Lemma 2.1
20R5Let \mathbf{M}\in\hyperlink{LCset}{\mathcal{LC}}, then \omega_{\mathbf{M}}\in\hyperlink{omset0}{\mathcal{W}_0} holds and om6(\omega_6) for \omega_{\mathbf{M}} if and only if \mathbf{M} has mg(\operatorname{mg}).
We continue the study of the known equivalent reformulations of the classical moderate growth condition for weight sequences in the mixed setting; i.e. when dealing with two different sequences. This approach is becoming crucial in the weight matrix setting and also, in particular, when dealing with weight functions in the sense of Braun-Meise-Taylor. It is known that a full generalization to the mixed setting fails, more precisely the condition comparing the growth of the corresponding sequences of quotients and roots is not clear. In the main result we prove a new characterization of this property in terms of the associated weight function; i.e. when the given weight function is based on a weight sequence.
Let \mathbf{M}\in\hyperlink{LCset}{\mathcal{LC}}, then \omega_{\mathbf{M}}\in\hyperlink{omset0}{\mathcal{W}_0} holds and om6(\omega_6) for \omega_{\mathbf{M}} if and only if \mathbf{M} has mg(\operatorname{mg}).
Let \omega:[0,+\infty)\rightarrow[0,+\infty) be nondecreasing and \lim_{t\rightarrow+\infty}\omega(t)=+\infty; i.e. \omega is a weight function in the sense of [genLegendreconj] (and of [index]). Then the following are equivalent:
(a) \omega satisfies om6(\omega_6).
(b) \omega satisfies
(c) \omega satisfies
(a)\Rightarrow(b) and (c)\Rightarrow(a) are clear (take a:=2). For (b)\Rightarrow(c) let b>1 be arbitrary, then choose n\in\mathbb{N}_{>0} such that a^n\ge b with a appearing in (b), and iterate the estimate n-times.
Let \mathbf{M}\in\hyperlink{LCset}{\mathcal{LC}}, then the moderate growth index g(\mathbf{M}) is defined by
and let us set g(\mathbf{M}):=+\infty if (?) is violated.
Let \omega\in\hyperlink{omset0}{\mathcal{W}_0} be given and \mathcal{M}_{\omega}=\{\mathbf{W}^{(\ell)}: \ell>0\} the associated weight matrix. Then
Lemma 3.1 holds, in particular, if \omega=\omega_{\mathbf{M}} for some \mathbf{M}\in\hyperlink{LCset}{\mathcal{LC}}, and in this situation the conclusion follows for \mathbf{M}=\mathbf{W}^{(1)}; see (?).
Next note that (3.1) is equivalent to
Indeed, this equivalence holds by the correspondences c=d and y=xc. On the other hand, (3.2) can be shown directly analogously as (3.1) by involving [modgrowthstrange, (4.10), (5.2)] and hence using the estimates
Summarizing, if for some index x_0>0 one has that \mathbf{W}^{(x_0)} satisfies (?) with some d\in\mathbb{N}_{>0}, then (?) also holds for any \mathbf{W}^{(cx_0)} and \mathbf{W}^{(x_0/c)} with a precise modification of the parameter d.
Let \mathbf{M},\mathbf{N}\in\hyperlink{LCset}{\mathcal{LC}} be given and consider the matrices \mathcal{M}_{\omega_{\mathbf{M}}}=\{\mathbf{M}^{(x)}: x>0\} and \mathcal{M}_{\omega_{\mathbf{N}}}=\{\mathbf{N}^{(x)}: x>0\}. If \mathbf{M}\hyperlink{preceq}{\preceq}\mathbf{N}, then
consequently \mathcal{M}_{\omega_{\mathbf{M}}}\hyperlink{Mroumpreceq}{\{\preceq\}}\mathcal{M}_{\omega_{\mathbf{N}}} and \mathcal{M}_{\omega_{\mathbf{M}}}\hyperlink{Mbeurpreceq}{(\preceq)}\mathcal{M}_{\omega_{\mathbf{N}}} is valid. More precisely, it holds that
and
Here, C in (3.3) can be chosen to be a constant such that M_p\le C^pN_p holds for all p\in\mathbb{N} and in (3.3) set C_x:=CM^{(\frac{1}{x})}_{2x}.
