Abstract
We introduce a new class of ultradifferentiable pseudodifferential operators on the torus whose calculus allows us to show that global hypoellipticity, in ultradifferentiable classes, with a finite loss of derivatives of a system of pseudodifferential operators, is stable under perturbations by lower order pseudodifferential operators whose order depends on the loss of derivatives. The key point in our study is our definition of loss of derivatives. We also give an easy proof of the fact that if a system of pseudodifferential operators is globally \({\mathcal {M}}\)-hypoelliptic then its transpose is globally solvable in \(D'_{\mathcal {M}}\left( {\mathbb {T}}^N\right) \). Finally we present an application of our results.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
If A is an elliptic properly supported pseudodifferential operator of order m in a \(C^\infty \) manifold then \(Au \in H^{\text {loc}}_{(s)}\) implies \(u\in H^{\text {loc}}_{(s+m)}\) (cf. Hörmader, [8], Theorem 18.1.29). There are cases where A is not elliptic but the condition \(Au \in H^{\text {loc}}_{(s)}\) implies \(u\in H^{\text {loc}}_{(s+m-r)}\) for some \(r>0\). When \(Au \in H^{\text {loc}}_{(s)}\) implies that \(u \in H^{\text {loc}}_{(s+m-r)}\) for some \(r\ge 0\) we say that the operator A is hypoelliptic with loss of r derivatives. There are many results in the literature on this problem. For instance, we refer the reader to Parmegiani [12], Ferra and Petronilho [6], Chinni and Cordaro [5] and to the references in these papers.
Our main purpose in this paper is to study the above problem in the periodic ultradifferentiable frame, which includes the periodic Gevrey case, as P. D. Cordaro suggested to us.
We start by recalling that in the paper “On Global Analytic and Gevrey Hypoellipticity on the Torus and the Métivier Inequality”, see [5], the authors G. Chinni and P. D. Cordaro introduced a new theory about analytic pseudodifferential operators on the N-dimensional torus \(\mathbb {{\mathbb {T}}}^N\). One question analyzed by them is the following: assuming that P(x, D) is a linear partial differential operator defined on \(\mathbb {{\mathbb {T}}}^N\) with real-analytic coefficients, that P(x, D) is \(\epsilon \)-subelliptic for some \(\epsilon >0\) and that P(x, D) is globally analytic hypoelliptic on \(\mathbb {{\mathbb {T}}}^N\) they ask when is it true that P(x, D) remains globally analytic hypoelliptic when one adds to it an analytic pseudodifferential operator on \(\mathbb {{\mathbb {T}}}^N\) of order \(< \epsilon \).
In order to start working on the Cordaro’s question we first, in Sect. 2, describe the theory of the ultradifferentiable functions on the torus that we will need. We would like to point out that throughout this paper we work with the space of the periodic ultradifferentiable functions of Roumieu type and we do not distinguish if our weight sequence is quasianalytic or non-quasianalytic. In Sect. 3 we introduce a new class of \(\mathcal {M}\)-ultradifferentiable pseudodifferential operators defined on \(\mathbb {{\mathbb {T}}}^N\) (cf. Definition 3.3), where \(\mathcal {M}\) is a weight sequence (see Sect. 2), that generalizes the class of analytic pseudodifferential operators introduced by [4] and [5], in fact, for \({\mathcal {M}}=\{n!\}_{n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+}\) we obtain the class of [5]. Also, our class does not involves finite differences in the dual variable \(\xi \), as in Ruzhansky M. and Turunen V., see [14].
We would like to point out that in [4] and [5] the authors work in the distributions frame while we work in the \(\mathcal {M}\)-ultradistributions classes, and therefore we generalize [4] and [5] even in the analytic case. Due to the fact that we are working with \(\mathcal {M}\)-ultradistributions, in general, the statements and proofs are more sophisticated.
For example, in order to use the calculus of our class of ultradifferential pseudodifferential operators to study the perturbation problem the first key point is the appropriate caracterization of the \(\mathcal {M}\)-ultradistributions introduced in Sect. 4, (see the subspace \((\mathcal {DE})_{\left\{ \mathcal {M}, \delta , \sigma \right\} } ({\mathbb {T}}^N)\) of the \(\mathcal {M}\)-ultradistribution given in the Definition 4.1), where we have defined a norm and we characterize when a \(\mathcal {M}\)-ultradistribution belongs either to the space \((\mathcal {DE})_{\left\{ \mathcal {M}, \delta , \sigma \right\} } ({\mathbb {T}}^N)\) or belongs to the space \({\mathcal {E}}_{{\mathcal {M}}}\left( {\mathbb {T}}^N\right) \) (see Definition 2.2). This norm, in some sense, mixes the ultradifferentiable usual norm and the usual Sobolev norms (see Definition 4.1). The ultradifferentiable usual norm appears, for instance, when we take the Gevrey classes, i.e., we replace \(m_n\) by \(n^{s-1}\) in the expression of the supremum in Definition 4.1. We also would like to stress that in the Definition 4.1 we allow \(\delta <0\) in order to introduce norms in the space of ultradistributions.
The other key point is to use the norm introduced in the space \((\mathcal {DE})_{\left\{ \mathcal {M}, \delta , \sigma \right\} } ({\mathbb {T}}^N)\) to give our definition of a system of ultradifferentiable pseudodifferental operators to be globally \({\mathcal {M}}\)-hypoelliptic with loss of \(r\ge 0\) derivatives, (cf. Definition 5.1), given in Sect. 5. In order to compare our Definition 5.1 with the definition of a properly supported pseudodifferential operator to be hypoelliptic with loss of r derivatives (with \(0\le r <\infty \)), in the local smooth case, we refer, e.g., the reader to Parmegiani [12]. We now point out that in [4] and [5] some of the results are for \(\varepsilon -\)subelliptic operators with \(\varepsilon = m -r>0\) where m is the order of the operator and \(r\ge 0\) is the loss of derivatives while in our case we can have \(m-r<0\) as one can see in our application in Sect. 6.
The theory introduced in Sects. 4 and 5 and the important Proposition 5.5 allows us to prove the following result: “Assuming that \({\mathcal {A}}=\{a_j(x,D)\}_{j=1}^m\) is a system of \(\mathcal {M}\)-ultradifferentiable pseudodifferental operators of order \(\sigma \in \mathbb {{\mathbb {R}}}\), where \(\mathcal {M}\) is a weight sequence (cf. Definition 3.3) that is globally \({\mathcal {M}}\)-hypoelliptic on \(\mathbb {{\mathbb {T}}}^N\) with loss of \(r\ge 0\) derivatives (cf. Definition 5.1) and therefore is \({\mathcal {M}}\)-hypoelliptic on \(\mathbb {{\mathbb {T}}}^N\) (cf. Theorem 5.3), then the system \({\mathcal {C}}\dot{=}\{c_j(x,D)=a_j(x,D) + b_j(x,D)\}_{j=1}^m\) in \(\mathfrak {D}_{\mathfrak {p}_{\sigma }}^{\mathcal {M}}({\mathbb {T}}^N)\) is globally \({\mathcal {M}}\)-hypoelliptic, provided that \({\mathcal {B}}=\{b_j(x,D)\}_{j=1}^m\) is a system of pseudodifferential operators in \(\mathfrak {D}_{\mathfrak {p}_{\tau }}^{\mathcal {M}}({\mathbb {T}}^N)\), with \(\tau <\sigma -r.\)
In Sect. 6 we turn attention to our second question: it is well known that if P(x, D) is a linear partial differential operator that is hypoelliptic, then its transpose is locally solvable, see, e.g., Treves [15], Theorem 52.2. For the Gevrey version of this result, see Albanese, Corli and Rodino [1] and for a global version in the Gevrey classes, see Albanese and Zanghirati [2]. In this section, we shall prove an analogue of these results for our class of ultradifferentiable pseudodifferential operators. More precisely we prove that if \({\mathcal {A}}=\{a_j(x,D)\}_{j=1}^m\) is a system of \(\mathcal {M}\)-ultradifferentiable pseudodifferental operators of order \(\sigma \in \mathbb {{\mathbb {R}}}\) is globally \({\mathcal {M}}\)-hypoelliptic on \({\mathbb {T}}^N\) then \(\ker {\mathcal {A}}\) has finite dimension and \(\;\!^t {\mathcal {A}}\) is globally solvable in \(D'_{\mathcal {M}}\left( {\mathbb {T}}^N\right) \) (cf. Theorem 6.7), thus extending to the ultradifferentiable classes the well-known analogous result in the corresponding Gevrey classes (see [2]) and our proof is easier than the proof given by [2]. We would like to point out that the proof of this last result was inspired by [3].
Finally in the last section we present an application of our results which was considered in [5] in the analytic frame. More precisely we consider a linear partial differential operator that is of constant strength (cf. [7] Vol. II) given by \(a(x,D)=P_0(D)+\sum _{j=1}^ma_j(x)P_j(D),\) where \(a_{j}(x) \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\) and \(P_{0}(D), P_{1}(D) \ldots P_{m}(D)\) are linear partial differential operators with constant coefficients in \({\mathbb {T}}^N\) and we prove that if \(\beta _0\) is the order of operator \(P_0(D)\) then a(x, D) is globally \(\mathcal {M}\)-hypoelliptic on \({\mathbb {T}}^N\) with loss of \(\beta _{0} + M\) derivatives, where M is the constant that appears in the Greenfield-Wallach condition on \(P_0(D)\). As a consequence of our results we conclude that a(x, D) is globally \(\mathcal {M}\)-hypoelliptic on \(\mathbb {T}^N\), its kernel has finite dimension and its transpose \(\;\!^t {\mathcal {A}}\) is globally solvable in \(D'_{\mathcal {M}}(\mathbb {T}^N)\). Furthermore, these properties remain valid if we perturb the operator a(x, D) by any ultradifferentiable pseudodifferential operator of order \(\tau < -M\). (cf. Corollary 7.3).
