Spectrum of composition operators on S(R) with polynomial symbols

https://doi.org/10.1016/j.aim.2020.107052Get rights and content

Abstract

We study the spectrum of operators in the Schwartz space of rapidly decreasing functions which associate each function with its composition with a polynomial. In the case where this operator is mean ergodic we prove that its spectrum reduces to {0}, while the spectrum of any non mean ergodic composition operator with a polynomial always contains the closed unit disc except perhaps the origin. We obtain a complete description of the spectrum of the composition operator with a quadratic polynomial or a cubic polynomial with positive leading coefficient.

Introduction

Composition operators on Fréchet spaces of smooth functions on the reals have attracted the attention of several authors recently ([7], [8], [12], [16], [17], [18], [19]) but to our knowledge very little is known about the spectra of composition operators in this setting. See for instance [3] or [5], where the spectrum of composition operators and other classical operators on spaces of real smooth functions is investigated. This contrasts with the large number of existing articles studying spectral properties of composition operators in Banach spaces of analytic functions on the unit disc.

We study the spectrum of composition operators defined in the Schwartz space S(R) of smooth rapidly decreasing functions, Cφ:S(R)S(R),ffφ, in the case that φ is a non constant polynomial. Recall that S(R) consists of those smooth functions f:RC with the property thatπn(f):=supxRsup0jn(1+x2)n|f(j)(x)|< for each nN. S(R) is a Fréchet Montel space when endowed with the topology generated by the sequence of seminorms (πn)nN. In particular, a sequence (fj)j in S(R) converges to f if and only if P(x)(fjf)(n)(x) converges to zero uniformly on the real line for each polynomial P and each nN. This space, introduced by Schwartz [20], plays a central role in the theory of Linear Partial Differential Equations and also in Harmonic Analysis (see for example [11]). We refer to [14] for background on Fréchet spaces.

A smooth function φ:RR is said to be a symbol for S(R) if Cφ maps S(R) continuously into itself. The symbols for S(R) were completely characterized in [8, Theorem 2.3]. It follows from that characterization that any non constant polynomial φ is a symbol for S(R). The results of the present paper complement our study in [7] where we investigate dynamics and the spectrum of some particular composition operators. Concrete examples were given where the spectrum coincides with the open unit disc, the unit circle or C{0}. In particular, the spectrum of translation and dilation operators was analyzed in [7, Examples 5-6]:

Example 1.1

  • (a)

    Let φ(x)=x+1. Then σ(Cφ)={λC:|λ|=1}.

  • (b)

    Let φ(x)=ax where a0 and |a|1. Then σ(Cφ)=C{0}.

However, in [7] we did not obtain any result concerning the spectrum of the composition operator with a polynomial of degree greater than one. This is precisely the objective of this work. It turns out that some dynamical properties of the composition operator are characterized by the spectrum of the operator. For example, the spectrum of a mean ergodic composition operator is always contained in the closed unit disk ([7, Corollary 4.5]). This result can be improved when the symbol is a polynomial. As we prove in Theorem 2.5, the spectrum of a composition operator which is mean ergodic and whose symbol is a polynomial with degree greater than one coincides with {0} while the behavior of non mean ergodic composition operators with polynomial symbols is different (Theorem 2.8). For strictly decreasing symbols, not necessarily polynomials, the containment of D in the spectrum of the operator is equivalent to its mean ergodicity.

In [5] the spectrum of composition operators on A(R), the space of all real analytic functions, is investigated for the case that the symbol is a quadratic polynomial. For quadratic polynomials, we have a complete characterization of the spectrum of the corresponding composition operator depending on the number of fixed points of the polynomial. As expected, the spectrum of the operator depends on the space where it is considered. To give an example, when φ is a quadratic polynomial without fixed points then σA(R)(Cφ)=C, whereas σS(R)(Cφ)={0}.

Theorem 4.1 contains a complete description of the spectrum of a composition operator with a polynomial of degree three whose leading coefficient is positive. For polynomials with negative leading coefficient some partial results are available but we lack a complete characterization.

The final section contains some results concerning the spectra of composition operators with (non polynomial) monotone symbols.

We recall that given an operator T:XX on a Fréchet space X, σ(T), the spectrum of T, is the set of all λC such that TλI:XX does not admit a continuous linear inverse. Unlike for Banach spaces, the spectrum σ(T) need not be compact and it may happen that σ(T)=. An operator T is said to be power bounded if {Tn(x):nN} is bounded for each xX. A closely related concept to power boundedness is that of mean ergodicity. Given TL(X), the Cesàro means of T are defined as T[n]=k=1nTk/n. T is said to be mean ergodic when T[n] converges to an operator P, which is always a projection, in the strong operator topology, i.e. if (T[n](x)) is convergent to P(x) for each xX. There is a vast literature studying dynamics of linear operators on Banach or Fréchet spaces. See for instance [1], [4], [10], [13], [15] and the references therein.

From now on φn=φφ denotes the n-th iteration of φ.

The following results will be used in what follows.

Lemma 1.2

[7, 3.10] Let φ be a polynomial of even degree without fixed points. Then there is NN such that ψ=φN has neither zeros nor fixed points. Moreover, for every K>0 there is m0N such that|ψm+1(t)|K(ψm(t))2mm0,tR.

Theorem 1.3

[7, 3.11] Let φ be a polynomial with degree greater than or equal to two. Then, the following are equivalent:

  • (1)

    Cφ is power bounded.

  • (2)

    Cφ is mean ergodic.

  • (3)

    The degree of φ is even and it has no fixed points.

Section snippets

Polynomial symbols

Two polynomials φ,ψR[x] are linearly equivalent if there exists (x)=ax+b with a,bR and a0 such that ψ=1φ. Then, for every λC,CφλI=C1(CψλI)C, from where it follows that σ(Cψ)=σ(Cφ).

The first result follows immediately from this observation and [7, Examples 5-6] as each polynomial of degree one other than the identity is linearly equivalent to a translation or to a dilation.

Proposition 2.1

Let φ(x)=ax+b a polynomial with a,bR and a0. Then

  • (a)

    For a1, φ is linearly equivalent to ψ(x)=ax. Hence σ(Cφ)

Quadratic polynomials

Next we apply the previous results to discuss the spectrum of Cφ in the case that φ is a quadratic polynomial. Such a polynomial φ(x)=a0+a1x+a2x2 (a20) is linearly equivalent to ψ(x)=x2+c where c=a0a2+a12a124. In fact, take (x)=ax+b where a=a2,b=a12. It is routine to check that φ=1ψ. We observe that 0σ(Cψ)=σ(Cφ) since the range of Cψ consists of even functions.

c>14 implies that φ and ψ lack fixed points, hence σ(Cφ)={0} (Theorem 2.5). In the case c<14 we have that φ (and also ψ) has

Cubic polynomials

Let us now consider polynomials φ of degree 3 with limx+φ(x)=+.

Theorem 4.1

Let φ be a polynomial of degree 3 with positive leading coefficient. Then C{0}σ(Cφ) unless φ has a fixed point of multiplicity 3 in which case σ(Cφ)=D{0}.

Proof

According to its fixed points the following cases can occur:

  • (i)

    φ has three different real fixed points,

  • (ii)

    φ has two different real fixed points, one with multiplicity two and the other is simple,

  • (iii)

    φ has only one real fixed point, the other two being complex conjugate numbers,

  • (iv)

    φ has

Monotone symbols

We recall that the symbols for S(R) were completely characterized in [8, Theorem 2.3]. The aim of this section is to provide some information regarding the spectrum of composition operators defined by monotone symbols. Then, let us assume that the symbol φ is strictly monotone and let us denote by ψ its inverse and by ψn its n-th iterate. For λ0, and f,gS(R), the relation Cφfλf=g implies that (5) holds for every n, that isf(φn(x))=λnf(x)+k=0n1λn1kg(φk(x)), which impliesf(x)=λnf(ψn(x))+k=

References (21)

There are more references available in the full text version of this article.

The present research was partially supported by the projects MTM2016-76647-P and Prometeo2017/102 (Spain).

View full text