On the conjugate weight function and ultradifferentiable classes of entire functions Authors: Gerhard Schindl Published: 2026 ABSTRACT AbstractWe introduce the new notion of a conjugate weight function and provide a detailed study of this operation and its properties. Then we apply this knowledge to study classes of ultradifferentiable functions defined in terms of fast growing weight functions in the sense of Braun–Meise–Taylor and hence violating standard regularity requirements. Therefore, we transfer recent results shown by the author and D.N. Nenning from the weight sequence to the weight function framework. In order to proceed and to complete the picture we also define the conjugate associated weight matrix and investigate the relation to conjugate weight sequences via the corresponding associate weight functions. Finally, as it has already been done in the weight sequence case, we generalize results by M. Markin from the small Gevrey-setting and show how the corresponding non-standard ultradifferentiable function classes can be used to detect boundedness of normal linear operators on Hilbert spaces (associated with an evolution equation problem). On the one hand, when involving the weight matrix here the crucial information concerning regularity of the weak solutions can be expressed in terms of only one weight, namely of the given weight function. But, on the other hand, for the connection to the weighted entire setting the required conditions on the weight function are too restrictive in the general case.Keywords: Weight functions and weight sequences, Associated weight functions, Conjugate weight function, Growth and regularity properties for functions and sequences, Boundedness of linear operators Introduction This article is the direct continuation of the author’s (joint) works [12, 16, 17], and the aim is to transfer the main notions and results from the weight sequence case treated in [12] to the (ultradifferentiable) weight function setting. In [12] we have been interested in introducing and studying ultradifferentiable function classes defined in terms of non-standard (small) weight sequences M. The goal was to emphasize the meaning of such “exotic classes” which are strictly smaller than the space of all real-analytic functions and, in particular, to generalize results by M. Markin from [10] dealing with small Gevrey sequences, i.e. when Gs:=(p!s)p∈N with 00}, i.e. a one-parameter family of weight sequences with parameter ℓ>0. Mω can be used equivalently to describe the corresponding ultradifferentiable spaces; this is shown in [14, 20] and see Sect. 5.2 for precise definitions and a summary. Thus it seems to be possible and reasonable to consider F=Mω in the results from [12, Sect. 4] generalizing M. Markin’s approach. In this situation the crucial information concerning regularity of (weak) solutions from (1.1) can be expressed in terms of one given weight function ω in a unified way. From this point of view the weight function setting seems to be more natural for the problem studied by M. Markin since by involving the matrix parameter the weight ω admits naturally more flexibility in order to test regularity of the solutions from (1.1). Related to these comments and in order to complete the picture it is also the problem to see how the two notions of conjugation in the weight sequence and weight function setting are related to each other and how they are interacting. More precisely, the goal is to understand the connections between the functions ωM∗ and ωM∗ and between the matrices Mω∗:={(W(ℓ))∗:ℓ>0} and Mω∗. The paper is structured as follows: First, in Sect. 2 we gather all relevant definitions and growth conditions concerning weight functions. In Sect. 3 we define the conjugate weight function ω∗ and provide a detailed study of this operation: In Sect. 3.1 we investigate basic properties of ω∗ and introduce also the double-conjugate ω∗∗; in Sect. 3.2 we show how the conjugate operation is interacting with the known growth relations between weight functions from Sect. 2.2. Then, in Sect. 3.3 we study the interaction of the conjugate with the generalized lower and upper Legendre envelopes ⋆ˇ and ⋆^ from [17], see the summary in Sect. 2.4 and the main result Proposition 3.9. In Sect. 3.4 we show how the conjugate operation is modifying the growth indices γ(·) and γ¯(·) for weight functions defined and studied in [6], see Theorem 3.12 (for precise definitions of γ(·) and γ¯(·) see Sect. 2.3). Section 4 is dedicated to the weight sequence setting; i.e. when ω≡ωM and M∈R>0N satisfies basic growth properties. In Corollary 4.10 it is shown that both ωM∗ and ωM∗ are simultaneously well defined. The main result is Theorem 4.15 (and its Corollary 4.17) where the expected relation between ωM∗ and ωM∗ is established. Analogously as before, in Lemma 4.11 we investigate the interaction with growth relations (between weight sequences) and Theorem 4.13 provides information about the interplay with the generalized lower and upper Legendre envelopes. Then, in Sect. 5 we apply all this information to a BMT-weight function ω and its associated matrix Mω. We introduce the corresponding conjugate associated matrix Mω∗ (see (5.7)) and study its behavior in Proposition 5.5: On the one hand this set is always well defined but, on the other hand, the identity Mω∗=Mσ for a BMT-weight σ already implies the restriction that both matrices Mω and Mω∗ are constant, i.e. all sequence are equivalent, and hence one loses flexibility w.r.t. the matrix parameter. Similarly, this happens in the main results Theorems 5.6 and 5.7 where it is shown that assumption γ¯(ω)<1 is convenient to obtain more information but also restrictive. In view of this it turns out that whenever the conjugate objects enjoy desired standard growth conditions, then this forces already for the originally given weight (matrix) the restriction mentioned before and the fact that the (ultradifferentiable) classes defined via ω (resp. equivalently Mω) can already be described and defined via a single weight sequence; see the comments in Remark 5.8 and Sect. 5.4. All results mentioned so far can have more applications for other analogously defined weighted settings since we exclusively investigate weights and their growth properties. Moreover, in Sect. 6 we provide a concrete application and transfer the results from [12, Sect. 4] to the BMT-weight function setting. On the one hand it turns out that some statements hold without any further assumption which is basically due to the fact that (each) W(ℓ) is automatically very regular; i.e. it belongs to the set LC (see Sect. 4.1). But, on the other hand, the technical issue described in the previous paragraph (concerning Sect. 5) also yields difficulties and restrictions of the generality in this direction: By the explanations in (d) in Sect. 6 the technical result [12, Prop. 4.5] has to be modified resp. generalized accordingly; see Proposition 6.2. This result is required for constructing a crucial specific (uniform bound) sequence a for the matrix Mω which is then used to obtain Corollary 6.5 and Theorem 6.6 even in the general (non-constant) case. (Proposition 6.2 (also) slightly improves the results in [12, Sect. 4].) In Proposition 6.9 we construct a “non-constant example” of a weight function to which Theorem 6.6 can be applied; here ω=ωM with M satisfying appropriate growth conditions and Theorem 6.6 holds for the non-constant matrix MωM. However, in the second main result Theorem 6.8 when combining the information from Theorem 6.6 and the equivalence from [12, Thm. 3.4], see also Corollary 4.16 and Sect. 5.4, then by the required assumptions on the weight in the latter result establishing the isomorphism to certain weighted spaces of entire functions, again we have to restrict to the constant case. Therefore, the constructed example also illustrates the different nature of Theorems 6.6 and 6.8. Finally, in Appendix A we study the property that ωM is slowly varying (see (2.3)) in terms of the defining sequence M; see Proposition A.1. This question is inspired by the reflexivity of a crucial growth relation from Sect. 2.2 and concrete applications of this technical statement are expected, especially when constructing (counter-)examples. Weight functions and growth properties General notation We write N:={0,1,2,⋯}, N>0:={1,2,⋯}, occasionally also R>0:=(0,+∞), and for x∈Rd the expression |x| denotes the usual Euclidean norm on Rd. Basic definitions and growth relations First, let us recall the general notion of a weight function in the sense of [6, 17]. Definition 2.1 ω:[0,+∞)→[0,+∞) is called a weight function if ω is non-decreasing and limt→+∞ω(t)=+∞. We are going to use the following notation for any (weight) function ω:[0,+∞)→[0,+∞) and arbitrary α>0: set ωι(t):=ω(1t), t>0, and ω1/α(t):=ω(t1/α), t≥0. Note that ω1/α is obtained by a power-substitution; ω is a weight function according to the previous definition if and only if ω1/α is so for some/any α>0. Moreover, (ωι)1/α(t)=(ω1/α)ι(t) for all t>0 and finally write id1/α for t↦t1/α. The function id1/α is also called the Gevrey weight with index α>0. The next two conditions are natural for weight functions and appear frequently in the literature for Braun–Meise–Taylor weight functions, see e.g. [2], and the abbreviations are motivated by the notation already used in [20]. We also refer to Sect. 5 for more details: ω(2t)=O(ω(t)) as t→+∞, i.e. ∃L≥1∀t≥0:ω(2t)≤L(ω(t)+1). ∃H≥1∀t≥0:2ω(t)≤ω(Ht)+H. Next, let us recall some crucial growth relations between weight functions, see [18, Sect. 2.3]: We write σ⪯τ if 2.1τ(t)=O(σ(t))ast→+∞, σ⊲τ if 2.2τ(t)=o(σ(t))ast→+∞. σ and τ are called equivalent, written σ∼τ, if σ⪯τ and τ⪯σ. ⪯ is obviously reflexive, relation ⊲ implies ⪯ but ⊲ is not reflexive because each weight functions tends to infinity. Moreover, we write σ⪯cτ if ∃h,C>0∀t≥0:τ(t)≤σ(ht)+C, σ⊲cτ if ∀h>0∃Ch>0∀t≥0:τ(t)≤σ(ht)+Ch. We write σ∼cτ if σ⪯cτ and τ⪯cσ. ⪯c is reflexive and ⊲c implies ⪯c but ⊲c is in general not reflexive: If σ⊲cσ holds then σ has to be slowly varying, i.e. 2.3∀u>0:limt→+∞σ(tu)σ(t)=1; see [1, (1.2.1)], [6, Sect. 2.2], the comments at the end of [18, Sect. 2.3] and Appendix A for more details. Growth indices γ(ω) and γ¯(ω) We briefly recall the definitions of the growth indices γ(ω) and γ¯(ω); see [6, Sect. 2.3 (6), Sect. 2.4 (7)] and the references in these sections. Weight functions in the sense of Definition 2.1 are sufficient and convenient in order to introduce and work with the crucial indices γ(ω) and γ¯(ω). Let ω be a weight function and γ>0. We say that ω has property (Pω,γ) if If (Pω,γ) holds for some K>1, then (Pω,γ′) is satisfied for all γ′≤γ with the same K since ω is non-decreasing. Moreover we can restrict to γ>0, because for γ≤0 condition (Pω,γ) is satisfied for any weight function ω, again because ω is non-decreasing and K>1. Then put and if none condition (Pω,γ) (with γ>0) holds then set γ(ω):=0. Analogously, for γ>0 we say that ω has property (P¯ω,γ) if If (P¯ω,γ) holds for some A>1, then (P¯ω,γ′) is satisfied for all γ′≥γ with the same A since ω is non-decreasing. Moreover, we can restrict to γ>0 because for γ≤0 condition (P¯ω,γ) is never satisfied for any weight function (ω is assumed to be non-decreasing and A>1). Then set We obtain (0≤)γ(ω)≤γ¯(ω), see [6, Sect. 2.3 and 2.4], and by definition the following identities are valid: The indices γ(·) and γ¯(·) are preserved under ∼; this follows from the characterizations in the main results [6, Thm. 2.11 and 2.16], see also [6, Rem. 2.12] and the comments after [6, Thm. 2.16]. Moreover, conditions (ω1) and (ω6) are equivalent reformulations of these growth indices as follows. Remark 2.2 Let ω be a weight function. Then ω has (ω1) if and only if γ(ω)>0, see [6, Thm. 2.11, Rem. 2.12, Cor. 2.14]. And by [6, Thm. 2.16, (7), Cor. 2.17] condition (ω6) holds if and only if γ¯(ω)<+∞. Generalized lower and upper Legendre conjugates Let σ,τ be weight functions according to Definition 2.1. We introduce the generalized lower Legendre conjugate by and the generalized upper Legendre conjugate by We also have σ⋆ˇτ(0)=σ(0)+τ(0) (see [17, Rem. 3.1]) and set σ⋆^τ(0):=σ(0)-τ(0) which is justified by [17, Lemma 4.1 (b)]. These operations and their effects on the growth indices γ(·), γ¯(·) are studied in detail in [17, Sect. 3 and 4] for (general) weight functions and in [17, Sect. 5] the weight sequence setting is studied; i.e. when σ=ωM, τ=ωN and so focusing on associated weight functions. Concerning this setting we refer to Sect. 4.2 for more details. ⋆ˇ yields again a weight function but, in general, for ⋆^ this is not clear; see requirements (A) and (B) in [17, Sect. 4.2]: First, in order to ensure σ⋆^τ(t)≥0 for all t it suffices to assume that τ(0)=0, see [17, Lemma 4.2]. Second, in order to guarantee that σ⋆^τ is well defined, i.e. that