Ultradifferentiable regularity properties of a class of hypoelliptic operators Authors: Stefan Fürdös Published: 2026 ABSTRACT AbstractWe investigate the ultradifferentiable regularity of solutions and vectors of a class of hypoelliptic differential operators first introduced by Hörmander. Our results extend results of Hashimoto–Morimoto–Matsuzawa and Bolley–Camus–Metivier.Keywords: Ultradifferentiable hypoellipticity, Theorem of Iterates Introduction The question of regularity of vectors of differential operators, commonly referred to as the problem of iterates, has recently experienced a surge in interest, see e.g. [7–9, 13, 15] or [16]. The survey [5] gives an overview of the older literature and the connection of the problem of iterates with the problem of hypoellipticity. We recall that a smooth function f∈E(Ω) defined in an open set Ω⊆Rn is in the Gevrey class Gs(Ω) of order s≥1 (The case s=1 corresponds to the class of real-analytic functions) if for all compact sets K⊆Ω there are constants C,h>0 such that for all α∈N0n. If P is a differential operator of order d with real-analytic coefficients on Ω then a distribution u∈D′(Ω) is an s-Gevrey vector, s≥1, if Pku∈Lloc1(Ω) for all k∈N0 and for all compact K⊆Ω there are constants C,h>0 such that for all k∈N0. The space of s-Gevrey vectors of P is denoted by Gs(Ω;P). If P is an elliptic differential operator with real-analytic coefficients on an open set Ω⊆Rn then Gs(Ω;P)=Gs(Ω), cf. [3] (for s=1 see also [18] and [20]). According to [21] the converse is also true for s>1. On the other hand it was shown in [1] for hypoelliptic differential operators P of principal type that G1(Ω;P)=G1(Ω) and there is a loss of regularity for u∈Gs(Ω,P). Similar results were proven by Derridj, see [8], for sum of squares of real vector fields of finite type. This brief note can be viewed as a continuation of the recent paper [12] of Schindl and the author, where a novel approach to the problem of iterates was introduced. In particular, we proved, [12, Theorem 1.1], that for hypoelliptic operators of principal type an ultradifferentiable theorem of iterates hold with respect to certain ultradifferentiable classes given by weight functions in the sense of Braun–Meise–Taylor [6]. We focus our study in this paper on a class of hypoelliptic differential operators which were originally introduced by Hörmander, for a definition see the next section. In [2] Bolley–Camus investigated the Gevrey hypoellipticity and the regularity of Gevrey vectors of these operators. Their results on the Gevrey hypoellipticity were improved by Hashimoto–Morimoto–Matsuzawa [14], whereas Bolley–Camus–Metivier [4] sharpened the results on the regularity of Gevrey vectors. We will analyze the hypoellipticity and regularity of vectors of operators in this class in the general ultradifferentiable setting given by weight matrices [22] which generalize the setting of weight sequences, cf. e.g. [19], and the classes given by weight functions, see [6]. In particular, we will prove the theorem of iterates for these operators for the same family of Braun–Meise–Taylor classes which have been considered in [12, Theorem 1.1]. Note that by [13, Remark 2.8] an analogous statement cannot hold for Denjoy–Carleman classes. Statement of main results We focus in this article on a class of linear partial differential operators P(x, D) of some given order d, with real-analytic coefficients in some open set Ω, whose symbols p(x,ξ) satisfy the following conditions: There are constants ρ and δ such that 0≤δ<ρ≤1 and for all compact K⊂Ω there are constants C,A>0 and M>0 such that |∂ξα∂xβp(x,ξ)|≤CA|α|+|β||α|!