2 Ultradifferentiable Functions on Torus
Throughout this paper \(\mathbb {T}^N\) is going to denote the N-dimensional torus. We say that a sequence of positive real numbers \({\mathcal {M}}=\{m_n\}_{n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+}\) is a weight sequence if it satisfies the following properties:
We recall that a weight sequence \({\mathcal {M}}=\{m_n\}_{n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+}\) is called quasianalytic if
If the sum if finite, then \({\mathcal {M}}=\{m_n\}_{n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+}\) is called a non-quasianalytic weight sequence.
Remark 2.1
Throughout this paper we do not distinguish if our weight sequence is quasianalytic or non-quasianalytic.
Definition 2.2
Let \(\mathcal {M}=\{m_n\}_{n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+}\) be a weight sequence. We say that a function \(f \in C^\infty \left( {\mathbb {T}}^N\right) \) is periodic ultradifferentiable of class \(\{{\mathcal {M}}\}\) if there exist constants \(C,h>0\) such that
The space of the periodic ultradifferentiable functions of class \(\{{\mathcal {M}}\}\) will be denoted by \({\mathcal {E}}_{{\mathcal {M}}}\left( {\mathbb {T}}^N\right) \). This class is also known as the space of the periodic ultradifferentiable functions of Roumieu type.
We would like to point out that for \(m_n=n!^{s-1}\) we have the spaces of periodic Gevrey functions, in particular for \(s=1\) we have the space of periodic analytic functions.
The reader should compare this definition with the one given in [10], where the author used the sequence \(M_n=m_n \cdot n!\) to define the space of ultradifferentiable functions (see also the next remark).
Remark 2.3
Note that (2.2) is equivalent to say that the sequence \(\left\{ \frac{m_n}{m_{n-1}}\right\} _{n\in {{\,\mathrm{{\mathbb {N}}}\,}}}\) is increasing. Also if the sequence \(\{m_n\}_{n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+}\) satisfies (2.2) then the sequence given by \(M_n=m_n n!\) also satisfies (2.2). Thus our condition (2.2) is stronger than the condition (M.1) of Komatsu [ [10], p.26], which says that the sequence \(\{M_n\}_{n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+}\) satisfies (2.2). It is not difficult to see that our condition (2.3) is equivalent to condition (M.2) of [10] given by
We also point out that several basic properties that we shall need are also valid if we replace (2.3) by the weaker condition
which is equivalent to say that the space \({\mathcal {E}}_{{\mathcal {M}}}\left( {\mathbb {T}}^N\right) \) is closed under differentiation and it is also equivalent to condition \((M2)'\) of [10].
For a weight sequence \(\mathcal {M}=\{m_n\}_{n\in {{\,\mathrm{{\mathbb {N}}}\,}}}\) we define the associated weight function \(\omega _{\mathcal {M}} : [0,\infty ]\longrightarrow [0,\infty ]\) by
We now list some basic properties of weight sequences that we will need.
Proposition 2.4
If \(\mathcal {M}=\{m_n\}_{n\in {{\,\mathrm{{\mathbb {N}}}\,}}}\) is a weight sequence then we have the following properties:
-
(i)
\(m_n \ge 1\) for all \(n \in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\);
-
(ii)
the sequence \(\{m_n^{1/n}\}_{n\in {{\,\mathrm{{\mathbb {N}}}\,}}}\) is increasing;
-
(iii)
\(m_j m_k \le m_{j+k}\) for all \(j,k \in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\);
-
(iv)
for each \(k \in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\) we can find a constant \(A_k\ge 1\) such that
$$\begin{aligned} m_{n+k} (n+k)! \le A_k^{n+1} m_n n!, \ \forall \ n \in {{\,\mathrm{{\mathbb {Z}}}\,}}_+; \end{aligned}$$(2.7) -
(v)
for each \(t \in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\) the supremum in (2.6) is assumed in \(n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\) that depends on t;
-
(vi)
for each \(k>0\) we have
$$\begin{aligned} \omega _{\mathcal {M}}(k\rho )-\omega _{\mathcal {M}}(\rho ) \ge \frac{\ln (k) \ln (\rho )}{\ln (H_0)}, \ \ \forall \ \rho>0,\,\,\text {for some}\,\,H_0>1. \end{aligned}$$(2.8)
For more results on weight sequences we refer the reader to Komatsu [10], Pilipović [13], Kirilov and Victor [9] and to the references in these papers.
Remark 2.5
It is clear from item (i) of Proposition 2.4 that \({\mathcal {E}}_{{\mathcal {M}}}\left( {\mathbb {T}}^N\right) \) contains the space of all periodic analytic functions \(C^\omega \left( {\mathbb {T}}^N\right) \) since \(|\alpha |! \le m_{|\alpha |} |\alpha |!\) for all \(\alpha \in {{\,\mathrm{{\mathbb {Z}}}\,}}_+^N.\)
Remark 2.6
Let \({\mathcal {M}}=\{m_n\}_{n\in {\mathbb {Z}}_+}\) and \({\mathcal {L}} = \{\ell _n\}_{n\in {\mathbb {Z}}_+}\) be two weight sequences. It can be proved that \({\mathcal {E}}_{{\mathcal {M}}}\left( {\mathbb {T}}^N\right) \subset {\mathcal {E}}_{{\mathcal {L}}}\left( {\mathbb {T}}^N\right) \) if and only if
Definition 2.7
For \(h>0\) we set
where \(\left\| f \right\| _{{\mathcal {M}},h} := \sup \left\{ \frac{|D^{\alpha } f(x)|}{h^{|\alpha |} m_{|\alpha |} |\alpha |!} : x \in {\mathbb {T}}^N,\, \alpha \in {{\,\mathrm{{\mathbb {Z}}}\,}}_+^{N} \right\} .\)
It can be proved that \({\mathcal {E}}_{{\mathcal {M}}, h}({\mathbb {T}}^N)\) is a Banach space. Furthermore, if \(h<h'\) , \({\mathcal {E}}_{{\mathcal {M}}, h}({\mathbb {T}}^N) \subset {\mathcal {E}}_{{\mathcal {M}}, h'}({\mathbb {T}}^N)\) with compact inclusion. We equip \({\mathcal {E}}_{{\mathcal {M}}}({\mathbb {T}}^N) = \displaystyle \bigcup _{j=1}^\infty {\mathcal {E}}_{{\mathcal {M}}, h_j}({\mathbb {T}}^N)\) with the inductive limit of any strictly increasing \(\{h_j\}_{j \in {{\,\mathrm{{\mathbb {N}}}\,}}}\) that tends to infinity. In particular, \({\mathcal {E}}_{{\mathcal {M}}}({\mathbb {T}}^N)\) is a DFS-space (i.e. locally convex spaces whose topology can be described as injective limits of sequences of Banach spaces with compact inclusion maps).
Now we define the ultradistributions with respect to the sequence \(\{m_n\}_{n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+}\): the space \(D'_{\mathcal {M}}\left( {\mathbb {T}}^N\right) \) is the topological dual of \({\mathcal {E}}_{{\mathcal {M}}}({\mathbb {T}}^N)\), that is \(u \in D'_{\mathcal {M}}\left( {\mathbb {T}}^N\right) \) when \(u:{\mathcal {E}}_{{\mathcal {M}}}({\mathbb {T}}^N)\longrightarrow {{\,\mathrm{{\mathbb {C}}}\,}}\) is a continuous linear functional.
Finally we end this section with a characterization of the space of ultradifferentiable functions and ultradistributions in terms of the Fourier transform. For \(u \in D'_{\mathcal {M}}\left( {\mathbb {T}}^N\right) \), we set
which is well defined by Remark 2.5.
Theorem 2.8
Let \(\mathcal {M}=\{m_n\}_{n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+}\) be a weight sequence.
-
(a)
A function \(\varphi \in C^\infty \left( {\mathbb {T}}^N\right) \) belongs to \({\mathcal {E}}_{\mathcal {M}}\left( {\mathbb {T}}^N\right) \) if and only if there exists \(C,h>0\) such that
$$\begin{aligned} \left| {\hat{\varphi }}(\xi )\right| \le C \displaystyle \inf _{n \in {\mathbb {Z}}_{+}}\left( \frac{h^n m_n n!}{(1+|\xi |)^n} \right) , \ \ \forall \ \xi \in {{\,\mathrm{{\mathbb {Z}}}\,}}^N. \end{aligned}$$(2.10)Moreover, if \(\{C_\xi \}_{\xi \in {{\,\mathrm{{\mathbb {Z}}}\,}}^N}\) is a sequence such that (2.10) holds true with \(C_\xi \) in place of \({\hat{\varphi }}(\xi )\), then there exits an unique function \(\varphi \in {\mathcal {E}}_{\mathcal {M}}\left( {\mathbb {T}}^N\right) \) such that \({\hat{\varphi }}(\xi )=C_\xi \) for all \(\xi \in {{\,\mathrm{{\mathbb {Z}}}\,}}^N\).
-
(b)
If \(u \in D'_{\mathcal {M}}\left( {\mathbb {T}}^N\right) \), for each \(\epsilon >0\) we can find \(C_\epsilon >0\) such that
$$\begin{aligned} \left| {\hat{u}}(\xi )\right| \le C_\epsilon \sup _{n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+} \left( \frac{\epsilon ^n (1+|\xi |)^n}{m_n n!}\right) , \ \forall \ \xi \in {{\,\mathrm{{\mathbb {Z}}}\,}}^N. \end{aligned}$$(2.11)Moreover, if \(\{C_\xi \}_{\xi \in {{\,\mathrm{{\mathbb {Z}}}\,}}^N}\) is a sequence such that (2.11) holds true with \(C_\xi \) in place of \(\hat{u}(\xi )\), then there exits an unique ultradistribution \(u \in D'_{\mathcal {M}}\left( {\mathbb {T}}^N\right) \) such that \({\hat{u}}(\xi )=C_\xi \) for all \(\xi \in {{\,\mathrm{{\mathbb {Z}}}\,}}^N\) and we have that
$$\begin{aligned} u(x) = \sum _{\xi \in {{\,\mathrm{{\mathbb {Z}}}\,}}^N} {\hat{u}}(\xi ) e^{i\left\langle x,\xi \right\rangle } \end{aligned}$$where the limit is taken in the weak topology of \(D'_{\mathcal {M}}\left( {\mathbb {T}}^N\right) \).
