p-1.
(d) Let I E Ap,o, P = p, Q(z) = cxozP + QI (z), deg QI :::: p - 1. Show that
and
lim -I exp
R~oo PR
( -QI + -zP
p
L
IZnl:o R
z;/ ) g = g
holds in Ap,o.
Use (c) and (d) to prove that all ideals in Ap,o are localizable.
7, Let V = V(fl" .. , In), II, ... , In jointly invertible in Ap(C). Show that V is a
universal interpolating variety if and only if V is an interpolating variety.
8. Let p(z) = IzI P , r > 0, kl > 1, k2 > 0, compute the function r(z) > 0 that has the
property: pel;) :::: kIP(z) + k2 if and only if l; E B(z, r(z)).
9. Suppose V = (Zk> mkh,,1 satisfies condition (1) of Proposition 2.6.22, i.e., for some
C, D >0
i o
rk n(z
k>,
t
t V)
dt :::: C(P(Zk) + I) (k :::: 1).
Show that there are e > 0, A > 0 such that for any j f:. k
IZk - Zj I ~ ee-Ap(zkl.
10. (a) Use Proposition 2.6.23 to show that the lattice Il + ill is interpolating for Ap
if p = 2.
(b) Give a direct proof of the same result by using the Weierstrass a function [Mark].
CHAPTER 3
Exponential Polynomials
where aj E C, Pj E C[z]. We assume that the aj are distinct and the polynomials
Pj not zero. The Pj are called the coefficients of f and aj the frequencies.
(Sometimes the aj are called the exponents, especially in the Russian literature.
In some contexts aj = i)'j, Aj E JR, and the Aj are called the frequencies and
Tj = 27r/Aj (when Aj =j:. 0) the periods; clearly eiAjz is periodic of period Tj.) It
is immediate that there is a unique analytic functional T whose Fourier-Borel
transform 'J(T) coincides with f, i.e., 'J(T)(z) = (T~, eZ~) = fez). Namely, T
is representable as a distribution in C of the form
T = ~a· o(v)
~ j,V Uj'
j.V
where OUj is the Dirac measure at the point aj, o~~) is a "holomorphic" derivative
of order v, o~~) = (a/azYOUj' and the aj,v are complex constants. In the case
where the frequencies aj are purely imaginary, i.e., aj = iAj, Aj E JR, then f is
the Fourier transform of a distribution /L E £'(JR). That is, we let
dV
/L := ~ aJ· viv -d 01.
L...J, xv'
j,v
with OAj the Dirac mass at the point Aj E JR (acting on Coo functions in JR) and
198
3.1. The Ring of Exponential Polynomials 199
3.1.1. Proposition. Let fez) = LO::;j::;m Pj(z)e ajZ , g(z) = LO::;k::;n Qk(z)e fikZ be
two exponential polynomials of the form (*). If g divides f in E, then the frequen-
cies f31, ... , f3n of g are linear combinations with rational coefficients of the
frequencies al,···, am of f.
Proof. Since g divides f in E there is h E E*, h (z) = Lo::;/::;p R/e Y/ z , such that
f = hg. Since we are assuming Yo ~ YI ~ ... ~ YP' it is immediate that Yo = O.
If P = 0, then the frequencies of f and g are the same, and the coefficients are
related by Pj = RoQj. We can assume therefore that p ::::: 1.
Let 0 = Wo ~ WI ~ ... ~ Wr represent all the frequencies of the form f3k + Yl,
O:s k :s n, 0 :s I :s p. Clearly lao, ... , am} S; {wo, ... , wr }.
In order to proceed we need an elementary, albeit important, lemma.
3.1.2. Lemma. Let F(z) = LI::;j::;n ({Jj(z)e AjZ be a nonzero exponential polyno-
mial with frequencies Al ~ A2 ~ ... ~ An, and coefficients ({Jj(z) = ajzm j + ... ,
polynomials of degree mj ::::: 0 (aj E C*). For every j, 1 ::; j ::; n, we have
Hence
d )m+1
( dz - A (P(z)e AZ ) = O.
L Qk(z)R/(z) = { ~j(z) if Wi
if Wi
= aj,
f/. {ao, ... ,an}.
k,/
fh+y/=Wi
Continuing this way we find a unique f3kl denoted f3*, such that
f3* = L Uiei,
O:'::i:"':l
whose successive coefficients Ui are maximal.
Proceeding in the same way with YI, 0 ::: t ::: P, we find y* = Yk"
y* = L Viei,
O:'::i:"':l
whose successive coefficients are maximal. Here Vo ?: 0 since Yo = o.
Let w* := f3* + y* = LO<i<1 Wiei. Then Wo = Uo + Vo ?: 1. Assume w* =
f3k + Yl for some pair k, t. The successive coefficients of f3k and Yl must be
maximal, otherwise they could not add up to the coefficients of w*. Therefore,
f3k = f3* and Yl = y*. Hence the sum
since Pkl =I- 0, Rli =I- 0.1t follows that w* E {ao, ... ,an}. This means that Wi = 0
for i fj. {1, ... , q}. In particular, Wo = O. This is a contradiction. We conclude
that every f3k is in V, and the statement of Proposition 3.1.1 is correct. 0
In fact, let fez) = 1 + aleOlIZ + ... + ameOlmZ , 0 --< al --< •.• --< am, be simple.
Since Re ai ?: 0 for all i, there is 'A E C, Re'A ?: 0, such that
ai =qi'A,
qi E Q, qi > O. Let N be the common denominator of all the qi, then we can
write qi = Pi/N, Pi E N*, and
ai = PilL, IL = 'A/N, 0< PI < ... < Pm.
202 3. Exponential Polynomials
3.1.6. Lemma. Let 0 -< al -< ... -< am and P := dimlQ V, with V = EI::;j::;mlQaj,
the IQ-vector space spanned by al, ... ,am. Let IQ+ = IQ n ]0,00[. One can find a
basis /LI, ... ,/Lp olV such thataj E EI::;k::;plQ+/Lko I::: j::: m.
Aj = L ajlml,
1::;/::;p
Since we want tlk approximately equal to Reml, then the rational numbers
bjk := EI::;/::;p ajltlk are approximately equal to Re(EI::;/::;p ajlml) = Re Aj > O.
In other words, if we choose tlk sufficiently close to RemJ, I ::: k ::: p, we can
guarantee that bjk E IQ+ and, clearly, this can be done so that simultaneously
the determinant det(tlk>I,k =f:. O.
3.1. The Ring of Exponential Polynomials 203
We now let f1k = ei8 Mb 1 ~ k ~ p. They form a basis for V and (Xj E
L I :sk:sp Q+ f1k· 0
We augment the collection f11, ... , f1p to a basis f11, ... , f1q of the Q-vector
space generated by (XI,.··, (Xm, f31, ... , f3n, YI,···, Yr' Among Yo, ... , Yr we
find y* whose successive coefficients with respect to f11, ... , f1q are minimal.
Let
y* = Vlf11 + ... + vqf1q,
3.1.8. Remark. It follows from Proposition 3.1.7 that the Yj are also Q-linear
combinations of the f1k with nonnegative coefficients.
We shall accept the following result of Ritt about factorization in the ring Eo.
The proof is entirely algebraic and rather long (see [Ril], [Go], [Schi]).
204 3. Exponential Polynomials
(3) Every nonzero exponential sum f can be written in a unique way in the
form
with ae az E U(E o), ai, ... , as normalized, simple, sharing no common nonzero
frequencies, and 7rt. ... ,7rr , normalized irreducible exponential sums.
is a ring isomorphism. o
3.1.13. Lemma. Let I, g E Eo, then there is a g.c.d. in Eo.
Proof. As a consequence of Theorem 3.1.9 we have to consider the following
three cases:
(i) I and g are simple;
(ii) I and g are irreducible;
(iii) I is simple and g is irreducible.
206 3. Exponential Polynomials
Case (i). Let I(z) = 1 + alea1Z + ... + ame amz , 0 -< al -< ... -< am, ai =
PiA, Pi E N*, and g(z) = 1 +bleP1Z + .. , +bne Pnz , 0 -< b l -< ... -< bn, bj =
qjf.J" qj E N*.
We have two possibilities: either al I {31 E Q or not. The first case is equivalent
to ai I {3j E Q and to AI f.J, E Q. Thus, in this case, there is v E C* such that
ai = ri v, ri, Sj E N*, 1~ i ~ m, 1~ j ~ n.
Moreover, we have rl < r2 < ... < rm and SI < '" < Sn. Let l; = e Vz , and
consider the two polynomials in l;
F(l;) := 1 + all;r1 + ... + am l; rm ,
G(l;) := 1 + bll;sl + ... + bnl;Sn.
Let H be the g.c.d. of F and G in ((:[l;]. It is clear that h(z) = H(e VZ ) divides
I and g in Eo. Furthermore, h is normalized and simple. The Bezout identity
H=FU+GV
for some U, V E ((:[l;] shows that
h = ul + vg,
where u(z) := U(e VZ ) and v(z) := V(e VZ ) belong to Eo. Hence if ii E Eo divides
I and g, it follows ii divides h. Therefore, in this case, h is the g.c.d. of I
and g.
We claim that if AI f.J, It' Q, then 1 is the g.c.d. of I and g. In fact, if h
divides both I and g, h is normalized and not I, then by Proposition 3.1.7
all the frequencies of h are integral multiples of A, hence h is simple. By the
same proposition, all the frequencies of h must also be multiples of f.J,. Hence
AI f.J, E Q, which is impossible.
Case (ii). If I and g are irreducible, then either 1= ag, a E U(Eo), in which
case I = g (if they are normalized), or 1 is the g.c.d. of I and g.
Case (iii). If I is simple and g is irreducible, the g.c.d. must be g or 1. If g
divides I, then it is simple by Proposition 3.1.7. Accordingly, the g.c.d. is 1.
This concludes the verification that Eo has the property (~). 0
Proof. Let fez) = LO:::;j:::;1 jj(z)zj, jj E Eo, and lex) = LO:::;j:::;1 jj(z)xj, the
corresponding polynomial in Eo[X]. Then, there are unique polynomials
ii"J, ... , iir, primitive and irreducible in Eo[X], and nl, ... , nr E N* such that
I = c(f)ii~1 ... ii:'.
Let 7rj be the exponential polynomials in E that correspond to iij via the
isomorphism Eo[X] ::::: E. Since this is a ring isomorphism, it follows that 7rj
are irreducible in E. This proves the factorization. The uniqueness follows the
same way. 0
Proof. We apply Ritt's Theorem 3.1.9 to c(f) and Lemma 3.1.13. The isomor-
phism E ~ Eo[X] allows us to conclude the factorization is unique. 0
3.1.18. Theorem. The ring E also satisfies the condition (Ll): any two exponen-
tial polynomials have a g.c.d. in E.
We now use Lemma 3.1.2 to obtain some inequalities about exponential poly-
nomials. We first note that if f(z) = Ll:::::i;'ON Pj(z)e AjZ , Al -< A2 -< ... -< AN,
Pj a nonzero polynomial of degree mj ::: 0, then these coefficients are uniquely
determined cy the function f and, in particular, we can define the degree of f,
dOf := ml + ... + mN + N - 1. We can also consider the supporting function
H of the convex hull K ofthe points Xl, ... , XN , conjugates of the frequencies,
namely,
,
H(z) := max Re(Ajz) = m!lx(Xjlz),
,
where (,1,) denotes the Euclidean inner product in ~.2. This function is also
uniquely determined by f. Recall that M(f, z, r) = max{lf(s)l: IS - zl s r}.
3.1.19. Proposition. Let f(z) = Ll;'Oj;'ON Pj(z)e AjZ E E*. For any ro > 0, there
is a constant C > 0 such thatfor every z E C and every 0 < r S ro, the inequality
holds.
Cj,l E C. Then
eAjZ = ~ L
c, O;'Ol;'OdOf
Cj,d(l)(z) = L l!c!,l
O;'OI;'OdOf27rIC,
.1 1';-zl=r
(s - z)dOf-1 !;S10/+l d S'
(s)
This shows that given ro > 0 there is a constant K > 0 such that
ro)dOf z, r)
eH(z) :::: K ( -
r
max If(s)1
Iz-';I=r
= C M(f,
r
dOf . o
eAjZ = C.
1
L Cj,J/(l) (z),
j O~I~dOf
hence
leAjZI :::~? ICj,t/Cjl L If(I)(z)l,
j. O~I~dOf
3.1.23. Remark. If we replace d(z, V) by d(z, V) = inf(1, d(z, V», then one
can take JL = 0 in Theorem 3.1.22 when I E Eo, at least when V #- 0. When
V = 0, I(z) = aeJ..z, a E C*, then d(z, V) = 00 but d(z, V) = 1, v = 0, JL = 0,
=
H(z) Re(J..z) and the inequality is trivially satisfied.
3.1. The Ring of Exponential Polynomials 211
From previous remarks we see that we have a rather precise knowledge of all
the constants C, f.1" v in the Lojasiewicz inequality. It is rather annoying that f.1,
could be positive. We shall show in what follows that one can take JL = 0 (after
replacing d(z, V) by d(z, V)), but at the cost of losing track of the nature of
C. This is an argument due to Grudzinski [Grudl].
Let us consider a vector a = (aI, ... ,aN) E Iff such that al -< a2 -<
... -< aN, m = (ml, ... , mN) E NN. Set Iml = ml + ... + mN, f.1, = Iml + N-
1, lIall = maxI:5j:5N lajl. Let E(a, m) be the vector space of all exponential-
polynomials of the form I(z) = EI:5j:5N Pj(z)eCXjZ, deg Pj :s mj, 1 :s j :s N,
where we allow some of the Pj to be zero. Denote Pj (z) = EO<k<m. - - ) Pj.kzk
and 11/11 := Ej,k IPj,kl. It is clear that II!II is a norm in the finite-dimensional
vector space E(a, m). We also denote aj = aj(f), the leading term of Pj , with
the possible choice aj = 0 if Pj == O.
Clearly, dOl :s f.1, when I E E(a, m), hence the multiplicity of any zero of I
is bounded by f.1" unless I == O.
3.1.24. Lemma. There is a constant c = c(a, m) > 0 such that lor any IE
E(a, m) and any r, 0 < r 1, :s
M(f, 0, r) 2: cr lL II !II.
Proof. We have
(M(f, 0, r»2 2: _1
2n Jo
r 27r
I/(re ili )1 2 de = L
I~O
11(1)(0)
I!
12 r21
2: L I(~!(O) (L
O:5/:5IL
1 12 r21 2:
O:5/:5IL
1 I(~!(O) 12) r2IL.
On the other hand, the map I f-+ (LO:5 / :5IL 1I(I)(0)/1!12)1/2 clearly defines a
seminorm in E(a, m), but the previous observation about the multiplicity of
zeros for functions in E(a, m), shows that it is really a norm. Since any two
norms in E(a, m) are equivalent, there is a constant c = c(a, m) > 0 such that
is also a nonn in E(a, m). Moreover, the previous proof yields for r ~ 1
-I
1/2
fCl) (0) 2)
M(f, 0, r) ~ IL~J-l-!
(
rlL,
3.1.27. Corollary. For any r E ]0, 1], I E E(a, m), and z E C, we have
M(f, z, r) ~ crlL max {laj(f)leReajZ} ~ c(f)rlLeH(z),
l~j~N
are the roots of g, then for k E N* the Newton sums of the roots are given by
L k
zl =-
1 1, Zkg'(Z)
--dz,
0::;19 2rri Izl=p g(z)
and the coefficients of Qg are fixed polynomials in the Newton sums. Therefore,
one can find a fixed number K > 0 such that
In order to estimate this last quantity, let us prove the following simple lemma:
3.1.28. Lemma. There is a constant Cv > 0 such that for any monic polynomial
Q of degree v and any p > 0, there exists pi E [p /2, p] for which
IQ(z)1 :::: Cvpv if Izl = p'.
Proof. (This lemma is also a consequence of the Cartan - Boutroux lemma
[BG, §4.5.13].) Decompose the annulus p/2 :::: Izl :::: p into v + 1 concentric
annuli using circles such that the difference of the successive radii is exactly
p/2(v + 1). One of these annuli contains no roots of Q. Let pi be the radius
of the median circle of one such annuli. Let Q(z) = (z - ZI) .•. (z - zv). For
Izl = pi we have
3.1.29. Proposition ([Grud1]). Let f E E(Ot, m), 0 < r :s 1, and let K(f) =
cominl~j~N /aj(f)/. Then, for any z E C and any holomorphic function g in
a neighborhood of 8(z, r), we have
M(fg, z, r) 2: K(f)riLeH(z)/g(z)/.
Proof. Let h = Lz(g), then h(O) = g(z), and
We shall now show that the exponential polynomials are not only slowly
decreasing in the sense of the previous chapter, but that one can give a very
precise description of the size of the components of S(f, B, C) (see Defini-
tion 2.2.13).
S(f, 81, C 1) = {Z E C: If(z)1 < 81e-Cdzl} one has IZ1 - z21 < ce- Clzd for every
Zl,Z2 E o.
Proof. We can assume 8 S 1. Fix z E 0 and let 0 < r S 1, to be chosen later
(depending on z). Consider l; E aB(z, r) such that If(nl = M(f, z, r). As
pointed out after the proof of Lemma 4.2.11, the Minimum Modulus Theorem
yields p E ]~r, 2r[ such that for every w, Iw -l;1 = p, one has
l oIf(w)1
g - - > -8 log M ( -f f ,l;,4er ) .
If(l;)1 (n
Hence,
inf If(w)l> If(l;)1 9 •
Iw-(I=p -M(f,l;,4er)8
We know that for some C > 0, IL > 0,
If(nl = M(f, z, r) 2: CriLeH(Z),
and, for some D > 0,
M(f, l;, 4er) S M(f, z, Ser) S D(1 + IzI)MeH(z).
Therefore,
ErveH(z)
inf If(w)l> ------,:-
Iw-(I=p - (1 + Izl)N
for some E > 0, N 2: 0, v > o.
Let us now choose r = r(z) = ce- Clz1 /6e c . We are going to choose 81 > 0,
C 1 > 0 so that for any z E C one has
Er(zYeH(z)
8 e-Clizi < ---...,..,-
1 - (1 + Izl)N .
To verify that this choice is possible, let us observe that there is ex > 0 such
that for every z E C
H(z) 2: -alzl.
3.1.32. Remark. In the case where all the frequencies of f are purely imaginary,
then H(z) ::: -al Imzl for some a > 0 and we can replace Izl throughout the
proof by p(z) = log(1 + Iz12) + I Imzl. In this case
SU, EJ, C1) = (z E C: If(z)1 < Ele-C1P(z)}
EXERCISES 3.1.
Conclude that f is of completely regular growth and that if /-L E .;tf' (C) is such that
=
;;(/-L) f, then /-L*: .;tf(Q + K) --+ .;tf(Q) is surjective for any open and convex subset
Q ofC.
°
What is the corresponding statement for p(z) = IzIP, P > I? Could V be an interpolation
variety when < p < 1 if f has more than one frequency?
4. Let f be an exponential polynomial, V = V(f) = (Zb mkk!lo P::: 1. Show that
°
the following two statements are equivalent:
(i) There is A > such that IZk - Zj I ::: e-A(lzkl+lzjl)P for every pair (k, j), j "# k.
(ii) There is one e, B > 0 such that for every k, If(mk)(Zk)1 ::: ee-BlzkIP.
Given P > 1 find an exponential polynomial f satisfying (ii) for this value p but for
no p' < p.
5. Let f be an exponential polynomial with purely imaginary frequencies, and p(z) =
I Irnzi + log(1 + Izi), V = V(f) = (Zb mk)k;::l. Show that V is an interpolation variety
3.2. Distributions of Zeros of an Exponential Polynomial 217
for Ap(C) if and only if there are e > 0, A > 0, such that
for every pair (k, j), k # j.
Equivalently, show this condition can be replaced by the existence of /) > 0, B > 0, such
that for every k
If(mk)(zk)1 ::: /)e-Bp(zkl.
Let 8 denote the ray starting at 0 and making an angle cp + JT /2 with the
positive real axis. By T (e) we denote the angular sector of width 2e and having
8 as a bisectrix. (See Figures 3.1 and 3.2.)
If /-t = 0, V (cp, 0, H) is a half-strip limited by two half-lines parallel to 8 and
equidistant from 8 at a distance H. In this case, it is a straight half-strip.
If /-t i= 0, the two unbounded portions of the boundary are curves asymptotic
to lines paralled to 8. We then say that V(cp, /-t, H) is a logarithmic half-strip.
Let us choose argz E [0, 2JT[, then
JT
lim argz = cp + -,
Izl~oo 2
zeV (rp.ll. H)
which means that, for every e > 0, there is an r > 0 such that
V(cp, /-t, H) n B(O, r)C S; T(e).
3.2.1. Lemma. The intersection o/two half-strips V (CPI, /-tl, HI) n V (CP2, /-t2, H2)
is an unbounded set if and only if CPI = CP2 and /-tl = /-t2.
218 3. Exponential Polynomials
Figure 3.1
T(9)
Figure 3.2
3.2. Distributions of Zeros of an Exponential Polynomial 219
Proof. From the previous observation, if V = V(fPI, ILl, HI) n V(fP2, IL2, H2) is
unbounded, then
7r 7r
lim arg z=fPI+- 2'
2 =fP2+-
ZEV
Izl->oo
Moreover,
Re(ze- icp !) . Re(ze-CP2)
ILl = lim = hm = IL2· o
ZEV log Izl ZEV log Izl
Izl->oo Izl->oo
We shall now prove a very famous result of P6lya concerning the distribution
of zeros of exponential polynomials. Following the work of Dickson [Dil] we
shall give simultaneously the result for a larger class of functions, those that are
asymptotically exponential polynomials (AEP) (see also [RoD.
3.2.2. Definition. We say that 1 E AEP if there is a compact set K such that
1 E .rf(C\K) and has the form
I(z) = L Ajzmj (1 + 8j(z))e WjZ ,
I:::;j:::;n
Note that any exponential polynomial with more than one frequency is in
AEP. It is enough to write the polynomial coefficients Pj as
Pj(z) = ajzmj(l + Cj(z)),
where mj = deg P, 8j(Z) = Pj(z)/ajZmj - 1, which is holomorphic for Izl » 1.
respectively, then 0 ::: 1/11 < 1/12 < ... < 1/Iu < 21l'. Note that for z E hk one has
Re(wkz) = Re(wk+lz) for 1 ::: k ::: u - 1 (with the obvious extension to the case
k = u).
Let
SI := {z E C: 1/Iu ::: argz ::: 1/11 + 21l'}
and, for k = 2, ... , u,
Sk := {z E C: 1/Ik-1 ::: argz ::: 1/Ik}.
For a fixed 9 verifying 0 < 29 < min{ 1/11 - 1/1u + 21l', 1/Ik - 1/Ik-1 (k = 2, ... , u)},
let us define
SI := {z E C: 1/Iu + 9::: argz ::: 1/11 + 21l' - 9},
Sk := {z E C: 1/Ik-1 + 9 ::: arg z ::: 1/Ik - 9} (2::: k ::: u).
Note that because n ::: 2 one has 0 < 9 < 1l'/2. Finally, let us denote
Tk := {z E C: 1/Ik - 9 ::: arg z ::: 1/Ik + 9} (1 ::: k ::: u).
(See Figure 3.3.)
Figure 3.3
3.2. Distributions of Zeros of an Exponential Polynomial 221
3.2.4. Lemma.
(a) Ifz E Sk, l; E P, then Re«wk -l;)z) ~ O.
(b) Ifz E Sk, l; E P, then Re«wk -l;)z) ~ Izllwk -l;1 sin8.
(c) If z E Sk n Tk and Wj fj. Lb then there is a value 80 > 0 such that
Re«wk - Wj)z) ~ Izllwk - wjl sin 80.
(d) If z E Sk+1 n Tk and Wj fj. Lk. then there is a value 81 > 0 such that
Re«wk+1 - Wj)z) ~ Izllwk+1 - wjl sin81.
Proof. Recall that (·1·) denotes the Euclidean scalar product on ]R2 and that for
any a, bEe, Reab = (alb) = (alb).
We shall give the proof of the lemma for 2 ~ k ~ a-I, leaving it to the
reader to adapt it to the cases k = 1, k = a.
Part (a). We observe that if h-I denotes the ray perpendicular to h k- I, chosen
so that the angle from h-I to hk-I is rr /2, and h, the ray perpendicular to hb so
that the angle from hk to h is rr /2, then the sector Sk is complementary to the
two quadrants determined by h-I and h k- I and by hk and h. The point ~ E P
and, therefore, ~ - Wk belongs to the closed sector determined by It-I and It.
This is just the sector opposite to Sk. Hence, the angle between any z E Sk and
~ - Wk is at least rr /2. In other words,
Re(z(l; - wd) = (zl~ - Wk) ~ O.
This is equivalent to statement (a).
Part (b). Here the angle between z E Sk and ~ - Wk is ~ rr /2 + 8, hence
(zl~ - Wk) ~ -Izll~ - wkl sin8.
Part (c). Since Wj fj. Lk, one can immediately see there is a 8' > 0 such that
rr, _ _ 3rr
1frk + "2 + () :5 arg(wj - wd :5 1frk-1 + 2'
On the other hand, z E Sk n Tb so that
1/Ik-1 + () < 1/Ik - 8 ~ arg z ~ 1frk.
Hence, the angle between z and Wj - Wk lies between rr /2 + 8' and 3rr /2 - 8.
Letting 80 = min{8, 8'}, we obtain (c).
Part (d). This time there is 8" > 0 such that
rr _ _ rr"
+ "2 + "2 -
,I,
'f'k+1 ~ arg(wj - Wk+l) ~ 1/Ik 8 ,
and
1/Ik ~ argz ~ 1/Ik +8 < 1/Ik+1 - 8.
Choosing 8 1 = min{8, 8"} we can proceed as before. D
3.2.5. Lemma.
(a) Jfz E Sk and j =j:. k, then zmjeWjZ = e(z)zmkeWkZ.
(b) Ifz E Tk n Sk and Wj fj. Lb then zmjeWjZ = e(z)zmkeWkZ.
222 3. Exponential Polynomials
°
3.2.7. Corollary. Under the same conditions of the previous lemma, there is
r > such that
We remark that this lemma shows that all the zeros of f, of large absolute
value, must lie within some sector Tk. The condition on T is just < T <
infk IAk1/2. On the other hand, the definition of Sk depends on e, hence the
°
value ro of this lemma depends on e. We proceed to study what happens in the
sector Tk •
3.2.9. Lemma. Let Jk be the set of indices j such that Wj ELk. For z E Tko we
have
f(z) = LAj zmj(1 +8j(z))eWjZ ,
jeJk
3.2. Distributions of Zeros of an Exponential Polynomial 223
with
lim Ej(Z) = O.
zeTk
z .... oo
Proof. This is a corollary of Lemma 3.2.5(b) and (c), the functions Ej are the
same as in (*) for j =j:. k, k + 1, those with these two indices will change. 0
with otk,l = Re Wk,1 = Re Wk,I. Note that with our conventions otk.! > otk,2 >
... > otk,ak' Due to the convexity of Qb these slopes satisfy
-00 < ILk,1 < ILk,2 < ... < ILk,a-1 < 00.
Figure 3.4
3.2.10. Lemma. There is v > 0 such that for every Tk,p ¢ Lk,l then
mk,p - mk,l - J.Lk,l(ak,p - ak,l) < -v,
mk,p - mk,l+1 - J.Lk,l(ak,p - ak,l+d < -v.
Proof. Because of our choice of coordinates it is natural to consider variablt:s
a, m, so that the a-axis is the real axis. In these variables, the line Lk,l can be
represented by either of the two equations
m - mk,l - J.Lk,1 (a - ak,l) = 0,
In these coordinates, the point Tk,p is given by (ak,p, mk,p)' Since Qk is convex
and Tk,p ¢ Lk,l, this point lies below the line Lk,l, hence it satisfies the inequal-
ities
mk,p - mk,l - J.Lk,/(ak,p - ak,/) < 0,
mk,p - mk,l+1 - J.Lk,l(ak,p - ak,l+d < O.
Since there are only finitely many possible Tk,p, the existence of v > 0 follows.
o
(a) If j = 1, assume Tk,p ¢ Lk,}; if2 :::: j :::: ak - 1, assume Tk,p ¢ Lk,} U Lk,}-I;
and if j = at. assume Tk,p ¢ L k,}_" then we have
with
lim s(z) = O.
z.... oo
ZEUk,j
3.2. Distributions of Zeros of an Exponential Polynomial 225
(b) If 1 :::; j :::; Uk - 1 and 'Ck,p E Lk,j, or if2 :::; j :::; Uk and 'Ck,p E Lk,j-J. then
Izmk. p e Wk .pZ 1 :::; Izmk,j eWk,jZ le-Hlwk,p-wk,j I.
Proof. (a) If j = 1, ... ,Uk - 1, there is a It < Itk,j (It depends on z) such
that x + It log Izl = O. If j = 2, ... , Uk. there is a It > Itk,j-l with the same
property. (Just use the definition of Uk,j') Since Im(Wk,p - Wk,j) = 0 and
Re(wk,p - Wk,j) = Ci.k,p - Ci.k,j,
= Iz Imk,p-mk,j-/L(ak,p-ak,j).
Consider the case when Ci.k,p - Ci.k,j < O. Then definitely j i= Uk. hence 'Ck,p ¢
Lk,j, and It < Itk,j' The previous lemma implies that
mk,p - mk,j - 1t(Ci.k,p - Ci.k,j) < mk,p - mk,j - Itk,j(Ci.k,p - Ci.k,j) < -v
for some v > O.
In case Ci.k,p - Ci.k,j > 0, we have j > 1 and use the hypothesis that 'Ck,p ¢
Lk,j-l when j = 2, ... ,Uk. We obtain
mk,p - mk,j - 1t(Ci.k,p - Ci.k,j) < mk,p - mk,j - Itk,j-l (Ci.k,p - Ci.k,j) < -v.
Finally, if Ci.k,p - Ci.k,j = 0, we must have that mk,p < mk,j' Choosing v
smaller if necessary, we have in any case, le(z)1 < Izl- v •
(b) If j = l,oo.,uk - 1 and 'Ck,p E Lk,j, then mk,p - mk,j = Itk,j(Ci.k,p - Ci.k,j)'
On the other hand, we have x + It log 1z 1 = 0 (with the same choice of It
as in (a)) and x + Itk,j log Izl :::: H. Hence (ltk,j - It) log Izl :::: H and, since
Itk,j - It > 0, Izl :::: exp(H /(ltk,j - It)). Because 'Ck,p E Lk,j, we have Ci.k,p :::;
Ci.k,j and therefore,
= Iz I/Lk,j(ak,p-ak,j)-/L(ak,p-ak,j)
so that
If3.::: j .::: ab then <k.l ¢ L k.j- h and by Lemma 3.2. 11 (a), we have zmk,leWk,lZ =
s(z)zmk,jeWk,jZ, hence the inequality (a) holds if Izl » 1.
The proof of (b) is similar, with <k,Uk replacing <k.l' 0
3.2.14. Lemma. There are constants ro > 0, Ho > 0, such that if r ~ ro, and
H ~ Ho, then lor any z E Uk,j n Tk n B(O, r)C one has
I/(z)l> tIAk,jllzmk,jeWk.jzl.
Proof. We use the asymptotic development of I given in the previous lemma,
and then use Lemma 3.2.11(b). We obtain
I/(z)1 ~ IAk,jzmk,jeWkj(1 + Sk,j(z»l- L IA k,pzmk,p(1 + sk,p(z»eWk,pZI
p#j
= IAk,jZmk,j eWk,j I
It is clear that by choosing rand H sufficiently big, the required estimate holds.
o
From nOw On we will assume r ~ ro ~ 1 and H ~ Ho, so that the conclusion
of Lemma 3.2.14 holds. Note that the regions Tb and correspondingly the value
ro, depend on the opening e. On the other hand, we can take Ho independent
of e. Namely, it is enough to choose it so that
3.2.15. Lemma. Let 0 < r < infk,j IAk,j 1/2. If z E Uk,j n Tko Izl ~ ro, and ~ E
P, then
le-('z f(z)1 > r.
Proof. If z E Uk,j n Tk n Sk we have from the previous lemma and from
{)emma 3.2.12(a)
le-('z f(z)1 ~ 1IAk,jZmk,jeWk,jZe-('zl
IAk,j
> - -z II Imk 'I > r
- 2 '
where the previous to last ineqUality uses Lemma 3.2.4(a).
If z E Uk,j n Tk n Sk+h the reasoning is the same, with mk,l replaced by mk,<1k'
D
3.2.16. Proposition. For r » 1,0 < r < infk,j IAk,j 1/2, by switching the condi-
tions ~ E P and z E (Uk,j Vk,j)C n B(O, rY, then
le-('z f(z)1 > r.
Hence,
If(z)1 > reH(z),
A consequence of the proposition is that all the zeros of f with large absolute
value lie in Uk,j Vk,j' Note that because the Vk,j do not depend on 8, the rl
can be considered independent of 8. The reader should verify that for those
228 3. Exponential Polynomials
Proof. Since z E Vk,j we have x + JLk,j log Izl = K, for some K such that
IKI :::: H. By Lemma 3.2.10, if T:k,p ¢ Lk,j, we have
Izmk,p-mk,i e(Wk,p-Wk,i)Z I = Iz Imk,p-mk,i eX(ak,p-ak,i)
= Iz Imk,p-mk,i -J.Lk,i (ak,p-ak,i) e(ak,p-ak,i)K
:::: Izl-ve(ak,p-ak,i)K.
Similarly, if z E Vk,j n Sk+b the same argument holds with mk,1 replaced
by mk,ak' D
°
where 8k,p -+ as z -+ 00 within '0,k n Tk, and Log z represents the principal
value of the logarithm.
Proof. When Tk,p E Lk,j, we have mk,p - mk,j = /.Lk,j(Wk,p - Wk,j)' We then
apply Corollary 3.2.19, and we are done. D
3.2.22. Remark. If we do not make the assumption that 1/Ik = 17: /2, and hence
that Im(wk,p - Wk,j) = 0, then we can introduce ({Jk = 1/Ik - 17:/2 in the expan-
sion of f as follows:
f(z) = zmk,jeWk,jZ L Ak,p(1 + 8k,p(Z»
Tk,pELk,j
3.2.23. Proposition. Let g(z) = l:1::J:o;n Aje WjZ , z = x + iy, Aj =f:. 0, Wj E JR,
WI < W2 < ... < Wn. For H > 0, h > 0, and YI E JR, consider the rectangles
R := {z E C: Ix I ~ H, Iy - YII ~ h}. There is a constant Ho ~ 0, independent
of YI and h, such that every zero of g lies in the strip Ixl ~ Ho. Moreover, if
H ~ Ho, the number N (R) of zeros of g in the rectangle R satisfies
4,
with limx->oo f31 (z) = O. Hence, if x » 1 we have 11 + f31 (z)1 ::: and G cannot
have any zeros. Similarly, G(z) = (1 + f32(Z», where f32(Z) ~ 0 as x ~ -00.
Hence, there is Ho > 0 such that if I Rezl > Ho, G(z) =I O.
Clearly, if H ::: Ho the number N(R) is independent of H. Let us fix e > O.
Choose H ::: Ho such that
and
on the lines x = ±H, j = 1,2. For a given h > 0, YI E~, let us choose
8 > 0, 0 < 8 < min{h, Jre/2Yn}, with property that G has no zeros on the
boundaries of the rectangles RI := [- H, H] X [YI - h - 8, YI + h + 8] and
R2 := [-H, H] x [YI - h + 8, YI + h - 8]. This is possible because the zeros
of G are isolated.
Let us investigate now the variation of the argument of G along aRI' In
the vertical portion x = - H, it is at most rr e 12. On x = H, it is at most
rrel2 + (2h + 28)Yn. On the horizontal sides, the function Re G(z) has the form
By a similar reasoning
hYn
N(R) ::: N(R2) ::: - - n + 1 - e.
rr
Since e > 0 was arbitrary, we obtain the proposition. D
The reader can easily verify that the example g(z) = 1 +e Z shows the esti-
mate in Proposition 3.2.23 is optimal.
3.2. Distributions of Zeros of an Exponential Polynomial 231
The tenn 2K(z)/p, is bounded, while -2x/p, -+ +00 as Izl -+ 00, hence
Iy/xl -+ 00 as Izl -+ 00, z E Vk,j'
as Izl -+ 00, to conclude that IWI - w21 :::: IVI - v21 :::: 8/2 if rl » 1.
If IYI - Y21 < 8, let q = YI - Y2. We have
and Log(zI!z2) --+ 0 as rl --+ 00. Hence, in this case, we can also conclude that
IWl - w21 ~ 8/2 if rl » 1.
Proof. From Proposition 3.2.16 we know the existence of Ho > 0 and <0 > 0
such that if we fix H ~ Ho there is ro > 0, so that Izl > ro, z rf. Uk,j Vk,j, implies
that If (z) I ~ <oeH(z). Therefore, it is enough to consider z E Vk,j, Iz I --+ 00.
Following Remark 3.2.22, we can expand f as follows:
where W = ze- itpk + ILk,j logz, {h,p = eitpk(wk,p - Wk,j), and YJk,p(W) = ek,p(z).
Clearly, YJk,p(W) --+ 0 as Iwi --+ 00, and if d(z, Z(f)) ~ 8, we have
d(w, Z(gl)) ~ 8/2 as long as Izl » 1. Here Z(f) = {z: Izl > R, f(z) = OJ,
Z(gl) = {w: gl (w) = OJ, the variables z, W being restricted to z E Vk,j' Izl > R,
and W to the corresponding region. Note that as a consequence, I Re wi :::: Hand
1m W --+ 00, as z --+ 00. We are thus led to introduce the following exponential
sum, with real frequencies:
go(W) =L Ak,pefh,pw,
where the summation takes place over the same set of indices as above. From
what we have just said, if z --+ 00 in Vk,j, then
Ig(w)1 ~ c8~
3.2. Distributions of Zeros of an Exponential Polynomial 233
Let us define, with the preceding notations for a function f E AEP, a > 0,
s > 0, H > 0,
Rk,j(a, S, H) :={z E C : Im(ze-il"k) + /-Lk,j argz E [a, a + s),
I Re(ze-il"k) + JLk,j log Izll s H},
with argz E [cpk. CPk + n), and let N(Rk,j(a, s, H)) be the number of zeros of
f in Rk,j(a, s, H).
We are now ready to prove the P6lya-Dickson theorem on the distribution
of zeros of functions that are asymptotically exponential polynomials.
gk,j(W) = L Ak,pefh,pw.
rk,pELk,j
°
Assume H > so that the conclusion of Proposition 3.2.23 holds for every gk,j'
Note that the quantity IfJk,j+1 - fJk,j I = IWk,j+1 - Wk,j I is the distance between
the furthermost frequencies of gk,j'
Choosing rl » 1 we can guarantee that for different choices of (k, j) the
sets V(qJt, JLk,j, H + 1) n B(O, rl)C are disjoint. (Recall Vk,j = V(qJb JLk,j, H).)