Concerning Lemma 3.3 we comment:
(*) In general, the converse implication in this result is unclear: \mathcal{M}_{\omega_{\mathbf{M}}}\hyperlink{Mroumpreceq}{\{\preceq\}}\mathcal{M}_{\omega_{\mathbf{N}}} implies
and so M_p^x\le C^{xp}N_{xp}. Similarly, \mathcal{M}_{\omega_{\mathbf{M}}}\hyperlink{Mbeurpreceq}{(\preceq)}\mathcal{M}_{\omega_{\mathbf{N}}} implies
Thus, in order to verify \mathbf{M}\hyperlink{preceq}{\preceq}\mathbf{N}, either \mathbf{M} or \mathbf{N} is required to satisfy moderate growth; see the proof of [orliczpaper, Thm. 4.4, (iii)\Rightarrow(ii)].
(*) Lemma 3.3 strengthens one part of the conclusions in [modgrowthstrange, Lemma 4.9 (i)]: There it has been shown that for equivalent sequences \mathbf{M},\mathbf{N}\in\hyperlink{LCset}{\mathcal{LC}} under the assumption that (at least one) sequence is strongly nonquasianalytic, see (M.3) in [Komatsu73] and e.g. (\beta_1), (\gamma_1) on [modgrowthstrange, p. 5], we have both R- and B-equivalence between the associated weight matrices. Lemma 3.3 shows that this assumption and the arguments involving (?) are superfluous.
However, in [modgrowthstrange, Lemma 4.9 (i)] even \omega_{\mathbf{M}}\hyperlink{sim}{\sim}\omega_{\mathbf{N}} was shown and which implies stronger estimates than (3.3) and (3.3): the exponential growth factors (in the variable p) can be replaced by a constant; see also e.g. (?).
Let \mathbf{M},\mathbf{L}\in\hyperlink{LCset}{\mathcal{LC}} be given. Then the following are equivalent:
(i) It holds that
(ii) It holds that
(iii) It holds that
i.e. in view of (?) and (?),
Concerning the parameter a, the proof shows the following correspondences: In (i) and (iii) we can take the same a; if (i) holds with a then (ii) with 2a and if (ii) holds with a then also (i).
Let \mathbf{M}\in\hyperlink{LCset}{\mathcal{LC}} be given. Then the following are equivalent:
(i) \omega_{\mathbf{M}} satisfies (4.1) for \mathbf{M}=\mathbf{L},
(ii) \omega_{\mathbf{M}} satisfies om6(\omega_6),
(iii) \mathbf{M} satisfies mg(\operatorname{mg}).
Let \mathbf{M},\mathbf{L}\in\mathbb{R}_{>0}^{\mathbb{N}} be given such that M_0=1=L_0 and such that \mathbf{L} is log-convex. Recall also the notation \lambda_p:=\frac{L_p}{L_{p-1}}, p\in\mathbb{N}_{>0} and \lambda_0:=1. Then the following are equivalent:
(i) It holds that
(ii) It holds that
(iii) It holds that
i.e. the mixed variant of (?) between \mathbf{L} and \mathbf{M} resp. the mixed variant of (?) between \mathbf{L} and \widetilde{\mathbf{M}}^a.
Concerning the parameter a\in\mathbb{N}_{>0} the proof shows the following correspondences: In (i) and (ii) one can choose the same a; if (i) holds with a then also (iii); if (iii) is valid for a then (ii) with 2a.
Let \mathbf{M},\mathbf{L}\in\hyperlink{LCset}{\mathcal{LC}} be given.
(i) Any of the equivalent assertions in Proposition 4.1 implies the equivalent assertions listed in Proposition 4.3 and the proof shows that if (4.1) holds with some a, then (4.3) as well.
(ii) When \mathbf{M}=\mathbf{L}, then the converse of (i) is not valid in general; in particular the q-Gevrey sequences M_p:=q^{p^2}, q>1, provide a (counter-)example.
