CC BY
0Hindawi Publishing Corporation Abstract and Applied Analysis Volume 2012, Article ID 879073,26 pages doi:10.1155/2012/879073
Research Article
Generalized Carleson Measure Spaces and Their Applications
Chin-Cheng Lin1 and Kunchuan Wang2
1 Department of Mathematics, National Central University, Chung-Li 320, Taiwan
2 Department of Applied Mathematics, National Dong Hwa University, Hualien 970, Taiwan
Correspondence should be addressed to Chin-Cheng Lin, clin@math.ncu.edu.tw Received 10 October 2011; Revised 20 February 2012; Accepted 12 March 2012 Academic Editor: Stevo Stevic
Copyright © 2012 C.-C. Lin and K. Wang. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
We introduce the generalized Carleson measure spaces CMOa'q that extend BMO. Using Frazier and Jawerth's ^-transform and sequence spaces, we show that, for a € R and 0 < p < 1, the duals
of homogeneous Triebel-Lizorkin spaces fp>'q for 1 < q < g and 0 < q < 1 are CMOq/pyq/q)
and CMO;a+(n/p)-n'~ (for any r € R), respectively. As applications, we give the necessary and sufficient conditions for the boundedness of wavelet multipliers and paraproduct operators acting on homogeneous Triebel-Lizorkin spaces.
1. Introduction
In 1972, Fefferman and Stein [1] proved that the dual of H1 is the BMO space. In 1990, Frazier and Jawerth [2, Theorem 5.13] generalized the above duality to homogeneous Triebel-Lizorkin spaces Fpq. More precisely, they showed that the dual of F1q is F^ for a € R and 0 < q < g, where q' is the conjugate index of q. Throughout the paper, q' is interpreted as q' = g whenever 0 < q < 1, and q' = q/(q - 1) for 1 < q < g.
BMO = FGg2. For a € R, 0 <p< 1, and 0 <q< g, it is known (cf. [2-4]) that the dual of F^ is
F<£+(l//P n'G. Here, we will give another characterization for the duals of FP'q in terms of the generalized Carleson measure spaces for a € R, 0 <p < 1, and 0 < q < g.
We say that a cube Q c Rn is dyadic if Q = Qjk = {x = (x\, x2xn) € Rn : 2-jki < Xi < 2-j (ki +1), i = 1,2,...,n} for some j € Z and k = (ki,k2,...,kn) € Zn. Denote by £(Q) = 2-j the side length of Q and by xq = 2-jk the "left lower corner" of Q when Q = Qjk. We use supp and Xp to express the supremum and summation taken over all dyadic cubes P, respectively. Also, denote the summation taken over all dyadic cubes Q contained in P by ^qcp . For any dyadic cubes P and Q, either P and Q are nonoverlapping or one contains the other. For any
function f defined on R", j e Z, and dyadic cube Q = Qjk, set
/q(x) = \Q\ ^/(XQ1) = 2in/2f(2ix - k),
fj (x) = 2jnf ( 2jx), f(x) = f-x).
It is clear that gj * f (xq) = | Q| X/1{f,gQ), where (J,g) denotes the paring in the usual sense for g in a Frechet space X and f in the dual of X.
Choose a fixed function y in Schwartz class S = S(R"), the collection of rapidly decreasing Cg functions on R", satisfying
supp(<^) c jl: 2 <\l\< 2}, 3 5
|£(l)| > 0 0 if 5 <\l\< 3.
For a e R and 0 < p,q < +<x>, we say that f belongs to the homogeneous Triebel-Lizorkin space tp,q if f eS'/P, the tempered distributions modulo polynomials, satisfies
P a,q Fp
Z(2kaWk * f I)
for 0 <p < GO,
if G V^
\\P\-1 E (2ka|^k * f (x)ßqdxV < GO for p = G.
J Pk=-log2^(P) j
When 0 < p < go and q = to, the above ¿"?-norm is modified to be the supremum norm as usual, and t^T is defined to be B, which is
to» := sup sup 2kaIfk * f (x)| « sup\Q\-(a/n)-(1/2)
keZ xeQ Q
e(Q)=2-k
Kf'fQ}1 < G
We now introduce a new space CMO^ as follows.
Definition 1.1. Let y e S satisfy (1.2). For a,r e R and 0 < q < g, the generalized Carleson measure spaces CMOp'q is the collection of all f e S'/P satisfying ||f ||cmom < g, where
su4\P\-f E( \ Q \ -(a/n)-(1/2) Kf'VQ)IlQ(x))qdx
P PQ P
sup sup Q
-(a/n)-(1/2)K f,yQ) I = sup \ Q \-(a/n)-(1/2)K f,yQ) |, Q
0 < q < go, q = g,
and xq denotes the characteristic function of Q.
Remark 1.2. By definition, we immediately have CMO"'œ = Fœœ for a,r e R, and it is easy to check CMOar'q = {0} for r < 0 and 0 < q < œ. Note that the zero element in CMO/ means the class of polynomials. Also note that CMO^ = F^ with equivalent norms for a e R and 0 < q < œ. It follows from Proposition 3.3 that CMO^ = for a e R and 0 < q < œ. In particular, CMO^'2 = BMO, and hence the spaces CMO^ generalize BMO.
Remark 1.3. For a dyadic cube P, denote by kP = -log2£(P); that is, kP is the integer so that i(P) = 2-kp. In [5, 6], Yang and Yuan introduced the so-called "unified and generalized" Triebel-Lizorkin-type spaces f7^ with four parameters by
:= sup IP
E {2ka\n * f (x)\)q
for a,T e R, p e (0, to), q e (0, to], and f e S/P. Note that in [5] the space was defined for t e [0, to), p e (1, to), and q e (1, to]. It follows from [6, Theorem 3.1] that
sup|P rjf E(lQI(-a/")-(1/2)Kf,^Q >\XQ(x))
P \J P .QcP
It is clear that CMO^ = F^"1 for 0 <q < to, and hence CMO^ "looks like" a special case of Fpq. In fact, it was proved in [7, 8] that the space F^ is the "same" as the space
CMO"'q1 , .
rq+1-q/p
The definition of CMOaq is independent of the choice of y e S satisfying (1.2). To show that, we need the following Plancherel-Polya inequalities.
Theorem 1.4 (Plancherel-Polya inequality for 0 < q < to). Let e S satisfy (1.2). For a,r e R and 0 <q < to, if f eS' / P satisfies
k=-log2^(P) QCP
e(Q)=2-k
E E 2kasup\^k * f (u)\ IQI
k=-log2^(P) QCP
e(Q)=2r<
IP r Ë E (2kasup\^k * f (u)\) IQI
IP Ir E E [2kaUnQ \(pk * f (u)\) IQI
k=-log2£(P) QCP \ ueQ
e(Q)=i-k
Theorem 1.5 (Plancherel-Polya inequality for q = to). Let eS satisfy (1.2). For a,r e R, if f eS'/P satisfies
sup IQI
<a/n)-rsup\ykQ * f («)\ ) < TO,
(1.10)
sW |Q|-(a/n)-r sup \(^kQ * f (M)\j « sup( IQ|-(a/n)-rMnQ* f (M)\) . (1.11)
Remark 1.6. Let e S satisfy (1.2). Denote by CMO^(y) the collection of all f e S'/P satisfying ||f ||CMO«-q(^) < to defined in Definition 1.1 with respect to y. Then, by Theorem 1.4,
CMor < ^F1'
IPI-r E E (2kasup\^k * f (m)\ ) IQI
k=-log/(P) QçP \ «eQ /
e(Q)=2-k
< C sup
IPI-r E E 2kainf* f (u)\) IQI
k=-log2^(P) QçP V "eQ
£(Q)=2-k
(1.12)
< Cllf llcMo^-q w for0 <q< TO.