|β|!|p(x,ξ)||ξ|-ρ|α|+δ|β| for x∈K, ξ∈Rn with |ξ|≥M, α,β∈N0n. There are constants ρ and δ such that 0≤δ<ρ≤1 and for all compact K⊂Ω there are constants C,A>0 and M>0 such that |∂ξα∂xβp(x,ξ)|≤CA|α|+|β||α|!|β|!|p(x,ξ)||ξ|-ρ|α|+δ|β| for x∈K, ξ∈Rn with |ξ|≥M, α,β∈N0n. There exists some d′>0 and for all compact subsets Ω⊆K there are constants C>0 and L>0 such that |ξ|′≤C|p(x,ξ)| for x∈K, ξ∈Rn with |ξ|≥L. There exists some d′>0 and for all compact subsets Ω⊆K there are constants C>0 and L>0 such that |ξ|′≤C|p(x,ξ)| for x∈K, ξ∈Rn with |ξ|≥L. We denote the space of operators satisfying conditions (1) and (2) by Aρ,δd,d′(Ω). As noted above we want to discuss the ultradifferentiable hypoellipticity of operators in Aρ,δd,d′(Ω). In the literature there are mainly two closely related families of ultradifferentiable classes: first, Denjoy–Carleman classes with weight sequences as defining data, and Braun–Meise–Taylor classes given by weight functions. Definition 2.1 A weight sequence is a sequence M=(Mk)k of positive real numbers such that M0=1 and A function f∈E(Ω) is ultradifferentiable of class {M}, respective of class (M), if for all compact sets K⊆Ω there are constants C,h>0, respective for all h>0 there exists C>0, such that The spaces of functions of class {M} and (M) are denoted by E{M}(Ω) and E(M)(Ω), respectively. Definition 2.2 A weight function is a continuous, increasing function ω:[0,∞)→[0,∞) such that ω[0,1]≡0 and the following conditions are satisfied: If we consider also the conjugate function of φω given by φ∗(t)=sups≥0(st-φ(s)) then a smooth function f∈E(Ω) is of class {ω} if for all compact sets K⊆Ω there are constants C,h>0 such that On the other hand f is of class (ω) if for all compact K⊆Ω and every h>0 there exists C>0 such that the estimate (2.1) is satisfied. In the following we will use the notation [∗]={∗},(∗) where ∗=ω,M. If P is a differential operator then we say that P is [∗]-hypoelliptic if singsupp[∗]Pu=singsupp[∗]u for every u∈D′(Ω). Here singsupp[∗]u is the singular support of u with respect to the class E[∗](Ω), which is defined in the obvious way. Definition 2.3 A weight sequence M is semiregular if Mk/k!k→∞ for k→∞ and An example of a semiregular weight sequence is the Gevrey sequence Gσ=(k!σ)k of index σ≥1. We are now able to formulate the main statements of this paper: Theorem 2.4 Let P∈Aρ,δd,d′(Ω) and θ=(ρ-δ)-1. Then the following statements hold: P is {M}-hypoelliptic for any semiregular weight sequence M such that supkkθ/Λk<∞ where Λk=Mkk is the quotient sequence associated to M. P is {M}-hypoelliptic for any semiregular weight sequence M such that supkkθ/Λk<∞ where Λk=Mkk is the quotient sequence associated to M. P is (M)-hypoelliptic for any semiregular weight sequence M with limk→∞kθΛk=0. P is (M)-hypoelliptic for any semiregular weight sequence M with limk→∞kθΛk=0. P is {ω}-hypoelliptic for every weight function ω with ω(t)=O(tρ-δ), t→∞. P is {ω}-hypoelliptic for every weight function ω with