Note that rl can also be chosen so that for z E V(qJt, JLk,j, H + 1) n B(O, rlY
we can take arg z E [qJt, qJk + Jl'] and that for a convenient choice of ao
Im(ze-i<{Jk) + JLk,j arg z ~ ao > 0.
+
in the region n (arg z E [qJk, qJk Jl'D. The transformation F k,{ z ~ W =
U + i v := ze-i<{Jk + JLk,j log z maps n conformally into the half-strip lu I :::
H + 1, v ~ ao. The regions Rk,j(a, s, H + 1) are mapped conformally onto
the rectangles
R(a, s, H + 8) = [-H - 8, H + 8] + i[a, a + s],
for a ~ ai, 8 > 0, 0::: 8 ::: 1.
Let /k,j(z):= J(z)z-mk,je-Wk,jZ and hk,j be defined by hk,j Fk,j = Jk,j'
°
0
° °
It follows from Proposition 3.2.23, applied to the gk,j, that there is a constant
M > such that for any a E JR, < 8 ~ 1,
Ngk,j (R(a, 8, H + 8» ~ M.
for any point W EA. (The line segment A depends clearly on 8, a, and (k, j).)
It is enough to divide the rectangle in parallel horizontal strips of equal width
1/(M + 1), and take as A the bisectrix of a strip void of zeros of gk.j'
Fix (k, j), s > 0, and a:::: al + 1, let Ra,s be the rectangle whose
vertical sides lie in u = ±(H + 8) and whose horizontal sides are given
by the line segments Al and )..2 corresponding to R(a - 8, 8, H + 8) and
R(a + s, 8, H + 8), respectively. We have that Ra,s ;2 R(a, s, H) and that, for
every point W E 0 Ra,s one has
8
d(w, Z(gk,j)) :::: 2(M + 1)
From 3.1.3 we conclude that there is • = .(8) > °such that
Igk,j(w)1 ::::. on oRa,s,
If V:::: Vo » 1, we have l1Jk,j(w)1 :::: ./2, hence, for a :::: a2 :::: al + 1, we have
Ngk,/Ra,s) = Nhk,j(Ra,s).
We remark that the choice of a2 depends on 8, and eventually on B. The height
of Ra,s lies between s and s + 28, and now from Proposition 3.2.23 we infer
that
SI 8I w k,j+1 - wk,jl
l Ngk,j (Ra,s) - 27r IWk,j+1 - Wk,j I :::: nk,j - 1+ 7r .
°
Given B > 0, we can chose < 8 :::: 1 so that the left hand side does not
exceed nk,j - 1 + B. for any (k, j). Hence
and
, S
Ngk,j (Ra,s) :::: 27r IWk,j+1 - Wk,j I - nk,j + I-B.
This concludes the proof of the P6lya - Dickson Theorem. o
3.2.26. Remarks. (1) We have already used in the proof of Theorem 3.2.25 the
factthatforany HI > H, and as soonasa » 1, the number Nf(Rk,j(a, s, H)) =
Nf (Rk,j (a, s, HI)).
(2) Taking B = 1 in the statement of Theorem 3.2.25, one gets the bound
Taking B = ~, for those s such that (s 12:7l') IWk,j +I - Wk,j 1 E N* one gets the
sharper bound nk,j - 1 for the above quantity.
In any case, we get
Let L be a side of the P6lya polygon P(g) of g, and denote by l(L) its
°
length. Let us denote by 1/IL the direction of the exterior normal of P(g) on this
side L. Set CPL = 1/IL + :7l'/2. Let < 0 < :7l'/2 be fixed and denote
TdO) := {z E C: 1argz -1/ILI < O}.
°
3.2.28. Proposition. If there is ro > such that the function fig is holomorphic
in TdO) n (ij (0, roW, then there is a side L * of the Polya polygon P (f) of f
such that L * is parallel to Land l(L) ::: l(L *).
Proof. Let Ph, 1 ::: h ::: a', be the vertices of P(g), L = Li. = [Ph, PHd, and
1/1L = 1/Ih. Let Q h(g) be the corresponding polygonal region, constructed as in
Lemma 3.2.9, resting on the side L, <h,p' 1 ::: p ::: a h, the vertices of Qh(g) lying
outside L'h' we also denote by Lh,p = [<h,p' <h,p+d, Vh,p = V(cpiz, J.,Li.,p' H), the
corresponding half-strips, and Th(O) = {z E C: largz -1/Ihl < O} = TdO).
We are going to show that if fig is holomorphic in Th(O) n (B(O, r)y,
then there is a side L * = Lk of P (f) such that Lk is parallel to Li. and the
corresponding vertices ak, ak+1 satisfy IPHI - Ph 1::: lak+1 - ak I·
We assume that H » 1 and r » 1 so that we can use the conclusion of
Theorem 3.2.26, if necessary (for both f and g) .
. For r» 1 we can also assume that V(cpiz, J.,Li.,p' H) S;; Th(O) n (B(O, r)Y.
Since fig is holomorphic in this region, then all the zeros of g in V (cph' J.,Lh,p' H)
are also zeros of f. Therefore, f has infinitely many zeros, in V(cph' J.,Lh,pH).
3.2. Distributions of Zeros of an Exponential Polynomial 237
It follows from the previous lemma that there are ({Jt, I-tk.j such that ({J~ = ({Jk
and 1-t'".P = I-tk.j. Therefore, there is a side Lk of P(f) directed by ({Jt, hence
parallel to L'". By the same reason, for every P, 1 S P S af, - 1, there is
j E {I, ... , ak - I} such that I-t'",p = I-tk,j'
For a » 1, s > 0, let
'R~,p(a, s, H) = {z E V~,p: Im(ze-iqJ~) + I-t~,p argz E [a, a + sH.
It coincides with the region 'Rk,j(a, s, H), previously defined for f, and since
every zero of g is a zero of f (counting multiplicities) then
Ng('R~,p(a, s, H» S Nj ('Rk,j (a, s, H».
By Theorem 3.2.26 we have
s - -
Ng('R~,p(a, s, H» = 2n l,Bh,p+1 - ,Bh,pl + 0(1),
Recall that the .Bh,p lie in the segment L'" and are ordered in such a way that
So that
f(L~) S L lak,j+1 - ak,j I s L lak.HI - ak,j I = f(Lk),
i.e., the length of the projection of P(f) onto JR, 8R(f) ::: O. By PR[ai, ai+d
we denote the length of the projection onto JR of the segment [ai, ai+d.
238 3. Exponential Polynomials
xl =qg+r,
with either r = 0 or ojR(r) < OjR(g).
Proof. Let u" ... , up be real numbers such that UI < ... < up (if p 2: 2) and
let {UI,"" up} = {Re(aj): 1:::; i :::; mI. Let us define I j := {i: Re(a;) = Uj}
and
<Pj(z) = LAk(z)exp[i(Imak)z].
kElj
Then
I(z) = L <Pj(z)eUjz .
I~J~p
Note that the frequencies of the <Pj are all distinct and purely imaginary. The ring
<5 of exponential polynomials with purely imaginary frequencies is a subring
of E.
Let us consider the set
F:= {h E E*: h(z) = L Eh(z)eYkZ,O:::; YI < ... < Ys, ek E <5\{0}} U {OJ.
l~k9
Let us write g(z) = I:1~j~n \II) (z)e WjZ , WI < ... < Wn, \II) E <5\{0}. Assume
first that UI = WI = O. Then I, g E F, OjR(f) =
up. and OjR(g) Wn. =
If up < W n , then we can choose X = 1, q = 0, 1= r, and we are done.
If up 2: Wn and n = 1, we can take X = \fI" q = I, r = 0, and we are also
done.
Hence, we assume up 2: Wn, n 2: 2. Let up, ... , Up-k be all the Uj 2: Wn.
Consider
ro(z) = \fin (z)/(z) - (<Pp(z)e UpZ + ... + <Pp_k(z)eUp-kZ)e-WRZg(z),
qo(z) = (<I>p(z)e UpZ + ... + <l>p_k(z)eUp-kZ)e-W,z.
Then we have
\fInl = qog + ro,
qo E F, ro E F. and either ro =
0, or ojR(ro) :::; up - (w n - Wn-I). If ro = 0 or
if ojR(ro) < Wn• we are done. If not, repeat the procedure for ro and thus obtain
qlo rl E F. such that
\fInro qlg + rl,=
where either rl = 0 or ojR(rd :::; ojR(ro) - (w n - Wn-I). Since Wn - Wn-I > 0
we can continue to the point we find qo, ... ,qs-I E F, rs E F, and either rs = 0
or ojR(rs) < Wn, and one can write
= (\fI~-lqo + \fI~-2ql + ... + qs-I)g + r s,
\fI~1
which proves the lemma in the case UI = WI = O.
3.2. Distributions of Zeros of an Exponential Polynomial 239
Let us now consider the case of two arbitrary values UI, WI. The functions
F(z) := f(z)e- U1Z , G(z) := g(z)e- W1Z , are now in the previous situation. Hence,
°
there are 1/r E 6, q, rEF, such that either r = or o]R(r) < O]R(G), and
1/rF = qG + r.
It follows that
1/r(z)f(z) = [q(z)e(ul-wilz]g(z) + eU1Zr(z).
Clearly, either eU1Zr = ° o]R(eU1Zr) = o]R(r)
or < O]R(G) = O]R(g). 0
3.2.31. Theorem. Let f, gEE, g =1= 0, and assume the coefficients of g are
relatively prime. If there is R ::: 0, and an angular sector S of opening strictly
bigger than lr such that fig is holomorphic in S n (B(O, R)y, then there is
h E E such that f = gh.
Proof Rotating coordinates if necessary, we can assume that the sector is
°
S = {z E C: - y/ < Argz < lr}, with < y/ < lr.
The proof proceeds in three stages. First, we shall show that if fig is
holomorphic in {1m z > 0, Iz I > R}, then fig = q 11/r, with q, 1/r E E, and 1/r has
only purely imaginary frequencies (i.e., 1/r E 6* in the notation of the previous
lemma.) We show later that q I 1/r being holomorphic in {z E C: I Arg z I < y/}
implies ql1/r = alb, a E E, bE C[z]. Finally, using that the coefficients of g
are coprime we show that q I a E E.
3.2.32. Lemma. If fig is holomorphic when Izl > R, Imz > 0, then there are
q E E, X E 6* such that xf = gq.
Proof of Lemma 3.2.32. Replacing z by z + i R we can assume fig
is holomorphic in 1m z > O. This transformation is compatible with the
condition 1/r E 6*.
We show first the following:
u,
Figure 3.5
the same size relation, let us say PJR[Pko Pk+d ::::: PJR[aik, ait+l]' Therefore,
Appealing now to the preceding Lemma 3.2.30, we infer there are XES"',
and exponential polynomials q and r such that
xl =qg+r,
and, either r = 0, or 8JR(r) < 8JR(g). We can rewrite the previous equation as
-gr =x-Ig -q,
which shows that rig is holomorphic in {Imz > OJ. From what we have just
shown we must have
b~(z)h(z) = q*(iz)1/r(z). o
3.2.34. Lemma. Let f, g, h E E and c(f), c(g), and c(h), respectively, the
g.c.d. of their coefficients. If fg = h, then c(f)c(g) = c(h).
Proof of Lemma 3.2.34. Let f(z) = LI:5i:5m Ai (z)e(li; g(z) = LI:5j:5n Bj(z)efJj ;
and h(z) = LI:5k:5P We can assume that al -< ... -< am, fh -< ... -<
Ck(z)e Ykz •
fJn, and YI -< ... -< Yp·
Consid.er first the case where c(f) = c(g) = 1. We want to show that then
the Ck must be coprime, i.e., c(h) = 1.
Let WI, ... , Wq be the different values taken by ai + fJj, 1 :::: i :::: m, 1 :::: j ::::
n. Assume WI -< lV2 -< ... -< wq • We have
From the independence of {eAZhec over C[z] (Corollary 3.1.3) we obtain that
p :::: q, that for each frequency Yk == wit' for some It, and that
Ck(Z) = L Ai (z)Bj(z),
(li+fJj=W1k
0= L Aj (z)Bj (z),
(li+fJk=WI
Let us suppose that c(h) 1:- 1. This means there is a value z a which is a =
common zero of all polynomials Ck. Since c(f) = =
1, z a is not a common
zero of all Ai. Let io be such that Aj(a) = 0 for 1 :::: ; < ;0 and Aio(a) 1:- o. Let
WI = aio + fJI, whether I E {II. ... , Ip} or not, the corresponding sum vanishes
at z = a. On the other hand, if ai + fJj = aio + fJlo it is not possible that; > io
242 3. Exponential Polynomials
Since the left-hand side vanishes at z = a, and the second sum also does, we
have Aio(a)BI(a) = O. In other words, BI(a) = O.
Let us now assume Bj(a) = 0 for 1 :::: j < k :::: n. We want to show that
Bk(a) = O. We consider
The total sum vanishes at z = a, as well as the sum for those indices (i, j)
with i < io. If i > io, then Clio -< Cli and, since Cli + f3j = Clio + 13k we conclude
f3j -< 13k. i.e., j < k. Hence the sum with indices i > io also vanishes at z = a.
For i = io we have only one term AioBk' Therefore, Bk(a) = O.
We have arrived at the conclusion that Bk(a) = 0, 1 :::: k :::: n, thus c(g) i= 1,
contrary to the hypotheses. Therefore, is must be the case that c(h) = 1.
We consider now the case where c(f), c(g) are arbitrary nonzero polynomials.
Then F = flc(!) E E, G = gjc(g) E E, c(F) = c(G) = 1. Hence, H = FG
also has c(H) = 1. H = hj(c(f) c(g)) is an exponential polynomial, and by
Corollary 3.1.3 (i.e., the independence of {eAZhelC over C[z]), we conclude that
c(f) c(g) divides all the coefficients of h. That is, c(f) c(g) divides c(h). On
the other hand, since c~H) = 1, we conclude that we have c(f) c(g) = c(h). 0
Let us now return to the proof of Theorem 3.2.31. By Lemma 3.2.32 we have
that there is X E 6*, h E E, such that
xf = gh.
It follows from the hypotheses that h j X is holomorphic in {Iz I > R, I Arg zI <
lJ}. We can apply Lemma 3.2.33 and conclude there are b E C[z]*, q E E
such that
bh = qx.
That is,
xbf = Xgq·
Because X i= 0 we have
bf = gq.
From the assumption c(g) = 1 and Lemma 3.2.34, we conclude that b divides
c(q). Hence, qjb E E and, afortiori
LEE,
g
as asserted by the theorem. o
3.2. Distributions of Zeros of an Exponential Polynomial 243
. log If(re
if)
)1
11m
r;:¢r r
= H(e if)
).
when rk < r ~ rk+ 1. Hence, given 6 > 0, let j so that 1/j < 6, then
1" n(Z, rj) n(Z, r)
- L..J Pn ~ +6---,
r rl::::lz.19 r r
which shows that E has zero relative measure. o
3.2.37. Corollary. Let n(V, Zo, r) denote the number of zeros (counted with
multiplicities) of an entire function f E AEP in B(zo, r) and l(P) the length
of the boundary of the Polya polygon of f. Then
lim n(V, zo, r) = l(P) .
r-+oo r 2]l'
Proof. This follows from properties of functions of completely regular growth.
See [Lev, p. 288]. 0
n(V, zo, r) 4
~ -Or
]l'
+ 2d°f,
where 0 = max{laj I: aj frequency of fl. Using a geometric argument, he also
obtained
l(P)
n(V, zo, r) ~ 2]l' r + vdof,
where v denotes the number of vertices of the P6lya polygon.
Let us start by recalling some notation. If f is a meromorphic function in an
open set U of the complex plane, f ¢. 0, then for Zo E U, v(zo) = v(zo, f) is
the index of the first nonvanishing coefficients of the Laurent development of f
about Zo, i.e., if v(zo) < 0 then Zo is a pole of f and -v(zo) is the order of this
pole, if v(zo) > 0, then it is the order of Zo as a zero of f. finally, if v(zo) = 0,
then Zo is neither a zero nor a pole of f.
Following [Voor] we introduce a generalization of the total variation of the
argument of f along a closed interval [a, b] of the real axis.
3.2. Distributions of Zeros of an Exponential Polynomial 245
3.2.39. Lemma.
lim A(a, b, f,,)
,,--+0
= A(a, b, f).
Proof. By the additivity of the functional A we can assume that f has no zeros
or poles in ]a, b[, so that for some rJo > 0 we have
f'(z) ex f3
f( z) = -z-a + -
z-
b + g(z),
For 0 < e « 1, a < ~ < a + e, the last two tenus are bounded for all values of
y/,O < IY/I < Y/o. So that
A(a, a + e, fry) = 1 a s
+ I Im(f~(~)lfry(~»1 d~
= O(e) 1 a +£
+ laY/I a y/2 + (~ _ a)2
a~
1IIm(f/(~)lf(~))!d~ = 1 IImg(~)ld~ +
a
b
a+s
b s
- O(e).
1 Im(f~(~)lfry(~»1 _l Img(~)1 d~
1
b s
-
a+s
I
a+s
b s
- I I
: : 1 IIm(f~(~)lfry(~» Img(~)ld~
b s
- -
1 + - + 1
a+s
b- s d~ b- s d~
::: laY/I a+£ y/ 2 (~ a)2 1.8Y/1 a+s y/ 2 + (~ - b)2
+ 1a+s
b s
- I Im(g(~ + iy/) - g(~»1 d~
::: (Ial + 1.8I)(b - a)IY/le- 2 + O(IY/i)·
We infer that
IA(a, b, fry) - A(a, b, f)1 :::0 (c) + O(IY/I) + 0(1Y/l/e 2)
+ (lal + 1.81)1 arctan(e/lY/1) - 7T121.
If we let IY/I = 0(e 3), then all the tenus above are O(e), which proves that
lim A(a, b, fry)
ry .... O
= A(a, b, f). o
and
(f;)' II; = k(f~IfT/)'
hence the identity (ii) is clear for IT/ instead of I, and the estimates (i) are a
consequence of the triangle inequality for IT/' gT/. Using the continuity at rJ = 0
proved in the previous lemma, we conclude the corollary holds. D
where O(z):= I Arg(f'(z)II(z)) I (Arg~ is, as usual, the principal value 01 the
argument 01 ~).
Proof. Note first that I Arg ~ I is a continuous function in C* with values in [0, rr].
Hence,O is a continuous function in a neighborhood of a and of b. Moreover,
for small 7} real,
d d
dl; O(~) = ak dg Im(Log(f'(~)II(g)))
= ak 1m (:g LOg(f'(~)II(I;)))
= aklm(f"(g)II'(g) - 1'(I;)II(g))·
l
~
X
k+1 I Im(f'(g)II(~))I dg = ak l ~
X
k+1 Im(f'(~)II(g» dl;
248 3. Exponential Polynomials
so that
n [Xk+!
= ~ iXk Uk Im(f"(~)/f'(~» d~ - (O(b) - O(a»
!!...
dz
[1/In!!...
dz
[... !!...
dz
[1/I1!!...
dz
I] ...]] = 0,
where 1/11, ... , 1/In are meromorphiclunctions on [a, b]. (For simplicity we shall
drop the brackets in the future applications 01 this proposition.) Then
On the other hand, the previous corollary and Corollary 3.2.40(i) and (ii) yield
One can prove, using this proposition, that if ({JI, ••• ,({Jm are linearly inde-
pendent meromorphic functions on [a, b], and CI, ... , Cm are arbitrary constants
such that I = L Cj({Jj ¢ 0, then an upper bound for A(a, b, f) is independent
of the values of the constants. The same holds for an upper bound of the total
number of zeros and poles of I on [a, b] (see Exercise 3.2.8).
We are now ready to go back to exponential polynomials. We recall from
Section 3.1 that we denote
I(z) = L pj(z)eajZ ,
I~j~n
deg Pj = mj ~ 0, dO I = m I + ... + mn + n - 1.
3.2.44. Proposition. Let a < b, N(a, b, f) be the number olzeros (counted with
multiplicity) 01 the exponential polynomial I in the segment la, b[. Set
Then
where D = d/dz (see Exercise 3.2.6 and the proof of Lemma 3.1.2). We are
using the notation from the statement of Proposition 3.2.43. Further, one can
write D2 = DID, D3 = DIDID, ... , where I is the constant function 1 acting
as a multiplication operator. Hence (*), has the form described in Proposi-
tion 3.2.43, applied to l(x)e- aIX , where we have a total dOl functions 1/Ik> some
250 3. Exponential Polynomials
1/Ik == 1, and the others 1/Ik(X) = exp«ak - ak+l)x). From Proposition 3.2.43 we
conclude that
A(a, b, f(x)e-aIX):s L A(a, b, e(ak-ak+I)X) + 7rdof,
l~k~n-l
19~n-l
since A(a, b, 1) = 0 and A(a, b, eWX ) = (b - a)1 Imwl (Exercise 3.2.3). Since
1m ak is increasing, we have
L 1Im(ak - ak+l)1 = L (Imak+l - Imak)
l~k~n-l l~k~n-l
= 1m an - Imal = J -I,
and hence,
A(a, b, f(x)e- aIX ) :s 7rdof + (b - a)(J -I).
On the other hand,
ib Im(f'(x)lf(x» dx ~ (b - a) 1m an - i
b
1 Im«(f'(x)lf(x» - an)1 dx
Proof. For s > 0 we have N(a, b, f) ~ N(a - s, b + s, f). From the previous
proposition we have
(b - a + 2s)I + rr N(a - s, b + s, f) - rrdof
~ (b - a + 2s)J + rrdof - rr N(a - s, b + s, f),
so that
2rr N(a, b, f) ~ 2rr N(a - s, b + s, f) ~ 2rrdof + (b - a + 2s)(1 - /).
We can obtain from Proposition 3.2.44 inequalities holding in any line segment
of the complex plane.
3.2.46. Corollary. Let a, bEe and denote by [a, b] the line segment joining
these two distinct points. If f has no zeros on la, b[, then
1m (r
i[a,b]
f'(z) dZ)
f(z)
~ rrdof + maxlm«b -
}
a)cxj)
Proof. Let
r g'(~) d~ = t (b _ a) f'(a + (b -
1 a)~) d~
io g(~) io f(a + (b - a)~)
= r
i[a,b]
f'(z) dz.
f(z)
Hence, from the second inequality of Proposition 3.2.44 we infer that
1m (r
i[a,b]
ff'(Z) dZ)
(z)
~ rrdof + max{lm«b -
}
a)cxj)}.
Since P = cv{a, ... , an}, we have maxj {lm«b - a)cxj)} ~ maxzeP 1m (z(b - a».
D
Then,
1
NQ ::: 2dof + 2rr {(b - a)tJ. x + (e - d)tJ. y }.
NQ < NQ = _1_ (
- •
f'(z) dz = _11m
2rri JaQ. f(z) 2rr
(rJaQ, f'(z) dZ) .
f(z)
We divide aQe into four segments and apply Corollary 3.2.46 to each of
them. We obtain
+ l:::oj:::on
max (1m «a - b - 2e)aj»
+ l:::oj:::on
max (Im(i(d - c - 2e)aj»
+ l:::oj:::on
max (Im(i(e - d - 2e)aj»
Proof. It is easy to see that there is R > 0 such that f has no zeros in the
region IRezl > R. Let Q = {z E C, -R::: Rez::: R, e::: Imz::: d}, and apply
the previous theorem. 0
3.2. Distributions of Zeros of an Exponential Polynomial 253
Clearly,
~y ~ 2Q,
hence,
4Qr
n(V, zo, r) ~ 2dof + -.
1t
o
Finally, we need a little geometric argument to obtain Voorhoeve's strength-
ening of the P6lya estimate involving the perimeter l(P) of the P6lya polygon
of f.
= L Im(fik(ak+1 - ad)
I::;j::;v
= L Im(ak(fik+1 - fid).
I::;k::;v
Similarly,
maxlm(z(fJk+l - fJk»
zeP
= Im(ak(fJk+l - fJk» = Im(ak+l(fJk+l - fJd)·
L fik+lak+l = L fikak,
= 1m ( L fik+lak+l - L fik+l ak )
l~k~v l~k~v
= 1m ( L
l~k~v
fikak - L
l~k~v
fik+lak)
= 1m ( L (fik -
l~k~v
fik+l)ak)
= L Im(ak(fJk+l - fJk»
l~k~v
= L maxlm(z(fJk+l - fJk».
zeQ
l~k~v
1 1
zeros of f. Therefore, if N Q denotes the number of zeros of f in Q, we have
Figure 3.6
256 3. Exponential Polynomials
From the proof of Lemma 3.2.50 we know the maxima are achieved at any
point of the segment [fit. fik+l). Since Q circumscribes B(zo, r) we can choose
a point Z that belongs to aQ so that Ii - iol = r and
max Im«i - iO)(ak+l - ak» ::: rlak+l - akl.
zeQ
see [Ba]. Note that in this case the frequencies of f are ±i ± iA, which are
all algebraic numbers if A E Q. Moreover, the coefficients of f written as an
exponential sum are also algebraic numbers. So that for f (z) = sin Zsin AZ,
A E Q, guarantees that the zeros are well separated for any weight.
If we consider f(z) = sin(z - a) - sin("fiz - a), a > 0, then the frequen-
cies are again algebraic, but the coefficients are combinations of sin a and cos a
with algebraic coefficients, which in general are transcendental numbers. On the
other hand the zeros form the sequences
2kn: 2a + (2k + I)n:
M and r;:;' k E 7L..
l-v2 l+v2
Again, depending on the choice of a we can make the zeros well separated
or not.
Moreover, note that from the P6lya-Dickson Theorem 3.2.25 it follows that
when the frequencies of an exponential polynomial f are purely imaginary, then
there is a positive constant A such that the zeros of f lie in the logarithmic strip
I Imzl ~ A(1 + 10g(1 + Izl»
so that, if the zeros of f are well separated for the weight p(z) = IImzl +
10g(1 + Izl), it follows that there are 6 > 0, N > 0, such that
6
IZj - zkl ::: (IZkl + IZjl)N
for any two distinct zeros Zj, Zk.
These considerations led to the following conjecture of Ehrenpreis [Eh5] ,
[BY2]:
EXERCISES 3.2.
1. Using the notation from the proof of the P6lya-Dickson theorem show that:
(a) If z = x + iy E Vk • j , then limz ..... "" Iy/xl = 00.
(b) For z E Vk •j , lim z..... "" Arg z = 7r /2.
(c) Show that the curves of equation y + iLkj Arg z = c in V k•j are asymptotic to
y = c - iLk.j 7r /2 as Iz I --+ 00. Similarly, the curve x + iLk,j log Iz I = H is asymptotic to
x + iLk.j logy = H.
(d) Let w = z + iLj.k logz, z E Vk,j' Show that if IZI - z21 ~ 8> 0, then for Zl, Z2
large one has IWI - w21 ~ 8/2.
(e) For some R » I, (U Uk,/) U (U Vk./ ) ;2 (B(O, R)C n {y ~ O}).
2. Assuming the Conjecture 3.2.52 is correct, show that if II, fz are exponen-
tial polynomials with algebraic frequencies and algebraic coefficients, which have no
common zeros, then IE l(fl, fz), the ideal generated by II, fz in Ap(C)(p(z) = Izl or
I Imzl + log(l + Izl».
3. Let A(a, b, f) be the functional defined in Definition 3.2.38. Evaluate it in the
following cases:
(i) I(z) == C E IC;
(ii) I(z) = e''';
(iii) I (z) = e'P(z), rp holomorphic; and
(iv) If I is a polynomial of degree m, show that A(a, b, f) ::: 7rm.
(b) Let / be a meromorphic real valued function in [a, b], show that
6. Prove that if /(z) = EI::;j::;n pj(z)e ajz is an exponential polynomial, degpj = mj'
Then
where D = d/dz.
7. Use the example /(z) = (2cosz)n-1 to show that the estimate in Corollary 3.2.45
is sharp.
8. Let ~I,"" ~m be meromorphic functions in the disk B(O, R), and let Wk =
W(~10 ... , ~k) be their Wronskian, i.e.,
~I ~k
~; ~k
Wk = det
~:i-I) ~kk-I)
(a) Show that for any other meromorphic function h, W(h~1o ... h~k) = hkWk.
(b) Show that ~I, ... , ~k are linearly dependent if and only if Wk == O.
(c) Show that for any k and I E N*
W(W(~I' ... , ~ko ~HI)' W(~10 ... , ~ko ~k+2)' ... , W(~I' ... , ~ko ~H/» = wt- I WH/ .
(Hint: Prove it by induction on k.)
(d) Let / = EI::;k::;1 Ck~ko Ck E C and ~I, ... ,~m linearly independent. Show that
D W~_I D W~_2 D .. . D Wi D wl DL = O.
Wm Wm- 2 Wm - I Wm - 3 W3 WI W2 WI
(Hint: Consider the auxiliary functions hk = W(~I' ... , ~k-I, f), for 2 :::: k :::: m + 1.)
(e) Let / be as in (d) and let N(a, b, f) be the number of zeros and poles of /
(counting multiplicities) in the closed interval [a, b], when -R < a < b < R. Give an
upper bound of N(a, b, f) independent of the choice of constants Clo ••• , Cm.
CHAPTER 4
G(T)(z) := / TI;,
\
1
1- ze
1;)'
4.1.1. Proposition. If T is an analytic functional carried by the compact convex
set K cc Q, then G(T) belongs to Jro(Q(K». Moreover, its Taylor series
development about z = 0 is given by
G(T)(z) = L J(T)(n)zn,
and its Laurent development about z = 00 is
G(T)(z) = _ "J(T)(-n).
L...J zn
n:;:O
260
4.1. The G-Transfonn 261
Proof. The proof depends on the relations between the Cauchy and
Fourier-Borel transfonns of an analytic functional which were established in
Section 1.3. To see that G(T) is holomorphic in Q(K) we can appeal to the
representation
1
(T, h) = 2i 1 y T(w)h(w) dw,
A
G(T)(z) = -2.
1
I
1 y
T(w) 1
A dw
-ze W
'
G(T)(z) =L
n::::O
(;i 1 y
T(w)e nW dW) zn
is the Taylor development of G(T) about z = 0, convergent at least for Izl < rl.
Similarly, for Izl > r2 we have
so that
G(T)(z) = _ '" J(T)(-n)
L.J zn
n::::!
is the Laurent development of G(T) about z = 00, convergent for Izl > r2. D
and y the loop y = e- a lying in e-U\e- K • Then, for every analytic functional
T carried by K and every h E .1f(U) we have
(T,h)=-~
2HZ Jry G(T)(z)h(-Logz)dzz =~ rG(T)(e-S)h(s)ds.
2HZ Ja
Proof. Let z = e- s , for s E U, hence, s = - Log z. Therefore,
r
Ja
G(T)(e-S)h(s) ds =- r
Jy
G(T)(z)h( _ Log z) dz .
z
On the other hand,
1
- - = --U1 +qJ(u),
1- e U
G(T)(e- S) = / Tw,
\ 1- eW
1_) = _ / _1_) +
S \
Tw,
W - S
O(s),
and O(s) := (Tw, qJ(w - s») is well defined, since W - s E 2Q and 0 E .1f(Q).
It follows from Cauchy's theorem that
1 O(s)h(s) ds = O.
_1_) =
Since
/ Tw , HT(S),
\ s- W
1 1
we have
- -1. G(T)(z)h(-Logz)- 1
dz = ----;- T(s)h(s)ds = (T,h).
A
D
2HZ y z 2z a
1 1
have
J(T)(~) = --.
1 dz
G(T)(z)e-SLogz_ = -.
1 G(T)(e-S)e SS ds.
2HZ y z 2m a
(T,h) = -~
2HZ
r g(z)h(_Logz)dz,
Jy z
which is independent of the choice ofloop y in C\e- K , as long as it has index 1
with respect to every point of e- K • We can thus assume it is a Jordan curve. It
4.1. The G-Transfonn 263
= - -1.
2m
1
y
I dz
g(z)(1-c;exp(-Logz))--
z
= --1-.1
2m y
g(z)_I_ dz
z -C;
= g(l;).
The last identity holds because Indy (C;) = -1 and g(oo) = O. D
4.1.5. Corollary. A necessary and sufficient condition for two analytic func-
tiona Is T I , T2 E .J'f' (Q) to coincide is that there is no E Z such that the identity
~(TI)(n) = ~(T2)(n) for every n ~ no, n E Z.
Proof. Evidently there is a convex compact set K ~ Q such that both TI and T2
are carried by K. The analytic functionals Sj = enoz 1j (i.e., (Sj, h) = (1j, enoz h))
are also carried by K. Moreover, ~(SI(n)) = ~(S2)(n) for all n ~ O. By Propo-
sition 4.1.1, G(SI)(l;) = G(S2)(C;) for C; in a neighborhood of O. Since Q(K)
is connected, G(SI) == G(S2). The previous corollary implies now that SI = S2,
hence TI = T2. D
and
(ii) F(s) = f(O) + lim ' " B(s + 2imr) in C\ U(K + 2irrj) ,
2 N-+oo L.J
Inl~N jEZ
Proof. Since f = J(T) for a uniquely determined T E .jf'(K), (a) and (b) are
immediate consequences of the previous two propositions and their corollaries.
Furthermore, ZU) = G(T). It is also clear that F(s) = G(T)(e- S ) and, hence,
F admits an analytic to continuation to the connected open set
The statement (i) of part (c) is now obvious from Proposition 4.1.1. To prove
(ii) we need to prove first a little lemma.
l(w)'= lim L 1 = -I +2 L w
• N-+oo w - 2rrin w w 2 + 4n 2rr2
Inl~N n::-:I
1
lim
Rew-->-oo 1 - e- W
= O.
Fix some value 0 < CL < rr/2.1f WE Q and Iwi » 1, we have I Argw 21 ::: CL.
Then A := 2rr/w will also satisfy I ArgA 21 ::: CL and
~+2" w =~ 1+2" 1
w ~ w2 + 4n 2rr2 2rr ~ 1 + n 2A2
n~l n~l
IAI (
::: 2rr 1 +
2
~
in
n-l
dt
1 + (J2/2)t2IAI2
)
< ~
- 2rr
(1 + 2~ Jo IAI
roo ~)
1 + u2
IAI (
::: 2rr 1 +
rr~) '
TAl
which is bounded when IAI « 1, i.e., when Iwl » 1. These two observations
imply that k - I is bounded in 11m wi ::: rr, hence everywhere. Therefore, the
difference k( w) - I (w) = =
C constant, which can be evaluated at w O. It =
=
turns out that C = O. We conclude that k I, which proves the first part of the
lemma. The second part follows from the first by letting w = s - ~. We simply
take e > 0 sufficiently small, so that the sets Ve(K) + 2rrij remain pairwise
disjoint (recall that Ve(K) = (z E C: d(z, K) ::: eD, then ~ - s remains in a
compact set disjoint from 2rriZ when ~ E Ve(K). 0
Returning now to the proof of Theorem 4.1.6. We have already seen that
= /\ T, 1~) lim L
= !3'(T)(O) + N-->oo / T,
F(s)
1- e -s
Inl~N
\ S + 2rrln
1.
- ~
) .
On the other hand, from the diagram following Definition 3.3.12 we see that
/ T,
\ s +
2 1.
7rln - ~
) = rrT(s + 2rrin)
= C(T)(s + 2rrin)
=8(f)(s + 2rrin) = 8(s + 2rrin).
Thus,
F(s) = -1(0)
2
.
+ N-->oo
hm
L 8(s + 27rln), .
Inl~N
as we wanted to prove. o
266 4. Integral Valued Entire Functions
dk
( k!1 dl;k 1 )
G(T)(z) = 1 - ze~ ~=o
1 1
G(T)(e- S ) = 1- e- S
= 4+ N--+oo
lim '"
~ S -
.
27rln
(s ¢ 27riZ).
Inl::::N
1
=L
nEZ
(s - 27rin)k+l
(s ¢ 27riZ),
for Res> 0,
Res < o.
4.1.9. Remark (On the Inversion Formula for the G-Transform). Let y be a
loop in e-Q\e- K of index 1 with respect to the points of e- K , we denote by
9Jt(g), the Mellin transform of g E Jt"o(C\e- K ), the entire function
9Jt(g)(z) := - - .
27r1
1 1 y
-
g(w)
wz+l
dz,
where
Exp(K) := {f E Jf(C) : "1£ > 0, 3Cs ::: 0
such that If(z)1 ~ CseHK(z)+slzl, z E q.
We also consider Exp(Q) := UKccn Exp(K).
We would like to find now some explicit formulas for the inverse m- i of the
bijective map m: Jfo(Q(K)) ~ Exp(K).
Let f = 'J(T). For 0 < ({J+ < rr /2 and -rr /2 < ({J- < 0 we denote by y+ =
y+(a, ({J+) (resp. y_ = y_(a, ({J-)) the half-lines of origin a E ] - 1, O[ making
an angle ({J+ (resp. ((J-) with the positive real axis. Let y be the broken line y_y+.
As we shall now see, for Izl sufficiently small we have the integral representation
G(T)(z) =~
21
1y
(w) e-irrwz w dw
SInrrw
(0 < Izl « 1).
For n ::: 0 let Yn be the loop of base point a suggested by Figure 4.1.
Let 0 < z « 1, the residue theorem yields
1
--:-
21
1v
,n
_.few) . z W dw
_ _ e- I1rW
SInrrw
= rr " Res (f(W)'
'~
O~k~n
_._ _ e- I1rW z W , w
SInrrw
= k)
= L f(k)zk.
O~k~n
a
-1 n + 1/2
Figure 4.1
268 4. Integral Valued Entire Functions
Let us now verify that the integrand is integrable along y+. In fact, we let
z = re ie
and w = peiC(J+ + a. The denominator is bounded below, since
for some c = c(<p+, a) > O. Hence, when w E y+ we have for arbitrary e > 0,
If (w)e-ilfWz WI C eP(HK(ei~+)+e)ra+cosC(J+e-pe sinC(J+elfP sinC(J+
-'-'------'-_ _--'- < e
I sin:rrw I c exp(:rrp sin <p+)
On the other hand,
() sin<p+ - Hk(eiC(J+) - e - (cos<p+)logr > 0
for all 0 < r « 1, since cos <p+ > O. Therefore, the integrand satisfies
If(w)e-ilfWzwl _
-'-'---------'- < c' e p8
I sin:rrwl - ,
for some c' > 0 and /) > 0 that can be chosen independent of r, (), as long as
o< r « 1. This confirms that the integral over y+ converges uniformly and
absolutely for all z E B(O, ro)\{O}. The same is true for y_.
On the other hand, consider the vertical segment rn of Yn, w = n + ~ + iu,
(n + ~ - a)tan<p_ :::: u :::: (n + ~ - a)tan<p+. We have
and
elf1ul
I sin:rrwl I sin(:rr(n + ~) + iu)1 = I coshul ::: -2-'
so that
for some C, B' > O. Clearly, the upper bound tends to zero when n --+ 00 if
0< r < ro « 1.
