A result of Strichartz on a generalization of Wiener’s characterization of continuous
measures revisited
B´ erenger Akon KPATA
Laboratoire de Math´ ematiques et Informatique, UFR des Sciences Fondamentales et Appliqu´ ees, Universit´ e Nangui Abrogoua,
02 BP 801 Abidjan 02, Cˆ ote d’Ivoire kpata akon@yahoo.fr
Received: 19.8.2014; accepted: 18.5.2015.
Abstract. We give the reasons for which the continuous part of a measure that belongs to the Wiener amalgam space M
2does not contribute to the Bohr mean of its Fourier transform.
Keywords: Fourier transform, Radon measure, Wiener amalgam space MSC 2010 classification: primary 42B10, secondary 42B35
1 Introduction
In this paper, for 1 < p < ∞, we denote by p 0 the real number satisfying
1
p + p 1
0= 1. A Radon measure µ on R d belongs to the Wiener amalgam space M p , (1 ≤ p < ∞) if 1 kµk p < ∞ with
r kµk p =
X
k∈Z
d|µ|(I k r ) p
1 p
, r > 0
where I k r = Q d
i=1 [k i r, (k i + 1)r) for k = (k 1 , . . . , k d ) ∈ Z d and |µ| denotes the total variation of µ.
The Fourier transform on amalgam spaces has been studied by various authors including F. Holland ([8], [9]), J. Stewart [11], J. P. Bertrandias and C. Dupuis [2], J. J. F. Fournier ([6], [7]) and I. Fofana [5].
In [8], the Fourier transform f 7→ b f defined on the usual Lebesgue space L 1 by
f (x) = (2π) b −
d2Z
R
df (y) e −ixy dy, x ∈ R d
http://siba-ese.unisalento.it/ c 2015 Universit` a del Salento
has been extended by F. Holland to the spaces (L q , l p ) defined for 1 ≤ q, p ≤ ∞ as follows:
(L q , l p ) = n
f ∈ L 0 | 1 kf k q, p < ∞ o
where L 0 stands for the space of (equivalence classes modulo the equality Lebesgue almost everywhere of) all complex-valued functions defined on R d and for r > 0,
r kf k q, p =
X
k∈Z
df χ I
krq
p
1 p
if 1 ≤ p < ∞, sup
x∈R
df χ J
xrq if p = ∞,
where J x r = Q d
i (x i − r 2 , x i + r 2 ) for x = (x 1 , . . . , x d ) ∈ R d , χ I
rkdenotes the characteristic function of I k r and k · k q is the usual Lebesgue norm. In the same paper, he extended to the spaces M p (1 < p ≤ 2) the Fourier transform µ 7→ µ b defined on the space M 1 of finite Radon measures on R d by
µ(x) = (2π) b −
d2Z
R
de −ixy dµ(y), x ∈ R d .
In fact, he proved that if µ belongs to M p (1 < p ≤ 2), then there exists a unique element µ ∈ (L b p
0, l ∞ ) such that for any sequence (r n ) n≥1 of positive real numbers increasing to ∞, the sequence ( \ µbJ 0 r
n) n≥1 converges in (L p
0, l ∞ ) to b µ, where µbJ 0 r
nis the measure defined by (µbJ 0 r
n) (N ) = µ (J 0 r
n∩ N ) for any Borel subset N of R d . In addition,
Z
R
dg (x) µ (x) dx = b Z
R
db g(x) dµ (x) , g ∈ L p , l 1 . In [5], I. Fofana has proved the following Hausdorff-Young inequality :
r −
p0dr k µk b p
0, ∞ ≤ C
1r
kµk p , r > 0 (1.1) where the real constant C does not depend on µ and r.
The relation between the properties of a Radon measure and the asymptotic behavior of its Fourier transform is an important topic in Harmonic Analysis.
A well-known theorem of Wiener [14] states that if µ ∈ M 1 , then
r→∞ lim r −d Z
J
0r| µ(y)| b 2 dy = X
a∈D
|µ ({a})| 2 , (1.2)
where D = { a ∈ R d | µ({a}) 6= 0 } is a countable set.
