doi: 10.2969/jmsj/06120551
A generalization of Miyachi’s theorem
By Radouan DAHER, Takeshi KAWAZOEand Hatem MEJJAOLI (Received June 9, 2008)
Abstract. The classical Hardy theorem on R, which asserts f and the Fourier transform offcannot both be very small, was generalized by Miyachi in terms of L1þL1 and logþ-functions. In this paper we generalize Miyachi’s theorem for Rd and then for other generalized Fourier transforms such as the Che´bli-Trime`che and the Dunkl transforms.
1. Introduction.
Classical Hardy’s theorem [7] asserts the following: let a; b >0 and f a measurable function on R satisfying jfðxÞj Ceax2 and jfðyÞj ^ Ceb2. Then f0 if ab >1=4, f is a constant multiple of eax2 if ab¼1=4, and there are infinitely manyf ifab <1=4. Considerable attention has been devoted to finding generalizations to new contexts for Hardy’s theorem. Especially, M. Cowling and J. Price [5] obtained anLp version of the theorem. As further generalizations, A.
Bonami, B. Demange, P. Jaming [2] extend it in a Beurling form, and Miyachi [9]
obtains anL1þL1 version: Letab¼1=4and f2L1ðRÞsatisfy eax2fðxÞ 2L1ðRÞ þL1ðRÞ
and
Z þ1 1
logþjf^ðÞeb2j
C d <1
for someC >0. Thenf is a constant multiple ofeax2, where L1ðRÞ þL1ðRÞis the set of functions of the formf¼f1þf2,f12L1ðRÞ,f22L1ðRÞ, andlogþx¼ logx ifx >1 andlogþx¼0 ifx1. In this paper we shall generalize Miyachi’s 2000Mathematics Subject Classification. Primary 43A85; Secondary 43A62, 43A32, 44A12, 42A38.
The second auther was supported by Grant-in-Aid for Scientific Research (C) (No. 20540188), Japan Society for the Promotion of Science.
Key Words and Phrases. Hardy’s theorem, Miyachi’s theorem, Radon transform, Dunkl transform, Che´bli-Trime`che transform.
J. Math. Soc. Japan
Vol. 61, No. 2 (2009) pp. 551–558
theorem forRdand furthermore, we shall obtain an analogue in a general measure spaceðX; d!Þ equipped with Fourier, Radon, and dual Radon transforms. As a special case of this general setting, we can deduce Miyachi’s theorem for the Che´bli-Trime`che, the Dunkl transforms, and the Jacobi-Dunkl transform.
Especially, we can obtain alternative proofs in [6] for the Jacobi transform and in [3] for the Dunkl transform.
2. Miyachi’s theorem onRd.
For l¼1;2; , we denote by Fl the Fourier transform on Rl. Miyachi’s theorem is generalized onRd as follows.
THEOREM1. Let ai; bi>0 and aibi¼1=4 for 1id. We put A¼diagða1; a2; ; adÞ and B¼diagðb1; b2; ; bdÞ. If a measurable function on Rd satisfies
ehAx;xifðxÞ 2L1ðRdÞ þL1ðRdÞ ð1Þ
and
Z
Rd
logþjFdfðÞehB;ij
C d <1 ð2Þ
for someC >0, then f is a constant multiple ofehAx;xi.
PROOF. For ðx1; x0Þ ¼ ðx1; x2; ; xdÞ and 0¼ ð2; ; dÞ, we shall con- sider
Gðx1; 0Þ ¼Fd1ðfðx1;ÞÞð0Þ ¼ Z
Rd1
fðx1; x0Þeih0;x0idx0;
wheredx0¼dx2 dxd. Sincef 2L1ðRdÞby (1),Gðx1; 0Þis well-defined and is in L1ðRÞ L1ðRd1Þ. Moreover, since f¼eða1x21þa2x22þþadx2dÞðu1þu2Þ, where u12L1ðRdÞandu22L1ðRdÞby (1), it follows that
jGðx1; 0Þj ea1x21X2
k¼1
Z
Rd1
eða2x22þþadx2dÞjukðx1; x0Þjdx0:
and thus, as a function of x1, ea1x21Gðx1; 0Þ 2L1ðRÞ þL1ðRÞ. We note that F1ðGð; 0ÞÞð1Þ ¼FdfðÞand substitute it in (2). Then Fubini’s theorem implies that there exists a subset E of Rd1 with positive measure such that for
0¼ ð2; ; dÞ 2E, Z þ1
1
logþjF1ðGð; 0ÞÞð1Þeb121j
Ceðb222þþbd2dÞ d1<1:
Therefore, Miyachi’s theorem onRyields thatGðx1; 0Þ ¼Cð0Þea1x21for 02E.