ω(t)=O(tρ-δ), t→∞. P is (ω)-hypoelliptic for every weight function ω such that ω(t)=o(tρ-δ), t→∞. P is (ω)-hypoelliptic for every weight function ω such that ω(t)=o(tρ-δ), t→∞. If P is a differential operator with analytic coefficients of order d and ω a weight function then a distribution u∈D′(Ω) is an element of E{ω}Ω;P if Pku∈Lloc1(Ω) for all k∈N0 and for every V⋐Ω there are C,h>0 such that On the other hand, u is an element of E(ω)(Ω;P) if Pku∈Lloc1(Ω) for all k∈N0 and for every V⋐Ω and all h>0 there is C>0 such that (2.2) is satisfied. Theorem 2.5 Let ω be a weight function such that Then for any P∈Aρ,δd,d′(Ω) we have that Ultradifferentiable hypoellipticity If M and N are two weight sequences then we set and Definition 3.1 A weight matrix M is a family of weight sequences such that for each pair M,N∈M we have either M≤N or N≤M. Let Ω⊆Rn be an open set and M a weight matrix. A smooth function f∈E(Ω) is ultradifferentiable of class {M} if for each compact set K⊆Ω there are constants C,h>0 and some M∈M such that We say that f is of class (M) if for each compact set K⊆Ω and every h>0 and M∈M there is a constant C>0 such that (3.1) is satisfied. We write E[M](Ω) for the space of functions of class [M]. If M and N are two weight matrices then we write M{⪯}N if for all M∈M there is some N∈N such that M⪯N and M(⪯)N if for all N∈N there is M∈M such that M⪯N. Moreover, M{⊲)N if M⊲N for all M∈M and all N∈N. Then M[⪯]N implies that E[M](Ω)⊆E[N](Ω) whereas E{M}(Ω)⊆E(N)(Ω) when M{⊲)N. Definition 3.2 Let M be a weight matrix. We say that M is R-semiregular if the following conditions hold: 3.2∀M∈Mlimk→∞Mkk!k=∞3.3∀M∈M∃N∈M∃Q>0Mk+1≤Qk+1Nk∀k∈N0. We say that M is R-semiregular if the following conditions hold: 3.2∀M∈Mlimk→∞Mkk!k=∞3.3∀M∈M∃N∈M∃Q>0Mk+1≤Qk+1Nk∀k∈N0. The weight matrix M is B-semiregular if (3.2) and 3.4∀N∈M∃M∈M∃Q>0Mk+1≤Qk+1Nk∀k∈N0 hold. The weight matrix M is B-semiregular if (3.2) and 3.4∀N∈M∃M∈M∃Q>0Mk+1≤Qk+1Nk∀k∈N0 hold. Remark 3.3 Condition (3.2) means that G1(Ω)⊊E[M]. Condition (3.2) means that G1(Ω)⊊E[M]. Condition (3.3) (resp. (3.4)) implies that the Roumieu class E{M}(Ω) (resp. the Beurling class E(M)(Ω)) is closed under derivation. Condition (3.3) (resp. (3.4)) implies that the Roumieu class E{M}(Ω) (resp. the Beurling class E(M)(Ω)) is closed under derivation. In fact, if M is [semiregular] then E[M](Ω) is closed under composition with real-analytic mappings: Let Φ:Ω1→Ω2 be a real-analytic mapping then f∘Φ∈E[M](Ω1) for all f∈E[M](Ω2). In fact, if M is [semiregular] then E[M](Ω) is closed under composition with real-analytic mappings: Let Φ:Ω1→Ω2 be a real-analytic mapping then f∘Φ∈E[M](Ω1) for all f∈E[M](Ω2). We are going to need the following characterization of E[M](Ω) from [11]. Theorem 3.4 ([11, Proposition 5.1]) Let u∈D′(Ω), M be a weight sequence and p0∈Ω. Then the following holds: If M is R-semiregular then u is of class {M} near p0 if and only if for some neighborhood V of p0 there is a bounded sequence uk∈E′(Ω) such that uk|V=u|V and there are a constant h>0 and some M∈M so that 3.5supξ∈Rnk∈N0|ξ|k|u^k(ξ)|hkMk<∞. If M is R-semiregular then u is of class {M} near p0 if and only if for some neighborhood V of p0 there is a bounded sequence uk∈E′(Ω) such that uk|V=u|V and there are a constant h>0 and some M∈M so that 3.5supξ∈Rnk∈N0|ξ|k|u^k(ξ)|hkMk<∞. If M