3 Pseudodifferential Operators on Torus
In this section, we introduce a new class of ultradifferentiable pseudodifferential operators defined on the N-dimensional torus \({\mathbb {T}}^N\) which map \({\mathcal {E}}_{{\mathcal {M}}}({\mathbb {T}}^N)\) into itself and whose calculus allows us to study the corresponding perturbation problem in a far more general context. Our theory was inspired by the papers [4] and [5] where the authors introduced an analytic class of pseudodifferential operators on torus and by the work [6] where the smooth case was considered.
We start by recalling that the discrete symbol of a continuous and linear operator \(A:C^\infty ({\mathbb {T}}^N) \longrightarrow C^\infty ({\mathbb {T}}^N)\) is the function \(a:{\mathbb {T}}^N\times {\mathbb {Z}}^N\longrightarrow {\mathbb {C}}\) defined by \(a(x,\eta ) = e^{-i\left\langle x,\eta \right\rangle } A\left( e^{i\left\langle x,\eta \right\rangle }\right) \) and we shall use the notation \(A=a(x,D)\). Notice that if \(\varphi \in C^\infty ({\mathbb {T}}^N)\) then by linearity and continuity we have
We first note that if \(\{\widehat{a}(\xi ,\eta )\}_{\xi \in \mathbb {Z}^N}\) denotes the sequence of the Fourier coefficients of the function \(x \mapsto a(x,\eta )\) then it is easy to check that
and, if \(k \in D'\left( {\mathbb {T}}^N\times {\mathbb {T}}^N\right) \) denotes the Schwartz distribution kernel of a(x, D) then we have the following important relation
We now will show a relation between the Schwartz distribution kernel k(x, y) and the discrete symbol of the operator a(x, D). More precisely we have the following:
Proposition 3.1
Let \(a(x, D): C^{\infty }({\mathbb {T}}^N) \rightarrow C^{\infty }({\mathbb {T}}^N)\) be a continuous, linear operator and consider \(k \in D'({\mathbb {T}}^N \times {\mathbb {T}}^N)\) its distribution kernel. If \(\mathcal {M}=\{m_n\}_{n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+}\) is a weight sequence and \(\sigma \in {\mathbb {R}}\), then the following statements are equivalent:
-
(i)
There exist positive constants \(C_1\) and \(h_1\) such that
$$\begin{aligned} |D_{x}^{\alpha } a(x, \eta )| \le C_1 h_1^{|\alpha |} m_{|\alpha |} |\alpha |! (1+|\eta |)^{\sigma },\ \forall \,\, x \in {\mathbb {T}}^{N}, \ \forall \,\, \eta \in {\mathbb {Z}}^N, \ \forall \,\, \alpha \in {{\,\mathrm{{\mathbb {Z}}}\,}}_+^{N}.\nonumber \\ \end{aligned}$$(3.4) -
(ii)
There exist positive constants \(C_2, h_2 >0\) such that
$$\begin{aligned} |{\widehat{a}}(\xi , \eta )| \le \frac{C_{2} h_{2}^{k} m_{k} k! (1+ |\eta |)^{\sigma }}{ (1+|\xi |)^{k} }, \ \forall \,\, k \in {{\,\mathrm{{\mathbb {Z}}}\,}}_+, \ \forall \,\, (\xi , \eta ) \in {\mathbb {Z}}^{2N}. \end{aligned}$$(3.5) -
(iii)
There exist positive constants \(C_3, h_3 >0\) such that
$$\begin{aligned} |{\widehat{k}}(\xi , \eta )| \le \frac{C_{3} h_{3}^{k} m_{k} k! (1+ |\eta |)^{\sigma }}{(1+|\xi + \eta |)^{k}}, \ \ \forall \,\, k \in {\mathbb {Z}}_+, \ \forall \,\, (\xi , \eta ) \in {\mathbb {Z}}^{2N}. \end{aligned}$$(3.6)
Proof
We begin the proof by showing that (3.4) implies (3.5). If \(\xi =0\) then
For \(\xi \ne 0\) we choose \(p \in \{1,\ldots ,N\}\) such that \(\displaystyle \max _{1\le q \le N} |\xi _q| = |\xi _p|\). If \(k \in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\) and \(e_p\) is the p-th vector of the canonical basis of \({{\,\mathrm{{\mathbb {R}}}\,}}^N\), then \(\beta = k e_p \in {{\,\mathrm{{\mathbb {Z}}}\,}}_+^N\) and \(|\beta |=k\). We also have
The last inequality, (3.4) and the fact that we have \(\xi \ne 0\) give us that
Hence, if we take \(C_2=C_1\) and \(h_2=\max \{1,2\sqrt{N} h_1\}\), the last inequality and (3.7) give us (3.5).
Reciprocally, let us prove that (3.5) implies (3.4). We have
By using (3.5) it follows from the last inequality that
By taking \(C_{1} = A_{2N} C_{2} h_{2}^{2N}\displaystyle \sum _{\xi \in {\mathbb {Z}}^{N}} \displaystyle \frac{1}{(1+|\xi |)^{2N}}\) and \(h_{1} = h_2 A_{2N}\), we obtain (3.4).
The equivalence between (3.4) and (3.6) follows immediately from (3.3). \(\square \)
We will now proceed in the opposite direction to that we have done so far in this section and discuss classes of symbols that correspond to an operator as that one defined in (3.1).
We start by the following definition.
Definition 3.2
Let \(\mathcal {M}\) be a weight sequence and \(\sigma \in {\mathbb {R}}\). We consider \(S^\sigma _\mathcal {M}({\mathbb {T}}^N\times {{\,\mathrm{{\mathbb {Z}}}\,}}^N)\) as the space of all functions \(a(x,\eta ) \in C^\infty ({\mathbb {T}}^N, {\mathbb {Z}}^N)\), i.e., when function \(a(\cdot , \xi )\) is smooth on \({\mathbb {T}}^N\) for all \(\xi \in {\mathbb {Z}}^N\), satisfying (3.4).
Definition 3.3
If \(a(x,\xi ) \in S^\sigma _\mathcal {M}({\mathbb {T}}^N\times {{\,\mathrm{{\mathbb {Z}}}\,}}^N)\) we define, in \(C^\infty ({\mathbb {T}}^N)\), the operator
We say that a(x, D) is a \(\mathcal {M}\)-ultradifferentiable pseudodifferential operator of order \(\sigma \), i.e, \(a(x,D)\in {\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\).
By using the fact that \(\varphi \in C^\infty ({\mathbb {T}}^N)\) and estimate (3.4) one can easy prove the following
Proposition 3.4
Let \(\varphi \in C^\infty ({\mathbb {T}}^N)\). Then \(a(x,D)\varphi \) in (3.9) is well defined and \(a(x,D)\varphi \in C^\infty ({\mathbb {T}}^N)\). Moreover, operator \(a(x,D) : C^\infty ({\mathbb {T}}^N) \rightarrow C^\infty ({\mathbb {T}}^N)\) is continuous.
In the rest of this section we will present the main properties of this class of ultradifferential pseudodifferential operators and we will only present the proof of some of them. The first property is given by
Proposition 3.5
Let \(a(x,D) \in \mathfrak {D}^{\mathcal {M}}_{\mathfrak {p}_{\sigma }} ({\mathbb {T}}^N)\) and \(b(x, D) \in \mathfrak {D}^{\mathcal {M}}_{\mathfrak {p}_{\sigma '}}({\mathbb {T}}^N)\). The composition \(a(x,D) \circ b(x,D)\) is an element of \( \mathfrak {D}^{\mathcal {M}}_{\mathfrak {p}_{\sigma + \sigma '}} ({\mathbb {T}}^N)\) whose discrete symbol is given by
The next auxiliary result will be used in the proof that any pseudodifferential operator in \({\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\) can be extended as a continuous and linear operator in \(D'_\mathcal {M}\left( {\mathbb {T}}^N\right) \) and its proof will be omitted.
Lemma 3.6
-
(a)
There exists a positive constant \(\lambda \) depending only on the dimension N of \({\mathbb {T}}^N\) such that for all \(h>0\) and all \(f \in \mathcal {E}_{\mathcal {M}, h}({\mathbb {T}}^N)\) we have
$$\begin{aligned} |\widehat{f}(\xi )| \le \frac{\left\| f \right\| _{\mathcal {M}, h} (\lambda h)^{k} m_{k} k!}{(1+|\xi |)^{k} }, \ \ \forall \, k \in {{\,\mathrm{{\mathbb {Z}}}\,}}_+, \ \forall \, \xi \in {\mathbb {Z}}^{N}. \end{aligned}$$(3.11) -
(b)
For each \(C,h>0\) we can find positive constants \(C'\) and \(h'\) depending only on h, \({\mathcal {M}}\) and the dimension N of \({\mathbb {T}}^N\) such that if \(g \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\) and g satisfies
$$\begin{aligned} |\widehat{g}(\xi )| \le \frac{C h^{k} m_{k} k!}{(1+|\xi |)^{k} }, \ \ \forall \, k \in {{\,\mathrm{{\mathbb {Z}}}\,}}_+, \ \forall \, \xi \in {\mathbb {Z}}^{N}, \end{aligned}$$(3.12)then \(\left\| g \right\| _{\mathcal {M}, h'} \le C'C\).
Theorem 3.7
If \( a(x,D) \in {\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\), then a(x, D) defines a continuous, linear operator on \(\mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\).
Proof
In order to prove this result it suffices to show that given \(h>0\) there exists \(h'>0\) and \(C>0\) such that \(\left\| a(x,D) f \right\| _{\mathcal {M}, h'} \le C\left\| f \right\| _{\mathcal {M}, h}\), for all \(f\in \mathcal {E}_{\mathcal {M}, h}({\mathbb {T}}^N)\).