The left-hand side of (1.2) is the so-called Bohr mean of µ. Wiener’s result b has found many applications in various areas of mathematics, including ergodic theory and optimal control theory. Since then, Wiener-type characterizations of continuous measures are extensively investigated (see for example [1], [3], [4], [12] and [13, Section 12.5]). Strichartz has established that equality (1.2) continues to hold if the measure µ belongs to M 2 (see Theorem 4.4 in [12]).
The aim of this note is to give, by a new proof, the reasons for which the continuous part of a measure belonging to M 2 does not contribute to formula (1.2). To this end , we prove in Section 2 the following result: if µ is a continuous measure that belongs to M p , then
r→0 lim r kµk p = 0.
In Section 3, we show that if µ is a discrete measure that belongs to M p , then
r→0 lim r kµk p = X
a∈D
|µ ({a})| p
!
1p
.
Since any Radon measure on R d has a decomposition into discrete and continu- ous parts, we combine these above results with inequality (1.1) to establish, in Section 4, the generalization of Wiener’s theorem obtained by Strichartz.
2 M p -estimate of a continuous measure
Let us recall the definition of a continuous Radon measure.
Definition 1. A Radon measure µ on R d is continuous if for any x ∈ R d we have µ({x}) = 0.
Notice that for any Radon measure µ on R d and any x ∈ R d , we have
|µ({x})| = |µ|({x}). Therefore, a Radon measure on R d is continuous if and only if its total variation is continuous.
The following proposition will be useful in the proof of the main result of this section.
Proposition 1. [10]. Let 1 ≤ p < ∞. If µ ∈ M p and 0 < r < s < ∞, then (i) s kµk p ≤ Int r s + 2
p0d2
dpr kµk p , where Int s r
denotes the greatest integer not exceeding r s ;
(ii) r kµk p ≤ 3
p0d2
dps kµk p .
We can now prove the following result.
Proposition 2. Suppose that 1 ≤ p < ∞ and µ is a continuous measure that belongs to M p . Then we have
(i) lim r→0 sup k∈Z
d|µ|(I k r ) = 0, (ii) p > 1 =⇒ lim r→0 r kµk p = 0.
Proof. For each non-negative integer n, set s n = sup
k∈Z
d|µ|(I k 2
−n).
(i) Notice that (s n ) n is a decreasing sequence of positive real numbers. Suppose that lim n→∞ s n = s > 0. For each non-negative integer n, let
K n = n
k ∈ Z d | |µ|(I k 2
−n) > s 2
o . Let us notice that for each positive integer n,
0 < card K n ≤ 2
s 2
−nkµk p
p
≤ 2
s 1 kµk p
p
< ∞ and
k ∈ K n+1 =⇒ ∃ l ∈ K n : I k 2
−(n+1)⊂ I l 2
−n,
where card K n denotes the cardinality of K n . So there exists a sequence (k n ) n of elements of Z d such that for each non-negative integer n we have
k n ∈ K n and I k 2
n+1−(n+1)⊂ I k 2
n−n. It follows that ∩ n I k 2
−nn
is a one point set and |µ|(∩ n I k 2
−nn
) > 0. This is in contra- diction with the fact that µ is a continuous measure. So
n→∞ lim s n = 0.
Now let us consider a real number ε > 0. There exists an integer n 0 such that s n
0< 2 ε
d. Let us notice that for each element (r, k) of (0, 2 −n
0) × Z d , the set I k r intersects at most 2 d elements of { I l 2
−n0| l ∈ Z d }. So
|µ|(I k r ) ≤ 2 d s n
0, (r, k) ∈ (0, 2 −n
0) × Z d . It follows that
sup
k∈Z
d|µ|(I k r ) < ε, r ∈ (0, 2 −n
0).
Therefore,
r→0 lim sup
k∈Z
d|µ|(I k r ) = 0.
(ii) Suppose p > 1. Let us consider a real number t > 0. There exists a finite subset L of Z d such that
X
k∈Z
d\L
|µ|(I k 1 ) p < t.
Let us set
E = [
k∈L
I k 1
and for each non-negative integer n, L n = n
k ∈ Z d | I k 2
−n⊂ E o
, α n = X
k∈L
n|µ|(I k 2
−n) p and t n = X
k∈Z
d\L
n|µ|(I k 2
−n) p .