is B-semiregular then u is of class (M) near p0 if and only if for some neighborhood V of p0 there is a bounded sequence uk∈E′(Ω) such that uk|V=u|V and (3.5) holds for all h>0 and every M∈M. If M is B-semiregular then u is of class (M) near p0 if and only if for some neighborhood V of p0 there is a bounded sequence uk∈E′(Ω) such that uk|V=u|V and (3.5) holds for all h>0 and every M∈M. For the definition and theory of the wavefront set WF[M]u associated to ultradifferentiable classes given by weight matrices we refer to [11] and [10]. Theorem 3.5 Let P∈Aρ,δd,d′(Ω). Then the following holds: P is {M}-hypoelliptic in Ω for any R-semiregular weight matrix M such that Gθ{⪯}M. P is {M}-hypoelliptic in Ω for any R-semiregular weight matrix M such that Gθ{⪯}M. P is (M)-hypoelliptic in Ω for any B-semiregular weight matrix M such that Gθ{⊲)M. P is (M)-hypoelliptic in Ω for any B-semiregular weight matrix M such that Gθ{⊲)M. Proof Following the arguments in the proof of [14, Theorem 3.1], see also [23, Theorem 3.26] and [4], one sees that for any U⋐Ω there are operators F,R:E′(U)→D′(U) such that FP=Id+R in U. Moreover, the kernel of R is in Gθ(U) whereas the kernel of F, which we also denote by F, satisfies singsuppθF⊆{(x,x):x∈U}⊆U×U. Now, if M is an R-semiregular weight matrix such that Gθ{⪯}M, then singsupp{M}Fu⊆singsupp{M}u and Ru∈E{M}(U) for all u∈E′(U) by [10, Theorem 3.14]. Thence for all u∈E′(U), which proves (1). The second part is proven completely analogously. □ Theorem 3.5 immediately implies Theorem 2.4(1),(2). The other parts of Theorem 2.4 also follow from Theorem 3.5 if we also consider the Remark below. Remark 3.6 If ω is a weight function then the weight matrix W associated to ω consists of the weight sequences Wλ, λ>0, given by Then W has the following properties: In fact, E[ω](Ω)=E[W](Ω) as topological spaces, cf. [22]. Moreover it is well-known, that ω(t)=O(t1/s), s≥1 is equivalent to Gs{⪯}W and Gs(⪯)W. In the same manner, ω(t)=o(t1/s) is equivalent to Gs{⊲)W. Ultradifferentiable scales and a general theorem of iterates If P is a differential operator with analytic coefficients of order d and M is a weight matrix satisfying (3.2) then the space E{M}Ω;P consists of those distributions u∈D′(Ω) such that Pku∈Lloc1(Ω) and for all V⋐Ω there are M∈M and C,h>0 such that whereas a distribution u∈D′(Ω) is an element of E(M)(Ω;P) if Pku∈Lloc1(Ω) for all k∈N0 and for all V⋐Ω, every M∈M and each h>0 there is a constant C>0 such that (4.1) holds. If ω is a weight function with ω(t)=o(t) and W is the weight matrix associated to ω then, in view of Remark 3.6, we have that E[ω](Ω;P)=E[W](Ω;P) for any differential operator P with analytic coefficients. The rest of this paper is dedicated to the proof of Theorem 2.5 which will be in the same spirit as the proof of [12, Theorem 1.1]. For this we need to recall the definition of ultradifferentiable scales from [12, Subsection 4A]. However, in order to be able to use respective adapt the arguments of the proof of [2, Theoreme 1.2] we have to modify the definition of ultradifferentiable scales a little bit. Definition 4.1 Let Λ be a totally ordered set. A mapping ζ:Λ×[0,+∞)⟶[0,∞) is said to be a generating function if for each λ∈Λ the function ζλ=ζ(λ,.) is continuous and ζλ(t)/t is increasing on t≥1 and the following