A L
E S
E D L ’IN IT ST T U
F O U R
ANNALES
DE
L’INSTITUT FOURIER
Loïc GAILLARD & Pascal LEFÈVRE
Lacunary Müntz spaces: isomorphisms and Carleson embeddings Tome 68, no5 (2018), p. 2215-2251.
<http://aif.cedram.org/item?id=AIF_2018__68_5_2215_0>
© Association des Annales de l’institut Fourier, 2018, Certains droits réservés.
Cet article est mis à disposition selon les termes de la licence CREATIVECOMMONS ATTRIBUTION–PAS DE MODIFICATION3.0 FRANCE. http://creativecommons.org/licenses/by-nd/3.0/fr/
L’accès aux articles de la revue « Annales de l’institut Fourier »
(http://aif.cedram.org/), implique l’accord avec les conditions générales d’utilisation (http://aif.cedram.org/legal/).
cedram
Article mis en ligne dans le cadre du
LACUNARY MÜNTZ SPACES:
ISOMORPHISMS AND CARLESON EMBEDDINGS
by Loïc GAILLARD & Pascal LEFÈVRE
Abstract. — In this paper we prove thatMΛp is almost isometric to `p in the canonical way when Λ is lacunary with a large ratio. On the other hand, our approach can be used to study also the Carleson measures for Müntz spacesMΛp when Λ is lacunary. We give some necessary and some sufficient conditions ensuring that a Carleson embedding is bounded or compact. In the hilbertian case, the membership to Schatten classes is also studied. When Λ behaves like a geometric sequence the results are sharp, and we get some characterizations.
Résumé. — Dans cet article, nous montrons queMΛpest presque isométrique à `p, et ce de façon naturelle, lorsque Λ est lacunaire avec une raison grande.
Par ailleurs, notre approche permet aussi d’étudier les mesures de Carleson pour les espaces MüntzMΛp lorsque Λ est lacunaire. Nous donnons des conditions né- cessaires et des conditions suffisantes qui permettent d’assurer qu’un plongement de Carleson est borné ou compact. Dans le cadre hilbertien, nous étudions aussi l’appartenance de ce plongement aux classes de Schatten. Nous obtenons des ca- ractérisations complètes lorsque Λ se comporte comme une suite géométrique.
1. Introduction
Let m be the Lebesgue measure on [0,1]. For p ∈ [1,+∞), Lp(m) = Lp([0,1], m) (sometimes denoted simply by Lp when there is no ambigu- ity) denotes the space of complex-valued measurable functions on [0,1], equipped with the norm kfkp = (R1
0 |f(t)|pdt)p1. In the same way, C = C([0,1]) is the space of continuous functions on [0,1] equipped with the usual sup-norm. We shall also consider some positive and finite measures µ on [0,1) (see the remark at the beginning of Section 2), and the as- sociated Lp(µ) space. For a sequence w = (wn)n of positive weights, we denote`p(w) the Banach space of complex sequences (bn)n equipped with
Keywords:Müntz spaces, Carleson embeddings, lacunary sequences, Schatten classes.
2010Mathematics Subject Classification:30B10, 47B10, 47B38.
the norm kbk`p(w) = (P
n|bn|pwn)1p and the vector space c00 consisting on complex sequences with a finite number of non-zero terms. All along the paper, when p∈(1,+∞), we denote as usualp0 = p−1p its conjugate exponent.
The famous Müntz theorem ([3, p. 172],[7, p. 77]) states that if Λ = (λn)n∈N is an increasing sequence of non-negative real numbers, then the linear span of the monomials tλn is dense in Lp (resp. in C) if and only ifP
n>1 1
λn = +∞ (resp. and λ0 = 0). We shall assume that the Müntz conditionP
n>1 1
λn <+∞ is fulfilled and we define the Müntz spaceMΛp as the closed linear space spanned by the monomials tλn, where n ∈ N. We shall moreover assume that Λ satisfies the gap condition: infn λn+1− λn
>0. Under this later assumption Clarkson–Erdös theorem holds [7, Thm. 6.2.3]: the functions inMΛp are the functionsf inLpsuch thatf(t) = Pantλn (pointwise on [0,1)). This gives a class of Banach spacesMΛp (Lp of analytic functions on (0,1).