Then, for each non-negative integer n,
2
−nkµk p = (α n + t n )
1p, α n ≤ s p−1 n |µ|(E) and t n ≤ t.
Let us notice that
|µ|(E) ≤ card L 1 kµk p < ∞
(where card L denotes the cardinality of L) and ( 2
−nkµk p ) n is a decreasing sequence. It follows that
n→∞ lim 2
−nkµk p ≤ t
p1.
Since the above inequality holds for an arbitrary real number t > 0, we deduce that
n→∞ lim 2
−nkµk p = 0.
Further, for each non-negative integer n, it follows from Proposition 1 (ii) that
0 ≤ r kµk p ≤ 3
p0d2
dp2
−nkµk p , r ∈ (0, 2 −n ).
So
r→0 lim r kµk p = 0.
QED
3 M p -estimate of a discrete measure
This section is devoted to the proof of the following result.
Proposition 3. Suppose that 1 ≤ p < ∞ and µ is a discrete measure that belongs to M p . Then
r→0 lim r kµk p = X
a∈D
|µ({a})| p
!
1p,
where D = { a ∈ R d | µ({a}) 6= 0 }.
Proof. Let us consider a real number ε > 0. There exists a finite subset K of Z d such that
X
k∈Z
d\K
|µ| I k 1 p
1 p
< ε 2
dp.
Set
E = [
k∈K
I k 1 , D n =
a ∈ D | |µ({a})| ≥ 1 n
and µ n (B) = µ(B ∩ D n )
for any positive integer n and any Borel subset B of R d . Then, for each positive integer n and each element r of (0, 1), we have
r kµk p ≤
X
k∈Z
d|µ| (I k r ∩ E) p
1 p
+
X
k∈Z
d|µ| (I k r \ E) p
1 p
≤
X
k∈Z
d|µ| (I k r ∩ E) p
1 p
+
2 d X
k∈Z
d\K
|µ| I k 1 p
1 p
≤
X
k∈Z
d|µ| (I k r ∩ E) p
1 p
+ ε
≤
X
k∈Z
d|µ n | (I k r ∩ E) p
1 p
+ |µ| (E ∩ (D \ D n )) + ε
≤
X
k∈Z
d|µ n | (I k r ) p
1 p
+ |µ| (E ∩ (D \ D n )) + ε.
Since for each positive integer n the set D n is finite, then there exists an element r n of (0, 1) such that for all a, b ∈ D n ,
a 6= b =⇒ ∃ i ∈ { 1, . . . , d } : |a i − b i | > r n and for each element r of (0, r n )
X
k∈Z
d|µ n | (I k r ) p
1 p
= X
a∈D
n|µ({a})| p
!
p1≤ X
a∈D
|µ({a})| p
!
1p.
Therefore, for each element r of (0, r n ) we have
r kµk p ≤ X
a∈D
|µ({a})| p
!
1p+ |µ| (E ∩ (D \ D n )) + ε.
Since |µ|(E) < ∞ and (D \ D n ) n is a decreasing sequence that converges to the empty set, we get
lim sup
r→0 r kµk p ≤ X
a∈D
|µ({a})| p
!
1p+ ε.
As this inequality holds for an arbitrary real number ε > 0, we deduce that
lim sup
r→0 r kµk p ≤ X
a∈D
|µ({a})| p
!
1p.
Further, for all real numbers r > 0, we have
X
a∈D
|µ({a})| p
!
1p=
X
k∈Z
d
X
a∈D∩I
kr|µ({a})| p
1 p
≤ r kµk p .
So
r→0 lim r kµk p = X
a∈D
|µ({a})| p
!
1p.
QED
4 Generalization of Wiener’s theorem
Throughout this section, for 1 < p < ∞ and µ ∈ M p , we set D = { a ∈ R d | µ({a}) 6= 0 }.
Proposition 4. Let 1 < p < ∞. If µ ∈ M p then
r→0 lim r kµk p = X
a∈D
|µ({a})| p
!
1p.
Proof. Let us write µ = µ c + µ δ , where µ c is a continuous measure and µ δ is a discrete measure. For each real number r > 0, we have
r kµ δ k p ≤ r kµk p ≤ r kµ c k p + r kµ δ k p .
The desired result follows from Propositions 2 and 3.