Let \(h>0\) be given. For \(f \in \mathcal {E}_{\mathcal {M}, h}({\mathbb {T}}^N)\) we shall analyze the term \(\widehat{(a(x,D)f)}(\xi )\) given in (3.2). Take \(p\in {\mathbb {Z}}_+\) such that \(\sigma \le p\). By Lemma 3.6 item (a) and (3.5) there exist \(\lambda , C_2, h_2>0\) such that, for \(j,k\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\), we have
Since \(\sigma < p\) and taking advantage from (2.7), Proposition 2.4, we have
where \(C_3 = C_2 A_{p} (\lambda h)^p\).
Now we will analyze the term \((I)\, \dot{=} \displaystyle \sum _{\eta \in {{\,\mathrm{{\mathbb {Z}}}\,}}^N} \frac{h_2^{j} m_{j} j!}{(1+|\xi -\eta |)^{j}} \frac{(A_p \lambda h)^k m_{k} k!}{(1+|\eta |)^{k}}.\)
Note that
If \(|\eta | \le \frac{|\xi |}{2}\) then we have \(|\xi | \le |\xi -\eta |+\frac{|\xi |}{2}\) which in turns implies that \((1+|\xi |)\le 2(1+|\xi -\eta |)\). Thus, by taking \(k=2N\) we can infer that
If \(|\eta | > \frac{|\xi |}{2}\) then we have \((1+|\xi |) \le 2(1+|\eta |)\) and by taking \(j=2N\), we obtain
since
If we denote \(C_4 = \Big (\displaystyle \sum _{\eta \in {{\,\mathrm{{\mathbb {Z}}}\,}}^N}\frac{1}{(1+|\eta |)^{2N}}\Big ) (A_p \lambda h)^{2N} m_{2N} (2N)!\), \(C_5=\Big (\displaystyle \sum _{\eta \in {{\,\mathrm{{\mathbb {Z}}}\,}}^N}\frac{1}{(1+|\eta |)^{2N}} \Big )h_{2}^{2N} m_{2N} (2N)!\), and \(h_1=\max \{2 h_{2}, 2 A_p\lambda h\}\), it follows from (3.14) and (3.15) that
From the last inequality and (3.13) we conclude that
Hence, if we set \(C=C_3(C_4+C_5)\) we conclude that for all \(f \in \mathcal {E}_{\mathcal {M},h}({\mathbb {T}}^N)\) we have
If \(f\not \equiv 0\), then we use item (b) of Lemma 3.6 with \(h=h_1\) in order to obtain \(C',h'>0\) (also take \(g=a(x,D){\tilde{f}}\), where \({\tilde{f}} = f/ \left\| f \right\| _{\mathcal {M},h}\)) such that \(\left\| a(x,D) f \right\| _{\mathcal {M}, h'} \le C' C \left\| f \right\| _{\mathcal {M}, h}.\) \(\square \)
We now recall that the transpose operator, \({}^ta(x, D)\), of the operator \(a(x,D) \in {\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\) is given by
and we will prove the following result.
Proposition 3.8
If \(a(x,D) \in {\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\) then its transpose \({}^ta(x,D)\) given by (3.16) belongs to the space \({\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\).
Proof
We denote by \(\left\langle \cdot ,\cdot \right\rangle _0\) the inner product in \(L^2({\mathbb {T}}^N)\). The formal adjoint in \(L^2\) of a(x, D) is the operator \(a^*(x,D)\) such that
Note that if \(f,g \in C^\infty ({\mathbb {T}}^N)\) then \(\left\langle f,g\right\rangle _0 = \int _{{\mathbb {T}}^N} f(x) \overline{g(x)} dx = \left\langle f, {{\overline{g}}}\right\rangle .\)
Since \(\langle f,\;\!^ta(x,D) g \rangle = \left\langle a(x,D) f, g\right\rangle = \left\langle a(x,D) f, {\overline{g}}\right\rangle _0 = \left\langle f, a^*(x,D) ({\overline{g}})\right\rangle _0 = \left\langle f, \overline{a^*(x,D) ({\overline{g}})}\right\rangle \) for all \(f,g \in C^\infty ({\mathbb {T}}^N)\), it follows that \(\;\!^ta(x,D) g = \overline{a^*(x,D) {\overline{g}}}\).
If we take \(g_\xi (x) = e^{-i\left\langle x,\xi \right\rangle }\) in the last equality we obtain
Thus
Since \(\widehat{a^*}(\xi ,\eta ) = \overline{{\hat{k}}(\eta ,-(\xi +\eta ))}\), (see [5]), it follows from (3.3) that
Since \(a(x,D) \in {\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\) then it follows from Proposition 3.1 that we can find \(C,h>0\) such that for all \(j\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\) we have
Now we take \(p\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\) such that \(|\sigma | \le p\). By using (2.7) and the following inequality
we conclude that, for all \(n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\) we have
which concludes the proof by Proposition 3.1. \(\square \)
Remark 3.9
It follows from the last proposition and from the fact that \({\mathcal {E}}_{{\mathcal {M}}}({\mathbb {T}}^N)\) is a Montel space that we can extend \(a(x,D) \in {\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\) to a continuous and linear operator on \(D'_\mathcal {M}\left( {\mathbb {T}}^N\right) \) by setting
and the same can be done with \({}^ta(x,D)\) instead of a(x, D).
4 Properties of the Space \((\mathcal {DE})_{\left\{ \mathcal {M}, \delta , \sigma \right\} } ({\mathbb {T}}^N) \subset D_{\mathcal {M}}'({\mathbb {T}}^N)\)
In this section, we shall introduce a key definition that will allow us prove our main result of the next section. We start by setting the following
Definition 4.1
For \(\delta , \sigma \in {\mathbb {R}},\) with \(\delta \in {\mathbb {R}}\), we denote by \((\mathcal {DE})_{\left\{ \mathcal {M}, \delta , \sigma \right\} } ({\mathbb {T}}^N)\) the space of all ultradistributions \(u \in D_{\mathcal {M}}'({\mathbb {T}}^N)\) such that:
where \(r(\delta ) = sgn(\delta ) ={\left\{ \begin{array}{ll} 1,\quad \text{ if } \delta \ge 0\\ -1,\quad \text{ if }\, \delta <0. \end{array}\right. }\)
Remark 4.2
-
(i)
For \(\delta =0\), in the definition 4.1, we recover the Sobolev spaces that are used to study the \(C^\infty \) global hypoellipticity, which we are not interested here, since it has been done in [6]. Thus, here we will work with the case \(\delta \ne 0\).
-
(ii)
For \(m_n=n!^{s-1}\), with \(s\ge 1\), we are in the Gevrey spaces and we can prove that for \( \delta >0\) the condition \(\left\| u \right\| _{\left\{ \mathcal {M}, \delta , \sigma \right\} } < +\infty \) is equivalent to existence of \(\delta _1 > 0\) such that \(\left\| u \right\| _{\left\{ \delta _1, s, \sigma \right\} } = \sum _{\xi \in {\mathbb {Z}}^N} e^{2 \delta _1 |\xi |^{1/s}}(1+|\xi |)^{2 \sigma }|\widehat{u}(\xi )|^2 <+\infty .\)
We shall need the following basic property whose proof will be omitted.
Proposition 4.3
Let \(\delta _{1}, \delta _{2}, \sigma _{1}, \sigma _{2}\) be real numbers with \(\delta _{1} \le \delta _{2}\) and \(\sigma _{1} \le \sigma _{2}\), with \(\delta _1, \delta _2 \ne 0\). Then
Moreover, if \(u \in (\mathcal {DE})_{\left\{ \mathcal {M}, \delta _{2}, \sigma _{2} \right\} } ({\mathbb {T}}^N)\), then
We now will prove the following important properties.
Proposition 4.4
Let \(u \in D_{\mathcal {M}}'({\mathbb {T}}^N)\). Then the following statements hold true:
-
1.
\(u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^{N})\) if and only if there exist \(\delta > 0\) and \(\sigma \in {\mathbb {R}}\) such that \(u \in (\mathcal {DE})_{\left\{ \mathcal {M}, \delta , \sigma \right\} }({\mathbb {T}}^N)\).
-
2.
\(u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^{N})\) if and only if there exists \(\delta > 0\) such that \(u \in (\mathcal {DE})_{\left\{ \mathcal {M}, \delta , \sigma \right\} }({\mathbb {T}}^N)\) for all \(\sigma \in {\mathbb {R}}\).
-
3.
\(u \in (\mathcal {DE})_{\left\{ \mathcal {M}, \delta , \sigma \right\} }({\mathbb {T}}^N)\), for all \(\delta <0\) and \(\sigma \in {\mathbb {R}}\).
Proof
First note that if we prove that the condition in item 1 is sufficient to have \(u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^{N})\), then the condition in item 2 will also be sufficient to have \(u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^{N})\). On the other hand, if we prove that the condition in item 2 is necessary to have \(u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^{N})\), then the condition in item 1 will also be necessary to have \(u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^{N})\).
Sufficiency of condition 1: By hypothesis there exist \(C, \delta >0\) and \(\sigma \in {\mathbb {R}}\) such that
Let \(p\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\) such that \(-\sigma <p\) and, for \(k\in {\mathbb {Z}}_+\), we take \(n=k+p\). It follows from the last equality that
Now by using (2.7) we obtain
By setting \(C_1=\frac{C A_p}{\delta ^p}\) and \(h=\frac{A_p}{\delta }\) it follows from the last inequality that \(u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\) by Theorem 2.8, item (a).
Necessity of condition 2: Let \(u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\), then again it follows from Theorem 2.8, item (a), that there are positive constants C and h such that
For \(\xi \in {\mathbb {Z}}^{N}\) and \(\sigma \in {\mathbb {R}}\), we denote
where \(\delta >0\) will be chosen later.