As a consequence, we obtain that for 0 < Iz I < ro « 1
1
---:-
21
1y
_.f(w).
_ _ e- l1fW zWdw
Slll:rrw
= "~ f(n)zn = G(T)(z),
n~O
since the series also converges for small Izl. Note that the value of the integral
we obtained does not depend on the choice of arg z. On the other hand, for
z = 0, the value of the integral obtained setting Ow = 0 does not necessarily
coincide with G(T)(O) unless f(O) = O.
We are going to use these remarks to show that the integral converges and
represents G(T) for all z ¢ (e- L U {OD, L := K + ] - 00,0].
4.1. The G-Transfonn 269
For that purpose we consider separately the convergence of the integral over
y+ and over y_. Let z ¢ (e- L U (O}), z = re ili , then the integral over y+ is
absolutely convergent if there is an 8 > 0 such that
Thus, a + if3 E L if and only if the inequalities (**) hold for -n /2 :s ({J :s n /2.
Since z ¢ (e- L U (O}), then the points -log Izl - i arg z ¢ L for any choice of
argz. Clearly, if a + if3 ¢ L, then a + if3 + t ¢ L for any t 2: O. Therefore, for
the given ({J+(O < ({J+ < n/2), since the inequality (**), with ({J = ({J+, determines
the supporting line of L slope tan(n /2 - ({J+), we can choose a determination
of arg z such that the point - log Iz I - i arg z lies on the open side of this line
and the set L on the other. (Namely, if argz E ] - n, n] does not work, we can
choose argz E ] - (2n + 1)n, -(2n - 1)n] for n » 1.) (See Figure 4.2.) It is
clear that we can find a half-strip R+ = {a + if3: a > ao > -00, f31 < f3 < f32,
f32 - f31 < 2n} such that -log Iz I - i arg z E R+ and there is 8 > 0 such that
for every a + if3 E R+
a cos ({J+ - f3 sin ({J+ 2: HK(ei'P+) + 28.
The image of R+ by the map a + if3 f-* e-(a+ifj) will be the open angular
sector S+ = R E C, 0 < Izl < e- ao , -f32 < arg~ < -f3d and z E S+. Since
(**) holds now for every point ~ E S we see that the integral over y+ defines
a holomorphic function of ~ in the sector S+. In exactly the same way we
construct a sector S_ such that the integral over y_ defines a holomorphic func-
tion of t;. Now z E S = S+ n S_ and we have a holomorphic function defined
in S, which for 0 < Izl < ro coincides with G(T)(~) since, as we said before,
1t
-1t
Figure 4.2
270 4. Integral Valued Entire Functions
for those values the integral representation does not depend on the choice of
arg ~. We conclude that the integral representation is now valid everywhere in
C\(e- L U (O}) because G(T) is holomorphic and single valued there.
In the same fashion, if we take ({J+ E ]1l', 1l' /2[, ({J- E ] - 1l', -1l' /2[ and
consider the path Y«({J+, ({J-, a) = Y defined as before, then the integral
-~
21
1 !(w) (_z)W dw
JI S1D1l'W
Figure 4.3
and the radius of convergence of the series Ln::-:o f (n )zn is exactly exp( -hf (0».
Hence,
lim sup v'lf(n)1 = exp(hf(O»
n-->oo
and
· log If(n)1
11m sup = hf (0) .
n--+oo n
This is a theorem originally due to P6lya which can be easily derived from the
previous Remark 4.1.9.
We are going to use the G-transform to study some classical results about
holomorphic functions. The first question to study is that of the convergence of
4.1. The G-Transfonn 271
the Newton series of an entire function. (An excellent reference for this subject
is [Ge].)
We recall from [BG, p. 237] that if I is an entire function, its nth divided
difference Don = Don (f) is given by
Clearly, if I is a polynomial, the series terminates and its sum coincides with
I. The empty product has the value one, as always.
Let U = {z E Q: le z - 11 < I}. This set is an unbounded open convex subset
of Q whose boundary is the curve
7t
Log 2
-7t
Figure 4.4
(T,(el;-W) = L (~)(-I)n-j(T,el;j}=f:1.n(f).
O:;;J:;;n
L
R(R + 1) ... (R + n - 1) r n
__________________
n!
n~l
has radius of convergence 1. Letting Izl .::: R, r = A(K, s), this ensures that the
Newton series converges uniformly over compact subsets of Co 0
1['\ {-I}, let us call/3 this subspace. In A we have the subspace A of those
analytic functionals for which there is a distribution R E £' (] - rr, rr [) such that
(T, h) = (Ry, h(iy» for every h E Je(Q).
The relations between these spaces are given by the following statement:
4.1.13. Proposition. The map <I> which to every TEA associates the element
S E B defined by
C(S)(z) = -G(T)(z), z E C\1[',
is a linear isomorphism of A onto B that maps A onto 13.
Proof. Let us recall (Definition 1.6.9) that the Fantappie transform of S E Je' (1[')
is the pair (Sl, S2) given by
so that Sl(Z) = -rrS(z) = -C(S)(z) for Izl < 1 and S2(Z) = rrS(z) = C(S)(z)
for Izl > 1. Moreover, the map S f-+ (Sl, S2) is an isomorphism of Je'(1[') onto
Je(B(O, 1)) x Jeo(C\B(O, 1».
If qJ is a function holomorphic in the complement of a compact set L £;
1['\{-1}, which is zero at 00, then the functional S E JeI(1[') defined by
Sl = qJIB(O, 1) and S2 = -qJl(C\B(O, 1) is carried by L. Let y be a loop
in C\] - 00,0] of index 1 with respect to L, then we can define T by
(T,h) = --I-.1qJ(Z)h(-LOgZ)dz.
2rrl y z
By Corollary 4.1.4, T is carried by any compact convex set K such that L £;
e- K , in particular, TEA. Furthermore, qJ = G(T).
=
Conversely, if TEA, the function qJ G(T) allows us to define S E B such
=
that C(S) -qJ, and hence, S <I>(T). =
= = =
If S <I>(T) 0, then qJ -C(S)
°A =
0, hence G(T) 0, and since T is =
carried by {O} + i[ -a, a], for some :os a < rr. we obtain T = 0. We conclude
that <I> is a linear isomorphism from onto B.
If T can be represented by a distribution R E £10 - rr, rrD and S = <I>(T),
we have
1~(T)(n)1 = I(TI;, enl;}1 = I(R y , einY}1 :os C(1 + Inl)D
for some C, D > 0. Since S = -(1/rr)G(T), we can apply Propositions 1.6.10,
1.6.14, and 4.1.1 to conclude that the Fourier coefficients Cn (S) satisfy
Icn(S)1 :::: C(1 + Inl)D
and S E V'(1['). Hence <I>(A) £; 13.
Now let S E /3, then Sl and S2 have slow growth on a neighborhood of 1[',
i.e.,
274 4. Integral Valued Entire Functions
=
for some M ~ 0, k ~ O. As a consequence, if G(T) -C(S), we have that
G(T)(e- S ) satisfies a similar inequality in a neighborhood of i] - rr, rr[. In the
proof of Proposition 4.1.2 we showed that
G(T)(e- S ) = rrT(s) + O(s), I Imsl < rr,
with 0 holomorphic in 11m s I < rr. Therefore, T has slow growth near i] - rr, rr [.
Since T is holomorphic inside a compact subset of i] - rr, rr[, the Edge-of-the-
Wedge Theorem [BG, Theorem 3.6.23] ensures the existence of distribution R
with compact support in ] - rr, rr [ such that
= (R y , h(iy»)
(T, h)
for every h E Jr(Q). We conclude that <I>(A) = B. o
4.1.14. Proposition. Let T E Jr'(Q). A necessary and sufficient condition for
the existence of a distribution R E e' (] - rr, rr [) such that
(T, h) = (R y , h(iy»), hE Jr(Q),
Iffor some pEN, f(n) = O(lnIP),for all nEZ, then there is a E [0, rr[ and
C > 0 such that
(z E C).
There are elements of Exp(Q) such that their zero sets contain the complement
of a set of the form Z\pZ, for some p E N*. For instance, if w = e 2rri / 3 is the
cubic root of unity, consider the function
+ e WZ + eW Z)g(z)
2
fez) := (1
f = J«o + Ow + o(2) * T)
and
G(wz) = L f(3k) (wz)3k = L f(3k)z3k = G(z).
k:o:O
1 .:5 k .:5 p.
276 4. Integral Valued Entire Functions
J(T)(~) = - -1. ~
2rrl
"1
I::::k::::p Yk
G(T)(z)e-SLogz_.
dz
z
Since any point in Yk is of the form z = wk-Iu, where u E YI, and Logz =
Log u + [2(k - l)rri]1 p, we have
1
G(T)(z)e-SLogzdz = e-[2(k-I)7ri/p]S
Yk z Yl
1
G(T)(u)e-pogudu.
u
On the other hand, there is an analytic functional S carried by (II p)Q such
that
(S,h} = --I-.1G(T)(e-iO-,/P)7rZ)h(-LOgZ)dz,
2rrl Y z
rc/3
rc
rc/3
Do
0
-rc/3
-rc -rc/3
(p = 3)
Figure 4.5
4.1. The G-Transform 277
y = ei(l-l/p)Jr yl . Hence,
~(S)«() = __1_.ei(1-l/P)Jr~
2Jrl
1YI
G(T)(z)e-~Logzdz
z
and
~(T)(n = ei(l-l/p)Jr~ ( L e-[2(k_l)Jri/ PR ) ~(S)«().
l:;:k:;:p
We can now state the following proposition:
G(z)
N(z)
= -D(z) =
ao + alz + ... + as_Iz s- 1
= -zn
L Un
bo + biz + ... + bs ZS n:::1
°
where the values Zj =I are the poles of G of respective mj 2: 1, 1 :::: j :::: M.
The advantage of writing G in this form depends on the fact that the Laurent
expansion at 00 of the different terms is very simple. In fact, for v 2: 0, a =I 0,
Un = L Qj(n)zj
15;h'iM
280 4. Integral Valued Entire Functions
for some collection of nonzero polynomials Qj, deg Qj ::: mj - 1, and complex
numbers Zj =f. 0, 1 ::: j ::: M. We can appeal to the fact that (P\!}\!~o is a basis
of the vector space C[w] to express each Qj as
Qj = L Aj,\!P\!,
O~\!~mj-l
for some uniquely determined complex numbers Aj,\!. Then, for Izi » 1 we have
=L (L L Aj,\!p\!(n)z;) z-n
n~1 l~j~M O~\!~mj-I
That is,
- = (boi
ao + alz + ... + as-Iz s + ...s
+ bsz) (UI
-; + -;
U2 + . .. ) ,
which implies
bsUI = as-I,
bs-IUI + ... + bsU2 = as-2,
(which is said to be of order s when bobs,," 0), with the initial conditions
UI =as_J/bs ,
U2 = (as-2 - bs-IUI)/bs,
with bs "" 0 and UI, ... , Us-I given, then the Z-transfonn Ln>1 unz- n is the
Laurent development at 00 of a rational function. To see this, we define
ao, ... ,as-I by the fonnulas
as-I := bsUI,
as-2 := bsU2 + bs-IUI,
Proof. It only remains to show that the condition is sufficient.· For that purpose
we need a preliminary lemma.
4.2.2. Lemma. Let A = (aij h~i,j~n be a square matrix with complex coefficients
and let Aij be the cofactor ofaij. Denote D = detA and d = det(aijh~i,j~n-I'
then we have the relation
Al1Ann - AlnAnl = Dd.
Proof of Lemma 4.2.2. One can easily verify the following matrix identity:
al1 al2 al n AI,I 0 0 Al,n
A2,1 1 0 A2,n
A3,1 0 1· .. 0 A3,n
An-I,1 0 1 An-I,n
ani an2 ann An,l 0 0 An,n
al,2 al,n-l
a2,2 a2,n-1
(I :1
=
an-I,2 an-l,n-I
an,2 an,n-I
and conclude that
D(Al1Ann - AlnAnl) = D 2d.
If D i= 0 we obtain,
Al1Ann - AlnAnl = Dd
as we wanted. Since the invertible matrices are dense in the space of all n x n
matrices, this formula remains valid by continuity. 0
(t)
If for some fixed n one has Hj = 0 for all j 2: jo then one obtains from
these identities that H:+ I = 0 for all j 2: jo, hence H~ = 0 for all j 2: jo and
all m ~ n.
Let s be the smallest integer such that H~ = 0 for all n sufficiently big. By
hypothesis such an s exists. If s = 0, then Un = 0 for all sufficiently big n, and
then the conclusion of the proposition is true.
If s = 1, then we conclude from Lemma 4.2.2 that u~ - Un-IUn+1 = 0 for all
n sufficiently big. Hence, either uno = 0 for some no »
1 and then Un = 0 for
4.2. Integral Valued Entire Functions 283
all n ::: no, in which case we are done, or Un -=1= for all ° n ::: no » 1. In this
case the quadratic equations leads to the relations
~ = Un+1 = Un +2 = ... = 11- -=1= 0,
Un+1 Un+2 Un+3
r
Hence, for n ::: N we can consider the system of s equations and s + 1
unknowns
+ + bs-Iun+s-I + bsu n+s =0,
u + bs-Iu n+s + bsUn+s+1
(tt)
bOUn+1
,
+ =0,
Un Un +2
n s
Un+1 Un +3 U +
Un+s-I )
(
bl = (-l))..det :
Un Un+1 Un+s-I )
bs = (-1)s Adet ( : = (-1) s AH:- I •
Un+s-I Un+2s-2
Thus, the solution is nontrivial if and only if A -=1= 0. For such a solution we also
have
boun+s + blun+s+1 + ... + b sUn+2s = 0,
since
Un
s ( •
0= Hn = det .
Un+s-I
U n+s
284 4. Integral Valued Entire Functions
( _1)s+1
= ). [boun+s + ... + bsUn+2S] .
Therefore, exactly the same (bo, ... , bs) chosen above will be a solution of the
system
bOUn+1 + ... + bsUn+s+1 = 0,
{ Un+2 + ... + bsUn+s+2 = 0,
bO
4.2.3. Corollary. Let the series G(z) = En>1 unz-n be convergent for large Izl.
A necessary and sufficient conditionfor G to-be the Laurent expansion at z = 00
of a rational function, which vanishes at 00, is the existence of an integer s ::: 1
such that the Hankel determinants Ht vanish/or all k ::: s.
Proof. Namely, the linear recurrence bou n + ... + bsu n+s = 0 can be rewritten as
boun + ... + bsun+s + OUn+s+1 + ... + OUn+k = 0
for k ::: s. Thus, as we already know that if G is a rational function then H: = 0
for n ::: 1 and some s ::: 1, we conclude that H: = 0 for all k ::: s and all n. In
particular Hf = 0 for all k ::: s.
=
Conversely, if Hf 0 for all k ::: s ::: 1, then from (t) we will have (Hf)2 =
=
Hf Hf - H~+I H;-I 0 for all k ::: s, etc. We conclude that H: 0 for all =
n ::: 1 and all k ::: s. The preceding proposition can now be used to finish the
proof. 0
=
4.2.4. Theorem (Fatou). Let G(z) En>o unz- n be the expansion about z 00 =
of a rational function. Let s ::: 1 be the smallest of the integers k ::: 1 for which
there is a polynomial Nko deg Nk ::: k - 1, and a polynomial Dko deg Dk = k,
such that G = Nk/ Dko and let
N(z) ao + alz + ... + as_Iz - s 1
G(z) = --
D(z)
= - -------
bo + bIZ + ... + bsz s
be such a minimal degree representation of G. Then, if all the Un are integers we
can choose the denominator D(z) = bo + bIZ + ... + bsz s to be a polynomial
4.2. Integral Valued Entire Functions 285
with integral coefficients and bs = 1. In this case, the numerator N will also
have integral coefficients.
Proof. We want to show that bs = 1. For that purpose, choose integers CI, ••• , Cs
such that if Wn := CIU n + ... + CsUn+s-1o then WI = W2 = ... = Ws =I- O.
From the factthatthere is a representation G = Nsf Ds, with deg (Ds) = s, we
know that it follows that H:
= 0 for n 2: 1, and hence that there are nontrivial
solutions bo, ... , bs of the system
as-I := bsUI,
as-2 := bsU2 + bs_IUI,
4.2.5. Lemma. Let P(z) = Lk<1 Pkzk, Q(z) = Lk<m qk zk , R(z) = Lk<n rk zk
be three formal Laurent series with integral coefficients, PI =I- 0, qm =I- 0, and
PQ = R. If neither of the sequences (pkh:sl, (qkh:sm' has a common prime
factor, then the same holds for the sequence (rkh:sn.
Proof of Lemma 4.2.5. Since PI =I- 0 and qm =I- 0, then n = I + m and rn =I- O.
Assume there is a prime d which divides all the rko k ::::: n. If d does not divide
PI, then we have rn = Plqm and this implies that dlqm. From the next identity,
rn-I = Plqm-I + PI-Iqm, we conclude that dlqm-I. Continuing in this fashion
we see dlqk for all k ::::: n. Therefore, we can assume dlpl, and hence there is
286 4. Integral Valued Entire Functions
j ::: 0 such that dlpl,.'" dlpl_j and dlpl_j_l. Consider now the identity
Let us return to the proof of Theorem 4.2.4. Since N and D are coprime,
there are polynomials A, B E Z[z] and an integer m E N* such that
AN+BD=m.
Mn(E) = inf{max
zeE
IP(z)l: P(z) = zn + alz n - I + ... + an}.
One knows that this infimum is achieved for some monic polynomial of degree
n whose roots line in cv(E). Any polynomial that achieves this minimum is
called a Cbebyscbev polynomial for E.
The transfinite diameter of E is the quantity
If Y is a union of Jordan curves YI, "', YI, with disjoint interiors, we will denote
Int(y) = UI::J:::I Int(Yj)·
We shall need the following fact about the transfinite diameter. For every
e > 0 there is a rectifiable curve Ye, which is the union of finite many Jordan
4.2. Integral Valued Entire Functions 287
This curve Y is a rectifiable union of finitely many Jordan curves with disjoint
interiors and
Int(Ye) = {z E C: ITm(z)1 < (r(E) + 8)m}.
For z E E we have (Tm(z)) = Mm(E) ::: (r(E) + 8/2)2, hence E ~ Int(Ye).
Consider now the monic polynomials of degree mn
Pn(z) = (Tm(z))n.
Un = -1.
2TlI
1
y
Z n-l G(z)dz.
G(Z)Pk(Z) = L Pkz~~:)) .
n2:0
Note that if we let Po(z) == 1, Po«u n)) = Un, the last formula remains valid
for k = O.
Consider the Hankel determinants
UO Ul
( Ul U2
H~ := det : n::::1.
Un
288 4. Integral Valued Entire Functions
that is, the composition of the operators corresponds to the product of the poly-
nomials. Therefore, we also have
Po«Uo» P1«ud)
Pn«U n» )
Ho = det ( PO«Ul» Pn«~n+l»
Moreover,
Vj,k = Pj Pk«U n» = -1-.1
2rrl y
Pj(Z)Pk(Z)G(Z) dz.
z
After these preliminaries we can state and prove the following theorem due
to P6Iya-Carlson:
Vj,k = -.
1
27rl
1 y
dz
G(z)1j(Z)Tk(Z)-,
Z
then
Let M = max ZEY IG(z)/zl and L be the length of y. Since the Tk are
Chebyschev polynomials for Int(y), there is a constant A > 0 (independent
of k) such that for z E Int(y)
ITdz)1 ~ A(r(lnt(y» + tf ~ A(rY.
Therefore,
and
IBnl
o <- IT_ (Iv- j,O
1+'" + lv-j,n I) <
A n +1
1
- (1 _ r,)n+1
(r,)n(n+I)/2.
OSjsn
Since r' < 1 there is an no such that for all n :::: no we have
A 1 (r,)n/2
- - - < 1.
1- r'
Hence, for n :::: no we have
IHol < 1.
On the other hand, the Un E Z for all n, whence Ho
E Z for all n. We conclude
that Ho= 0 for n :::: no. This implies that G is a rational function. D
It is now immediate from the discussion at the beginning of this section that
the following result holds:
where the Zj are algebraic integers lying in E, together with all their conjugates,
and the Qj are polynomials with algebraic coefficients in the field generated by
Zl, ... , ZN and their conjugates.
290 4. Integral Valued Entire Functions
In order to be able to apply the preceding theorem and its corollary we need
to find an elementary way to estimate1:(E) for rather simple compact sets E.
In particular, we assume that E has no holes, i.e., S2\E is simply connected.
It is known (see [BG, Chapter 4]) that there is a conformal map cp from E e
onto B(O, r)C, r = 1:(E), cp(oo) = 00, of the form
(see, e.g., [BG, Exercise 4.9.6]). Let us assume for simplicity that E is connected
and its boundary aE is regular of class Coo. We know that 1/f, cp admit a
Coo extension up to the boundary of B(O, r) and E, respectively (see, [BG,
Theorem 4.8.17]).
= =
For w re i8 we have dz 1/f'(w)dw, and the arc-length parameter s in aE
is given by
ds = rl1/f'(re i8 )ldO.
On the other hand, one can compute by residues
~
27l'1
1. iwi=r
1/f'(w) dw
w
=1
or, equivalently,
fo21r 1/f'(rei8 )dO = 27l'.
Hence, for the length L of aE one has
A =! r
iaE
x dy - y dx
°
as an elementary computation shows. (Parametrize aE, as earlier, using "".)
Equality holds if and only if an = for all n ~ 1, i.e., if and only if aE is a
circle. It follows that
~::::: r = r(E).
L
~- < r(E) < -
n - - 2n'
and observed that any of the two equalities implies that E is a disk of radius
r(E) and one has equality throughout. One can relax the hypothesis of Coo
regularity of aE to C 1 regularity ([Porn]).
T = "\'
L...J ak .J·8(k)
Zj ,
k,j
where the values log Zj are algebraic integers lying together with their conjugates
in B(O, 1).
Proof. The last part is just a consequence of Theorem 4.2.4 and computation of
G-transforms in Example 4.2.8. 0
4.2.11. Lemma (Fekete). Let E be a compact subset ofe with r(E) < 1, then
there can be at most a finite number of algebraic integers with the property that
they and all their conjugates lie in E.
292 4. Integral Valued Entire Functions
have the property that r(E) = limHoo onCE). If we had infinitely many
algebraic integers lying together with their conjugates in E, there is an increasing
sequence of integers nk -+ 00 that for any nk there is a family ZI,···, znk
of distinct algebraic integers that contains all its conjugates. The product
TI'::;j<h::;nk (Zj - Zh)2 is a symmetric function on the roots of a certain polynomial
with integral coefficients, hence it is a nonzero integer. In order words,
II IZj - zhl 2 2: 1.
'::;j¥-h::;nk
It follows that
and thus
r(E) 2: 1. D
4.2.13. Remark. The previous two lemmas are due to Fekete [Fe], [Bu], who
also showed that if r(E) < 1 there is a monic polynomial P E Z[z] such that
E £ S = {z E C: IP(z)1 < I} (see the construction of Ye after Corollary 4.2.6).
4.2.14. Proposition. Let K be a convex compact subset ofQ = {z: 1Imzl < rr}
such that reeK) < 1. Then, every entire function f satisfying f(N*) £ Z and
such that for every e > 0 there is Ce > 0 such that
If(z)1 ::: CeeHdzl+elzl (z E C),
4.2. Integral Valued Entire Functions 293
about z = 00.
From the previous results of this section we conclude that G(T) is a rational
function, all whose poles Zl, ... , ZN are algebraic integers, which belong to e K
together with all their conjugates, and
v ao + alz + ... + as_Iz s-
1
G(T)(z) = ~---...:---~
bo + biz + ... + bsz s
ao, ... , as-I, bo, ... , bs E Z, bs = 1.
Applying the inversion formula 6.1.3, we have
Proof. The compact e K cc B(m, 1), hence r(e K ) < 1. Moreover, by Lemma
4.2.12, m is the only algebraic integer that together with its conjugates belongs
to B(m, 1) = {z: Iz - ml < I}. The remark atthe end of the proof of Proposition
4.2.14 implies that the polynomial P has rational coefficients. 0
294 4. Integral Valued Entire Functions
1 1 -
rk E ] - 2' 2[nQ, Pk E Q[z].
Proof. There is e > 0 such that eK £ B(1, 1) n {z: Rez > e}, which ensures
-r(e K ) < 1. We can now apply Lemma 4.2.12 to P(z) = z - 1. The nonzero
roots of (p(z»m = 1 have the form 1 + e21rik/n, -n12 < k < n12. 0
where Ck are algebraic integers in B(O, ex) (together with their conjugates) and
Pk E Q[z]. Moreover,
{ 3+iJ3 3-iJ3}
(i) if ex < 0.8 then Ck E 1,2, 2 ' 2 ;
The reader will find other interesting extensions of the results of this section
in [Gra], [Grum], [Bez], [Bu].
T = L L Ak.v8~~),
l:sk:SpO:::1J:::mk
l:sk:sp
'"
gl(t,Z)=L...Cn(t)Zn= 1 - tz 2'
O 1 - 2tz + z
n~
g2(t, z) = 'L...
" Un(t)z n =
1-
1
2tz + z
2'
n~
O
g3(t, z) = 'L...
" Ln(t)z n = 1 ,
n~O
v'1 - 2tz + z2
(iv) Gegenbauer polynomials, C;(A > 0)
(a) For t E ] - 1, 1[ let 0, E ]0, 11'[ be such that cosO, = t, and define the convex sets
r
K l " = {OJ, K j " = i[-O" 0,] (j =2,3,4). Let j " be a positively oriented Jordan curve
in C\] - 00,0] enclosing exp( -Kj ,,). Show that the relations (for Izl sufficiently small)
~(T;,,)(z) = --2'
I
11'1
1
rj.r
gj(t, ~)e- zLo~d~
g -;-
'>
(l ::: j ::: 4)
1
p.(t) = 211'i
r d~
}y g(t,~) ~n+l'
where y is a Jordan curve enclosing the origin such that exp( - K 2,,) is contained in the
exterior of y.
(c) Let K be a convex compact subset of n and J E Jt'o(n(K» be such that its
Taylor series at the origin is Ln>o an zn. Show that for every t E ] - 1, 1[, j = 1, 2,
such that K + Kj" £; n and K - K + Kj" (or Kj" - Kj" + K £; n) there is an analytic
functional Sj., carried by K + Kj" with the property that:
Use residue calculus to obtain (i)-(iv). If Kj" - Kj" + K £; n just replace K by Kj" in
the previous construction.)
(d) Let t, S E ] - 1, 1[ be such that 20s + 0, < 11' (or Os + 20, < 211'), and let
Q(t, s, z) = 1 - 4tsz - 2(1 - 2t 2 - 2s 2)Z2 - 4tsz 3 + Z4. Show that for Izl < 1
1- z2
L Un (t)Un(s)z' = Q(t,s,z )'
n~O
"Un(t)CA(s)zn
fUn- 2(t) [t (~) C~m)Cn-2m(S)]
= n_=_O_ _---'=-m_=_O_ _ _ _-,---_ _ _--=-__
(-z)"
L...J
n~O
n Q(t
'
S Z)A
,
r! I: (' + 1) Cn(t)(-z)n
"(n + 1)··· (n + r)Cn(t)zn = ---"n_=o'---_n_ _--:---,-_
L...J (1 - 2tz + z2),+1
1- Z2 ~ (:!~) Un(t)(-z)n
L(n + 1) ... (n + r)Un(t)zn = r! (1 - 2
tz + z2) r +1
n~O
4. Let 1 be an entire function of exponential type such that for some convex compact
K ~ n and every e > 0 there exists a constant Ce > 0 so that 1 satisfies the inequality
G(T)(z) = P(z) .
1- zm
(c) Deduce that if y is a Jordan curve in C\] - 00,0] enclosing the mth roots of unity
different from -1, then
I(z) = L I(n)km(z - n)
!
O~n<m
where
sin7rz
if m is odd,
. (7rZ)
msm -
km(z) = m~ "L...J e2rrij(zjm) = . ( m 7rZ)
sm (m - 1)-
Ijl<mj2 m
if m is even .
. (7rZ)
m sm -;;
298 4. Integral Valued Entire Functions
Thus, with m' = [m/2] - 1 if m is even and m' = [m/2] if m is odd, we have
f(z) = "L.J CkexP (27rikZ)
--;;;- .
Ikl~m'
Conclude that f == O. (Hint: If If(z)1 ~ Ae Blzl , let m = [Blrl/7r] + I, and apply the
previous exercise to g(z) = f(rz/m).)
6. Let f be an entire function of exponential type and a, f3 E C such that
Im(a - f3) =F O. Assume that for some no E N we have
Ref(n +a) = Ref(n + f3) = 0 for all n::: no,
and, moreover,
lim f(n) = O.
Inl-oo
neZ
Summation Methods
299
300 5. Summation Methods
One of the essential points of the theory is contained in the following simple
lemma. (Compare with Theorem 1.3.11.)
°
5.1.1. Lemma. Let < R < 00 be the radius of convergence of the series f (z) =
L:n~o anz n • In the disk B(O, R) we have the integral representation
f(z) = 1 00
e- t 131(f)(tz)dt,
where 131 (f)(z) := L:n~oanzn In!, which is an entire function of exponential type
exactly II R.
Proof. In fact, we have lim sUPn-+oo V'la n I = II R. Hence, for every e >
is an integer ne such that
° there
lanllzl n ) -Izl n
- - - < ( -+e
1
n! - R n!
This inequality shows that 131 (f) is an entire function and, moreover, it is of
exponential type at most 1I R. We leave to the reader the task of checking that
the type is exactly 1I R.
From the obvious identity
n! = 100
e-tt n dt, n ::: 0,
we obtain
On the other hand, for z fixed, Izl < R, we can choose e > °with the property
that f.L := Izl(11 R + e) < 1. Hence, for n ::: n e , t > 0,
5.1.2. Lemma. With the same notations as in the previous lemma, assume that
Zo E C* is such that the function t 1-+ e- t 131(f)(tzo) is integrable in [0, 00[.
Then:
5.1. Borel and Mittag-Leffler Summation Methods 301
(i) for every A, 0 < A :::: 1, Z = AZo, the function t f-+ e- t 13\(f)(tz) is inte-
grable on [0, oo[ and
1 00
e- t 13\(f)(tz)dt = s 1 00
e-st13\(f)(tzo)dt,
with s = 1I A ~ 1;
1
(ii) the function
e- st 13\ (f) (tzo) dt
00
S f-+ S
°
z E K zoo Hence, fzo is holomorphic in K Zo by (ii) and coincides with f on the
segment AZo, < Amin{1, Rllzl} by (i). Therefore, fzo = f on B(O, R) n K zo '
and the conclusion of the lemma is correct. 0
This lemma justifies the definition of the Borel polygon B(f) as the set
of those of ~ E C such that there is a compact neighborhood V~ of ~ and a
nonnegative function g~ E L \ ([0, oo[) with the property that for every z E V~
the inequality
a.e. t > 0,
holds.
Proof. Let us show that M(f) ~ B(f). Let Zo E M(f), then Kzo CC D(f),
therefore we can find a positively oriented circle y in D(f) enclosing Kzo'
For ~ E y, we have ~ =f. 0 and ~ f-+ Re(zog) is a continuous function strictly
smaller than 1 - 8, for some 0 < 8. Choosing y conveniently we can make 8
as small as necessary.
Hence, we can write for ~ E Y
--
1 - =
1 - zog
10
00
e-Ue(zog)u duo
302 5. Summation Methods
We denote still by ! the analytic continuation of the series Ln~o anz n to D(f).
-1-.1 !(~)
Then
!(zo) = ( ('" e-Ue(zo!t;)u dU) d~.
27C1 Y ~ 10
Let A := maxy I!(~)gl < 00, hence for (~, u) E Y x [0, oo[ one has the esti-
mate
~
27C1
1y
!(~)e(Zo/l;)ud~ = ~ {
2m 111;1=8
~
!(~)e(Zo/l;)ud~ = 131(f)(uzo),
~
where the last identity is an elementary computation of residues. Moreover, this
identity yields the estimate,
!(zo) = 100
e-U131(f)(uzo) du,
!(z) = 1 00
e- t 131(f)(tz)dt
holds for Z E Vzo ' U E [0,00[. This shows that Zo E B(f). We conclude that
M(f) ~ B(f).
Let us now prove the other inclusion, B(f) ~ M(f). Let Zo E B(f), and
°
ex > small enough so that B(zo, ex) ~ B(f) and there is a nonnegative inte-
grable function gzo such that e- t IB1 (f)(tz)1 ::: gzo(t), a.e. t > 0, z E B(zo, ex).
Consider the open set
f(z) = 1
00
e- t 8,(f)(tz)dt, z E B(f).
=
It is easy to see that E, (z) eZ and that Ep is a function of exponential type,
for instance, from the decay rate of its coefficients (see [BG, Chapter 4]). We
need somewhat more precise asymptotic information. For future use we shall
restrict ourselves to p :::: 1.
Let us recall that the entire function 1/ r can be expressed in terms of the
Hankel integral
1 1 ( et
J
r(z) = 27ri ya .• fz dt.
where Ya,a is the Hankel contour in Figure 5.1, the indices satisfy 0< a, 7r/2 <
a < 7r, and t Z = eZ Log t (see [BG, Exercise 5.3.3]).
304 5. Summation Methods
Figure 5.1
= uP = ePLog u, we obtain
1
Changing variables in the integral, t
_1__ ~ uP -pz+p-l d
f( ) - 2' e u u,
z Jrl Y~.a
= =
where Y~,a is in fact Yb,{J, with b a 1/ p , f3 a/ p. Since a > 0 is arbitrary
and alp E ]Jr/2p, Jr/p[, we can replace Y~,a by the Hankel contour Ya,lI, with
() E ]Jr /2p, Jr / p[. In particular,
1 p ( du uP
Ep(z) p
= -. euP -du
-,
2Jrl Yo.9 U - z
where 0 < a, Jr/2p < () < Jr/p.
Proof. Let u = rei<p, r ~ a, cp = ±(). Then uP = r Pe±ipll, Jr /2 < p() < Jr. Hence,
for u in one of the half-rays of Ya,lI we have leuP I = erP cospll, cos p() < 0, r > a.
If Izl < a is fixed. we have Izl/Iul :::: f..t < 1 for any u E Ya,lI. Then on both
half-rays we have
L Iun+!
zn eu pi <
__ 1 erPcosp/l
- (1 - f..t) r '
n;:::O
5.1. Borel and Mittag-Leffler Summation Methods 305
L Iu::1 e I S A <
n"::O
UP
00.
1 (1
Therefore,
-p e uP -du
-= L -p e uP -du- ) Zn
2ni Ya,e U - Z n"::O 2ni Ya,e un + 1
=L
Zn
f(1 + n/ p) = Ep(z). o
n"::O
1
5.1.7. Lemma. The function
Zf-+-
p uP -
e du-
2ni Ya,e U - Z
is holomorphic in C\Ya,(J.
Proof. It is left to the reader. o
Let us say that Z is to the left of Ya,(J, if it belongs to the component of C\Ya,(J
which contains the origin. To the right, otherwise.
1
(i) For Z to the left ofYa,(J we have
Ep(z) = -p. e uP du
2n[ Ya,e U - Z
Figure 5.2
(ii) lim
Izl-H)O
IEp(z)1 = o.
Jr/2p+e:::1 Arg zl:::Jr
Proof. Note that the result holds for p = I as E I (z) = e Z • Let 0 < 8 « I so
that I Argzl ::: rr/2p::: rr/p -8. Thus we have Ep(tz) = pe(IZ)P +qJ(tz), with
limHoo qJ(tz) = O. It follows immediately that
· log IEp(tz)1 -- Rez
I1m ( P) ,
1-+00 tP
and this limit is uniform for Izl 2: 1. In particular, for z = Izlei'l', we have that
if 8 > 0 there is Re » I so that for Izl 2: Re
1
For rr /2p ::: I Arg zl ::: rr we can write
I
E (z)
p
= - rO-I/p) -zI + -2rri
p
Ya,e
e
uP udu
z(u - z)
,
since
I I u
- - =--+
u- z z z(u - z)
,
and
-p
2rri
1 Ya,e
eU P du = r(1-I/p)
I ,
because p > 1. It follows that
E z -- 1 -1 0 ( - I 2 )
p()- rO-I/p)z+ Iz1 '
everywhere. o
5.1.11. Lemma. In the open set {z E C: Lp(z) < I} (p ~ 1) we have the relation
-1- =
l-z
1 0
00
e- tP t p- 1Ep(tz) dt.
Proof. Let K CC {z: Lp(z) < I}, then we can find e > 0 such that
sup(Lp(z) + elzl P - 1) = -k < O.
zeK
Figure S.3
5.1. Borel and Mittag-Leffler Summation Methods 309
This integral defines hence a holomorphic function in {z: Lp(z) < I} and,
furthermore, one can justify for Izl < 1 the following computation:
= ~zn = _1_.
L..J l-z
n~O
We are going now to return to the Borel method, but this time with the sets
Kf := {z E C*: 3t' E C* such that { = t'i', Lp(1/t'):::: I} U to},
for z E C*. It is easy to show that Kf = zDp, where
Dp := Ki = It' E C*: Lp(I/t') :::: I} U to},
which is a compact convex set with 0, 1 E 8Dp. Moreover, Dp is star-shaped
with respect to the origin, and the equation of 8 D p in polar coordinates is
r = (cos (1)i/p , iT
-- < (1 < - .
2p - - 2p
iT
Figure 5.4
310 5. Summation Methods
For -n/2p < e < n/2p one has, with r' = dr/de and r" = d 2r/de 2 ,
r2 + 2r' - rr" = (cose)2/p(I + p + (1 + p)(tanpe)2) > 0,
which shows that D p is convex.
For f(z) = Ln>oanz n, with radius of convergence 0 < R < 00 as above,
and p ::: 1 we denote by M p (f) the open set,
Mp(f) := {z E C*: Kf ~ D(f)}.
5.1.14. Lemma. Let Zo E C* be such that t f-+ e- tP t p- 1Bp (f) (tzo) is integrable
on [0, 00[. Then:
(i) For every z = )"zo, 0 < )" :s 1, t f-+ e- tP t p- 1Bp(f)(tz) is integrable on
[0, oo[ and
= psP 00
e-(US)P u p- 1Bp(f)(uzo) du,
=
with s 1/)".
1
(ii) The function
00
S f-+ psP e-(tsJPtp-1Bp(f)(tzo)dt,
fzo(z) := p(zo/z)P 1 00
e- tP (zo/z)Pt p- 1Bp (f) (tzo) dt
5.1. Borel and Mittag-Leffler Summation Methods 311
D(f) = UBp(f).
p~l
x 2y' + y =x
admits a hyperfunction solution. On the other hand, if we try to find a formal
power series solution we find
f(x) = x L(-Itn!x\
n~O
which converges only for x = O. Apply the Borel summation method of Lemma
5.1.1 to the formal series. We obtain
1
F(z) = L(-Wzn = - .