σ⋆^τ(t)<+∞ for all t, it is required to have This condition corresponds to (B) in [17, Sect. 4.2] resp. to [17, (4.4)] with t0=+∞. It is immediate and mentioned at the end of [18, Sect. 2.3] that we have (2.11) if and only if τ⊲cσ; i.e. this relation gives the fact that σ⋆^τ is well defined on [0,+∞). The conjugate weight function Definition and basic properties Let ω be a weight function according to Definition 2.1. Then introduce the conjugate function and this implies the following properties: By definition ω∗ is non-decreasing and convex since it is given as the supremum of convex functions (indeed the supremum of affine lines). One has 3.2∀s≥0:ω∗(s)≥max{s2-ω(s),s-ω(1)}. (3.2) implies, in particular, that log(s)=o(ω∗(s)) as s→+∞ and hence lims→+∞ω∗(s)=+∞. ω∗(0)=-ω(0) because ω is non-decreasing and therefore, since also ω∗ is non-decreasing, in order to ensure ω∗(s)≥0 for all s≥0, one has to assume ω(0)=0. In this case ω∗(0)=0 holds, too. The next result characterizes the situation when ω∗ is well defined on [0,+∞). Lemma 3.1 Let ω be a weight function, then the following are equivalent: ω∗(s)<+∞ for all s≥0. t=o(ω(t)) as t→+∞; i.e. ω⊲id. Proof (i)⇒(ii) By assumption thus lim supt→+∞stω(t)≤1 for all s≥0 since ω is a weight function. As s→+∞ this verifies (ii). (ii)⇒(i) The relation t=o(ω(t)) precisely means Fix ϵ>0, then this estimate implies tϵ-ω(t)≤Dϵϵ for all t≥0 and so, by definition, one has ω∗(ϵ-1)≤Dϵϵ. Since ϵ>0 is arbitrary we are done. □ Let ω be a given weight function, then in view of the above observations and Lemma 3.1 in order to ensure that ω∗ is a weight function according to Definition 2.1 and satisfying ω∗(0)=0, one shall assume the following properties: ω(0)=0, t=o(ω(t)) as t→+∞, i.e. ω⊲id. Note that (C2) is obviously preserved under equivalence of weight functions and it is also stable w.r.t. ∼c. The next result states that ω∗ automatically has (C2). Lemma 3.2 Let ω be a weight function, then t=o(ω∗(t)) as t→+∞ is valid. Proof The desired relation is equivalent to supt≥0{t-ω(t)/s}=ω∗(s)s→+∞ as s→+∞. And this holds since for any given C≥1 large there exists some tC≥0 such that tC-ω(tC)/s≥C⇔s(tC-C)≥ω(tC) for all sufficiently large s≥0: e.g. take tC:=2C. □ Note: Formally, for the proof of Lemma 3.2 it is not required to assume (C2) for ω. If (C2) fails then ω∗(s)=+∞ for all sufficiently large s, see Lemma 3.1 (ω∗ is not a weight function) and in this case t=o(ω∗(t)) is formally trivial. Example 3.3 Let σ:=id1/α for 0<α<1 and so both (C1) and (C2) are satisfied. Then σ∗(s)=supt≥0{st-t1/α} and we compute this supremum by direct calculations: For s=0 we have σ∗(0)=0 since σ(0)=0 and for any s>0 let fs(t):=st-t1/α, t∈[0,+∞). Then fs′(t)=s-1αt1α-1=s-1αt1-αα, so fs′(t)=0 if and only if (sα)α1-α=t and this value gives the (unique) point where the maximum of fs is attained on [0,+∞). And since σ∗(s)=fs((sα)α1-α)=s(sα)α1-α-(sα)11-α=s11-α(αα1-α-α11-α) we see that σ∗∼id11-α. Note that by the general assumption on the parameter α we have that 0<1-α<1 as well and hence id1/(1-α) is a weight function, too. Note also that αα1-α-α11-α>0⇔αα>α⇔1>α1-α. This example implies, in particular, that up to equivalence the Gevrey-weights with indices 0<α<1 are preserved under taking the conjugate operation and that id2:t↦t2 is self-conjugate. This last fact should be compared with the information from [12, Lemma 2.6] and [17, Ex. 2.9]. By combining Lemmas 3.1 and 3.2 and motivated by the previous example, when ω is a given weight function satisfying (C1) and (C2), then it is natural to consider the double-conjugate ω∗∗ given by Thus ω∗∗ is again a weight function satisfying ω∗∗(0)=0. Moreover, one immediately infers that Remark 3.4 Let ω be a weight function such that (C2) is violated. Then there exists some s0∈[0,+∞) such that ω∗(s)=+∞ for all s>s0. Therefore one might set in this case ω∗∗(t):=sups∈[0,s0){ts-ω∗(s)} for all t≥0 and (3.3) still holds. Finally, Lemma 3.2 becomes in this case trivial. Next we investigate the converse estimate in (3.3) and ask: Under which assumptions on ω one can expect that at least ω∗∗∼ω holds? We prove the following result which is inspired by (the proof of) [13, Prop. 1.6]: Lemma 3.5 Let ω be a weight function satisfying (C1) and (C2). Moreover, assume that ω is convex, then Note: The weights treated in Example 3.3 are convex and thus this lemma is consistent with the computations made for these particular weights. Proof First, by (C1) and (C2) the function ω∗ is well defined and note that by the above comments and assumptions ω(0)=0=ω∗∗(0). So let now t>0 be given and fixed. Let u≥0 with u≠t, then by convexity one has that u↦ω(u)-ω(t)u-t is non-decreasing and since ω is non-decreasing the quotient is non-negative (for any u≠t). There exists s≥0 such that Multiplying these estimates with u-t we get that s(u-t)≤ω(u)-ω(t) for all u≥0 and note that for u=t this estimate is becoming trivial. Consequently infs≥0supu≥0{s(u-t)-ω(u)}≤-ω(t), equivalently -infs≥0supu≥0{s(u-t)-ω(u)}≥ω(t) for all t>0. The left-hand side yields -infs≥0supu≥0{s(u-t)-ω(u)}=sups≥0{-supu≥0{s(u-t)-ω(u)}}=sups≥0{st-supu≥0{su-ω(u)}}=ω∗∗(t) and so the converse of (3.3) for all t>0. Thus (3.4) is shown. □ On growth relations The aim is to study how the growth relations between weight functions are transferred under applying the conjugate operation. First, given a weight function ω with (C1) and (C2), then writing ωc(t):=ω(ct) for c>0 one immediately infers Lemma 3.6 Let σ and τ be weight functions. Assume that both weights satisfy (C1) and such that (C2) is valid for τ. If σ⪯τ, then σ∗ is also well defined, i.e. σ satisfies (C2) too, and 3.5∃C≥1∀s≥0:Cσ∗(s)≤τ∗(sC)+C. Conversely, (3.5) implies σ∗∗⪯τ∗∗ and the fact that σ∗ is well defined. If σ⊲τ, then σ∗ is also well defined and 3.6∀0c>0 (small) there exists some Dc≥1 such that τ(t)≤cσ(t)+Dc for all t≥0. Consequently, the conclusions follow then by the same computations as in (i). (a) W.l.o.g. one can assume that C≥2 in relation σ⪯τ and hence (3.5) implies 2σ∗(s)≤τ∗(sC)+C for all s≥0 which is the mixed version of (ω6). If τ∗ satisfies in addition (ω1), then choose n∈N>0 such that 2n-1≥C with C denoting the constant appearing in (3.5). Iteration of (ω1) implies Cσ∗(2s)≤τ∗(2Cs)+C≤τ∗(2ns)+C≤Lτ∗(s)+L+C for some L≥1 and all s≥0. Similarly, if σ∗ satisfies in addition (ω1), then Cσ∗(2s)≤CLσ∗(C-1s)+CL≤Lτ∗(s)+CL+CL for some L≥1 and all s≥0. Both derived estimates yield the conclusion. (b) Since σ⊲τ is stronger than σ⪯τ, the mixed variant of (ω6) follows by (a) and the mixed variant of (ω1) holds by choosing c:=12 in (3.6). □ Note: In the proof of (a) any choice n∈N>0 such that 2n-1≥C holds is sufficient to conclude and, indeed, one may select the minimal n. This comment applies to any argument involving iteration techniques; in particular we use this in several proofs in this article. Corollary 3.7 Let ω be a weight function satisfying (C1) and (C2). Moreover, assume that ω is equivalent to a convex weight function σ satisfying (C1). Then ω and ω∗∗ are equivalent; more precisely: Proof First, recall that (C2) is preserved under equivalence and hence σ shares this property, too. Then (i) in Lemma 3.6 gives that ω∗∗ and σ∗∗ are equivalent and by Lemma 3.5 we infer that σ(t)=σ∗∗(t) for any t≥0. Combining everything yields the desired relations stated in (3.7). □ Next we treat ⪯c, ⊲c and verify that these relations are reversed when taking the conjugate. Thus these relations can be considered to be more natural when dealing with conjugate weight functions. Lemma 3.8 Let σ and τ be weight functions. Assume that both weights satisfy (C1) and such that (C2) is valid for τ. If σ⪯cτ, then 3.8∃h,C≥1∀s≥0:σ∗(s)≤τ∗(sh)+C; i.e. τ∗⪯cσ∗. Moreover, σ∗ is also well defined, i.e. σ satisfies (C2), too. Conversely, (3.8) implies σ∗∗⪯cτ∗∗ and that σ∗ is also well defined. If σ⊲cτ, then 3.9∀0h>0 (small) there exists some Ch≥1 such that τ(t)≤σ(ht)+Ch for all t≥0. Consequently, the conclusions follow by the same computations as in (i). (a) Note that in general h>2 and hence the conclusion does not follow automatically. However, by iterating (ω1) for the particular weight one is able to conclude analogously as in the proof of Lemma 3.6. (b) We choose h:=12. □ On the generalized Legendre envelopes We study the interaction of the conjugate operation and the generalized Legendre envelope operators ⋆ˇ and ⋆^. Proposition 3.9 Let σ and τ be given weight functions such that σ(0)=0=τ(0), i.e. both weights satisfy (C1). Moreover, assume that σ⋆ˇτ satisfies (C2), then and Proof σ⋆ˇτ is a weight function such that σ⋆ˇτ(0)=σ(0)+τ(0)=0, see [17, Rem. 3.1, Lemma 3.2 (a)&(b)], and by assumption (σ⋆ˇτ)∗ is well defined. Note that (C2) for σ⋆ˇτ precisely means t=o(σ⋆ˇτ(t)) as t→+∞ and, in view of [17, Lemma 3.2 (a)], this property implies (C2) for σ and τ. Let t>0 be given, then: Note that for t=0 we have (σ⋆ˇτ)∗(0)=-σ⋆ˇτ(0)=0, whereas by [17, Lemma 4.1 (b)] and (C1) it holds that τ∗⋆^σ(0)=-τ(0)-σ(0)=0. Summarizing, the first part in (3.10) is verified. And since ⋆ˇ is commutative, see [17, Lemma 3.2 (c)], analogously one infers the second one, too. Finally, by this computation(C2) for σ⋆ˇτ implies that both τ∗⋆^σ and σ∗⋆^τ are well defined. Concerning (3.11) let t>0, then For t=0, again by (C1) and [17, Lemma 4.1 (b)] we infer (σ∗⋆^τ)(0)=-σ(0)-τ(0)=0, ((id⋆^τ)⋆^σ)(0)=(id(0)-τ(0))-σ(0)=0, and hence equality for all t≥0. □ The conjugate operation and growth indices We study now the effect of taking the conjugate operation defined in (3.1) on the growth indices γ(ω) and γ¯(ω) defined in Sect. 2.3. First, note that σ⪯σ (reflexivity) is clear and one can take in the proof of (i) in Lemma 3.6 any constant C≥1 and σ=τ. Thus, by applying (a) in this result to σ=τ and in view of Remark 2.2 we immediately get the following result: Lemma 3.10 Let σ be a weight function satisfying (C1) and (C2). Then σ∗ satisfies (ω6), i.e. γ¯(σ∗)<+∞. Moreover, the second part of (a) in Lemma 3.6 becomes trivial, however (b) cannot be applied to σ=τ since relation ⊲ is not reflexive. On the other hand we get that assumption (C2) also implies a certain restriction of the value γ(σ): Lemma 3.11 Let σ be a weight function. Then (C2) implies γ(σ)≤1; i.e. γ(σ)∈[0,1]. Proof By [6, Thm. 2.11 and Rem. 2.15] it follows that γ(σ)>1 implies ω(t)=O(tα) as t→+∞ for some α∈(0,1) which obviously contradicts (C2). □ The next main result relates the growth indices of σ and σ∗ in a precise manner and should be compared with [12, (A.1)and(A.2)]. Theorem 3.12 Let σ be a weight function satisfying (C1) and (C2). (i)It holds that max{0,1-γ¯(σ)}≤γ(σ∗). It holds that max{0,1-γ¯(σ)}≤γ(σ∗). (ii)It holds that γ¯(σ∗)≤1-γ(σ). It holds that γ¯(σ∗)≤1-γ(σ). (iii)Assume that in addition σ∗∗∼σ holds. Then the following is valid: It holds that γ(σ)=1-γ¯(σ∗).It holds that 1-γ¯(σ)=γ(σ∗).It holds that γ¯(σ)=1-γ(σ∗).It holds that γ(σ)=1-γ¯(σ∗). In this situation all indices belong to the closed unit interval [0, 1]. Assume that in addition σ∗∗∼σ holds. Then the following is valid: It holds that γ(σ)=1-γ¯(σ∗).It holds that 1-γ¯(σ)=γ(σ∗).It holds that γ¯(σ)=1-γ(σ∗).It holds that γ(σ)=1-γ¯(σ∗). In this situation all indices belong to the closed unit interval [0, 1]. It holds that γ(σ)=1-γ¯(σ∗). It holds that γ(σ)=1-γ¯(σ∗). It holds that 1-γ¯(σ)=γ(σ∗). It holds that 1-γ¯(σ)=γ(σ∗). It holds that γ¯(σ)=1-γ(σ∗). It holds that γ¯(σ)=1-γ(σ∗). It holds that γ(σ)=1-γ¯(σ∗). It holds that γ(σ)=1-γ¯(σ∗). Note: If γ¯(ω)<1, then the estimate in (i) in Theorem 3.12 implies, in particular, that γ(σ∗)>0 and so (ω1) for σ∗. If γ¯(σ)≥1, then this property for σ∗ is not clear. In view of Lemma 3.11 in (ii) one has γ(σ)∈[0,1] and hence the estimate is natural and consistent with the growth limitation obtained in Lemma 3.11: γ(σ)>1 would imply γ¯(σ∗)<0 but this is impossible for any weight function. Proof First, note that by assumption on σ the conjugate σ∗ is also a weight function satisfying (C1) and (C2) and σ∗∗∼σ implies γ(σ∗∗)=γ(σ), γ¯(σ∗∗)=γ¯(σ). (i) Since σ∗ is a weight function we get γ(σ∗)≥0 and hence the case γ¯(σ)≥1 is trivial. So let now (0≤)γ¯(σ)<1 and let γ>0 be such that γ<1-γ¯(σ)(≤1) and such a choice is possible by assumption on the index γ¯(σ). Since 1-γ>γ¯(σ) by definition of this index and we estimate as follows for all s≥0: Summarizing, this estimate gives σ∗(Aγ+ϵs)≤A1+ϵσ∗(s)+H for all s≥0. Now fix ϵ′∈(0,ϵ0) and let ϵ>0 be such that ϵ<γϵ′<ϵ′<ϵ0. Then set K:=A1+ϵ′(>A1+ϵ), γ′:=γ+ϵ1+ϵ′ and so γ′<γ⇔ϵ<γϵ′ holds. Therefore, the verified estimate implies (Pσ∗,γ′) with this choice for K and so by definition γ(σ∗)≥γ′. (Indeed, in view of [6, Thm. 2.11] we even have a strict inequality here.) Then let ϵ′→0, consequently ϵ→0 and hence γ(σ∗)≥γ holds. Since 0<γ<1-γ¯(σ) was chosen arbitrarily we have shown that 1-γ¯(σ)≤γ(σ∗). (ii) First let γ(σ)∈(0,1) and let 0<γ<1 be such that γ(σ)>1-γ(>0), i.e. γ>1-γ(σ), and note that such values γ exist by assumption on γ(σ). Then by definition we get Hence for all s≥0: This proves σ∗(Kγ-ϵs)≥K1-ϵσ∗(s)-H for all s≥0. Let ϵ′∈(0,ϵ0) be fixed, then choose ϵ>0 such that ϵ<γϵ′<ϵ′<ϵ0 and put A:=K1-ϵ′(1 and γ′>γ⇔ϵ′γ>ϵ and the above estimate verifies that (P¯σ∗,γ′) is valid for this choice of A. Then let ϵ′→0, hence ϵ→0 too and so γ′→γ which verifies γ¯(σ∗)≤γ. Finally, we let γ→1-γ(σ) and therefore γ¯(σ∗)≤1-γ(σ) is shown. If γ(σ)=1, then the same proof gives γ¯(σ∗)=0: Indeed, here in the arguments we can consider any γ>0 and we restrict again to γ∈(0,1) and then let γ→1-γ(σ)=0 which gives γ¯(σ∗)=1-γ(σ)=0. Finally, if γ(σ)=0, then follow the proof above with γ>1 and choose ϵ,ϵ′∈(0,ϵ0) such that ϵ<ϵ′<ϵ′γ<ϵ0. Again, as ϵ′,ϵ→0 we get γ¯(σ∗)≤γ and then let γ→1(=1-γ(σ)). (iii)(a) First, (ii) implies γ¯(σ∗)≤1-γ(σ)≤1. Then apply (i) to σ∗ which gives 1-γ¯(σ∗)≤γ(σ∗∗)=γ(σ). For these arguments note that σ∗ is a weight function satisfying (C1) and (C2) (recall Lemma 3.2). Summarizing, 1-γ¯(σ∗)=γ(σ) holds as desired. (iii)(b) We apply (ii) to σ∗ and which gives (0≤)γ¯(σ)=γ¯(σ∗∗)≤1-γ(σ∗)(≤1). Second, using this (i) implies 0≤1-γ¯(σ)≤γ(σ∗) and combining everything yields 1-γ¯(σ)=γ(σ∗). (iii)(c) Replace σ by σ∗ and σ∗ by σ∗∗ in the proof of (iii)(a). (iii)(d) Replace σ by σ∗ and σ∗ by σ∗∗ in the proof of (iii)(b). □ Remark 3.13 Consider the weight(s) from Example 3.3: First, note that γ(id1/α)=α=γ¯(id1/α) for any α>0 and thus Theorem 3.12 holds for resp. can be applied to all α∈(0,1). However, as verified in this Example, since for such indices α one infers (id1/α)∗∼id1/(1-α) it follows that the (limiting) case α=1 cannot be considered in Theorem 3.12. The weight sequence setting In this section we focus on the case ω=ωM, i.e. when the weight function is expressed in terms of a given weight sequence M. ωM denotes the corresponding associated weight function and due to this additional information we can consider ωM∗ and ωM∗, M∗ denoting the conjugate sequence introduced and used in [12]. Thus the goal is to establish a precise connection between both notions: the conjugate in the weight function and in the weight sequence framework. Frequently, we use the following estimates (with the convention 00:=1): The first estimate is clear and the second one follows, as recognized by one of the anonymous referees in the report, by the power series expansion of the exponential function since ep=∑k=0+∞pkk!≥ppp! for all p∈N. Indeed, formally Stirling’s formula is not required to prove (4.1) and note that these useful estimates appear regularly in this context in the literature and also in some of the author’s works. Weight sequences Given a sequence M=(Mp)p∈R>0N we also use the notation μ=(μp)p with μp:=MpMp-1, p≥1, μ0:=1. Moreover, we set mp:=Mpp! and obtain the sequence m=(mp)p∈N. Analogously these conventions are used for all other sequences under consideration. M is called normalized, if 1=M0≤M1 is valid. For the pointwise product sequence we simply write M·N and MN for the pointwise quotient sequence; i.e. these sequences are given by Mp·Np and MpNp respectively. M is called log-convex if equivalently if μ is non-decreasing. M is called strong log-convex if even m is log-convex which is equivalent to p↦μpp being non-decreasing. If M is log-convex and normalized, then μp≥1 for all p∈N, both mappings p↦Mp and p↦(Mp)1/p are non-decreasing, (Mp)1/p≤μp for all p∈N>0 and finally MpMq≤Mp+q for all p,q∈N; see e.g. [19, Lemmas 2.0.4 and 2.0.6]. M satisfies moderate growth, abbreviated in the following by (mg), if In view of [8] this condition is also known under (M.2) or stability under ultradifferential operators. In case M0=1, then it suffices to consider Cp+q in the estimate. We introduce the value and recall the fact that for any log-convex sequence one has Mι=limp→+∞(Mp)1/p=limp→+∞μp∈(0,+∞], see e.g. [25, Sect. 2.2, Lemma 2.1], and we comment on a minor technical issue in this result: Remark 4.1 By inspecting the proof of [25, Lemma 2.1] it turns out that one shall assume there limp→+∞μp=C∈(0,+∞) and note that the case C=0 is not excluded by the general assumption M∈R>0N. However, when limp→+∞μp=0 then the result is still valid by following the arguments and only considering the upper estimates with C=0: So for any ϵ>0 there exists pϵ∈N>0 such that for all i∈N>0 one obtains (0<)(Mpϵ+i)1pϵ+i≤ϵipϵ+i(Mpϵ)1pϵ+i. Then, as i→+∞ we infer ϵipϵ+i(Mpϵ)1pϵ+i→ϵ and since ϵ>0 was chosen arbitrary it follows that limp→+∞(Mp)1/p=0, too. But, finally, we recall that for any log-convex M∈R>0N this case can never occur since p↦μp is non-decreasing and each μp is non-negative. Let M,N∈R>0N be given, then write M≤N if Mp≤Np for all p∈N, M⪯N if supp∈N>0MpNp1/p<+∞, M⊲N if limp→+∞MpNp1/p=0; i.e. if ∀h>0∃Ch≥1∀p∈N:Mp≤ChhpNp. M and N are called equivalent, formally denoted by M≈N, if M⪯N and N⪯M. Obviously, M⊲N implies M⪯N, but ⊲ is never reflexive. Condition moderate growth is obviously preserved under equivalence of sequences. An important example, also for the considerations in this work, are the Gevrey sequences Gs:=(p!s)p∈N, s>0. Note that Gs is equivalent to G¯s with G¯s:=(pps)p∈N by (4.1) and each Gs is normalized, log-convex and satisfies (mg). For our purpose it is convenient to consider weight sequences in the following (general) sense: Definition 4.2 M=(Mp)p∈N is called a weight sequence if If M is a log-convex weight sequence, then in view of [25, Sect. 2.2, Lemma 2.1] and see also the above Remark 4.1, we get limp→+∞μp=+∞, too. Moreover, occasionally we consider the following (sub)set of weight sequences: We see that M∈LC if and only if 1=μ0≤μ1≤⋯ and limp→+∞μp=+∞ (see e.g. [14, p. 104]) and there is a one-to-one correspondence between M and μ=(μp)p by taking Mp:=∏i=0pμi. Obviously, Gs∈LC for any s>0. For any given sequence a=(ap)p∈R>0N the upper Matuszewska index α(a) is defined by and the lower Matuszewska index β(a) by β(a)>0 implies, in particular, limp→+∞ap=+∞. For these definitions we refer to [6, Sect. 3.2] and the citations there. Crucially we also mention [6, Thm. 3.10] where the connection γ(M)=β(μ) is established even for any arbitrary M∈R>0N. Here, γ(M) denotes the growth index introduced by V. Thilliez in [27]. Note that in [6] concerning the sequence of quotients a different notation has been used: There, m denotes μ, more precisely one has the correspondence mp=μp+1 for all p. But this index shift does not effect our considerations; we refer to [6, Rem. 3.8] and the comments given after [6, Cor. 3.12]. Associated weight functions For the following definition we refer to [9, Chapitre I], [8, Sect. 3] and the more recent work [25] in which more properties and explanations are summarized and non-standard cases are treated. Let M=(Mp)p∈N∈R>0N be given, then the associated function ωM:[0,+∞)→[0,+∞)∪{+∞} is defined as follows: with the conventions 00:=1 and log(0)=-∞. This ensures ωM(0)=0 and ωM(t)≥0 for any t≥0 since t0M0M0=1 for all t≥0. (4.2) corresponds to [8, (3.1)] and we immediately have that ωM is non-decreasing and satisfies limt→+∞ωM(t)=+∞. However, the fact that ωM:[0,+∞)→[0,+∞), i.e. that ωM is well defined, is equivalent to assume limp→+∞(Mp)1/p=+∞; see [25, Lemma 2.2] and compare this with Definition 4.2. For any s>0 we have id1/s∼ωGs; for a proof we refer e.g. to [17, Ex. 2.9]. Let M be a log-convex weight sequence, then one can compute M by involving ωM as follows, see [9, Chapitre I, 1.4, 1.8], [8, Prop. 3.2] and more recently [25]: If M is a weight sequence which is not log-convex, then the right-hand side of (4.3) yields Mplc with Mlc denoting the log-convex minorant of M and then Mlc is a log-convex weight sequence. Moreover, for any log-convex weight sequence let us introduce the counting function and recall the following integral representation formula see [9, 1.8. III], [8, (3.11)] and [25, Lemma 2.5]. Consequently, and so for any M∈LC the function ωM vanishes on [0, 1]. Next let us recall [25, Lemma 2.4]. Lemma 4.3 Let M=(Mp)p∈N be a log-convex weight sequence and hence Mι=limp→+∞(Mp)1/p=limp→+∞μp=+∞. Then For concrete applications, when one is (only) interested in the associated function w.l.o.g. one can assume that the defining sequence is log-convex. This fact holds since the geometric regularizing procedure introducing ωM gives, in particular, that ωM=ωMlc; see again [25]. Remark 4.4 A log-convex weight sequence with M0=1 is a weight sequence in the sense of [6]; see [6, Sect. 3.1]. However, if M is a (log-convex) weight sequence with M0≠1, then introduce M~ with M~p:=MpM0. So M~ is a (log-convex) weight sequence with M~0=1 and M~≈M holds but even μ~p=μp for all p is valid. Moreover, ωM~=ωM since Thus when involving results from [6] the assumption M0=1 is not restricting the generality since we apply the statements to M~ and by the above equality of associated functions and quotients they transfer to M immediately. Finally, when M is log-convex then by applying [19, Lemmas 2.0.4 and 2.0.6] to M~ we get that p↦(Mp/M0)1/p is non-decreasing, MpMq≤M0Mp+q for all p,q∈N and (Mp/M0)1/p≤μp for all p∈N>0. The conjugate sequence M∗ Let us recall some definitions and properties listed and stated in [12, Sect. 2.5 and 2.6]; see also there for more details and explanations. Let M∈R>0N be given, then the conjugate sequence M∗=(Mp∗)p∈N is defined by i.e. M∗=1m. The corresponding sequence of quotients μ∗=(μp∗)p∈N is given by μp∗=pμp, p∈N>0, and set μ0∗:=1. We gather several immediate properties: Set M∗∗:=(M∗)∗, then M∗∗=M and M∗·M≡G1. Set M∗∗:=(M∗)∗, then M∗∗=M and M∗·M≡G1. The following are equivalent: (*)M∗ is log-convex,(*)m is log-concave, see [12, (2.5)], i.e. ∀p∈N>0:mp2≥mp+1mp-1.(*)p↦μpp is non-increasing. If a sequence N is log-concave, then following the arguments in [19, Lemma 2.0.4] we get that p↦(Np/N0)1/p is non-increasing. Consequently, if M∗ is log-convex and M0∗=m0=1, then p↦(mp)1/p is non-increasing. The following are equivalent: (*)M∗ is log-convex,(*)m is log-concave, see [12, (2.5)], i.e. ∀p∈N>0:mp2≥mp+1mp-1.(*)p↦μpp is non-increasing. If a sequence N is log-concave, then following the arguments in [19, Lemma 2.0.4] we get that p↦(Np/N0)1/p is non-increasing. Consequently, if M∗ is log-convex and M0∗=m0=1, then p↦(mp)1/p is non-increasing. (*)M∗ is log-convex, M∗ is log-convex, (*)m is log-concave, see [12, (2.5)], i.e. ∀p∈N>0:mp2≥mp+1mp-1. m is log-concave, see [12, (2.5)], i.e. ∀p∈N>0:mp2≥mp+1mp-1. (*)p↦μpp is non-increasing. p↦μpp is non-increasing. If both M and M∗ are log-convex, then M∗·M≡G1 implies MιMι∗=+∞. If both M and M∗ are log-convex, then M∗·M≡G1 implies MιMι∗=+∞. It is immediate that M⊲G1⇔limp→+∞(mp)1/p=0⇔limp→+∞(Mp∗)1/p=Mι∗=+∞, and this is equivalent to the fact that ωM∗ is well defined; recall [25, Lemma 2.2] applied to M∗. Thus, whenever involving ωM∗ it is natural to assume this growth property on M. If M∗ is also log-convex, then M⊲G1 if and only if limp→+∞μp∗=+∞; see [25, Sect. 2.2]. It is immediate that M⊲G1⇔limp→+∞(mp)1/p=0⇔limp→+∞(Mp∗)1/p=Mι∗=+∞, and this is equivalent to the fact that ωM∗ is well defined; recall [25, Lemma 2.2] applied to M∗. Thus, whenever involving ωM∗ it is natural to assume this growth property on M. If M∗ is also log-convex, then M⊲G1 if and only if limp→+∞μp∗=+∞; see [25, Sect. 2.2]. M≤N if and only if N∗≤M∗ and M⪯N if and only if N∗⪯M∗. Thus M, N are equivalent if and only if the conjugate sequences M∗, N∗ are so. Moreover, M⊲N if and only if m⊲n if and only if N∗⊲M∗. M≤N if and only if N∗≤M∗ and M⪯N if and only if N∗⪯M∗. Thus M, N are equivalent if and only if the conjugate sequences M∗, N∗ are so. Moreover, M⊲N if and only if m⊲n if and only if N∗⊲M∗. M∗ satisfies (mg) if and only if mpmq≤Cp+q+1mp+q if and only if p+qpMpMq≤Cp+q+1Mp+q for some C≥1 and all p,q∈N. If M is log-convex, then Remark 4.4 gives p+qpMpMq≤2p+qM0Mp+q≤max{2,M0}p+q+1Mp+q for all p, q, so the desired estimate holds with C:=max{2,M0} and M∗ satisfies (mg). M∗ satisfies (mg) if and only if mpmq≤Cp+q+1mp+q if and only if p+qpMpMq≤Cp+q+1Mp+q for some C≥1 and all p,q∈N. If M is log-convex, then Remark 4.4 gives p+qpMpMq≤2p+qM0Mp+q≤max{2,M0}p+q+1Mp+q for all p, q, so the desired estimate holds with C:=max{2,M0} and M∗ satisfies (mg). Next, we verify the converse of (f) and first, motivated by comment (b) above, we prove the following technical result. Lemma 4.5 Let M∈R>0N be given and recall the notation mp:=Mpp!, p∈N. Then the following are equivalent: (i)M satisfies 4.7∃H≥1∀p∈N>0:(mp+1)1/(p+1)≤H1/(p+1)(mp)1/p. M satisfies 4.7∃H≥1∀p∈N>0:(mp+1)1/(p+1)≤H1/(p+1)(mp)1/p. (ii)M satisfies 4.8∃A≥1∀p∈N>0:μp+1≤A(Mp)1/p. M satisfies 4.8∃A≥1∀p∈N>0:μp+1≤A(Mp)1/p. The proof shows the following correspondences: If (4.7) holds with some H, then in (4.8) we choose A:=2eH and, conversely, (4.8) implies (4.7) with H:=A. Proof (i)⇒(ii): Let p∈N>0, then we have Moreover, (p+1)(mp)1/p≤2p(mp)1/p≤2e(p!)1/p(mp)1/p=2e(Mp)1/p is valid by (4.1). Combining both estimates