In full generality, the Müntz spaces are difficult to study, but for some particular sequences Λ, we can find some interesting properties of the spaces MΛp. Let us mention that lately these spaces received an increasing atten- tion from the point of view of their geometry and operators: the monograph of Gurariy–Lusky [7], and various more or less recent papers (see for in- stance [1, 2, 4, 9, 10]).
We shall focus on two different questions on Müntz spaces. The first one is related to an old result: Gurariy and Macaev proved in [8] that, inLp, the normalized sequence ((pλn+ 1)1ptλn)n is equivalent to the canonical basis of`p if and only if Λ is lacunary (see Theorem 2.3 below). More recently, the monograph [7] introduces the notion of quasi-lacunary sequence (see Definition 2.1 below), and states that MΛp is still isomorphic to `p when Λ is quasi-lacunary. On the other hand, some recent papers discuss about Carleson measures for Müntz spaces. In [4], the authors introduced the class of sublinear measures on [0,1), and proved that when Λ is quasi- lacunary, the sublinear measures are Carleson embeddings forMΛ1. In [10], the authors extended this result to the case p = 2 but only when the sequence Λ is lacunary.
In this paper, we introduce another method to study the lacunary Müntz spaces: for a weightwand a measureµon [0,1), we define TΛ,µw :`p(w)→ Lp(µ) by TΛ,µw (b) = P
nbntλn for b = (bn) ∈ `p(w). The operator TΛ,µw depends on w, µ, p and Λ, and when it is bounded we shall denote its norm by
TΛ,µw
p. We shall see that an estimate of TΛ,µw
p can improve
Gurariy–Macaev theorem, and allows to generalize former Carleson embed- ding results to lacunary Müntz spacesMΛp for anyp>1.
The paper is organized as follows: in Section 2, we specify the miss- ing notation and some useful lemmas. The main result gives an upper bound for the approximation numbers of TΛ,µw (see Proposition 2.9). In Section 3, we focus on the classical case: we fix the weightw(p) defined by wn(p) = (pλn+ 1)−1 and consider TΛw(p) =TΛ,mw(p) : `p(w(p))→ MΛp, the isomorphism occuring in Gurariy–Macaev theorem. For p > 1, we prove that TΛw(p) is bounded exactly when Λ is quasi-lacunary. On the other hand, when Λ is lacunary with a large ratio, we also get a sharp bound for
TΛw(p)−1
p (see Theorem 3.5 below). Our approach leads to an asymp- totically orthogonal version of Gurariy–Macaev theorem exactly for the super-lacunary sequences. In Section 4, we apply the results of Section 2 for a positive and finite measureµ on [0,1) with the weightwn =λ−1n , in order to treat the Carleson embedding problem. When Λ is lacunary, we give an estimate of the approximation numbers of the embedding operator ipµ:MΛp →Lp(µ). In Section 5, we focus on the compactness ofipµusing the same tools as in Section 4. In the casep= 2, this leads to some control of the Schatten norm of the Carleson embedding and some characterizations when Λ behaves like a geometric sequence.
As usual the notation A .B means that there exists a constant c >0 such thatA6cB. This constant c may depend on Λ (or sometimes only on its ratio of lacunarity), on p . . . . We shall specify this dependence to avoid any ambiguous statement. In the same way, we shall use the notation A≈B orA&B.
Acknowledgments. We wish to thank the referee for the very careful reading of the paper and many suggestions in order to improve its quality.
2. Preliminary results
Let us first give a few words of explanation about our choice of mea- sures on [0,1). The measures involved (if considered on [0,1]) must satisfy µ({1}) = 0. Indeed, we focus either on the Lebesgue measurem(satisfying of course m({1}) = 0) or on measures such that the Carleson embedding MΛp→Lp(µ) is (defined and) bounded, so that testing a sequence of mono- mialsgn(t) =tλn we must have
µ({1}) = limkgnkpLp(µ).limkgnkpLp(m)= 0.
Moreover, by Clarkson–Erdös theorem, the value at any point of [0,1) of any function inMΛp can be defined without ambiguity.
We shall need several notions of growth for increasing sequences.
Definition 2.1.
• A sequenceu= (un)nof positive numbers is calledlacunaryif there existsr > 1 such thatun+1 >run, for everyn∈N. We shall say that such a sequence isr-lacunary and thatris a ratio of lacunarity of this sequence.
• A sequenceuis calledquasi-lacunaryif there is an extraction(nk)k such thatsupk∈N(nk+1−nk)<+∞,and(unk)k is lacunary.
• A sequenceuis called quasi-geometricif there are two constants r andR such that1< r6 uun+1n 6R <+∞, for everyn∈N. Such a sequence is lacunary.
• A sequenceuis called super-lacunary if uun+1
n −→+∞.
Remark 2.2. — It is proved in [7, Prop. 7.1.3, p. 94] that a sequence is quasi-lacunary if and only if it is a finite union of lacunary sequences.
The following result is due to Gurariy and Macaev.
Theorem 2.3([7, Cor. 9.3.4, p. 132]). — Letp∈[1,+∞). The following assertions are equivalent
(1) The sequenceΛ is lacunary.
(2) The sequence
tλn ktλnkp
in Lp is equivalent to the canonical basis of`p.
In particular, whenΛ is lacunary, we have for anyb∈c00
Xbntλn p≈
X |bn|p pλn+ 1
1p
and the underlying constants depend onpandΛonly.
We shall recover and generalize partially this result: for a given sequence of weights (wn)n and a positive finite measure µ on [0,1), we study the boundedness of the operator
TΛ,µw :
`p(w) −→ Lp(µ)
b 7−→ P
bntλn.
Example 2.4. — In the case of the Lebesgue measureµ=mand when the weights are wn(p) = (pλn+ 1)−1 or in a simpler way (if we do not care about the value of the constants) wn =λ−1n , Theorem 2.3 states in particular thatTΛw and TΛw(p) are bounded on`p(w) or `p(w(p)), when Λ is lacunary.
Remark 2.5. — In the casep >1, a (rough) sufficient condition ensuring the boundedness ofTΛ,µw is
Z
[0,1)
X
n
w−
p0
n ptp0λn
!pp0
dµ <∞.
Indeed, this comes from the estimate TΛ,µw
p= sup
a∈B`p a∈c00
sup
g∈BLp0 (µ)
Z
[0,1)
X
n
anw−
1
nptλng(t) dµ
6 sup
g∈BLp0(µ)
Z
[0,1)
|g(t)| sup
a∈B`p
a∈c00
X
n
anw−
1 p
n tλn
dµ.
For standard weights, wn ≈λ−1n and for a quasi-geometric sequence Λ, this condition can be reformulated with the help of Lemma 2.10 below as
Z
[0,1)
1
1−t dµ≈ Z
[0,1)
1
1−tp0 dµ <∞.
Such a condition will be considered later (see Proposition 5.5 below for instance).
To get a sharper estimate, we introduce the sequence Dw,pΛ,µ(n)
ndefined forn∈Nandp>1, with a priori values inR+∪ {+∞} by
Dw,pΛ,µ(n) = Z
[0,1)
w−
1 p
n tλn X
k>0
w−
1 p
k tλk
!p−1
dµ
!1p .
