SIAM J. MATH. ANAL.

()1993 Society for Industrial and Applied Mathematics

Vol. 24, No. 1, pp. 263-277, January 1993 017

INEQUALITIES OF LITTLEWOOD-PALEY TYPE FOR FRAMES

Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

AND WAVELETS*
CHARLES K. CHUI AND XIANLIANG SHI$
Abstract. Inequalities of Littlewood-Paley type for frames in both the wavelet and Weyl-
Heisenberg settings, and those for any unconditional basis of the form Cj,k(x) 2(2Jx-
k), are
established. In particular, if {j,k } is a semi-orthogonal basis, then the Littlewood-Paley identity is
obtained. A similar identity for the "biorthogonal wavelets" of Cohen, Daubechies, and Feauveau is
also obtained.

Key words. Littlewood-Paley inequalities, frames, frame bounds, wavelets, biorthogonal

wavelets

AMS(MOS) subject classifications. 41A17, 42C15

1. Introduction and results. The objective of this paper is to establish cer-

tain inequalities and identities of Littlewood-Paley type and to discuss some of their
important consequences. This section is devoted to introducing the necessary defini-

_ _
tions and notation and to a discussion of the main results in this paper. For the sake
of clarity, it will be divided into two subsections: with the first one on frames and the
second one on wavelets.
1.1. Inequalities for frames. The notion of flames was introduced by Duffin
and Schaeffer [6] in their work on nonharmonic analysis. For the Hilbert space L 2 :-
L2(-c, c), a family of functions Ck E L 2, k E Z, is said to be a frame of L 2 if there
exist two positive constants, C1 and C2, with 0 < C1 _< C2 < cx), such that

(i.1) Clllfll [(f,k)l : C211fll

kZ

for all f e L 2. Here and throughout, II/11 denotes the L2-norm of f. A frame
is called a tight frame, if C1 C2 (cf. [5], [6]). Note that even a tight frame with
C1 C2 1 is not necessarily a basis of L 2. For instance, if {Yk}, k Z, is an
orthonormal basis of L 2 and {ak}, k Z, is any sequence of real numbers, then the
family {T]k COStk, r/k sinak}, k Z, which is certainly not a basis of L 2, is, however, a
tight frame of L 2 with C1 C2 1. Observe that {ak} may be so chosen that every
function in this nonbasis tight frame is nontrivial.
In this paper, we will only consider frames that are generated by a single function.
Two types of such frames are of particular interest:
(i) s.t. frames (or frames generated by scaling and translation of a function
E L 2) defined by:

(1.2) (Sj,k)(x) :-- a1/2(aJx kb), j,k e Z,

*Received by the editors May 28, 1991; accepted for publication (in revised form) April 21,
1992. This research was supported by National Science Foundation grants DMS-89-01345 and INT-
87-12424.
Center for Approximation Theory, Department of Mathematics, Texas AzM University, Col-
lege Station, Texas 77843-3368.
:Permanent address, Department of Mathematics, Hangzhou University, Hangzhou, Zhejiang,
310028, People’s Republic of China.
263
 264

_ CHARLES K. CHUI AND XIANLIANG SHI

where a > 1 and b > 0 are (fixed) constants, and

(ii) w.h. frames (or frames of Weyl-Heisenberg type, generated by a function
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

E L 2) defined by

(1.3) (Hj,k)(x) := eiJPx(x- kq), j,k e Z,

where p, q > 0 and pq 2r.

1.1.1. s.t. frames. Let us first study s.t. frames. For any function b L2, set
(1.4) Cj (x) "= aJ (-aJ x), je Z,
where again a > 1 is fixed. Then we have the so-called "semidiscrete integral wavelet
transform""

(1.5) (Wjf)(x) (f Cj)(x), je Z, f e L 2,

where denotes the integral convolution on (-(x), oc) (cf. [8], [9]). For this transform
to have any practical value, it must be "stable"; and by this, we mean the existence
of constants A and B, with 0 < A < B < oc, such that

-
(1.6) AIIfll 2 < IIWjfll 2 < BIIfll 2, f e L2.

On the other hand, in signal analysis, since has the property of a bandpass filter,
in order to be able to reconstruct the original signal from its wavelet transform (1.5),
the function must satisfy

(1.7) A I(a)[2 B a.e.,

jz

where is the Fourier transform of , defined by

(w) e-ix.’(x)dx.