Hence FdfðÞ ¼F1ðGð; 0ÞÞð1Þ ¼Cð0Þeb121 for 02E. Since FdfðÞ has a holomorphic extension onCn(see (1)), we can prolong the precedent relation as FdfðÞ ¼Cð0Þeb121 for all02Rd1. HenceGðx1; 0Þ ¼Cð0Þea1x21 for all0 2 Rd1. Here we put hðxÞ ¼ea1x21fðxÞ. Then, as a function of x0¼ ðx2; ; xdÞ, it belongs to L1ðRd1Þ (see (1)) and Fd1ðhðx1;ÞÞð0Þ ¼ea1x21Fd1ðfðx1;ÞÞð0Þ ¼ ea1x21Gðx1; 0Þ ¼Cð0Þ for 02Rd1. Since Cð0Þ is independent of x1, it follows that h is also independent of x1. We obtain that f is of the form fðx1; x2; ; xdÞ ¼ea1x21hðx2; ; xdÞ. We can obtain the desired result by
repeating this argument.
3. Generalization of Miyachi’s theorem.
Let ðX; d!; Þ be a topological space with a positive measure d! and a distance function :X!Rþ. In what follows we assume the existence of an integral operatorR that satisfies the following properties: R is of the form
RfðyÞ ¼ Z
X
fðxÞdyðxÞ; f2CcðXÞ; ð3Þ
(A1) yis a positive measure with the support infx2Xj kyk ðxÞg, (A2) R is bounded fromL1ðX; d!Þto L1ðRdÞ,
(A3) Rgives an isomorphism betweenSðXÞandSðRdÞ, whereSðXÞis a suitable Schwartz class onX and SðRdÞis the Schwartz class onRd. We here define the dual operatorR:S0ðRdÞ !S0ðXÞofR by
Z
Rd
RfðyÞgðyÞdy¼ Z
X
fðxÞRgðxÞd!ðxÞ: ð4Þ
We furthermore assume that (A4) R is of the form
RfðxÞ ¼ Z
Rd
fðyÞdxðyÞ; f2CðRdÞ; ð5Þ
where x is a positive measure on Rd with support in fy2 Rdj kyk ðxÞg.
Then we see thatRð1ÞðxÞ C0for allx2Xbecause of (4) and (A2). We denote the Euclidean Fourier transform onRd byFd and put
FX¼FdR:
We call FX a generalized Fourier transform on X, which is an isomorphism betweenSðXÞand SðRdÞ(see (A3)). We define Gaussian type functions haðxÞ andhaðxÞ,x2X; a2Rþ, as follows:
haðxÞ ¼ ðRÞ1ðeakk2ÞðxÞ and haðxÞ ¼Rðeakk2ÞðxÞ:
Since eakk22SðRdÞ, haðxÞ is well-defined (see (A3)), positive on X and FXhaðyÞ ¼Cekyk2=4a. Since haðxÞ eaðxÞ2Rð1ÞðxÞ C0eaðxÞ2, haðxÞ is well- defined and positive onX. Especially,ðhaÞ1ðxÞis positive onX.
THEOREM2. Let a >0 and ab¼1=4. If a measurable function f on X satisfiesf ¼f1þf2 such that
haðxÞf1ðxÞ 2L1ðX; d!Þandh1a ðxÞf2ðxÞ 2L1ðX; d!Þ ð6Þ
and
Z
Rd
logþjFXfðÞebkk2j
C d <1
for someC >0, then f is a constant multiple ofha.
PROOF. We show thatRðfÞsatisfies
eakyk2RfðyÞ 2L1ðRdÞ þL1ðRdÞ:
Actually, by (6) fðxÞ ¼f1ðxÞ þf2ðxÞ ¼ ðhaÞ1ðxÞv1ðxÞ þhaðxÞv2ðxÞ, wherev12 L1ðX; d!Þand v22L1ðX; d!Þ. Then
keakyk2Rf1kL1ðRdÞ ¼ Z
Rd
eakyk2RððhaÞ1v1ÞðyÞdy
Z
Rd
eakyk2RððhaÞ1jv1jÞðyÞdy
¼ Z
X
Rðeakk2ÞðxÞðhaÞ1ðxÞjv1ðxÞjd!ðxÞ(see(4))
¼ Z
X
jv1ðxÞjd!ðxÞ
and
eakyk2jRf2ðyÞj kv2k1eakyk2RhaðyÞ ¼ kv2k1:
Hence eakyk2RfðyÞ belongs to L1ðRdÞ þL1ðRdÞ. Since FXf¼FdðRfÞ, it follows from Miyachi’s theorem on Rd (see Theorem 1) that RfðyÞ ¼Ceakyk2.