Similarly, ||f ||CMO«-qy < C||f |CMO«-q(^ by interchanging the roles of y and Hence, the definition of CMO^y) is independent of the choice of y and, for short, denoted by CMO^. Also, Theorem 1.5 shows that CMOa,TO is independent of the choice of y satisfying (1.2) in the same argument.
Remark 1.7. The classical Plancherel-Polya inequality [9] concludes that if {xk} is an appropriate set of points in R", for example, lattice points, where the length of the mesh
is sufficiently small, then
E\f(*k)\M «Hfl. (1.13)
for all 0 <p <to with a modification if p = to.
Using the CalderOn reproducing formula (either continuous or discrete version), several authors obtain the variant Plancherel-Polya inequalities [10-13]. These inequalities give characterizations of the Besov spaces and the Triebel-Lizorkin spaces. Moreover, using these inequalities, one can show that the Littlewood-Paley ^-function and Lusin area S-function are equivalent in Lp-norm.
Define a linear map Sv from S'/P into the family of complex sequences by
Sf = { if'V^Q- (1-14)
Let So denote the family of f e S satisfying Jxkf (x)dx = 0 for all k e (N U {0})". For g e CMO-a'q, define a linear functional Lg by
Lg (f) = S(g), S,(f )> = Z(g'fQXf'VQ> for f e So. (i 15)
We now state our first main result as follows. Theorem 1.8 (duality for f^)- Suppose that a e R, 0 <p < 1, and 0 < q < to.
(a) For 1 <q < to, the dual of Fpq is CMO^/,^^) in the following sense.
(i) For g e CMOj;pHq,/q), the linear functional Lg given by (1.15), defined initially on S0, extends to a continuous linear functional on tO1 with ||Lg || < C\\g\\ O-a,q .
CMO(q'/p)-(q'/q)
(ii) Conversely, every continuous linear functional L on Fp'q satisfies L = Lg for some g e CMO-;'i)-(q'/q) With llgllCMO-' ( '/ ) < C"L".
(q' /p)-(q/q)
(b) For 0 <q < 1,the dual of ta'q is CMO-a+("/p)-"'TO (any r e R) in the following sense.
(i) For g e CMO-a+("/p)-"'TO, the linear functional Lg given by (1.15), defined initially on S0, extends to a continuous linear functional on F^ with ||Lg|| <
C|lgllCMO-a+("/p)-"'TO.
(ii) Conversely, every continuous linear functional L on f^ satisfies L = Lg for some
g e CMO-a+("/p)-"'TO with 11g11CMO-a+("/p)-",TO < CllLll.
Remark 1.9. For 0 < p < 1 and 0 < q < 1, it follows immediately from [2, 3] (Verbitsky [4] corrected a gap of the proof) and definition that (t^)' = = CMO-a+("/p)-"'TO
(any r e R). Theorem 1.8 (b) shows a different approach to the duality and includes the case of p = 1.
For p = 1 < q < to, we have CMO-"^/^ = (fa'q)' = F^TOa'q. For 0 < p < 1 < q < to,
CMO-a'q = (fa'q)' = f-a+("/p)-"'TO and hence CMO-a'q = CMO-a+("/p)-"'TO That CMO(q'/p)-(q'/q) (Fp ) Fto , and hence CMO(q'/p)-(q'/q) CMOr . That
is, each CMOaq,/p)-{q/ql) coincides with CMOa+("/p)-"'TO for a,r e R and 0 <p < 1 <q< to.
Remark 1.10. In Remark 1.2 we are aware that CMOa'q generalize BMO by the viewpoint of spaces directly. Choosing a = 0 and q = 2 in Theorem 1.8, we immediately have (Hp)' = (tp-2)' = CMO02/p)-1 for 0 < p < 1. In particular, BMO = CMO0'2. Once again, we obtain
that CMOa'q generalize BMO by the viewpoint of duality. It was also proved in [14] that the dual of the multiparameter product Hardy space is the generalized multiparameter Carleson measure space (cf. [14] for more details).
Remark 1.11. For a,r e R, in order to make each index works, we defined CMOa/<x> to be supP\P\-rsupQrp\Q\-(a/n)-(i/2)\(J,fQ)\ in our earlier 7
for 0 < p,q < 1, the dual of F^ would be CMO-*/^. In this paper, however, we follow the referee's suggestion and adopt a more "natural" definition of CMO^ in Definition 1.1, that is, the limit of CMO^ as q ^ to. The sequence space c^ given in Definition 2.1 has a similar story as well.
As applications, we first recall the Haar multipliers introduced in [15, 16]. Given a sequence t = [tI}j, where the I's are dyadic intervals in R, a Haar multiplier on L2(R) is a linear operator of the form
Htf (x) := £ ti{f,hi)hi(x), f e L2(R), (1.16)
where hI are the Haar functions corresponding to I.
Using Meyer's wavelets, we may generalize the above Haar multiplier to Rn and obtain a necessary and sufficient condition for the boundedness on Triebel-Lizorkin spaces. Let [yv} for i e E := [1,2,...2n - 1} be Meyer's wavelets (cf. [17], [18, pages 71-109]). Then, [yQ}, where i e E and Q's are dyadic cubes in Rn, is a frame for Fpq for a e R and 0 <p,q <to; that is, \\f \\paq a £ieE \\[(f,yQ)}QII aq for f e Fpq. For t = [tQ}Q, define a wavelet
p Q Jp
multiplier Xt on Rn by
Ttf) = EE \Q\-1/2tQ f,yQ )yQ (1.17)
for f e S/P such that the above summation is well defined. Theorem 1.12. Suppose that a,p e R, 0 <p < 1 and 0 < q < to. Then,
(a) for 1 < q < to, Tt is bounded from Fpq into Fa^1 if and only if t e ^/y^/y
(b) for 0 < q < 1 and r e R, Tt is bounded from F(p'q into Fp^1 if and only if t e cf+(n/p)-n,TO, where c^ is given in Definition 2.1.
We consider another application. Let y and y in S satisfy (1.2) and (3.1). Choose a function O e S supported on [0,1]n and J® = 1. For a e R and g e FTOtoTO, define the paraproduct operator ng by
ng (f) = E{g,yQ )\Q\-1/2{f, Oq )yQ. (1 18)
Thus, the adjoint operator ng is
ng(f) = Z{g,yQ)\Q\-m(f,yQ) Oq. (1.19)
Then, ng 1 = g and ng 1 = 0 since (1,Oq) = \Q\1/2 and (1,yQ) = 0. Also, if g e F°TOTO, then
both ng and ng are singular integral operators satisfying the weak boundedness property. Moreover, ng is a Calderon-Zygmund operator (i.e., ng is bounded on L2(R")) if and only if g e fTO2 by David-Journe's T1 theorem [19] (also see [12, Theorems 5.4 and 5.8]). The authors showed a more general type of paraproduct operators in [12, page 688], which were derived from the discrete Calderon reproducing formula.