In fact, it is easy to see that (1.6) and (1.7) are equivalent, with the same constants
A and B.
To generate an s.t. frame, we further discretize the transform Wjf in (1.5) by
defining Sj,k as in (1.2), using another parameter b > 0. Then analogous to the
stability condition (1.6) for semidiscrete integral wavelet transforms, we require

(1.8) Cy,k(X) :-- (Sj,k’g,)(x) a1/2"d2(aJx- kb)

to satisfy the frame condition:

(1.9) A’llfll 2 < I(f, Cy,)l e -< B’llfll 2, f e L 2,

j,kZ

where 0 < A < B < cx. Our result on s.t. frames is then an analogue of (1.7), as
follows.
 LITTLEWOOD--PALEY INEQUALITIES FOR FRAMES AND WAVELETS 265

-
THEOREM 1. Let {j,k }, as defined in (1.8) .for some a > 1 and b > O, be a frame
of L with frame bounds A’ and B’ as in (1.9). Then satisfies
2
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

1
(1.10) A’ <_ l(aw [2 <_ a.e.,

.for the same constants A and B .

From (1.10), we can easily derive other interesting inequalities. For instance, by
integrating each term in

over 1 <_ [w[ _< a, we have

2A’ log a dw <_ 2B’ log a,

which immediately yields

(1.11) A’ _<
1
2bloga J_ ](w)]
oc
2
dw <_ B’.

We remark that the so-called "compactibility condition" (1.11) for s.t. frames was also
derived by Daubechies [5] by using techniques from trace-class operators. In addition,
Daubechies [5] also observed that under the assumptions of Theorem 1, the quantity

is bounded from above and below by some positive constants. The contribution in
Theorem 1 is that these constants are given by the frame bounds.
1.1.2. w.h. frames. For a function E L 2, consider the semidiscrete window
Fourier transform (also known as short-time Fourier transform)"

(1.12) (Fyf)(x) e-iiPt(t x)f(t)dt, f e L 2,

where p > 0 is a fixed constant. Then analogous to (1.6), the stability of the transform
(1.12) is defined by the requirement"

(1.13) CIIfll 2 [IFyfll 2 Dllfll 2, f e L 2,

jz

for some 0 < C _< D < oc, independent of f. It is not difficult to see that (1.13) is
equivalent to

(1.14) C <_ I(w jp)l 2 <_ D a.e.,

jz
 266 CHARLES K. CHUI AND XIANLIANG SHI

for the same constants C and D. Further discretization of the transform Fj f in (1.12)
results in introducing
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

(1.15) J,k(X) := (Hj,k)(x) eiJPX(x- kq),

as defined by

(1.16)

(1.17) C’Ilfll
_ _
(1.3), where q > 0 is the second discretization constant satisfying

(cf. [5] for the requirement of (1.16)). For (d,k}

0 < C’ _< D’ < x), such that

j,kez
0

I<f, ,>1
< pq <_ 27r
to be a frame, we need constants

D’llfll , f e L2.

Our result on w.h. flames in this paper is the following inequalities, which are along
the same line of (1.14).
THEOREM 2. Let (j,k}, as defined in (1.15) .for some p, q > 0 satisfying (1.16),
be a frame of 5 2 with frame bounds C and D’ as in (1.17). Then and satisfy

1
(1.18) C’ <_
q jz
i(w_ jp)l 2 <_ D’, and

(.) c,< :--P I(- J)l < D, ..,

jez

for the same constants C’ and D’.

1.2. Results on wavelets. In the following, we will set a 2 and b 1 in the
definition of i,k in (1.8); that is, we will consider

(1.20) j,k(x) 2z(2Jx- k), j,k e Z.

Furthermore, we will also assume that {j,k } in (1.20) is an unconditional basis of L 2

with constants 0 < K _< L < oc, namely: it is complete and satisfies

(1.21) K laj,l 2 <

j,kZ

for all {aj,k} E 12(Z2). However, this assumption does not guarantee that the dual
basis {,k }, relative to {j,k }, defined by

(1.22)

same manner as {i,k} from .

is obtained by dyadic dilations and integral translations of a single function in the
We will give an elementary proof of this somewhat
surprising result in 3. If it so happens that

(1.23) ;,k CJ,k, j, k e Z,

 LITTLEWOOD-PALEY INEQUALITIES FOR FRAMES AND WAVELETS 267

where

.,
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

(1.24)
for some E L 2, then we will call the "dual" of Since {j,k} is clearly an

then is self-dual with .

unconditional basis of L 2, it follows that is the dual of also. Observe that if
is an "orthonormal wavelet" in the sense that {j,k} is an orthonormal basis of L 2,