1+ z
n~
O
IRn(x)1 ~ 1 00
sne-s/x ds = n!xn+l.
That is, the Borel summation represents the hyperfunction solution f in an
asymptotic sense.
These elementary remarks show that, at least in this example, there is a relation
between hyperfunctions, summability, and asymptotic developments. In fact, in
recent years these ideas have been carried out to develop a very beautiful theory,
which has already been used to solve several difficult problems. We refer the
reader to [Ec] , [MR], [Mal], and [Si] for these new results. Our presentation
owes much to [Lin].
EXERCISES 5.1.
1. Show that, a priori, it always makes sense to talk about the largest domain, star-
shaped with respect to the origin, into which a power series f admits an analytic
continuation.
n
{AZk: A ~ f}, and - -
M(f) = (z: Re«z - Zk)Zk) < OJ.
l::;k::;N
3. Show that if f(z) = En>o anz n has radius of convergence R, 0 < R < 00, then the
entire function F(z) =
En~o(anln!)zn has exponential type exactly II R.
4. Show that for ex > 0, the set
is open, and that Uzo U 8(0, R) is star-shaped with respect to the origin (see proof of
Proposition 5.1.4).
5. Justify the proof of Lemma 5.1.8, and prove Corollary 5.1.9.
6. Verify that D p is a convex compact set, M p (f) is open, as well as provide the
proofs of statements Lemmas 5.1.12-5.1.14 and Propositions 5.1.15 and 5.1.16.
7. Let rp be a holomorphic function in the closed half-plane Re z ::: ex, ex > O. Assume
there is '[', 0 ::::: '[' < 1(, with the property that for every e > 0 there is a constant c. > 0
such that 1(
IArgzl::::: 2'
(a) Let f3 ::: ex and m an integer so that m - 1 < f3 < m. Prove that for every integer
L rp(n)zn = 1erp(~)z{
N:::m
N
.~ d~, 2
n=m '" -1 YR
where YR is the contour described by the Figure 5.5. We denote rR the arc of circle
in YR.
5.1. Borel and Mittag-Leffler Summation Methods 313
R=N+1I2-13
m-l
Figure 5.5
(b) Let 1/10 E ]0,11"/2[. Show that there are constants Ro ::: 0, kJ > 0, k2 > 0, such that
for 0 < r < I, ~ = f3 + Re i1/t, one has
and
qJ(~)(-r)~ 1 < k
1 e21ri~ _ 1 -
fJ
2r e
-R(lrsin"'o-r-E)
if 1/Io::S 11/11 ::s ~.
(c) Choose 1/10, ro so that r < 11" sin 1/10 < 11", 0 < ro < I, (-logro) cos 1/10 = 11" sin 1/10.
Let k = max(k J, k2) and show that for any 0 < r < ro, ~ ErR, one has
and that
(d) Let ~ = f3 + it, a > 0 very small, r +a < () < 211" - r - a, r > 0, and z = re i8 .
Write (}J = () - 11" and show that
Z ~ l fJ+ iOO
(J-ioo e2"'~
rp(s)z~
,ds
- 1
is holomorphic in the angle r < arg z < 2l'l' - r.
(e) Let
00
F(z) := L:>(n)zn.
n=O
Show that F is holomorphic for Iz I < e- r • Moreover, if m > 0 and m - 1 < fJ < m,
l
then
L
{J+iOO rp(Oz~
F(z) = rp(n)zn - , e2"i~ _ 1 di;,
O:5n:5m-l (J-lOO
and for m < 0, -m - 1 < fJ < -m,
" rp( -n) r{J+iOO rp(i;)z~
F(z) = - L- -z-n- - J{J8-' e2"i~ _ 1 di;.
l~n~-m roo
Conclude that F has an analytic continuation to the angle r < arg z < 2l'l' - r.
8. The following are essentially direct applications of the previous exercise:
(a) Let 0 < a < 2, 0 < fJ < 1, and rp(z) := z-uz. Show that
zn r{J+iOO i;-u~z~
L nun = 1 -
n~O
J{J8 '
-roo
e2"i~ _ 1 ds
,,-=-
Deduce that
zn 1
lim
u.....o+L- nun 1- z
n~O
9. Let cp be a holomorphic function in C\ {aI, ... , aN}, aj E C\Z, such that there is a
value Ro > max{lajl: 1.::: j .::: N} with the property that for any e > 0 there is C, > 0
so that if Izl ~ Ro then
Icp(z)1 .::: C,e'lzl.
Use the results of this section and Exercise 5.1.8 to prove that:
(a) The series I(z) = L:.>ocp(n)z' has an analytic continuation to C\[1, 00[.
(b) Let No E N be such that all aj satisfy Re(aj) > -No, and define
<I>(z) := 27r1.L.....
"
Res (cp(~)z~
e2Jri~ _ l' aj )
l'5j'5N
which depends on the choice of argz. Show that if z E C\[1, 00[,0 < argz < 27r, and
-No - 1 < f3 < -No then the function I of part (a) satisfies
(c) Let z E] - 00, -1[, f3 = -(No + 1/2), and ~ = f3+ it, then, for some e(f3) > 0,
with the additional property that if we let No ~ -00 then e(f3) ~ O. Conclude that
(d) Let n = {z E C: Izl > 1, z ¢ [1, oo[} and 0 < argz < 27r for ZEn. With this
choice of arg z in n, let
I(z) := - L-
cp(-n)
- -
z·
<I>(z) (z En).
n~l
Then
(log z)k-l
Fl (z) = F(z) + 2rri (k _ I)! .
5.2.3. Lemma. Let (It «h, (h-OJ <rrlp. Assume that hf,p(ei()l) ::::hJ and
hf,p(eifh ) ::::: h2. Let
hJ sin P(02 - 0) + h2 sin p(O - OJ)
g(O) : = . .
smp(02 - OJ)
Then,for every 0, OJ ::::: 0 ::::: ~,
"()
hf,p(e' ) ::::: g(O).
Proof of Lemma 5.2.3. Let I) > 0 and let ga(O) = ap cos pO + bp sin pO be the
trigonometric function that takes the values hj + I) at 0 = OJ. Define
h(z) := f(z)e- i(a8+ib8 )zP
5.2. The Lindelof Indicator Function 317
The function h is of order p and finite type in the angle (h :::: 0 :::: 02, and
it converges to zero along the sides of this angle. By the Phragmen-Lindelof
principle, Ih I is uniformly bounded in the whole angle. It follows that
"9
ht,p(e' ) :::: g8(0), 0, :::: 0 :::: 02,
= =
Let us point out that if ht ,p(e i91 ) ht,p(eilh.) -00 and O2 - 0, < 1l/ p, then
the proof of the lemma implies that ht,p(ei9 ) = -00 for 0, :::: 0 :::: 02.
5.2.4. Lemma. Let 0, < O2 < 03, 03 - 0, < 1l / p. Let g (0) = a cos pO + b sin pO
be such that ht,p(ei91 ) :::: g(O,), ht,p(eilh.) ~ g(02). Then
ht,p(e'"93) ~ g(03).
Proof of Lemma 5.2.4. We argue by contradiction. Let 8 > 0 be such that
ht ,p(ei 9)) :::: g(03) - 8. Consider the auxiliary function
which is impossible. D
The same proof shows that if ht,p(eilh.) ~ g(02) and ht,p(ei9)) :::: g(03), then
ht,p(ei91 ) ~ g(O,).
The rest of the proof the continuity in Proposition 5.2.2 is analogous to that
of Proposition 1.1.15. D
5.2.5. Proposition. Under the same hypotheses as Proposition 5.2.2, for any
e > 0 there is a constant C e > 0 such that,for all Z E C,
It follows immediately that there is a constant C(e, zo) > 0 such that for every
z E C· so that z/Izl E B(zo, al) one has
log If(z)1 ~ (hf,p(z) + elzl P ) + C(e, zo).
It is easy to finish the proof using the compactness of the unit circle. 0
h(z) := II (1 - ;j) .
j~l
4~ 2j <
- 2
j - !4<- Iz I -< 2j + !2-2
< ~ 2j •
Hence,
Ih(z)1 ::: 2-
j
II 11 - I::: 2- II
Zk
j
(i 2j - k - 1) II (1 - ~ 2j - k )
4 k<Fj 2 4 l~k<j bj
II II 2
j-3
> -C.- (3- 2J.- k - 1) > -C k > Cl > 0,
- 2J+2 4 - 32 -
l~k<j k=l
In fact, there is at most one no~ro term for each z, so g is a Coo function.
Moreover, g is constant on every disk B(2i , !).
For the Coo function f =
g - h v to be entire we must have
of og ov
o = oi = oi - h oi .
The function
og (z)
_1_. for z ¢ U B (7 j , !) ,
w(z) = { ~(Z) oi j~l
fQr z E U B (2 j , !) ,
j~l
5.2. The LindelOf Indicator Function 319
:!
is a Coo function. Furthennore, for some constant K > 0 we have
" . log2.
~ 1 + (log 2) L- (h + 2 - J) = 1 + -2-(} + 1)(}. + 1)
l~k~j+1
~ C2 (log2(1 + Izl) + 1)
for some C2 > 0, and this estimation holds everywhere if C2 is sufficiently
large. Let
q(z) := p(z) + C2(log2(1 + Izl) + 1) 2: p(z)
1
then
Iv(z)12 -2q d
----'::--::-e m < 00.
c (1 + IZ12)2
If(z)1 = I~
7r
r
18(z.1)
f(~)dm(~)1 ~ ~
7r
r
18(z, l)
If(~)le-ql(~)eqIR)dm(n
320 5. Summation Methods
The only thing to verify is that the Ue are subharrnonic. Let us show first that
for each r > 0 the function
the function q;o(z) = q;(z) := u(ZI/p) is convex. The previous proposition allows
us to assume first that u is Coo in C*.
From the homogeneity of degree 1 of q; we conclude that
aq; aq;
x-+y-=q;,
ax ay
a2q; a2q;
ax2 + ay2 ::: O.
Hence,
322 5. Summation Methods
For a given function cp, subharmonic and homogeneous of degree p > 0, let
uS introduce the Banach space Btp of all entire functions f in C such that
If(z)le-tp(z) = 0(1)
with the norm IIflltp = sUPzeC If(z)e-tp(z) I.
Let CPn(z) := cp(z) + IzlP In, then we also consider
Etp = n
n~l
Btpn'
this space of holomorphic functions has countably many norms IIflitp. and it is
a Frechet space with the topology defined by these norms.
It follows from Proposition 5.2.5 that f E Etp if and only if
implies
Un(Z) ~ A(u n , z, r),
En ~ E""
since I E En implies that hU(z) ~ un(z) ~ l/I(zn) - lin for Z E B(zn, lin),
and we know from Proposition 5.2.5 that there is a function In E E",.such that
hU(zn) = l/I(zn). Moreover, the inclusion map
in: En ~ E",
Let lEE", \ Un> I En, we claim that hf,p = l/I. If not, let ~ be such that
=
hf,p(~) < l/I(~), I~r 1. Choose 8 > 0 so that hf,p(~) < l/I(~) - 8. By conti-
nuity, there is a 8> 0 so that hf,p(z) < l/I(z) - 8/2 for all Z E B(~, 8). There is
a point zno of the sequence such that zno E B(~, 8). Since the same point appears
infinitely often, there will be an index n I ~ no such that
and
2. Let f be a holomorphic function of order p and finite type in the angular region
=
81 ::: argz :::~, ~ - 81 < '!rIp, and h hl,p, the indicator function of f of order p.
(a) Show that for 81 < 8 < 82 the following inequality is correct:
h(8) - h(81) h(~) - h(8d
-,-:......:.....--:....::...<----
sinp(8 - 8t> - Sinp(82 - ( 1)
8 -
+h(81)sinp ( -2-
2 9) seep (9- 2 -
2 - 81 ) seep (9-2-
-81 ) .
324 5. Summation Methods
Conclude that
() 1-+ h«(}) - h«(}I) + k«(} - (}I)
sin p«(} - (}I)
is an increasing function. Deduce the existence of the right and left derivatives h~«(}I)
and h'-
«(}t>.
(c) Show that h'-«(}I) :::: h~«(}I).
(d) Show that if (}o is a local maximum of h and I() - (}ol :::: 7f / P then
h«(}) ~ h«(}o) cos p(}o.
3. (a) Let I be an entire function of order 0 < P < 1 and finite type, so that
i(z) = cz m n (1 - -=-)
k~1 ak
colzl m n (1 -J:L)
k~1 lakl
:: I/(z)1 :::: colzl m n (1 + J:L)
k~1 lakl
.
(b) Let
and let hi be the indicator function of order P of II. Show that hi (7f) = SUP8 hi «(}).
(c) Let m(r) = min{l/(z)l: Izl = r}, M(r) = max{l/(z)l: Izl = r}.
Show that
.
11m log m(r) 1. log III (r)1
sup ~ 1m sup ::---=:"":':--:--'--'--7
HOO 10gM(r) r_oo 10gl/l(-r)1
4. Let I
be a holomorphic function of order p and finite type in the angular sector
f3. Assume -7f < a < 0 < f3 < 7f.
a:::: Argz ::::
(a) Show that for every y > 0 there is a sequence rn --+ 00 such that
(c) Let rpn(z) := I(rn + z)/I(rn). Show that for Izl ~ 2e8r,
log Irpn(z)1 ~ 3y(rn + Izl)p(rn+1zi).
Using the Minimum Modulus Theorem ([BG], [Lev]), infer that, for 0 < w < 1 and
H(t) := 2 + log(3e/2t), t > 0, the inequality
holds in Izl ~ 8rn , outside an exceptional set of finitely many disks, the sum of whose
radii does not exceed w8rn.
(d) Conclude that in the interval (1 - 8)rn ~ r ~ (1 + 8)rn one has the inequality
N(r) = _1_
rP 2:n:r P
r
10
log I/(reiB)1 de _ log 1/(0)1
rP
and conclude that
lim sup -
r-->oo
N(r)
-
rP
~ -1
2:n:
1 0
2"
hf(e) de.
lim sup -
n(r)
= 1= -
ep 121f hf(e) de.
r-->oo rP 2:n: 0
6. Let 1 be an entire function of order p, 0 < p < 1,/(0) = 1. Assume the only zeros
of 1 are simple, located at z = -rn , 0 < r1 < r2 < "', and such that for some A > 0,
n(t) ~ AlP as t ~ 00,
326 5. Summation Methods
C\] -
n
(a) Show that for z E 00, 0[,
1 z 1
o
00
n(t) - At P I
t(t + z)
dt < Izl
-
lA
0
n(t) + (A + e)t P dt + elzl
tit + zl
1
0
00 tP
- - dt.
tit + zl
(c) Prove that using the principal branch of zP
z 1 00
- - dt
o t(t + z)
tP
= 7rZ P coseC7rp.
(d) Conclude that for -7r < () < 7r one has
Log(f(re iO » ~ eip°7rA(cosec7rp)rP
as r -+ 00.
5.3.1. Definitions.
(1) For a nonempty subset K of C let
Hp,K(Z) := sup Lp(~z)
t;eK
(v) ( UAi)*P =
leI ieI
A? n
Proof. Item (i) and the first part of (ii) are evident. From them it follows that
«A*P)*P)*P £ A*P. Moreover, from the first part of (ii), A*P £ «A*P)*P). So
(ii) holds. Part (iii) is a consequence of the homogeneity of Lp.
It is clear that
R E C: Lp(~z) = 0, Vz E A} £ R E C: Lp(~z) < 1, Vz E A} = A*P.
On the other hand, if ~ E A*P and aA £ A, then for a > 0 and Z E A we have
Lp(~az) < 1,
so that
Since in part (iv), a > 0 is arbitrary, we obtain that Lp(~z) ::'S 0 for every ~ E
A*P, Z E A. As pointed out in Lemma 5.1.10, for p > 1, the function Lp ~ 0,
hence we conclude that A*P £ {~ E C: Lp(~z) = 0, Vz E A}.
From Ai E UjeI Aj and (i) we conclude that neI A? ;2 (UieI Ai)*p. Con-
versely, if ( E niEI A? and Z E UiEI Ai, then Z E Aj for some j E I and, hence,
Lp«(z) < 1 since ~ E A? Therefore, ~ E (Uiel A;)*P. D
Let us remark that the definition of p-convexity implies that every p-convex
set contains the origin and that, for every K £ C one has
5.3.3. Proposition.
(i) If K is p-convex, then K = (K*P)*P and, moreover, if K is also compact,
then
K*P =R E C: Hp,K(n < I}.
(ii) If K is compact and K = (K*P)*P, then K is p-convex.
(iii) The intersection of an arbitrary number of p-convex sets if p-convex.
(iv) If K =j:. 0 there is a smaller p-convex set containing K. It is called the
p-convex hull of K and denoted cVp(K). Moreover,
cv p(K) = {~ E C: Lp(z~) ::'S Hp,K(Z), Vz E K}.
328 5. Summation Methods
(v) For any two closed p-convex sets KI and K 2, the inclusion KI ~ K2 is
equivalent to Hp.K :::: Hp.K2 .
(vi) For any nonempty bounded set K the/unction Z 1-* Hp,K(Z) is continuous.
Proof. (i) From the previous proposition we know that K ~ (K*P)*P. Assume
that K is p-convex. If there was I; E (K*P)*P such that I; rt K, then one could
find Zo E C such that Lp(l;zo) > Hp,dzo). Hence Lp(l;zo) > 0 and
Therefore,
Zo
- - - - - : - ; - E K*P,
(L p(I; zo» II p
whence, as I; E (K*P)*P, we have
Let 1;0 rt (K*P)*P. There is then Zo E K*P such that Lp(zol;o) ~ 1. On the other
hand, since Zo E K*P we have Lp(zol;) < 1 for alII; E K. By the compacity of
K it follows that Hp,K(ZO) < 1. Hence 1;0 rt A.
(iii) Let K = niEi K i , all Ki being p-convex. Let A be the set defined in (ii).
We need to show that A ~ K. Let I; E A, then for any z E C, i E I,
(z E C),
The function ({J is hence continuous and differentiable except at to, where it has
right and left derivatives. Clearly
It follows that
and so
hence
which shows that Hp,K is Lipschitz on any bounded subset of C. The continuity
is now obvious. 0
330 5. Summation Methods
E- := S2\ {I ~: Z E }
E ,
Proof. Let us define Az := {w E C: Lp(wz) < I}. First, let us show that if ~
A z • If W rt A" then Lp(wz) :::: 1, hence wz =f. O. Let 1: = 1/(wz), ~ = n, then
Lp(zg) = Lp(l/r) ::::: 1, which means ~ E Kf. Since w~ = rwz = 1, we have
W ¢ if.
UKf = (B*P)-.
zeB
G= (U Kf) - nif
zeB
=
zeB
= (V E c: Lp(~z) < 1, Vz E B}
= B*P.
Hence,
o
5.3. The Fourier-Borel Transfonn of Order p of Analytic Functionals 331
K= U Kf.
ZEK*P
where y is any piecewise C' loop in K C , which has index 1 with respect to K
and lies in the domain of holomorphy of h.
Proof. Since T is carried by K, and, for 8 > 0, K s/2,p is a neighborhood of K,
there is a constant As > 0 such that
where 8(e') = e' SUP~EK£/2.p I~IP. Choosing e' > 0 so that 8(e') < el2 accom-
plishes the desired estimate.
Conversely, consider the integral
<;o(t) := P 1 00
e- t- Pt p- 1 f(t~) dt.
so that
e- tP tp-llf(t~)1 ::; Ce t p- 1e-(1-v)t p •
We conclude that K*P ~ Bp(<;O) = Mp(<;O) (see Section 5.1). Therefore we have
Kf ~ Bp(<;O) for any z E K*P. It follows from Proposition 5.1.15 that <;0 has an
analytic continuation to UZEK*P Kf = K.
Now let y be a piecewise C 1 loop of index 1 with respect to K and contained
in the domain of holomorphy of h E JIf(K). We know that <;0 admits the Taylor
development
<;O(~) = .!. L r
P n":O
(1 + ~) an~n
P
~
p
(T)(z) =~ (_I_.j
n~o 2:rr1 y
r(1 +nlp) an~n-m-ld~) zn
r(1 +mlp)
= Eanz n = f(z),
n":O
as we wanted to prove. o
5.3.11. Corollary. Let P :::: 1 and T an analytic functional in C, then there is a
smallest p-convex carrier of T.
5.4. Analytic Functionals with Noncompact Carrier 333
Proof. For p = 1 it was done in Theorem 1.3.5. For p> 1, let K be the
intersection of all p-convex carriers of T. If ~p(T) = En>oan~n, consider
1/f(~) = En>or(1 +n/p)ang n+ l • Then 1/f has an analytic continuation to K C •
Therefore, we can represent T by
(T, h) = -2
1 .1
7r1 Y
h(~)1/f(~)d~, h E .1t'(K),
EXERCISE 5.3.
1. Use Definition 1.3.12 to show that for p = 1 and T E .1t'(I(:),
zeL,
(1n n1)
limit
Q(L,k):=
I
hm
•
Q L, -,k,I - ,
~
n .... oo
334 5. Summation Methods
so that Q(L, k') is a DFS space. Denote Q'(L, k') the dual space of Q(L, k').
An element of this space is, by definition, an analytic functional carried by
L and type::::: k'. Recall that a linear functional 8 on Q(L, k') is continuous if
and only if, for every s, s' > 0 there is C = C(s, s') ~ 0 such that
1(8, f)1 ::::: C sup If(z)e(k'+e')zl
zeL,
for every f E Q(L, s, k', s').
We shall frequently use the following version of the Phragmen-Lindelof
principle:
- - _ . R(L,r,s,k',s')
1 1
R(L r [, k', s') - - - - . - . R(Lr,k',s')
5.4. Analytic Functionals with Noncompact Carrier 335
The horizontal arrows are compact maps. We consider now the projective limits
R(C\L; k', s') := lim R(L, r, s, k', s')
(
r----+-oo,k' ,£----+0
These two spaces are FS. Lemma 5.4.1 shows that the second one is a closed
subspace of the first.
If 0 < s; < s', we have the commutative diagram of continuous linear injec-
tions
R(C\L; k', si) - - _ . R(C\L; k', s')
1 1
R(C; k', si) ----~, R(C;k',si)
This diagram induces a continuous linear mapping among the quotient spaces
R(C\L; k', si) R(C\L; k', s')
----~---+ .
R(C; k', si) R(C; k', s')
because
e-(e'-e;)lwll F(w)e-(k'+e;)w I ::: e-(e'-e;) Re w!F(w )Ie-(k'+<;) Re w
::: IF(w)e-(k'+<')wl.
where the maps needed for the projective limit are the earlier maps (*).
An element [F] of H (C, k') is determined by a family of functions (Fe' )e'>O,
where Fe' E R(C\L; k', e') and if 0 < e~ < e', then -
Fe; - Fe' E R(C; k', e').
Let us assume 1 E Q(L, k') and [F] E H(C, k') are given. Choose eo, eo> 0
such that 1 E Q(L, eo, k', eo), then for 0 < e < eo, 0 < e' < eo, the integral
JaL.
r l(w)Fe,(w) dw, Fe' E [F],
and, if we let
B:= sup IF(w)e-(k'+e')wl < 00,
weL2e \L.
then
IF(w)1 .::: Be(k'+e')Rew,
This reasoning not only implies the absolute convergence of the integral, but it
can be used to show that its value is independent of e and e'. In this way any
[F] E H(C, k') defines a continuous linear functional on Q(L, k') denoted by
Int([FD,
(Int[F], f) := - r
JaLe
l(w)Fe,(w) dw.
[F] ~ Int([FD,
We are going to verify that the map Int is bijective by exhibiting the
inverse map.
k £
I
a-£ o
Figure 5.6
5.4.6. Definition. For S E Q'(L, k') its Cauchy transform is the family 8 of
functions (8e, )e'>O defined by
1 ( e(k'+e')(w-z) )
= --2
A
Se'(w) . Sz, .
HZ W - Z
5.4.7. Proposition. Let 8 E Q'(L, k'). For any e' > 0 the function 8e, is holo-
morphic in C\L, and for every e > 0 it satisfies the inequality
sup 18e, (w)e-(k'+e')w 1< 00.
w¢L,
Proof. The functional S can be represented by a convenient measure. Using
Lemma 5.4.4 and the theorem of Morera it is easy to conclude that the function
Se' is holomorphic in C\L. Furthermore, for every s > 0, e' > 0, there is a
=
constant C C(e, s') ::: 0 such that for every f E Q(L, e, k', e')
I(S, f>1 s C sup If(z)e(k'+e')zl.
zeL,
338 5. Summation Methods
and thus,
I
sup ISe'(w)e-(k'+s'lw = sup _1 I( St. _e-(k'+e')Z)
_ __ <-
C
w¢L2' w¢L2e 2:rc w- z - 2:rc8'
wE C\L,
admits an analytic continuation to an entire function satisfying the estimate
In other words, z ~ G(w - z) belongs to Q(L, 28, k', 8"). Thus, the function
1
F(w) = - - . (Sz. G(w - z)}
2:rc1
is defined for w E Le and, by the continuity of S, satisfies
sup W(w)! ::: Cle(k'+e'lu + C2e(k'+e"lu,
weL,
5.4.9. Definition. The Cauchy transform C(S) of S E Q'(L, k') is the element
[S] E H(C, k') defined by the family S = (Se)e>O.
5.4. Analytic Functionals with Noncompact Carrier 339
(S, f) = -1 aLB
f(W)Se,(w)dw,
w~---
w-z
belongs to Q(L, So, k'-oso). In fact, it is continuous in Leo, holomorphic in to
(even holomorphic in L2eo)' and one can verify that
-(W-Z)2 I
sup 1e e(k'+e')w < 00.
weL.o W - Z
(t) 1 aL.o
(Ss'(W) - Fe,(w»
e-(w-z)2
w- z
dw = O.
1
Let us introduce an auxiliary function G, holomorphic in L4so' by
I A e-(w-z)2
G(z) := -2' (Ss'(w) - Fe,(w» dw.
7r1 aL4<o w- z
If z e L3so \L2so' we can apply (t) and Cauchy's theorem to obtain
G(z) = Ss,(z) - Fs'(z).
340 5. Summation Methods
A 0
Thus, the function Se' - Fe' admits an analytic continuation G e' to L4eo by G,
hence G e' is an entire function. Moreover, one can prove without difficulty that
le-(k'+e')zGe,(z)1 remains bounded in L3eo. It follows that G e, E R(C; k', e'),
which finally proves that
C(lnt[FD [F]. = o
We are now going to extend the Fourier-Borel transform to Q'(L, k').
5.4.12. Lemma. The function z ~ eZ~ belongs to Q(L, k') if and only if
Re~ < -k'.
Proof. It is elementary and left to the reader. o
5.4.13. Definition. The Fourier-Borel transform of SEQ' (L, k') is the func-
tion ~(S) defined in the half-plane Re ~ < -k' by
for every e, e' > 0, where HL is the supporting function of the half-strip L.
It is easy to see that letting ~ = ~ + i T/
+oo if ~ > 0,
Hd~) = supRe(~z) = { a~ - klT/ if ~ :::: 0, T/ ~ 0,
zeL a~ - k2T/ if ~ :::: 0, T/:::: o.
°
Proof. Let S E Q'(L, k'), hence for every e, e' > there is a constant C =
C(e, e') ~ such that for ~ = ~ + iT/, ~ :::: -k' - e',
I~(S)(~)I = I(Sz, eZ~)1 :::: C sup lez~e(k'+e')zl
zeL,
Ce(Hk'+e')(a-e)-kl'l if T/ ~ 0,
{
:::: Ce(Hk'+e')(a-e)-k2'1 if T/ :::: o.
This estimate is precisely of the form
I~(S)(~)I :::: c' eHd~)+el~l, Re ~ :::: -(k' + e').
We are now trying to prove that ~ is a bijection between the two spaces consid-
ered above. Let F E Exp(] - 00, -k'[ + ilR) and fix two complex numbers ~o =
~o + iT/o,~' =~' + iT/', such that ~o < -k', I~'I = 1,~' :::: O. The ray starting at
5.4. Analytic Functionals with Noncompact Carrier 341
Figure 5.7
342 5. Summation Methods
then
wen = UW£(n.
£>0
5.4.16. Lemma. For any ~o E C, ~' E C, I~'I = 1, Re~' :s 0, the function given
by w 1-+ F(w, ~o, n
is holomorphic in the open half-plane Wen.
U W«() = C\L.
1~'1=1
Re~'~O
r F(r)e- rW dr = r F(r)e- rW dr
J~o+~'oo J~oH"oo
as we wanted to prove. o
5.4.18. Lemma. Let ~o = -(k' + 8'). One can define a holomorphic function
F(w, ~o) in C\L by
G(w) = -.j
1
27T1
-(k'+e")
-(k'+e')
F(r)e- TW dr. o
5.4.20. Definition. For F E ExpO - 00, k'[ + ilR, L) we define for every s' > 0
a function B(F)e' by
The family (B(F)e')e'>O determines an element B(F) of H(C, k'), called Borel
(or Laplace) transform of F. Denote
5.4.21. Example. The function F(~)=ezl; belongs to ExpO - 00, -k'[ +iR L)
if and only if Z E L. The Borel transform of F is the "Cauchy kernel" used in
Definition 5.4.6,
1 e-(k'+e')(z-w)
B(F)e'(w) = -.
27T1
-- --
Z - W
(w E C\L).
Jo Int oB = idEXp(]_oo,-k'[+ijR,L)'
Proof. Let F E ExpO - 00, -k'[ + ilR, L), S = Int([B(F)]). From the defini-
tion of Int we have, for Re ~ < -k' - s',
J(S)(~) = - r
JaL,
B(F)e,(w)e-wl; dw.
Decompose the path aLe into three parts following Figure 5.8.
Denote ~o = -k' - s', and since in the portion I, we have 1m w = k2 +s >
k2, we can let Fe,(w, ~o) = F(w, ~o, -i). Then
344 5. Summation Methods
W2~---+----~-----------------
IT W2 (a - £) = + i(k2+ E)
m
Wl~---+-----+-----------------
Figure 5.8
=~ lew~
27r1 I
(1 ~o-ioo
F(r)e- WT dr) dw
1
27r I ~o-ioo I
1 e{~-T)Wl
= ----. F(r) dr.
27r1 ~o-ioo ~ - r
Similarly, if 1m ~ i- 0 we have
f e~w F(w, ~o) dw = f e~w F(w, ~o, -1) dw
~I ~
= ---
1 1 e{~-T)Wl - e{~-T)W3
dr.
27ri ~o-l.oo ~ - r
1 1
Thus, for Re ~ < -k' - 8', 1m ~ i- 0, we have
1 e{~-T)Wl 1 e{~-T)W3
J(S)(~) = ---. F(r) dr + ---. F(r) dr,
27r1 C1 ~ - r 2m . Cz ~ - r
where Clo C2 are the paths suggested in Figure 5.9.
5.4. Analytic Functionals with Noncompact Carrier 345
k' E'
Figure 5.9
1
An application of Cauchy's theorem yields
I e(~-~)Wl -F(~) if Im~ < 0,
°
{
- F(r) dr =
I1
27r i C1 /; - -r if 1m/; > 0,
- F(r)
e(~-~)W3
dr =
{O if Im~ < 0,
27ri C2 ~ - r -F(~) if Im~ > 0.
Therefore,
J(S)(~) = F(~)
whenever Re~ < -k' - s'. This concludes the proof. o
5.4.23. Proposition. The following diagram is commutative. Each arrow is a
bijective linear map:
Q'(L, k')
~ c • EXr - 00, -k'[ + fIR, L)
.... ~ H(C,k').
(In fact, all the maps are topological isomorphisms [Mol] [Mo2].)
346 5. Summation Methods
Proof. We have already shown that C and Int are inverses of each other and
JoInt B = id. It is enough to show B J = C to conclude the proof. Let
0 0
S E Q'(L, k'), F(~) = J(S)(~), ~o = -(k' + 8'). For W E W(~') one has
F(w,
- ~o) = -.
1
2rrl
1
~o+~'oo
F(r)e-W'I: dr = -.
1
2rrl
1~oH'oo
(Sz, eZ'l:)e-W'I: dr
= _1_ / S 1
2rri \ z, ~oH'oo
e'l:(Z-W) dr) = __1_
2rri
(s "
e-(k'+e')(Z-W»)
z- w
= -Se'(w). 0
We are now going to extend the results of Sections 4.1 and 4.2.
Let 0:::: k' < 1, ~ E e- L , then the function Z 1--* 1/(1 - ~eZ) belongs to
Q(L, k') (as the reader can easily verify). Thus, for any T E Q'(L, k') we can
define a function G(T) by
G(T)(~) := / Tz , 1 ) ,
\ 1 - ~ez
which is hence defined for ~ E Q (L) := C\e- L and has the following properties:
5.4.25. Proposition. Let L = [a, oo[ + i[kb k2] with k2 - kl < 2rr and let 0 ::::
k' < LIfT E Q'(L, k'), then for every 8 such that 0 < 28 < 2rr + kl - k2 and
every 8' such that 0 < 8' < 1 - k' there is a constant C ~ 0 with the property that
C
IG(T)(~)I :::: 1~Ik'+e'
5.4. Analytic Functionals with Noncompact Carrier 347
e-a
Figure 5.10
for every
Proof. From the continuity of T there is a constant C' :::: 0 such that if r; f/.
e- L '/2, then
1 e(k'+E')Z
IG(T)(r;)1 = I(Tz, z}1 ~ c' sup
1 - r;e zeL,/2 1 - r;e z
e(k'+e'-l)z
~ C'sup
zeL e- Z - r;
so that
inf 11 - ;ezi :::: sin(s/2).
zEL,;2
Let us keep the hypotheses of the preceding proposition from now on. Denote
Ho(n(L), k') the space of all holomorphic functions rp in n(L) such that:
(i) lim rp(O = 0; and
I~I-c>oo
(ii) sup{lrp(;Rk'+s'I: ; E A(-kl + s, -k2 + 2rr - s)} < 00, whenever 0 < 2s
< 2rr + kl - k2, 0 < s' < 1 - k'. This space provided with the seminorms
sup{lrp(;)I: 1;1 > e- a +S } < 00,
is a space of Frechet-Schwartz.
1
(i) For p > 0 the function
hp(z)
1
= -. hew) dw
2rrl aL"p 1 - eZ - W
belongs to Q(L, 1) (and hence to Q(L, k')), with the contour aLs,p =
aLs n {w:Rew::: pl.
(ii) For Z E t
h(z) = - 1 hew) dw.
2rr i aL, 1 - eZ - w
1
5.4. Analytic Functionals with Noncompact Carrier 349
11 h(w)
- - - d w :::: C
aEe(P) 1 - e Z - W
I e-(k'+e')Rew
aEe(p) 1 - e Rez - p
dlwl,
and it is easy to see that the right-hand side converges to zero as p --+ 00.
(iii) We need to show that if Ye,p = aLe \aLe,p, then
1 Y€,p
_h(_w_)_ dw------+O
1 - eZ - W p-+oo
in Q(L, k').
sup e(k'+e'/2)z
e
ZEL /2
1 h(~~w dw
Ye,p 1 - e
sup
e
ZEL /2
1 Ye,p
h(w)(e- Z - e- w)-k'-e'/2(1 - e Z - W )-l+k'H'/2 dw
350 5. Summation Methods
~ CI Ye,p
Ih(w)le(k'+e'/2)Rew dlwl
<
-
c'l Ye,p
e-e'/2(Rew) dlwl
p ..... 1t o. o
5.4.28. Proposition. Let T E Q'(L, k'), h E Q(L, k'). Choose 8, 8' > 0
sufficiently small so that 0 < 28 < 2rr + kl - k2' 0 < 8' < 1 - k', and h E
Q(L, 8, k', 8'). Then
(T, h) = 21. r
rrl JaLe
G(T)(e-W)h(w) dw.
_1 r /
2rri JaLe,p \
Tz , 1
1 - eZ -
\ h(w)dw
W /
= (Tz, _1 r h(w)
2rri JaLe,p 1 - eZ - W
dW).
_1_.
2m JaLe
r G(T)(e-W)h(w) dw
by Proposition 5.4.25. o
As a consequence of this proposition, we can recover T from G(T). We can
apply this to obtain a generalization of the Carlson theorem (see Theorem 4.1.6).
5.4.29. Theorem. Let 0 ~ k' < 1, L = [a, oo[ + i[k 1 , k 2], k2 - kl < 2rr. If the
function F E Exp(] - 00, -k'[ + ilR., L) is such that F( -n) = 0 for all n E N*,
then F == O.
Proof. There is T E Q'(L, k') such that ~(T) = F. The hypotheses and the
connectedness of Q(L) imply that G(T) == O. Thus, T = 0 and F == O. 0
EXERCISES 5.4.
1. For L = [a, oo[ + i[ -Jr /2, Jr /2], k' E ~, define a functional T by
(c) Show that for any R E JR, M > 0, 6 > 0, 6 > 0, there is c > such that
3. (a) Show that for ({J E 'H.o(O (L), k'), L = [a, oo[ + i[ -k, k] (a E JR, k > 0), the
1
integral
M«({J)(z) := - .
1 -((J(W) dw
211"1 8(exp(-L.» wz+1
c+;oo Irnz
-5 -4 -3 -2-1
Rez
c -;00
Figure 5.11
352 5. Summation Methods
I 1
---:-
C
+iOO _.F(z)
__ (_w)Z dz (k < largwl ~ rr),
={
1~(z)
M-1(F)(w) 21 c-ioo smrrz
-~ (-w)' dz
21 r smrrz
where r is the path suggested by Figure 5.11 and -1 < c < -k'.
Show that for every 8 > 0, 8' > 0, there is a constant C > 0 such that, with x = Re z
and y = Imz,
F(z)(-w)Z I < C exp«a - + (k + 8)lyl + x log Iwl - y arg( -w».
I
8)X
sinrrz - (sin2(rrx) + sinh2(rry»1/2
Conclude that the two integrals defining M- 1(F)(w) are holomorphic functions in their
respective regions. (Note, so far there is no claim that M- 1(F) is well defined.)
(c) Prove that for every 8 > 0, 8' > 0, there is a constant D > 0 such that
Conclude that the two integrals defining M- 1(F)(w) coincide whenever k < I arg wi ~ rr
and Iwl > e- a •
(d) Show that for Iwl > e- a
M-1(F)(w) =- L F(-n)wn
n~l
Harmonic Analysis
The ongms of Hannonic Analysis lie in the work of Euler [Eu] and the
Bemoullis who proposed to write periodic functions in tenns of the exponen-
tials e inx , nEZ, in their study of the vibrating string. It is known that every
COO-function which is 2rr-periodic in the real line has an expansion of the fonn
2::;:"00 aneinx (we remind the reader one can estimate these coefficients an very
precisely, and that we do not need to restrict ourselves to COO-functions). It was
the work of Fourier [Fo] on heat conduction that showed, once and for all, the
importance and the interest of such expansions, and since then they have been
called Fourier expansions. It is clear that another way of saying that a function
! is periodic with period r is to say that ! satisfies the convolution equation
(8 r - 8) *! = O.