(4.1) and (4.3) should be compared with the “weak separativity condition” (W.S.) in [matsumoto, p. 669] between the sequences \mathbf{L} and \widetilde{\mathbf{M}}^a. Indeed, for given \mathbf{L}\in\mathbb{R}_{>0}^{\mathbb{N}} this condition reads as follows:
(4.5) for the particular choice \mathbf{M}=\widetilde{\mathbf{M}}^a obviously implies both (4.1) and (4.3). If \mathbf{L} is log-convex, then (4.5) gives \lambda_p^p\le\lambda_{p+1}\cdots\lambda_{2p}=\frac{L_{2p}}{L_p}\le A^pM_p and so \lambda_p\le A(M_p)^{1/p} for all p\in\mathbb{N}_{>0}; i.e. the mixed variant of (?) resp. (?) between \mathbf{L} and \mathbf{M}.
Let \mathbf{M},\mathbf{L}\in\hyperlink{LCset}{\mathcal{LC}} be given. Consider the following assertions:
(i) (4.3) holds; i.e.
(ii) It holds that
(iii) It holds that
(iv) It holds that
(v) It holds that \mathbf{L}\hyperlink{preceq}{\preceq}\widetilde{\mathbf{M}}^{2a}\star\widetilde{\mathbf{M}}^{2a}, so
(vi) It holds that
Then one has the following implications resp. equivalences:
Finally, concerning (vi)\Rightarrow(i) we can show a partial converse: If \mathbf{L} satisfies (?) for some d\in\mathbb{N}_{>0} and
then (4.3) is valid with a:=db.
Let \mathbf{M} be a given log-convex sequence such that M_0=1 and satisfying (?) for some d\in\mathbb{N}_{>0}. Then any log-convex sequence \mathbf{N} being equivalent to \mathbf{M} (and with N_0=1) has (?) too; i.e. (?) is preserved under equivalence of log-convex sequences.
Indeed, if \mathbf{M} satisfies (?) with d, then for any equivalent log-convex \mathbf{N} we can choose 2d.
By combining Theorem 4.7 and Lemma 3.3 and in view of [modgrowthstrange, Prop. 4.4] we have the fact that both quotient-root comparison properties are preserved when switching to a (R- and B-equivalent) weight matrix which is generated by an equivalent weight sequence. Therefore, the conclusion in [modgrowthstrange, Lemma 4.9 (ii)] holds even without part (b) in [modgrowthstrange, Thm. 4.8]. Thus this part and also [modgrowthstrange, Lemma 4.6] are becoming superfluous.
Let \omega\in\hyperlink{omset0}{\mathcal{W}_0} be given and let \mathcal{M}_{\omega}=\{\mathbf{W}^{(\ell)}: \ell>0\} be the associated weight matrix. Consider the following assertions:
(i) \omega satisfies om6(\omega_6).
(ii) It holds that
(iii) It holds that
(iv) It holds that
Then (i)\Rightarrow(ii)-(iv) and (iii),(iv)\Rightarrow(ii); if in assertion (ii) for the parameters the relation \ell_1>\ell a holds, then (ii)\Rightarrow(i).
Let \omega\in\hyperlink{omset0}{\mathcal{W}_0} be given and let \mathcal{M}_{\omega}=\{\mathbf{W}^{(\ell)}: \ell>0\} be the associated weight matrix. If
then \omega satisfies om6(\omega_6).
Let \mathbf{M},\mathbf{N}\in\hyperlink{LCset}{\mathcal{LC}} be given, then the following are equivalent:
(i) It holds that
(ii) It holds that
The parameter \ell can be chosen to be the same in (5.1) and (5.1); moreover we have the correspondences A_1=e^C and B=A.
Let \mathbf{M},\mathbf{N}\in\hyperlink{LCset}{\mathcal{LC}} be given, then
with
Let \mathbf{M},\mathbf{N},\mathbf{L}\in\hyperlink{LCset}{\mathcal{LC}} be given, then the following are equivalent:
(i) It holds that
(ii) It holds that
The proof shows that we can take in both estimates the same parameter a and note that when (5.3) holds with some a, then also for all a'\ge a.