Proposition 2.6. — Let p ∈ [1,+∞). Assume that Dw,pΛ,µ(n)
n is a bounded sequence of real numbers. Then we have for everyb∈`p(w),
X
n>0
bntλn Lp(µ)6
X
n>0
|bn|pwn DΛ,µw,p(n)p
1 p
.
Proof. — Ifp= 1 the result is obvious. Assume now thatp >1. For any t∈[0,1) andn∈N, we have
bntλn=bnw
1 pp0
n tλnp ×w−
1 pp0
n tλnp0 , By Hölder’s inequality, we get
Xbntλn
6 X
n
|bn|pw
1 p0
n tλn
!p1 X
k
w−
1 p
k tλk
!p10
.
We conclude Z
[0,1)
Xbntλn
p
dµ6 Z
[0,1)
X|bn|pwn.w−
1 p
n tλn X
k
w−
1 p
k tλk
!p−1
dµ
=X
n
|bn|pwnDw,pΛ,µ(n)p.
Assume that DΛ,µw,p(n)
nis a bounded sequence of real numbers. We may define the bounded diagonal operator
Dw,pΛ,µ:`p(w)→`p(w)
acting on the canonical basis of`p(w) whose diagonal entries are the num- bersDw,pΛ,µ(n). In other wordsTΛ,µw andDΛ,µw,pare bounded, and we have
∀b∈`p(w), kTΛ,µw (b)kLp(µ)6kDΛ,µw,p(b)k`p(w).
This gives informations about the approximation numbers ofTΛ,µw . Let us recall some definitions.
Definition 2.7. — For a bounded operator S : X → Y between two separable Banach spacesX, Y, the approximation numbers (an(S))n of S are defined forn>1by
an(S) = inf{kS−Rk,rank(R)< n}.
The essential normofS is defined by
kSke= inf{kS−Kk, K compact}.
It is the distance fromS to the compact operators.
We shall use in the sequel the following notions of operator ideals.
Definition 2.8.
• An operatorS:X →Y isnuclearif there is a sequence of rank-one operators(Rn)satisfyingS(x) =P
nRn(x)for every x∈ X with P
nkRnk<+∞.Thenuclear normofS is defined as kSkN = inf
( X
n
kRnk,rank(Rn) = 1,X
n
Rn=S )
.
• An operator S : X → Lp(µ) is order bounded if there exists a positive function h ∈ Lp(µ) such that for every x ∈ BX and for µ−almost everyt∈Ωwe have |S(x)(t)|6h(t).
• Forr >0 and whenX, Y are Hilbert spaces, we say that a (com- pact) operatorS:X →Y belongs to theSchatten classSr if
X
n
an(S)r
<+∞.
In this case, we define itsSchatten normbykSkSr= P
n an(S)rr1 . Recall that nuclear and Schatten class operators are always compact.
Of course, the Schatten norm is really a norm when r>1. TheS2 class is also called the class ofHilbert–Schmidtoperators.
We shall be interested in how far from compact (the essential norm) or, on the contrary, how strongly compact (possibly Schatten in the Hilbert framework) are the Carleson embeddings.
For technical reasons, we introduce the following notation: for a bounded sequence (un)n in R+, we define (u∗N)N thedecreasing rearrangement of (un)n by
u∗N = inf
A⊂N
|A|=N
sup{un, n6∈A}.
We have limN→+∞u∗N = lim supn→+∞un. Now, we can state,
Proposition 2.9. — If Dw,pΛ,µ(n)
nis a bounded sequence of real num- bers, then we have
(1) aN+1(TΛ,µw )6 DΛ,µw,p∗
(N).
(2) kTΛ,µw kp6supn∈NDw,pΛ,µ(n).
(3) kTΛ,µw ke6lim supn→+∞DΛ,µw,p(n).
(4) ∀p>1,kTΛ,µw kN 6P
n>0w−
1
np tλn
Lp(µ). (5) Ifp= 2, for any r >0,kTΛ,µw kSr 6 P
n>0 Dw,2Λ,µ(n)r1r .