As a consequence of Theorem 1, we have the following inequalities of Littlewood-

Paley type.
COROLLARY 3. Let {j,k}, as defined in (1.20), be an unconditional basis of L2,
with bounds K and L as given in (1.21). Then the Fourier transform of satisfies

(1.25) K _< l(2Jw)l 2 <_ L i.e.,

_ _
and consequently,

I(w)l
(1.26) K<
1
21og2 /_ Iw----V 2
-dw -< L.
Furthermore, if is a wavelet with dual as defined in (1.22)-(1.24), then {j,k} is
also an unconditional basis of L 2 with bounds L-1 and K -I, and consequently,

(1.27) 5 -1 1/(2J0)l 2 K -1 i.e.,

and

(1.28) L_ <
1
21og2 /_ I(w)l 2 dw
iw
K -1

There are two special cases that are of particular importance. We will discuss
them separately.

,
1.2.1. Semi-orthogonal wavelets. Let {y,k} be an unconditional basis of L 2
generated by some function as governed by (1.20) and (1.21). For each j E Z, set

(1.29) Wi closL2 span{/,k: k Z}.

Then we say that is a "semi-orthogonal wavelet" if

(1.30) Wy_I_W, j,gZ, j.

The dual of a semi-orthogonal wavelet is easily obtained via the Fourier transform,
namely

(1.31)
=
(w)-
(w),
 268 CHARLES K. CHUI AND XIANLIANG SHI

where
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

By giving up the orthogonality of (0,k: k E Z} it is possible to construct compactly

supported with certain desirable properties. For instance, the compactly supported
spline-wavelets of Chui and Wang in [1] are symmetric for splines of even order and
anti-symmetric for splines of order. In addition, explicit formulas of compactly sup-

.
ported semi-orthogonal spline wavelets and their duals were given in Chui and Wang
[1], and a characterization of all the compactly supported ones in [2]. For this type of
wavelets, we have the following result.
THEOREM 4. Let be a semi-orthogonal wavelet with dual Then

(1.33) E (2Jw)(2Jw)= 1 a.e.,

and consequently,

(w) (w) dw
(1.34) 2 log 2.

In particular, if
transform satisfies
is.an orthonormal wavelet so that , then its Fourier

(1.35) E 1(2iw)12 1 a.e.,

and

C’= o Iwl
dw=21og2.

1.2.2. Nonorthogonal wavelets. Wavelets without any orthogonality struc-

ture have even more flexibility. For instance, examples of those and their duals in
Cn(-oc, oc), where n E Z+ is arbitrary, both with symmetry and compact support,
have been constructed recently by Cohen, Daubechies, and Feauveau [4]. Following
Cohen [3], we let too(w) and 0(w) be two 27r-periodic Lip(a) continuous functions,
0 < c < 1, satisfying

mo(w)Cno(w) + rno(w + )o(w + ) 1;

(1.37)
rno(0) o(0) 1; mo() ,= o(r) 1.

Suppose that and , defined by

w) m0
j=l
(1.38)
)= o N
 LITTLEWOOD--PALEY INEQUALITIES FOR FRAMES AND WAVELETS 269

satisfy
C
I(x) + 14(z)l <_
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

for some e, C > 0. Then is a wavelet with dual , where

(1.39)

in the sense that {j,k} and {j,k}, defined by (1.20) and (1.24), are both uncondi-
tional bases of L 2, such that

(cf. Cohen [3]). Although this is not semi-orthogonal, its Fourier transform still
satisfies the same Littlewood-Paley identity, as follows.
THEOREM 5. Let be a wavelet with dual as defined by (1.37)-(1.39). Then

(1.40) E (2Jw) (2Jw)- 1, wE R,

and consequently,

(1.41) /_ (w) (w) dw 2 log 2.

Remark. For an orthonormal wavelet ,

identity (1.35) is called the Littlewood-
Paley identity. Hence, the results in (1.33) and (1.40) may be called Littlewood-Paley
identities also. The significance of these identities is that when is considered as a
bandpass filter, so that {y,k} gives rise to a filter bank decomposition, the wavelet
can be used for perfect reconstruction. In other words, the pair (, ) constitutes
an allpass filter as demonstrated by (1.33) and (1.40). We also remark that the value
of C in (1.36) is needed in the reconstruction formula (from the integral wavelet
transform) of Grossmann and Morlet.
2. Proofs. In this section, we establish all the results stated in 1.
2.1. Proof of Theorem 1. By (1.2), we have, for any f E L 2,

