An Interpolation Problem for Coefficients of H Functions Author(s): John J. F. Fournier Source: Proceedings of the American Mathematical Society, Vol.

42, No. 2 (Feb., 1974), pp. 402-408 Published by: American Mathematical Society Stable URL: http://www.jstor.org/stable/2039516 Accessed: 22/11/2010 06:31
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=ams. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

American Mathematical Society is collaborating with JSTOR to digitize, preserve and extend access to Proceedings of the American Mathematical Society.

http://www.jstor.org

OF THE PROCEEDINGS SOCIETY MATHEMATICAL AMERICAN Volume 42, Number 2, February 1974

AN INTERPOLATION PROBLEM FOR COEFFICIENTS OF HO FUNCTIONS'

JOHN J. F. FOURNIER HI ABSTRACT. denotes the space of all boundedfunctionsg on the unit circle whose Fouriercoefficientsg(n) are zero for all negative n. It is known that, if {nfk},O is a sequenceof nonnegativeinte12 gers with nk+l > (1 +5)nk for all k, and if Z0-0 IVk < OO, then there is a functiong in HI with A(nk)=vk for all k. Previous proofs of this

fact have not indicatedhow to constructsuch HI functions. This paper contains a simple, direct construction of such functions. The constructiondependson propertiesof some polynomialssimilar to those introducedby Shapiro and Rudin. There is also a connection with a type of Riesz product studied by Salem and Zygmund.

We shall present the construction in ?1 and discuss its relation to other results in ?2. First we recall some notation. For p < xo, LI denotes the space of all measurable functions on the unit circle whose p'th power is integrable; as usual, functions are identified if they agree almost everywhere. L' denotes the space of all essentially Ioo bounded measurable functions on the unit circle; for g in L', 1ig denotes the essential supremum of Igl.The Fourier coefficients of an LI function g are defined by
g(n)
=

g(eiO)e-nodO

(n

0, 41, 42**

Finally, H' is taken to be the space of all g in L' which have g(n) =0 for all n<O. It can be shown [4, p. 39] or [5, Theorem 3.12, p. 88] that this version of H' is isomorphic to the space of all bounded analytic functions in the open unit disc, but this fact will not be used here. 1. The construction depends on the fact that, for complex numbers a, b, and v,
(1) la + vbI2 +

lb

ea12 =

(1 +

IV12)(la12

lb12).

Received by the editors March 13, 1972.

AMS (MOS) subject class fications (1970). Primary 30A10, 30A78, 42A44; Secondary

26A82, 30A80, 42A04, 42A16, 42A80. Key wordsandphrases. Interpolationby Fourier coefficients, HI function, RudinShapiro polynomials, Hadamardset, A(2) set, Riesz product. 1 Research partially supported by National Research Council of Canada operating grant numberA-4822.
?
American Mathematical Society 1974

402

COEFFICIENTS OF HO1 FUNCTIONS

403

This identity plays the same role here as the parallelogram law does in the method of Shapiro and Rudin [10, p. 35] and [7, pp. 855-856].
THEOREM. Let {nk}k'O be a sequence of nonnegative integers so that, for some 6> 0, nk+1 >(I +6)nk f or al k. Let O IVkI2<X. Then there is a function g in H' with g(nk)=vk for all k.

PROOF. We begin with the basic case, when 6>1. Let z=etO. Define trigonometric polynomials gk and hk by:
go(z) = vozlo, ho(z) 1,

and, for k>O,

gk(Z)
=

gk-1(Z)

VkZ fkhk(Z)

and
hk(Z) = hk-1(Z) jk Zzl gk(Z).

We claim that the polynomials (a) (b)

(c)

gk

have the following properties:

< 2tgkIIO

11 (1
3=0

+ ivj12).

gk E H. gk(ni) = vj

for all j < k.

Suppose for a moment that these assertions hold. Because .3= Iv,12<x, the product H,S=o(1 +Ivj12) converges and the sequence {1lgkK1}o is bounded. Therefore the sequence {gk}k -O has a weak star limit point, g say,
in L'. Since g(n) is a limit point of {g
(nf)}'0

we have, by (b), that g is in

H' and, by (c), that g(nj)=vj for allj. Thus g has the desired properties. To verify assertion (a) we note that, by (1),
|gk(z)|' + Ihk(Z) 12 = (1 + IVk 12) (Igk_.1(Z)12 +

Ihk-1 (Z) I 2).