By choosing \(p \in {\mathbb {Z}}_+\) such that \(\sigma \le p\) we can estimate (4.2) as
It now follows from the last inequality and (4.1) that
By using (2.3) and (4.1), it follows from the last inequality that
If \(B\dot{=}(C^{2} h^{2(p+N)} m_{2(p+N)} (2(p+N))!)\), then B depends only on u, \(\mathcal {M}\), \(\sigma \) and N. By taking \(0<\delta \le \displaystyle \frac{1}{2 h H},\) we obtain
from where we conclude that
It proves that \(u \in (\mathcal {DE})_{\left\{ \mathcal {M}, \delta , \sigma \right\} }({\mathbb {T}}^N)\) for every \(\sigma \in {\mathbb {R}}\) and \(0<\delta \le \displaystyle \frac{1}{2 h H}\).
We proceed then to item 3: let \(u \in D_{\mathcal {M}}'({\mathbb {T}}^N)\), \(\delta < 0\) and \( \sigma \in {\mathbb {R}}\). Our goal is to prove that \(u \in (\mathcal {DE})_{\left\{ \mathcal {M}, \delta , \sigma \right\} }({\mathbb {T}}^N)\). From Theorem 2.8, item (b), for each \(\varepsilon > 0\) there exists a positive constant \(C_{\varepsilon }\) such that
Fixed \(\xi \in {\mathbb {Z}}^N\) and recalling that for \(\delta <0\) we have \(r(\delta )=-1\) we write
Again we take \(p\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\) satisfying \(\sigma \le p\). By using (4.4) and Proposition 2.4, item (v), we have for some \(n_0\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\) depending on \(\varepsilon \) and \(\xi \) that
Since \(\left[ \displaystyle \sup _{n \in {\mathbb {Z}}_+} \displaystyle \frac{|\delta |^{n} (1+|\xi |)^{n}}{m_{n} n!} \right] ^{-2}\le \left[ \displaystyle \frac{|\delta |^{n} (1+|\xi |)^{n}}{m_{n} n!} \right] ^{-2}\) for any \(n\in {\mathbb {Z}}_+\) then it follows from the last inequality that
Now we use (2.7) and we obtain
Now we choose \(\varepsilon = \displaystyle \frac{|\delta |}{A_{p+N}}\), which depends only on \(\sigma , \delta , N\) and \({\mathcal {M}}\). Then, for a new positive constant \(C_1=C_{1} (\sigma , \delta , N, {\mathcal {M}})\) we have
From (4.5) and (4.6) it follows that
from where we conclude that \(u \in (\mathcal {DE})_{\left\{ \mathcal {M}, \delta , \sigma \right\} }({\mathbb {T}}^N)\).
Since \(\delta <0\) and \(\sigma \) are arbitrary, the proof is complete. \(\square \)
Another important property that we need is given by the following:
Lemma 4.5
If \(f \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^{N})\), then we can find positive constants \(\delta _{0}, B, \rho \) such that
Proof
For \(\xi \in {\mathbb {Z}}^{N}\), \(\delta _0 >0\) and \(k,n \in {\mathbb {Z}}_+\), we write
where \(\delta _0\) will be chosen later.
Then
Applying (2.3), the fact that \(f \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\), twice, and (2.7) we conclude that there are positive constants C, H and h satisfying
Now we consider \(\delta _{0} = \displaystyle \frac{1}{2 h H}\) in the definition of (I) and we obtain
It follows from (4.7) and (4.8) that
Hence, if we set \(B \dot{=} \displaystyle \sum _{\xi \in {\mathbb {Z}}^N} \displaystyle \frac{1}{(1 + |\xi |)^{2N}}, B_1 \dot{=} (C^2 h^{2N} A_{2N} B)^{\frac{1}{2}}\) and \(\rho = h 2H A_{2N}\) we conclude that
and now we use the Proposition 4.3 in order to obtain
which concludes the proof. \(\square \)
5 Systems of Ultradifferentiable Pseudodifferential Operators and Global Hypoellipticity with Loss of Derivatives
From now on, if \(a_1(x,D), \ldots , a_m(x,D)\) belong to \({\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\), then we will denote by \({\mathcal {A}}\) the system associated to these operators and we say that \({\mathcal {A}}=\{a_j(x,D)\}_{j=1}^m\) is a system of pseudodifferental operators in \({\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\).
We now state our key definition.
Definition 5.1
Let \({\mathcal {A}}=\{a_j(x,D)\}_{j=1}^m\) be a system of pseudodifferental operators in \({\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\). We say that \({\mathcal {A}}\) is globally \({\mathcal {M}}-\)hypoelliptic with loss of \(r\ge 0\) derivatives if there exist constants \(B>0\), \(C > 0\), \(\delta _{0} <0\) and \(k_{0} \in {\mathbb {R}}\) such that
for any \(k \in {\mathbb {N}}_{0}, \ u \in D_{\mathcal {M}}'({\mathbb {T}}^N)\) and \( \delta _{0}< \delta < 0. \)
We also have a standard notion of global ultradifferentiable hypoellipticity for systems of pseudodifferential operators given by
Definition 5.2
Let \({\mathcal {A}}=\{a_j(x,D)\}_{j=1}^m\) be a system of pseudodifferental operators in \({\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\). We say that \({\mathcal {A}}\) is globally \({\mathcal {M}}-\)hypoelliptic on \({\mathbb {T}}^N\) if the conditions \(u \in D_{\mathcal {M}}'({\mathbb {T}}^N)\) and \(a_j(x,D)u \in \mathcal {E}_\mathcal {M}\left( {\mathbb {T}}^N\right) \) for all \(j=1,\ldots ,m\) imply \(u \in \mathcal {E}_\mathcal {M}\left( {\mathbb {T}}^N\right) \).
As in the \(C^\infty \) frame (cf. [12], p.229 for the local case and cf. [6] for the global case) we will prove the correspondent result for the ultradifferentiable version.
Theorem 5.3
If \({\mathcal {A}}=\{a_j(x,D)\}_{j=1}^m\) is a system of pseudodifferental operators in \({\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\) that is globally \({\mathcal {M}}-\)hypoelliptic with loss of \(r\ge 0\) derivatives, then \({\mathcal {A}}\) is globally \({\mathcal {M}}-\)hypoelliptic on \({\mathbb {T}}^N\).
Proof
Let us fix \(u \in D_{\mathcal {M}}'({\mathbb {T}}^N)\) such that \(a_j(x,D)u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\) for all \(j=1,\ldots ,m\). We will prove that \(u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\). It follows from our hypothesis and Lemma 4.5 that one can obtain \(\delta _0 <0\), positive constants \(C, B, B_1, \rho \) and \(k_0\in {\mathbb {R}}\) such that
for all \(k\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\) and \(\delta _0<\delta <0\). Since \(u \in D_{\mathcal {M}}'({\mathbb {T}}^N)\) it follows from Proposition 4.4, item 3, that \(\left\| u \right\| _{\left\{ \mathcal {M}, \delta , k_{0} \right\} }< \infty .\)
Thus, if we set \(C_{1} = \max \left\{ C, C B_1 \right\} \) and \(h_{1} = \max \left\{ \rho , B \right\} \) then we have
For \(j \in {\mathbb {Z}}_+, \ \xi \in {\mathbb {Z}}^N\) and \(\delta _{0}< \delta < 0\), we denote
If \(q \in {\mathbb {Z}}_+\) satisfies \(r - \sigma \le q\) then we can use (5.1) to obtain
From (2.7),
Since the right hand side of the last inequality does not depends on \(\xi \), if we set \(C_2=C_{1} h_{1}^{q} A_q\) and \(h_2 = h_1 A_q\) and by taking the square root in the last inequality, we conclude that
and therefore
for every \(\xi \in {\mathbb {Z}}^{N}\) and \( \delta _{0}< \delta < 0\). By taking the sum in j in the last inequality, we obtain
which ensures that, for every \(\xi \in {\mathbb {Z}}^{N}, \ j \in Z_+\) and \( \delta _{0}< \delta < 0\), we have
It follows from Proposition 2.4 (item (v)) that we can find \(n_0 \in {\mathbb {Z}}_+\) depending on \(\xi \) and \(\delta \) such that
Now we use (5.3) and conclude that, for all \(k \in {\mathbb {Z}}_+\) and \(\delta _{0}< \delta < 0\) we have
in turns implies that
It follows from (2.3) and the last inequality that
Let us fix \(\delta \) such that \(\delta _0< \delta < 0\) and \(-\delta =|\delta | \le \frac{1}{4 H h_{2}}\). Then \((4h_{2} H |\delta |)^{n_{0}} \le 1\) for any \(n_{0}\). Thus we can rewrite last inequality as
in turns implies that
Now by summing in \(\xi \in {\mathbb {Z}}^N\) we conclude that
It follows from Proposition 4.4 item 1. that \(u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^{N})\) and proof is now complete. \(\square \)
For the proof of our main result of this section we need the following auxiliary result whose proof is easily seen and it will be omitted.
Lemma 5.4
For any \(\lambda >0, \ \delta <0\), \(\rho< \sigma < \tau \) and \(u \in D_{\mathcal {M}}'({\mathbb {T}}^N)\) we have
The next result is a version of Proposition 1.1 of [CC] with our norms in the ultradifferentiable case and its proof presents several thecnical difficults. So we are going to present a detailed proof of this result here.