Theorem 1.13. Suppose that ¡5 e R, 0 <r < 1 and 0 <p < r < q <r/(1 - r).
(i) For a < 0, ng is bounded from tp,''1 into F;+5'r if and only if g e CMO5''q-/;()q/pr()q-r).
(ii) If a e R with a + ¡5 > 0 and g e CMO¡;rP/pr()q-r), then ng is bounded from tp' into
■a+fi/r
Remark 1.14. When r = 1,0 <p < 1 < q < to, and ¡5 e R, Theorem 1.13 says that ng is bounded from p;" into fa+5'1 if and only if g e CMO;/p)-(q'/q) for a < 0, and ng is bounded from p;" into Fa,+5'1 for a > -5 provided g e CMO^p)-(qVq). In 1995, Youssfi [20] showed that, for ¡5 e R, 1 < p < to, 1 < q < 2, and g e fTO'TO, ng is bounded from tp' into tp,'v if and only if g e fTOp. The special case of Theorem 1.13(i), p = r, generalizes Youssfi's result to 0 < p < 1. More precisely, for a < 0, 5 e R, 0 <p < 1, and p < q <p/(1 - p), ng is bounded from P,'' to
f;+5'p if and only if g e CMOfq/(q-p) = fTOpq/(q-p).
The paper is organized as follows. In Section 2, we introduce the discrete version of the generalized Carleson measure spaces c;'q and show that the duals of sequence Triebel-Lizorkin spaces fP'q for 1 < q < to and 0 < q < 1 are c;/y/) and cr;+("/p) "'TO (for any r e R), respectively. In Section 3, we prove the duals of homogeneous Triebel-Lizorkin spaces Fpq for 1 < q < to and 0 < q < 1 to be the generalized Carleson measure spaces CMO(;'/p)-(q,/q)
and CMO-;+("/p) "'TO (for any r e R), respectively. In Section 4, we prove the Plancherel-Polya inequalities that give us the independence of the choice of y for the definition of the generalized Carleson measure spaces. In the last section, we show the boundedness of wavelet multipliers and paraproduct operators. Throughout, we use C to denote a universal constant that does not depend on the main variables but may differ from line to line. Also, Q and P always mean the dyadic cubes in R", and, for r > 0, we denote by rQ the cube concentric with Q whose each edge is r times as long.
2. Sequence Spaces
In this section, we introduce sequence spaces c;'q and then characterize the duals of fp'q by means of c;'q. Let us recall the definition of these sequence spaces fp'q defined in [2]. For a e R and 0 <p; <to, the space fp'q consists all such sequences s = {sq }q satisfying
|s|f;'q :=
X(lQ|-(a/")-(1/2)|sQ|XQ)
< to if 0 <p < to,
Lp , 1/q (^
sup I IP Z (|Q|(-a/")-(1/2) | Sq | XQ (x) )qdx\ < TO if p = TO.
As before, the previous £q-norm is modified to the supremum norm for 0 <p < to and q = to. For p = q = to, we adopt the norm
J := supiQr^-^lsQl. (2.2)
Note that \\s\\qaq is equivalent to the Carleson norm of the measure
E( i Q i-(a/n)-(1/2)|sd)q i Q i <W(Q)), (2.3)
where 6(x t) is the point mass at (x,t) e R++1. See [2] for the details
To study the duals of fp^,q, we introduce a discrete version of the generalized Carleson
measure spaces ca'q.
Definition 2.1. For a,r e R and 0 < q < to, the space cr is the collection of all sequences t = [tQ}q satisfying WtW^q < to, where
'sup \ P \ ~r[ E ( \ Q \-(a/n)-(1/2)|tQ|XQ(x))qdx for 0 <q< TO,
P PQ P
lltllc-q :=
supsup iQ r(a/n)-(1/2)ltQl = sup iQ i-(a/n)-(1/2)|fQ| for q = to.
P QCP Q
It is obvious that
\P\-r E (\ Q \-(r/n)-(1/2)+(1/q)|tQ|)^ for 0 <q< TO (2.5)
and M^to = \\t\\fTOTO for a,r e R. Using embedding theorem, Frazier and Jawerth [2, equation (5.14) and Theorem 5.9] obtained that, for a e R and 0 < q < to, the dual of
fP,q is fTO-^+(n/p)-n,TO when 0 <p < 1, and the dual of fpq is f^ . Note that c^ = [0} for r < 0
T^TorC» TA70 rri^70 fVl Q H 1 1 Q 1 f01 d \ O V* 1 t~\ CDHIIDnrD cncipoc -f
and 0 < q < to. Here we give the dual relationship between sequence spaces fp,q and c^.
Theorem 2.2 (duality for fp,q). Suppose that a e R, 0 <p < 1, and 0 < q < to.
(a) For 1 < q < to, the dual of fp,q is c-^)^, /) in the following sense.
(i) For t = [tQ }q e cq*)W the linear functional ¿t on fp" given by £t(s) = Xq SQtQ is continuous with \\£t\\ < C\t\for s = [sq}q e fp,q.
(q'/p)-(q'/q)
(ii) Conversely, every continuous linear functional i on f p^ satisfies i = £t for some
t e %/v)-(q'/q) with m^ < C\\i\\.
(q'/p)-(q'/q)
(b) For 0 <q < 1,the dual of fp,q is c-r+(n/p)-n,TO (any r e R) in the following sense.
(i) For t = {tQ }q e c/+("/p) "'TO, the linear functional 4t on fP'q given by 4t(s) =
Xq SQtQ is continuous with ||4t|| < C|t|c-a+(»/p)-»'TO for s = {sq}q e fp'q. (ii) Conversely, every continuous linear functional 4 on fP'q satisfies 4 = 4t for some t e c-a+("/p)-"'TO with ||t||c-a+(»/p)-»'» < C||4||.
Remark 2.3. For a e R and 0 < q < to, sequence spaces cp' = fTO'q and c;'TO = fTO'TO (for any
r e R) by definitions. Theorem 2.2 shows that (fP'q)' = f-;'q, which gives a different but simpler proof of Frazier-Jawerth's result for the duality of f,'q (cf. [2, Theorem 5.9]).
Proof of Theorem 2.2. For s = {sq }q e fp'q and t = {tQ}Q e c-a';, set s = {Sq}q and t = {?q}q to
be Q Q Q Q
Sq = ¿q = IQIa/"tQ. (2.6)
Then, £f(s) = 4(s). Also,
llf = fp^' |i||c0'q, = 1111 . (2.7)
Without loss of generality, we may assume that a = 0.