1+ As 1go(z)12+1ho(z)12= Iv012,we have by induction on k that

k Igk(Z)12 + Ihk(Z)12
=

IJ
j=0

(1

Ivjj2),

and (a) follows. Now let Gk denote the support of kk and We shall prove by induction on k that
(2)
Ck

Hk

denote the support of A7.

[0, nk]

and (3)
Hk [nk,

01

404

J. J. F. FOURNIER

[February

These assertions clearly hold when k=O. Supposing that (2) and (3) hold for some index k, considei the formula
gk+l(Z) = gk(Z) +
Vk+?Z k+lkh(Z)

gk(Z)

+ fk+1(Z),

say.
[nk+l-nk
nk+1].

By (3), the support offk+1 is contained in the interval Hence (4)

Gk+1

G Gk U [fk+1-

nk, nk+41c [0, nk+1],

as required.

Similarly Hk+1c [-nk+1, 0]. Thus assertions (2) and (3) hold for all k. It follows from (2) that gk E H'; so (b) holds. Next, because nk+1>2nk, the intervals [0, nk] and [nk+1-nk,nk+1] are disjoint. Therefore kkand fk+1 have disjoint supports and for all n<n Similarly hk?l(n)=hk(n) for all n_-nk. It follows that h (0)=I for all k. By (2), kk(nk+1)=O; hence
ki+(n)
=

(5)

g (n)

(6)

gk+l(nk+l) -fk+l(nk+l)

= Vk-I hl(O)

= Vk?l-

Combining (5) and (6) we see that kk(nj)=vj for all j_ k; so (c) holds. This completes the proof for the case when *S>1. 1 Before taking up the case when *S< we mention two more facts about the polynomials gk. First, it follows from (5) that the weak star limit g discussed above is unique and that gk iS the partial sum of order nk of the Fourier series of g. Indeed, since the subsequence of partial sums fgk }'=k is uniformly bounded, it follows, without appeal to weak star compactness, that the corresponding trigonometric series must be the Fourier series of a bounded function [12, vol. 1, p. 148]. Second, it is clear from (4) that Gkc {no}U [n1-nD, nl]U**. u nkj- Therefore, the support of k is contained in {n0}U [nk-nk_1
[nl-no, nl]u..U
t[nk-nk-1,

nJU-9

- -

When 6 < I, we must split {nk} into a finite number of subsequences, so that the above method works for each subsequence, and so that the functions g corresponding to different subsequences have disjointly supported transforms. The procedure is like the one for Riesz products [5, pp. 108-109] or [12, vol. II, p. 131]. Let q be the smallest integer such that o(I + )"-I>1. Split {n,} and {Vk} into q subsequences in each of which the index k runs through an arithmetic progression with period q. Let {nkJ} , be such a subsequence. Since ki71I-k =q, we have 1(1 +
n

> (1 +

nk

(70?+

n>2k.

2nki

19741

COEFFICIENTS H' FUNCTIONS OF

405

by the choiceof q. So the methodusedabove appliedto this subsequence yieldsa functiong in H' withg(nk )=Vki for all i. Moreover is supported g It is easy to verify by {nklU [Ink, l-nk0 fnkIU*-)..) n[ki-nk, nlki-]U
that
nk-nki __>nki-1,

so that the only term of the sequence

{fnk}

for all the terms , nk1] is nk . Hencek(nk)=O lying in the interval[nki-nk1 of {nk} exceptthe nki. Thus we can solvethe interpolation for problem each and add the solutions to obtain an H' function of the q subsequences which solves the interpolationproblemfor {nk} and {vk}. This completes the proof. We can sharpenthe conclusionof the Theoremin two standardways.
First, the function g in H' can be chosen so that JJgJJ._(qe)112liV112

where IIvII2=(2=o IvkI2)1/2 and q is the integer defined above. To see this we begin with the basic case when 6? I and q = l. There is nothing to prove if l1vI12=O; let 11vI12>O. g be the weak star limit point of so Let the polynomials By (a) gk.
00

llgll0

171
k=O

(1

IVkl2)1/2 <

exp(IIVII2/2).