Proposition 5.5
Let \(a(x, D) \in \mathfrak {D}_{\mathfrak {p}_{\sigma }}^{\mathcal {M}}({\mathbb {T}}^N)\). Then there exist positive constants C and h such that for every \(\varepsilon > 0\), we can find \(\delta _{\varepsilon } < 0\) in such way that, for every \(u \in D_{\mathcal {M}}'({\mathbb {T}}^N)\), \(k \in {\mathbb {Z}}_+\) and \(\delta _\varepsilon<\delta <0\) we have
Proof
Let \(u \in D_{\mathcal {M}}'({\mathbb {T}}^N), k\in {\mathbb {Z}}_+\) and \(\delta <0\) be given. By using (3.2) and Remark 3.9 we have
Noticing that
it follows from the (5.5) that
It follows from the Minkowiski’s inequality and Schwarz’s inequality that we can estimate the last term by
Let \(q \in {\mathbb {Z}}_+\) such that \(q \ge |\sigma |\). Now we use the Proposition 3.1 and inequality
in order to conclude that, for some positive constants \(C_2\) and \(h_2\),
Now we set \(C_{3} = C_{2} h_{2}^{q+2N} \Big (\displaystyle \sum _{\eta \in {\mathbb {Z}}^{N}} \displaystyle \frac{1}{(1+|\eta |)^{2N}}\Big )\) and use (2.7) to obtain
From (2.3), it follows that
where \(C_4^2=C_{3} A_{q+2N}\) and \(h_3= 2 H h_2 A_{2q+N}\). From this inequality and (5.6) we conclude that
where
Let us analyze the term (B). First, note that by applying inequality (5.7), with the roles of \(\xi \) and \(\eta \) changed, we obtain
where
For the term (C), observe that by fixing \(s \in {\mathbb {N}}\), we can write
By using the fact that the supremum that appears on the expression of (C1) is greater or equal to 1 and Proposition 3.1, item (ii), there exist positive constants \(C_2\) and \(h_2\) such that
We notice that \(|\xi |\le \frac{s}{s+1}|\eta |\) implies that \(|\xi - \eta | \ge \frac{|\eta |}{s+1}\) in turns implies that \(\frac{1}{1+|\xi -\eta |} \le \frac{1}{1+\frac{|\eta |}{s+1}}\) and it follows from the above inequality that
where \(C_5=C_{2} h_{2}^{2N + q} \displaystyle \sum _{\xi \in {\mathbb {Z}}^N}\displaystyle \frac{1}{(1+|\xi - \eta |)^{2N}}\) and therefore does not depend on \(\eta \) and \(n \in {\mathbb {Z}}_{+}\). We use again (2.3) and (2.7):
Now we choose \(\delta <0\) such that \(|\delta | \le \displaystyle \frac{1}{(s+1) h_{2} A_{2N+q} 2H}\). Then
which implies that
for all \(\eta \in {\mathbb {Z}}^N\), since \(C_6\dot{=}C_{5} A_{2N+q}\) and \(\delta \) does not depend on \(\eta \).
We now are going to estimate the term (C2). By hypothesis,
Thus
and then
By Proposition 3.1, item (ii), we have
and then by setting \(B_{2} \dot{=} C_{2} h_{2}^{q + 2N} m_{q+2N} (q +2N)! \Big (\displaystyle \sum _{\xi \in {\mathbb {Z}}^{N}} \displaystyle \frac{1}{(1+|\xi |)^{2N}} \Big )\) we obtain
From (2.6), we may rewrite the last inequality as
Now we use (2.8) with \(k = \displaystyle \frac{s}{s+1}\) and \(\rho = |\delta | (1+|\eta |)\), in order to get
Multiplying both sides of the last equation by \(-2\) and taking the exponential, we obtain
If \(|\delta |\le 1\) then
Now, for a given \(\varepsilon > 0\), we take \(s \in {\mathbb {Z}}_+\) such that \(\displaystyle \frac{\log \left( 1 + \frac{1}{s}\right) }{\log H_0} \le \varepsilon \) and the last inequality tells us that
Puting (5.12) into (5.13) and using (2.6), we obtain
In short, (5.10) , (5.11) and (5.14) give us
where \(B_3^2= \max \left\{ C_{6}, B_{2} \right\} \) for all \(-1<\delta <0\), with \(|\delta | \le \displaystyle \frac{1}{(s+1) h_{2} A_{2N+q} 2H}\). Now, by associating (5.9) and (5.15), we conclude that
which, together with (5.8), ensures us that
Let us fix \(1 \le j \le k-1\). Since \(\sigma + \varepsilon< k - j + \sigma + \varepsilon < k + \sigma + \varepsilon \), we use Lemma 5.4 with \(\lambda _{j} = \displaystyle \frac{1}{\left( {\begin{array}{c}k\\ j\end{array}}\right) (2h_{3})^{j} m_{j} j!}\) to conclude that
By the choice of \(\lambda _j\),
Because \(\frac{k!}{(k-j)!} \le k^j\) and \(k!\le k^k \le e^k \cdot k!\), it follows from Proposition 2.4 (item (ii)) that
where \(h_{4} = 2 h_{3} e\).
Now we join the informations of (5.16), (5.17), (5.18) and the fact that \(h_3\le h_4\), to conclude that
By taking \(C \dot{=} 2(B_3 C_{4} 2)\) we have from last inequality:
for all \(-\delta _\epsilon< \delta <0\), where \(\delta _{\varepsilon } = \min \left\{ 1, \displaystyle \frac{1}{(s+1) h_{2} A_{2N+q} 2H}\right\} \) (recall that s is a natural number with \( \displaystyle \frac{\log \left( 1 + \frac{1}{s}\right) }{\log H_0} \le \varepsilon \)). The proof is now complete. \(\square \)
One of the main reasons for introducing this class is the following result:
Theorem 5.6
Let \({\mathcal {A}}=\{a_j(x,D)\}_{j=1}^m\) be a system of pseudodifferential operators in \(\mathfrak {D}_{\mathfrak {p}_{\sigma }}^{\mathcal {M}}({\mathbb {T}}^N)\) that is globally \({\mathcal {M}}\)-hypoelliptic with loss of \(r\ge 0\) derivatives. If \({\mathcal {B}}=\{b_j(x,D)\}_{j=1}^m\) is a system of pseudodifferential operators in \(\mathfrak {D}_{\mathfrak {p}_{\tau }}^{\mathcal {M}}({\mathbb {T}}^N)\), with \(\tau <\sigma -r\), then the system \({\mathcal {C}}\dot{=}\{c_j(x,D)=a_j(x,D) + b_j(x,D)\}_{j=1}^m\) in \(\mathfrak {D}_{\mathfrak {p}_{\sigma }}^{\mathcal {M}}({\mathbb {T}}^N)\) is globally \({\mathcal {M}}\)-hypoelliptic.
Proof
Let \(u \in D_{\mathcal {M}}'({\mathbb {T}}^N)\) such that \(c_j(x,D)u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\) for all \(j=1,\ldots ,m\). We will prove that \(u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\). By hypothesis about \({\mathcal {A}}\) and the triangular inequality, one can find constants \(C_{1}, \ B_1 > 0, \ \delta _{0} <0\) and \(k_{0} \in {\mathbb {R}}\) such that:
for all \(k \in {\mathbb {Z}}_+\) and \(\delta _{0}< \delta < 0\).
It follows from Lemma 4.5 the existence of positive constants \(\beta _0, C_2, h_1\) such that for \(\delta \le \beta _0\) we have
in particular it also holds true for \(\delta _0<\delta <0\).
Hence, if we set \(C_{3} = \max \{C_1,C_2\}\) and \(h_2 = 2\max \{h_1, B_1\}\), we can rewrite the last inequality as
for all \(\delta _0< \delta <0\) and \(k \in {\mathbb {Z}}_+\).
Let us analyze the term \(\max _{1\le j \le m}\left\| b_j(x,D)u \right\| _{\left\{ \mathcal {M}, \delta , k \right\} }\). By taking \(\varepsilon >0\) such that \(\tau + \varepsilon < \sigma - r\), it follows Proposition 5.5 that there exist constants \(C_4, h_{3} > 1\) and \(\delta _{\varepsilon } < 0\) such that
Thus, we conclude that
for all \(\delta _1\dot{=}\max \{\delta _0, \delta _\varepsilon \}< \delta <0\) and \(k \in {\mathbb {Z}}_+\).
We may suppose, without loss of generality, that \(k_{0} \ge \tau + \varepsilon \). By Proposition 4.3,
for all \(k \in {\mathbb {Z}}_+\) and \(\delta _1<\delta <0\), \(C_5=\max \{2C_3C_4, 2C_3, 1\}\) and \(h_4=\max \{h_2,h_3\}\).
By choice of \(\varepsilon \), we have \(\tau + \varepsilon - 1< k + \tau + \varepsilon < k + \sigma - r\) for any \(k\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+\). By taking \(\lambda = \displaystyle \frac{1}{2C_{5}}\), Lemma 5.4 ensures us that
where \(n_{0}\) is a fixed natural number satisfying \(n_{0} \ge \displaystyle \frac{1}{\sigma - r - (\tau + \varepsilon )}\) (also note that \(C_5\ge 1\)). By setting \(h_5=(2C_5)^{n_0}\), one can infer that
Putting (5.21) into (5.20) we obtain
Since \(m_k\ge 1\) for all \(k\in {\mathbb {Z}}_+\), if we take \(h_{6} = \max \left\{ h_{4}, h_5, 1 \right\} \) then it follows that
By taking \(C_{6} = 4C_5 h_6\), we obtain from (5.22):
Since (5.1) is similar to (5.23) then following the lines of the proof of Theorem 5.3 we are able to show that \(u \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\). The proof is complete. \(\square \)
6 The Solvability Theorem
It is well known that if P(x, D) is a globally hypoelliptic differential operator on \({\mathbb {T}}^N\), its transpose is globally solvable, see, e.g., Treves [15]. For the Gevrey version of this result, see Albanese, Corli and Rodino [1] and for a global version in the Gevrey classes, see Albanese and Zanghirati [2]. In this section, following the lines of [3] where the author requires a weak version of global hypoellipticity in Gevrey classes, we prove an analogue of these results in ultradifferentiable classes, on the torus \({\mathbb {T}}^N\), for our class of ultradifferential pseudodifferential operators.