we have verified (4.8) with A:=2eH. (ii)⇒(i): Let p∈N>0 be arbitrary, and then Since (p!)1/pp+11/(p+1)≤1⇔(p!)1/p≤p+1 is clear for any p∈N>0 the estimate (4.7) is shown with H:=A as desired. □ Concerning the appearing conditions in Lemma 4.5 we point out: Remark 4.6 The choice H=1 in (4.7) precisely means that p↦(mp)1/p is non-increasing, whereas it is known that (4.8) is equivalent to (mg) if M∈R>0N is log-convex; we refer to [15, Lemma 2.2 (1)⇔(2), Rem. 2.3, (2.10)], the proof of [11, Thm. 1] and also to the detailed discussion in [23]. Note that for both implications in this equivalence log-convexity is required in the arguments but, however, limp→+∞(Mp)1/p=+∞ is not necessary for this part. On the other side, this characterization breaks down in general in the mixed setting as it is shown in [23, 26]. Corollary 4.7 If both M∗ and M are log-convex, then M has to satisfy (mg). Proof Via comment (b) above, if M∗ is log-convex, then p↦(mp/m0)1/p is non-increasing and so Since m0-1/p≤(1-ϵ)-1⇔(1-ϵ)p≤m0 holds for any p∈N>0 when ϵ∈(0,1) is chosen sufficiently close to 1, we get that H1/(p+1)≥m0-1p(p+1)⇔H≥m0-1/p holds for H chosen sufficiently large and so (4.7) is valid. Then Lemma 4.5 and Remark 4.6 yield the conclusion. □ Finally, we briefly recall the technical but crucial and useful comments from [12, Lemma 2.3 (a), Rem. 2.4, Rem. A.5 (iv)]. Remark 4.8 Let M∈R>0N be given and assume that μppp is almost decreasing; i.e. Note that by definition α(μ)<1 implies (4.9) whereas this property implies α(μ)≤1. Then set and introduce the sequence L=(Lp)p∈N via Lp:=∏i=0pλi. Then we get (see [12, Rem. 2.4]): L0=1 holds. l=(lp)p∈N given by lp:=Lpp! is log-concave; i.e. p↦λpp is non-increasing. M≈L is valid. If limp→+∞(mp)1/p=0 resp. limp→+∞(Mp)1/p=+∞ holds, then by this equivalence l resp. L shares the particular growth property, too. If M is log-convex, then L, too. Moreover, in view of the general construction in [12, Rem. A.5 (iv)], if L is a log-convex weight sequence, then by applying a technical modification for L for finitely many indices (at the beginning), one can even achieve that there exists L~∈LC such that L~0=1=L~1, l~ is log-concave, M≈L~ and, finally, L~,L~∗∈LC. Indeed, by the above comments this situation applies when M is log-convex, μppp is almost decreasing, limp→+∞(mp)1/p=0 and limp→+∞(Mp)1/p=+∞. In this case in order to lighten notation we simply write again L, L∗ for these sequences. The importance of the previous remark is given by the fact that for given M satisfying some growth requirements one can find an equivalent and very regular sequence L such that L∗ is also very regular (and equivalent to M∗). Conjugate associated weight functions We are interested in the situation when ωM∗ is well defined and characterize this in terms of M. Lemma 4.9 Let M be a log-convex weight sequence. Then the following are equivalent: ωM∗ is well defined. ωM satisfies (C2); i.e. t=o(ωM(t)) as t→+∞ equivalently ωM⊲id. ωG1(t)=o(ωM(t)) as t→+∞; i.e. ωM⊲ωG1. M⊲G1 holds; i.e. 4.11limp→+∞Mpp!1/p=limp→+∞(mp)1/p=0. The crucial growth condition (4.11) is non-standard in the theory of ultradifferentiable function classes and related weighted settings. Proof First, (i)⇔(ii) holds by Lemma 3.1 applied to ωM∗ and (ii)⇔(iii) is valid because id∼ωG1. This equivalence is well-known; for an explicit proof we refer e.g. to the case s=1 in [17, Ex. 2.9]. And this equivalence implies, in particular, that ωG1 satisfies both (ω1) and (ω6) since these conditions are obviously true for id. (iii)⇒(iv): The assumption precisely means that for all ϵ>0 there exists some Cϵ≥1 such that ωG1(t)≤ϵωM(t)+Cϵ for all t≥0. Then let h>0 small be given and choose n∈N>0 such that h-1≤2n. Since ωG1 has (ω1) by iteration we obtain ωG1(h-1t)≤ωG1(2nt)≤LωG1(t)+L for some L≥1 depending on h and for all t≥0. We apply the previous relation to ϵ:=L-1 and obtain: Set L1:=LCL-1+L, then by using this estimate and involving (4.3) we obtain for all p∈N: For the last equality note that G01=0!=1. Summarizing, which gives M⊲G1. (iv)⇒(iii): By assumption hence by definition ωG1(t)≤ωM(ht)+log(Ch/M0) for all t≥0. Let now A≥1 large be given and fixed. Choose n∈N>0 such that A≤2n and by iterating (ω6) we obtain AωG1(t)≤2nωG1(t)≤ωG1(Ht)+H for some H≥1 depending on chosen A and for all t≥0. When applying the above relation to h:=H-1 we arrive at hence lim supt→+∞ωG1(t)ωM(t)≤1A. Since A≥1 can be chosen arbitrary large this shows ωG1(t)=o(ωM(t)) and we are done. □ Note: Condition (4.11) admits the possibility to check the desired growth property for given explicit examples directly and remark that log-convexity is only required for (iii)⇒(iv). By comment (d) in Sect. 4.3 and in view of Lemmas 4.9 and 3.1 we obtain the following consequence: Corollary 4.10 Let M be a log-convex weight sequence, then the following are equivalent: limp→+∞(mp)1/p=0 is valid. ωM∗ is well defined. ωM∗ is well defined. Note that in Corollary 4.10 it is not required necessarily to have log-convexity for M∗, i.e. log-concavity for m. Next we investigate the growth relations: For given weight sequences M, N the relation M⪯N precisely means which gives by definition i.e. ωM⪯cωN. Thus M≈N yields ωM∼cωN. Similarly, M⊲N implies ωM⊲cωN. Using this we can show the following: Lemma 4.11 Let M, N be log-convex weight sequences. Assume that M⪯N and that limp→+∞(np)1/p=0 holds. Then ωN∗ and ωM∗ are well defined and 4.13∃D>0∃h>0∀t≥0:ωM∗(t)≤ωN∗(ht)+D; i.e. ωN∗⪯cωM∗. Moreover, ωM∗ and ωM∗ are well defined and 4.14∃D>0∃h>0∀s≥0:ωM∗(s)≤ωN∗(sh)+D; i.e. ωN∗⪯cωM∗. Assume that M⊲N and that limp→+∞(np)1/p=0 holds. Then even ωN∗⊲cωM∗ and ωN∗⊲cωM∗ and all (associate) weight functions under consideration are well defined. Therefore, when M≈N and either limp→+∞(mp)1/p=0 or limp→+∞(np)1/p=0, then ωM∼cωN, ωM∗∼cωN∗, ωN∗∼cωM∗ and all (associate) weight functions under consideration are well defined. Concerning this additional information note that equivalence is preserved under taking the conjugate operation; recall comment (e) in Sect. 4.3. Proof (a) First, by comment (e) in Sect. 4.3 we have N∗⪯M∗. Second, by limp→+∞(np)1/p=0 and M⪯N also limp→+∞(mp)1/p=0. Thus both ωN∗ and ωM∗ are well defined, see Corollary 4.10, and similarly like (4.12) estimate (4.13) follows. And Corollary 4.10 also implies that both ωM∗ and ωM∗ are well defined. Since ωM⪯cωN holds we have that (i) in Lemma 3.8 implies ωN∗⪯cωM∗. (b) The conclusions follow analogously by taking into account (ii) in Lemma 3.8, comment (e) in Sect. 4.3 and Corollary 4.10. □ The conjugate sequence and the generalized upper and lower Legendre envelopes We study the interaction of the conjugate operation and the Legendre envelopes introduced in Sect. 2.4 in the weight sequence setting when involving the corresponding conjugate sequence. First we verify an auxiliary result which should be compared with [17, Lemma 3.4, Thm. 4.14 and 4.15]. Lemma 4.12 Let M, P, N, Q be weight sequences. If M⪯P and N⪯Q, then ωM⋆ˇωN⪯cωP⋆ˇωQ holds whereas M⊲P and N⊲Q imply ωM⋆ˇωN⊲cωP⋆ˇωQ. If M⪯P and Q⪯N, then ωM⋆^ωN⪯cωP⋆^ωQ holds whereas M⊲P and Q⊲N imply ωM⋆^ωN⊲cωP⋆^ωQ. Concerning (ii) note that in order to give this part a meaning one shall naturally assume that ωM⋆^ωN is well defined, see [17, Prop. 5.2, Def. 5.5]. Recall that by [17, Prop. 5.2, (I)(ii)⇒(i)] we get that N⊲M ensures this fact and one obtains a characterization provided N is log-convex. The relations between the sequences stated in (ii) give that N⊲M implies Q⊲P and thus this result is consistent with [17, Prop. 5.2]. Proof (i) By assumption we can find C1,C2,h1,h2>0 such that Mj≤C1h1jPj and Nj≤C2h2jQj for all j∈N. Set C:=max{C1,C2} and h:=max{h1,h2} and the definition of associated weight functions gives ωP(t)≤ωM(ht)+log(CP0/M0) and ωQ(t)≤ωN(ht)+log(CQ0/N0) for all t≥0. Thus (2.9) implies for all t≥0: This finishes the first conclusion; the second one dealing with the stronger growth relations is completely analogous. (ii) Similarly as before, one infers that ωP(t)≤ωM(ht)+log(CP0/M0) and ωN(t)≤ωQ(ht)+log(CN0/Q0) for all t≥0. Then (2.10) yields for any t>0: Again, the second part is analogous and note that for t=0 we even get equality with the value 0; see Sect. 2.4. □ We formulate and prove now the main statement in this section and verify the meaning of the conjugate sequence within the weight sequence setting when involving the generalized Legendre envelopes ⋆ˇ and ⋆^. Theorem 4.13 Let M be a weight sequence and M∗ the corresponding conjugate sequence. Then the following are valid: Assume that limp→+∞(mp)1/p=0 holds and that there exist sequences P, Q such that P≈M≈Q, with P being log-convex whereas q being log-concave. Then ωM⋆ˇωM∗∼id. Let M be equivalent to a strong log-convex sequence S and such that even limp→+∞(mp)1/p=+∞, then ωM⋆^ω(M∗)-1∼id. Let M be log-convex, then ωM2M∗⋆^ωM∼id. Assume that limp→+∞(mp)1/p=0 and limp→+∞(Mpmp)1/p=+∞ holds and that there exist sequences P, Q such that P≈M≈Q, with P being log-convex and q being log-concave. Assume also that P·q is log-convex, then ωM⋆^ωM∗∼cωM·m=ωG1·m2. Let M be log-convex, limp→+∞(mp)1/p=0, m is log-concave, G1M-2 is log-convex and satisfying limp→+∞(p!Mp-2)1/p=+∞. Then ωM⋆⋆^ωM=ωM∗M=ωG1M-2. Proof (i) First, note that both P and Q∗ are log-convex weight sequences by assumption, see (b) and (d) in Sect. 4.3, and so [17, Thm. 5.1] can be applied to M≡P and N≡Q∗. Moreover, by relation M·M∗≡G1 we get P·Q∗=:G1~≈G1 and so Pι·Qι∗=+∞; see (c) in Sect. 4.3. Comment (d) there and Sect. 4.2 yield that ωP and ωQ⋆ are weight functions. Thus, when gathering this information we obtain: Next recall that ωG1∼id is valid by [17, Ex. 2.9] (for s=1) and that the equivalence between G1~ and G1 implies ωG1~∼cωG1; see (4.12). Moreover, ωM⋆ˇωM⋆∼cωP⋆ˇωQ⋆ by taking into account (e) in Sect. 4.3 and (i) in Lemma 4.12 and which implies ωM⋆ˇωM⋆∼cωG1. Finally, since ωG1 satisfies (ω1) by [18, Prop 2.2 (ii)] we obtain ωM⋆ˇωM⋆∼ωG1 and hence the conclusion. (ii) Apply [17, Thm. 5.7] to M≡S and N≡(S∗)-1 and obtain Note that in this case the quotient sequence SN is equal to G1 and thus clearly log-convex. Moreover, in view of (4.6) we have Np=Spp!=sp for all p∈N and consequently the log-convexity for N is precisely assumption strong log-convexity for S. Therefore, all appearing sequences are log-convex by assumption and definition. Mι=+∞=Nι since even limp→+∞(mp)1/p=+∞ holds. Finally, relation N⊲S amounts to have which is equivalent to requiring limp→+∞(p!)1/p=+∞ and this is clear by (4.1). Thus the result follows by taking into account comment (e) in Sect. 4.3, (ii) in Lemma 4.12 and involving again [18, Prop 2.2]: Since ωG1 satisfies both (ω1) and (ω6) both equivalence relations ∼ and ∼c hold simultaneously. (iii) Apply [17, Thm. 5.7] to M1≡M2M∗=G1M and N1≡M and thus M1N1=G1. All sequences G1M, M and G1 are clearly log-convex and consequently, analogously as in (ii), we get Note that here (M1)ι=+∞=(N1)ι holds since M is a weight sequence and relation N1⊲M1 amounts to have and which is again equivalent to limp→+∞(p!)1/p=+∞. (iv) Apply [17, Thm. 5.7] to P and N≡Q∗ and note that, like in (i) before, both are log-convex weight sequences by assumption. Then, by recalling (e) in Sect. 4.3, (4.6), (4.12) and (ii) in Lemma 4.12 we obtain For this note that P·q≡PQ∗ is assumed to be log-convex and Q∗⊲P if and only if M∗⊲M which amounts to have i.e. requiring limp→+∞(Mpmp)1/p=+∞. (v) Apply [17, Thm. 5.7] to M≡M∗ and N≡M. Again both sequences are log-convex weight sequences by assumption and in view of (4.6) one has The log-convexity M∗M is ensured since G1M-2 is log-convex and limp→+∞(p!Mp-2)1/p=+∞ precisely gives M⊲M∗. □ Concerning (iv) and (v) in Theorem 4.13 we comment on some explicit example(s); more precisely we consider small Gevrey-sequences. Remark 4.14 Take M≡Gα with α∈(0,1). Then m≡Gα-1, so M is a log-convex weight sequence and m is log-concave such that limp→+∞(mp)1/p=0. Moreover, G1·m2≡G2α-1 is log-convex and limp→+∞(p!mp2)1/p=+∞ provided that α∈(12,1). On the other hand, G1M-2≡G1-2α is log-convex and limp→+∞(p!Mp-2)1/p=+∞ provided that α∈(0,12). However, in this setting the index α=12 has to be excluded: M≡G1/2 yields M≡M∗ but ωN⋆^ωN is not well defined with N being any log-convex weight sequence (thus Nι=+∞); for this recall [17, Cor. 5.3]. Conjugate sequences and conjugate associated weight functions Let M be a log-convex weight sequence satisfying limp→+∞(mp)1/p=0. In view of Corollary 4.10 we have that both ωM∗ and ωM∗ are well defined and hence it makes sense to consider two conjugate operations: either the conjugate sequence M∗ or the conjugate associated weight function ωM∗. The aim is to establish a precise connection between both notions, to relate and to compare ωM∗ and ωM∗ and we prove the following main statement. Theorem 4.15 Let M be a given weight sequence such that limp→+∞(mp)1/p=0. Then the following holds: Assume that there exist