conditions are satisfied: We associate to a generating function a scale (Mλ)λ of weight sequences by setting The matrix Mζ=M={Mλ:λ∈Λ} is the weight matrix associated to the scale (Mλ)λ. We see that Mζ satisfies (3.2) due to (4.4). We are going to need the following conditions: Note that (4.5) (resp. (4.6)) implies that Mζ satisfies (3.3) (resp. (3.4)), cf. [24]. In order that every weight sequence Mλ is itself semiregular we need to require that Furthermore, we also consider the following conditions Finally, for σ≥1 we say that the ultradifferentiable scale (Mζλ) is superordinated to Gσ if Clearly that means that {Gσ}{⊲)Mζ. We are now able to begin with the proof of Theorem 2.5. To this end assume that P∈Aρ,δd,d′(Ω). We are making use of the parametrices AN of the iterated operators PN, N∈N, constructed in [4]. Here we only briefly summarize the results we are going to need to analyze the regularity of ultradifferentiable vectors of P. The symbols aN of the operators AN have asymptotic expansions aN(x,ξ)∼∑j≥1aN,j(x,ξ) with the symbols aN,j being constructed in [4, II−1.2]. In fact, by [4, Equation (1.10)] the symbols aN are of the form for x∈V⋐Ω and ξ∈Rn. Here λ≥1 to be specified later on. Moreover χ~j(η)∈C∞(Rn) is a sequence such that χ~0(η)=0 for |η|≤L and χ~0(η)=1 if |η|≥2L and for j≥1 we have that for a constant C>0 independent of j, cf. [4, pp. 57–58]. According to [4, Corollaire 2.7] there are λ>0, ε>0 and C>0 such that for all ξ∈Rn, |ξ|≥C, all N∈N, ψ∈C0∞(V¯) and all u∈D′(Ω) with Pku∈Lloc1(Ω) for k∈{1,⋯,N} we have that where Sk are error terms (defined in [4, p. 63]) satisfying for v∈Lk(V) and |ξ|≥2λL. Now let x0∈V⋐Ω and we fix also ψ∈C0∞(V) equal to 1 near x0. We choose also a sequence φN∈D(V) such that φN=1 near x0 and φNψ=φN and for all β∈N0n there is a constant Cβ so that for all N≥1. Moreover, we set χN(x)=χ~qN(x/2λ) where q=[2d′/ρ]+2. Then we have for u∈D′(Ω) with Pku∈Lloc1(Ω) for all k∈N0. We need to estimate the right-hand side of (4.10) first for u∈D′(Ω) satisfying the following estimates: There are a weight sequence M and constants C,h>0 such that and we recall from above that Λk=Mkk denotes the sequence of quotients associated to M. By the proof of [4, Corollary 2.9] we obtain that there is a constant C1>0 such that for |ξ|>C1 and 1≤j≤k we have that On the other hand, observe that if |ξ|d′/d≥2Λdk. Thus we have proven the following statement. Lemma 4.2 There exists a constant C2>0 such that for |ξ|≥2C2ΛdNd/d′. Regarding the third term we have by the proof of [4, Proposition 2.10] that there are constants C3,C4>0 such that for |ξ|≥C3 and j=1,⋯,k and k∈N. As above we thus get: Lemma 4.3 There exists a constant C5>0 such that for |ξ|≥C5ΛdNd/d′. For the first term, we observe that by [4, Proposition 2.12] there is a constant C7>0 such that for all N∈N and |ξ|≥C7 we have that Finally, regarding the second term the proof of [4, Proposition 2.11] gives that Assume now that M=Mζλ for some generating function satisfying (4.5). Iterating (4.5) shows that for each λ∈Λ and α>0 there are λ′∈Λ and γ>0 such that where ν∈N is the smallest integer such that α≤2ν. Thus we obtain Summarizing we have proven the following fact: if (Mζλ) is an ultradifferentiable scale superordinate to Gθ such that (4.5) holds and u∈E{Mλ}Ω;P (resp. u∈E(Mλ)(Ω;P) then there exist λ′∈Λ and a constant c>0 and some C0,h0>0 (resp. for each h0>0 there is a constant C0) such that for |ξ|≥cexp(d/(d′)2ζλ(d′N)) and N≥1. Since the sequence (φNu)N is bounded in E′(Ω) the Banach-Steinhaus theorem implies that there is a constant μ≥0 such that that there is a constant C9>0 such that Thence, in fact, there are C0,h0>0 (resp. for every h0>0 there exists C0>0) such that for N∈N and ξ∈Rn. In order to continue for N∈N we set where ⌈ρ⌉ is the smallest integer larger than ρ. In particular note that N+d+n+μ≤d′N^≤N+d+n+μ+d~, where d~=⌈d′⌉. By the above estimates we obtain for |ξ|≥1. If (4.8) holds then we have that there exists λ~∈Λ such that there are C0,h0>0 (resp. for each h0>0 there is C0>0) so that and therefore, if (4.7) holds then there are C0,h0>0 (resp. for every h0 there is C0>0) such that Thence using Theorem 3.4 we obtain that x0∉singsupp[Mλ~]u and since x0∈V⋐Ω have been chosen arbitrarily we have proven the following theorem. Theorem 4.4 Let P∈Aρ,δd,d′(Ω) and (Mζλ) be an ultradifferentiable scale superordinate to Gθ, θ=1ρ-δ. If the generating function ζ additionally satisfies (4.5), (4.7) and (4.8) then for any λ there is some λ~ such that In fact, since our arguments are completely local, we can easily see that also the following statement is true. Theorem 4.5 Let P∈Aρ,δd,d′(Ω) and (Mζλ) be an ultradifferentiable scale superordinate to Gθ, θ=1ρ-δ. If (4.5) and (4.8) hold then E{Mζ}Ω;P=E{Mζ}(Ω). If (4.5) and (4.8) hold then E{Mζ}Ω;P=E{Mζ}(Ω). If (4.6) and (4.9) hold then E(Mζ)(Ω;P)=E(Mζ)(Ω). If (4.6) and (4.9) hold then E(Mζ)(Ω;P)=E(Mζ)(Ω). Theorem 2.5 follows now from Theorem 4.5 and the following Remark. Remark 4.6 Let ω be a weight function. According to [12, Section 5A] ζω(t,λ)=1λφω∗(λt) is a generating function. The convexity of φ∗ implies that Hence both (4.5) and (4.6) are satisfied. Moreover, by [17, (A.2)] condition (Ξ) implies that Arguing as in [17, p. 2172] we see that we can conclude that (Ξ) gives that Since we have that φ∗(s)+φ∗(t)≤φ∗(s+t), cf. [22], we obtain in fact Thence, if ω satisfies (Ξ) then both (4.8) and (4.9) hold for the generating function ζω. Moreover, by [13, Remark 2.8(3)] we obtain that if ω is a weight function such that (Ξ) holds then ω(t)=o(tγ) for any 0<γ<1. Now according to [25, Corollary 5.10] the fact ω(t)=o(tγ), 0<γ<1, is equivalent to the fact that Gσ⊲Wλ for all λ>0, where σ=1/γ. That means that, if (Ξ) holds for ω then We want to replace k∈N in the limit with t>0. To this end observe that both tσ, σ>1, and the function t↦1tφω∗(t) are increasing by [6]. Thence we obtain for t∈[k,k+1], k∈N, and s=⌈σ⌉ that by (4.11). Therefore we have proven that if ω is a weight function satisfying (Ξ) then the ultradifferentiable scale generated by ζω(t,λ)=1λφω∗(λt) is superordinate to Gσ for all σ≥1. Acknowledgements This work was funded in whole or in part by the Austrian Science Fund (FWF) 10.55776/PAT1994924. Funding Open access funding provided by University of Vienna. Data availability No datasets were generated or analysed during the writing of this article. Declarations Conflict of interest The author has no conflict of interest to declare that are relevant to the content of this article. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement No datasets were generated or analysed during the writing of this article.