- Home
- Documents
*Javier Jimآ´enez-Garrido, Javier Sanz and Gerhard Schindl ... Javier Jimآ´enez-Garrido, Javier...*

prev

next

out of 26

View

0Download

0

Embed Size (px)

ar X

iv :1

60 7.

08 02

7v 2

[ m

at h.

C A

] 1

5 Fe

b 20

18

Log-convex sequences and nonzero proximate orders

Javier Jiménez-Garrido, Javier Sanz and Gerhard Schindl

October 11, 2018

Abstract

Summability methods for ultraholomorphic classes in sectors, defined in terms of a strongly regular sequence M = (Mp)p∈N0 , have been put forward by A. Lastra, S. Malek and the second author [10], and their validity depends on the possibility of associating to M a nonzero proximate order. We provide several characterizations of this and other re- lated properties, in which the concept of regular variation for functions and sequences plays a prominent role. In particular, we show how to construct well-behaved strongly regular sequences from nonzero proximate orders.

Keywords: Log-convex sequences; regular variation; proximate orders; Carleman ultraholo- morphic classes.

AMS 2010 Classification: 26A12, 40A05, 26A51, 30D60.

1 Introduction

A general, common treatment of summability in Carleman ultraholomorphic classes in sec- tors, which extends the powerful theory of k−summability dealing with Gevrey formal power series, has been put forward by A. Lastra, S. Malek and the second author [10]. The tech- nique consisted in the construction of pairs of kernel functions with suitable asymptotic and growth properties, in terms of which to define formal and analytic Laplace- and Borel-like transforms which allow one to explicitly recover the sum of a summable formal power series in a direction. The main inspiration came from the theory of moment summability methods developed by W. Balser in [1, Section 5.5], and required the notion of proximate order, appearing in the study of growth properties of holomorphic functions in sectors.

The Carleman ultraholomorphic classes ÃM(G) we deal with are those consisting of holomorphic functions f admitting an asymptotic expansion f̂ =

∑ p≥0 apz

p, in a sectorial region G with vertex at 0, with remainders suitably bounded in terms of a sequence M = (Mp)p∈N0 of positive real numbers: for every bounded and proper subsector T of G, there exist CT , AT > 0 such that for every p ∈ N0 and z ∈ T , one has

∣∣∣f(z)− p−1∑

k=0

akz k ∣∣∣ ≤ CTApTMp|z|p.

We will mostly consider logarithmically convex sequences M with quotients mp := Mp+1/Mp tending to infinity, and frequently our attention will focus on strongly regular sequences as defined by V. Thilliez [22], of which the best known example is that of Gevrey classes, appearing when the sequence is chosen to be (p!α)p∈N0 , α > 0.

The second author introduced in [19] the constant

ω(M) = lim inf p→∞

log(mp)

log(p) ∈ (0,∞),

1

http://arxiv.org/abs/1607.08027v2

measuring the rate of growth of any strongly regular sequence M. Whenever the associated function dM(t) = log(M(t))/ log t, where

M(t) := sup p∈N0

log ( tp Mp

) , t > 0, (1)

is a nonzero proximate order, one can provide nontrivial flat functions in optimal sectors of opening π ω(M), and this is the crucial point for the success in putting forward a satisfactory summability theory in ÃM(G) (see [10]). So, it seemed important to characterize the fact that dM is a nonzero proximate order in a simple way, what was achieved in the paper [8] by the first two authors. The main result was the following.

Theorem 1.1 ([8], Th. 3.14). Let M be a strongly regular sequence, then the following are equivalent:

(i) dM(t) is a proximate order,

(ii) limt→∞ dM(t) = 1/ω(M),

(iii) limp→∞ log(mp)/ log(p) = ω(M),

(iv) limp→∞ log ( mp/M

1/p p

) = ω(M).

These conditions are satisfied for every strongly regular sequence appearing in the appli- cations (ODEs, PDEs and difference equations; see the introduction of [8] and the references therein for a non-complete account). However, it was not clear for us whether these condi- tions held true for any strongly regular sequence, so we investigated on some pathological examples. To our surprise, we found an example (see Example 3.12 in this paper) satisfying (iii) but not (iv). It turns out that in our (wrong) proof that (iii) implies (iv), a very recent criterion by F. Moricz [14] for the convergence of a sequence summable by Riesz means played a prominent role.