Hence we obtain thatf¼Cha.
REMARK3. If there exists a functionðxÞonXsuch that haðxÞhaðxÞ ðxÞ;
then the conditionð6Þin Theorem 2 can be replaced by
h1a ðxÞfðxÞ 2L1ðX; d!Þ þL1ðX; d!Þ: ð7Þ
4. Examples.
4.1. Che´bli-Trime`che hypergroups.
We refer to [1], [10] and [12] for general notations and basic facts on Che´bli- Trime`che hypergroups. Let >1=2 and AðxÞ ¼x2þ1BðxÞ be a Che´bli- Trime`che function on Rþ. Then ðX; d!; Þ ¼ ðRþ; AðxÞdx;j jÞ and the Che´bli- Trime`che transform FX gives an isometry between L2ðX; d!Þ and L2ðR; dÞ wheredðÞ ¼ jCðÞj2dand supported on even functions onR. We recall that FX¼FR where R is the Weyl type integral transform
RfðyÞ ¼ Z 1
y
fðxÞKðx; yÞAðxÞdx; y0
andRis the Riemann-Liouville type integral transform
RfðxÞ ¼ Z x
0
fðyÞKðx; yÞdy; x0:
The multiplicative functions on the hypergroup coincide with spherical functions ðxÞ,2C. LetSeðRÞdenote the space of even rapidly decreasing functions on RandSðXÞ ¼0ðxÞSeðRÞ, where0 is the spherical function with¼0. Then R satisfies the assumptions (A1) - (A4) in Section 3 (see [12]), where dyðxÞ ¼Kðx; yÞ½y;1ÞðxÞAðxÞdx, y0, and dxðyÞ ¼Kðx; yÞ½0;x ðyÞdy, x0.
Hence, by introducing the Gaussian type functionsha andha, Theorem 2 holds for the Che´bli-Trime`che transformFX. We shall obtainðxÞin Remark 3. We recall from [12] that
haðxÞ C 1 ffiffiffiffiffiffiffiffiffiffiffi pBðxÞeax2:
On the other hand, sinceeax2¼eaðx =2aÞ2þ x 2=4a,
haðxÞ Ceaðx =2aÞ2Rðe yÞðxÞ Ceaðx =2aÞ2i ðxÞ Ceax2 x:
Hence
haðxÞhaðxÞ C 1 ffiffiffiffiffiffiffiffiffiffiffi pBðxÞe x
and
C 1
ffiffiffiffiffiffiffiffiffiffiffi BðxÞ
p e xd!ðxÞ ¼Cx2þ1 ffiffiffiffiffiffiffiffiffiffiffi pBðxÞ
e xdx
In particular, if X is the Bessel-Kingman hypergroup, then AðxÞ ¼x2þ1 and d!¼Cx2þ1dx. If X is the Jacobi hypergroup, then AðxÞ ¼cðsinhxÞ2þ1 ðcoshxÞ2þ1, 1=2, and d!¼CðtanhxÞþ1=2xþ1=2dx. Hence the con- ditionð7Þin Remark 3 is nothing but the one used in [6, Theorem 3.1].
4.2. Dunkl analysis.
We refer to [8] and [11] for general notations and basic facts in Dunkl transform. LetðX; d!; Þ ¼ ðRd; !kðxÞdx;k kÞ wherekis a multiplicity function on the root system. Then the Dunkl transform FX gives an isometry between L2ðX; d!ÞandL2ðRd; Ckd!Þ. We recall thatFX¼FdR. HereR and its dual operatorRare given as
RfðyÞ ¼ Z
Rd
fðxÞdyðxÞ;
whereyis a positive measure onRdwith the support infx2Rdj kxk kykgand RfðxÞ ¼
Z
Rd
fðyÞdxðyÞ;
where x is a positive measure of probability on Rd with the support in fy2 Rdj kyk kxkg(see [11]). LetSðRdÞdenote the Schwartz space onRd. ThenR satisfies the assumptions (A1) - (A4) in Section 3 (see [11]). Moreover, we note thathaðxÞ ¼eakxk2 and
haðxÞ eakxk2Rð1ÞðxÞ ¼eakxk2:
HencehaðxÞhaðxÞ 1and ðxÞ ¼1inð7Þ. Hence we can obtain the correspond- ing Theorem 2 for the Dunkl transform.