Proof. — We first prove (1). Forn∈N, we denoteϕ∗n :`p(w)→Cthe functional on`p(w) defined byϕ∗n(b) =bnfor a sequenceb= (bn)n ∈`p(w).
We define alsogn ∈Lp(µ) bygn(t) =tλn. For any integer N and A⊂N with|A|=N, we have
aN+1(TΛ,µw )6
TΛ,µw −X
n∈A
ϕ∗n⊗gn .
By Proposition 2.6, forb∈`p(w),
TΛ,µw (b)−X
n∈A
ϕ∗n(b)gn
=
X
n6∈A
bntλn Lp(µ)
6sup
n6∈A
DΛ,µw,p(n)
kbk`p(w)
and so (1) holds.
Assertions (2) and (3) are direct consequences of (1).
Assertion (4) follows easily from the natural decomposition TΛ,µw (b) = P
nϕ∗n(b)gn and the fact thatkϕ∗nk=wn−1p.
For (5): if (Dw,2Λ,µ(n))n6∈`rthen the result is obvious. If (DΛ,µw,2(n))n ∈`r, we have in particularDΛ,µw,2(n)→0 whenn→+∞. Since for allε >0, the set{n, DΛ,µw,2(n)>ε} is finite, there exists a bijectionβ :N→Nsuch that for anyn∈N,DΛ,µw,2(n)∗=Dw,2Λ,µ(β(n)). We have
X
N
aN+1(TΛ,µw )r6X
N
DΛ,µw,2(N)∗r
=X
n
Dw,2Λ,µ(β(n))r
=X
n
Dw,2Λ,µ(n)r
.
Lemma 2.10. — Letα ∈R∗+. Assume that Λ is a quasi-geometric se- quence. Then there are two constants C1, C2 ∈ R∗+ such that for any t∈[0,1)we have
C1
1 1−t
α
6X
n
λαntλn6C2
1 1−t
α
·
Proof. — Since Λ is quasi-geometric, it is r-lacunary for some r > 1, so there exists a constant C = (r−1)−1 such that for any n ∈ N, λn 6 C(λn+1−λn). Moreover, there is a constantR >1 such that λn+16Rλn
and hence we have
λαn ≈(λn+1−λn)α≈λαn+1
where the underlying constants do not depend onn. We obtain X
n
λαntλn≈X
n
(λn+1−λn)αtλn≈X
n
X
λn6m<λn+1
(λn+1−λn)α−1tλn
≈X
n
X
λn6m<λn+1
mα−1tλn
Formsuch thatλn6m < λn+1, we havetm6tλn6tmR and so we obtain X
n
λαntλn. X
m>0
mα−1tmR . 1
1−tR1 α
. 1
1−t α
·
On the other hand we have X
n
λαntλn& X
m>1
mα−1tm&
1 1−t
α
·
Remark 2.11. — If Λ is only lacunary, the right inequality in Lemma 2.10 still holds. Indeed, the above proof can be easily adapted. We can also notice that there exists a quasi-geometric sequence Λ0 = (λ0n)n which contains Λ, and we have
X
n∈N
λαntλn6X
n∈N
λ0αntλ0n 6C2
1 (1−t)α·
A new proof of the upper bound part in Gurariy–Macaev theorem (The- orem 2.3) follows from the next proposition.
Proposition 2.12. — Let p ∈ [1,+∞). Assume that the weights are given bywn=λ−1n or(pλn+ 1)−1. IfΛ is lacunary andµis the Lebesgue measure, then DΛ,µw,p(n)
n is a bounded sequence.
