betaFeedback

WorksMathematics

Spectral and mapping properties of the inverse Ornstein–Uhlenbeck operator

Abstract

We investigate the spectral and boundedness properties of the inverse Ornstein–Uhlenbeck operator AQ,B\mathcal{A}^{Q,B} and its associated semigroup in LpL^p spaces with respect to the inverse Gaussian measure. We establish strong continuity and contractivity of this semigroup in LpL^p for p∈[1,∞)p \in [1, \infty), and obtain a complete characterization of the spectrum of AQ,B\mathcal{A}^{Q,B}. For p∈(1,∞)p \in (1, \infty), the spectrum is independent of pp, consists entirely of eigenvalues and exhibits a spectral gap. For p=1p=1, the spectrum is instead the left half-plane. Furthermore, we discuss ultracontractivity and hypercontractivity properties of the semigroup.

Introduction

Throughout this paper, we consider two real n×nn \times n matrices QQ and BB fulfilling the following two conditions:

(H1) QQ is symmetric and positive definite;

(H2) all the eigenvalues of BB have negative real parts.

For each t∈(0,+∞]t \in(0,+\infty], we introduce the covariance matrices

Qt=∫0tesBQesB∗ ds.Q_t = \int_0^t e^{sB} Q e^{sB^*}\,ds .

Each QtQ_t is well defined, symmetric and positive definite. Thus QtQ_t increases in the sense of quadratic forms and converges to Q∞Q_\infty.

We consider Rn\mathbb{R}^n endowed with the inverse Gaussian measure

dγ−∞(x)=exp⁡(12⟨Q∞−1x,x⟩) dx.(1)d\gamma_{-\infty}(x)=\exp\left(\frac{1}{2}\langle Q_\infty^{-1}x,x\rangle\right)\,dx . \tag*{(1)}

Our aim is to study spectral properties of the inverse Ornstein–Uhlenbeck operator, given by

AQ,B=12tr⁡(Q∇2)−⟨Bx,∇⟩,(2)A^{Q,B}=\frac{1}{2}\operatorname{tr}(Q\nabla^2)-\langle Bx,\nabla\rangle, \tag*{(2)}

in Lp(γ−∞)L^p(\gamma_{-\infty}). Here the symbols ∇\nabla and ∇2\nabla^2 denote the gradient and the Hessian, respectively.

The first three authors are members of the Gruppo Nazionale per l’Analisi Matematica, la Probabilità e le loro Applicazioni (GNAMPA) of the Istituto Nazionale di Alta Matematica (INdAM). Part of this research was carried out while the fourth author was GNAMPA Professore Visitatore at the University of Padova in Italy. The fourth author also profited from a grant from the Royal Society of Arts and Sciences in Gothenburg and another from the Royal Swedish Academy of Sciences. He is also grateful to the University of Padova for its hospitality during several visits.

The semigroup generated by AQ,B\mathcal{A}^{Q,B} will be denoted by the symbol

HtUO=etAQ,B,t>0.\mathcal{H}^{UO}_t=e^{t\mathcal{A}^{Q,B}},\qquad t>0.

As in [6], we use the superscript “UOUO” to highlight the relationship with the Ornstein–Uhlenbeck setting, where the superscript “OUOU” is used. We recall that the classical Ornstein–Uhlenbeck semigroup HtOU\mathcal{H}^{OU}_t is generated by the Ornstein–Uhlenbeck operator

LQ,B=12tr⁡(Q∇2)+⟨Bx,∇⟩,(3)\mathcal{L}^{Q,B}=\frac{1}{2}\operatorname{tr}(Q\nabla^{2})+\langle Bx,\nabla\rangle, \tag*{(3)}

where QQ and BB are the same as in (1.2). For extensive information on recent harmonic analysis concerning the Ornstein–Uhlenbeck semigroup, we refer the reader to [8, 9, 11, 12, 13] and the references therein.

The operator AQ,B\mathcal{A}^{Q,B} has been introduced by Salogni in [26] in the standard case Q=−B=InQ=-B=I_n, where InI_n denotes the identity matrix; there, the author studied LpL^p boundedness properties of the maximal operator for the associated semigroup. Under the same assumption Q=−B=InQ=-B=I_n, the first and the fourth authors of this paper studied boundedness properties of some Riesz transforms associated to HtUO\mathcal{H}^{UO}_t (see [5, 7]). Other results related to AI,−I\mathcal{A}^{I,-I} may be found in [1, 2, 3, 4, 21]. More recently, we introduced a more general version of AQ,B\mathcal{A}^{Q,B}, with QQ and BB only fulfilling (H1) and (H2), and proved the weak type (1,1) for the associated maximal function [6]. Notice that the measure used in [6] differs from dγ−∞d\gamma_{-\infty} by an irrelevant constant factor.