We first consider the case 1 < q < to. Let t e c;'/) and define a linear functional
4 on f°'q by
4(S) = X sQtQ for s e fpq. (2.8)
For S = {sq}q e f^,let
vq{x) :=(|] (|Q|-1/2|sq|xq(x^ ) . (2.9)
For k e Z, let
Qk : = {x e Rn :2k < Vq(x) < 2k+^,
Qk : = { x e R" : Mxak (x) > 1J' (2.10)
Bk := {dyadic Q : |Q n > Q' |Q n j1 < Q for some j > k
where M is the Hardy-Littlewood maximal function. Then, for each dyadic cube Q, there exists exactly a k e Z such that Q e Bk. For every Q e Bk, let Q denote the maximal
dyadic cube in Bk containing Q. Then all of such Q's are pairwise disjoint. Thus, by Holder's inequality for q and the inequality (a + b)p < ap + bp for 0 <p < 1,
ZsQtQ Q
ee e( iQ i-(1/2)+(1/?)|sqI)( i Q i (1/2)-(1/q)m)
k(=-Z QB QÇQ
E E I E ( iQ i-(1/2)+(1/q)M)q I I E (i Q r(1/2)+(1/q,)M)q
kG~Z QeBk\ QÇQ I \ QCQ
Wfp)-W fq)
EE|E( i Q i -(1/2)+(1/q)|sqi)qI |Q|
teZ QpP„. V qcQ I
p/q 1fP
q 1 ^|i-(p/q)
QeBk \ QÇQ QB
(2.11)
Since Q e Bk implies Q c k, the disjointness of Q's and Holder's inequality yield
p/q1/p
E sq^q Q
< lltll
(q'/p)-(q'/q) ^kez
Îe|ôkr/q)(E (i Q i -(1/2)+(1/q)|SQ0'
I kez \Q(=-Bk
(2.12)
We claim that £QeBk ( \Q\-(1/2)+(1/q) \sq\)q < C2kq\£k\ for k e Z and 0 < q < to. Assume the claim for the moment. The weak (1,1) boundedness of M gives \££k\ < C\£k \, and hence
E SQ^Q Q
< C||t|| oq
(q'/p)-(q' fq)
(e|ö k
1-(p/q)
E 2kp i ik i
< cpy^ _
(q'fp)-(q'fq) \kez
< C|t|c0^ \\Vq\\Lp
(qlfp)-(ql fq)
= C|t|c0,ql ||s|/0,q.
(q'fp)-(q' fq) Jp
To prove the claim, we note that, for k e Z and 0 < q < to, 2q(k+1)|£k 1 >
(2.13)
k > (Vq(x))qdx
}Q k\U f=k+1 ï
E (|QI-1/2|sqIiq(x))qdx
J ikM]^ ï Q v 7
(|QI-1/2|sq|)q| (ïk \ j n q| for some j > k + 1,
Abstract and Applied Analysis which implies
2q(fc+1)|6fc|> (|Q|
-(1/2)+(1/q)
(2.15)
For 0 < q < 1, with a modification, we have
ZSQtQ Q
(i/p)-i
<XX X(lQl1/2|sQO(lQI-(1/p)+(1/2)|tQl)v M ,
keZ Q eB^Q-Q \|Q|/
<l|t||
,(n/p) -n,c
XXIX IQI1/2|SQ|
QeBk \ Q-Q
< C||t||c(n/p)-n. ( X|"fc|1-^2fc|ñJ)p
< c|t|c(n/P)-n,~ y vqy^
(1/p)-1
(2.16)
On the other hand, suppose that £ is a continuous linear functional on fP'q. For each dyadic cube P, write eP = {(ePto be the sequence defined by
1 if Q = P, Q 10 if Q/ P•
(2.17)
Let tP = £(eP) and t = {tP}P. Then, for s = {sq}q e ,
¿(s) = X sQtQ = 4(s)
(2.18)
Fix a dyadic cube P. For 1 < q < to, let X be the sequence space consisting of s = {sq}qcp, and define a counting measure on dyadic cubes Q c P by da(Q) = |Q|/|P|(q /p)-(q /q). Then,
(q'/p)-(q'/q)
X (QI-(1/2)+(1/q')|tQ|)q
\\s\\eq (X'da)
IP|(q/p)-(q/q) qU IQI SQ|Q|-1/2|tQ|
<\\£\\ sup
\\eq(X'do)
(q'/p)-(q'/q)
Note that
{| pPS-/'^} QÇP - I PI (q'/l)-(q'/q)^EE Q ^^ ^ ^ ^ }
f0,q I I \QÇP / I (2.20)
< C||s||m?(X,dc).
1 q \ 1/q
__L_ , Q j(-l/2)+(l/q') |tQ|)^ < Cm. (2.21)
and hence t e c0qq/p)-(qyq). For 0 < q < 1, consider eP defined before. Then, \\ePW^ =
\ p\-(i/2)+(i/p) and
( \ P \ (1/2)-(1/p)\ tp\)||ep|| = \ tp\ = |^(ep) | < \m\||ep|| (2.22)
Hence, HtH^/p)-^ = supP \ P\(1/2)-(1/p)\ tP\ < \\e\\. This completes the proof. □
3. Proof of the Main Theorem
Let us recall the y-transform identity given by Frazier and Jawerth [2]. Choose a function y eS satisfying (1.2). Then there exists a function y eS satisfying the same conditions as y such that Xy(2-j£)y(2-j£) = 1 for = 0. The y-transform identity is given by
f = E{f,yQ)yQ, (3.1)
where the identity holds in the sense of S' / P, So, and Fpq -norm.
Define a linear map Sy from S' /P into the family of complex sequences by
SyJ) = {{J,yQ)}Q, (3.2)
and another linear map Ty from the family of complex sequences into S'/P by
^(Mq) =x SQyQ. (3.3)
Then, Ty o Sy\paq is the identity on Fp,q by [2, Theorem 2.2].
Proposition 3.1. Suppose that a e R and, 0 <p,q < +œ, and y,y in S satisfy (1.2) and (3.1). The
7y • Fp " Jp and Ty • Jp
linear operators Sy : FJq i—> jp'^ and Ty : fa i—> Fa* defined by (3.2) and (3.3), respectively, are
F aq Fp
Dual relation by Theorem 2.2 (a)
(q '/p)-(q/q)
Dual relation by Theorem 1.8(a)_ cmo-"'17'
CMO(q'/p)-(q'/q)
Figure 1: Diagram for spaces and maps for 1 < q < to.
faq jP
Dual relation by Theorem 2.2 (b)
(n/p)-U'!X
a+(n/p)-
Figure 2: Diagram for spaces and maps for 0 < q < 1.
bounded. Furthermore, Tf o Sf is the identity on p;". In particular, ||f Hp;; « ||Sy(f)Wfa; and t;'q can be identified with a complemented subspace of fP'q.
Figures 1 and 2 illustrate the relationship among P,'', fP'q, CMO;'q, and c;'q.
One recalls the almost diagonality given by Frazier and Jawerth [2]. For a e R and 0 <p'q < to, let J = n/(min{1,p,q}). One says that a matrix A = {aQP}qp is (a,p^)-almost diagonal if there exists e> 0 such that
sup |aQP I < +TO' (3.4)
Q'P wQP (e)
() (e(Q)YA |xQ - xp | VJ-£ . f/4(Q)\("+£)/2 / 4(P) x(("+-)/2)+J-"| WQP(e)n^j V + max(*(P),e(Q))J im) I
Lemma 3.2. For a,r e R and 0 < q < to, an (a + nr,q,q)-almost diagonal matrix is bounded on c;'q. Furthermore, when r > 0,an (a + nr' <tov TO)-almost diagonal matrix is bounded on c;'TO.
We postpone the proof of Lemma 3.2 until the end of Section 4.