Proof. — From Lemma 2.10 and Remark 2.11 we get Dw,pΛ,µ(n)p
=λ
1
np
Z
tλn X
k∈N
λ
1 p
ktλk
!p−1
dt
.λ
1
np
Z 1 0
tλn 1
1−t p10
dt
=λ
1 p
n
Z 1−λn1 0
tλn 1
1−t p10
dt+λ
1 p
n
Z 1 1−λn1
tλn 1
1−t p10
dt
6λ
1
npλ
1 p0
n
Z 1 0
tλndt+λ
1
np
Z 1 1− 1
λn
(1−t)−p10dt
6 λn
λn+ 1+λ
1 p
n
p λ
1
np
6p+ 1.
From Proposition 2.6, we obtain as claimed
X
n∈N
bntλn p
. X
n∈N
|bn|p λn
!p1 ,
for anyb∈c00, when Λ is lacunary.
From Lemma 2.10 and Gurariy–Macaev’s Theorem, one can easily get an estimate of point evaluations onMΛp.
Proposition 2.13. — LetΛbe a quasi-geometric sequence andp>1.
For anyt∈[0,1), the point evaluationf ∈MΛp7−→δt(f) =f(t)satisfies δt
(Mp
Λ)∗≈ 1 (1−t)p1
· WhenΛis only lacunary, we only have
δt
(Mp
Λ)∗. 1
(1−t)1p ·
Proof. — Since Λ is in particular lacunary, Gurariy–Macaev theorem gives
δt
(Mp
Λ)∗= sup
f∈BMp Λ
|f(t)| ≈ sup
a∈B`p
X
n>0
λ
1
npantλn
where the underlying constants depend onpand Λ. Ifp >1, the last term is
P
n>0λ
p0 p
n tp0λnp10
and we conclude with Lemma 2.10.
Ifp= 1, we have fortclose to 1 (sayt>exp(−1/λ1)) δt
(M1
Λ)∗≈ sup
a∈B`1
X
n>0
λnantλn
= sup
n>0
λntλn
6 sup
s>λ0
sts= 1
e|ln(t)| ≈ 1 (1−t)· Moreover, if Λ is quasi-geometric then there exists someδ∈(0,1) (depend- ing on Λ only) and some integerntsuch thatλnt ∈ δ/|ln(t)|,1/|ln(t)|
so that
sup
n>0
λntλn >λnttλnt > δ
e|ln(t)| ≈ 1
(1−t)·
3. Revisiting the classical case
We consider the Lebesgue measure µ=mon [0,1]. We define the oper- ator
TΛw(p):
`p(w(p)) −→ MΛp
b 7−→ P
nbntλn
where the weightsw(p) = (wn(p))n∈Nare given bywn(p) = (pλn+ 1)−1= ktλnkpp. In particular, if we denote by (ek)k the canonical basis of`p(w(p)), we have
∀k∈N, kTΛw(p)(ek)kp=kekk`p(w(p)).
Gurariy–Macaev theorem says that TΛw(p) is an isomorphism if and only if Λ is lacunary. By our Propositions 2.6 and 2.12, we recover thatTΛw(p)is bounded when Λ is lacunary.
Since TΛw(1)
1 = 1, we focus mainly on the case p > 1. We shall also prove thatTΛw(p)is bounded if and only if Λ is quasi-lacunary (for p >1).
We shall refine the method used in Proposition 2.12 and get a sharper estimate of the norm. Our approach is different from the original one (which was based on some slicing of the interval (0,1)). We control the norm with explicit quantities depending only on the ratio of lacunarity (andp). As a
consequence, we shall get that for p∈ (1,+∞), the operator TΛw(p) is an asymptotical isometry if and only if Λ is super-lacunary.