QEDAs an immediate consequence of (1.1) and Proposition 4, we have the fol- lowing result.
Corollary 1. Let 1 < p ≤ 2. There exists a real constant C > 0 such that for any µ ∈ M p we have
lim sup
r→∞
r −
d
p0
r k µk b p
0, ∞ ≤ C X
a∈D
|µ({a})| p
!
1p.
In the case p = 2, we obtain the following generalization of Wiener’s theorem.
Proposition 5. If µ ∈ M 2 then
r→∞ lim r −
d2Z
J
0r| µ(y)| b 2 dy
!
12= X
a∈D
|µ({a})| 2
!
12.
Proof. a) Let us write µ = µ c + µ δ , where µ c is a continuous measure and µ δ is a discrete measure. It follows from Propositions 2 and 3 that
r→∞ lim
1rkµ c k 2 = 0 and lim
r→∞
1rkµ δ k 2 = X
a∈D
|µ({a})| 2
!
12< ∞.
Note that for each element (r, x) of (0, ∞) × R d , we have Z
J
xr| µ(y)| b 2 dy = Z
J
xr| c µ δ (y)| 2 dy + 2Re Z
J
xrc µ δ (y) µ b c (y) dy + Z
J
xr| µ b c (y)| 2 dy,
where Re Z
J
xrc µ δ (y) µ b c (y) dy denotes the real part of Z
J
xrc µ δ (y) µ b c (y) dy.
By the H¨ older’s inequality and (1.1), we have
Z
J
xrc µ δ (y) µ b c (y) dy
≤ Z
J
xr| c µ δ (y)| 2 dy
!
12Z
J
xr| µ b c (y)| 2 dy
!
12≤ r k c µ δ k 2, +∞ r k µ b c k 2, +∞
≤ C 2
1r
k µ δ k 2
1r
k µ c k 2 r d and
Z
J
xr| µ b c (y)| 2 dy ≤ r k µ b c k 2 2, +∞ ≤ C 2
1 rkµ c k 2 2 r d .
It follows that
lim sup
r→∞
r −
d2r k µk b 2, ∞ = lim sup
r→∞
r −
d2r k c µ δ k 2, ∞ and
lim inf
r→∞ r −
d2r k µk b 2, ∞ = lim inf
r→∞ r −
d2r k c µ δ k 2, ∞ .
Therefore, it suffices to consider the case where µ is a discrete measure to prove the desired result.
b) Suppose now that µ is a discrete measure.
Let us consider an element ε of
0, 1 3 P
a∈D |µ({a})| 2
12. For each positive integer n and each Borel subset B of R d , set
D n =
a ∈ D | |µ({a})| ≥ 1 n
and µ n (B) = µ(B ∩ D n ).
There exists a positive integer N such that
X
a∈D\D
N|µ({a})| 2
1 2
< ε
3(C + 1) , (4.1)
where C is a real constant as in Corollary 1. So
lim sup
r→+∞
r −
d2r k µ − b µ c N k 2, ∞ ≤ C
X
a∈D\D
N|µ({a})| 2
1 2
< ε 3 .
Thus, there exists a real number R 1 > 0 such that for any r ≥ R 1 and any x ∈ R d ,
r −
d2Z
J
xr| b µ(y)| 2 dy
!
12− r −
d2Z
J
xr| µ c N (y)| 2 dy
!
12< ε
3 . (4.2)
Since µ N is a finite measure, it follows from (1.2) that
r→∞ lim r −
d2Z
J
0r| µ c N (y)| 2 dy
!
12=
X
a∈D
N|µ({a})| 2
1 2
.
Thus, there exists a real number R 2 > 0 such that for any r ≥ R 2 ,
r −
d2Z
J
0r| µ c N (y)| 2 dy
!
12−
X
a∈D
N|µ({a})| 2
1 2
< ε
3 . (4.3)
Let R = max{ R 1 , R 2 }. From inequalities (4.1), (4.2) and (4.3), we obtain
r −
d2Z
J
0r| µ(y)| b 2 dy
!
12
− X
a∈D
|µ({a})| 2
!
12
< ε, r ≥ R.
Hence
r→+∞ lim r −
d2Z
J
0r| µ(y)| b 2 dy
!
12= X
a∈D
|µ({a})| 2
!
12.
QED