11x+r/2
IL*!(X)=- !(t)dt=O,
r x-r/2
where IL = (ljr)X[-r/2,r/2). This function! is periodic of period r (as it can be
seen by differentiation of the above convolution equation) and has zero "mean"
(average) over any interval of length r, whence the name mean-periodic.
353
354 6. Harmonic Analysis
6.1.1. Definition. Let 0 =f:. JL E e'(R). We say that a function f E e(R) is JL-
mean-periodic (or simply mean-periodic) if it satisfies the convolution equation
JL * f = O.
6.1.2. Proposition. The set VJt = (f E e(R): JL * f = O} is a closed subspace
of e(R) which is invariant (under translations).
Proof. We first recall that given hER one defines the translation operator rh
by rh(f)(x) = f(x - h). To say that VJt is invariant under translations means
that rh (VJt) ~ VJt for every hER
Since rh (f) = Oh * f, the fact that VJt is invariant is clear. The other prop-
erty is also immediately apparent from the fact that the convolution operator
JL*: e(R) -+ e(R) in continuous. 0
6.1.3. Proposition. If JL =f:. 0 and JL =f:. COa for any a E Rand C E C*, the space
VJt = Ker JL* is a proper invariant subspace of e(R).
Proof. First we remark that VJt ~ Ker jJ" where jJ, is considered as a continuous
linear functional on e(R) (and not as a convolutor). This follows from the
definition of convolution:
0= JL * f(O) = (jJ" f).
Hence, if JL =f:. 0, then VJt =f:. e(R). On the other hand, if JL is not of the form
coa , then the entire function FJL has zeros [BG, §4.S.11]. Let J... be such a zero
and f(x) = exp(iJ...x), then
JL * f(x) = FJL(J...)f(x) = o.
Therefore VJt =f:. {O}. o
6.1.4. Remark. This proof shows that the exponential eiJ..x of frequency J... lies
in VJt if and only if J... is a zero of F JL. There are other simple functions in
VJt, namely, the exponential monomials xkeiJ..x, when J... is a zero of FJL of
mUltiplicity m > k. In fact, one has the identity
k
(JLr, (x - tleiA(x-t)} = eiJ..x 2: (~) x k- j (-i)j (JLr, t j e- iAt )
j=O }
= eiAX t
j=O
(~) x k-
}
j (-i)j (F JL)(j) (J...).
6.l. Convolution Equations in ~ 355
6.1.5. Proposition. Let rot be a closed invariant subspace of £(JR), and let
(vJt).L = {IL E £' (JR): (IL, /) = 0, 'If E rot}. Then (vJt).L is a closed ideal of the
convolution algebra £'(JR). Moreover, (vJt).L = {IL E £'(JR): IL * f = O,for all
f E rot}.
If'J is a closed ideal of £' (JR), then (J).L = {f E £ (JR): (j'L, f) = 0, 'IlL E 'J} is
a closed invariant subspace of £(JR) which coincides with the subspace of £(JR)
given by {f E £(JR): IL * f = 0, 'IlL E 'J}.
Furthermore, under these conditions, (((vJt).Ln.L = rot and (((J).Lf).L = 'J.
Proof. It is evident that (vJt).L is closed even if rot were not closed. Once we
*
show that (vJt).L = {IL E £'(JR): IL f = 0, 'If E rot} is true, it becomes clear
that (vJt).L is an ideal of £'(JR).
Let IL E (vJt).L, f E rot then
(IL * f)(x) = (ILl, f(x - t»)= (ILl, (Lx (f)f(t») = 0,
since LAf) E rot. Hence (vJt).L £ {IL E £'(JR): IL * 1= 0, 'If E rot}. The other
inclusion is evident.
We also have, for IL E 'J and f E (J).L,
'CAin = (IL * 'CAf) (0) = (IL * Ox * f)(O) = (IL * ox, /) = 0,
(IL,
since IL * Ox E 'J for every x E R Hence 'Cx (f) E (J).L. The identification of
(J).L is similar to the previous case.
The final statement is a consequence of the Hahn-Banach theorem [Hor]. 0
Now let 0 =f:. IL E £' (JR) and consider rot = Ker IL*. We can identify (vJt).L
immediately as follows. Let 1= F«vJt).L). Then I is a closed ideal in the
algebra Ap(C). From Remark 6.1.4 we know that if (A, m) E V(FIL), the multi-
plicity variety associated to FIL, then xke iAX E rot for 0:::: k :::: m - 1. Hence,
if v E (vJt).L we must have v * xke iAx = O. This implies that Fv vanishes at
the point A with multiplicity at least m. Therefore Fv = </>FIL for some entire
function </>. In the terminology of Chapter 4, if we let f be the algebraic ideal
in Ap(C) generated by FIL and let (FIL) be the algebraic ideal generated by
FIL in Jf(C), we see that 1= Ap(C) n (FIL). In other words, I = floc.
If we define rot o as the invariant closed subspace generated by the expo-
nentials monomials xke iAx , with 0:::: k:::: m -1 and (A,m) E V(FIL), then the
preceding argument shows that 10 := F«vJto).L) coincides with floc. The reason
is that to show that I = floc we only used the exponential monomials in rot. It
follows, from Proposition 6.1.5 that roto = rot. That is, the exponential polyno-
mials solutions of IL * f = 0 are dense in the family of all solutions.
356 6. Hannonic Analysis
Lemma 6.1.8 and Theorem 6.1.9 extend this result to arbitrary closed invariant
subspaces rot Before we proceed, let us point out that there is another way to
obtain invariant subspaces.
6.1.6. Definition. Let I E [(lR). We denote by 'r(f) the closure of the space
spanned by rx(f), x E R
This space is the smallest closed invariant subspace of [(lR) containing I.
=
Proof. It is clear from the previous Proposition 6.1.5 that 'r(f) [(lR) if and
only if the ideal ('r(ff)1. is the zero ideal. Ifnot, let 0 =j:. f,L E ('r(ff)1.. Using
the same proposition we conclude that f,L * I = O. 0
We have just seen that the exponential polynomial solutions of a single convo-
*
lution equation f,L I = 0 are dense in the family of all solutions of the same
equation. Of course, we could not hope to obtain Fourier-type expansions of
the solutions unless this density property were true. We will show in detail how
to accomplish this expansion in Theorem 6.1.10 under some restrictions on f,L.
What is not immediately apparent is whether a given function could not have
two different Fourier expansions. For instance, the reader could ask himself what
happens with a function I which is both I-periodic and 2-periodic. How about
I-periodic and J"2-periodic? More generally, we could ask what happens with
the solutions of systems of homogeneous convolution equations. Is it possible
for such a system to have some nonzero solutions and, at the same time, have
no· exponential solutions?
This problem is a particular case of toe spectral analysis problem, which
asks to decide whether given a nonzero closed invariant subspace rot there is
or there is not an exponential monomial xkeiJ...x E rot. Let us denote by roto the
closure of the span of the exponential monomials in rot. The spectral analysis
problem becomes:
Does rot =j:. 0 imply roto =j:. O?
From the previous Remark 6.1.4 and Proposition 6.1.3 we know that when
rot = Ker f,L* then ooto =j:. {OJ if and only if f,L is not of the form c8a , in which
case rot = roto = {OJ.
We also have seen for rot = Ker f,L* that rot = roto always hold. This question
can be posed for arbitrary nonzero closed invariant subspaces rot. The spectral
synthesis problem for rot is:
Is roto = rot?
If this holds for every closed invariant subspace rot, we say that the spectral
synthesis property holds for [(lR). It is a famous result of L. Schwartz [Schw1]
that the spectral synthesis property holds, in fact for [(lR).
One can generalize these concepts to [(JRn) and find the surprising fact that
the spectral synthesis property does not hold for [(JRn), n ~ 2 [Gu1], [Gu2].
6.1. Convolution Equations in IR 357
We are sure the reader recognizes the relation between this question and the
Wiener-Tauberian theorem for the algebra L I (JRn ).
We need a few more definitions before proceeding to prove Schwartz's
theorem. Let rot and roto be as above, denote by I, 10 the closed ideals in
Ap(C) given by .r«vJt).L) = I, 10 = .r«Mo).L). Let V be the multiplicity
variety associated to I, it is called the spectrum of rot. We recall also that Iloc
is a closed ideal.
For a function f, its spectrum is simply the spectrum of'r(f).
From this lemma we can conclude that rot admits spectral analysis if and only
if hlC #- Ap(C), which is equivalent to the statement that V #- 0. On the other
hand, from the same lemma we can conclude that rot admits spectral synthesis
if and only if I = hoc. Namely, I = Iloc if and only if I = 10. Equivalently, if
and only if, (vJt).L = (vJto).L.
1:
Let f E I1., thus T:yf E I1. for any y E R Since PnJ-L E I we have
Therefore, 0 = T:y f (0) = f (- y) and so, f == O. This proves that I1. = {OJ and
I = £'(JR), a contradiction. 0
It is clear now that to conclude the proof of Theorem 6.1.9 we need to prove
that if cP E I then its derivative cp' E I also. Recall from [BG, §4.6.15] the
Hadamard canonical product of cp is
cp'(z) = ~ +a
cp(z) Z
+ Lmk
k?:l
(_1_ +~),
Z - Zk Zk
Therefore, if we could prove that the series (*) converges in Ap(C) it would
follow that q/ E I. This convergence might not hold, so we use the trick of
introducing an extra element of 1 in each term of the series. Namely, we rewrite
(*) as
mp(z)
(**) cp'(z) = -z- + acp(z)
+ "L.t" ( mk--
cp(z) + -cp(z)
mk + mkzCP(z))
2 - (""
L.t 2mk) zcp(z).
k::::l z - Zk Zk zk k::::l zk
This is similar to the Mittag - Leffler procedure for the expansion of merom orphic
functions. On the other hand,
liZ) cp(z) z2
mkCP(z) ( - - + - + -
Z - Zk Zk z;
= mZk ---,
- Zk z;
and
Returning to convolution equations, if we let rot = Ker J.t*, J.t =f:. 0, we have
that the spectrum of rot coincides with V = V (FJ.t). For any f E [(lR.) satis-
*
fying J.t f = 0, we can find finite sets An <.;Z(FJ.t) and polynomials Pn,J..,
).. E An, deg Pn,).. < multiplicity of ).. as a zero of F J.t, such that
where the limit is in the topology of E(lR). This is nothing other than the state-
ment 9J10 = 9J1. In fact, one can find an actual representation of f in the form of
a series of exponential polynomials in 9J10 . These exponential polynomials tum
out to be unique but the proof that this representation exists is rather delicate. It
was originally given by Schwartz [Schwl], who used it to obtain the first proof
of the Spectral Synthesis Theorem. We will restrict ourselves, for simplicity, to
a special case of convolutors JL such that FJL is slowly decreasing and V is
an interpolation variety. Our proof can be generalized to the case of arbitrary
convolutors in E' (JR) and even further to the case of several variables, althought
the spectral synthesis property is false for E(JRn ), n ~ 2. We refer the interested
reader to [BT2], [BT3], [BYl], [Eh5], [Pa], [BSt2].
6.1.11. Theorem. Let JL E E' (JR) be such that F JL is slowly decreasing and let
V = V(FJL) is an interpolation variety for Ap(C). Then, every CC'O-solution f
of the convolution equation JL * f = 0 can be written in the form
The coefficients aA,j are uniquely determined. The last series is convergent
in the topology of the space E(JR), i.e., absolutely and uniformly convergent on
every bounded interval and the same is true for every derivative of this series,
which then converges to the corresponding derivative of f.
Moreover, the coefficients aA,j satisfy the estimates
With this notation Ap(V) is the space of sequences b = (bk,t) such that, for
some C > 0
IIblic = supe-Cp(zkl Ibk,tl < +00.
k~l
6.1.12. Lemma.
(i) The space Ap(V) is an DFS space that can be represented as the inductive
limit of the Banach spaces Ap,c(V) (c > 0), where
Ap,c(V) = {b = (bk,lh~!
O="I="mk-!
: IIbli c = supe-cp(zkl (
k O="I="mk-!
I:
Ibk'lr) < +oo}.
(ii) The dual space (Ap(V»' can be identified to the space of the sequences
a= (ak,[h~! such that,for every B > 0,
O~[~mk-!
Proof. The space Ap(V) was described, as a set, as the union of the sequences
of spaces Ap,c(V) in Chapter 2 (§2.7.1). We have only to show that the
quotient map
p: Ap(C) -+ Ap(V)
I
becomes a topological isomorphism when Ap(V) is described as the inductive
limit of Ap,c(V)' It is clear that the spaces Ap,c(V) are Banach spaces and that,
362 6. Hannonic Analysis
if Cl > C2, the canonical injection Ap,cl (V) ~ Ap,C2(V) is a compact map. It is
also evident that if we restrict the value of C to a strictly increasing sequence
converging to +00 the corresponding .c~-topology is stronger that the topology
of the coordinatewise convergence for Ap(V). Then this topology is the same
as the one obtained using all values of c. It follows that this topology is DFS.
Since the map p is continuous and surjective from Ap(C) to Ap(V), and Ap(C)
is also a DFS space, it follows from the Open Mapping Theorem that p is a
topological isomorphism.
To prove the second part, let us show that any sequence a = (ak,l) satisfying
the estimates from (ii) defines a continuous linear functional on Ap (V) by
In fact, we have
I(a, b}1 :::: L L laklbkil
k:::l O~I~mk-1
Choose any c > B, then it follows that b = (bk,/) E Ap,c(V), (It is this reasoning
that allows maxI lak,ll and LI lak,ll be interchangeable in the definition of
(Ap(V))'.) Therefore, all the finite sequences
satisfy
= E eBp(zkl E /ak.!/
1::;I::;n O::;I::;mk-1
by definition of f3kl. Thus
E eBp(zkl E /ak.!/ < +00.
k::; 1 O::;I::;mk-1
This shows that the sequence a = (ak.!) satisfies the desired estimates. It also
follows that, for any b E Ap(V), we have
6.1.13. Lemma.
(i) la: (Ap(V»' -+ £(JR.) is a topological isomorphism onto its image.
(ii) 1m la = (Kera)-L = {f E £(JR.): J1, j = OJ. *
Proof. We know that a: £'(JR.) -+ Ap(V) is a continuous surjective map between
two DFS spaces. It follows from Section 1.4 that 1m la is closed and la is a
topological isomorphism from (Ap(V»' onto 1m la.
The equality 1m la = (Kera)-L is a result from functional analysis (see [Hor]).
On the other hand, .r(Kera) = I, hence Kera = J1, £'(JR.). From Proposition *
6.1.5 we get
(Kera)-L = {f E £(JR.): J1, * j = OJ. o
As we observed at the beginning of the proof, when J1, satisfies the hypotheses
of the theorem Lemmas 6.1.12 and 6.1.13 imply that for any J1,-mean-periodic
function I there is a unique sequence a E (Ap(V»' such that
I(x) = E E AklxleixZk,
k:;::l O::;I::;mk-1
where Ak,! = ak,/(-i)'II!. To see that the series converges in £(JR.) one notes
that
(~) m(xleixZk) = ( . L
O::;J::;mf(l,m)
(7) l(l- 1) ... (l - j + 1)x l - j (iZk)m- j ) eiXZk .
For Ixl :::: R and IZkl ~ 1, whenever k ~ ko we obtain the following inequal-
ities:
364 6. Hannonic Analysis
since
and
This shows that the differentiated series converges absolutely and unifonnly
over all compact subsets of R This concludes the proof of Theorem 6.1.11. 0
(
ILk,/,xe
s -izrx) _
-
{O(_1)1 if (r, s) '" (k,I),
if (r,s) = (k,l),
where V = V(FIL) = {(Zk, mk), k ~ I}, 0 ::: 1::: mk - 1, and 0::: s ::: mr - 1.
6.1.14. Lemma. Given an arbitrary 0 '" IL E [ ' OR) one can find explicit distri-
butions ILk,l satisfying the above properties. Moreover, if [a,,8] is the smallest
closed interval containing the support of IL, then supp(lLk,d S;; [a, ,8].
We note that for any polynomial Pk,l(Z) = L:j aj(z - Zk)j of degree strictly less
than mk, the function
Pk,/(Z)FIL(Z)
(z - Zk)m k
is entire, belongs to Ap(iC), and has the required order of vanishing at Zr for
r '" k. We claim that for a convenient choice of the polynomial Pk,l we obtain
6.1. Convolution Equations in ~ 365
the Fourier transform of the distributions J-Lk,1 we are looking for. To verify the
claim it is enough to observe that the Taylor development of F J-L about Zk is
(z Z )mk
FJ-L(z) = - k (FJ-L)(mkl(Zk) + O«z _ Zk)mk+l).
mk!
It is immediate from this expression that there is only one polynomial Pk,l
that satisfies all conditions. In fact, it coincides with
Pk'!(~ ~) Tk,
where FTk(z) = FJ-L(z)/(z - Zk)m k. This last distribution is easy to compute.
We leave as an exercise to the reader to show that
6.1.15. Corollary. Let f E g(~) and let J-L E £'(lR), Assume that
f(x) =L L aklxleizkX,
k::;:IO::;:I::;:mk- 1
where V(FJ-L) = {(Zk, mk)} and the series converges in the topology of £(~).
Then the coefficients akl can be computed by the formula
akl = (J-Lk,l, j).
In particular, if F J-L is slowly decreasing and V (F J-L) is an interpolation variety
then b = (I! akl) E (Ap(V»'.
Proof. Due to the convergence of the series, the function f is J-L-mean-periodic
because that is true for each term. The formula for the coefficients is an imme-
diate consequence of Lemma 6.1.14. Finally, if J.L satisfies the hypotheses of
Theorem 6.1.11, then we know that the coefficients of f must satisfy the prop-
erty (l! akl) E (Ap(V»'. 0
6.1.16. Remark. It is clear that this corollary holds even if the series converges
in £ (~) only after some summation procedure has been applied. In fact, the
theorem of Schwartz [Schwl], [BTl], [BT2] shows that every J-L-mean-periodic
function has such a series representation which is convergent after grouping
terms and Abelian summation.
It is also clear that if J-L is a distribution of order N, one needs the convergence
to be in the topology of eN (~). This is the case since all the distributions J-Lk,1
366 6. Hannonic Analysis
have order at most N. Moreover, one can prove that the expansion theorems
*
hold in V'(JR), i.e., if f.L f = 0, f E V'(JR), then f has a Fourier expansion
of the same kind as before which converges in V'(JR).
6.1.17. Corollary. Let f.L E e' (JR) satisfy the hypotheses of Theorem 6.1.11,
supp(f.L) £ [a, ,8], a < ,8. Let f E e(JR) be f.L-mean-periodic which vanishes
identically on [-,8, -a]. Then f is identically zero in JR.
Proof. It is a known fact of the theory of distributions ([Schw2, Theorem
XXVIII]) that if v E e'(JR) has supp(v) £ [a, ,8], then (v, = O. From the 1>
previous corollary, the coefficients ak,l = (f.Lk,l, = O. Thus f == O. 1> 0
6.1.18. Remark. It is also clear from the proof and Remark 6.1.16 that the
restrictions on f.L of F f.L being slowly decreasing and V an interpolation variety
are not necessary. It is also clear that f is determined by its values on any interval
of length L = ,8 - a (if L > 0). When L = 0, f.L is an ordinary differential
operator with constant coefficients (up to a translation). In this case, one needs
the values of f and a certain number of derivatives at one point to vanish. We
will assume that L > 0 unless stated otherwise.
6.1.19. Proposition. Let fEe (JR) be the solution of two convolution equations
f.L * f = v * f = 0, and assume that f has two representations
f(x) = L P)..(x)eiJ..x = L Qa(x)e iax
()...m)eV(.1"/L) (a.n)eV(.1"v)
which converge in e(JR). Then each a in the second series for which Qa t= 0 is
a zero of Ff.L and moreover Qa = Pa• and a similar statement holds for ).. in the
first series.
Proof. If Qa(x) contains the nonzero term aa.lx1e iax let Va = va,o be the distri-
bution defined in Lemma 6.1.14. It is easy to see that
Hence
f.L * (Qa(x)e iax ) = (f.L * Va * f)(x) = 0,
which implies that Ff.L(a) = O. In the same way the zeros).. of Ff.L which
appear in the first representation must be zeros of Fv. The fact that Pa = Qa
is due to the uniqueness of the coefficients in Corollary 6.1.15. 0
and only if A belongs to the spectrum 01 I with multiplicity bigger than or equal
to (degree P'A) + 1.
Proof. It is enough to observe that in the previous proposition one does not
need to know that I has a representation with respect to f.L to conclude that F f.L
vanishes at A to the order (deg P'A) + 1 when f.L * I = O. 0
6.1.21. Examples.
(1) If a function I E e(lR) is periodic of period t" > 0, then I is the solution
of the convolution equation
ak = (ILk. /) =.!.
t"
r 1(_s)ei (2rrk/T)Sds = .!.1°
Jo t"
l(s)e-27riks/Tds
-T
=.!. r l(s)e-27riks/Tds.
t" Jo
This is the classical Fourier coefficient of I, f is represented by the series
L
00
I(x) = ake27riks/T,
k=-oo
and the estimates from Theorem 6.1.11 are
L
00
eBp(Zt)lakl < 00
k=-oo
for every B > O. Here P(Zk) = log(2 + 1(2rri/t")kI 2), hence we have
FIL()..) = L:f=D Ck(i)..)k = 0 and the exponential monomial solutions are x j ei)..x
if ).. is a root of F IL of multiplicity bigger than j. This is the classical result of
Euler.
(3) As a final example let us consider the difference-differential equation
!'(x) = f(x + 1).
This equation corresponds to convolution with IL = 80- L 1, its Fourier trans-
form is FIL(z) = iz - eiz and (FIL)'(z) = i(1- eiz ). The zeros of FIL are all
simple as we can thus see easily. Let V = V(FIL) = {zd, since (FIL)'(Zk) =
i + Zk, then V is an interpolation variety by Proposition 2.7.8.
In fact we can describe reasonably well the roots Zk as follows. Let iz = pe iO ,
then the equation defining V becomes
{
pcose = logp, -n < e :::: n,
p sine = e + 2kn, k E Z.
The first equation implies that cos e > 0 for p > 1, this restricts e to the interval
-n /2 < e < n /2. Furthermore, we can see that
p2 _ (logp)2 = (e + 2kn)2,
{
tane = e + 2kn.
logp
From the first equation Pk ::: 2lkln. Hence, the second equation becomes
2kn
tan e ::: ,
log21kln
which shows that for k -+ +00 one has a root ek -+ n /2 and for k -+ -00 one
has a root ek -+ -n /2. Therefore, one can improve the estimate of Pk to
and
arg Zk ::: Arctan ( 2kn ) - ::..
log21kln 2
It follows that the Zk lie in the lower half-plane. For k -+ +00, arg Zk -+ 0
and, for k -+ -00, argzk -+ -no Furthermore, one can see I Imzkl = Log Izk!
hence p(zd ::: Log IZk I ::: log 2lkln. From these considerations we obtain that
any COO-solution of this difference-differential equation has the form
L
00
f(x) = akeizkX
k=-oo
JLk = JLk,O,
ak = (JLk. j) = ~
I +Zk
{1°eiZkS f(-s)ds -
-1
-/- f(O)}
IZk
=. -Zk. [te-iZkSf(N)(S)ds_~(N)(O)].
(I + Zk)(/Zk)N Jo IZk
The last line is obtained using the equation and integration by parts. It shows
the coefficients have the correct rate of decay.
6.1.22. Proposition. Let 0 i= JL E e'(lR). Then the operator JL*: £(lR) ~ e(lR)
is surjective if and only if F JL is invertible.
Proof. The surjectivity of JL* is equivalent to the injectivity and closed range
of the transpose operator iJ-*: e'(lR) ~ £'(IR). By Fourier transformation the
last statement is equivalent to the fact that the multiplication operator by F iJ-
in Ap(1C) ~ Ap(1C) must be injective and have closed range. The injectivity is
obvious since JL i= 0 and the range is the principal ideal i generated by F iJ- in
Ap(lC). Since FJL is invertible then [ = (FJL)Ap(1C) is closed. Since [ is closed
if and only if i is closed we have proved the surjectivity.
Conversely, if JL* is surjective, then i is closed, and by the spectral synthesis
property [ = floc, which is precisely the definition of the invertibility for FJL. D
(2) There exists a constant a ::: 1 such that lor each x E JR. there is x' E JR. so
that:
(i) Ix - x'i .::: a log(2 + Ix I);
(ii) I/(x')1 ::: (a + Ix'l)-a.
(3) There exists a constant C > 0 such that lor every ZE C there is z' E C so
that:
(i) Iz - z'l .::: Cp(z);
II (z') I ::: e-Cp(z').
(ii)
(4) [IvE &'(JR.), :Fv/I E Jt'(C), then :Fv/I E Ap(C).
(5) [ = [Ioc (i.e., I is invertible).
(6) [ is closed.
(7) f.L*: &(JR.) ~ &(JR.) is surjective.
(8) For every subset B £; &' (JR.), if f.L * B is bounded in &' (JR.), then B is bounded
in &'(JR).
Proof. We will first show that conditions 0), (2), and (3) are equivalent. Then
the proof of Proposition 6.1.2 shows that (6) implies (7) and (7) implies that
[L*: &' (JR.) ~ &' (JR.) is an isomorphism onto its image. The definition of bounded
sets in a topological vector space shows that this last condition implies that
whenever B £; &'(JR.) verifies that [L * B is bounded in &'(JR), then B must also
v
be bounded in &' (JR.). Since v 1-+ is an isomorphism of &' (JR) onto itself then
condition (7) implies (8). Finally, to conclude the proof we will show that (8)
implies (2).
Assume x E JR. and I/(x)1 < (a + Ixl)-I, then x E S(I/I, s, c). Let 0 be the
connected component of SOil, s, c) containing the point x and let x + iy be a
point in ao
n (x + iJR.). Then
Iyl .::: p(x + iy) .::: A1P(x) + A2,
that is,
and
I/(x + iy)1 = e-cp(x+iy) ::: 8(2 + Ixl)-Y
for convenient 8 > 0, y > O. From the Minimum Modulus Theorem [BG,
§4.5.14] or Lemma 2.2.11 above, we have that for any R > 0 there is a value
r, R/4 .::: r .::: R/2, such that
min log I/(~)I ::: 9y log(2 + Ix I) + 9 log 8 - 8 log M,
1~-(x+iY)I=r
6.1. Convolution Equations in R 371
where
M = max
1~-(x+jY)I:::2eR
If(~)I.
Let us choose R = 51yl. then for any ~ in B(x + iy. 2eR) we have that
I1m ~ I ::: 311yl. I Re ~ I ::: Ixl + 301yl. Thus. for any such ~.
If(~)1 ::: CleC2P(~) ::: (2 + Ixl)BI
for a convenient constant BI > O. since
p(~) ::: 311yl+ log(2 + 611yl + Ixl)
::: 31A I log(2 + Ixl) + 31A2 + log(2 + 61A I log(2 + Ixl) + 61A 2)
::: Bolog(2 + Ixl).
Therefore there exists KO > 0 such that
min log If(~)1 ::: KO log(2 + Ixl)
1~-(x+jY)I=r
and since r ::: ilyl there is x' E JR so that lx' - (x + iy)1 = r. then it follows
that Ix'-xl::: r::: ~IYI. Thus
lx' - xl ::: ~(AIlog(2 + Ixl + A2 ) ::: a log(2 + Ixl)
and
If(x')1 ::: (2 + Ixl)-KQ ::: (a + Ix'l)-a
for a > 0 conveniently chosen. Hence. we have just proved that (1) implies (2).
To show (2) implies (3) given a point z = x + iy. using (2) choose x' E JR.
lx' - xl ::: a log(2 + Ixl). and If(x') ::: (a + Ix'l)-a. We can assume that a ::: 1.
thus Iz - x'i ::: a log (2 + Ixl) + Iyl ::: ap(z). Therefore z' = x' satisfies (3).
We are going to prove that (3) implies (1). Given a point z we know there
exists a point z' such that Iz - z'l ::: Cp(z) and If(z')1 ::: e-Cp(z') for some C :::
1. Let R = 51z - z'l. then for I~ - z'l = 2eR we have
I Im~1 ::: I Imz'l + I~ - z'l ::: I Imz'l+ 281z - z'l
::: I Imzl + 291z - z'l ::: 30Cp(z).
Similarly. we have
I~I ::: Iz'l + I~ - z'l ::: Iz'l + 2eR ::: Iz'l+ 281z - z'l
::: Izl + 291z - z'l ::: Izl + 29Cp(z).
Hence.
p(~) = I Im~1 + 10g(2 + I~I)
::: 30Cp(z) + log(2 + Izl + 29Cp(z» ::: AIP(z)
for some Al > O.
Applying the Minimum Modulus Theorem as done earlier we obtain a circle
r of center z' and radius r ::: Rj4 so that z E Int(r) and on r we have an
inequality of the form
If(~)1 ::: ee-cp(~)
372 6. Harmonic Analysis
for a convenient choice of E > 0, c > O. Moreover, the same computation shows
there is A2 > 0 such that for any two points w, w' of Int(f) one has p(w) ::5
A2P(W'). 0
Proof of Lemma 6.1.25. Since f is not slowly decreasing then condition (2)
cannot be satisfied for any choice of a ::: 1. Hence, for any j E N* there is
Xj E JR so that:
First, we observe that all the functions h j are of exponential type rr. Hence,
the same is true for the gj. On the real axis, we have the following properties:
(a) hj(O) = 1;
(b) Ihj(x)1 ::5 1 (x E R); 'and
(c) Ihj(x)1 ::5 (j/rrlxl)j for Ixi =f. O.
We claim that the sequence (gj) is not bounded in Ap(C). For that, it is
enough to show that there is no value n > 0 such that Igj (Xj) I ::5 IXj In for every
j ::: 1. In fact,
We want to show now that the sequence fgj is bounded in Ap(C). All
these functions are of exponential type ::5 rr + (type of f). From the Phragmen-
Lindelof theorem we conclude that it is enough to prove that f gj are uniformly
bounded on R by a function of the form (2 + Ixl)N for some N.
For Ix - Xj I ::5 j log IXj I we have from (b) that
while
so that
and
2j ~ Ix!.
Therefore, for the same range of x, Ix - xjl ~ j log IXjl, one has that
for some N. This concludes the proof that the family (f gj) j is bounded in
Ap(C) while (gj)j is not. 0
The last lemma shows that condition (8) implies I is slowly decreasing. This
concludes the proof of Theorem 6.1.23. 0
6.1.26. Remark. Using the same type of argument the reader can find in [Eh3]
a few other important equivalences of the slowly decreasing condition. For
instance, I is slowly decreasing if and only if:
(9) JL* 1J'(IR) = 1J'(IR);
(10) IL*: 1J(IR) -+ 1J(IR) is an isomorphism onto a closed subspace of 1J(IR);
*
(11) for any v E e'(IR), if IL v E 1J(IR), then v E 1J(IR).
These equivalences are also valid in IRn. The reader should consult [Eh3] or
[H02] for details.
For IL slowly decreasing such that cv(supp IL) = [a,,8] let us consider the
Cauchy Problem
{
IL*I =g,
il[a,.8] = h,
where g E e(IR), h E e([a, .8]). From Corollary 6.1.15 we know that if V(FIL)
is an interpolation variety for Ap(C) then (*) has at most one solution I E
e(IR). (Remark 6.1.16 implies that the uniqueness is true for any IL '# 0.) The
obstruction to find a solution is only to be able to solve the simpler problem
{ IL * 10 = 0,
iol[a,,8] = ho,
for ho E e([a, ,8]). Namely, to solve the Cauchy problem (*) we choose an
*
arbitrary solution II of IL Ii = g, which exists by Proposition 6.1.21. If I is
374 6. Hannonic Analysis
6.1.28. Remark. The theorem is valid without the assumption that V(FIL) is an
interpolation variety [Eh3]. The proof is similar to the one given below except
for the grouping of terms.
Proof. Let 9J1 = Ker IL* and E = {t/> E £([a, ,8]): (IL, t/>(n)} = 0, "In EN}. Since
the map 1----+ il[a,,8] is an injective continuous map from 9J1 into E, it is
° °
surjective if and only if it is a topological isomorphism. Hence, if IL is hyperbolic
and € > is fixed, there are C 1 > and N E N such that for any I E 9J1
sup I/(x)1 ::: C 1 sup 1/(j)(y)l.
xE[a-s,,B+sl YE[a,,Bl
O:'Oj:'ON
:FJL(O) = 0 and we can then reduce ourselves to have at most one exceptional
value, Zo = 0.) We claim that the series
g(Z) = - L 1/1k(Z)
k:::O
converges in Ap(1C) and defines a distribution with support in [a, ,8], In fact, for
each B > 0, q can be chosen so that the map :FS 1-+ g is a linear continuous
map from Ap,B(1C) into :F(£[a,t!](lR».
Namely, for Iz - zkl ?: 2rko if k is not an exceptional value, then
AeBp(ztl
l4>k(Z) I ::: I
elzk q
(1 + IZkl)N ::: A 1(2+ IZkI)M-q,
if Iz - wi::: I. Therefore, by the maximum principle, the above estimate for Vtk
holds inside the disk B(zk, 2rk) up to a small modification. Since the series
it follows that:
:FS(w) _
wq:FJL(W) - 2rri
-1-1 ~q:FJL(~)(~
y
:FS(~)
- w)
d
~,
6.1. Convolution Equations in lR 377
hence
(t)
1aB(Zk,rk)
FS(~)
~qFJL(~)(~ - z)
d~ = 1 aB(zk,s)
FS(~)
~qFJL(~)(~ - z)
d~;
l/!k(Z) = zqFJL(z)
2rri
1aB(zk,s)
FS(~)
~qFJL(n(~ - z)
d~.
Since the two sides in this expression are holomorphic in C\B(zk, s) they coin-
cide at the point w. Hence, the above expression (t) for FS(w) can be rewritten
as
FS(w) = -l/!k(W) + FJL(w)hdw),
with
hk(w) = -
wq
2rri
1aB(zk,rd
FS(~)
~qFJL(~)(~ - w)
d~,
S = JL * So + Sl.
To conclude the proof we first observe that, from the expression of ifJk as a
linear combination of l/[(z - Zk)j], 1 :::: j :::: rnk, it follows that each 1/Ik is a
linear combination of F«d/dx)q JLk,I), 0:::: I :::: rnk - 1, for k ::: 1, and similarly,
of F(JLO,I), 0:::: I :::: rno - 1, if Zo = 0 is a zero of FJL. Here, the JLk,[ are the
distributions introduced in Lemma 6.1.14.
Furthermore, if S = Ox for some x E JR., the computation of the residues that
define ifJk shows that the coefficients of l/[(z - Zk)j] are linear combinations
of xl e- iXZk with 0 :::: I :::: rnk - 1. We note that the coefficients of these linear
combinations depend on q, which can be taken to be the same for all x, Ix I :::: R.
This concludes the proof of Lemma 6.1.29. 0
Now let us return to the proof of Theorem 6.1.27. We need to show that the
inequalities on the zeros of FJL imply the hyperbolicity. To see how to proceed,
378 6. Harmonic Analysis
8x = /-t * So + Sl.
First, we need to show that this definition of j is independent of the choice
of decomposition of 8x . In fact, let
8x = /-t * So + Sl = /-t * To + Tl
with cv(supp Sl) ~ [a, ,8] and cv(supp Tl ) ~ [a, ,8]. Then
Tl - Sl = /-t * (So - To)
and, from the support theorem,
[a, ,8] + cv(supp(So - To» = cv(supp(Tl - St}) ~ [a, ,8].
Therefore, cv(supp(So - To» = {OJ, which indicates that So - To = P(d/dx)(80)
for some polynomial P. Hence
The limit
. 8x+I - 8x
11m ~I
= 0
1-+00 t x
6.1. Convolution Equations in lR 379
1-+00 t
Repeating this procedure we see that
j<n)(x) = ('R.(8!n». h).
To conclude the proof of this theorem we need to verify that the function f
*
is a solution of /L f = O. This follows from the last part of Lemma 6.1.29
I
which tells us that (x) is given by an infinite series of the form
L L Cklxle-ixzt
k O:::I:::mt-1
with coefficients Ckl independent of x as long as Ix: I < R. and the series
converges in £(] - R. RD. This representation shows that (/L f)(x) = 0 if *
Ixl < R - Max(lal.I,BI). Since R is arbitrary. f E rot D
=
For instance, we can take 8n .fii and nk = k 6 for k ::: ko where ko is a
convenient integer. (The construction holds for other sequences En. Note that
En ::: 1 and Lk>l (EnJnk)2 < 00.)
We define -
This function has zeros at the integers n E Z*\A and at the points ±(n + iEn)
for n E A. Therefore. it is a function of exponential type by Lindelof's theorem
(see [Lev. Chap 1. Theorem 15]). To show that f E Ap(C) it is enough to show
that it grows at most like a power of x for x E JR.. This fact and the fact that f
is slowly decreasing will be a consequence of the following lemma:
6.1.31. Lemma. Let n E A and 8n• En ::: 1 as above. For x E JR. we have:
(1) 1- (- -X)2
- > n2 I1- (X)21
-
n + iEn - n 2 + E~ n'
380 6. Hannonic Analysis
1
1---x-l= In-x+iEnl > In-xl> n 1 1 -::1
n+iEn In+iEnl - In+iEnl -In+iEnl n
the same inequality holds when we replace x by -x. Multiplying the two
inequalities we obtain (1).
(2) Remark that because En ~ 1,
1- ( - -X)2
.-
n + I En
:::: n
2
-2--2
n + En
( En)2 1- (X+i)2
1+2-
8n
--
n
. D
6.1. Convolution Equations in lR 381
The function I being even we can limit ourselves to x ::: O. Consider first
the case where, for some n E A, we have Ix - nl .::: 8n . By property (3) of
the sequence A we have Ix - kl > 8k for any k E A \{n}. Therefore, using the
definition of I and the inequalities (2) and (3) of the previous lemma we have
I/(x)1 = II
mtf-A
m#O
1 (;rl· II
1-
keA\{n)
1- (-
k+
X)Z
-
iBk
·1--- (X)Z
n + iBn
.: : II 1- +')Z
(~ . II (kZ) 2)Z
-z--z ( 1+~ 1- (+
~')Z
mtf-A m k +
keA\{n) 8Bk k k
m#O
X 2
nZ
zlxl
Z
1-
(.)2
x
--
+1
.
n +Bn n
Proof. From the definition of I and the first inequality in Lemma 6.1.31, we
obtain
I/(x)1 ::: II ( n n2)
neA +
II 1 -
-Z--2
Bn n#O
1 (X)21
-
n
= sin1l'x
AI 1- -1
1l'X
382 6. Hannonic Analysis
with Ai> 0, since the infinite product is convergent. Given any point x E IR
there is y E IR such that Ix - y I :::: 1 and I sin n y I = 1. This is clearly sufficient
to show that I is slowly decreasing. 0
It is clear from Corollaries 6.1.32 and 6.1.33 that I is the Fourier transform of
a distribution f..t which satisfies cv(suPPf..t) ~ [-n,n]. A simple application of
the Squires theorem (Remark 2.6.24) shows that V (f) is an interpolation variety.