Lemma 3.1. — Let α ∈ (0,+∞), p ∈ (1,+∞) and (qn)n be an r- lacunary sequence. We have
sup
n∈N
X
k∈N k6=n
q
1
npq
1 p0
k qn
p +qpk0
α
6 p0α
rαp −1+ pα rpα0 −1
·
Proof. — Letn∈N. For k < n, we have q
1 p nq
1 p0 qn k
p+qk
p0 6p
qk
qn
p10
6pr−n−kp0 · We obtain
n−1
X
k=0
q
1 p
nq
1 p0
k qn
p +qpk0
α
6pα
n−1
X
k=0
1 r
(n−k)α p0
6 pα rpα0 −1
·
Similarly, whenk > n, we use q
1 p nq
1 p0 qn k
p+qk
p0 6p0
qn
qk
1p
6p0r−k−np . For p ∈ [1,+∞) we consider the sequence Dw,pΛ,µ(n)
n defined in Sec- tion 2, but since we focus on the caseµ=mandw=w(p), we lighten the notation and write
D(p)Λ (n) = Z 1
0
(pλn+ 1)1ptλn X
k
(pλk+ 1)1ptλk
!p−1
dt
!1p .
Proposition 3.2. — Letp >2 and Λ be a (lacunary) sequence such that(pλn+ 1)n isr-lacunary. Then we have
kTΛw(p)kp6 1 + 2pp−11 rp(p−1)1 −1
!p10
.
Proof. — Forj ∈N, we denoteqj =pλj+ 1 andfj(t) =q
1 p
j tλj = tλj
ktλjkp
· We have
DΛ(p)(n)p
= Z 1
0
fn X
k
fk
!p−1
dt=
X
k
fk
p−1
Lp−1(fndt)
.
Sincep−1>1, the triangle inequality gives D(p)Λ (n)p0
6X
k
kfkkLp−1(fndt)=X
k
q
1
npq
1 p0
k
Z 1 0
tλn+(p−1)λkdt p−11
.
Forn, k∈N, we have q
1
npq
1 p0
k
Z 1 0
tλn+(p−1)λkdt= q
1
npq
1 p0
k
λn+ (p−1)λk+ 1 = q
1
npq
1 p0
k qn
p +qpk0
·
By Lemma 3.1, we obtain for anyn∈N D(p)Λ (n)p0
6X
k∈N
q
1
npq
1 p0
k qn
p +qpk0
!p−11
61 + 2pp−11 rp(p−1)1 −1
sincep>p0and the term indexed byn=kis 1. Thanks to Proposition 2.6, we have
kTΛw(p)kp6sup
n
DΛ(p)(n).
Let us point out that the operatorsTΛw(p):`p(w(p))→MΛp ⊂Lp(m) are not defined on the same scale ofLp-spaces, since the weightw(p) depends on p. We cannot apply directly the Riesz–Thorin theorem withTΛw(1)andTΛw(2) to estimate the norm ofTΛw(p)whenp∈(1,2), even not its weighted version.
The next result gives a bound different from the one in Proposition 3.2;
they coincide whenp= 2.
Proposition 3.3. — Letp∈[1,2]andΛbe a (lacunary) sequence such that(pλn+ 1)n isr-lacunary. Then we have
kTΛw(p)kp6 1 + 4 r12 −1
!p10
·
Proof. — Our proof is adapted from the classical proof of Riesz–Thorin theorem, with an additional trick.
Let θ = p20 ∈ (0,1). We have 1p = 1− θ2· As usual, for z ∈ C such that 0 6 Re(z) 6 1, we define p(z)1 = 1− z2 and p01(z) = z2· We have p(θ) = p and p0(θ) = p0. We fix a = (an)n a sequence in R+ with a finite number of non-zero terms and a positive functiong∈Lp0, such that kak`p(w(p))=kgkp0 = 1.Finally we define
F(z) =X
n∈N
a
p p(z)
n
Z 1 0
tp(z)p λng(t)
p0 p0(z)dt .
This is actually a finite sum, andF is a holomorphic function on the band {z∈C|Re(z)∈(0,1)}. Forx∈R, we have
|F(ix)|6X
n∈N
apn Z 1
0
tpλndt=X
n∈N
apn
pλn+ 1 = 1.