Let r,r e R. For q — œ, we have car'™ — fT and CMOr'œ = FT. Thus, Sy • CMOr'œ ^ ca,co and Ty • cr,œ ^ CMOr-œ are bounded by Proposition 3.1. For 0 <q< œ and f e CMO^, let s — (sq}q — Sy(f). Then, the y-transform identity (3.1) shows that f — ^QsQyQ and Wfiicmo^ — l^y^W^q — Mcy.In particular, \\f \\CMOr'q — l^yWW^q — WSy(f)Wp « \\fh%<. Furthermore, for s e c°r'q,
\\Tf (s)\\cMOaq " Zspyp =
P cMoa,q
Yl^p'Vq P I JQ
= ||As||cM, (3.6)
where A := [(yP,yQ)}qp is (a + nr,q,q)-almost diagonal (cf. [2, Lemma 3.6]) and hence A is bounded on cp'q by Lemma 3.2. Therefore, Sy is bounded from CMOp'q to cp'q and Ty is bounded from cp'q to CMOp'q.
We summarize that Ty o Sy\CMOpq is also the identity on CMOp'q.
Proposition 3.3. For (a,r,q) e R x R x (0, to) or (a,r,q) e R x R x[to}, the linear operators Sy : CMOp'q — c^ and Ty : cp'q — CMOp'q are bounded. Furthermore, Ty o Sy is the identity on CMOp'q and \\f hcmo^ = \\Syf H^q. In particular, \\f hcmo^ = \\Sy(f = \\Sy(f )\\p a \\f H^ for a e R and 0 < q < to, and \\f \\CMOa-TO = \\Sy(f )\\caTO = \\Sy(f )\\j^,to a \\f \\p«,to for a,r e R.
Theorem 1.8 can be proved as a consequence of Propositions 3.1-3.3 and a duality result between two sequence spaces.
>-a,q'
Proof of Theorem 1.8. First let us consider the case for 1 < q < to. Let g e CMO(q,/p)-(q,/q) by Proposition 3.3, \\g\\CMO-pq' = \\Sy(g)\\ -r,q . It follows from Theorem 2.2 that Cs
(q'/p)-(q'/q) _ fy /p)-(q'/q)
is a continuous linear functional on fp'q and \\£Sy(g) \\ a \\Sy (g) \\ -aq . Hence, for f e S0,
c(^/p)-(q'/q)
lLgf l < C\\Sy (g )\\
c V , II 9
(q fp)-(q fq)
sy(f) \U < C\\g\\
cmo a,q
Wfp)-Wfq)
Fa,q. Fp
Since J>0 is dense in Fp'q, the functional Lg can be extended to a continuous linear functional
on F^ satisfying №gN < CHgN
cmo:y, )
(q fp)-(q fq)
Conversely, let L e (FP'q)', and set ê — L ◦ Ty on fp'q. By Proposition 3.1, £ e (fp'q)'.
ÇJpMqi/q) sUchthat
Thus, by Theorem 2.2, there exists t — (îq}q e c^,
— X sQlQ for {qq e f
and HtH^ « UH < CHLH. For f e Fp'q, we have
(q'fp)-(q'fq)
m ◦ Sy(f ) = L ◦ Ty ◦ Sy(f ) = L(f ). (3.9)
So, for f eSo and letting g = T9 (t) = £Q ïq^q,
Lf = e ◦ Sf = £ {f,yQ)tQ = <t,S,(f)>. (3.10)
It follows from [2, equations (2.7)-(2.8)] that (g,f> = (Sv(g),S(p(f)> and (t,S9(f)> = (Tv (t),f > for f e So and g e S'/P. This shows that L(f ) = (Ty(t),f > = Lg(f) for f e So. Proposition 3.3 and Theorem 2.2 give
||g||CMO-«q < CHtnc^ < C||L||. (3.11)
(qVp)-(q' /q) (q'/p)-(q'/q)
A similar argument gives the desired result for 0 < q < 1 with a slight modification, and hence the proof is finished. □
Remark 3.4. As pointed out by one of the referees, Yang and Yuan [8, Theorem 1] show that if t > 1/p and 0 < p,q < œ, then f7^ = Fœ+nT (n/p)/x>, where the definition of Fpq is given in Remark 1.3. Thus, for0 <p < 1 and 1 < q < œ,
(FT)' = F^"^ = fqJ(1/p)-(1/q) = CMO^/p )-(q'/q), (3.12)
which demonstrates a different approach to the duality.
4. Proofs of the Plancherel-Polya Inequalities
In this section we demonstrate the Plancherel-Polya inequalities.
Proof of Theorem 1.4. Without loss of generality, we may assume that a = 0. By (3.1), we rewrite <pj * f (u) as
< * f(u) = ^(f^Q) <(u - x)fQ(x)dx
E X iQi(f,Vk('-xQ)) (u - x)Wk(x - xQ)d'
^ z Q ^
~xQ)) I < (u - x)wk(x - xQ)dx.
e(Q)=2-k
Using the inequality [2, page 151, equation (B.5)]
^(^j (u - x)fk (x - xq)dx
< C2-Klj-fci-
2-(jAfc)
(2-(jAfc) + |u - xq i)
where j A k = min{ j, k} and K > 1 + nr, we obtain
i _ i 9-(jAk)
|*j * f (u)|< ^ E rK|j-k||Q| TTjk I-^ ^ * I- (4.3)
kez Q (2-(jAk) + |u - Xq|) v ;
e(Q)=2-k
Thus, for e(Q) = 2-j,
sup | d>j * f (u)|) < C[ V V 2-K|j-k|| Q |---|Sk * f (xq)|
uuQ % nV ~ ^ Q ^ (2-(jAk) + Xq- - XQ|)n+1 m fQn
e(Q)=2-k / (4.4)
2-(jAk)
< ^ E 2-K|j-k|Q|7j-2-^^*f(Xq)|
kez Q (2-(jAk) + | Xq- Xq ^
e(Q)=2-k
where the last inequality is followed by Holder's inequality and
2-(jAk)
£ -^ < c (4.5)
Q (2-(jAk) + | Xq- Xq |)
e(Q)=2-k
Denote Tq by
TQ := S|<^k * f (u)|q- (4.6)
Since xq can be replaced by any point in Q in the last inequality,
2-(jAk)
/ i~ iV 2-(jAk)
(sup|fc * f (u)|) < ^ E 2-K|j-k||Q|72-"Ak)—|-^TQ- (4.7)
\ueQ- '/ kez Q (2-(jAk) + | Xq- Xq |)
Given a dyadic cube P with e(P) = 2 ko, the above estimates yield
■x, / \ q
E E sup|<^j * f (u)| |Q'|
j=ko QQP \ueQ- /
e(Q-)=2-j
2-(jAk) (4.8)
_T1 i i v '
x - jAk
< E 2-K|j-k||Q'|--—-2-^tq|q|
j=ko Q'CP kez Q (2-(jAk) + |xQ, - XQ|)
e(Q-)=2-j e(Q)=2-k
:= CAi + CA2,
& O-(j'Ak)
* = E E E E -Tq 1 Q 1 -
j=k° Q'CP k>k° Q (2-(jAk) + \XQ'- Xq |)
e(Q')=2-J rn)=2-k , .
& 2-(jAk) v '
* E 2-K|j-k||Q'|-2--r-2-^Tq | Q | .
j=k° Q'CP k<ko Q (2-(jAk> + | XQ' Xq|)
)=2-J e(Q)=2-k
Then, A1 can be further decomposed as
& O-dAk)
*i = E E E E 2-K 1 d-k llQ'l72d^+2-^ TQ i Q i
j=k° Q'CP k>k° QC3P (2 (jAk) + | XQ' XQ |)
e(Q')=2-i e(Q)=2-k
K1 d-k iQ'i • n+1
j=ko QCP k>ko Qr3P=0 (2-(jAk) + \XQ - XQ^"+
e(o' )=2-J e(Q)=2-k
:= *ii + *12.