(We give a direct proof below.) Clearly the zeros of I of the form = n + ien,
with n E A, do not satisfy
Consider first the case where Zk is a real zero of I, by the first part of Lemma
6.1.26 we see that if we replace every zero n + ien, n E A, by n we obtain
1- ( - -
n+ien)2 .
k + iek
For k E A \{n} we compare 11 - [(n + ien)/(k + iek)]1 with 11 - n/ kl. We have
n +-ien
\1 - - \ = I 1--
n 1\ -k- \\ 1+1. (Ck
-- -cn) \
k + iCk k k + iek k- n
Similarly,
n +-iCn
\1 + - \ > I 1+-n 1\ 1 \
k+iek - k l+i(ck/k)
6.1. Convolution Equations in R 383
On the other hand the convergence of the series 2:kEA (sd k)2 implies there
is a constant A such that
1~ IT 11+(~rl~A<OO
kEA\{n}
independently of n. Therefore
IT 1- (nk ++i
kEA\{n}
Sn ) 2 >
iSk -
~
A
IT 11 - (~r
kEA\{n} k
I.
For k E N*\A we have, by the same method,
1- ( -n +k.)2
-
ISn
~ I1-(k) 21 .
n
It follows that
k#n
C1
>---
In+isnl
with C 1 > O. Therefore, for every zero Zk of f we have If'(Zk)1 ~ D/lzkl for a
convenient choice of D > O. This proves that VU) is an interpolating variety.
o
6.1.34. Remark. In the paper [Eh3, p. 333] one finds an example of f.-L E £' (lR)
such that all zeros of F f.-L are real but f.-L fails to be hyperbolic because F f.-L is
not slowly decreasing. In this example
= IT ( cos dz )d
n
Ff.-L(z)
n~l n
EXERCISES 6.1.
Unless explicitly mentioned every function belongs to t:(lR), convergence is understood
in the sense of t:(lR), and every distribution belongs to t:'(lR).
1. Show that if P is a polynomial, P(x)e iAX is JL-mean-periodic if and only if xke iAx
is JL-mean-periodic for 0 ::s k ::s deg P.
2. Show that the spectrum 'r(f) of a function f is either empty, discrete, or the whole
complex plane.
3. Let M be (closed) invariant subspace of t:(lR). Show that (d/dx)(M) ~ M,
JL *M ~ M. If P is polynomial and P(x)e iAx E M, conclude that xke iAX E M for
O::s k::s degP.
4. Find all finite dimensional invariant subspaces of t:(lR).
5. An invariant subspace M is minimal if M #- {OJ and there is no proper invariant
subspace N ~ M. Find all the minimal invariant subspaces of t:(lR). Show every
invariant subspace contains a minimal subspace.
6.2. Convolution Equations in C 385
15. Let /L E £'(~) be slowly decreasing, the zeros of .r/L all real and simple, V(.r/L)
an interpolation variety. Using Theorem 6.1.11 show that every f E £(~) which is
/L-mean-periodic can be written as f(x) = Lk ckeiOlkx, Lk ICkl < 00. (In fact, the esti-
mations on Ck are even better than this one.) In particular, f is almost periodic.
16. Show that if f E C(~) is almost periodic then f E LOO(~).
*17. Using Theorem 6.1.11 show that if /L = 81H + 81-1- + 81--1 + L1_1- and A is a
Liouville number, then there exist f E £(~) which are t-t-mean-periodic but not almost
periodic.
18. Let f E £(~) (f E C(~) is enough) be mean-periodic with respect to a measure /L
of support in [a, b]. Define f+ = fx[o,oo[, f- = f - f+. Show that if g := f+ * /L then:
(a) g = - f- * /L and suppg S; [a, b];
(b) the functions Lb(g) = G, Lb(/L) = M (Lb is the bilateral Laplace transform) are
entire functions.
Show that if v is another measure of compact support such that v * f = 0, h := f + * v,
H = Lb(h), N = Lb(V), then the meromorphicfunctions G/ M and H / N coincide. (This
meromorphic function is called the Fourier-Carleman transform of f.)
We note that this definition is slightly different from the one used in the
previous section for distributions. Regretfully, both terminologies are standard.
We will therefore adhere to the definition (*) in this section. We also recall from
Chapter 2 that we can convolve analytic functionals T, S as follows:
(T * S, f) = (S, T * f) (f E Jf'(C))
and that the Fourier-Borel transformation
~(T)(z) = (T~, e~Z) (z E C)
is a topological isomorphism between the algebra Jf" (C) of analytic functionals
and the algebra Exp(C).
The considerations from the previous sections about the relations between
translations, invariants, (closed) subspaces of Jf'(C), and closed ideals of Exp(C)
hold verbatim. We summarize them in the following statements:
A subspace rot of the space Jf'(C) is said to be invariant if rot is closed and
fh(rot) ~ rot for all hE C, where fh(f)(Z) = f(z + h).
A subspace rot is invariant under differentiation if rot is closed and f' E rot
for every f E rot.
It is clear that if rot is an invariant subspace of Jf'(C), then it is invariant
under differentiation but the converse is also true because
f(z + h) = L
f(n)(z)h n
n!
n2!O
is a series convergent in the topology of Jf'(C). We leave it to the reader to
verify that this equivalence does not hold in the space t:(IR).
If T E Jf"(C) then Ker T* is an invariant subspace. Moreover, if ~(T)(a) = 0
then eaz E Ker T*, hence Ker T* =j:. 0 in and only if T =j:. C8a (c =j:. 0, a E C) and
Ker T* =j:. Jf'(C) if and only if T =j:. O. The following is an easy result:
We also note that zk eaz E rot if and only if there exists meN, m > k such that
(a, m) E V(l), [ = ~(rotol).
It follows that zj eaz e rot for all 0 ~ j ~ k.
6.2.2. Definition. An invariant subspace rot =j:. {OJ admits spectral analysis if
there exists a E C such that eaz e rot.
rot admits spectral synthesis if and only if [ = [loco
6.2. Convolution Equations in C 387
6.2.3. Theorem. Every nonzero invariant subspace rot of Jft?(C) admits spectral
analysis and spectral synthesis.
Proof. This follows from Chapter 2 where we have showed that every nontrivial
closed ideal of Exp(C) is localizable. 0
6.2.6. Theorem. Let V = V(J(T)) = {(at, mkh:::d, la,1 ~ la21 ~ .... Either
this sequence is finite or there exists a sequence of indices k, = 1 < k2 < ... such
that any T -mean-periodic function f in Jft?(C) admits the following expansion
convergent in Jft?(C) (that is the series in n converges absolutely and uniformly
on every compact of C):
where Pk is a polynomial of degree < mk. (If the sequences (ak) is finite then
(**) is a finite sum.) Moreover, the function f is identically zero if and only if
all the polynomials Pk are zero.
Proof. Let us denote J(T) by <I> and recall that there is a sequence 0 < R, <
+,
R2 < "', with Rn + 1 ~ Rn ~ 2Rn, n 2: 1, such that
(t)
for some e > 0 and A > 0 (see proof of Proposition 2.7.14). It follows that the
principal ideal (<I» of Exp(C) is closed.
The same proof of Lemma 6.1.12 shows that if fJ: Jft?'(C) -+ Exp(C)/(<I» is
the composition of the Fourier-Borel transform and the canonical quotient map,
then tfJ is a topological isomorphism of (Exp(C)/(<I>))' onto Ker T*. We need
therefore to identify the quotient space Exp(C)/(<I» to a subspace of A(V).
Since, in general V is not an interpolation variety for the space Ap(C) with
p(~) = I~ I, the identification of this quotient is more complicated and requires
the double summation in the statement (**) (i.e., grouping of terms) which did
not appear in Theorem 6.1.11.
388 6. Hannonic Analysis
Let us denote by Vn = {(at, mk): k n :::: k < kn+d the family of (at, mk) E V
such that ak belongs to the annulus Cn,
(n ~ 2),
while C1 is the disk C1 = {~ E c: I~I < Rd. In this case we have a canon-
ical interpolation formula. Namely, if g is a holomorphic function on Cn and
continuous up to the boundary, we consider
In fact, if h is another function with Pn(h) = a then In (g) = In (h), this implies
the second inequality. In future, we shall write In(a) for In(g). Remark that if
Vn = 0, then A(Vn) is the trivial vector space {OJ.
We define the space A.(V) of sequences with grouping in the following way:
given a E A(V) we can write it in a unique way as a sequence (an)n~l with
an E A(Vn), then
Let a = (an)n:::b lIaliD < +00, for some D > O. Choose a sequence (hn)n:::1
with h n E JIt'(Cn ) satisfying
Pn(hn) = an,
IIhn 1100 ::: 211a IIDe DR•.
Since the components of the set S(I4>I, 8, A) are contained in Un>1 Cn we
can apply the Semi-Local Interpolation Theorem 2.6.4 and obtain h EExp(C)
such that
p(h) = a,
Ih({)1 ::: MliaIlDe(D+N)I~I,
when we identify A(Vn) with Cd. by enumeration of the values g(l)(ak)/ I!,
an= Pn (g). Hence, for every D > 0 we have
IIbn II~ := L IIbnlI~eDRn < +00
n:::1
and
{a, b} = L{an , bn }.
n:::1
From these considerations it follows that to every f E JIt'(C} such that
T * f = 0 corresponds to a unique element b(f) E A: (V) in such a way that
for S E JIt" (C)
(~) (S, f) = (p(~(S», b}.
(The verification of this last identity is done the same way as in Theorem 6.1.11.)
Moreover, the continuity of f as a linear functional on JIt" (C) tells us that
the convergence of the series that are implicit on the right-hand side of {~} are
uniform for any bounded family of analytic functionals S. In particular, we can
take S = 8z , z belonging to a compact subset of C. Then
390 6. Harmonic Analysis
and
where
o
It is clear that the choice of grouping is not unique, though one can construct
examples showing that they are necessary for the convergence of the series
[Leo]. One natural way to group terms is to use the connected components
of S(I<I>I, e, A), this observation and basically the same proof as that of
Theorem 6.2.6 can be used to prove the expansion of Coo /L-mean-periodic
functions in JR when /L E e' (JR) is just slowly decreasing.
In order to find an explicit formula for the coefficients bk,I/ I! we need to
construct analytic functionals Tk,j such that:
(~(Tk I»(j) 1
p(~(Tk,/»i,j = :,
J.
(a;) = -1'. ~/,j . ~k,i'
Let () be a holomorphic function near w = 0 which vanishes at w = 0 with
order exactly m E N*. Then
1 A-m
--=-+--+ A-m+1
... A_I ....
+-+
B(w) w m w -
m I W
[B~~)L '
then, for 0 ::: 1 < m,
Wi ]
[- -
B(w) 0
=W - + ... +--
I (A-m
wm
A_I_I)
wl+ l
=w I (1
B(w) -
{A_I A_I+I A_I })
Wi + wl-I + ... + --;;; + ...
wi
= B(w) + HI(w),
where HI is holomorphic near w = O. Hence,
B(w) [~]
B(w) 0
= wi + B(w)HI(w),
6.2. Convolution Equations in C 391
for w near 0, and the order of vanishing of the term () HI at the origin is bigger
or equal to m. Therefore, when 0 :::: j < m,
T.
J( k,I)(~)'-
'- <I>
(~)
- ad]
[(~l!<I>(~) ak'
l
Tk,/. First we define a functional Tk by the formula
z
(Tko f) = ( Tz , eakZ I f(w)(z - w)mk-le-akw dw ) .
(mk - 1). Zo
One can see that this formula is independent of the c!toice of base point
zoo Changing Zo amounts to adding (Tz, eakZ P(z») where P is a polynomial of
degree less than or equal to mk - 1, but this quantity is zero because ak is a zero
of multiplicity mk of <1>. Using the same observation one obtains by integration
by parts
'1:'(T.)( ) - <I>(~)
v k ~ - (r., -ak )mk
p (r).-
k,l.,
(r -ak)mk [(~1!<I>(~)
.-.,
-ad] ak'
then
Moreover,
(Tk,l, f) = (p(J(Tk,I», b(f») = It·
bk I
(~)
392 6. Harmonic Analysis
where the series (in n) is uniformly and absolutely convergent on every compact
subset of C, the CXk are among the zeros of some function <1> of exponential type,
and the Pk are nonzero polynomials of degree dk < mk = multiplicity of CXk as
zero of <1>.
Proof. There exists aTE ye' (C) nonzero, such that T * f = O. From the
previous theorem we know that f has such a Fourier expansion with frequencies
CXk which are zeros of <1> = J(T). Since we are only considering those CXk
for which Pk =1= 0, the frequencies which appear in the representation do not
necessarily exhaust all the zeros of <1>. We want to show that both the frequencies
CXk and the polynomials Pk are independent of the choice of T.
Let Tk,O be the analytic functional defined above, then
and
= L
0:9~1
C) (J(Tk.O»(S)(f3)zl-se fJz ,
If f3 = CXj =1= CXk. since 1< mj, we have (J(Tk,o»(s)(CXj) = 0, and if f3 = CXk
then (J(Tk,o»(s)(CXk) = 0 for 0 < s ~ I ~ mk. Hence,
if j =1= k,
In other words,
0= L L Sk.O * (P/z)e
n j
ajz )
6.2.8. Definition. Let f E Jt"(Q) and Zo, Zl, ... , Zm be distinct points in n. The
mth divided difference of the function f at the points Zo, ... , Zm is
d m f(zo, ... , zm):= L f(zj) II(Zj - Zk)-l
O~j~m bf.j
The divided differences I::.. m! can be still defined by continuity when some
of the points coincide. For instance, for Zl, ... , Zk such that Zj =j:. Zj for i =j:. j,
=
each repeated mj + 1 times, m mj + ... + mk + k - 1
I::.. m!(Zlo ... , Zl, Z2, •.• , Z2,· •• , Zko· •. , Zk)
1 am1 +·+mk
" ,a a
= ml·m2··· ·mk· zlm1a z2m2 ... zkmk I::.. k- 1!(Zl, ... , Zk).
The interpolation formula has to be interpreted to mean that Pm interpolates
the values of ! and its derivatives, and the remainder formula remains true.
The reader will find other expressions for divided differences and applications
to interpolation in [Os], [MiT].
For our purposes we need to relate the size of the function!, the Newton poly-
nomial Pm, and the corresponding divided differences, that is, the coefficients
of the Newton polynomial.
Since I::.. m!(zo, Zl, ... , zm) = I::..m-1g(Zl, .•. , zm) the induction step is correct.
o
Let us denote by Vn the number of zeros counted with multiplicity of ~ =
~(T) in the annulus en. Denote by zo' ... , Z:._l these points, each of them
repeated according to their multiplicity. It is clear that the divided differences
I::.. j (g) for Pn (g) = an E A (Vn) are independent of the choice of g. For this
reason, we denote
6.2. Convolution Equations in C 395
Since Vn ~ KoRn for a constant KO > 0, which depends only on <1>, we have
II an lin ~ IIPnil oo ~ KoRnlllanllln.
M M
~ jllaIlDe 5(D+N)Rn ~ jllan IIn e(4D+N)Rn.
On On
Choose D = ~ to obtain
Illanlll n ~ Me(N+l)Rnllanlln'
We conclude that we could have defined the topology of A*(V) using the norms
IllalllD = sup IIlanlllne-DRn
Let us assume first that the points z8, ... , Z~n-I are distinct and consider
the map
J: an = (ao, ... ,avn-I) f-+!:!. = (!:!.(O) (a), ... , !:!.(vn-I)(a)).
6.2.10. Remark. The most important point is the observation that except for the
factor 8;;- j the computation of the norm II Ibn III~ is entirely computed from the
knowledge of the zeros zJ and the values bk,l. We also note that, in the definition
of the norm III 'III Dwe could replace the weight exp(-DRn) byexp(-DlzJD
for an arbitrary j and, up to a change of D, the norms are equivalent. This
depends on the fact that the weight p(z) = Izl does not change too much in the
region en.
Let
T = L (_1)1 ak,lo~~) ,
l::sk::sp
l::sl::sqk
where ak,!, r:k are complex numbers, r:j =ft r:k if j =ft k. Then
(T * f)(z) = L akd(l)(z + r:k)
l::sk::sp
l::sl::sqk
and
with
Qk(~) = L ak,l~/.
l::sl::sqk
°
As we pointed out after the proof of Theorem 6.2.6, the groupings in the
Fourier-expansion of solutions of the equation T * f = are really determined
by the connected components Qn, n ~ 1, of the sets S(I<l>I, e, A). In Proposi-
tion 3.1.31 we have shown that one can choose e > and A > such that the
diameter of every component is bounded by 1. It follows that the number of
° °
zeros of <l> in any such component is bounded by some constant v > 0, inde-
pendent of the component. It is then easy to see that the norms III . Illn can be
defined without introducing the diameter On in the definition. The topology of
the corresponding spaces are absolutely equivalent. In other words, the function
f has the expansion
where In is the set of indices of distinct zeros otk of <l> that are in Q n and
'" bkl I
Pk(Z) = LJ i!Z'
O::sl::smk
IIlbnlll~ = L ICjl,
O::Sj::svn-l
where Cn = t(J-l)(bn ).
Furthermore, we recall that in this case we have a large amount of information
about the asymptotic distribution of the frequencies otk (see Chapter 3).
398 6. Hannonic Analysis
In this case the zeros of <I> can be at most double. Double zeros different from
~ = 0 can only occur if alb E Q.
If r = alb E Q then the variety of zeros of <I> is an interpolation variety since
different zeros lie in a finite number of sequences
fJ1+ -2rri
fU.
'71
r
In fact, there are always double zeros when r E Q, for instance, if a = b all
zeros are double. In any case, if f is an entire function satisfying the equation
<*) f(z + a + b) - f(z + a) - f(z + b) + f(z) = 0,
then
f(z) = LPn(z)e anZ ,
n~l
Since
lea~ - 11 2: C(a(~, (2rrila)Z»,
where a(~, (2rrila)Z) = inf(l,~, (2rrila)Z) (see Theorem 3.1.22) and the
corresponding inequality for eb~ - 1 also holds, it follows that in the case there
are constants 8 > 0 and A > 0 such that
(H) 1
2: ik - 2:ill2: 8e-AOkl+llD (for all k, I E Z),
then, we can find 81 > 0 and AI > 0 such that the disks B(a, 8Ie- Atlal ), a E V,
are disjoint and I<I>(~)I 2: const. exp( -A 2 IaD on IJB(a, 8Ie-Allal). Therefore, V
is an interpolation variety and the same considerations of the case alb E Q hold
so that any entire solution f of (+) can be written as
6.2. Convolution Equations in C 399
with
. log Ibnl
lIm - -
n-+oo Ian I
= -00 (0 < laii ::: la21 ::: ... ).
For instance, the condition (:j::j:) is satisfied if alb is not real. Among the
real numbers alb that satisfy (:j::j:) we have all the irrational algebraic numbers.
Since, in this case,
la ml
b - -;; ~ n1+li
Cli
where the two infinite series are convergent in Jff(C), the first one corresponds
to the frequencies ak that can be isolated and the second one to the pair al,o
and al,l that have to be grouped together because (:j::j:) fails.
The coefficients bk satisfy the condition
lim log Ibk I -00, =
k-+oo lakl
and the coefficient pairs (bl,o, bl,l) satisfy the two conditions
lim log Ibl,o + bl,II = -00
1-+00 lal,o!
and
lim log lal,o - al,llIbl,II -00. =
1-+00 !al,ol
Note that the failure of (H) is exactly the condition
lim log lal,o - a/,l! = -00.
1-+00 lal.ol
Moreover, the conditions on bl. 1 and on bl.o are actually symmetric (just use
that Ibl.ol ::: 1- b/,ll + Ibl.o + bl.d).
It is very easy to see that the groupings are necessary in this case where
(§) holds. The two conditions on the coefficients bl.o and bl,l are satisfied by
=
taking bl.o 1 and bl,l = -1. If the series was convergent without grouping,
the general term of this series should be bounded at every point. Let us suppose
for simplicity that a > 0, b > 0, and set Z = -i, then
Ib1,oe-a/,oi I = e21tmtla,
which is unbounded for a certain sequence ml -+ +00. This example which
shows that groupings may be necessary is due to Leont'ev.
400 6. Hannonic Analysis
6.2.11. Remark. It follows from the proof of Theorem 6.2.6 that a sufficient
condition on the analytic functional T for the convergence of the Fourier series
representations fez) = Lk>l Pk(z)e ClkZ without groupings is that for some 8 >
0, A > 0, each component-of S(I4>I, 8, A) contains only one zero of 4>. This
happens precisely when V is an interpolation variety so that also the series
'""'
~ c k.J.zj eClkZ
k~l .
O::;j<mk
is absolutely and uniformly convergent over compact sets. In the previous formu-
lation we are really grouping all the terms (LO::;j<mk Ck.jzj)e Clk = Pk(z)e Cltz
together.
but it is not so clear how to find an explicit solution for an arbitrary entire
function g. Nevertheless, there is a method, due to Gelfond and Leont'ev,
which we will explain in detail in the following section for the very simple
equation fez + 1) - fez) = g(z). It consists of the following: Given g(z) =
Ln~o(an/n!)zn, there is an increasing sequence Rn > 0 such that the function
(*) {S *ff =
T
*
=h
g
where g, h E Jr(C) are given and we are looking for an entire solution f. There
is clearly a compatibility conditions, namely,
T*g=S*h.
Is this condition sufficient for solvability of the system (*)? To simplify we
consider only the case where 'J(T) and 'J(S) have no common zeros. We leave
the general case to the exercises.
6.2. Convolution Equations in C 401
6.2.12. Lemma. Assume that J(T) and J(S) have no common zeros. The neces-
sary and sufficient condition for the previous system to have a solution for every
pair satisfying the compatibility condition is the existence of SI, TI E .1f' (C) such
that
*
SI S + TI T = 8. *
Proof. Consider 'R-:= {(g, h) E (.1f(C»2: S * h - T * g = OJ. Then 'R- is a
closed subspace of (.1f(C»2. Let
a : .1f(C) ~ 'R-,
aU) := (S * f, T * f).
This map is injective as a consequence of the hypothesis that J(T) and J(S)
have no common zeros (see Proposition 6.2.7). Therefore a is surjective if and
only if a is an isomorphism, and this is equivalent to the fact that to' is also an
isomorphism. By the Hahn-Banach theorem every element of the dual space
'R-' can be represented as a pair (SI, T1), SI, TI E .1f'(C). Therefore there exist
Sh TI such that
* * =
The equation S1 S + TI T 8 is sometimes called the Bezout equation.
It is equivalent to finding functions ¢I, Vrl E Exp(C) such that
¢IJ(S) + VrIJ(T) = 1.
We have already considered this last problem in Lemma 2.6.15 and we can state
the following corollary:
6.2.13. Corollary. Under the assumptions of Lemma 6.2.12, the necessary and
sufficient condition for the existence of a solution f of the previous overdeter-
mined system for every pair satisfying the compatibility condition is the existence
of constants 8 > 0, C > 0, such that
IJ(S)(~)I + IJ(T)(~)I ~ 8e-Cl~1 (~ E q
(strongly coprime condition)
402 6. Harmonic Analysis
We would like to consider next the case in which ~(T) and ~(S) do not
have common zeros but they do not satisfy the strongly coprime condition. The
question is: What extra conditions should g and h satisfy in order that the system
is solvable?
To simplify the analysis let us assume all the zeros of ~(S) and ~(T) are
simple. Since the convolution operator S* is surjective, we can find a function
G E J'r(C) such that
*
S G = g.
Consider F = f - G, then letting H := h - T * G we have
S *F = 0,
T*F = H.
The compatibility condition becomes
S*H =S *h - S*T *G = S * h - T * g = O.
Hence F and H are S-mean-periodic. Let V = {ak} be the variety of zeros of
~(S), then
F(z) = LakeClkZ,
k;O:!
H(z) = LbkeClkZ ,
k;o:!
because we have assumed that ';j(T) and ';j(S) are not strongly coprime, and in
particular we have
. -log 1';j(T)(ak) I
11m sup = +00.
k..... oo lakl
If V is not an interpolation variety and we need groupings we can make a
similar analysis using the nonos III· Illn.
EXERCISES 6.2.
1. Verify the details of the example of Leont'ev given in the text about groupings for
any solution I E JtI'(C) of the equation IL I * = =
0, IL 8a+b - 8a - 8b + 80• Consider
also the case alb is not real.
* *
2. Study the system S I = g, T 1= h, as given in the text, in the cases (a) there
are no common zeros but some zeros of lV(S), lV(T) are multiple; (b) there are common
zeros to lV(S) and lV(T); and (c) there are no common zeros but the variety of zeros of
lV(S) is not an interpolation variety.
3. This exercise is a self-contained treatment of results from the recent manuscript
[BezGr].
(a) Show that JtI'(C) is a reflexive Fr~het space and that the transposed map
\Y': (Exp(C»' ---+- JtI'(C) is a topological isomorphism such that \Y'IJtI"(C) lV. Deduce =
that one can introduce a convolution product in (Exp(C»' by means of the formulas:
then
are entire functions of exponential type without common zeros and show there are 8 > 0,
C > 0, satisfying
for all z E C.
Prove that under these conditions any entire function f{J satisfying the system of equations
M N
A= o
F
o N
o F
such that Ap = O.
6.3. The Equation fez + 1) - fez) = g(z) 405
Let us recall from [BG] that we denote by C* = C\{O}, and by Log{, for
{ E C\] - 00, 0], the principal branch of the logarithm, i.e., the principal branch
of the argument of ~ is taken to be -rr < Arg ~ < rr. By arg ~, log ~, we denote
arbitrary branches of the argument and logarithm of ~ E C*. Let exp: C ~ C*
be the map z t-+ e1:Jriz, this map is the universal covering map of C*. We can
relate periodic meromorphic functions in C to meromorphic functions in C* by
means of the following lemma:
Proof. Since the group of automorphisms of the covering map exp is generated
by the transformation z t-+ z + I, and f being periodic is exactly the same as
saying that f is invariant under this group, then F is uniquely defined by passage
to the quotient. As exp is a local biholomorphic map, the other assertions follow.
o
Note that F is explicitly given by
F(~) = f (~IOg{)
2m
for any { ¢. P(F), the choice of branch of the logarithm is irrelevant.
6.3.4. Proposition. Any entire periodic function f admits a Fourier series expan-
sion
f(z) = Lcne2:n:inz,
neZ
Cn = l
Zo
zo +1
f(z)e- 2:n:inz dz
6.3. The Equation I(z + 1) - I(z) = g(z) 407
each of the series with the same kind of convergence properties as above.
Proof. It is clear it is enough to prove the assertions for periodic entire functions.
Let F (~) = f «(1 /2rr i) log ~), then F is holomorphic in C*. It admits a Laurent
expansion about ~ = 0,
c -
n -
-1-1
2 .
7rl Isl=p"
F(~) d
rn+1~' ° < p < 00.
We remind the reader that it is also possible for a merom orphic function to
have two independent periods. This fact leads to the beautiful theory of elliptic
functions. One of the particular systems of inhomogeneous equations (associated
to the two periods) leads to the study of theta functions. These subjects are
beyond the scope of this book, the reader can find an excellent introduction in
[Mark], [Ra].
Let us consider now the inhomogeneous difference equation
f(z + 1) - f(z) = g(z),
where g is an entire function. It follows from the general theory of the last
section that there are always entire solutions to this equation. Clearly any two
solutions differ by an entire periodic function and we have just studied them. We
are going to give here an explicit construction of a solution to (*). The original
idea of this construction is due to Guichard, Appell, and Hurwitz a century ago,
408 6. Harmonic Analysis
we refer the reader to [HuI], [Ap] and the beautiful monographs of Whittaker
[Whi] and Gelfond [Ge] for other related results.
We start by solving (*) in a few particular cases. We denote by 6 g, a sum
of g, any particular solution of the equation (*). Then g is called the difference
of f. It is similar to the concept of primitive and derivative. For instance, it is
easy to verify that one can take
61 =z, 6z = z(z -
1).
2
We leave other examples to the exercises. The one example important to us is
the following, let a > 1 be a fixed number and consider the entire function a Z
(fixing a determination of log a). Then it is easy to see that
aZ
6a z = - - .
a-I
Clearly the same result holds for any a E C\{O, I}. In particular, if we let a = et,
t ¢ 2rriZ, 6e tz = e tz I(e t - 1), where the sum is with respect to the variable z.
Consider the auxiliary function
tetz
B(z, t) =- - = t6e tz = 6(tetz ),
e -1
t
which satisfies
B(z + 1, t) - B(z, t) = tetz
and it is holomorphic, for every z fixed, for It I < 2rr. (This was the reason of
the multiplication by t.) If we expand B(z, t) in a Taylor series in t about t = 0,
we have
tn
B(z, t) = '"' Bn(z)-, It I < 2rr.
L..J n!
n:;:O
It is very easy to convince ourselves that Bn (z) is a polynomial of degree
exactly n (usually called the Bernoulli polynomial), but more interesting, by
comparison of the two sides of (**) we see that
Bn(z + 1) - Bn(z) = nzn-l,
so that
6zn = Bn+! (z)
n+I
for n ?: O. We also note that if 0 < p < 2rr, then Cauchy's formula for the
1
Taylor series coefficients gives the representation
n!
Bn(z) = - .
etz dt
-t----;; (0 < p < 2rr).
2rrl Itl=p e - I t
Let us assume g(z) = L:n:;:O gnzn, then, at least formally,
6g(z) = '"' gn
L..J --Bn+!(z).
n:;: O
n+I
6.3. The Equation I(z + 1) - I(z) = g(z) 409
If the series were to converge unifonnly over compact sets, then it would repre-
sent a solution of (*). Whether it does or not, depends on the behavior of the
coefficients gn of g. If we only make the assumption that g is entire, all we
have is the Hadamard estimate
lim
n~oo
yIjgJ = O.
On the other hand, we could modity each Bn by adding a periodic function,
and, as in the Mittag-Leffler expansions of meromorphic functions, force the
series to converge to a sum of g. One way to do this is to observe what happens
to the integral defining Bn if we make p larger. For instance, if 2n < p < 4n,
then we have to add the residues at t = ±2n i, so that
- n! 1 tz dt
-e - -
2ni Itl=p et - 1 t n
= Bn(z) +n! ( e-
2niz
(-2ni)n
2niz
e -) .
+-
(2ni)n
In other words, we obtain Bn (z) plus a periodic trigonometric polynomial. In
1
general, let
n!
Bn, k(Z)-- - e tz- -dt if 2n(k - 1) < p < 2nk,
- 2 . nl
t
Itl=p e -
1 tn'
6.3.5. Proposition. For any entire function g(z) = Ln~o gnzn, the series
f(z) = "" + 1 Bn+1,n+l (z)
~ n gn
n~O
converges locally uniformly and it defines an entire function solving the difference
equation
f(z + 1) - f(z) = g(z).
Proof. We need to find estimates for Bn+1,n+l so that the fonnal series of
f converges. Let us choose p = (2n + 1)n in the integral representation of
Bn+1,n+l' Note that there is a constant c > 0 independent of n such that for
It I = (2n + 1)n
as one can reduce this inequality to the Lojasiewicz inequality for sin z. Hence,
using Stirling's fonnula, we can prove the existence of A > 0 such that for
410 6. Hannonic Analysis
all z, n
This bound, together with the Hadamard estimate for gn, shows that the series
defining f converges absolutely and unifonnly on any disk of C. 0
6.3.6. Remarks.
(1) If g is a function of infraexponential type, i.e.,
lim 10gM(g, r) = 0,
r-->oo r
where M(g, r) = max{lg(z)I: Izl ::::: r}, then one can prove that the series
'"' gn
L..J n + 1 Bn+l (z)
n2:0
Returning to the equation (*) with g merom orphic , let us assume that there
is real number p so that all the poles of g lie in the half-plane Re z < p. The
idea is to try to find polynomials Yo, YI, ... so that the series
qJl(Z) := {yo(z) - g(z)} + {Yl(Z) - g(z + I)} + ...
converges to a meromorphic function in the whole plane. (Note that this function
is obtained by applying the Mittag-Leffler procedure to the right formal sum
of g.) In order to choose the aj, let kEN be the smallest value such that
p - k < O. Choose Yo = ... = Yk-I = O. The poles of g(z + n) lie in the half-
plane Rez < p - n, hence for any n ~ k we have that g(z + n) is holomorphic
in a neighborhood of the disk Izl ~ n - p and n - p > O. Choose Yn as the
partial sum of the Taylor series of g(z + n) about z = 0 such that
Ig(z + n) - Yn(z)1 ~ Tn for Izl ~ n - p.
For any R > 0 we have that the series
the same is true for f. The importance of this feature lies in the following
application. Recall that for a meromorphic q;, its logarithmic derivative q;' Iq;
has only simple poles with integral residues. Conversely, for any meromorphic
function <l> with this property we can find q; meromorphic such that <l> = q;' Iq;.
These two remarks together tell us that given a merom orphic function 1/1 there
is another merom orphic q; such that
q;(z + 1) = 1/I(z).
q;(z)
Retracing the steps we have shown that given a meromorphic function 1/1 one
can find meromorphic function q; such that (**) is satisfied. Clearly we can
multiply q; by an arbitrary periodic meromorphic function and obtain a new
solution of (**). All solutions are obtained this way. Being meromorphic, q; is
of the form e f Pt! P2, where f is entire and PI, P2 are Weierstrass canonical
products, but following the technique of the right and left formal sums used to
solve (**) one can anticipate the form of PI and P2. We shall illustrate this
below in the case of the r -function.
We can now state the main theorem of this section.
Proof. From the previous remarks we know there exists a nontrivial meromor-
phic function fl such that
CPr(z) =- L:-1
z +nO
n~
which has its poles at the points z = 0, -1, -2, .... From here we conclude
that a solution of the equation (!) is given by
e"'(z)
f(z) = z '
(1 + -)
00
z I1 e- z/ n
n=! n
where 1/f is entire and the denominator is the Weierstrass canonical product with
zeros at z = 0, -1, -2, .... Letting
= m->oo
lim (z + m + 1) exp [1/f(Z) -1/f(z + 1) - ~ .!.]
~ n
nl
n=!
-
The limit
y lim [~!
= m->oo log(m + 1)] ~ 0.557
~n
n=!
is the Euler-Mascheroni constant. Inserting this value and using equation (!)
we obtain
zf(z)
1 =
f(z + 1) =
exp[1/f(z) -1/f(z + 1) - y).
414 6. Hannonic Analysis
Hence,
1(1) = m--+oo
lim Im(l)
This function is identically zero for ~ > 0, continuous for ~ < 0, and it repre-
sents a distribution on IR if and only if there is k :::: 0 such that 1T}l k F(g + iT})
remains bounded in a punctured neighborhood of ;; = 0 (see [BG, Proposi-
tion 3.6.12]). In general ({! is a hyperfunction, as explained in Chapter 1. Intro-
ducing a change of variables;; = e27riz (or uniformizing parameter z), let us
define 1 by F(;;) = I(z), in other words,
1 (~IOg;;)
2Jrl
= F(;;).
Thus, 1 is an entire function of z which satisfies, for x = Re z = -!, y = 1m z,
the difference equation
I(z + 1) - I(z) = ({!(e 27riz ) = ({!(_e- 27rY ).
6.3. The Equation fez + 1) - fez) = g(z) 415
Note that with this notation it is natural to rewrite the original equation as
(; < 0).
(Equations of the form F(q;) - F(;) = cp(;). 0 < q, often arise in Number
Theory and are related to the theta functions mentioned at the beginning of this
section.)
Let us consider a special example, cp(;) = Log 1;1 for; < O. Then we have
cp(_e- 2Jl'Y) = -2rry for y E R By an easy inspection we see that fez) = rriz 2
satisfies
fez + 1) - fez) = -2rry (Rez = -4).
Hence,
F(n = -~Log2~,
4rrl
in C\] - 00,0]. Since cp E V'(IR), this answer could also have obtained with the
help of the Cauchy transform, as suggested by the general theory in Chapter 1
and in [BG, §3.6].
After these preliminaries, let us consider a type of question raised by Hurwitz
[Hu2, Vol. 2, p.752]. The homotopy group rrt (C*) of C* is isomorphic to Z and
it is generated by the unit circle I~ I = 1, oriented counterclockwise. This group
acts on S) by sending a function F to the function obtained by analytic continu-
ation along the unit circle. This is precisely what we have denoted by F(e 2Jl'in.
In this case F(e 2Jl' in n, n E N, means the new function in S) obtained from F by
analytic continuation along the unit circle traversed n times counterclockwise,
and for n < 0, traversed Inl times clockwise. Since the space S) is very large it
is natural to consider subspaces invariant under the action of rrt (C*). Typically
they are defined by functional equations. For instance, the equation
F(e2Jl'i~) = F(~)
describes all the functions in S) that are single valued, that is, the collection of
all Laurent series
convergent in C*.
Observe that if F E S), then its derivative F' also belongs to S). So that in S)
we have three types of operators acting:
D: F(n ~ F'(n;
A: F(n ~ A(~)F(~), multiplication by A E Jf(C*); and
r: F(~) ~ F(e2Jl'i~), "monodromy" operator.
It is not hard to see that r commutes with D and with A. For this reason Hurwitz
considered the following simple-looking equation in 1918:
(H)
DF = r F, that is,
F'(~) = F(e2Jl'i~).
The solutions of such an equation, if there are any, also form an invariant
subspace for the action of rrt (C*). It is clear that if we assume F is single
416 6. Hannonic Analysis
valued, then there is essentially only one solution for (H), namely F(l;) = ce s ,
for any c E C*. Hurwitz seemed to believe this was the only solution. Note
that when using the uniformizing parameter l; = e2rriz this equation becomes
essentially more complicated than those in Theorem 6.3.7. Namely,
d
(H*)
2 .
dzf(z) = 27rie rr/z fez + 1),
where fez) = F(e 2rriz ) as earlier. In 1975, Naftalevich [Na] used the method of
the formal sums to obtain many nontrivial solutions of (H*), i.e., different from
c exp(e 2rriz ). Earlier Hans Lewy had found the explicit solution
which is well defined and holomorphic for Re l; > O. To show that this function
belongs to the class S), let us consider for () E IR the auxiliary functions
FIJ(l;) :=
o
l
°oe;e 1
exp(-l;t + -. log2 t) dt.
47rl
In the integral defining FIJ we have t = re ilJ , 0 < r < 00. If I; = pe ia , p > 0,
then the integrand is absolutely convergent when cos«() + a) > 0, in particular,
if -7r /2 - () < a < 7r /2 - (). It follows that FIJ is holomorphic in the half-plane
-7r /2 - () < arg I; < 7r /2 - (). Note that if 0 < ()2 - ()\ < 7r, then the functions
FIJI and F~ have a common domain of definition, and applying Cauchy's
theorem they can be seen to coincide there. It follows that the collection (FIJ)IJEIR
defines a single function F E S). Moreover, for Re l; > 0 we have
F(e 2rri l;) = F2rr(I;).