Theorem 1.2 ([14], Th. 5.1). If a sequence (sk)k∈N of real numbers is Riesz-summable to some A ∈ R, then the ordinary limit exists (with the same value) if, and only if,

lim sup λ→1+

lim inf p→∞

1

(⌊pλ⌋ − p)Hp

⌊pλ⌋∑

k=p+1

sk − sp k

≥ 0 (2)

and

lim sup λ→1−

lim inf p→∞

1

(p− ⌊pλ⌋)Hp

p∑

k=⌊pλ⌋+1

sp − sk k

≥ 0, (3)

where ⌊·⌋ denotes the integer part, and Hp is the p−th partial sum of the harmonic series. By carefully inspecting this result, we noticed that its statement is not correct, although a

right one may be deduced from the final lemma 5.5 in Moricz’s paper, where the expressions in (2) and (3) are rewritten. Indeed, they should read as follows:

lim sup λ→1+

lim inf p→∞

1

H⌊pλ⌋ −Hp

⌊pλ⌋∑

k=p+1

sk − sp k

≥ 0

and

lim sup λ→1−

lim inf p→∞

1

Hp −H⌊pλ⌋

p∑

k=⌊pλ⌋+1

sp − sk k

≥ 0.

This fact made clear to us that several implications in Theorem 1.1 were false. In Section 3 the suitable corrections will be carefully described (see, in particular, Remark 3.11).

Apart from the necessity to recover from this mistake, there are some important points to note. Firstly, as indicated in [19, Remark 4.11(iii)], in order to obtain the summability

2

theory in ÃM(G) mentioned above, it is not crucial that dM is a nonzero proximate order, but rather that

(*) there exist a nonzero proximate order ρ(t) and constants A,B > 0 such that

A ≤ log(t) ( dM(t)− ρ(t)

) ≤ B for t large enough,

since these estimates allow for the obtention of the required kernels, integral transforms and asymptotic relations. Secondly, and in the same line of ideas, there is some flexibility in the definition of the space ÃM(G): If L = (Lp)p∈N0 is another sequence of positive real numbers and it is equivalent to M (in the sense that there exist C,D > 0 such that DpLp ≤ Mp ≤ CpLp for every p ∈ N0), then ÃM(G) = ÃL(G). So, even if dM is not a nonzero proximate order, it makes sense to wonder whether

(**) there exists a sequence L equivalent to M and such that dL is a nonzero proximate order.

The main aim of this paper is to provide statements, as accurate as possible (in the sense that they impose the least restrictive hypotheses on the sequence M), clarifying the equivalences or implications between the different properties (i), (ii), (iii), (iv), (*) and (**), together with the property of regular variation of the sequence of quotients, which already appeared in [8]. As a byproduct we obtain a new characterization of regularly varying sequences.

It will be specially interesting the fact that, following a classical idea of H. Komatsu [9], one can also go from nonzero proximate orders to well-behaved strongly regular sequences, so reversing the way from suitable sequences M to proximate orders dM. In this respect the results of L.S. Maergoiz [12], on the construction of holomorphic functions in sectors whose restriction to the positive real axis has a growth accurately specified by a given nonzero proximate order, will be extremely useful.

We will also show several examples that prove that, in some cases, our results are sharp. In particular, Example 3.12 provides a strongly regular sequence not satisfying (iv) which, nevertheless, is equivalent to a Gevrey sequence (for which the corresponding function d is known to be a proximate order). Example 4.16 gives a strongly regular sequence M for which the limit in (iii) exists and with equal indices ω(M) and γ(M) (this last constant was introduced by V. Thilliez [22]) and which, however, does not admit a proximate order. Finally, Example 4.18 shows a strongly regular sequence for which the limit in (iii) does not exist, so solving another open question.

2 Preliminaries

This section is devoted to provide all the necessary information regarding proximate orders and logarithmically convex sequences.

2.1 Proximate orders

We recall the notion of proximate orders, appearing in the theory of growth of entire func- tions and developed, among others, by E. Lindelöf, G. Valiron, B.Ja. Levin, A.A. Goldberg, I.V. Ostrosvkii and L.S. Maergoiz (see the references [24, 11, 6, 12]).

Definition 2.1. We say a real function ρ(r) defined on (c,∞) is a proximate order, if the following hold:

(A) ρ is continuous and piecewise continuously differentiable in (c,∞) (meaning that it is differentiable except possibly at a sequence of points, tending to infinity, at any of which it is continuous and has distinct finite lateral derivatives),

(B) ρ(r) ≥ 0 for every r > 0, (C) limr→∞ ρ(r) = ρ

In case the value ρ in (C) is positive (respectively, is 0), we say ρ(r) is a nonzero (resp. zero) proximate order.

Definition 2.2. Two proximate orders ρ1(r) and ρ2(r) are said to be equivalent if

lim r→∞

(ρ1(r) − ρ2(r)) lo