(f, Cj,k) a f(x)(aJx kb)dx

1 1/2 f(ow)(w)eibodw
Hence, by setting
271
(2.1) T
b’
 270 CHARLES K. CHUI AND XIANLIANG SHI

we have

-.
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

j,keZ

a)T2
.f(cd (w + gT))(w + iT) e

E .f(aJ (w / eT))(w + eT) dw.

so that for any

Now, consider ]
inequality becomes
M
--
Therefore, it follows from (1.9) and (2.2) that

A’]I]I[ 2 <_

aJ a-:iwO+
fJa-Jo-
EEz/(aJ(w + T))(w + t.T)l
M > 0, M E Z, and wo E (-cx, c), we have

E ](aJ (w + eT))(w +
(1/x/)Xio ....o+,1,
d B,

> 0. Then for sufficiently small

II]II 2,

, the above

M a-j(wq-)
aJ
bf
--_,(o_)
l()12d < B’,

and thus, by taking

(2.4)
0 and M
- cx consecutively, we have

On the other hand, for any wo, r/> 0, a positive integer M may be chosen so that
X)

aMwo(l+a) -1

Also, for

a+ lW0, T}
0<<min{a-1
the function ] (1/x/)Xio o+1 satisfies

](aJ (w + eT)) 0
 LITTLEWOOD--PALEY INEQUALITIES FOR FRAMES AND WAVELETS 271

for all e Z with I1 _> (e/aJT) + 1 and all w e [a-Jwo (T/2), a-Jwo + (T/2)]. Hence,
for this f, we have
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

_M

j=-
aj
ia_Jwo+
-o-
](( + T))5( + T)
--M

(e.) =- a-o
fa-wo+
I( + tT)lo ....o+(( + tT)) +

.
Since e
C
i=_
[
-a-i(w+)

< (a- 1/a + 1)wo, the intervals

are mutually disjoint;

(o-)

[a-J (wo e), a-J (coo + )],

and hence, by (2.5), we have
I()1 = + I()1 =

j6Z,

j----O0
f-’<<"+’:
d a-J (WO--e)
I(<,.,)ld<,., _<
L aMwo(1-Fa) -1
I(o)1 <

so that it follows from (2.6) that

2
T
aJ
E/(aJ (w + eT))(w +
2

j=-M+I
E ](a (w + eT))(w +
t6Z
-M
+C+
C
[wo+e =-o

-
o-

Therefore, by (2.3) and (2.7), we have

oo 2
T
(2.8) I :=
j=--M+I
E aJ
L E
w+e -M
_> A’ C
C
I
.o- =_
I(-)1.
On the other hand, for all sufficiently small e > 0, it is clear that

i=
j=-M+I
a-o-)I/()()1
1
2b o-e
I(-)1,
j=-M+I
 272 CHARLES K. CHUI AND XIANLIANG SHI

where / (1/x/)Xio
we may take
- ....
o+1" Hence, in
0 in (2.8) to arrive at
view of the boundedness property in (2.4),
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

o -U
1
(.9) [(a-Jwo)12 > A C/- C [(a-Jwo)[ 2
j=--M+I

for almost 11 w0 > 0. Since y > 0 is arbitrary, (2.4) and (2.9) together yield
(e 0) (-) > A,
for almost all w > 0. A similar argument holds for w < 0. Henee, by (2.4) and (2.10),
we have completed the proof of Theorem 1.

2.2. Proof of Theorem 2. Instead of (2.1), we now consider T 27r/q. Then

it follows from (1.15) that, for any f E L 2,
2
1
ei(-pl((w jp)f(co)da
,kZ ,keZ

:4.2 j _(w + T jp)f(w + T)_ ei

r

ez 12
function ]
=4r2

that (2.11) becomes

T21
"
As in the proof of Theorem 1,
(1/)X[o ....

C’ <
1
(w + IT jp)](w + gT) dw.

we fix any w0 and 0 < e < T/2, and consider the

o+]" Then it follows from the frame bounds in (1.17)

[wo+s 15(o
oo-
1
jp)ldo <_ D,,

and this implies (1.18).

To derive (1.19), we set T 27r/p and note that
T
I(f, ’)1 T, f(x + eT’)(x + eT’- kq)e’dx
j,kEZ j,kZ g.EZ

T’2 E7
kZ
1
oE T’

Z
f(x + eT)(x + eT- kq) dx.