This last integral can be computed from the observation that for r > 0
1 . 1 1 () ()2
- . log2(re'lJ ) = -. (Logr + j())2 = -. Log2 r + - Logr - - . '
47r1 47r1 47r1 27r 47r1
For () = 27r, we get
where the series converges locally unifonnly in the Riemann surface of log ~ .
This example indicates that it is natural to consider a wider class of equations
for multivalued holomorphic equations. Their natural name should be mono-
dromic differential equations, e.g., an equation of the fonn
L An,k(ODkr n F(O = 0
n.k
for F E 5), where the (finite) sum has coefficients A n .k E Jt"(C*).
More generally, we could consider infinite order differential operators or
consider other domains like C\{O, I}, C\{al, ... , ad, C\Z, etc. We remind the
reader that the case C\{O, I} is intimately related to the hypergeometric func-
tions. Here the group 1l'1(C\{O, I}) has two generators, and Riemann showed
that if a multivalued function F has the property that at any point ~ =J 0, 1, the
number of linearly independent branches of F is exactly 2, then F satisfies a
hypergeometric differential equation (see [BG, Exercise 5.15.2], [Tr] [Pool]).
The domain C\Z appears in the closely related theory of resurgence [Ec],
[Ram], [Mal].
EXERCISES 6.3.
1. Prove the converse of Lemma 6.3.2. Show also that if I is periodic meromorphic
and F(~) = 1((1/27ri) logO. the order of the corresponding zeros for I and F coincide.
When is f entire? When is the function F entire?
4. (a) Let I be a periodic meromorphic function such that limx-.oo I(x + iy) a for =
every y E R. Show that I(z) == a.
(b) Let I be a periodic meromorphic function such that the limits
limY-doo I(x + iy) = a± (possibly infinite) exist and are independent of x E [0,1], then
I is a rational function of eZrr;z.
5. Verify that the following sums are correct:
z)
(a) 6 ( n =
( z
n + l'
) (Z)
n
__ z(z - 1) ... (z - n + 1) n EN.
where
n."
418 6. Hannonic Analysis
8. Does Theorem 6.3.7 hold when either ao or at are identically zero (but not both)?
*9. Let g be an entire function of infraexponential type, g(z) = 2::n>O gnzn its Taylor
series about z = O. Show that the series fez) = 2::n>o[gnl(n + l)]Bn+~(z) converges to
a sum of g and it is also of infraexponential type. -
10. Study the solvability of the equation fez + 1) - af(z) = g(z), a E C\{O, I}, as
done in the proof of Proposition 6.3.5.
11. (a) Show that a solution of (z -1)f(z + 1) = 2z(z - 2)f(z) is the function
fez) = 2Z [r(z)/(z - 2)]. Find all the solutions of this equation.
(b) Let r(z) be a rational function. Find all the meromorphic solutions of the equation
f (z + 1) = r(z) f (z) with the help of the r -function.
12. Let 1/I(z) = r'(z)1 r(z). Show that
1 (l)m
6 - - = ----1fr(m)(z), mEN.
zm+t m!
Determine 6(z - a)-m. Verify that 6 log z = Log r(z) (at least for z E ]0,00[.)
13. Show that any function F E j) admits the expansion 2::n>O an 10gn C;, absolutely
and locally uniformly convergent in the Riemann surface of log f
14. For a E C solve the equation F(e 21Ci c;) = aF(C;), F E j).
15. For f E Jt'(C), D = dldz, a E C, one defines eaD fez) := 2::n~o(an In!)D n fez).
(a) Show that eaD fez) = fez + a).
(b) Fix P E Jt'(C) and define operators Land K in the space Jt'(C) by the formulas
Lf := e- Df - e P Df, Kf := e-D(e DDf). For an entire function rp define Urp to be the
formal series Urp = rp + Krp + K2rp + .... Show that formally
(c) Use Exercises 9 and 3 of this section to show that if f is a function of infraexpo-
nential type it cannot have any asymptotic period.
~(T)(s) =L ans n ,
n2:0
then
(T * f)(z) = Lanf(n)(z).
n2:0
In other words, one could write
T = L(-l)n an c5(n).
n2:0
We are interested in a particular class of operators of this kind, those for
which ~(T) is an entire function of exponential type zero, that is, T is carried
by {OJ. As we shall see, these operators have a number of interesting properties,
not shared by general convolution operators.
The function ~(s) = ~n>O ans n is in this case an entire function of expo-
nential type zero (see [BG, Chapter 4]) and we usually denote T f by *
~ (~) (f)(z) = Lanf(n)(z).
dz n2: 0
Recall that we denote Exp({O}) the space of entire functions of exponential
type zero.
is a continuous operator. The same is true lor this operator acting on the space
Oa = Jt'({a}), a E C.
Proof. The proposition is clearly correct once we know it holds for disks. On
the other hand, Propositions 2.4.2 and 2.4.3 have already shown this for any
open convex set Q. 0
6.4.4. Theorem. Let Q be a convex open set in C (resp. K convex compact) and
elI(d/dz) =I: 0 an infinite order differential operator. Then:
fn E Exp(f2) (the limit in the topology of Exp(f2», then f = <l>g for some func-
tion g E Exp(f2) (see Corollary 2.3.13). Since convergence in Exp(f2) implies
convergence in .if(C) we know there is a unique entire function g such that
f = <l>g.
From [BG, §4.S.S] (or the following lemmas), g is an entire function of expo-
nential type. We need to show that the indicator function of g is of the form
He, the supporting function of C, for some compact convex subset C of f2.
This depends on the following lemmas:
6.4.6. Lemma. Let <I> be an entire function of exponential type zero such that
<1>(0) #- 0, then,for every '). > 0 and every e > 0, there exists a constant De E lR
such that for z #- 0
Proof of Lemma 6.4.6. Let R = (I + ').)Izl. Since <I> is of exponential type zero,
given e > 0, there exist De E lR such that log 1<I>(u)1 :5 elul + De. Hence, the
422 6. Hannonic Analysis
subhannonic function
U t-+ log 1<I>(u)1 - eR - De
is negative in lui ~ R. On the other hand,
B(z, Alzl) ~ B(O, R);
therefore,
R2 A (log 1<1>1 - eR - De, 0, R) ~ (Alzl)2 A (log 1<1>1 - eR - De, z, Alzl).
By subhannonicity,
log 1<1>(0)1 - eR - De ~ A(log 1<1>1 - eR - De, 0, R).
Thus,
In other words,
1..2 A (log 1<1>1, z, Alzl) ::: (1 + 1..)2 log 1<1>(0)1 + (1..2 - (1 + A)2)(eR + De),
which implies the stated inequality. o
Proof of Lemma 6.4.5. Let e > °be given. We are going to show that
Ig(z)1 ~ Ae exp(HcCz) + elzl)
for some Ae > 0.
Up to translation we can assume that <1>(0) oj:. O. The hypothesis implies that
°
for every e, > there exists Eel> 0 such that
log I/(w)1 ~ HcCw) + e,lwl + Eel'
Let M := max{HcCu): lui ~ 1}, z oj:. 0, and A> 0, then
max
WEB(z,Alzl)
HcCw) ~ HcCz) + AMlzl.
Since log Igl is subhannonic we have
log Ig(z)1 ~ A(log Igl, z, Alzl) = A(log III, z, Alzl) - A(log 1<1>1, z, Alzl).
°
For e, > and e2 > 0, to be fixed later, there exist constants Eel and De2 given
by Lemma 6.4.6 such that:
log Ig(z)1 ~ HcCz) + AMlzl + e, (1 + A)lzl + Eel
- (1
+1..)2 log 1<1>(0)1 +
-1..- ((1 +1..)2 - 1) (e2(1 + A)lzl + De).
-1..-
Let
Ne = - (1+).)2
-).- log 1<1>(0)1 + ((1+).)2)
-).- - 1 De2 + Eel'
then
log Ig(z)1 ~ Hc(z) + slzl + N e •
In other words, there is R an analytic functional carried by C such that the
function g = F(R). 0
From Lemma 6.4.5 we can now conclude that part (i) of the theorem is correct.
(ii) Since (.1f(K))' is the space of analytic functionals carried by K the proof
of part (i) works verbatim.
(iii) Let J.1, E .1f' (Q) be orthogonal to M o, this implies that J(J.1,) is divisible
by <1>, J(J.1,) = <1> . h, h is an entire function. By Lemma 6.4.5 h E Exp(Q).
Hence there is R E .1f'(Q) with J(R) = h, thus
J.1,= T*R.
If f E M, then
(J.1"f) = (T*R,f) = (R,T*f) =0,
therefore Mo = M.
(iv) The same proof as (iii). o
We shall now prove that any solution f E .1f(Q) of a homogeneous differen-
tial equation of infinite order has a Fourier-expansion convergent in .1f(Q) of
the same kind as that found in the previous sections. It is clear that once we
have proved this result for an arbitrary open convex set Q it also holds in the
space Va. This Fourier representation goes back to the work of Valiron [Val],
Gelfond [Ge], Dickson [Di2]. On the other hand, the novelty of our proof is that
it obtains estimates for the coefficients of the expansion and it is essentially the
same as Theorem 6.2.7 up to technical details.
6.4.7. Lemma. Let <1> =I- °be an entire function of exponential type zero and let
°< s ~ ~. There exists R = Re °such that for any z with IzI ::: R there are
constants °< (I,s/8 < < sl4so thatforanyi; suchthatAlzl-(I
>
A Ii; -zl ~ ~
Aizi one has
while exp(-2slzl) ~ (I ~ 1.
Proof. Let us recall a property of entire functions of exponential type zero (see
° °
[Bo, Corollary 3.7.3, p. 52]), given 8 > and Y) > there exists ro > and a
measurable set E8 ~ [0, +oo[ such that if r ::: ro
°
m(E8 n [0, r]) ~ Y)r
424 6. Harmonic Analysis
Since cI> is of exponential type zero there is a constant D~ > 0 such that
Let us remark that the disk B(z, (e/4)r) S;; B. Since the sum of the diameters
of the Cj is 4Her < (e/8)r, there exist A, e/8 < A < e/4, such that
aB(z, Alz!) n Cj =0
for all j. On this circle we have
log 1cI>(~)1 ::: log IcI>(z') I - (2 + log(60))(48r + D~)
::: -8r(13 + 4 log 60) - (3Iog60)D~
whenever Iz - ~ I = Ar.
In order to obtain an estimate in the annulus, let us point out that the derivative
<1>' is also of exponential type zero, thus there is a value D~ such that
(~ E C),
therefore I<I>'(~)I :::: exp((1 + A)(£/4)r + (D~ + 1)) for U E B(O, (1 + A)r + 1).
Let w = z + le iIJ with Ar - a :::: 1 :::: Ar + a, then, for ~ = z + Are iIJ
1<I>(w) - <I>(~)I :::: a sup{l<l>'(z + reio)l: Ar - a :::: r :::: Ar + a}
:: a exp ((1 + A) ~r + (D~ + 1)) :::: aexp Gr + D~ + 1) .
Gr + D~ + 1) : :
We want
a exp e-er.
On the other hand, for r ~ r2 = 2(D~ + 1)/£ > 0, we have that the choice
In order to define the space A*(V) for V = V(<I», we first note that the
number of zeros n(r) of <1>, counted with multiplicity, in the disk B(O, r) satisfies
n(r)
lim - =0.
r--+oo r
If we denote the zeros repeated according to multiplicity by (Zn)n~l' with
Izd :::: IZ21 :::: .. " either they form a finite sequence or
. IZkl
hm -k
k--+oo
=+00.
Moreover, taking the constant Re from Lemma 6.4.7 bigger if necessary we
can assume
n(r) :::: £r for r ~ Re.
Assume <I> has infinitely many zeros, otherwise <I> is a polynomial.
426 6. Hannonic Analysis
Let us denote V = {(at. mk), k::: I}, lall ~ la21 ~ "', with distinct ak. Let
On = 2-n and Rn = max (Rsn , 2Rn_ l ) as given by Lemma 6.4.7.
We denote by kl,1 the first index k such that lakl ::: RI. Let AI,I > 0 associated
= =
by Lemma 6.4.7 to al,1 akl.l and 0 01. Denote
with Akl,2 chosen with respect al,2 := ak1,2 and 0 = 01. We let h,2 be the collec-
tion of indices of the ak E B I ,2 \BI,I. We continue in this fashion until we exhaust
all the zeros of <I> in this annulus.
We obtain in this way disks BI,j, 1 ~ j ~ Nlo and corresponding disjoint
index sets JI,j, 1 ~ j ~ Nlo with JI,j the collection of indices of the ak
belonging to the set
BI,j \ ( U
1~I~j-1
BI,I).
Observe that if ak does not belong to Uj BI,j and R3 > lakl ::: R 2 , then the
index k > kl,N I • We let k2,1 the first such index (if there is any such zero).
Denote
B2,1 := B(a2,1, A2,da2,1i)
with A2,1 chosen with respect to a2,1 := ak2,1 and 02 according to Lemma 6.4.7.
One can construct in this way a double indexed sequence of zeros (al,j),
1 ~ j ~ NI, disjoint index sets JI,j and closed disks BI,j.
Let Jo denote the collection of indices such that lakl < RI and also ak ¢
UI~j~NI BI,j.
This construction exhausts all the zeros of <I> and defines the groupings that
we need in the definition of III . III. Let us also point out that BI,j never intersects
Izl > RI+2' Let us now denote by VI,j the set of zeros aj of <I> counted according
to their multiplicity when i E JI,j, and let VI,j be the total number of points in
VI,j. Remark that we have
We recall that
6.4. Differential Operators of Infinite Order 427
6.4.8. Theorem. Let <I> be a nonzero entire function of exponential type zero,
then
Exp(Q) p
<I> Exp(Q) ---+ A*(V)
fez) = L(L
/o} aEJ/,j
p(AZ)e az ) ,
On the other hand, we can find l(p) ::: 1 such that for 1 ::: l(p), we have
Since there are only finitely many indices 1, j with 1 < 1(p) there is a constant
Bp > 0 such that
with
() = L (}I,j PI,j
I,j
coincides with Pl,j on a neighborhood of any ex E V1,j and there are good esti-
mates for a() / az, to be able to apply the idea of the semilocal interpolation
Theorem 2.6.4. The difficulty here is the need to estimate, very precisely, the
6.4. Differential Operators of Infinite Order 429
and an open set Uo which is the union of small disjoint disks surrounding each
ak with k E 10 and which is also disjoint from the U[,j'
We need to estimate the size of a region inside the aU[,j where we have good
lower estimates for <1>. This region is provided by the last part of Lemma 6.4.7,
For instance, for U1,2 we have a value O'l,l such that
and
I<I>«()I ::: e-EIIal,ll,
whenever AI,ilal,ll- 0'1,1:::: I( - al,11 :::: AI,llal,ll,
For U2 ,1 we have the subregion union of the two open subsets:
The construction of these regions for the general region U[,j is similar. We
shall call these regions collars. We need to determine what estimates we have
on these collars and their size. For this purpose, we recall that if two disks B[,j
and Bk,i, with I :::: k, intersect, the indices k and I differ at most by 1. Therefore,
at a point,
and
we have
and
For the same reason the width of that portion of the collar is CTl,j which satisfies
the inequality
Therefore we can find functions 8/,j E 'D(UI,j) which are identically 1 except
on the collar, 0 ::: el,j ::: 1, and
_ 1 _ 8el,j
1ef>(z) 8z
(Z)I < ei/2Ce24eda/,j\ < ei/2Ce48Bdz\.
- -
Since e = E/,j e/,j P1,j we are searching for a function Ili such that
g = e + \I1ef>
belongs to Exp(n). For this reason, we choose \11 as a solution of the equation
8\11 1 8e "" 1 8el j
(*) 8z = - ef> 8z = - L..-J
1 ' P1,j ef> 8i .
,} ,
[I
we use the fact that
1 8e 12 -q>(z) 2 2 [ dm
Jc ef>(z) 8z (z) edm ::: Dpillalilp Jc (l + Iz12)2
::: D~lIlalll;
to choose \11 so that it solves (*) and
!
C
(1
11fr(z)12
+ Iz12)2
e-q>(z)dm < ID' IIla1l1 2.
- 2 P P
6.4. Differential Operators of Infinite Order 431
Now, we introduce
CPo(z) = cp(z) + 2 log (1 + Iz12) + Eizi
and we obtain
We can choose E = T/p/2 and find a new constant Bp > 0 such that
Ig(z)1 :::: BplllalllpeHp(z)+'1pIZI
:::: BplllalllpeHp+l(z).
This concludes the proof of the isomorphism between Exp(Q)j<l> Exp(Q) and
A.(V).
The rest of the proof of Theorem 6.2.7 and subsequent considerations, includ-
ing estimates of the coefficients of the Fourier expansion hold verbatim. The only
thing that requires a minor remark is the fact that if f has two Fourier-expansions
converging in Jt'(Q) the coefficients are the same because the operators Tk used
in the proof of Proposition 6.2.8 are also infinite order differential operators in
this case, since :F(Tk)(z) = <I>(z)/[(z - ak)m k ] are entire functions of exponen-
tial type zero. 0
(k 2: 1).
432 6. Hannonic Analysis
We say that am has the property 'P if there exists n so that {am, an} forms a
pair (i.e., it satisfies condition (iii».
of A 1 has the property 'P and that there is a constant e satisfying 0 ::: e < T{ 12,
so that for every n E N
either I arg An I ::: e or I arg An - ]I" I ::: e.
Then there exists an even entire function Fl of exponential type zero, vanishing
only at the points z = ±An and satisfying "Ie > 0:
(a) I/1F1(re ia )1 = O(e Br ) whenever Ire ia ± Ani::: c/8 as r -+ 00.
(b) I/IF{(A n )1 = O(eBIAnl) as n -+ 00.
that follows is not affected by this assumption. Denote by L the number of the
terms in B distinct from AN. Let B(z) be the finite product
( z)
B~)= 2 .
I1 1--
AneB,n"#N A~
Thus,
Observe that for every n there are at most two terms equal to IN - nl. Hence,
[4
the last product has at least L] distinct terms, so that
II IN - nl ~ ([L/2]!)2.
AneB,n"#N
S (_4_)L e(LlogL)lzle3Le-logL
ccos&
= (_4_)L e3Lellzl(L/lzlloglzl/L)}.
ccos&
L Izl (L)I/2
-log - < -
Izl L - Izl
<
-
h vIee ,
so that
436 6. Harmonic Analysis
for a convenient constant I.t > O. The last point of this step of the proof is to
bound the term involving AN. From (*) we obtain
I( 1 - I
.::.) > cosOlz - ANI
A~ - 21z1 '
so that, altogether,
Step 2. This time we are going to estimate IFA (z)1 from below. Similarly
to the previous step, let a be the value of the index n that minimizes IAn - zl
among An E A. We see, as earlier, that when An E A we have
1
2 )M 1
::s ( - - IzlM ([MI2]!)
cose
2 ::s r M e3M exp[Mlnlzl- MlnM],
for some constant r > O. Again we conclude that for a convenient constant
a > 0 and all Izl » 0 we have
IA(z)1 ::s e<TJElzl.
On the other hand, we have that IAal ::s (1 - e)lzl and, hence, we also have
Iz - Aal 2: IZI-IAal. Thus,
1 IAal2 1 IAal 1 IAal
= < -- < - - --'---'---
II - z2/A~1 Iz - Aallz + Aal - cosO Iz - Aal - cose Izl-IAal
1 (1 - e)lzl 1 (1 - e)lzl 1 (1 - e)
<-- --- =-----
- cosO Izl- (1- e)lzl cosO elzl cosO e
Combining this inequality and the bound for A, we obtain for some constant,
1] > 0, the lower estimate
Step 3. Finally we have to consider the function Fe. When An E C one has
Izl/IAnl < 1/(1 + e) < 1, so that
1 £tl = £t.
1-
IAnl2
1-
IAnl 2
Thus, if r = Izl, rn =lAnl, we have
IIe (1 - r:)
rn
~ II
e
11 - Anz: 1= lFC(z)l·
1 1
It is immediate that
"e .rn!.
~
<
2-
00
(l+e)r
dnAI (u) <
U
2 -
00
(l+e)r
2nAI (u)
du
u3 '
21
where the last inequality was obtained after integration by parts. Note that
00
2nAI (u) d nAI (u)
r u <r max - -.
(I+e)r u3 - u~(1+e)r U
IIe (1 -r2/r2)
1 " .!.))
~ exp ({3r 2 (~ r2 ~ e2{3er.
n e n
This inequality implies that
lFC(z) I ::: e-2{3e lzl•
This concludes the third step of the proof.
It is clear that, after convenient renormalization of the constants, we have
shown that given e > 0 there are constants rE > 0 and y > 0 such that for
Izi ::: rE we have
Ylz - ANle- elzl ~ IFI(z)1 ~ eelzl.
Since F{(AN) = lim --+Z AN FI(z)/(z - AN), the two properties (a) and (b) follow.
Let us now observe that, because the counting function nA 2(r) = o(r), given
8 > 0 if r > 0 is sufficiently large then
This implies that adding the other element of the pair, if it is not already included
in the sum (**), does not change the nature of the problem. On the other hand,
if Am, An is a pair we have
1 1 I lAm + Ani c 1
1
Am + An = IAmAn I ~ 21 AmAnI'
Since the series L::l (l/IAn 12) is convergent, the lemma is completely proved.
o
The main point of the next lemma is that we do not make any assumption on
the arguments of the sequence.
1
(b) = O(eEIAnl) as n ~ 00.
IF~ (An)1
Proof. If the genus of the sequence were zero, we could use a simpler construc-
tion of F2, but we prefer to give a unified formula, irrespective of the genus.
Using Lemma 6.4.13 define w := limHoo L:IAnl~,(1/An). Let us consider the
function
F2(Z) = e- wz IT (1 - ~) An
e Z / An := lim e-wz
'-'>00
II (1 - ~)
An
e Z / An ,
n=1 IA.19
where the infinite product is defined by the limit of the finite products. The
reason is that the IAn I are not necessarily increasing. Note that in the genus
zero case, this product is not the Weierstrass canonical product [Lev], while
this is true when the genus is 1. Nevertheless, the usual estimates show that
F2 E Exp({O}) (see [Lev, Theorem 15, p. 26].) Thus, we need to concentrate in
6.4. Differential Operators of Infinite Order 439
II 1
< (1 +£)LlzIL2L
1
II
AnEB.n#N 11 - Z IA nI - AnEB.n#N IA N - AnI
As was done in Lemma 6.4.12, we can now find a constant K ~ 1 such that for
Izl » 1 we have
Step 2. It is clear that the same proof used in Step 2 of Lemma 6.4.12 yields
the following estimate:
II 11 -: I ~ e-K'v'elzl,
An EA n
Step 3. Let us now assume that Izl is sufficiently large so that £Izl > e12. It
follows that if an element of C is such that the other member of the pair is not in
C, then it must lie in B. Let us consider here the subset D of elements of C such
that the other member of the pair is not in C. We shall denote by E the remaining
portion of C.1t is clear that An ED implies that IAnl :'S (1 + £)Izl + e12, hence,
D is finite, and if we denote by P the number of its elements, then
P :'S ni\2((1 + £)Izl + e12) :'S £Izl,
for Izl » 1.
Let As denote an element of D that minimizes Iz - Ani among An ED. Then
(1 + £)Izl :'S IAsl :'S (1 + 2£)lzl, so that
----=
1
II - z/Asl
IAsl
< IAsl
<
lAs - zl - IAsl - Izl - (1 + £)Izl - Izl -
(1+2£lzl
<-- 1+2£
£ .
On the other hand, I1A nED.n#s 11 - z/Anl can be estimated from below as in
Step 2 of Lemma 6.4.12. As a consequence, we obtain that there is p > 0 such
440 6. Harmonic Analysis
Step 4. In this part, we are going to estimate the product of all exponential
factors with frequencies Am ED. As a consequence of the estimate of P in the
previous step, we have
"'--<
~D
Am E
1
IAml - (1 +e)lzl -
P
<e,
and thus,
11 eZ/ 2: e-eiz i.
Am
Am ED
Step 5. In this part we take care of the exponentials of the terms which are
"near" the origin, that is, with frequencies An E A U B. From the definition w
we have that for Iz I » 1
1
~ ;:- - w ::: e.
An EAUB n
Therefore,
e{J)Z 11 e- Z/An 2: e-eizi.
An EAUB
Step 6. Finally, we are left with the task of estimating the infinite product
over the set E. In this case,
J:l < _1_ < 1.
IAnl - 1 + e
Hence, we can apply [Lev, Lemma 3, p. 11] to obtain
log 11 (1 -
E
~) e-
An
Z/ An 2: -12elzI 2 (Xl nE!t) dt.
JIZ I t
2:
Since, for t Izl » 1, we have nE(t) ::: et, we can estimate the integral by
elzl- 2 /2. Thus,
11 (1 - ~) eZ/ 2: e-6eelzl. An
E An
The collection of estimates we have obtained concludes the proof of Lemma
6.4.14. D
Proof. We introduce the pairing of Definition 6.4.11, then the tenns of the
sequence {An}~1 are divided into two sets:
Further generalizations of the theorem can be found in [BLV] (see also [GR]
for related results).
For the applications of infinite order differential operators to the question of
overconvergence of Dirichlet series we need first a generalization of Theorem
6.4.8 to convolution operators.
442 6. Hannonic Analysis
is a continuous map.
This map is surjective for arbitrary Q if and only if the entire function defined
by ~(T) = 4> is of completely regular growth. We refer the reader to Section 2.4
for the details.
(ii) For every e > 0 there exist A e, Be > 0 such that the diameters 8n.e of the
connected components Se.n of the set
Se := (z E C: d(z) < Ae exp( -elz!)}
satisfy
where V(4)) V = = {(akt md: k ~ 1}, Pk are polynomials of degree strictly less
than mk. The convergence and the estimates are similar to the previous theorems
of this kind.
Sketch of proof. As in the proof of Theorem 6.4.8 the main point is to define
correctly the groupings and the corresponding norm III . III. The principle is the
6.4. Differential Operators of Infinite Order 443
where ~n,j is the diameter of Un,j, which can be estimated in terms of the
quantities Bn , and i runs over the sequence i = 0, ... , Vn,j - 1.
Let (Kp)p?:1 be an exhaustion by compact convex sets of Q and define
Illalilp = sup{lila/,jlll/,j exp(-HKp+K(a/,j))}'
Using the fact that the convex sets Kp +K exhaust Q + K and
HKp+K = HKp + Hko
we can prove that
Exp(Q + K) ~A*(V)
cI> Exp(Q)
is a topological isomorphism. The rest of the proof is the same as in
Theorem 6.4.8. 0
6.4.18. Lemma. Assume further that for some 0 < ~ < () there are at most
finitely many zeros of cI> in the set
f
f
f
f
f
Figure 6.1
and A**(V) are isomorphic. In fact, this means the dual spaces coincide, and so
the coefficients of the Fourier expansion of f in Q + K are also the coefficients
of the Fourier expansion of some entire function f, T * f = O. It is clear, that
f is the analytic continuation of f to the whole plane.
In order to prove this isomorphism all we need to show is that, for every
compact convex set L in C, we can find L J, a compact convex subset of Q + K
such that
The reader can convince himself that it is enough to take a convenient closed
vertical segment in r ll (sufficiently far to the right). (See Figure 6.1.) 0
If we do not have the extra assumption of Lemma 6.4.18 we can still conclude
the following:
6.4.19. Lemma. Let ro = {z E C*: 77:/2 + e < Arg z < 377:/2 - e}. Assume
*
Z(<1» 5; ro u (r;)C for some 0 < 8 < e. Let f E Jf"(Q + K), T f = 0, be
such that for some 8 > 0 the polynomial coefficients Pa of f in the Fourier
expansion vanish for all a in rD. Then f is an entire function.
Proof. This time we only have that every entire solution of the convolution
equation is a solution in Q + K but the converse does not necessarily hold. For
the f of the statement what we have to show is that the sequence b E A*(V)'
o'
that corresponds to f belongs to A**(V). For al.} E r we have bl ,} = O. For
6.4. Differential Operators of Infinite Order 445
the other indices (l, j) the construction of the compact L! in the previous lemma
still works, that is,
Hdal,j) S H L , (a/,j)
if al,j ~ r;. Remarking that the groupings can be made in such a way that no
points in ro and (r;y are in the same group, the last inequality shows that b
is a linear continuous linear functional on A**(V). 0
P (:J f=O.
S u B(D,e)
Figure 6.2
446 6. Hannonic Analysis
is also holomorphic in Se. Now, for any a E C and any given polynomial q(z) =
EO~j~d CjZ j we have
we have
q(z + 'l')ea(zH) = 1/1 (:z) (q(z)e az ).
Since akllakl --+ ±eirp when k --+ ±oo, respectively, then, for any 8> 0 we can
find D8 such that
so that
Let
u~ = {z E S: z + i~e-irp E S},
6.4.21. Theorem. Let T satisfy the hypotheses of Proposition 6.4.17 and Lemma
6.4.19, r/l c Q + K C {z E C: Rez > OJ. Assume further that the zeros of <I>
wtisfy a gap condition in rij, that is there are finitely many unit vectors e irpj ,
1 ::.: j ::.: r, pointing into rij such that the zeros in rij form sequences (aj,kh:::l
so that
lim -)'- = e,rpj
a'k .
(1 ::.: j ::.: r)
k-+oo laj,kl
and
. laj,kl
11m - - =00.
k-+oo k
Moreover, assume that the functions
are entire functions of exponential type zero whose varieties V(Pj ) are inter-
°
polation varieties for the space Exp({O}). Under these conditions, any function
f E Jt'(Q + K) that satisfies the equation T * f = and has an analytic contin-
uation to a disk B(O, e) for some e > 0, has also an analytic continuation to the
open set
f+ = L: qa(z)e°/Z,
aE(r;l'
f- = L: qa(z)eaZ ,
aEro
Let
which is holomorphic in
So = ro U B(O, e)
and satisfies the equation
Then
Thus,
P3 (:z)··· (:z)
Pr h2 = g2·
In other words,
Theorem 6.4.15 appears for the first time in [BStl]. The classical Fabry Gap
Theorem (see [Levs], [DieD is a particular case of Theorem 6.4.15. It states that
if 0 < Al < A2 < ... is a sequence which satisfies the conditions:
(i) limn~oo(Anjn) = 00.
(ii) For some constant C > 0 one has An+1 - An ::: C, then, every Dirichlet
series
fez) = Lane- Anz ,
n;::1
which has abscissa of convergence aa, -00 < aa < +00, has no regular
point on the line Re z = a a .
In fact, one recognizes that f is a solution in Re z > aa of the infinite order
differential equation P(djdz)f = 0 with
6.4.22. Lemma. Let <l> be a nonzero entire function of exponential type zero and
K, L two convex compact sets such that K £ Land
(~ E Z(<l>)).
450 6. Hannonic Analysis
Then,/or every entire/unction / E Exp(L) and every e > 0 one can find entire
functions g and h, such that g E Exp(Ke ), h E Exp(L e ), and
/ = cl>h + g,
where K e, Le are the open e-neighborhoods 0/ K and L, respectively.
Proof. One can certainly assume that K =F L and that cl> is not a polynomial.
(In the latter case it is enough to take g a polynomial that interpolates the values
of / on V(cl».)
In the general case, when Z ( cl» is infinite, we use the construction of the sets
UI,j from the proof of Theorem 6.4.8 and corresponding standard functions Ol,j'
Introduce now the function
o := "'I)OI,d).
I,j
I 1_1_
og cl>(~) a~ (~)
aOl,j 1-< 48 I~ I + C
21 0
(~ Eq.
Given e > 0 we can choose 71 = e/16 and 11 (e) ::: 10(e/16) so that if I ::: It (e),
then
1 e
21 (2ML + 4ML + 271 + 48) ::::: 4'
Hence. there is a constant C 1 ::: 0 for which
e
log IO(~)1 ::::: HK(~) + 21~1 + Cl
6.4. Differential Operators of Infinite Order 451
and
1
10g<1>(S)
\
ae \
a~ (l;) ::: HK(l;) + "2e 1l;1 + C 1•
We can conclude that there is a Coo function 1/1 such that the function
g= e + 1/1<1>
is an entire function satisfying the inequality
e
log Ig(l;) I ::: HK (l;) + "2 1l; I + 410g(1 + Il; 12) + C2
for some C2 ::: O. Namely, if
p(z) := 2Hdz) + elzl + 210g(1 + Iz12)
there is a solution 1/1 of the equation
a1/1 1 ae
=
az <1> az
satisfying
where the union takes place over all compact convex subsets of U. Let us note
that r A(U) is a convex open set. In fact, each YA (K) is clearly a closed convex
set, and by taking an exhaustion (Knk::l of U by compact convex sets we
see that r A (U) is an increasing union of convex sets and hence convex itself.
°
Moreover, if e > is sufficiently small so that Ks = K + B(O, e) £; U, then
YA(Ks) 2 YA(K) + B(O, e) = (YA(K))s,
which is a neighborhood of YA (K).
452 6. Harmonic Analysis
8 A(K) = U8 A,r(K),
r"::O
8 A (U) = U8 A (K).
KccU
These three sets are convex and 8 A (U) is open. In fact, for K CC U let
B =BK > °
such that Ks CC U. Then, for every r ~ 0,
8 A,r(Ks) :2 8 A,r(K) + B(O, B).
Hence,
therefore,
o ..£...
(where X denotes the interior of X, while X is the interior of the closure) and
we denote by f3(A) the subset of the unit circle '][' of points ~ for which there
is a subsequence (~jkh"::l with
6.4.23. Lemma. Let A = (~j)j"::l' limj ..... oo I~jl = +00, and let U be an open
convex set in C. Then
If we translate the set U keeping the set of directions f3 (A) fixed, then B A (U)
and C.(A)(U) are translated by the same amount. Therefore, ~e can assume that
o E U. Let K be a compact convex subset of U with 0 E K, then there is a
constant Ak such that for 1 < A < AK, the set AK is a neighborhood of K still
contained in U. We fix such a A.
Observe that there is an e > 0 such that the assumptions
«( E q
because 0 E k. From this inequality, it is easy to see that there exists 8 > 0
such that the assumptions
z E K, ( E a(A), and
imply that
Hence, if
( E a(A), Re(z() .::: HKC(), and
then,
(*) Re(z7]) .::: H K (7]) + H K «( - 7]) + Re(z(7] - () .::: Ha(7]).
Denote
a(A,8) := {7] E C: 3( E a(A) with I( - 7]1.::: 817]1}.
The inequality (*) implies that
Ya(A)(K) ~ Ya(A,8)(K).
It is also clear from the definition of f3(A) that there exists r > 0 such that
Ar := A n (B(O, r»C ~ a(A, 8).
It is then immediate that
n
Hence,
Ya(A)(K) ~ ABA(K) = BA(K)
A>1
and
o
We finally have here the theorem about analytic continuation of solutions of a
differential operator of infinite order due to Sebbar and Aoki (see also [BOV]).
hence,
(~ E Z(<I»).
We are thus in the situation of Lemma 6.4.22 so that for a convenient e > 0
we have
~(T) = <l>g + h
6.4.25. Corollary. Let n, 0, and <I> be as in Theorem 6.4.24. Assume that the
functions h E Je(O) and g E Je(n) are such that
<I>(:Z)G=h.
so that
<I> (:z) (g - (Goln» =0 in n.
456 6. Hrumonic Analysis
Q= ra(A)(Q) = (
sE,8(A)
n (z E C: Re(z~) :::: Hn(~)}) .
In comparison with the Fabry Gap Theorem 6.4.15, one can see that if the
sequence Ao = (adk2:1' where the ak are nonzero complex numbers such that
lim lakl = 00
k~oo k
(but not necessarily distinct), and
f3(Ao) f: { e"P:
. 31f}
e + "21f :::: qJ :::: 2" -e
for some ° < e < 1f /2, then any function
f E Jt'(re U B(O, e)) = Jt'(re,E)
which satisfies the equation
with
Observe that the set -f3(Ao) does not really intervene in the construction of the
domain Qe,E since
r -,8(Ao) (re) = C.
The domain Qe,E' where all the solutions f extend, can be found as follows:
/
/
/
Figure 6.3
EXERCISES 6.4.
*1. Use the Minimum Modulus Theorem to prove the density result used in the proof
of Lemma 6.4.7. Namely, let <I> '# 0 be an entire function of exponential type zero and
denote m(r) := min{log I<I>(~)I: I~ I = r}. Then
-m(r) = o(r),
except on a set of linear density zero. That is, given 8 > 0, if £8 := (r ~ 0: m(r) < -'-8r},
=
for any" > 0, there is ro ro(8,,,) such that ml (£8 n [0, r)) ::: "r for any r ~ O. (Here
ml denotes the Lebesgue measure on R)
*2. From the construction of the components Vt,i in the text preceding Theorem 6.4.8,
conclude the following result holds for any entire function <I> '# 0 of exponential type
zero: Let Z = (z: <I>(z) = O}, then for any e > 0 there is a value Re > 0 such that
dist(z, Z) ~ elzl implies 1<I>(z)1 ~ e-e1z l •
(Hint: For R~ »1 any connected component of the set S(<I>, e) = (z E C: I<I> (z) I <
exp(-elzl)} that contains any point with Izl ~ R;
must have diameter not exceeding
458 6. Hannonic Analysis
elzl/2. Assume dist(z, Z) ~ elzl, Izl ~ R., z E S(cI>, e), to conclude that log IcI>(~)1 is
hannonic in the component that contains z. Use the minimum principle for hannonic
functions to conclude the proof.) Note that this result can be used to give a slightly
simpler proof of Lemma 6.4.16.
3. Let cI>(z) = cos 0, Z = {z E c: cI>(z) = OJ. Find the asymptotic cone a(z) and f2
from Theorem 6.4.24, in the case
0=06 = {z E C: Re(e i6 z) > O} (0::: (} < 2rr) and 0 = B(O, I).
6.5. Deconvolution
The problem of solving an equation of the form
f.L * f = g,
where f.L is a measure with compact support in JR, appears often in engineering
and other applications of mathematics. As we know, this equation does not have
a unique solution in the spaces E(JR), 'D'(JR), not even if we assume f E LOO(JR).
For instance, the Fourier transform :Ff.L could have a real zero a and then
*
f.L eiax = O. For that reason, it is customary to restrict the domain of the f
considered, to a space like L2(JR) or LI(JR), so that there is at least uniqueness
of the solution.
In fact, if f ELI (JR) then the Fourier transform :Ff is a continuous function
*
which tends to zero at infinity, thus f.L f = 0 implies
(~ E JR).