Let \({\mathcal {A}}=\{a_1(x,D),\ldots ,a_m(x,D)\}\) be a system of pseudodifferential operators in \({\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\). We also use \({\mathcal {A}}\) to denote the map \({\mathcal {A}}:\mathcal {E}_{\mathcal {M}}\left( {{\mathbb {T}}^N}\right) (D_{\mathcal {M}}'({\mathbb {T}}^N))\longrightarrow \left( \mathcal {E}_{\mathcal {M}}\left( {{\mathbb {T}}^N}\right) \right) ^{m} \left( \left( D_{\mathcal {M}}'({\mathbb {T}}^N)\right) ^m\right) \) given by
By using the identification \(\left( \left( \mathcal {E}_{\mathcal {M}}\left( {{\mathbb {T}}^N}\right) \right) ^{m}\right) '= \left( D'_\mathcal {M}\left( {\mathbb {T}}^N\right) \right) ^m\), the transpose operator of \({\mathcal {A}}\), denoted by \(\;\!^t {\mathcal {A}}:\left( D'_\mathcal {M}\left( {\mathbb {T}}^N\right) \right) ^m \longrightarrow D'_\mathcal {M}\left( {\mathbb {T}}^N\right) \), is given by
Suppose that \(w \in D'_\mathcal {M}\left( {\mathbb {T}}^N\right) \) and \(\;\!^t {\mathcal {A}}(u_1,\ldots ,u_m) = w\) for some \((u_1,\ldots ,u_m) \in \left( D'_\mathcal {M}\left( {\mathbb {T}}^N\right) \right) ^m\). If \(\varphi \in \ker {\mathcal {A}}\), with \(\varphi \in \mathcal {E}_{\mathcal {M}}\left( {{\mathbb {T}}^N}\right) \), then \(a_j(x,D)\varphi =0\) for all \(j=1,\ldots ,m\). Hence
That is, \(w \in (\ker {\mathcal {A}})^\perp \). In other words,
Also recall that by arguments of functional analysis, \({{\,\mathrm{ran}\,}}\;\!^t {\mathcal {A}}\) is closed if and only if \({{\,\mathrm{ran}\,}}(\;\!^t {\mathcal {A}}) = ({{\,\mathrm{ker}\,}}{\mathcal {A}})^\perp \) (see Lemma 2.2 of [3] and Theorem 9 of [11]). This motivates the following
Definition 6.1
We say that the transpose \(\;\!^t {\mathcal {A}}\) is globally solvable in \(D'_\mathcal {M}\left( {\mathbb {T}}^N\right) \) if \(\;\!^t {\mathcal {A}}\) has closed image, which is equivalent to say that \({{\,\mathrm{ran}\,}}(\;\!^t {\mathcal {A}}) = ({{\,\mathrm{ker}\,}}{\mathcal {A}})^\perp \).
Remark 6.2
The definition above means that for all \(w\in D'_\mathcal {M}\left( {\mathbb {T}}^N\right) \) such that \(\langle w, \varphi \rangle =0,\,\,\text {with}\,\,\varphi \in \mathcal {E}_{\mathcal {M}}\left( {{\mathbb {T}}^N}\right) \) and \({\mathcal {A}} \varphi =0\), there exists \(u=(u_1,\dots ,u_m)\in \left( D'_\mathcal {M}\left( {\mathbb {T}}^N\right) \right) ^m\) such that \(\;\!^t {\mathcal {A}} (u)=w.\)
In order to use the technique introduced in [3], which is the key point in the proof of our main result of this section, we need some notations and results.
If \({\mathcal {M}}=\{m_n\}_{n\in {{\,\mathrm{{\mathbb {Z}}}\,}}_+}\) is a weight sequence, we set
Since \(\mathscr {L}_{\mathcal {M}}\) is given by the product of two weight sequences it is easy to see that \(\mathscr {L}_{\mathcal {M}}\) is a weight sequence. Furthermore,
It follows from Remark 2.6 that \(\mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N) \subset \mathcal {E}_{\mathscr {L}_{\mathcal {M}}}({\mathbb {T}}^N)\). In fact we have a more precise information, given by the
Lemma 6.3
\(\mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N) \subset \mathcal {E}_{\mathscr {L}_{\mathcal {M}}, 1}({\mathbb {T}}^N)\) and the inclusion \(i:\mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N) \hookrightarrow \mathcal {E}_{\mathscr {L}_{\mathcal {M}}, 1}({\mathbb {T}}^N)\) is continuous.
Proof
It suffices to prove that for a given \(h >0\), \(\mathcal {E}_{\mathcal {M}, h}({\mathbb {T}}^N) \hookrightarrow \mathcal {E}_{\mathscr {L}_{\mathcal {M}}, 1}({\mathbb {T}}^N)\) continuously.
If \(f \in \mathcal {E}_{\mathcal {M}, h}({\mathbb {T}}^N)\) we have
Since \(\displaystyle \lim _{k \rightarrow +\infty } \displaystyle \frac{h^k}{k!} = 0\), we can find \(k_{0} \in {\mathbb {Z}}_+\) such that \(k! \ge h^{k}\) for every \(k \ge k_{0}\) and therefore (6.1) reads as
By taking \(C = \max \{1, \max \{h^p: 0\le p \le k_0\}\}\), it follows from (6.1) and (6.2) that
which implies that
The proof is complete. \(\square \)
We now introduce a weak version of global hypoellipticity in ultradifferentiable classes that was inspired by [3].
Definition 6.4
Let \({\mathcal {A}}=\{a_1(x,D),\ldots ,a_m(x,D)\}\) be a system of pseudodifferential operators in \({\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\). We say that \({\mathcal {A}}\) is globally \((\mathscr {L}_{\mathcal {M}}-\mathcal {M})\)-hypoelliptic on \(\mathbb {\mathbb {T}}^N\) if the following condition holds true:
Remark 6.5
It follows immediately from Lemma 6.3 that \(D'_{\mathscr {L}_{\mathcal {M}}}({\mathbb {T}}^N) \hookrightarrow D'_\mathcal {M}\left( {\mathbb {T}}^N\right) \) and therefore every globally \(\mathcal {M}\)-hypoelliptic system of pseudodifferential operators in \({\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\) is also globally \((\mathscr {L}_{\mathcal {M}}-\mathcal {M})\)-hypoelliptic on \(\mathbb {\mathbb {T}}^N\).
Lemma 6.6
Let \({\mathcal {A}}\) be a globally \((\mathscr {L}_{\mathcal {M}}-\mathcal {M})\)-hypoelliptic on \(\mathbb {\mathbb {T}}^N\) system of pseudodifferential operators in \({\mathfrak {D}}^{\mathcal {M}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\) and let \(\Gamma \) be the graph of \({\mathcal {A}}: \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N) \rightarrow \left( \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\right) ^m\). Then \(\Gamma \) is a closed subspace of \(\mathcal {E}_{\mathscr {L}_{\mathcal {M}}}({\mathbb {T}}^N) \times \left( \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\right) ^m\).
Proof
By noticing that \(\mathcal {E}_{{\mathscr {L}}_\mathcal {M}}({\mathbb {T}}^N) \hookrightarrow D'_{\mathscr {L}_{\mathcal {M}}}({\mathbb {T}}^N),\, D_{\mathscr {L}_{\mathcal {M}}}'({\mathbb {T}}^N) \hookrightarrow D_{\mathcal {M}}'({\mathbb {T}}^N)\) and that the operators \(a_j(x,D)\in {\mathfrak {D}}^{\mathscr {L}_{\mathcal {M}}}_{{\mathfrak {p}}_\sigma } ({\mathbb {T}}^N)\), one can complete the proof easily. \(\square \)
Now we can state and prove our main result of global solvability of this section.
Theorem 6.7
If \({\mathcal {A}}\) is globally \({\mathcal {M}}\)-hypoelliptic on \({\mathbb {T}}^N\) then \(\ker {\mathcal {A}}\) has finite dimension and \(\;\!^t {\mathcal {A}}\) is globally solvable in \(D'_{\mathcal {M}}\left( {\mathbb {T}}^N\right) \).
Proof
If \({\mathcal {A}}\) is globally \({\mathcal {M}}\)-hypoelliptic on \({\mathbb {T}}^N\) then \({\mathcal {A}}\) is globally \(({\mathscr {L}}_\mathcal {M}- \mathcal {M})\)-hypoelliptic on \(\mathbb {\mathbb {T}}^N\) (see Remark 6.5). In particular, we may apply Lemma 6.6 and conclude that the graph \(\Gamma \) of \({\mathcal {A}}: \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N) \rightarrow \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)^m\) is closed when we view \(\Gamma \) as a subspace of \(\mathcal {E}_{\mathscr {L}_{\mathcal {M}}}({\mathbb {T}}^N) \times \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)^m\). Hence, it follows from our Lemma 6.3 and Theorem 2.5 in [3] \(\big (\text {putting}\,\, E_0=\mathcal {E}_{\mathscr {L}_{\mathcal {M}},1}({\mathbb {T}}^N)\big )\), that \(\ker {\mathcal {A}}\) has finite dimension and ran\(({\mathcal {A}})\) is closed in \(\left( \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\right) ^m\). Now applying Lemma 2.2 in [3] we conclude that ran\((\;\!^t {\mathcal {A}})\) is closed in \(D'_{\mathcal {M}}({\mathbb {T}}^N)\) and therefore \(\;\!^t {\mathcal {A}}\) is globally solvable in \(D'_{\mathcal {M}}\left( {\mathbb {T}}^N\right) \). The proof is complete. \(\square \)
We finish this section with a result that is consequence of Theorem 5.6 and Theorem 6.7.
Corollary 6.8
Let \({\mathcal {A}}=\{a_j(x,D)\}_{j=1}^m\) be a system of pseudodifferential operators in \(\mathfrak {D}_{\mathfrak {p}_{\sigma }}^{\mathcal {M}}({\mathbb {T}}^N)\) that is globally \({\mathcal {M}}\)-hypoelliptic with loss of \(r\ge 0\) derivatives. If \({\mathcal {B}}=\{b_j(x,D)\}_{j=1}^m\) is a system of pseudodifferential operators in \(\mathfrak {D}_{\mathfrak {p}_{\tau }}^{\mathcal {M}}({\mathbb {T}}^N)\), with \(\tau <\sigma -r\), then the kernel of the system \({\mathcal {C}}\dot{=}\{c_j(x,D)=a_j(x,D) + b_j(x,D)\}_{j=1}^m\) has finite dimension and \(\;\!^t {\mathcal {C}}\) is globally solvable in \(D'_{\mathcal {M}}\left( {\mathbb {T}}^N\right) \).