2-(dAk) (4.10)
EEH 2-K|d-k||Q'|-—--+tTq | Q|
There are 3n dyadic cubes in 3P with the same side length as P, so
E TqIQI < 3n sup E TqIQI. (411)
Qc3P P'C3P QCP' (4.t1)
e(Q)<e(P) e(P)=e(P) e(Q)<e(P')
& J-(jAk)
Ipi-rAn <cipiE E E 2-Kli-klQ'|72--r2-^ttqiQI
j=ko Q'CP k>k° QC3P (2 (jAk> + ^Q- XP |)
e(Q'),2-l t(Q)=2-k (4 12)
< Csup|P'|-r E E inQ№k * f (u)^IQI.
P' k=-log2£(P') QCP' ueQ
e(Q)=2-k
Next we decompose the set of dyadic cubes {Q : Q n 3P = C(Q) = C(P)} into {B{} according to the distance between each Q and P. Namely, for each i e N,
Bi := {P' : P'n 3P = 0, £(P) = £(P'), 2i-ko < ^p - yP| < 2i-k°+T},
(4.13)
where yQ denotes the center of Q. Then, we obtain
\p\-ra12 < \pVZ E E E 2-кjj-kj |q'\
2"(jAfc)
i=1 P'eBi j=ko Q'CP k>ko QCP'
e(Q')=2-i ¿(Q)=2-k
(4.14)
(2-(jAk) + \xp, - xpj)
Tq\Q\.
Since X Q-cP |Q'| = |P| for each j > k0 and | xp - xp| « 2i-k° for P' e Bj, the right-hand side of
e(o' )=2-j
(4.14) is dominated by
^E EjP\ 2(i-ko)(n+i) E ( E20
¿=1 P'eBi 2 k>ko\j=ko
-(jAk)-jk-jj
PT E Tq\Q\
¿(Q)=2-k
(4.15)
There are at most 2(t+2)n cubes in B¿, and hence
\P\-rA12 < C\
sup|P'rs E Tq\Q\
P' k>ko QCP'
e(Q)=2-k
oo 2-ko
e\P L.......2"
2(i-ko)(n+1) '
= Csup|P'|-r E E inQI£k */(«)T\Q\
k=-log2^(P') QCP'
i(Q)=2-;
(4.16)
To estimate A2, for i e N and k < ko, set
E,k := {Q : ¿(Q) = 2-k, xq e 2iP \ 2i-1p}.
(4.17)
Then, \xq - xp\« 2i-ko for Q e Ei,k and
a2 = eee e
2-K\j-k\\p \ 2-(jAk)
j=ko k<ko ¿=1 QeEi,k \QP (2-(jAk) + |xq - xp|)'
-\Q\-rTQ\Q\. (4.18)
Since, for Q e Ei/k,
\ Q \-rTQ \ Q \ < sup|P'|-r E E Tq'IQ'I (4.19)
pl m=-log2^(P') Q'CP'
¿(Q')=2-m
and the number of dyadic cubes contained in Ei/k is at most 2(i+k-k0)n,
| PI r A2 < C
sup|p'|-r E E tqQ
2-k(n+1)
P' m=-log2£(P) Q'CP'
e(o' )=2-m
X V V V2(ko-k)nr2K(k-j) _2_2(i+k-ko)n (4.20)
Z-i Zj Zj 2(i-k0)(n+1)
j=ko k<ko i=1
= C sup|p'r E E fm * / (M)I^IQ'I,
p' m=-log/(P') Q'CP' "e<Q
e(Ql)=2-m
where the condition K > 1 + nr is used in the last equality. Combining the estimates of A1 and A2, we prove Theorem 1.4. □
By modifying the proof above, we may easily show Theorem 1.5. Detailed verifications are left to the reader.
We now return to show Lemma 3.2.
Proof of Lemma 3.2. For r < 0, c^ = {0}, and hence the result holds. For r = 0, c^ = /a'q, and so the matrix is bounded by [2, Theorem 3.3]. To complete the proof, it suffices to show the boundedness of (a + nr, q, q)-almost diagonal matrices for the case r > 0.
We may assume that a = 0 since the case implies the general case. The proof is similar to the proof of Theorem 1.4. Here, we only outline the proof. First let us consider the case for q > 1. Let A = {aQP}q p be an (nr,q,q)-almost diagonal matrix. Then, for £(Q) = 2-k,
| (As)q| < C£ E 2(j-k)(nr+((n+^)/2))(1 + 2j|xq - xp|)~n-£ispi' jez e(p)=2-i
(I Q r1/2|(As)o|)q < C E E 2(j-k)(nr+(^2))(1 + 2j|xq - xp|p( I P|-1/2 I sp I )q
jez i(P)=2-j
(4.21)
due to Holder's inequality. Given a dyadic cube R with £(R) = 2 s,
EE (I Q I-1/2|(As)q|) q I Q I < CI + CII' (4.22)
k>5 QCR
e(Q)=2-'
I = E E E E 2(j-k)(nr+n+(^/2))(1 + 2j|xq - xp|)-n-^IPI-1/2 I sp i)qIPl
k>S QCR j>s £(P)=2-i
e(Q)=2-k
II =E E E E 2(j-k)(nr+n+(£/2)^1 + 2j | xq - xp|)-n-^IPI-1/2 I sp i)qIPI.
k>s QCR j<s e(p)=2-l e(Q)=2-k
Then, I can be further decomposed as
I = E E E E ^H™^))/! + 2j|xq -„|)"~(|P|-1/2 |Sp|V| P|
k>6 QCR j>6 PC3R
e(0)=2-k e(P)=2-i
+ E E E E 2(j-k)(nr+n+(e/2))(1 + 2j 1 Xq - xp|)-"-£(|p|-1/2|sp|)?|p| (4-24)
k>6 QCR j>6 Pn3R=0
l(Q)=2-k i(P)=2-J
:= I11 + Ill-
The same argument showed in the proof of Theorem 1.4 for the term A1 gives us
|R|-rI < C||s|| V (4.25)
To estimate II, for i e N and j < 6, let
Ehj := {Q : £(Q)= 2-j, xq e 2iR \ 2i-1^. (4.26)
Then, using the same argument as Theorem 1.4 for A2, we have
|R|-rII < C||Sy V (4.27)
Both estimates for I and II show the desired result for q > 1.
When q < 1, we modify the previous proof by replacing Holder's inequality with q-triangle inequality to get the result.
When q = to and r > 0, the space c^ = f^, and hence an (a + nr, to, w)-almost diagonal matrix is bounded on c^ by Proposition 5.3. □
Remark 4.1. Note that caq = /TO'q. By a duality argument and [2, Theorem 3.3 and page 81], one can show that the (a + n, q, q)-almost diagonal matrix is bounded on /TO,q. When q > 1 and r > 1, we can prove Lemma 3.2 by duality in Theorem 2.2. Let A = {aQP}qp be an (nr,q, q)-almost diagonal matrix. Also define the transpose of A by A' = {aPQ}Q P. For q > 1 and r > 1, let p = (q + q')/(q'r + q). Then, p < 1. Since A is (nr,q,q)-almost diagonal, A' is (0,p, q')-almost diagonal by a calculation for a different value of e. Thus, by Theorem 2.2 (a) and Proposition 5.3, A' is bounded on c^.