sequences L, N such that L≈M≈N, L is log-convex whereas n is log-concave. Then ∃C≥1∀s≥0:ωM∗(s)≤ωM∗(Cs)+1. Assume that M is log-convex and that μppp is almost decreasing. Then there exists a log-convex weight sequence L such that L≈M, l is log-concave and ∀s∈[0,λ1∗]:ωL∗(s)≥ωL∗(s)=0,∀q∈N>0:ωL∗(λq∗)≥ωL∗(λq∗), and ∀λq∗0:ωL∗(2s)≥ωL∗(s). Summarizing, we infer which gives for M, M∗ the estimates Therefore, ωL∗∼cωL∗ and ωM∗∼cωM∗ is valid. Proof First, in view of (d) in Sect. 4.3 the function ωM∗ is well defined. Second, for s=0 one obtains ωM∗(0)=supt≥0{-ωM(t)}=0 which immediately gives ωM∗(0)=0=ωM∗(0) and hence equality. So from now on we focus on s>0. (i) By (i) in Theorem 4.13 we have ωM⋆ˇωM∗∼id which does imply resp. with u:=ts that Note that this estimate is clear for s=0. Now put v:=uC and so we have shown which gives by definition ωM∗(v)=sups≥0{sv-ωM(s)}≤ωM∗(Cv)+1 for all v≥0 and therefore the conclusion. (ii) By Remark 4.8 we have that there exists an equivalent and log-convex weight sequence L such that l is log-concave and limp→+∞(lp)1/p=0 (by equivalence) and so L∗ is a log-convex weight sequence, too. Moreover, in view of comment (e) in Sect. 4.3 also M∗ and L∗ are equivalent. We start with the proof of the desired estimates: First, on the one hand ωL∗(s)=0 for all s∈[0,λ1∗] follows by Lemma 4.3 applied to L∗ and, on the other hand, since ωL(λ1)=0 by definition ωL∗(s)≥sλ1≥0 for all s≥0. Second, let q∈N>0 and by taking into account definition (4.6), Lemma 4.3 applied to both L∗ and L, and (4.1) (used in the second inequality) we estimate as follows for all q∈N>0: Finally, let s>0 be such that qλq0. For each s>1λ1 such q exists since limp→+∞qλq=+∞ by assumption L⊲G1 and comment (d) in Sect. 4.3. Hence 2s>2qλq≥q+1λq≥q+1λq+1 and the last estimate is valid because L is log-convex. Then, by taking into account the fact that ωL∗ and ωL∗ are non-decreasing and the estimate shown before we get: Combining this part with (i) applied to M≡L≡N we have shown (4.15) and from this (4.16) follows by taking into account (4.13) and (4.14) and recall for this the equivalences between M and L, M∗ and L∗. □ We proceed with two immediate consequences of this theorem; concerning the notation in the next corollary we refer to [12, Sect. 3] and [24, Sect. 2.5, (2.14)]. Corollary 4.16 Let M be a log-convex weight sequence such that μppp is almost decreasing and limp→+∞(mp)1/p=0. Then in the main result [12, Thm. 3.4], instead of the dilatation-type system vM∗,c(t):=exp(-ωM∗(ct)), c>0, one can equivalently involve the (dilatation-type) weight function system Proof This follows by (4.16) which transfers to the above defined weight functions in the weighted entire setting as follows: □ Note: In Corollary 4.16 the assumptions on M are stronger compared with [12, Thm. 3.4 and Rem. 3.5] since here we require more regularity and growth assumptions on M. Indeed, apart from the inverse normalization condition 1=M0≥M1 in [12], which is a minor technical issue, here in addition crucially it is required that M is a log-convex weight sequence. However, since we are involving ωM directly in the definition of the new weight function system, the above Corollary 4.16 gives more information and recall: The assumption that M is a log-convex weight sequence ensures that ωM is well defined, see Definition 4.2 and Sect. 4.2, and hence this additional assumption on M is natural when involving ωM. Corollary 4.17 Let M be a log-convex weight sequence such that limp→+∞(mp)1/p=0 and α(μ)<1. Then: The assertions listed in Remark 4.8 are valid; Both ωM∗ and ωM∗ are well defined and satisfy in addition (ω1); ωM∗∼ωM∗ is valid; M satisfies (mg). Proof First, by definition α(μ)<1 obviously implies that μppp is almost decreasing; thus Remark 4.8 holds and (ii) in Theorem 4.15 can be used. Second, by applying [6, Thm. 2.16] to ωM and [6, Cor. 4.6 (ii)] to M we get where β(·) for a given weight function is defined analogously as for a weight sequence; we refer to [6, Sect. 2.2, Thm. 2.4] and the citations there. Then (i) in Theorem 3.12 applied to σ=ωM gives 0<1-γ¯(ωM)≤γ(ωM∗) and thus ωM∗ has (ω1); recall Remark 2.2. Theorem 4.15 implies ωM∗∼cωM∗ and then ωM∗∼ωM∗ follows by taking into account [18, Proposition 2.2 (ii)]. Since (ω1) is preserved under equivalence also ωM∗ satisfies this property. Finally, via (4.17) we obtain γ¯(ωM)<1, which implies by taking into account Remark 2.2 that, in particular, ωM has to satisfy (ω6). And this condition is then equivalent to (mg) for M; see [15, Lemma 2.2], [11, Thm. 1] and [23]. □ Remark 4.18 Similarly, let M∈R>0N be a log-convex weight sequence and assume that μppp is almost decreasing. Then the assertions listed in Remark 4.8 are valid, α(μ)≤1 and via (4.17) we obtain then γ¯(ωM)≤1. And this implies again that, in particular, ωM has to satisfy (ω6) and (mg) holds for M. The Braun–Meise–Taylor weight function setting We focus now on a certain and crucial subclass of weight functions, more precisely we consider weight functions in the sense of Braun–Meise–Taylor (BMT-weights for short); see [3]. Weight functions in the sense of Braun–Meise–Taylor Let ω:[0,+∞)→[0,+∞) be continuous, non-decreasing, ω(0)=0 and limt→+∞ω(t)=+∞. If ω satisfies in addition ω(t)=0 for all t∈[0,1], then ω is called normalized and this can always be assumed w.l.o.g. For convenience we write that ω has (ω0) if it satisfies all these assumptions; see e.g. [6, Sect. 2.1] and [7, Sect. 2.2]. Apart from (ω1) and (ω6) already treated in Sect. 2.2 we consider the following conditions; again the abbreviations have already been used in [20]. log(t)=o(ω(t)) as t→+∞. φω:t↦ω(et) is a convex function on R. In the literature one can find different assumptions for BMT-weights. However, some of the conditions are basic and for convenience we introduce the set In the forthcoming W0 is understood to be the set of (normalized) weight functions in the sense of Braun–Meise–Taylor. Then, for any ω∈W0 we define the Legendre-Fenchel-Young-conjugate of φω by with the following properties, see e.g. [3, Rem. 1.3, Lemma 1.5] and [18, Sect. 2.9]: It is convex and non-decreasing, φω∗(0)=0, φω∗∗=φω, limx→+∞xφω∗(x)=0 and finally x↦φω(x)x and x↦φω∗(x)x are non-decreasing on [0,+∞). Note that by normalization we can extend the supremum in (5.1) from y≥0 to y∈R without changing the value of φω∗(x) for a given x≥0. We mention the following known result, see e.g. [21, Lemma 2.8] resp. [7, Lemma 2.4] and the references mentioned in the proofs there. Lemma 5.1 Let M∈LC. Then ωM∈W0. Furthermore, (ω6) holds for ωM if and only if M satisfies (mg). Associated weight matrix Before recalling some results concerning the associated weight matrix Mω we start with the following: Definition 5.2 A weight matrix is the set of sequences such that M(ℓ1)≤M(ℓ2) for all 0<ℓ1≤ℓ2. We call M standard log-convex if M(ℓ)∈LC for each ℓ>0 and constant if M(ℓ1)≈M(ℓ2) for all ℓ1,ℓ2>0. Now let us summarize some facts which are shown in [14, Sect. 5] and are needed in this work; all properties listed below are valid for ω∈W0 except (5.4) for which (ω1) is crucial. More basic properties and conditions for abstractly given weight matrices M can be found e.g. in [14, Sect. 4]. The idea was that to each ω∈W0 one can associate a weight matrix Mω:={W(ℓ)=(Wp(ℓ))p∈N:ℓ>0} by Wp(ℓ):=exp1ℓφω∗(ℓp). We have that W(ℓ)∈LC for each ℓ>0, i.e. Mω is standard log-convex, and we even have the stronger order relation 5.2∀ℓ2≥ℓ1>0:ϑ(ℓ1)≤ϑ(ℓ2), with ϑ(ℓ) denoting the corresponding sequence of quotients; see [15, Sect. 2.5]. “Usually”, for abstractly given weight matrices M:={M(ℓ):ℓ>0} it is sufficient to assume the pointwise order M(ℓ1)≤M(ℓ2) for all 0<ℓ1≤ℓ2; see [14, Sect. 4]. Mω satisfies 5.3∀ℓ>0∀p,q∈N:Wp+q(ℓ)≤Wp(2ℓ)Wq(2ℓ). (ω6) for ω holds if and only if some/each W(ℓ) satisfies (mg) if and only if W(ℓ1)≈W(ℓ2) for each ℓ1,ℓ2>0. Thus (ω6) is characterizing the case when Mω is constant. In case ω has in addition (ω1), then Mω also satisfies 5.4∀h≥1∃d≥1∀ℓ>0∃D≥1∀p∈N:hpWp(ℓ)≤DWp(dℓ). This estimate is crucial for verifying the equality of ultradifferentiable classes E[Mω]=E[ω] (as locally convex vector spaces). Concerning the definition of these spaces we refer to [3, 14] and the notation [·] denotes the convention meaning either the Roumieu case {·} or the Beurling case (·). We have ω∼ωW(ℓ) for each ℓ>0, more precisely 5.5∀ℓ>0∃Dℓ>0∀t≥0:ℓωW(ℓ)(t)≤ω(t)≤2ℓωW(ℓ)(t)+Dℓ; for a proof see [20, Theorem 4.0.3, Lemma 5.1.3], [14, Lemma 5.7] and also [7, Lemma 2.5]. Note that, on the one hand, for proving (5.5) the convexity condition (ω4) is indispensable but, on the other hand, (ω4) is only required for the second estimate in (5.5). We finish this section by recalling some comments summarized in detail in [18, Sect. 2.9]. Remark 5.3 By inspecting the proofs in [14, 20] it turns out that for any non-decreasing ω satisfying (ω3) it makes sense to associate the well-defined matrix Mω; we refer to [18, Sect. 2.9] and also to the comments in [20, Sect. 3.1], [16, Sect. 2.7, (a)-(c)] and [26, Sect. 6 (III)]. And via (ω1) and (5.4) also the equality between the corresponding weighted spaces holds when the classes are introduced directly (by definition) via weighting the derivatives in terms of φω∗. Indeed, note that the proof of (5.4) does not require necessarily (ω4) but this condition is crucial for showing (5.5) (for obtaining the second estimate). Moreover, (ω4) is needed for the equivalences listed in (iii). Inspired by these comments any non-decreasing function ω satisfying (ω3) is called matrix admissible and the equivalence between matrix admissible functions transfers in a precise way to the corresponding associated weight matrices (yielding equivalent weight matrices) and which ensures the equality of the corresponding (matrix) weighted spaces; see the relations introduced in [14, Sect. 4.2, p. 111] and the estimate [26, Sect. 6 (III), (6.6)]. Conjugate associated weight matrices We investigate the effects of taking the conjugate operations from the previous sections in the BMT-weight function setting: Let ω be a given BMT-weight function, then consider ω∗, ω(ℓ)∗ according to (3.1) and the conjugate weight sequences (W(ℓ))∗ w.r.t. (4.6). Our aim now is to study the correspondences of these notions in detail. First we start with some new definitions and observations which are based on these comments. In this section let us consider ω∈W0 with associated weight matrix Mω={W(ℓ):ℓ>0}. Since any such weight satisfies by definition ω(0)=0 we always have (C1). In order to deal with ω∗ we require that this function is well defined and hence we verify now when (C2) is valid. Lemma 5.4 Let ω∈W0 be given with associated weight matrix Mω={W(ℓ):ℓ>0}. Then the following are equivalent: ω satisfies (C2); i.e. ω∗ is a well defined weight function. Some/each ω(ℓ)∗ is well defined. Some/each ωW(ℓ) satisfies (C2). Some/each ω(W(ℓ))∗ is well defined. For some/each ℓ>0 one has the relation W(ℓ)⊲G1; i.e. 5.6∃ℓ>0/∀ℓ>0:limp→+∞Wp(ℓ)p!1/p=limp→+∞(wp(ℓ))1/p=0. Proof (i)⇔(iii) holds by (5.5) since (C2) is obviously preserved under equivalence of weight functions. And (ii)⇔(iii)⇔(iv)⇔(v) follows from Lemma 4.9 and Corollary 4.10 applied to some/any sequence W(ℓ). □ Let ω∈W0 be given with associated weight matrix Mω={W(ℓ):ℓ>0}, then in view of Lemmas 3.10, 4.9 and 5.4, and Corollary 4.10 we summarize: Assume that t=o(ω(t)) as t→+∞ holds. Then ω∗, ωW(ℓ)∗ and also ω(W(ℓ))∗ are well defined for any ℓ>0 and thus are at least weight functions in the sense of Definition 2.1. However, in this situation ω∗ satisfies (ω6). Let us consider the conjugate associated weight matrix5.7Mω∗:={(W(1/ℓ))∗:ℓ>0}, and note: Comment (e) in Sect. 4.3 implies that (W(ℓ2))∗≤(W(ℓ1))∗ for all 0<ℓ1≤ℓ2 and hence to make this fit with Definition 5.2 we have to invert the matrix parameter for ensuring the monotonic pointwise order (W(1/ℓ1))∗≤(W(1/ℓ2))∗ for all 0<ℓ1≤ℓ2. Comment (e) in Sect. 4.3 also gives that Mω is constant if and only if Mω∗ is so. Since each W(ℓ)∈LC one has that each (W(ℓ))∗ satisfies (mg); see comment (f) in Sect. 4.3. We are interested in the question whether Mω∗ is associated with an matrix admissible (BMT) weight function since this means that taking the conjugate operation (on the matrix level) “preserves the weight function setting” and one might expect the identity Mω∗=Mω∗. Proposition 5.5 We get the following: (i)Let ω be a weight function (in the sense of Definition 2.1) and assume that (C1) and (C2) holds. Then ω∗ is matrix admissible and hence Mω∗ is well defined. Let ω be a weight function (in the sense of Definition 2.1) and assume that (C1) and (C2) holds. Then ω∗ is matrix admissible and hence Mω∗ is well defined. (ii)Let ω∈W0 with associated weight matrix Mω and assume that (C2) holds. Moreover, let Mω∗ be the corresponding conjugate matrix from (5.7). Assume that there exist σ∈W0 with associated weight matrix Mσ={S(ℓ):ℓ>0} and