Let \mathbf{M},\mathbf{L}\in\hyperlink{LCset}{\mathcal{LC}} be given. Then the following are equivalent:
(i) (4.3) holds; i.e.
(ii) It holds that
(iii) (4.3) holds; i.e.
In (i) and (ii) we can take the same a; if (i) holds with a then also (iii); if (iii) is valid for a then (i) with 2a.
Let \mathbf{M}\in\hyperlink{LCset}{\mathcal{LC}} be given, then the following are equivalent:
(i) \mathbf{M} satisfies (?).
(ii) The associated weight function satisfies
If (?) holds with some d\in\mathbb{N}_{>0}, then choose a:=2d in (5.5); if (5.5) is valid for some a\in\mathbb{N}_{>0}, then choose d:=a in (?).
If (5.5) holds with a=1, then \mathbf{M}^a=\mathbf{M} and by (?) this estimate gives
i.e. om6(\omega_6) for \omega_{\mathbf{M}}. And by the known characterization \mathbf{M} satisfies (?) with d=1. Indeed, in view of (?) condition om6(\omega_6) is precisely (5.5) with a=1.
Let \mathbf{M},\mathbf{N},\mathbf{L}\in\hyperlink{LCset}{\mathcal{LC}} be given.
(i) If one has
then (5.3) holds for any a\in\mathbb{N}_{>0} with the same parameters A,C.
(ii) If (5.3) holds, then
Indeed, if (5.3) is valid with some a, then in (5.7) it suffices to choose 2a.
The previous result should be compared with the following comment: When \mathbf{M}, \mathbf{N} are log-convex and normalized then
see Section ?, and consequently the definition of the associated function gives
and a sufficient condition for having (?) is
By taking into account (?) this property turns out to be om6(\omega_6) and hence this is consistent with the known characterization.
Let \omega\in\hyperlink{omset0}{\mathcal{W}_0} be given and let \mathcal{M}_{\omega}=\{\mathbf{W}^{(\ell)}: \ell>0\} be the associated weight matrix. Then the following are equivalent:
(i) (?) is valid.
(ii) \mathcal{M}_{\omega} satisfies (?) and/or (?).
Let \omega\in\hyperlink{omset0}{\mathcal{W}_0} be given. Then (?) holds if and only if \omega satisfies om6(\omega_6).
Let \omega be nondecreasing and satisfying \lim_{t\rightarrow+\infty}\omega(t)=+\infty, i.e. \omega is a weight function in the sense of [genLegendreconj] (and of [index]), then we immediately get that om6(\omega_6) for \omega implies:
Note that this implication also holds by [genLegendreconj, Rem. 2.3, Thm. 3.4 (ii)] where the growth index \overline{\gamma}(\omega) from [genLegendreconj, Sect. 2.4] is involved in the arguments. However, in general the converse implication is not clear and note that for the arguments in Lemma 5.10 crucially (?) is used and for this, more precisely for the second estimate there, the convexity condition om4(\omega_4) is required.
Let \omega\in\hyperlink{omset0}{\mathcal{W}_0} be given. Then \omega satisfies om1(\omega_1) if and only if
Let \sigma\in\hyperlink{omset0}{\mathcal{W}_0} be given with associated weight matrix \mathcal{M}_{\sigma}=\{\mathbf{S}^{(\ell)}: \ell>0\}.
(i) Assume that there exists a weight matrix \mathcal{N}=\{\mathbf{N}^{(x)}: x\in\mathcal{I}\} such that \mathcal{M}_{\sigma}\{\approx\}\mathcal{N} and such that
(ii) \mathcal{N} satisfies the quotient-root comparison property of Roumieu type (?).
Then \mathbf{S}^{(1)} has to satisfy
If one assumes instead that \mathcal{M}_{\sigma}(\approx)\mathcal{N} and such that \mathcal{N} satisfies the quotient-root comparison property of Beurling-type (?), then \mathbf{S}^{(1)} has to satisfy
The text on this page is from arXiv:2505.08839v2 by Gerhard Schindl, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.
The record of each item is in analysis/objects/2505/2505.08839.jsonl.