Since :Ff.L = 0 is a discrete set, it follows that :Ff (~) is identically zero. There-
fore, f = 0 a.e. Nevertheless, the operator
f E LI(JR) ~ f.L *f E LI(JR)
and we are looking for a solution pair (ill, il2) E F(£'(IR)) = Ap(C) for the
weight p(z) = I Imzl + log(2 + Izl).
From Hormander's Theorem 2.8.15 (see also Theorem 2.8.16) we know that
the necessary and sufficient condition or the solvability of (* * *) is that there
are constants £ > 0, C > 0, such that:
(~E C).
In the engineering literature this condition is sometimes called "strong coprime-
ness." Generally speaking, this is simply a condition of how separated are the
zeros of ill from those of il2.
We would now like to construct explicitly deconvolvers VI, V2 for distributions
ILl, IL2, which satisfy a number of conditions that are often found in concrete
examples. We point out that most of the limitations that we are going to impose
in Proposition 6.5.1 can be eliminated by paying the penalty of more complicated
formulas for the deconvolvers.
(ii) All the zeros of ilj are simple and they are constants £ > 0, L > 0, such
that
ilj(O = 0 implies lilj(~)1 > £ L (j = 1,2).
- (1 + I~I)
(iii) There are constants 8 > 0, M > 0, such that
(~ E Yn, n 2: 1).
u q J.,tl (U)J.,t2(U)
A ,U = 0 .
Proof. Let
1 = _1_1_1_ = _1_
27Ti y. ~ - z 27Ti
r O(~)
Jy• O(~)(~ -
- O(z) d~
z)
+ O(z)
27Ti
r
Jy• O(~)(~ -
d~
z)
.
We observe that for any z fixed, this formula is valid as long as n is sufficiently
large (i.e., n 2: n z ). Let us now compute by the residue theorem the first integral:
27Ti
1 1 y.
O(~) - O(z) d
O(~)(~ _ z) ~ = J.,tl
A ( ) A
Z J.,t2 Z
( )P( )
Z
q 1 ft2(Z)
+z
A ' "
J.,tl(Z) ~ f3 qA (f3)AI(f3)Z_f3
#2 (fJ)=O J.,t 1 J.,t2
,BeInt(y.)
6.5. Deconvolution 461
~ A 2e A (3 + Izl)AeAllrnzl (z E C).
(z E C)
and
(z E C).
and, similarly,
and
1
L
il2(tl)=O
(1
+
1,8
I)
2 < +00,
(z E C).
462 6. Hannonic Analysis
To finish the proof we have to show the other integral goes to zero as the
integer n -+ +00. For n sufficiently large so that Clrn > Izl, we can use hypoth-
esis (iv) to obtain
O(Z) ( d~ I IO(z)ll(Yn) 1 IO(z) IEr;q+N+I
I
21ri Jr. O(~)(~ - z) ~ A21r(clrn -Izl) (clrn)q-N ~ 21rA(clrn -Izl)'
Condition (i) is the condition that both ILl and IL2 are hyperbolic distributions.
In this case the inequalities (ii) are exactly the condition that their zero sets are
interpolation varieties for .r(E'(lR.». Moreover, condition (iii) is exactly the
requirement that strong coprimeness holds on the set Z(ili/12). In particular,
the proposition says that if the distributions are hyperbolic, the zeros of ill and
il2 are simple, and the varieties V(ill) and V(il2) are interpolation varieties,
then the strong coprimeness on the zero set Z(illil2) implies strong coprimeness
everywhere.
We also note that if we define distributions VI, V2 by
VI (~) = ~q gl (~) + ~ P(~)il2(~)'
~(~) = ~qg2(~) + ~P(~)ill(~)'
then
VI * ILl + V2 * IL2 = B.
Moreover, due to the explicit formulas from Lemma 6.1.14 the distributions
VI and V2 are very explicit. It is enough to recall that
IILI(~)· IL2(~)1 ~ (1
6.5. Deconvolution 463
and the Jordan curves Yn can be constructed as follows. The distribution ILl * 1L2
is invertible, since each of them is surjective in C(lR), therefore, the function
P-IP-2 is slowly decreasing. For r e R fixed, Irllarge, we apply the Minimum
Modulus Theorem to construct an arc of a circle centered at r of radius between
2Tlog(2 + Irl) and 4Tlog(2 + Ir\) on which the function 11/P-I(~)P-2(~)1 is at
most C1lrlQ for some a > O. The intersection of this circle with the logarithmic
strip of hypothesis (i) will have two connected components, we choose the
outermost. We do the same for the point -r, and we complete a Jordan curve
by adding two semicircles lying outside the logarithmic strip.
Let us now see that under the assumption (Co), the crucial condition (iii) is just
a separation condition between the sets Z(P-l) and Z(P-2). It is the condition
they are well separated (see Section 3.1), that is, there, are constants e > 0,
N > 0, such that for every a e Z(P-l) and fJ e Z(P-2)
1
la - fJ I 2: e -(I-+-I-a-1+-lfJ-I)"=N .
and, by (.co),
C3
+ la1)N+N2
A
11L2(a) I 2: (1
for a convenient constant C3 > O. Conversely, if the zero sets are not well
separated there must be sequences an ~ 00, fJn ~ 00, of zeros of P-l, P-2,
respectively, such that
1
Ian - fJn I :5 (1 + Ian I + IfJn \)n
Since P-2 is also in F(C'(R», it follows that
B(2 + Ian I)A
11L2(an)l:5 (1 + Ian \)n
A
for some A > 0, B > 0, fixed. Hence condition (iii) cannot be satisfied.
In fact, in general, the condition (.co) cannot be expected to be true for P-,
even for exponential polynomials, unless the origin belongs to the interior of
cv(supp IL). For instance, for IL = al - a3 we have
P-(~) = e-i~ - e-3i~ = e-2i~ (i~ _ e-i~)
= 2ie-2i~ sin~.
We have
(~ e IC),
=
while le-2i~ I e21m ~, so that (.co) cannot hold for 1m ~ < o. Multiplying P- by
exp(2i~) solves the problem.
464 6. Hannonic Analysis
If (£1) holds, then the function (sin A{)/l({) will satisfy conditions (£0)' In
this case, we could apply the proposition to the products (sinA{) . /lj({).
Let us now consider some typical examples for which Proposition 6.5.1
applies.
6.5.2. Examples.
(1) Let ILj = X[-aj.aj] (aj > 0, j = 1,2), then
sinaj{
A
ILj {)
(
=- {
- ({ E q,
Z ILj)
(
A
= ajrr 'lI*
-ILJ •
These functions satisfy (Co) and the only condition we need to verify is
whether the zeros are well separated. That means there must exist constants
8>OandN>Osuchthat
This is exactly the arithmetical condition that the quotient at!a2 is badly approx-
imated by rational numbers. In particular
al ¢Q.
a2
We remind the reader that the theorem of Roth [Ba] implies that if al / a2 E Q\ Q
(that is, an irrational algebraic number) then, for every TJ > 0, there is I) > 0 such
I:: -~12:
that
(Ipl +~qI)2+~'
For a quadratic irrational we can take TJ = O. In any case, we can take N =3
in the earlier inequality whenever at!a2 E Q\ Q.
(2) Let us we consider a distribution IL of the form
IL = X[-a,a] + CI,
where CI is a C 2 function in R with SUpp(CI) £ [-a, a]. One can see that /l
satisfies the Lojasiewicz condition (£0) because
ILj * 1= 0
then I == 0 in ] - R, R[.
We remark that when R < 00, which we can assume, this proposition is not a
consequence of Proposition 6.5.1 (even assuming the extra conditions required
there). The reason is that the deconvolvers VI, V2, in general, will not have
support at a point, hence one cannot obtain information about I in the whole
interval] - R, R[.
Proof of Proposition 6.5.3. Let V(/1,I) = {(ak,md, k::: I}; from Lemma 6.1.24
we conclude that
I(x) =L Pk(x)e iakX
k;::1
It is clear that the proof we have just given cannot work if al + a2 > R, the
reason being that the convolution IL2 * Ok has too large a support. In fact, the
result could be false as the following example shows:
Let
6.5. Deconvolution 467
Let b = R - ../21f < 1f and let ({J i= 0 be a Coo function with compact support
in the open interval ]b, 1f[. We can extend it to IR as a periodic function <l> of
period 21f. Therefore we can expand <l> in a Fourier series
<l>(x) =L cne inx ,
nEZ
I(x) = Lbneinx
nEZ
6.5.4. Examples.
(1) Generalization 01 Morera's Theorem.
Let 1 be a continuous function in the disk B(O, R) and let To be a closed
triangle (or a square, or a simple polygon, etc.) such that
To 5; B(O, R/2).
Assume that
r I(z)dz = 0
JaT
for every T 5; B(O, R) which is congruent to To (i.e., T is obtained from To by
a transfonnation of the fonn z ~ ei () z + a).
The conclusion is that 1 is holomorphic in B(O, R).
!(z)=A(f,z,r)=_1
2rr
r
Jo
2
:rr !(z+rei(})dO.
E = {~l: .jiiJO(~j) - 1 = 0 (j
~2
= 1,2) and ~l
~2
> o} .
(Here Jo is the Bessel function.)
We observe that 1 E E, just take ~l = ~2; in general, the numbers ~j are
complex and we only consider those whose quotients are real and positive.
Delsarte has shown that this set is finite [De2].
Now let rl > 0, r2 > 0, be such that
rl rf. E
r2
and
6.5. Deconvolution 469
Under these two conditions, if f E C(B(O, R» satisfies the mean value prop-
erties
f(z) =
AU, z, rl) for Izl < R - rl,
and
These two examples were included to give the reader an inkling of an area of
current research in integral geometry in which the reader will find a profitable
use of the subjects covered in this volume and the preceding one.
EXERCISES 6.5.
1. Let a > 0 and consider the lattice Za generated by a. Show that Za is dense in JR
if and only if a is irrational.
2. Use the previous exercise to give a direct proof that if at. a2 > 0, a./a2 ¢ Q, then
the only continuous function in JR, such that J~~. f (x + t) dt = 0 for all x E JR, j = I, 2,
J
is the zero function. How does this relate to the deconvolution problem?
3. Let ILl = X[-a,a), a > 0, 1L2(X) = eilJx X[-a,a) (x), 0 < (J < If. Are there solutions of
the Bezout equation ILl * VI + 1L2 * ~ = Ii, with Vt. V2 E t:"(JR)?
We only make explicit reference to the main papers and books we have consulted. In many
of the subjects of the book there are extensive bibliographic references already available,
e.g., the relation between harmonic and complex analysis is very well documented in the
the surveys [BSt2] and [Nil]. In any case, the literature on the subjects covered in this
book is extremely vast, and it is quite possible that we have inadvertently omitted some
relevant references but we hope the reader will have a good start for further study with
the bibliography below.
[Ad] R.A. Adams: Sobolev Spaces, Academic Press, 1975.
[ABCP] M. Agranovsky, C.A. Berenstein, D.C. Chang, and D. Pascuas: Theoremes de
Morera et Pompeiu pour les groupe de Heisenberg, C. R. Acad. Sci. Paris Ser. 1, 315
(1993), 655-658.
[Ai] L.A. Aizenberg: Carleman Formulas in Complex Analysis, Kluwer Academic, 1993.
[Am] E. Amar: On the Corona problem, J. Geom. Anal. 4 (1991), 291-305.
[AC] J. Anderson and 1. Cima: Recovering HP functions from partial boundary data, to
appear in Complex Variables.
[AKS] 1.M. Anderson, D. Khavinson, and H.S. Shapiro: Analytic continuation of
Dirichlet series, Preprint, Royal Institute of Technology (Stockholm), 1993.
[Ao] T. Aoki: Existence and continuation of holomorphic solutions of differential
equations of infinite order, Adv. in Math. 72 (1988), 261-283.
[Ap] P. Appell: Sur les fonctions periodiques de deux variables, J. Math. Pures et Appl.
7 (1891),157-219.
[AM] M.F. Atiyah and I.G. Macdonald: Introduction to Commutative Algebra, Addison-
Wesley, 1969.
[Av1] V. Avanissian: Fonctionnelles analytiques liees aux polynomes orthogonaux clas-
siques, C. R. Acad. Sci. Paris 307 (1988),177-180.
[Av2] V. Avanissian: Quelques applications des fonctionnelles analytiques, Ann. Acad.
Sci. Fenn. 15 (1990), 225-245.
[AG1] V. Avanissian and R. Gay: Sur une transformation des fonctionnelles analytiques
portables par des convexes compacts de Cd et la convolution d'Hadamard, C. R. Acad.
Sci. Paris 279 (1974),133-136.
[AG2] V. Avanissian and R. Gay: Sur les fonctions entieres arithmetiques de type expo-
nentiel et Ie quotient d'exponentielle-polynomes de plusieurs variables, C. R. Acad.
Sci. Paris 279 (1974),161-164.
[AG3] V. Avanissian and R. Gay: Sur une transformation des fonctionelles analytiques et
ses applications aux fonctions entieres de plusieurs variables, Bull. Soc. Math. France
103 (1975), 341-384.
[AS1] V. Avanissian and R. Supper: Fonctionnelles analytiques et sommes de series de
puissances a coefficients produits de polynomes orthogonaux, C. R. Acad. Sci. Paris
312 (1991), 73-76.
471
472 References
[AS2] V. Avanissian and R. Supper: Sur les fonctions entieres arithmetiques au sens
d'Abel de plusieurs variables, C. R. Acad. Sci. Paris 312 (1991), 73-76, 781-784.
[Ba] A. Baker: Transcendental Number Theory, Cambridge University Press, 1979.
[BeC] R. Bellman and K. Cooke, Differential Difference Equations, Academic Press,
1963.
[BD1] C.A. Berenstein and M.A. Dostal: A lower estimate for exponential sums, Bull.
Amer. Math. Soc. 80 (1974), 687-691.
[BD2] C.A. Berenstein and M.A. Dostal: The Ritt theorem in several variables, Ark. Mat.
12 (1974), 267-280.
[BD3] c.A. Berenstein and M.A. Dostal: Analytically Uniform Spaces and Their Appli-
cations to Convolution Equations, Springer Lecture Notes in Mathematics, No. 256,
1972.
[BG] C.A. Berenstein and R. Gay: Complex Variables: An Introduction, Springer-Verlag,
1991.
[BG1] C.A. Berenstein and R. Gay: A local version of 2-circles theorem, Israel J. Math.
55 (1986), 267 - 288.
[BG2] C.A. Berenstein and R. Gay: Le probleme de Pompeiu local, J. Analyse Math. 52
(1989), 133-166.
[BGV] C.A. Berenstein, R. Gay, and A. Vidras: Division theorems in spaces of entire
functions with growth conditions and their applications to PDE of infinite order, to
appear in Publ. Res. Inst. Math. Sci. (Kyoto Univ.).
[BGY] C.A. Berenstein, R. Gay, and A. Yger: Inversion of the local Pompeiu transform,
J. Analyse Math. 54 (1990), 259-287.
[BL] C.A. Berenstein and B.Q. Li: Interpolation varieties for spaces of meromorphic
functions, to appear in J. Geom. Anal.
[BLV] C.A. Berenstein, B.Q. Li, and A. Vidras: Geometric characterization of interpo-
lation varieties for the (FN)-spaces A~ of entire functions, Can. J. Math. 47 (1995),
28-43.
[BSe] C.A. Berenstein and A. Sebbar: On a functional equation of A. Hurwitz, Adv. in
Math. 110 (1995), 47-87.
[BStl] C.A. Berenstein and D.C. Struppa: A Fabry-Ehrenpreis-Kawai gap theorem, Publ.
Res. Inst. Math. Sci. (Kyoto Univ.) 23 (1987), 565-578.
[BSt2] C.A. Berenstein and D.C. Struppa: Complex analysis and convolution equations, in
Encyclopaedia of Mathematical Sciences, Springer-Verlag, vol. 54 (1993), pp. 1-108.
[BSt3] C.A. Berenstein and D.C. Struppa: Dirichlet series and convolution equations,
Publ. Res. Inst. Math. Sci. (Kyoto Univ.) 24 (1988), 783-810.
[BTl] C.A. Berenstein and B.A. Taylor: A new look at interpolation theory for entire
functions of one variable, Adv. in Math. 33 (1979), 109-143.
[BTI] C.A. Berenstein and B.A. Taylor: Mean-periodic functions, Internat. J. Math. Math.
Sci. 3 (1980), 199-236.
[BT3] C.A. Berenstein and B.A. Taylor: Interpolation problems in en with applications
to harmonic analysis, J. Analyse Math. 38 (1980), 128-144.
[BY1] C.A. Berenstein and A. Yger: Le probleme de la deconvolution, J. Funct. Anal.
54 (1983), 113-160.
[BY2] C.A. Berenstein and A. Yger: On Lojasiewicz type inequalities for exponential
polynomials, J. Math. Anal. Appl. 129 (1988), 166-195.
[BZ] C.A. Berenstein and L. Zalcman: The Pompeiu problem in symmetric spaces,
Comment. Math. Helv. 55 (1980), 593-621.
[Bernd] B. Berndtsson: Some remarks on weighted estimates for ii, in Complex Analysis,
K. Diederich (ed.), F. Vieweg & Sohn, 1991,40-43.
[Berns] V. Bernstein: Le(;ons sur les Progres Recents de la Theorie des Series de Dirichlet,
Gauthier-Villars, 1933.
[Beur] A. Beurling: Analytic continuation across a linear boundary, Acta Math. 128
(1972), 153-182.
References 473
[Bez] J.P. Bezivin, Sur les points ou une fonction analytique prend des valeurs entieres,
Ann. Inst. Fourier 40 (1990), 785-809.
[BezGr] J.P. Bezivin and F. Gramain, Solutions entieres d'un systeme d'equations aux
differences, Ann.lnst. Fourier 43 (1993), 791-814.
[Bj] G. Bjorck: Linear partial differential operator and generalized distributions, Ark·fiir
Mat. 6 (1966), 351-407.
[Bo] R.P. Boas, Jr.: Entire Functions, Academic Press, 1954.
[BrK] A.V. Bratischev and J.F. Korobeinik: The multiple interpolation problem in
the space of entire functions of given proximate order, Math. USSR-Izv. 10 (1976),
1049-1074.
[Bra] R.W. Braun: Weighted algebras of entire functions in which each closed ideal
admits two algebraic generators, Michigan Math. 1.34 (1937), 441-450.
[BST] L. Brown, B.M. Schreiber, and B.A. Taylor: Spectral synthesis and the Pompeiu
problem, Ann.lnst. Fourier 23 (1973) 125-154.
[Bre] G.E. Bredon: Sheaf Theory, McGraw-Hill, 1967.
[Bu] R.C. Buck: Integral valued entire functions, Duke Math. 1. 15 (1948), 879-891.
[Cel] U. Cegrell: A generalization of the corona theorem in the unit disk, Math. Z. 203
(1990), 255-261.
[Ce2] U. Cegrell: Another generalization of the Corona theorem in the unit disk, Preprint,
Umea University, 1992.
[Chol] A.M. Chollet: Ensembles de zeros, ensembles pics pour A(D) et AOO(D), in
Complex Approximation, B. Aupetit (ed.), Birkhhiiuser, 1980, 57 -66.
[Choq] G. Choquet: Sur Ie fondement de la theorie fine du potentiel, C. R. Acad. Sci.
Paris 244 (1957), 1606-1609.
[Chou] C.C. Chou: La Transformation de Fourier Complexe et l' Equation de Convolution,
Springer Lecture Notes in Mathematics, No. 325, 1973.
[Del] J. Delsarte: Les fonctions moyenne-periodiques, 1. Math. Pures et Appl. 14 (1935),
403-453.
[De2] J. Delsarte: Lectures on Topics on Mean-Periodic Functions and the Two-Radius
Theorem, Tata Institute of Fundamental Research (Bombay), 1961.
[Dil] D.G. Dickson: Expansions in series of solutions of linear difference-differential
equations and infinite order differential equations with constant coefficients, Mem.
Amer. Math. Soc. 23, (1957).
[Di2] D.G. Dickson: Analytic mean periodic funtions, Trans. Amer. Math. Soc. 110
(1964),361-374.
[Die] P. Dienes: The Taylor Series, Dover, 1957.
[Dr] M.M. Dzhbarshian: Integral Transforms and the Representation of Functions in
Subdomains of C (in Russian), Nauka, 1966.
[Du] P.L. Duren: Theory of HP Spaces, Academic Press, 1970.
[Ee] J. Eealle: Les Fonctions Resurgentes, 3 vols., Publ. Math, d'Orsay, 1981-85.
[Ehl] L. Ehrenpreis: Mean periodic functions I, Amer. 1. Math. 77 (1955), 293-328,
731-733.
[Eh2] L. Ehrenpreis: Theory of infinite derivatives, Amer. 1. Math. 81 (1959), 799-845.
[Eh3] L. Ehrenpreis: Solutions of some problems of division IV, Amer. 1. Math. 82
(1960), 522-588.
[Eh4] L. Ehrenpreis: Solutions of some problems of division V, Amer. 1. Math. 84 (1962),
324-348.
[Eh5] L. Ehrenpreis: Fourier Analysis in Several Complex Variables, Wiley-Interscience,
1970.
[EhM] L. Ehrenpreis and P. Malliavin: Invertible operators and interpolation in AU
spaces, 1. Math. Pures et Appl. 53 (1974), 165-182.
[Er] A. Erdelyi et al.: Higher Transcendental Functions, McGraw-Hill, 1953.
[Eu] L. Euler: De integratione aequationum differentialium altorum graduum, Misc. Berol.
7 (1743), 193-242.
474 References
[Fe] M. Fekete: Ober die Verteilung der Wurzeln bei gewissen algebraischen Gleichungen
mit ganzzahligen Koeffizienten, Math. Z. 17 (1923), 228-249.
[Fi] S.D. Fischer: Function Theory on Planar Domains, Wiley-Interscience, 1983.
[FS] J.E. Fornaess and N. Sibony: On LP estimates for a, Proc. Sympos. Pure Math.,
vol. 52, part 3, Amer. Math. Soc., 1991, 129-163.
[Fo] J. Fourier: Theorie Analytique de la Chaleur, English translation, Dover, 1955.
[Gam] T.W. Gamelin: Wolff's proof of the Corona theorem, Israel J. Math. 37 (1980),
113-119.
[Ga] J.B. Garnett: Bounded Analytic Functions, Academic Press, 1981.
[Ge] A.O. Gelfond: The Calculus ofFinite Differences, GITTL (in Russian), 1952; Dunod
(in French), 1963.
[Gen] T.G. Genchev: Entire functions of exponential type with polynomial growth in Rft,
J. Math. Anal. Appl. 60 (1977), 103-114.
[Go] E. Gourin: On irreducible polynomials in several variables which become reducible
when the variables are replaced by powers of themselves, Trans. Amer. Math. Soc. 31
(1929),485-501.
[GrR] I.S. Gradshteyn and I.M. Ryzhik: Table ofIntegrals, Series and Products, Academic
Press, 1980.
[Ora] F. Gramain: Solutions indefiniment derivables et solutions presque-periodiques
d'une equation de convolution, Bull. Soc. Math. France 104 (1976), 401-408.
[GR] A.F. Grishin and A.M. Russakovskii: Free interpolation by entire functions, Teor.
Funktsil Funktsional. Anal. i Prilozhen. 44 (1985), 32-42 (in Russian).
[Grudl] O.v. Grudzinski: Einige elementare Ungleichungen fiir Exponentielpolynome,
Math. Ann. 221 (1976),9-34.
[Grud2] O.v. Grudzinski: Slowly decreasing entire functions and convolution equations
(Partial Differential Equations), Banach Center Publications 10 (1983), 169-184.
[Grum] L. Gruman: Proprietes arithmetiques des fonctions entieres, Bull. Soc. Math.
France 108 (1980), 421-440.
[GrL] L. Gruman and P. Lelong: Entire Functions o/Several Complex Variables, Springer-
Verlag, 1986.
[Gul] D.I. Gurevich: Generalized bases in some rings of holomorphic functions, Math.
USSR-Izv. 6 (1973), 564-578.
[Gu2] D.I. Gurevich: Counterexamples to a problem of L. Schwartz, Funct. Anal. Appl.
9 (1975), 116-120.
[Hajj] A. Hajj Diab: These, University, Bordeaux, 1976.
[Har] G.H. Hardy: Divergent Series, Oxford University Press, 1949.
[HW] G.H. Hardy and E.M. Wright: The Theory of Numbers, Oxford University Press,
1960.
[He] S. Helgason: Groups and Geometric Analysis, Academic Press, 1984.
[HRS] C.W. Henson, L.A. Rubel, and M.F. Singer: Algebraic properties of general
exponential polynomials, Complex Variables 13 (1989), 1-20.
[Heyl] R. Heyman: Interpolation of infinite order entire functions, Revista Matem.
Iberoamericana 10 (1994), 581-626
[Hey2] R. Heyman: Interpolation in £'(R), to appear in Complex Variables.
[Hof] K. Hoffman: Banach Spaces of Analytic Functions, Prentice-Hall, 1962.
[Hoi] L. Hormander: The Analysis of Linear Partial Differential Operators, Vol. 1,
Springer-Verlag, 1983.
[H02] L. Hormander: The Analysis of Linear Partial Differential Operators, Vol. 2,
Springer-Verlag, 1983.
[H03] L. Hormander: An Introduction to Complex Analysis in Several Variables, van
Nostrand, 1966.
[H04] L. Hormander: Generators for some rings of analytic functions, Bull. Amer. Math.
Soc. 73 (1967), 943-949.
[Hor] J. Horvath: Topological Vector Spaces and Distributions, Addison-Wesley, 1966.
References 475
[Hu1] A. Hurwitz: Sur l'integrale finie d'une fonction entiere, Acta Math. 20 (1897),
285-312.
[Hu2] A. Hurwitz: Mathematische Werke, 2 vols., Birkhauser, 1932-33.
[Ja] H. Jarchow: Locally Convex Spaces, B.G. Teubner, 1981.
a
[Jo] P. Jones: L OO estimates for the in a half-plane, Acta Math. 150 (1983), 137-314.
[Kahl] J.P. Kahane: Sur les fonctions moyennes-periodiques bomees, Ann.lnst. Fourier
7 (1957), 293-314.
[Kah2] J.P. Kahane: Lectures on Mean Periodic Functions, Tata Institute of Fundamental
Research (Bombay), 1957.
[Kan] A. Kaneko: Introduction to Hyper/unctions, Kluwer Academic, 1988.
[Kat] Y. Katznelson: An Introduction to Harmonic Analysis, Dover, 1968.
[Kawl] T. Kawai: On the Fourier hyperfunctions and its applications to partial differential
equations with constant coefficients, 1. Fac. Sci. Univ. Tokyo, 17 (1970), 467-517.
[Kaw2] T. Kawai: The Fabry-Ehrenpreis gap theorem and linear differential equations
of infinite order, Amer. 1. Math. 109 (1987), 57-64.
[KTl] lJ. Kelleher and B.A. Taylor: An application of the Corona theorem to some rings
of entire functions, Bull. Amer. Math. Soc. 73 (1967), 246-249.
[KT2] lJ. Kelleher and B.A. Taylor: Finitely generated ideals in rings of analytic func-
tions, Math. Ann. 193 (1971), 225-237.
[KT3] J.J. Kelleher and B.A. Taylor: Closed ideals in locally convex algebras of analytic
functions, 1. Reine Angew. Math. 255 (1972), 190-209.
[Ki] C.O. Kiselman: Prolongement des solutions d'une equation aux derivees partielles
a coefficients constants, Bull. Soc. Math. France 97 (1969), 329-356.
[Kom] H. Komatsu: On the index of ordinary differential operators, 1. Fac. Sci. Univ.
Tokyo, 18 (1971), 379-388.
[Ko1] P. Koosis: Introduction to HP Spaces, Cambridge University Press, 1980.
[K02] P. Koosis: The Logarithmic Integral, Cambridge University Press, vol. 1, 1988,
and vol. 2, 1992.
[Kr] I.F. Krasichov: Closed ideals in locally convex algebras of entire functions, Math.
USSR-Izv. 1 (1967), 35-55.
[Leb] N.N. Lebedev: Special Functions and their Applications, Dover, 1972.
[Leo] A.F. Leont'ev: The representation of functions by generalized Dirichlet series,
Russian Math. Surveys 24 (1969), 101-178.
[Lev] B.J. Levin: Distribution of Zeros of Entire Functions, Amer. Math. Soc., 1964.
[Levs] N. Levinson: Gap and Density Theorems, Amer. Math. Soc., 1940.
[Li] B.Q. Li: Ph.D. dissertation, University of Maryland, College Park, 1993.
[Lin] E. Lindelof: Le Calcul des Residus et ses Applications a la Theorie des Fdnctions,
Gauthier-Villars, 1905.
[Mal] B. Malgrange: Equations Dijferentielles a Coefficients Polynomiaux, Birkhauser,
1991.
[Mark] A.I. Markushevich: Theory o/Functions of a Complex Variable, Chelsea, 1977.
[Mart] A. Martineau: Oeuvres de Andre Martineau, Centre National de la Recherche
Scientifique (Paris), 1977.
[MR] J. Martinet and J.P. Ramis: Theorie de Galois differentielle et resommation, in
Computer Algebra and Dijferential Equations, E. Toumier (ed.), Academic Press, 1988,
117-214.
[MeTl] R. Meise and B.A. Taylor: Splitting of closed ideals in (DFN)-algebras of entire
functions and the property (DN), Trans. Amer. Math. Soc. 302 (1987), 341-370.
[MeT2] R. Meise and B.A. Taylor: Each nonzero convolution operator on the entire
functions admits a continuous linear right inverse, Math. Z. 197 (1988), 139-152.
[MeTV] R. Meise, B.A. Taylor, and D. Vogt: Characterization of convolution operators
on spaces of Coo functions which admit a continuous linear right inverse, Math. Ann.
279 (1987),141-155.
476 References
[Mer] A. Meril: Analytic functionals with unbounded carriers and mean periodic func-
tions, Trans. Amer. Math. Soc. 278 (1983), 115-136.
[MerS] A. Meril and D.C. Struppa: Equivalence of Cauchy problems for entire and
exponential type functions, Bull. London Math. Soc. 17 (1985), 469-473.
[MerY] A. Meril and A. Yger: Problemes de Cauchy globaux, Bull. Soc. Math. France
120 (1992), 87-11l.
[Mey] Y. Meyer: Algebraic Numbers and Harmonic Analysis, North-Holland, 1972.
[MiT] L. M. Milne-Thomson: The Calculus of Finite Differences, Macmillan, 1933.
[Mol] M. Morimoto: Analytic functionals with noncompact carrier, Tokyo. J. Math. 1
(1978),77-103.
[Mo2] M. Morimoto: A generalization of the Fourier- Borel transformation for the
analytic functionals with non compact carrier. Tokyo. J. Math. 2 (1979), 301-322.
[Mo3] M. Morimoto: An Introduction to Sato's Hyper/unctions, Amer. Math. Soc., 1994.
[MoYo1] M. Morimoto and K. Yoshino: A uniqueness theorem for holomorphic functions
of exponential type, Hokkaido Math. J. 7 (1978), 259-270.
[MoYo2] M. Morimoto and K. Yoshino: Some examples of analytic functionals with
carrier at infinity, Proc. Japan Acad. Ser. A. Math. Sci. 56 (1980), 357-36l.
[Mor] V.A. Morosov: Regularization Methodsfor Ill-Posed Problems, CRC Press, 1992.
[Na] A. Naftalevich: On a differential-difference equation, Michigan Math. J. 22 (1975),
205-223.
[NS] DJ. Newman and H.S. Shapiro: Fischer spaces of entire functions, Proc. Symp.
Pure Math. 11 (1968), 360-369.
[Nil] N.K. Nikol'skii: Closed ideals in certain algebras of entire functions, Siberian
Math. J. 9 (1968), 160-162.
[Ni2] N.K. Nikol'skii: Invariant subspaces in the theory of operators and theory of
functions, J. Soviet Math. 5 (1976), 129-249.
[Ni3] N.K. Nikol'skii: Treatise on the Shift Operator, Springer-Verlag, 1985.
[01] F.W.J. Olver: Asymptotics and Special Functions, Academic Press, 1974.
[Os] A. Ostrowski: Solution of Equations and Systems of Equations, Academic Press,
1966.
[Pal V.P. Palamadov: Linear Differential Operators with Constant Coefficients, Springer-
Verlag, 1970.
[PW] R.E.A.C. Paley and N. Wiener: Fourier Transforms in the Complex Domain, Amer.
Math. Soc., 1934.
[Pat] D.J. Patil: Representation of HP functions, Bull. Amer. Math. Soc. 78 (1972),
617-620.
[Pi] C. Pisot: Sur les fonctions arithmetiques analytiques acroissance exponentielle, C. R.
Acad. Sci. Paris 222, (1946), 988-990.
[Pol] G. P61ya: Geometrisches tiber die Verteilung der nullstellen ganzertranzendenter
Funktionnen, Munch. Sitwngsbl. 50 (1920), 285-290.
[PS] G. P61ya and G. Szego, Problems and Theorems in Analysis, 2 vols., Springer-Verlag,
1972.
[Porn] C. Pommerenke: Univalent Functions, Vandenhoeck and Ruprecht, 1975.
[Pool] E.G.C. Poole: Introduction to the Theory of Linear Differential Equations, Dover,
1960.
[Ra] H. Rademacher: Topics in Analytic Number Theory, Springer-Verlag, 1973.
[Ram] J.P. Ramis: Devissage Gevrey, Asterisque 59/60 (1978), 173-204.
[Ri1] J.F. Ritt: A factorization theory for functions ,£7=1 aje rtjZ , Trans. Amer. Math. Soc.
29 (1927), 584-596.
[Ri2] J.F. Ritt: Algebraic combinations of exponentials, Trans. Amer. Math. Soc. 31
(1929),654-679.
[Ri3] J.F. Ritt: On the zeros of exponential polynomials, Trans. Amer. Math. Soc. 31
(1929),680-686.
[Ro] L.I. Ronkin: Functions of Completely Regular Growth, Kluwer Academic, 1992.
References 477
[RuT] L.A Rubel and B.A. Taylor: A Fourier series method for meromorphic and entire
functions, Bull. Soc. Math. France 96 (1968), 53-96.
[Rud1] W. Rudin: Function Theory in Polydiscs, W.A Benjamin, 1969.
[Rud2] W. Rudin: Real and Complex Analysis, McGraw-Hill, 1966.
[Schae] H.H. Schaefer: Topological Vectors spaces, Springer-Verlag, 1970.
[Schap] P. Schapira: Theorie des Hyperfonctions, Springer-Lecture Notes in Mathematics,
No. 126. 1970.
[Schi] A Schinzel: Selected Topics on Polynomials, University of Michigan Press, 1982.
[Schm1] J. Schmetz: Equations differentiels et espaces d' hyperfonctions, Univ. de Liege,
1980.
[Schm2] J. Schmetz: Hyperfonctions et microfonctions d'une variable, Univ. de Liege,
1980.
[Schw1] L. Schwartz: Theorie generale des fonctions moyenne periodiques,Ann. ofMath.
48 (1947), 857-929.
[Schw2] L. Schwartz: Theorie des Distributions, Hermann, 1966.
[Se] A Sebbar: Prolongement des Solutions Holomorphes de certains operateurs
Differentiels d'Ordre Infini a Coefficients Constants, Springer Lecture Notes in
Mathematics, No. 822, 1980, 199-220.
[Sha] H.S. Shapiro: The expansion of mean periodic functions in series of exponentials,
Comm. Pure Appl. Math. 11 (1958), 1-21.
[Shi] A. Shields: On quotients of exponential polynomials, Comm. Pure Appl. Math. 16
(1963),27-31.
[Si] Y. Sibuya: Linear Differential Equations in the Complex Domain: Problems of
Analytic Continuation, Amer. Math. Soc., 1990.
[SkI] H. Skoda: Sous-ensembles analytiques d'ordre fini ou infini dans en, Bull. Soc.
Math. France 100 (1972), 353-408.
[Sk2] H. Skoda: Applications des techniques L 2 It la theorie des ideaux d'une algebre
de fonctions holomorphes avec poids, Ann. Sci. Ecole Norm. Sup. Paris 5 (1972),
545-579.
[Sq1] W.A Squires: Necessary conditions for universal interpolation in e'eR), Can. J.
Math. 3 (1981), 1356-1384.
[Sq2] W.A Squires: Geometric conditions for universal interpolation in e'eR), Trans.
Amer. Math. Soc. 280 (1983), 401-413.
[SuI] R. Supper: Exemples d'applications de transformation G des fonctionnelles analy-
tiques, Complex Variables, 18 (1992) 201-212.
[Su2] R. Supper: These, University of Strasbourg, 1993.
[Tal] B.A Taylor: Some locally convex spaces of entire functions, Proc. Sympos. Pure
Math. 11 (1966),431-466.
[Ta2] B.A Taylor: Linear extension operators for entire functions, Michigan Math. J. 29
(1982), 185-197.
[Ti] R. Tijdeman: On the number of zeros of general exponential polynomials, Indag.
Math. 33 (1971),1-7.
[Tr] M. Tretkoff: Picard-Fuchs differential equations for the quadratic periods ofAbelian
Integrals of the first kind, Springer Lecture Notes in Mathematics, No. 1277 (1987),
263-289.
[Tu] P. Tunm: On a New Method of Analysis and its Applications, Wiley-Interscience,
1984.
[Va] K. Viiiiniinen: On the arithmetic properties of certain values of the exponential
function, Studia Sci. Math. Hungar. 11 (1976), 399-405.
[Val] G. Valiron: Integral Functions, Chelsea, 1949.
[V] AJ. Van der Porten: On the number of zeros of functions, Enseign. Math. 23 (1977),
19-38.
[VT] AJ. Van der Porten and R. Tijdeman: On common zeros of exponential-
polynomials, Enseign. Math. 21 (1974), 58-67.
478 References
We follow the standard notation from [BG]. Here the reader will find the extra
symbols introduced and the first page where they appear.
A(Q) 39 E 330
A(V), Ap(V) 119 Exp({O}) 419
Ap(Q), Ap,B(Q) 109
A2(C) 74 Ir 2
Ap,oo(V) 120 f* 13
Ap(Q) 151 :FI 51
Ap(C) 170 3'(T) 58
Ap,o(C) 178 3'p(T) 326
AEP 219
G(T) 354
B(Q) 36
b(f) 36 ~P, HP 2
B 63 Jt'o(Q C ) 58
B(f) = B1(f) 300
HK 64
Bp(f) 310 65
hI, hj
Bn 408
hl,p 316
C(T) 58 Hp;K 326
cn(T) 99 H(C, k') 336
cVp(K) 327
C(S) 338 Iloc(fJ, ... ,1m) 122
Li 102 K(T) 54
dO 203 K(T) 58
D(f) 299 Kz 299
Dp 309 KPz 308
<SIR 237 K*P
z 326
Pr 119
R(L, r, s, k', s'),
Wp(Q) 3
R(Lr, k', s') 334
R(C\L; k', s'), Z(f) 263
R(C; k', s') 335 Z(f) 278
Index
481
482 Index