ℓ1,ℓ2>0 such that (W(1/ℓ1))∗≈S(ℓ2). Then σ has to satisfy (ω6) and Mσ is constant.Assume that there exists σ∈W0 such that Mω∗ is associated with σ. Then (also) ω has to satisfy (ω6) and Mω is constant. Let ω∈W0 with associated weight matrix Mω and assume that (C2) holds. Moreover, let Mω∗ be the corresponding conjugate matrix from (5.7). Assume that there exist σ∈W0 with associated weight matrix Mσ={S(ℓ):ℓ>0} and ℓ1,ℓ2>0 such that (W(1/ℓ1))∗≈S(ℓ2). Then σ has to satisfy (ω6) and Mσ is constant.Assume that there exists σ∈W0 such that Mω∗ is associated with σ. Then (also) ω has to satisfy (ω6) and Mω is constant. Assume that there exist σ∈W0 with associated weight matrix Mσ={S(ℓ):ℓ>0} and ℓ1,ℓ2>0 such that (W(1/ℓ1))∗≈S(ℓ2). Then σ has to satisfy (ω6) and Mσ is constant. Assume that there exist σ∈W0 with associated weight matrix Mσ={S(ℓ):ℓ>0} and ℓ1,ℓ2>0 such that (W(1/ℓ1))∗≈S(ℓ2). Then σ has to satisfy (ω6) and Mσ is constant. Assume that there exists σ∈W0 such that Mω∗ is associated with σ. Then (also) ω has to satisfy (ω6) and Mω is constant. Assume that there exists σ∈W0 such that Mω∗ is associated with σ. Then (also) ω has to satisfy (ω6) and Mω is constant. Proof (i) By assumption ω∗ is a well defined weight function and (ω3) follows by taking into account (3.2). (ii)(a) Each (W(ℓ))∗ satisfies (mg) as mentioned in (c) before and hence S(ℓ2) has (mg), too. Then the assertions follow by (iii) in Sect. 5.2. (ii)(b) By (ii)(a) the matrix S≡Mω∗ has to be constant, so Mω is constant too and this fact implies (ω6) for ω; again see (iii) in Sect. 5.2. □ However, note that (ω6) is a (too) restrictive assumption for general BMT-weight functions in the sense that then Mω has to be constant and the corresponding weighted spaces can already be described by a single weight sequence (take any arbitrary W(ℓ)); for this recall the comparison results from [2] and the statements from [14, Sect. 5]. (ω6) is equivalent to having γ¯(ω)<+∞; if we assume even a stronger resp. more specific bound on this growth index then we can show more crucial properties and so we formulate and prove the first main result in this section. Theorem 5.6 Let ω∈W0 be given such that t=o(ω(t)) as t→+∞ (i.e. (C2)). Let Mω be the associated weight matrix and Mω∗ the corresponding conjugate matrix. (i)One has 5.8∀ℓ>0∃Dℓ≥1∀s≥0:1ℓω∗(sℓ)≤ωW(ℓ)∗(s)≤12ℓω∗(2ℓs)+Dℓ2ℓ, with Dℓ being the constants from (5.5). One has 5.8∀ℓ>0∃Dℓ≥1∀s≥0:1ℓω∗(sℓ)≤ωW(ℓ)∗(s)≤12ℓω∗(2ℓs)+Dℓ2ℓ, with Dℓ being the constants from (5.5). (ii)Assume in addition that γ¯(ω)<1, then: ω∗ satisfies (ω1).ω∗∼ωW(ℓ)∗ for any ℓ>0 and each ωW(ℓ)∗ is well defined and satisfies (ω1).Each ω(W(ℓ))∗ is well defined, satisfies (ω1) and the following equivalences hold: 5.9∀ℓ>0:ω∗∼ωW(ℓ)∗∼ω(W(ℓ))∗.The matrices Mω and Mω∗ are constant and all sequences W(ℓ) and (W(ℓ))∗ have (mg).For each ℓ>0 there exists V(ℓ) such that V(ℓ)≈W(ℓ), V(ℓ) is log-convex and v(ℓ) is log-concave. Indeed, one can even achieve that V(ℓ),(V(ℓ))∗∈LC for each ℓ>0.If ω satisfies in addition (ω1) then γ¯(ω∗)<1 and hence, in particular, ω∗ has (ω6). Assume in addition that γ¯(ω)<1, then: ω∗ satisfies (ω1).ω∗∼ωW(ℓ)∗ for any ℓ>0 and each ωW(ℓ)∗ is well defined and satisfies (ω1).Each ω(W(ℓ))∗ is well defined, satisfies (ω1) and the following equivalences hold: 5.9∀ℓ>0:ω∗∼ωW(ℓ)∗∼ω(W(ℓ))∗.The matrices Mω and Mω∗ are constant and all sequences W(ℓ) and (W(ℓ))∗ have (mg).For each ℓ>0 there exists V(ℓ) such that V(ℓ)≈W(ℓ), V(ℓ) is log-convex and v(ℓ) is log-concave. Indeed, one can even achieve that V(ℓ),(V(ℓ))∗∈LC for each ℓ>0.If ω satisfies in addition (ω1) then γ¯(ω∗)<1 and hence, in particular, ω∗ has (ω6). ω∗ satisfies (ω1). ω∗ satisfies (ω1). ω∗∼ωW(ℓ)∗ for any ℓ>0 and each ωW(ℓ)∗ is well defined and satisfies (ω1). ω∗∼ωW(ℓ)∗ for any ℓ>0 and each ωW(ℓ)∗ is well defined and satisfies (ω1). Each ω(W(ℓ))∗ is well defined, satisfies (ω1) and the following equivalences hold: 5.9∀ℓ>0:ω∗∼ωW(ℓ)∗∼ω(W(ℓ))∗. Each ω(W(ℓ))∗ is well defined, satisfies (ω1) and the following equivalences hold: 5.9∀ℓ>0:ω∗∼ωW(ℓ)∗∼ω(W(ℓ))∗. The matrices Mω and Mω∗ are constant and all sequences W(ℓ) and (W(ℓ))∗ have (mg). The matrices Mω and Mω∗ are constant and all sequences W(ℓ) and (W(ℓ))∗ have (mg). For each ℓ>0 there exists V(ℓ) such that V(ℓ)≈W(ℓ), V(ℓ) is log-convex and v(ℓ) is log-concave. Indeed, one can even achieve that V(ℓ),(V(ℓ))∗∈LC for each ℓ>0. For each ℓ>0 there exists V(ℓ) such that V(ℓ)≈W(ℓ), V(ℓ) is log-convex and v(ℓ) is log-concave. Indeed, one can even achieve that V(ℓ),(V(ℓ))∗∈LC for each ℓ>0. If ω satisfies in addition (ω1) then γ¯(ω∗)<1 and hence, in particular, ω∗ has (ω6). If ω satisfies in addition (ω1) then γ¯(ω∗)<1 and hence, in particular, ω∗ has (ω6). (iii)Both matrices Mω and Mω∗ are constant (and thus, in particular, ω satisfies (ω6)) if one assumes in addition at least one of the following assertions: There exists some ℓ0>0 such that ϑp(ℓ0)pp is almost decreasing.ω satisfies (ω1) and there exists some ℓ0>0 and a log-convex weight sequence N such that W(ℓ0)≈N and such that νppp is almost decreasing.There exists some ℓ0>0 and a log-convex sequence N such that N≈W(ℓ0) and p↦(np)1/p is non-increasing. All assertions (a)-(c) imply γ¯(ω)≤1 and, moreover, (c) yields (a). Both matrices Mω and Mω∗ are constant (and thus, in particular, ω satisfies (ω6)) if one assumes in addition at least one of the following assertions: There exists some ℓ0>0 such that ϑp(ℓ0)pp is almost decreasing.ω satisfies (ω1) and there exists some ℓ0>0 and a log-convex weight sequence N such that W(ℓ0)≈N and such that νppp is almost decreasing.There exists some ℓ0>0 and a log-convex sequence N such that N≈W(ℓ0) and p↦(np)1/p is non-increasing. All assertions (a)-(c) imply γ¯(ω)≤1 and, moreover, (c) yields (a). There exists some ℓ0>0 such that ϑp(ℓ0)pp is almost decreasing. There exists some ℓ0>0 such that ϑp(ℓ0)pp is almost decreasing. ω satisfies (ω1) and there exists some ℓ0>0 and a log-convex weight sequence N such that W(ℓ0)≈N and such that νppp is almost decreasing. ω satisfies (ω1) and there exists some ℓ0>0 and a log-convex weight sequence N such that W(ℓ0)≈N and such that νppp is almost decreasing. There exists some ℓ0>0 and a log-convex sequence N such that N≈W(ℓ0) and p↦(np)1/p is non-increasing. There exists some ℓ0>0 and a log-convex sequence N such that N≈W(ℓ0) and p↦(np)1/p is non-increasing. When we put V:={V(ℓ):ℓ>0}, then (ii)(e) shows that in general the pointwise growth order V(ℓ1)≤V(ℓ2) might fail but the corresponding weighted spaces given by V and Mω coincide and apart from this technical detail V is standard log-convex. Proof (i) We use (5.5) and follow the proof of (i) in Lemma 3.6: This gives ω∗(s)≥2ℓωW(ℓ)∗(s2ℓ)-Dℓ and ωW(ℓ)∗(s)≥1ℓω∗(sℓ) for all s≥0 and ℓ>0. (ii)(a) By (i) in Theorem 3.12 one has 0<1-γ¯(ω)≤γ(ω∗) and so ω∗ satisfies (ω1); recall Remark 2.2. (ii)(b) Let ℓ>0 be arbitrary but fixed. Then the first estimate in (5.8) gives ℓωW(ℓ)∗(s/ℓ)≥ω∗(s) for all s≥0. Thus, for any ℓ≥1 the relation ω∗(s)=O(ωW(ℓ)∗(s)) as s→+∞ is clear. If 0<ℓ<1, then choose n∈N>0 such that 1ℓ≤2n holds and iterating (ω1) for ω∗ gives for some L≥1 (depending on ℓ via n) and all s≥0 and so ω∗(s)=O(ωW(ℓ)∗(s)), too. On the other hand, the second estimate in (5.8) implies ωW(ℓ)∗(s)=O(ω∗(s)) as s→+∞ for all 0<ℓ≤12. If ℓ>12, then we choose n∈N>0 to ensure 2ℓ≤2n⇔ℓ≤2n-1, iterate (ω1) and estimate as follows for some L≥1 and all s≥0: thus also in this case ωW(ℓ)∗(s)=O(ω∗(s)). Summarizing, the equivalence between ω∗ and each ωW(ℓ)∗ is verified and this implies the rest since (C2) and the growth index γ(·) resp. (ω1) are preserved under equivalence. (ii)(c) It remains to prove the second equivalence and we apply Corollary 4.17 to W(ℓ), ℓ>0 arbitrary. First, by assumption t=o(ω(t)) and Lemma 5.4 one has that each W(ℓ) is a log-convex weight sequence such that limp→+∞(wp(ℓ))1/p=0. Second, by (5.5) and the fact that the index γ¯(·) is preserved under equivalence we have for the first equality recall (4.17). Then Corollary 4.17 applied to W(ℓ) yields the conclusion. (ii)(d) By Remark 2.2 assumption γ¯(ω)<1 implies, in particular, that ω satisfies (ω6). Thus Mω is constant, recall (iii) in Sect. 5.2, and each W(ℓ) has (mg) (recall also Corollary 4.17). Comment (e) in Sect. 4.3 yields that Mω∗ is constant too and each (W(ℓ))∗ has (mg) by (f) in Sect. 4.3. (ii)(e) (5.10) implies that each ϑp(ℓ)pp is almost decreasing and then apply Remark 4.8 to W(ℓ). (ii)(f) Remark 2.2 and the assumptions give 0<γ(ω)≤γ¯(ω)<1. Then (ii) in Theorem 3.12 implies γ¯(ω∗)≤1-γ(ω)<1, hence (ω6) for ω∗ is valid again via Remark 2.2. (iii)(a) We apply Remark 4.18 to W(ℓ0) and get γ¯(ω)=γ¯(ωW(ℓ0))=α(ϑ(ℓ0))≤1. Then (ω6) follows for ωW(ℓ0) and by (5.5) also ω (and each ωW(ℓ)) satisfies (ω6). This implies that Mω is constant, see (iii) in Sect. 5.2 and consequently Mω∗ is constant as well. (iii)(b) By equivalence we get ωW(ℓ0)∼cωN, see (4.12), and via (ω1) for ω and (5.5) we get (ω1) for ωW(ℓ0) too. Then ωW(ℓ0)∼ωN follows, see again [18, Prop. 2.2 (ii)]. So (iii)(a) applied to N gives γ¯(ω)=γ¯(ωW(ℓ0))=γ¯(ωN)≤1 and the rest follows as before. (iii)(c) Since N is log-convex, by Lemma 4.5 and Remark 4.6 it follows that N satisfies (mg) and since this condition is preserved under equivalence the sequence W(ℓ0) has (mg), too. This implies via (iii) in Sect. 5.2 that the matrix Mω is constant and hence Mω∗ is constant, too. Finally, concerning the growth index note: N≈W(ℓ0) holds if and only if n≈w(ℓ0) and so there exists A≥1 such that for any 1≤p≤q one has (wq(ℓ0))1/q≤A(nq)1/q≤A(np)1/p≤A2(wp(ℓ0))1/p. Since W(ℓ0) satisfies (mg) the estimate ϑq(ℓ0)≤B(Wq(ℓ0))1/q=B(q!wq(ℓ0))1/q≤Bq(wq(ℓ0))1/q is valid for some B≥1 and all q∈N>0 (recall again Remark 4.6). Moreover, pe(wp(ℓ0))1/p≤(p!wp(ℓ0))1/p=(Wp(ℓ0))1/p≤ϑp(ℓ0) holds for all p∈N>0 by (4.1), by log-convexity and since W0(ℓ0)=1. Summarizing, ϑq(ℓ0)q≤A2Beϑp(ℓ0)p for all 1≤p≤q is shown and since both constants A and B are not depending on p, q the sequence ϑp(ℓ0)pp is almost decreasing and the rest follows as in (iii)(a). □ For the sake of completeness we remark: Let ω be given as in Theorem 5.6 and assume that in addition 0<γ(ω)<1. Then by (ii) in Theorem 3.12 we get γ¯(ω∗)≤1-γ(ω)<1 and hence, in particular, that ω∗ satisfies (ω6). By (5.8) and [18, Prop. 2.2 (i)] it follows that ω∗∼cωW(ℓ)∗ for all ℓ>0. We are now in position to prove the second main result. Theorem 5.7 Let ω∈W0 be given such that t=o(ω(t)) as t→+∞ (i.e. (C2)) and that γ¯(ω)<1 holds. Let Mω be the associated weight matrix and Mω∗ the corresponding conjugate matrix. Then as l.c.v.s. with V∗:={(V(1/ℓ))∗:ℓ>0} and V(ℓ) denoting the sequence from (ii)(e) in Theorem 5.6. Proof The first two equalities in (5.11) follow from (ii)(c) in Theorem 5.6 (via (5.9)), Remark 5.3 and (i) in Proposition 5.5: Note that ω∗ and each ωW(ℓ)∗ are automatically matrix admissible and the same holds for each ω(W(ℓ))∗ since associated weight functions have (ω3), recall Lemma 5.1. (Alternatively, by (5.9) condition (ω3) can be transferred between all appearing weight functions.) Concerning the third equality, first note that by (ii)(e) in Theorem 5.6 we get V(ℓ)≈W(ℓ), equivalently (V(ℓ))∗≈(W(ℓ))∗, for all ℓ>0. Then (4.12) yields ω(V(ℓ))∗∼cω(W(ℓ))∗ for any ℓ>0. Finally, since (each) ω(W(ℓ))∗ satisfies (ω1), see (ii)(c) in Theorem 5.6, by using [18, Prop. 2.2 (ii)] this relation implies ω(V(ℓ))∗∼ω(W(ℓ))∗ for any ℓ>0. For the fourth equality we apply [5, Thm. 5.4] to the matrix V∗ and note: By (ii)(c) in Theorem 5.6 and the previously verified equivalence each ω(V(ℓ))∗ satisfies (ω1) too and ω(V(ℓ1))∗∼ω(V(ℓ2))∗ for any ℓ1,ℓ2>0. Moreover, recall that each (V(ℓ))∗ satisfies (mg) because V(ℓ) is log-convex; see comment (f) in Sect. 4.3. Finally note that V∗ is apart from the pointwise order of the sequences formally standard log-convex as required in [5, Thm. 5.4]. For this note that (V(ℓ))∗∈LC for each ℓ>0; recall again Remark 4.8. But the failure of the pointwise order is only a minor technical issue and can be achieved when switching to a standard log-convex matrix U∗:={(U(ℓ))∗:ℓ>0} such that (V(ℓ))∗≈(U(ℓ))∗ for each ℓ>0. The fifth equality is clear by (ii)(e) in Theorem 5.6 whereas the last equality holds by the fact that Mω is constant; see (ii)(d) in this result. □ Remark 5.8 The equalities (as l.c.v.s.) stated in (5.11) are also valid when replacing the symbol/functor E by some other weighted and analogously defined spaces (e.g. weighted sequences of complex numbers). Similarly, the function classes can also be assumed to be H-valued; see the definitions in [12, Sect. 2.3 and 2.4] for the weight sequence case. This is due to the fact that in the proof above we exclusively deal with weights and their growth and regularity properties, while no information concerning the underlying weighted category is required. In view of Corollary 4.17 and the identities (4.17) and (5.10) in order to deal with the conjugate matrix Mω∗ and the corresponding (associated) weight functions ω∗, ωW(ℓ)∗, ω(W(ℓ))∗, it seems to be useful and natural to assume γ¯(ω)<1. However, in this case both matrices Mω and Mω∗ are constant and hence this assumption is a crucial restriction within the weight function setting. And (iii) in Theorem 5.6 provides more conditions under which both matrices are constant. Ultradifferentiable Braun–Meise–Taylor classes as weighted entire spaces We are interested in establishing an isomorphism between E[ω], defined via fast growing (non-standard) weights ω, and weighted spaces of entire functions expressed in terms of ω∗ or related conjugate (associated) weight functions. Thus the aim is to transfer the main statement [12, Thm. 3.4] from the weight sequence setting to the BMT-weight function setting. As in [12, Thm. 3.4] this isomorphism is provided via the extension operator E:f↦∑k=0+∞f(k)(x0)k!