By the same proof as above, we also have (1.19). This completes the proof of the
2

theorem.
2.3. Proof of Corollary 3. Let (,k} be the dual basis of L 2 relative to (5,k }.

(2.12)
*
L-1 E laj’kl2
j,kZ

-
Then (j,k) is also an unconditional basis of L 2 and it follows from (1.21) that

j,kZ j,kZ
 LITTLEWOOD--PALEY INEQUALITIES FOR FRAMES AND WAVELETS 273

for all {a.,k} e g2(Z2). Now, for any f e L 2, writing

_
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

S(x)- E a.i,k,k (x)’

we have

(2.12) yields

-
so that

](f,y,k)J 2 ]lf[] 2 K I(f,Y,k)[ 2

j,kz

which is equivalent to

(2.13) K]]f]]2 <- E ]<f’J,k>]2 -< LI]f]]2

j,kZ

for all f E L 2. Hence, (1.25) follows from Theorem 1with a 2 and b- 1. In

Cj,k, where is the dual of . .

addition, (1.26) follows from (1.11) for these specific values of a and b. Of course,
(1.27) and (1.28) are now the analogous consequences of (2.12) with ,k replaced by

2.4. Proof of Theorem 4. Before we go ahead to establish the theorem, let us

first say something about the dual of In the first place, we note that (0,k)
((.- k)} is an unconditional basis of W0, meaning, of course, that it is complete in
W0 and satisfies

kZ

for all {bk) E g2, where 0 < K _< L < cx are as in (1.21). Note that (2.14) is a
consequence of (1.21) with the same bounds. It is well known (cf. [7]) that (2.14) is
equivalent to

(2.15) K _< @(w) _< L a.e.

for the same constants K and L, where (w) was defined in (1.32). Hence, we have

jz

for some {cj } g2, so that

(x) E cj(x j)

is in W0. Secondly, from the definition of in (1.31), we see that

(2.16) E (w + 2rk)(w + 2rk) 1 a.e.,

 274 CHARLES K. CHUI AND XIANLIANG SHI

and by a standard argument, it can be shown that (2.16) is equivalent to

Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

(2.17) <(. j), (. k)> 5j,k, j, k e Z.

In the proof of Theorem 4, we orthonormalize {j,k } as usual by defining a func-
tion y, with

(2.18)
()1/2"
In view of (2.15), we have r/E W0, and

j,k Z,
constitute an orthonormal basis of L2. Hence,

]If, rlj,k)] 2 --]lf]], f e L 2,

j,k

and by Corollary 3 with K L 1, we have

(2.19) [(2Jw)12 1

so that (1.33) follows from (1.31), (1.32), (2.18), and (2.19). Furthermore, (1.34) is
a consequence of (1.33) by following the same method of derivation of (1.11). This
completes the proof of the theorem.

2.5. Proof of Theorem 5. By (1.39), we have

(2w) (2ca) m0(ca -4- r)ff0(w + r)(w) (w),

and hence it follows from (1.37) and (1.38) that

5() (): (- .0()0())5() ()

() 5() (:) 5(:).
Under the assumption on the decay property of and ,
we note that and are
both continuous at zero and converge uniformly to zero as Iwl oc. Therefore, by
telescoping, we have, from (2.20),

()(): (0)(0):

for all w, where the assumption m0(0) 0(0) 1 in (1.37) is used. This establishes
(1.40). Since (1.41) follows from (1.40) as in the derivation of (1.11), we have completed
the proof of the theorem.
 LITTLEWOOD--PALEY INEQUALITIES FOR FRAMES AND WAVELETS 275

3. Final remarks. In this paper we have established various inequalities and

identities of Littlewood-Paley type, among which are those for the following three
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

classes of functions:
(a) The collection .T" of all s.t. frames;
(b) The collection :R of all such that {j,k}, as defined in (1.20), is an uncon-
ditional basis of L2;
(c) The subcollection l/Y of functions in n that have duals as defined by (1.22)-
It was shown in 2.3 that 7 C ’, and, in fact, the bounds for the unconditional
basis {j,k} generated by any e n in (1.21) remain to be the frame bounds for the
frame {j,k}. Hence, we have

(3.1)
Observe that even for any E
)/Y C

’, the result
C .
from (1.11) already yields

(x)dx (0) 0,

of ,
provided that is continuous at zero, which follows from a very weak growth condition

,
such as E Ll(-oc, cx)). Hence, it is already reasonable to call any e
"wavelet." If, in addition, e )4; with dual "
a
then by using the notion of "integral
wavelet transform" as defined in (1.5), namely,