12. Given {Vk}, define a new sequence {vk} by Vk=VkIIIV Applying our methodto the sequence{vk} we obtaina functiong' in H' withg'(n )=vv for all k and lIg'IK,<e1'2. g=IIvII2g'. Then g(nk)=Vk for all k and Let as required.When 6<1, split {nk} into q subsequences jigL11<e112IIvI2, and apply this trick to each subsequence.Adding the resulting H' for all k. It follows functions gives a functiong in H' with g(nf)=vk from the Cauchy-Schwarz inequalitythat ligilo <(qe) 1211IV12. H" Second, we can arrangethat the interpolating functiong be continuous. To do this choose a functionf in L1 whose Fouriercoefficients tend to 0 so slowly that E =o IVklI(nk)12< C; given that 11v112< this is cc, possible by [5, Theorem4.1, p. 22] or [12, vol. I, Theorem 1.5, p. 183]. Apply the Theoremto the sequence{Vk/f(nk)})ko obtaining a functiong' in H" with g(nk)=Vklf (nlk) for all k. Then g=f * g', the convolutionof f and g', is a continuous H" function with g(nk)=vk for all k.
2. Let us say that a set of integers {nlk}k'=O has the LX interpolation propertyif, for every squaresummablesequence{Vk1=o, there is a function g in L' with g(nk)=Vk for all k. Similarly,we shall say that a set of if nonnegative integershas the H' interpolation property the set has the L' interpolationpropertyand the interpolatingfunction g can always be chosen to be in H'. Finally, we shall call a set of nonnegativeintegers {nk}k=oa Hadamard set if kk+1> (1 +a)nk for some (>0 for all k. We have just shown that every Hadamardset has the H' interpolationproperty. There are other ways to prove this fact, which do not give as much

406

J. J. F. FOURNIER

[February

information about the interpolating function g as our method does, but which allow a weakening of the assumption on the sequence {fnk}. First, one can use a method of Paley [6, Lemma 2, pp. 124-126] to show that if a set of a set of nonnegative integers has the LX interpolation property then it has the H' interpolation property. It is well known [8, p. 225] that a set of integers {nk} has the LX interpolation property if and only if it is a A(2) set, that is if and only if every LI function whose transform is supported by {nk} is actually in L2. Thus, in order that a set of nonnegative integers have the H' interpolation property it is necessary and sufficient that this set be a A(2) set. In fact, S. A. Vinogradov has shown that one can simultaneously solve the Rudin-Carleson interpolation problem and the above interpolation problem [11]. Unfortunately no arithmetic characterization of A(2) sets is known. It is known, however, that any union of finitely many Hadamard sets is a A(2) set and that some A(2) sets are not finite unions of Hadamard sets [8, p. 210]. Thus, in ?1, the hypothesis on {fnk} is stronger than necessary. But the proof given in ?1 is more elementary than those discussed above. The first proof that every Hadamard set has the LX interpolation property was indirect [1, Satz I, p. 212]. Subsequently, Salem and Zygmund used Riesz products to give a simple construction of LX interpolating functions [9]. We want to compare their construction with

the one given here. For simplicitysuppose that nk+l_3nk

00
I1kl2<

for all k; let

cc. Consider the product

00 00

11 (1
k=O

+ VkZflk -

vZ-1k)

1I
k= 0

[1 + 2i Im(vkz k)].

Expanding this product formally yields a trigonometric series which can

be shown to be the Fourierseries of a functionf in LX with f

(nk) =vk

for all k and lf12 <J_J'=o (1+41vk12), [9] or [12, vol. I, p. 211]. The support off is contained in the set F of all integers n having the form where kl<k2<-* *inkl, km. The function f is in n=?nk+?nk.-1

H' only if l1v112=0.