5. Applications
We define another wavelet multiplier on Rn by using ^-transform identity as follows. Let (p and f in S satisfy (1.2) and (3.1). For a sequence t = {tQ}q, where the Q's are dyadic cubes in Rn, define the wavelet multiplier Tt by
t.(/) = e \Q\-1/2tQ(f,yQWQ Q
for f e S/P such that the above summation is well defined. Thus, we have the following characterization.
Theorem 5.1. Suppose that a, ft e R, 0 <p < 1, and 0 <q < <x>. Then,
(a) for 1 < q< œ, Tt is bounded from Fpp,q into f^'1 if t g /y/y
(b) for 0 < q < 1 and r e R, Tt is bounded from Fp,'q into Fa+5,1 if t e c5+(n/p) n,œ.
Proof. We show the case a = 0 only, which implies the general case by (2.7). For ¡5 e R, 0 < p < 1, and 1 < q < œ, let f e FP'q and t e CqVpHqVq). It follows from Theorem 2.2 and Proposition 3.1 that
Tt(/)||ffi < C {|Q|-1/2fQ(f,yq)}q
CZ(\Qrß/n\tQ\)\{f,VQ )\ c||{(/,^q)}q||,o„ {\Q\-ß/ntQ}c
'(qVpHqVq)
< C||f Htf<
q ll'IL M
(q'/p)-(q'/q)
This shows that Tt is bounded from F^ into Ff1 and \\Tt\\ < Cyty^q yields the boundedness of Tt for the case 0 < q < 1.
(qVpHq'/q)
. A similar argument □
In order to prove Theorem 1.12, we demonstrate a similar result in sequence spaces first. For a sequence t = [tQjg, define Dt by
Dt(s) = j|Q| 1/2iQSQ} for s = ^sJwith finitely many nonzero terms. (5.3)
Theorem 5.2. Suppose that a, 5 e R, 0 <p < 1, and 0 <q < œ. Then,
(a) for 1 < q < <x>, Dt is extendible to be bounded from fp,q into f^^ if and only if t e rM
C(q'/p)-(q'/q),
(b) for 0 < q < 1 and r e R, Dt is extendible to be bounded from fp,q into f^^"1 if and only if
t e cP+(n/p)-n"x'
Proof. We still assume that a = 0. For p e R, 0 <p < 1, and 1 < q < to, let s = {sq}q e fpq and
t = {tQ }q e C
M (q'/pMq'/q)
. It follows from Theorem 2.2 that
IIA(s% = £(\Qrß/n\tQ\)\sQ\
< C||s|U,
{\Q\-ß/ntQ}i
W /o)~W/q)
= C||s|U ||t|U,'
(qVpHqVq)
Conversely, suppose that Dt maps from f^ into /f'1 boundedly. For t = {tQ}Q, let T = {|Q|-f/ntQ}q. Define a linear functional by
^t(s) = E sQtQ for s = ^sJwith finitely many nonzero terms.
(s)\ < E(\Q\-ß/n\tQ\)\sQ\ = ||Dt(s)|/f!. (5.6)
The assumption shows that is a continuous linear functional on fpq. Using Theorem 2.2,
we have t e clq'/ph(ql/q),and hence t e 4ipHqVq).
For 0 < q < 1, a similar argument gives the desired result of (b). □
Proof of Theorem 1.12. The "if" part follows from Theorem 5.1. To show the "only if" part, define Xi by
Tif ) = E IQr1/2fQfQ>Q. (5.7)
The boundedness of Tt says that T is bounded from pp,q into P'a+p' . Clearly,
Sfi ◦ T ◦ Tfi (s) = Dt(s) for s e f^. (5.8)
It follows from Proposition 3.1 that Dt is bounded from f^ into f^'1, and hence t e c|qq/p)_(q,/q) for 1 < q < to and t e C+(n/p)-n'TO for 0 < q < 1 and r e R by Theorem 5.2. □
In order to study the boundedness of the paraproduct operators acting on Triebel-Lizorkin spaces, we need more results described as follows.
Proposition 5.3 ([2, pages 54 and 81]). For a e R and 0 <p,q <to, an (a,p,q)-almost diagonal matrix is bounded on fp^'q.
Lemma 5.4. Define a matrix by G = {(fp,Oq)}q,p. Then, for a < 0 and 0 < p,q < +<x>, G is (a,p,q)-almost diagonal and hence is bounded on fp'.
Proof. For J(p) < J(Q), since J xYfP(x)dx = 0 for all j, by [2, page 150, Lemma B.1], we have
f O W< c(e(Q) \Y 1 + |XQ - xpI
^OQ)| <c{m) (1 + QQ)
J(P) N ((n+e)/2)+J-n
for s > 0 and a < J - n + (s/2), where J = n/min{1,p,q} and C is independent of P and Q. For £(Q) < £(P), by [2, page 152, Lemma B.2], we obtain
Kfp, Oq> | < c 1 +
1xq — xp1
-J—e
J(Q)\n/2 J(P)/
= cffl) X1
1xq —xp1
J (P )
-J—e
J(Q)\ J(P )/
(n—2a)/2
(5.10)
Choosing s = -2a, we obtain the result. We now can prove Theorem 1.13.
Proof of Theorem 1.13. To simplify notations, let q0 = qr / (q - r) and (1/p0) = (1/p) - (1/q) + (1/q0) . The requirement p < r < q < r/(1 - r) guarantees that p0 < 1 < q0. Now assume that g e CMO'
ß'0 (q0/p0) — (q0/q'0)
and f e Fp . To prove part (i), by (3.1) we rewrite ng(f) as
n(f) = X<^,^q>IQI—1l/2(Xf,yp)fp,®Q/fQ
= Z(g,fQ )IQ|—1/2(Gs)QfQ, Q
(5.11)
where s = {(f,yp )}p. Proposition 3.1 and Theorem 2.2 give
||n(f)11Fa+ß- < c {IQI—1/2 <g,fQ> (Gs)Q}
¿<x.a+,r
= C|( | Q |—(ß/n)—(1/2)+(1/2r)KgtVQ)|)r.( | Q |
—(a/n)—(1/2) + (1/2r)
|(gs)q|) r
{(|QI—(ß/n)—(1/2)+(1/2r)K 0r}
(a/n)—(1/2)+(1/2r)|(Gs)Q|)r}
,<Wr)'
~(('/r)'/(p/r))—(('/r)'/(q/r))
It is clear that
(ß/n)-(1/2) + (1/2r)
\(g><PQ)\) 1
o,(q/r)'
'((q/r)'/(0/r))-((q/r)'/(q/r))
= SuH \P\-r(q/r)'((i/p)-(i/q)) p l jpqcp
\ E (\Q\-(ß/n)-(1/2)\(g,PQ)\XQ(x))r{q/rYdx
Jpqcp x 7
S'Pq)}
qwj-Io
(qo/po) -(qo/p'0)
1/(q/r)'
(5.13)
-(a/n)-(1/2)+(1/2r)
(Gs)^r}
(a/n)-(1/2)
j o,(q/r)
\(gs)q\) XQ(x)
Hence, by Propositions 3.1 and 3.3, and Lemma 5.4,
(5.14)
lln (f ^ < c||i (Z'Pq)}q
IC10 ■■ "fp
(qo/oo)-(qo/q0)
||Gs||f;
< c||gllcMo^-qo , ||s||f;,q
(qo/po)-(qo/qo' )
< c||gwcMoßqo Wf WF;q.