On the other hand, for every real number x,
|F(1 +ix)|6X
n∈N
ap(1−n 12)
Z 1 0
tp(1−12)λng(t)p
0 2dt
= Z 1
0
g(t)p
0
2 X
n∈N
bntψndt
wherebn=a
p
n2 and Ψ = (ψn)n= pλ2n
n. Since (2ψn+1)nis alsor-lacunary we can apply Proposition 3.2. in the hilbertian case
X
n∈N
bntψn
26 1 + 4 r12 −1
!12 X
n
|bn|2 2ψn+ 1
!12
·
By Cauchy–Schwarz inequality, we get
|F(1 +ix)|6kgp
0 2k2×
X
n
bntψn 26
1 + 4 r12 −1
12 .
Now, the proof ends in a standard way and the three lines theorem gives
|F(θ)|6 1 + 4 r12 −1
!θ2 .
From this, we conclude easily that for arbitrarya∈`p(w(p)), we have kTΛw(p)(a)kp6
1 + 4 r12 −1
p10
kak`p(w(p)).
Now we can give a characterization of the boundedness of TΛw(p). Theorem 3.4. — Letp∈(1,+∞). The following are equivalent
(1) The sequenceΛ is quasi-lacunary ;
(2) The operatorTΛw(p):`p(w(p))→MΛp is bounded.
Proof. — Assume that Λ is a quasi-lacunary sequence. Using Remark 2.2, there existK >1 and lacunary sets Λj ⊂Λ (with j ∈ {1, . . . , K}) such that Λ = Λ1∪ · · · ∪ΛK. We define the operators
T(j):
`p(w(p)) −→ MΛp
b 7−→ P
nbntλn1Λj(λn) where1Λj is the indicator function of the set Λj.
We haveTΛw(p)=PK
j=1T(j). Since each Λj is lacunary, Proposition 3.2 and Proposition 3.3 (or Gurariy–Macaev theorem) imply that TΛw(p) is bounded.
For the converse, we assume that Λ is not quasi-lacunary. We denoteqk= pλk+ 1. For an arbitrarily largeN ∈N, the sequenceqN k is not lacunary since (N k)k∈Nhas bounded gaps. This implies lim infk→+∞q(k+1)Nq
kN = 1, so there existsk0 such that it is less than 2. Forn0=k0N we have
qn0+N 62qn0.
Let A={n0, . . . , n0+N −1}. Thanks to the inequality between arith- metic and geometric means, we have
kTΛw(p)(1A)kpp= Z 1
0
X
j∈A
tλj
p
dt>
Z 1 0
Np Y
j∈A
tpλjN dt.
We obtain
kTΛw(p)(1A)kpp> Np P
j∈A qj N
> Np
qn0+N > Np 2qn0· On the other hand,k1Akp`p(w(p)) =P
j∈A 1
qj 6 qNn0· SinceN is arbitrarily
large andp >1,TΛw(p)is not bounded.
The following is a refinement of Gurariy–Macaev theorem for lacunary sequences with a large ratio.
Theorem 3.5. — Letp >1. For anyε∈(0,1), there existsrε>1with the following property:
For any Λsuch that(pλn+ 1)n isrε-lacunary, we have
∀a∈`p(w(p)), (1−ε)kak`p(w(p))6
TΛw(p)(a)
p6(1 +ε)kak`p(w(p)). Remark 3.6. — If we denote q = max{p, p0}, the parameter rε =
1 +4q
1 q−1
ε
q(q−1)
is suitable for Theorem 3.5.
Proof. — Letq= max{p, p0}>2 andrε=
1 +4q
1 q−1
ε
q(q−1)
. In order to lighten the computation below, we shall writeω instead ofw(p) so that ωn=wn(p) = pλ1
n+1·
Let a be a sequence with kak`p(ω) = 1. Thanks to the above choice of rε, we have that
TΛω
p 6 1 + ε2p10
61 +2ε either by Proposition 3.2 if p>2, or by Proposition 3.3 ifp62.