7 Application: Operators with Constant Strenght
In this section, we consider a class of linear partial differential operators that was considered by [5] in the analytic setup, given by
where \(a_{j}(x) \in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\) and \(P_{0}(D), P_{1}(D) \ldots P_{m}(D)\) are differential operators with constant coefficients in \({\mathbb {T}}^N\), satisfying:
-
1.
There exist positive constants C, M and R such that
$$\begin{aligned} \left| P_{0} (\xi ) \right| \ge \displaystyle \frac{C}{(1+|\xi |)^{M}}, \ \ \forall \,\, \xi \in {\mathbb {Z}}^N,\, |\xi | \ge R. \end{aligned}$$(7.2) -
2.
For each \(j \in \left\{ 1, 2, \ldots , m \right\} \) there exist positive constants \(\beta _{j}, c_{j}\) such that
$$\begin{aligned} |P_{j}(\xi )| \le c_{j} \cdot \displaystyle \frac{|P_{0}(\xi )|}{(1+|\xi |)^{\beta _{j}}}, \ \ \forall \,\, \xi \, \in {\mathbb {Z}}^N, |\xi | \ge R. \end{aligned}$$(7.3)
Notice that under such conditions the operator a(x, D) defined in (7.1) is of constant strengh (cf. [7]).
Theorem 7.1
Let a(x, D) be the operator in (7.1) and \(\beta _0\) be the order of operator \(P_0(D)\). Then a(x, D) is globally \(\mathcal {M}\)-hypoelliptic on \({\mathbb {T}}^N\) with loss of \(\beta _{0} + M\) derivatives.
Proof
We start by noticing that operator a(x, D) given in (7.1) belongs to the space \(\mathfrak {D}^{\mathcal {M}}_{\mathfrak {p}_{\beta _{0}}} ({\mathbb {T}}^N)\).
For \(u \in D'_{\mathcal {M}}({\mathbb {T}}^N), \delta <0\) and \(k \in {\mathbb {Z}}_+\) we have
where R comes from (7.3).
We now use (7.2) and the last inequality in order to obtain
By setting \(B_{1} = \max \left\{ (1+R), C^{-1} \right\} \) it follows from the last inequality that
For any \(\ell >0\), we have
Let us consider \(\tau >0\) and \(C'>0\) satisfying
Thus
We also obtain from (7.3) that
If we choose \(B_{2} = \max \left\{ c_{1}, c_{2}, \ldots c_{m}, C' (1+R)^{\tau + M} \right\} \) and \(\gamma = \min \left\{ \beta _{1}, \beta _{2}, \ldots , \beta _{m} \right\} \) then we obtain from the last two inequalities that
Since we have \(a(x,D) = P_{0}(D) + \displaystyle \sum _{j = 1}^{m} a_{j}(x) P_{j}(D)\), by using the triangular inequality we can write
Now since \(a_{j}\in \mathcal {E}_{\mathcal {M}}({\mathbb {T}}^N)\), it defines an element of \(\mathfrak {D}_{\mathfrak {p}_{0}}^{\mathcal {M}}({\mathbb {T}}^N)\). In particular, we can use Proposition 5.5 (with \(\sigma =0\) and \(\varepsilon =\frac{\gamma }{2}\)) in order to conclude the existence of constants \(B_{3}, h > 0\) and \(\delta _{\gamma } < 0\), such that, for every \(k\in {\mathbb {Z}}_+\), \(\delta _{\gamma }< \delta < 0\), we have
From (7.6) and (7.7), we obtain
By using (7.5) twice, one with \(\ell =\frac{\gamma }{2}\) and the other with \(\ell =k+\frac{\gamma }{2}\), it follows from above inequality that
If we set \(B_{4} = \max \{2m B_{2} B_{3}, 2m B_{2} B_{3} B_{1}^{\frac{\gamma }{2}}, 1\}\) and \(h_{2} = \max \left\{ h, B_{1} \right\} \), then we deduce that
By taking \(\omega \in {\mathbb {R}}\) such that \(\omega < - \displaystyle \frac{\gamma }{2}\), we obtain \(\omega< k - \displaystyle \frac{\gamma }{2} < k\) for each k em \({\mathbb {Z}}_+\) and use Lemma 5.4 (with \(\lambda = \displaystyle \frac{1}{2B_{4}}\)) to conclude that
where \(C_1=(2B_4)^{-\frac{2\omega }{\gamma }}\) and \(r=(2B_4)^{\frac{2}{\gamma }}\).
Absorving the term \(\frac{ \left\| P_{0}(D)u \right\| _{\left\{ \mathcal {M}, \delta , k \right\} }}{2}\) in the left hand side of (7.10) we obtain
If we set \(B_{5} = 2 \max \left\{ C_{1}, B_{4} \right\} \) and \(s_{1} = \max \left\{ h_{2}, r \right\} \), then by using that \(\omega < - \displaystyle \frac{\gamma }{2}\) and Proposition 4.3 it follows from (7.11) that
It follows from (7.4) and (7.12) that
Since \(P_{0}(D) \in \mathfrak {D}^{\mathcal {M}}_{\mathfrak {p}_{\beta _{0}}} ({\mathbb {T}}^N)\), we have
Hence, if we take
we may utilize Proposition 4.3, (7.13) and (7.14) in order to infer that
for any \(k\in {\mathbb {Z}}_+\), \(u\in D'_{\mathcal {M}}({\mathbb {T}}^N)\), \(\delta _\gamma<\delta <0\) and the proof is finally complete. \(\square \)
Remark 7.2
In [5] the hyphotesis (7.2) is the following: There exist positive constants c, R and \(\varepsilon >0\) such that
thus our result generalizes that one in [5].
Corollary 7.3
Let a(x, D) be the operator in (7.1). Then a(x, D) is globally \(\mathcal {M}\)-hypoelliptic on \(\mathbb {T}^N\), its kernel has finite dimension and its transpose \(\;\!^t {\mathcal {A}}\) is globally solvable in \(D'_{\mathcal {M}}(\mathbb {T}^N)\). Furthermore, these properties remain valid if we perturb the operator a(x,D) by any ultradifferentiable pseudodifferential operator of order \(\tau < -M\).
Proof
It is a direct consequence of Theorem 7.1, Theorem 5.3, Theorem 5.6, Theorem 6.7 and Corollary 6.8. \(\square \)
References
Albanese, A.A., Corli, A., Rodino, L.: Hypoellipticity and local solvability in Gevrey classes. Math. Nachr. 243, 283–290 (2002)
Albanese, A.A., Zanghirati, L.: Global hypoellipticity and global solvability in Gevrey classes on the \(n\)-dimensional torus. J. Differ. Equ. 199, 256–268 (2004)
Araújo, G.: Regularity and solvability of linear differential operators in Gevrey spaces. Math. Nachr. 291, 729–758 (2018)
Braun, Rodrigues N., Chinni, G., Cordaro, P.D., Jahnke, M.R.: Lower order perturbation and global analytic vectors for a class of globally analytic hypoelliptic operators. Proc. Am. Math. Soc. 144(12), 5159–5170 (2016)
Chinni, G., Cordaro, P.D.: On Global Analytic and Gevrey Hypoellipticity on the Torus and the Métivier Inequality. Commun. Partial Differ. Equ. 42, 121–141 (2017)
Ferra, I. A., Petronilho, G.: Perturbations by lower order terms do not destroy the global hypoellipticity and solvability of certain systems of pseudodifferential operators. submitted
Hörmander, L.: The Analysis of Linear Partial Differential Operators, vol. II. Springer, New York (1983)
Hörmander, L.: The Analysis of Linear Partial Differential Operators, vol. III. Springer, New York (1985)
Kirilov, A., Victor, B.L.: Global and Partial Fourier Series for Denjoy-Carleman Classes on Torus. (2019). https://arxiv.org/pdf/1910.12605.pdf
Komatsu, H.: Ultradistribution I. Structure theorems and a characterization. J. Fac. Sci. Tokyo, Ser. IA. 20, 25–105 (1973)
Komatsu, H.: Projective and injective limits of weakly compact sequences of locally convex spaces. J. Math. Soc, Japan. 19, 366–383 (1967)
Parmeggiani, A.: A remark on the stability of \(C^\infty \)-hypoellipticity under lower-order perturbations. Pseudo-Differ. Oper. Appl. 6, 227–235 (2015)
Philipović, S.: Structural theorems for ultradistributions. Different aspects of differentiability. Diss. Math. 340, 223–235 (1995)
Ruzhansky, M., Turunen, V.: Pseudo-Differential Operators and Symmetries: Background Analysis and Advanced Topics. Besel - Boston - Berlin: Birkhäuser Verlag AG
Treves, F.: Topological Vector Spaces, Distributions and Kernels. Academic Press, New York (1967)
Acknowledgements
The second author was supported partially by São Paulo Research Foundation (FAPESP) under grant 2018/14316-3, CNPq under grant 303111/2015-1. The third author was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 001. The authors would like to thank the referees for constructive comments.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Fabio Nicola.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Ferra, I.A., Petronilho, G. & de Lessa Victor, B. Global \({\mathcal {M}}-\)Hypoellipticity, Global \({\mathcal {M}}-\)Solvability and Perturbations by Lower Order Ultradifferential Pseudodifferential Operators. J Fourier Anal Appl 26, 85 (2020). https://doi.org/10.1007/s00041-020-09799-7
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00041-020-09799-7
Keywords
- Ultradifferentiable classes
- Global hypoellipticity with loss of derivatives
- Solvability
- System of \(\mathcal {M}\)-ultradifferentiable pseudodifferential operators
- N-dimensional torus
- Lower order perturbations