(qofoo)-(qofqo ) 0
(5.15)
Next suppose that ng is bounded from Pp^ into P<a+fS,r. Without lost of generality, we may assume that a = 0. A computation yields
-qo((1/po) + (1/qo) -1)
IPQÇP = \p\-(1/p) + (1/q)
|pE(\Q\-(ß/n)-(1/2)\(g,PQ)\XQ (x))qo dx
|pE(\Q\-(ß/n)-(1/2)\(g,PQ)\XQ (x)) qr/iq~r)dx
f E ( \ Q \ -
(q-r) / qr
< c \ P\ -(1/^p E ( \ Q \ -{ß/nh(1/2) Kg,pq) \xq(x))rdx
Fix an integer N > (n/p) - n. Choose a function 0 e S(Rn) satisfying d(x) = 1 on [0,1]n, 0(x) = 0 if x / 3[0,1]n and JxY0(x)dx = 0 for all multi-indices y with |y| < N. By the
molecular theory [2, page 56], it follows that 0 e pp,q. For each dyadic cube P, define 0P by
0P(x) = 0
x - xP
(5.17)
Then, (0P,Oq) = j®Q(x)dx = |Q|1/2 for all dyadic cubes Q C P and ||0PHp* = C|P|1/p by the
translation invariance and the dilation properties of FP'q. By Proposition 3.1,
k (0P )W * ß,
{(g, Pq )\Q\-1/2( 0P, ®q)}
> ' ( \ Q \ -(ß/n)-(1/2)Kg,PQ)\XQ(x))rdx
(5.18)
and hence, by the boundedness of ng,
\ P \ -qo((1/po)+(1/qo)-1^ £ ^ \ Q \ -(ß/n)-(1/2)\(g,pQ )\XQ(x^ qo dx
(5.19)
Taking the supremum on P, we show that g e CMO'
(qo/po)-(qo/qo ).
To prove part (ii), assume that g e CMO' and s = {( f, fQ)}q. By Proposition 3.1,
(qo/po)-(qo/qo )
and f e F;,q. let t = {(g/ Pq >}q
E\ P \-1/2(g/PP )(®P/PQ )(f/fP )
= WGDts
>a+ß,r
(5.2o)
where (5 := {(®p,^q)}qp is the transpose of {(yP,®q)}q,p. Since a + ¡¡> 0, by Lemma 5.4, (5
is (a + ¡¡, r, r)-almost diagonal and hence is bounded on fa+5'r. Following the same argument as the proof of part (i), we get
|ng(f)||F«+p, < C||Dts||f;+p,
= C£(\ Q \ -(ß/n)-(1/2)+(1/2r)Kg,pQ)\)r.( \ Q \ -
< Cllg|lCMoß'
(;/n)-(1/2) + (1/2r)
\(f/fQ )\) '
(5.21)
(qo/po)-(qo/qo)
which completes the proof.
Acknowledgments
The authors are grateful to the referees for many invaluable suggestions. Research by both authors was supported by NSC of Taiwan under Grant nos. NSC 100-2115-M-008-002-MY3 and NSC 100-2115-M-259-001, respectively.
References
C. Fefferman and E. M. Stein, "Hp spaces of several variables," Acta Mathematical, vol. 129, no. 3-4, pp. 137-193,1972.
M. Frazier and B. Jawerth, "A discrete transform and decompositions of distribution spaces," Journal of Functional Analysis, vol. 93, no. 1, pp. 34-170,1990.
H. Triebel, Theory of Function Spaces, vol. 78 of Monographs in Mathematics, Birkhauser, Basel, Switzerland, 1983.
I. E. Verbitsky, "Imbedding and multiplier theorems for discrete Littlewood-Paley spaces," Pacific Journal of Mathematics, vol. 176, no. 2, pp. 529-556,1996.
D. Yang and W. Yuan, "A new class of function spaces connecting Triebel-Lizorkin spaces and Q spaces," Journal of Functional Analysis, vol. 255, no. 10, pp. 2760-2809, 2008.
D. Yang and W. Yuan, "New Besov-type spaces and Triebel-Lizorkin-type spaces including Q spaces," Mathematische Zeitschrift, vol. 265, no. 2, pp. 451-480, 2010.
C.-C. Lin and K. Wang, "Equivalency between the generalized Carleson measure spaces and Triebel-Lizorkin-type spaces," Taiwanese Journal of Mathematics, vol. 15, no. 2, pp. 919-926, 2011.
D. Yang and W. Yuan, "Relations among Besov-type spaces, Triebel-Lizorkin-type spaces and generalized Carleson measure spaces," Applicable Analysis. In press.
[9] M. Plancherel and G. Polya, "Function entires et integrales de Fourier multiples," Commentarii Mathematici Helvetici, vol. 9, pp. 224-248,1937.
10] Y. S. Han, "Calderon-type reproducing formula and the Tb theorem," Revista Matematica Iberoamericana, vol. 10, no. 1, pp. 51-91,1994.
11] Y.-S. Han, "Plancherel-Polya type inequality on spaces of homogeneous type and its applications," Proceedings of the American Mathematical Society, vol. 126, no. 11, pp. 3315-3327,1998.
12] C.-C. Lin and K. Wang, "Triebel-Lizorkin spaces of para-accretive type and a Tb theorem," Journal of Geometric Analysis, vol. 19, no. 3, pp. 667-694, 2009.
13] C.-C. Lin and K. Wang, "The T1 theorem for Besov spaces," preprint.
14] Y. Han, J. Li, and G. Lu, "Duality of multiparameter Hardy spaces Hp on spaces of homogeneous type," Annali della Scuola Normale Superiore di Pisa. Classe di Scienze. Serie V, vol. 9, no. 4, pp. 645-685, 2010.
15] N. H. Katz and M. C. Pereyra, "Haar multipliers, paraproducts, and weighted inequalities," in Analysis of Divergence, Appl. Numer. Harmon. Anal., pp. 145-170, Birkhauser, Boston, Mass, USA, 1999.
16] F. Nazarov, S. Treil, and A. Volberg, "The Bellman functions and two-weight inequalities for Haar multipliers," Journal of the American Mathematical Society, vol. 12, no. 4, pp. 909-928,1999.
17] P. G. Lemarie and Y. Meyer, "Ondelettes et bases hilbertiennes," Revista Matematica Iberoamericana, vol. 2, no. 1-2, pp. 1-18,1986.
18] Y. Meyer, Wavelets and Operators, vol. 37 of Cambridge Studies in Advanced Mathematics, Cambridge University Press, Cambridge, UK, 1992.
19] G. David and J.-L. Journe, "A boundedness criterion for generalized Calderon-Zygmund operators," Annals of Mathematics, vol. 120, no. 2, pp. 371-397,1984.
20] A. Youssfi, "Regularity properties of commutators and BMO-Triebel-Lizorkin spaces," Universite de Grenoble. Annales de l'Institut Fourier, vol. 45, no. 3, pp. 795-807,1995.
Copyright of Abstract & Applied Analysis is the property of Hindawi Publishing Corporation and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