On the other hand, it was shown in [3, Remark 8] that, under the same hypotheses on {nk} and {Vk}, there is a function g in H' with k(nk)= Vkfor all k and with the support of g contained in the set G of all integers n -. *f -nk2+nklk having the form n=nkm.-nk_ where the signs 1+nk-2 m is odd, and kl<k2< alternate, <km. The proof in [3] does not indicate how to construct such a function g. Since the set G is contained in the set F associated with the Riesz product f it is natural to consider the series itWeGf (n)z" and attempt to show that this is the Fourier series

1974]

COEFFICIENTSOF H*

FUNCTIONS

407

of a boundedfunction. In fact it is the Fourierseries of the function g constructed ?1. in It is not clear whetherthis relation betweeng and f can be used to prove that g is bounded. Instead we have modifieda method due independentlyto Shapiro [10, p. 35] and Rudin [7]. The idea of setting up two sequencesof trigonometricpolynomialswith prescribedcoefficients is due to them. Whatis new hereis the use of the identity(1), ratherthan the parallelogram law, to estimate the LX norms of the polynomials; this changeallows us to generatepolynomialswhose nonzerocoefficients do not all have the same absolutevalue. The methodused here seems to be dual to the Hilbertspacemethodof [3, Lemma 2], but has the advantageof producingthe interpolating function g explicitly. Otherwisethe two methods are equivalentin the sensethatanythingwhichcan be provedby one of the methodscan also be
proved by the other method. For instance, if nk+1>2nk for all k, and VkO0

for all k, then the functiong constructed ?1 has the propertythat the in supportof g is exactlythe set G definedabove, and in any case g is carried by G; it was shown in [3, Remark8] that there must be an interpolating function whose transformis carriedby G. Because of this equivalence many of the resultsin [3] can now be provedin two ways. In particular we now have two proofs of Theorem 1 of [3] but we still do not know whetherthe assumptionthat I is an exponentialgap systemimplies that 0 = 1 of ,u(mk)12< Xc for all ,uin Ml; see [3, ?2]for definitions theseterms. For furtherapplicationsof the method of this paper to problemsin severalvariables,see [2].
REFERENCES 1. S. Banach, Ubereine Eigenschaftder lacundren trigonometrischen Reihen, Studia Math. 2 (1930), 207-220. 2. J. Fournier, Extensionsof a Fouriermultipliertheoremof Paley. II, (to appear). 3. , Fouriercoefficientsafter gaps, J. Math. Anal. Appl. 42 (1973), 255-270. 4. K. Hoffman, Banachspaces of analyticfunctions, Prentice-HallSeries in Modern Analysis, Prentice-Hall,Englewood Cliffs, N.J., 1962. MR 24 #A2844. 5. Y. Katznelson, An introductionto harmonicanalysis, Wiley, New York, 1968. MR 40 #1734. 6. R. E. A. C. Paley, A note onpowerseries, J. London Math. Soc. 7 (1932), 122-130. 7. W. Rudin, Some theorems on Fourier coefficients,Proc. Amer. Math. Soc. 10 (1959), 855-859. MR 22 #6979. 8. , Trigonometric series withgaps, J. Math. Mech. 9 (1960), 203-227. MR 22 #6972. 9. R. Salem and A. Zygmund, On a theoremof Banach,Proc. Nat. Acad. Sci. U.S.A. 33 (1947), 293-295. MR 9, 88. 10. H. S. Shapiro, Extremalproblemsfor polynomialsandpower series, Thesis for S.M. degree, MassachusettsInstitute of Technology, 1951.

408

J. J. F. FOURNIER

11. S. A. Vinogradov, The Banach-Rudin-Carleson interpolation theoremand norms of embeddingoperatorsfor certain classes of analyticfunctions, Zap. Naucn. Sem. Leningrad, Otdel. Mat. Inst. Steklov. (LOMI) 19 (1970), 6-54=Sem. Math. V. A. Steklov Math. Inst. Leningrad19 (1970), 1-28. MR 45 #4137. 12. A. Zygmund, Trigonometric series, 2nd rev. ed., Vols. I, II, CambridgeUniv. Press, Cambridge,1959. MR 21 #6498.
DEPARTMENT OF MATHEMATICS, UNIVERSITY OF BRITISH COLUMBIA, VANCOUVER 8, BRITISH COLUMBIA, CANADA