4.3. Jacobi-Dunkl analysis.
We refer to [4] for general notations and basic facts in Jacobi-Dunkl transform. Let 1=2 and 6¼ 1=2. We put ¼þþ1>0 and A;ðxÞ ¼22 ðsinhjxjÞ2þ1ðcoshxÞ2þ1. Then ðX; d!; Þ ¼ ðR; A;ðxÞdx;j jÞ and the Jacobi-Dunkl transform FX gives an isometry between L2ðX; d!Þ and L2ðR; dÞ. Here dis of the form
dðÞ ¼ jj
8 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2
p cð ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2
p Þ21Rn ; ½ðÞd;
where 1Rn ; ½ is the characteristic function of Rn ; ½ and cðÞ is a meromorphic function on C. We recall that FX¼F R where R the Jacobi-Dunkl dual intertwining operator defined by
RfðyÞ ¼ Z
jxjjyj
Kðx; yÞfðxÞA;ðxÞdx
andRis the Jacobi-Dunkl intertwining operator defined by
RfðxÞ ¼ Z jxj
jxj
Kðx; yÞfðyÞdy if x2Rnf0g, fð0Þ if x¼0.
8>
<
>:
LetSðRÞdenote the Schwartz space onRandSðXÞ ¼ ðcoshxÞ2 SðRÞ. Then R satisfies the assumptions (A1) - (A4) in Section 3, where dyðxÞ ¼ Kðx; yÞ½jyj;1ÞðxÞA;ðxÞdx and dxðyÞ ¼Kðx; yÞ½0;jxj ðxÞA;ðxÞdx. Hence, by introducing the Gaussian type functions ha and ha, Theorem 2 holds for the Jacobi-Dunkl transformFX. As in Jacobi hypergroup case, the condition (7) in Remark 3 is given byd!¼CðtanhjxjÞþ12jxjþ12dx.
References
[ 1 ] W. Bloom and H. Heyer, Harmonic analysis of probability measures on hypergroups, Walter de Gruyter, Berlin, New-York, 1995.
[ 2 ] A. Bonami, B. Demange and P. Jaming, Hermite functions and uncertainty principles for the Fourier and the windowed Fourier transforms, Rev. Mat. Iberoamericana,19(2003), 23–55.
[ 3 ] F. Chouchene, R. Daher, T. Kawazoe and H. Mejjaoli, Miyachi’s theorem for the Dunkl transform, preprint, 2007.
[ 4 ] F. Chouchene, M. Mili and K. Trime`che, Positivity of the intertwining operator and harmonic analysis associated with the Jacobi-Dunkl operator onR, J. Anal. Appl. (Singap.),1(2003), 387–412.
[ 5 ] M. Cowling and J. F. Price, Generalizations of Heisenberg’s inequality. Lecture Notes in Math., 992, Springer Verlag, 1983, pp. 443–449.
[ 6 ] R. Daher and T. Kawazoe, Generalized Hardy’s theorem for Jacobi transform, Hiroshima Math.
J.,36(2006), 331–337.
[ 7 ] G.H. Hardy, A theorem concerning Fourier transforms,J. London. Math. Soc.,8(1933), 227–
231.
[ 8 ] M. F. E. de Jeu, The Dunkl transform,Invent. Math.,113(1993), 147–162.
[ 9 ] A. Miyachi, A generalization of theorem of Hardy, Harmonic Analysis Seminar held at Izunagaoka, Shizuoka-Ken, Japan 1997, pp. 44–51.
[10] K. Trime`che, Generalized wavelets and hypergroups, Gordon and Breach Science, 1997.
[11] K. Trime`che, The Dunkl intertwining operator on spaces of functions and distributions and integral representation of its dual, Integral Transforms Spec. Funct.,12(2001), 349–374.
[12] K. Trime`che, Cowling-Price and Hardy theorems on Che´bli-Trime`che hypergroups, Glob. J.
Pure Appl. Math.,1(2005), 286–305.
Radouan DAHER Department of Mathematics Faculty of Sciences and Informatics University Hassan II
B.P. 5366 Maarif, Casablanca Morocco
Takeshi KAWAZOE Department of Mathematics Keio University at Fujisawa Endo, Fujisawa
Kanagawa 252-8520 Japan
Hatem MEJJAOLI Department of Mathematics Faculty of Sciences of Tunis Campus, 1060 Tunis Tunisia