(z-x0)k, x0∈I fixed, z∈C. Here I⊆R denotes an interval and E is defined on E[ω](I,H); i.e. the functions f are considered to be H-valued with H a Hilbert space. These spaces are defined analogously to the usual scalar-valued Braun–Meise–Taylor classes from [3] and for precise definitions concerning the weight sequence spaces E[M](I,H) see [12, Sect. 2.3 and 2.4]. For the forthcoming comments and notions concerning weighted entire spaces, as in Corollary 4.16, we refer to [12, Sect. 3] and [24]. When inspecting the proof of [12, Thm. 3.4] it turns out that when choosing h:=1 there the first part can be transferred for any ω∈W0 satisfying t=o(ω(t)) as t→+∞: Indeed, by Lemma 5.4 each ω(W(ℓ))∗ is well defined and then, in the Roumieu case, for some A,ℓ>0 and all z∈C we have the estimate ‖E(f)‖≤2Aexp(ω(W(ℓ))∗(2|z|)). Similarly, in the Beurling case, for all ℓ>0 there exists A>0 such that this estimate is valid for any z∈C. Thus the operator E is continuous having a similar target space compared to the one from [12, Thm. 3.4]: More precisely, the parameter is subject to the weight sequences in the matrix Mω. Note that for ω one might expect the corresponding “exponential-type weight function system” as target space and which seems to be more natural: In the Roumieu case that there exist C,ℓ>0 such that the bound on the right-hand side is given by Cexp(ℓω∗(|z|)) for all z∈C whereas in the Beurling case for all ℓ>0 there exists C>0 such that the previous estimate holds for all z∈C. Again the crucial parameter ℓ>0 is subject to the associated weight sequences. But in view of Theorem 4.15 resp. Corollary 4.17, (ii) in Theorem 5.6 (especially (5.9)) a connection between both weighted entire notions without any additional assumption on ω is unclear. However, in any case, by inspecting the second part of the proof where the continuity of the inverse (restriction) mapping is studied, it turns out that the log-convexity of the conjugate sequence M∗ is indispensable, see (4.3) and Lemma 4.3, and by (b) in Sect. 4.3 in the weight function case one requires then the log-concavity for (each) w(ℓ). But (iii)(a) in Theorem 5.6 yields the fact that under this assumption both Mω and Mω∗ are constant and ω satisfies (ω6) and this is too restrictive in the general case. Indeed, for applying (iii)(a) in Theorem 5.6 even the more general assumption that some ϑp(ℓ)pp is almost decreasing is sufficient to conclude. On regularity results by M. Markin for BMT-weight functions The aim is now to combine all information and to see how the results from [12, Sect. 4] transfer to the BMT-weight function setting. For more explanations and the background concerning the small Gevrey setting we refer to [12, Sect. 4], the original work [10] and also to [4]. So let ω∈W0 be given with associated weight matrix Mω:={W(ℓ):ℓ>0}. (Indeed, for the next basic definitions it suffices to assume that ω is matrix admissible.) First, for a densely defined (and closed) operator A on a Hilbert space H we set where D(An) is the domain of definition of An, the n-fold iteration of A; we refer to [10, Sect. 2] and [4, Sect. 1.1] for this definition. Then consider the Roumieu-type class and the corresponding Beurling-type class is defined by Thus where E{W(ℓ)}(A) and E(W(ℓ))(A) are defined accordingly; see [12, Sect. 4.3]. These definitions are inspired by the Gevrey case treated in [10, Sect. 2], see also [4, Sect. 1.2] for more general sequences, and see the settings in [12, Sect. 4.3] and in [14, Sect. 4 and 5]. Indeed, note that the analogous spaces can be introduced for an abstractly given weight matrix M={M(ℓ):ℓ>0}. Moreover, inspired by the ultradifferentiable weight function setting in the sense of Braun, Meise, and Taylor from [3], we set and Next recall that via [4, Sect. 1.3, p. 74 and 75] a different description of E{W(ℓ)}(A) resp. of E(W(ℓ))(A) (for each ℓ>0) in terms of EA, the spectral measure associated to A, can be deduced; see the first part of [4, Thm. 1.3] and its proof on [4, p. 75]: and Indeed, these equivalent representations are valid as l.c.v.s. when being equipped with their canonical topologies and this statement holds for any weight sequence M in the sense of Definition 4.2: For this note that limp→+∞(Mp)1/p=+∞ is precisely condition [4, (1.4)] and the function ρ in [4, (1.5)] corresponds to exp∘ωM. We also remark that the assumption for M to be non-decreasing in [4, Thm. 1.3] is not required in the proof and also log-convexity is not necessary for this statement. (However, each W(ℓ) satisfies these additional properties; see (i) in Sect. 5.2.) Finally, recall that limp→+∞(Mp)1/p=+∞ is natural when working with ωM in order to ensure that this function is well defined; see [25, Lemma 2.2]. Write again [·] as a common notation for {·} and (·) and then, by involving the analogous growth controls in the definitions, (5.4) and (iii) in Sect. 5.2 we infer: Theorem 6.1 Let ω∈W0 be given and let Mω be the corresponding associated weight matrix. If ω satisfies in addition (ω1), then as l.c.v.s. and (ω6) implies as l.c.v.s. Finally, the identities (6.1) and (6.2) hold for any ℓ>0. By taking into account the idea that in the weight function setting canonically one deals with a certain one-parameter family of weight sequences belonging to the set LC, namely the associated weight matrix Mω, it is interesting to study the case F=Mω with F denoting the abstractly given family of weight sequences treated in [12, Sect. 4]. This is due to the fact that in this situation a precise controlled loss of regularity is possible in the sense that one describes all appearing sequences uniformly with a given single weight function ω∈W0. We summarize: [12, Thm. 4.1] transfers immediately to each sequence W(ℓ) since this result holds for arbitrary M∈R>0N. [12, Thm. 4.1] transfers immediately to each sequence W(ℓ) since this result holds for arbitrary M∈R>0N. Having F=Mω in the crucial result [12, Lemma 4.3], then by (5.5) it follows that [12, (4.2)], i.e. 6.3∀N∈F∃M∈F:ωM(2t)=O(ωN(t)),t→+∞, implies (ω1) for ω. Moreover, via the properties of the technical but crucial uniform bound sequence a=(ap)p∈N stated in [12, Lemma 4.3 (i) and (ii)] if such a sequence exists for Mω then one has, in particular, that condition (5.6) has to be satisfied and so Lemma 5.4 implies the fact that all conjugates ω∗, ωW(ℓ)∗, ω(W(ℓ))∗ are well defined. Conversely, (ω1) for ω implies via (5.5) condition (ω1) for each ωW(ℓ) and so (6.3) (resp. [12, (4.2)]) with M=W(ℓ)=N. Having F=Mω in the crucial result [12, Lemma 4.3], then by (5.5) it follows that [12, (4.2)], i.e. 6.3∀N∈F∃M∈F:ωM(2t)=O(ωN(t)),t→+∞, implies (ω1) for ω. Moreover, via the properties of the technical but crucial uniform bound sequence a=(ap)p∈N stated in [12, Lemma 4.3 (i) and (ii)] if such a sequence exists for Mω then one has, in particular, that condition (5.6) has to be satisfied and so Lemma 5.4 implies the fact that all conjugates ω∗, ωW(ℓ)∗, ω(W(ℓ))∗ are well defined. Conversely, (ω1) for ω implies via (5.5) condition (ω1) for each ωW(ℓ) and so (6.3) (resp. [12, (4.2)]) with M=W(ℓ)=N. However, (also) for F=Mω it is crucial to see whether such a uniform bound sequence a exists and one is interested in constructing this sequence needed in [12, Lemma 4.3] via the technical result [12, Prop. 4.5]. Note that a should satisfy (*)limp→+∞(ap)1/p=0,(*)a is a uniform bound for Mω; i.e. ∀ℓ>0∃C≥1∀p∈N:Wp(ℓ)p!=wp(ℓ)≤Cap. However, (also) for F=Mω it is crucial to see whether such a uniform bound sequence a exists and one is interested in constructing this sequence needed in [12, Lemma 4.3] via the technical result [12, Prop. 4.5]. Note that a should satisfy (*)limp→+∞(ap)1/p=0,(*)a is a uniform bound for Mω; i.e. ∀ℓ>0∃C≥1∀p∈N:Wp(ℓ)p!=wp(ℓ)≤Cap. (*)limp→+∞(ap)1/p=0, limp→+∞(ap)1/p=0, (*)a is a uniform bound for Mω; i.e. ∀ℓ>0∃C≥1∀p∈N:Wp(ℓ)p!=wp(ℓ)≤Cap. a is a uniform bound for Mω; i.e. ∀ℓ>0∃C≥1∀p∈N:Wp(ℓ)p!=wp(ℓ)≤Cap. Indeed, normalization of each sequence W(ℓ) and pointwise order, i.e. [12, Prop. 4.5 (i) and (ii)], are clear and [12, Prop. 4.5 (iii)] amounts to have (5.6) which holds if t=o(ω(t)) is valid; see again Lemma 5.4. But assertion [12, Prop. 4.5 (iv)], i.e. p↦(wp(ℓ))1/p is non-increasing for each ℓ>0, is not clear in general and in fact it is too strong resp. too restrictive in the general BMT-weight function setting: By (iii)(c) in Theorem 5.6 we obtain that both matrices Mω and Mω∗ are constant and hence in this case technically one deals with the single weight sequence situation (recall that even γ¯(ω)≤1 holds). More precisely, one can choose any arbitrary W(ℓ) resp. (W(ℓ))∗. Additionally, the fact that Mω is constant clearly violates assumption [12, Prop. 4.5 (v)]. (This problem is also suggested by (ii)(e) in Theorem 5.6: When involving V we have that p↦(vp(ℓ))1/p is non-increasing for any ℓ>0 provided that γ¯(ω)<1. However, by taking into account (ii)(d) in Theorem 5.6, see also Remark 5.8, under this assumption on the growth index again both matrices Mω and Mω∗ are constant.) Summarizing, the assumptions on F=Mω in [12, Prop. 4.5] are too restrictive in the general BMT-weight function setting and, indeed, they have been inspired by the family of all small Gevrey sequences; i.e. Gs with 00, is not clear in general and in fact it is too strong resp. too restrictive in the general BMT-weight function setting: By (iii)(c) in Theorem 5.6 we obtain that both matrices Mω and Mω∗ are constant and hence in this case technically one deals with the single weight sequence situation (recall that even γ¯(ω)≤1 holds). More precisely, one can choose any arbitrary W(ℓ) resp. (W(ℓ))∗. Additionally, the fact that Mω is constant clearly violates assumption [12, Prop. 4.5 (v)]. (This problem is also suggested by (ii)(e) in Theorem 5.6: When involving V we have that p↦(vp(ℓ))1/p is non-increasing for any ℓ>0 provided that γ¯(ω)<1. However, by taking into account (ii)(d) in Theorem 5.6, see also Remark 5.8, under this assumption on the growth index again both matrices Mω and Mω∗ are constant.) Summarizing, the assumptions on F=Mω in [12, Prop. 4.5] are too restrictive in the general BMT-weight function setting and, indeed, they have been inspired by the family of all small Gevrey sequences; i.e. Gs with 00N:β>0} be a one-parameter family of sequences N(β) which satisfies the following properties: N(β1)≤N(β2)⇔n(β1)≤n(β2) for all 0<β1≤β2 (pointwise order), limp→+∞(np(β))1/p=0 for each β>0. Then there exists a sequence a=(ap)p∈N such that p↦(ap)1/p is non-increasing, limp→+∞(ap)1/p=0, and that which implies, in particular, that a is a uniform bound for F. If in addition F satisfies limj→+∞Nj(β2)Nj(β1)1/j=limj→+∞nj(β2)nj(β1)1/j=+∞ for all 0<β1<β2 (large growth difference between the sequences), then F satisfies (6.3). Proof Fix two strictly increasing sequences of positive real numbers (ck)k∈N>0 and (dk)k∈N>0 such that c1=1=d1 and limk→+∞ck=+∞=limk→+∞dk. Put j1:=1 and then choose iteratively values jk+1∈N>0, with jk+1>jk, such that and For (6.5) and (6.6) we have only used properties (i) and (ii). Now introduce a:=(ap)p∈N as follows: We set a0:=1 and Thus we have by definition and (6.5) that j↦(aj)1/j is non-increasing and tending to 0. Indeed, the rate of approaching 0 is expressed in terms of the auxiliary sequence (ck)k∈N>0 and can be chosen as fast as desired. Let k0∈N>0 be given and from now on fixed. For j≥jk0+2 we can find k≥k0 such that jk+2≤j0 and can be chosen as fast as desired; concerning the arbitrariness of (ck)k∈N>0 and (dk)k∈N>0 we also refer to Remark 6.10 below. Finally, for each given β>0 by the pointwise order we can find some k0∈N>0 such that aj1/j(nj(β))1/j≥aj1/j(nj(k0))1/j for all j≥1 and hence a satisfies (6.4), too. Note that the definition of a differs from the one in [12, Prop. 4.5] by an index shift (in (6.6)). This is because Proposition 6.2 is more general and does not use [12, Prop. 4.5 (iv) and (v)]. The last part, i.e. showing (iii), follows as in [12, Prop. 4.5]. □ Remark 6.3 We recognize that (i) and (iii) in Proposition 6.2 are preserved when multiplying each N(β) pointwise with an arbitrary given (and fixed) M∈R>0N; i.e. consider F·M:={N(β)·M:β>0}. This technical fact has the following meaning: Assume that F:={N(β):β>0} and M are given such that (i) and (iii) in Proposition 6.2 are valid for F and such that N(β)⊲M for all β>0 which should be compared with (5.6) and note that (ii) is precisely this relation with M=G1. Then F·m-1 satisfies all assumptions (i)-(iii) and Proposition 6.2 applied to F·m-1 gives a uniform bound a for this matrix. For F the sequence a·m is then relevant and hence in (6.4) one should write apmp in the numerator. By taking into account Proposition 6.2, Lemma 5.4 and comment (b) in this section we infer: Theorem 6.4 Let ω∈W0 be given with associated weight matrix Mω={W(ℓ):ℓ>0} and assume that ω satisfies t=o(ω(t)) as →+∞ (i.e. (C2)). Then there exists a=(ap)p∈N such that p↦(ap)1/p is non-increasing, limp→+∞(ap)1/p=0, and a is a uniform bound for Mω, more precisely even If in addition (ω1) for ω holds, then for each ωW(ℓ) this condition is valid too and so (6.3) (i.e. [12, (4.2)]) with M=W(ℓ)=N, ℓ>0 arbitrary. Using this result we immediately get the following variant of the crucial [12, Lemma 4.3]: Corollary 6.5 Let ω∈W0 be given with associated weight matrix Mω={W(ℓ):ℓ>0} and assume that ω satisfies t=o(ω(t)) as →+∞ (i.e. (C2)) and (ω1). If one has as sets the equality then A is bounded. Proof By Theorem 6.4 we can apply the proof of [12, Lemma 4.3] to F=Mω. Recall that the first equality in (6.8) holds by Theorem 6.1. □ Thus, the crucial result [12, Thm. 4.8] turns into the following statement: Theorem 6.6 Let ω∈W0 be given with associated weight matrix Mω={W(ℓ):ℓ>0} and assume that ω satisfies t=o(ω(t)) as →+∞ (i.e. (C2)) and (ω1). If for any weak solution y of (1.1) on [0,+∞) there exists ℓ>0 such that y∈E{W(ℓ)}([0,+∞),H), then the operator A is bounded. Proof We follow the proof of [12, Thm. 4.8] and apply it to the family F=Mω. For this we involve [12, Thm. 4.1] applied to a/the sequence W(ℓ) and Corollary 6.5. Concerning the definition of the H-valued (local) weighted function class E{W(ℓ)}([0,+∞),H) we refer again to [12, Sect. 2.3 and 2.4]. □ Remark 6.7 One shall note that the assumption in Theorem 6.6 does not mean that each weak solution y belongs to E{ω}([0,+∞),H)=E{Mω}([0,+∞),H), where this equality is valid as l.c.v.s. by (5.4) via (ω1) and the arguments from [14, Thm. 5.14 (2)]. Recall that for E{W(ℓ)}([0,+∞),H) it is required to have the defining estimates for each compact set (interval) K⊆[0,+∞) with the same weight sequence W(ℓ) and hence uniformity w.r.t. the matrix parameter ℓ>0 on whole [0,+∞), whereas for the local classes E{ω}([0,+∞),H) naturally this parameter is allowed to depend on chosen K. Finally, we transfer [12, Sect. 4.4, Thm. 4.9] where a connection to the weighted entire setting is established: More precisely, that each weak solution y can be extended to an entire function satisfying a precise prescribed growth. For this again the previously described technical issue for F=Mω appears: Assertions [12, Sect. 4.4 (i)-(iii)] yield no problems, concerning (iii) recall Proposition 6.2. But [12, Sect. 4.4 (iv)] means log-concavity for each w(ℓ) and which is required in order to involve [12, Thm. 3.4]; see also Corollary 4.16 and the explanations in Sect. 5.4. As shown before this requirement is too restrictive in the sense that all sequences are equivalent and hence one loses the flexibility of the matrix parameter. However, for the sake of completeness when combining Theorem 6.6 with [12, Thm. 3.4] (and Corollary 4.16) let us formulate the final statement: Theorem 6.8 Let ω∈W0 be given with associated weight matrix Mω={W(ℓ):ℓ>0}. Assume that ω satisfies t=o(ω(t)) as →+∞ (i.e. (C2)), (ω1), each ϑp(ℓ)pp is almost decreasing. If for any weak solution y of (1.1) there is ℓ>0, C>0 and d>0 such that y can be extended to an entire function satisfying then the operator A is bounded. Alternatively, in (6.9) one can involve the bound Cexp(ω(ℓ)∗(d|z|)) and recall that (iii) implies γ¯(ω)≤1 and hence the restrictive condition (ω6). If assertion (iii) is replaced by the stronger condition γ¯(ω)<1, then in (6.9) alternatively one can use the bound Cexp(ω∗(d|z|)) and also the “exponential-type weight function system bounds” Cexp(dω(W(ℓ))∗(|z|)), Cexp(dω(ℓ)∗(|z|)), Cexp(dω∗(|z|)). Proof Since W(ℓ)∈LC for each ℓ>0, in view of assumption (iii) and Remark 4.8 we have that for each ℓ>0 there exists V(ℓ)∈LC such that W(ℓ)≈V(ℓ), (W(ℓ))∗≈(V(ℓ))∗ and (V(ℓ))∗∈LC. Equivalence W(ℓ)≈V(ℓ) preserves the (local) ultradifferentiable classes, i.e. E[W(ℓ)](I,H)=E[V(ℓ)](I,H) for each ℓ>0, and then apply [12, Thm. 3.4] (resp. Corollary 4.16) to (each) V(ℓ) and recall (4.12): (W(ℓ))∗≈(V(ℓ))∗ implies ω(W(ℓ))∗∼cω(V(ℓ))∗ and this relation preserves the bound in (6.9). Thus Theorem 6.6 gives the conclusion. The supplement concerning the alternative bound follows by Corollary 4.16 applied to W(ℓ). If even γ¯(ω)<1 holds, then (ii) in Theorem 5.6 implies the additional information on the bounds: In this case all conjugate functions are equivalent (see (5.9)) and satisfy both (ω1) and (ω6). □ We finish by constructing explicitly a weight function ω satisfying all requirements in Theorem 6.6 and such that the corresponding associated weight matrix is non-constant which means that under the assumptions on ω in this result the flexibility w.r.t. the matrix parameter can become crucial. This also illustrates then the difference between Theorems 6.6 and 6.8. In order to proceed we focus on ω=ωM. Proposition 6.9 There exists ω∈W0 satisfying (C2), (ω1) and such that the associated weight matrix Mω is non-constant (equivalently (ω6) for ω fails). Note: In view of Remark 2.2 and Lemma 3.11 the weight ω constructed in this proposition satisfies This also illustrates that in general in (i) in Theorem 3.12 one has 1-γ¯(σ)<γ(σ∗). Proof We construct a sequence M such that ωM satisfies all required properties. For this consider the corresponding sequence of quotients μ=(μp)p∈N with μ0:=1 and set Mp:=∏i=0pμi. Let us fix ϵ∈(0,1) and an arbitrary non-decreasing sequence of positive real numbers (dj)j∈N>0 satisfying dj≥1 for all j and limj→+∞dj=+∞. Next introduce several auxiliary sequences: First, consider (ap)p∈N by Second, let (cj)j∈N>0 be a strictly increasing sequence in N≥2 such that and define the corresponding sequence (pj)j∈N by Note that the choice in (6.11) is possible since 1+ϵ<2. Moreover, put Finally, the sequence μ of quotients is given by Claim I M∈LC holds and hence ωM∈W0 (see Lemma 5.1). By (6.13), (6.14) and the choice of (dj)j it follows that p↦bp is non-decreasing and limp→+∞bp=+∞. Normalization 1=M0≤M1 amounts to have μ1≥1 and which holds since μ1=μa0=b0=1. Claim II limp→+∞μpp=0 holds. By (6.15) we see that μq+1q+1<μqq for all ap≤q0 in (6.11). Summarizing, the claim is shown. Claim III limp→+∞(mp)1/p=0 holds and hence ωM satisfies (C2) (see Lemma 4.9 and Corollary 4.10). Since M∈LC we get (Mp)1/p≤μp⇔μ1⋯μp≤μpp for all p∈N>0 (and the second estimate is valid for p=0, too). Now, by (4.1) we obtain and so (mp)1/p≤eμpp for all p∈N>0 which yields the conclusion by using Claim II. Claim IV ωM satisfies (ω1). In order to conclude it suffices to check that lim infp→+∞μQpμp>1 for some Q∈N≥2 (see (A.1)); for a direct proof of this implication we refer to [2, Lemma 12, (2)⇒(4)] (cf. also [22, Prop. 3.4]) and see [6, Thm. 2.11, Cor. 2.14, Thm. 3.11, Cor. 4.6 (i)] involving growth index arguments. By definition in (6.15) one has μ2pμp≥1+ϵ for all p∈N>0 sufficiently large, since dj≥1+ϵ for all j∈N>0 large enough and so the desired property holds with Q:=2. Claim V ωM does not satisfy (ω6) which is equivalent to the fact that M violates (mg) (see Lemma 5.1) and equivalently MωM is non-constant (recall (iii) in Sect. 5.2). We use the characterization shown in [15, Lemma 2.2], recall again the discussion in [23], and verify supp∈Nμ2pμp=+∞: For this choose p=apj-1, j∈N>0, then by (6.14) and (6.15) we get μ2pμp=μapjμapj-1=djbpj-1bpj-1=dj and recall that dj→+∞. □ Remark 6.10 The proof of Proposition 6.9 is quite technical but it can directly be generalized when replacing j in (6.11) by fj such that (fj)j∈N>0 is a further given strictly (fast) increasing sequence. Of course, (fj)j∈N>0 via the choice of (cj)j∈N>0 will effect then the definition of M but we get that the conclusions in the above result even hold when p↦μpp resp. p↦(mp)1/p is required to tend faster to 0 and hence the rate of approaching 0 is not effecting the validity of this (counter-)example. On the other hand, the sequence (dj)j∈N>0 is used to achieve that M violates (mg) and hence MωM fails (iii) in Theorem 6.8. Acknowledgements The author of this article thanks the two anonymous referees for the careful reading and the valuable suggestions which have improved the presentation of the results. This research was funded in whole or in part by the Austrian Science Fund (FWF) 10.55776/PAT9445424. Appendix A. On slowly varying associated weight functions Motivated by the comment in Sect. 2.2 that for a given weight function σ the reflexivity σ⊲cσ implies the fact that σ has to be slowly varying, the goal of this appendix is to investigate this condition for ωM in terms of the defining sequence M. Proposition A.1 Let M be a log-convex weight sequence and recall the notation μp:=MpMp-1, p∈N>0. When M satisfies to equations (A.1) resp. (A.2) A.1∃Q∈N≥2:lim infp→+∞μQpμp>1 and (A.1) resp. (A.2) instead of (A.1) resp. A.2 in the appendix A.2limp→+∞μp(Mp-1/M0)1/(p-1)=limp→+∞μp(Mp-1)1/(p-1)=+∞, then ωM is slowly varying; i.e. (2.3): ∀u>0:limt→+∞ωM(ut)ωM(t)=1. When ωM is slowly varying, then (A.2) has to be satisfied. Proof First recall that for any weight function ω it suffices to check (2.3) for all u>1: When 01 large be given and fixed and let t≥μ1. Then there exist p,q∈N>0, q≥p, such that μp≤t<μp+1 and μq≤ut<μq+1. For this note that p↦μp is non-decreasing and limp→+∞μp=+∞. In view of Lemma 4.3 we have The estimate is valid since μp≤t<μp+1 and so by log-convexity Let ϵ>0 be given and then, in order to ensure that ωM is slowly varying, we put t=μp and so it suffices to require for all sufficiently large p≥2 that Now, since M satisfies A.1 we have the following: Iteration of this estimate yields μQnpμp≥(1+η)n for all p≥pη and all n∈N>0. Hence, when u>1 is given, then for n chosen sufficiently large to ensure u<(1+η)n we obtain: For any t≥μpη one can find p≥pη such that μp≤t<μp+1 and then ut1 and t>μ1 we rewrite and estimate the crucial quotient under consideration as follows: Since ωM is slowly varying, we get By combining both estimates, taking into account (again) Lemma 4.3 and finally the facts that p↦μp is non-decreasing and limp→+∞μp=+∞, one infers thus log(up)≤ϵlog(M0μpp/Mp). Similarly, as in part (i) this estimate is equivalent to having up/ϵMpM0=up/ϵμ1⋯μp≤μpp and so, finally, we arrive at As u>1 one has up(p-1)ϵ≥u1/ϵ for all p≥2 and so (A.2) is shown. □ We finish with the following observations (M being a log-convex weight sequence): (A.2) obviously violates (4.8) and hence (mg) fails, see the discussion in Remark (4.6). Indeed, the formal negation of (4.8) is supp∈N>0μp+1(Mp)1/p=+∞ and (A.2) is much stronger. Note that any slowly varying weight function ω cannot have ω6 since this condition implies ∃H0≥1∀H≥H0:lim inft→+∞ω(Ht)ω(t)≥2. Moreover, recall that condition (ω6) for ωM is equivalent to mg; again we refer to [15, Lemma 2.2], [11, Thm. 1] and [23]. Therefore, (ii) in Proposition A.1 is consistent with the known information concerning mg. Let ω be a slowly varying weight function, then it is straightforward to see γ(ω)=+∞. Thus γ¯(ω)=+∞ too and again ω6 fails by taking into account Remark 2.2. Finally, concerning (i) in Proposition (A.1) recall that (A.1) precisely corresponds to the condition appearing in [6, Thm. 3.11 (v)] with (β=0) (apart from the non-effecting facts that M0=1 might fail, see Remark 4.4, and the appearing index shift when considering the sequences of quotients as described at the end of Sect. 4.1). Thus via [6, Thm. (3.11)] assumption (A.1) precisely means for the growth index from [27] that γ(M)>0, and under the additional assumption (A.2) even γ(ωM)=+∞ follows as commented in (iii). Since γ(M)=+∞ is not clear, this part and the above comments (i), (ii) and (iii) are consistent with [6, Cor. 4.6 (i) and (iii)]. Funding Open access funding provided by University of Vienna. Data Availability The author confirms that during the preparation of this article no data have been generated. Declaration Conflict of interest The author confirms that there is not any conflict of interest. References Associated Data Data Availability Statement The author confirms that during the preparation of this article no data have been generated.