(W.f)(x) 2 /_ f(y)(2Jy- 2x)dy,

the coefficients

(3.2) aj,k 2 (Wj f) --]

of the "wavelet series" expansion

S(x)
j,kEZ

of any f L 2, contain very important information of the "signal" f in time-frequency

analysis.
Regarding (3.1), while it is obvious that T is a proper subset of ’, it is not
immediately clear that }/V is properly contained in 7. In the following, we give a very
simple proof of this fact.
Let 52 be such that {,Dj,k }, as defined in (1.20), is any orthonormal basis of
L 2, and consider the function
(x) :- (x) 221/2(2x),
 276 CHARLES K. CHUI AND XIANLIANG SHI

where z is any complex parameter with Iz[ < 1. Then it is clear that E T. In fact,
we have
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

(1-Izl) E laj,kl2 <-

j,kEZ
aj,k r]j,k _< (1 + Izl) 2 E laj,kl2
j,kEZ

for any {aj,k} e e2(Z2). Let {,k} be the basis dual to {rh,k }. Then it is easy to
verify that

Vo,o(X) zt-t,o(z),
(3.4) =0

?,1 (X) 0,1(X).

Hence, for r to be in l/Y, the family {,k} must be given by
o,0 (2x k),
and in particular,
(X + 1) v0,0(*),
(3.4) yields

-
so that

(x) ?,I(X 1) r,0(x (x) + E z-e,(x)’

=1

or
0.
=1

This is not possible, unless -,o(x) -0 or (x) 0 for all x.

The consideration of q,o
in (3.4) was motivated by the fundamental work of
Daubechies [5, p. 989] where a corresponding function h*o,o was constructed by using
the Meyer wavelet (cf. [10]) This function h*0,0 is not in Lp(-cx, o) for sufficiently
small p- 1 > 0 as shown by our derivation and (3.4) above. Note that this corrects
a mistake in [5], where it was erroneously stated that h*o,o was not in Lp(-x, cx) for
large p. On the other hand, the fact that 1/Y T also follows from an earlier result
of Tchamitchian [11], [12], as pointed out by Daubechies [5, p. 989] and Meyer [10,
p. 127].

Acknowledgment. We are grateful to the referees for their valuable comments

REFERENCES
[1] C.K. CHUI AND J. Z. WANG, On compactly supported spline-wavelets and a duality principle,
Trans. Amer. Math. Soc., 330 (1992), pp. 903-915.
[2] , A general framework of compactly supported splines and wavelets, J. Approx. Theory,
to appear.
[3] A. COHEN, Ondelettes, Analyses multiresolutions, et traitement numerique du signal, Ph.D.
thesis, University Paris-Dauphine, 1990.
 LITTLEWOOD--PALEY INEQUALITIES FOR FRAMES AND WAVELETS 277

[4] A. COHEN, I. DAUBECHIES, AND J. C. FEAUVEAU, Biorthogonal bases of compactly supported

wavelets, Comm. Pure Appl. Math., to appear.
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

[5] I. DAUBECHIES, The wavelet transform, time-frequency localization and signal analysis, IEEE
Trans. Inform. Theory, 36 (1990), pp. 961-1005.
[6] a. J. DUFFIN AND A. C. SCHAEFFER, A class of nonharmonic Fourier series, Trans. Amer.
Math. Sol., 72 (1952), pp. 341-366.
[7] S. (. MALLAT, Multiresolution approximations and wavelet orthonormal bases of L2(R), Trans.
Amer. Math. Soc., 315 (1989), pp. 69-87.
[8] S. (. MALLAT AND W. L. HWANG, Singularity detection and processing with wavelets, preprint,
1991.
[9] S. (. MALLAT AND S. ZHONG, Wavelet transform maxima and multiscale edges, in Wavelets
and Their Applications, M. B. Ruskai, G. Beylkin, R. Coifman, I. Daubechies, S. Mallat,
Y. Meyer, and L. Raphael, eds., Jones and Bartlett, Boston, 1992, pp. 67-104.
[10] Y. MEYER, Ondelettes, Vol. 1, Hermann, Paris, 1990.
[11] PH. TCHAMITCHIAN, Cacul symbolique sur les opdrateurs de Caldgron-Zygmund et bases in-
conditionnelles de L2(Rn), C.R. Acad. Sci. Paris, 303 (1986), pp. 215-218.
[12] , Biorthogonalitd et thdorie des opdrateurs, Rev. Mat. Iberoamericana, to appear.