The study of the inverse Ornstein–Uhlenbeck operator presents several challenges that do not arise in the classical Ornstein–Uhlenbeck setting. Most notably, although γ−∞\gamma_{-\infty} is an invariant measure for (HtUO)t>0(\mathcal{H}^{UO}_t)_{t>0}, such semigroup does not admit an invariant probability measure; and unlike the symmetric inverse setting Q=−B=InQ=-B=I_n considered in [26, 5, 7], neither the operator nor the associated semigroup is symmetric in general. Furthermore, the reference measure γ−∞\gamma_{-\infty} is locally doubling on admissible balls, see [5], but lacks the global doubling property, and it exhibits superexponential growth at infinity. This contrast between the global growth of the measure and its local Euclidean behavior strongly affects the boundedness and contractivity of the semigroup.

The main purpose of this paper is to provide a comprehensive analysis of the spectral and boundedness properties of AQ,B\mathcal{A}^{Q,B} and its associated semigroup. First, we establish the strong continuity and the contractivity of (HtUO)t>0(\mathcal{H}^{UO}_t)_{t>0} in Lp(γ−∞)L^p(\gamma_{-\infty}) for all p∈[1,∞)p\in[1,\infty). Building on these fundamental properties, we obtain a complete characterization of the spectrum of the generator AQ,B\mathcal{A}^{Q,B} in Lp(γ−∞)L^p(\gamma_{-\infty}). We prove that for p∈(1,∞)p\in(1,\infty) the spectrum is independent of pp, consists entirely of eigenvalues determined by those of the matrix B+QQ∞−1B+QQ_{\infty}^{-1}, and exhibits a spectral gap. In contrast, for p=1p=1 the spectrum is a left half-plane.

Furthermore, we investigate the smoothing behavior of the semigroup. We show that (HtUO)t>0(\mathcal{H}^{UO}_t)_{t>0} is ultracontractive with respect to both the Lebesgue measure and the inverse Gaussian measure. This is in contrast with the Ornstein–Uhlenbeck semigroup. Finally, we prove hypercontractivity with respect to γ−∞\gamma_{-\infty}.

Scheme of the paper

1.1. Scheme of the paper. Section 2 collects some preliminary facts about the inverse Gaussian measure, the inverse Ornstein–Uhlenbeck semigroup and the associated inverse Mehler kernel. In Section 3, we prove the strong continuity and the LpL^p-contractivity of the semigroup. Section 4 is devoted to the description of the spectrum of the generator AQ,B\mathcal{A}^{Q,B}. We also discuss the orthogonality or the failure of orthogonality of some generalized eigenspaces in the inverse setting. In Section 5, we establish the ultracontractivity of the semigroup and discuss a first counterexample concerning its short-time behavior. Finally, Section 6 provides a detailed analysis of the hypercontractivity of (HtOU)t>0(\mathcal{H}^{OU}_t)_{t>0}.

Notation

We shall adopt the “variable constant convention”, according to which positive constants are denoted by symbols such as c>0c>0 or C<∞C<\infty, and these are not necessarily equal at different occurrences. All constants will depend only on nn, QQ and BB, unless otherwise explicitly stated. For any two non-negative quantities aa and bb we write a≲ba\lesssim b (or b≳ab\gtrsim a) if a≤Cba\leq Cb for some CC; moreover a≃ba\simeq b stands for the conjunction of a≲ba\lesssim b and b≲ab\lesssim a. By N\mathbb{N} we mean {0,1,…}\{0,1,\ldots\}.

The inverse Gaussian measure

As is well known, the Gaussian probability measure

dγ∞(x)=(2π)−n2(det⁡Q∞)−12exp⁡(−12⟨Q∞−1x,x⟩) dxd\gamma_{\infty}(x)=(2\pi)^{-\frac{n}{2}}(\det Q_{\infty})^{-\frac{1}{2}}\exp\left(-\frac{1}{2}\langle Q_{\infty}^{-1}x,x\rangle\right)\,dx

is an invariant (probability) measure for the semigroup generated by LQ,B\mathcal{L}^{Q,B}, that is,

∫RnHtOUf(x) dγ∞(x)=∫Rnf(x) dγ∞(x)for all t>0 and f∈L1(γ∞).\int_{\mathbb{R}^{n}}\mathcal{H}^{OU}_{t}f(x)\,d\gamma_{\infty}(x)=\int_{\mathbb{R}^{n}}f(x)\,d\gamma_{\infty}(x)\qquad\text{for all }t>0\text{ and }f\in L^{1}(\gamma_{\infty}).

In fact, in [16 Section 11.2.3 page 313] it is proved that the existence of an invariant probability measure for the semigroup generated by an operator LQ,B\mathcal{L}^{Q,B} (for an arbitrary BB) defined as in (1.3) is equivalent to the spectrum of the matrix BB being contained in the open left half-plane (we refer also to [22] or [20]). Since the spectrum of −B-B is contained in the open right half-plane, HtOU\mathcal{H}^{OU}_{t} has no invariant probability measure.

We shall see in Proposition 3.1, however, that dγ−∞d\gamma_{-\infty} can be seen as an invariant measure for HtOU\mathcal{H}^{OU}_{t}, despite not being finite. An additional reason for which dγ−∞d\gamma_{-\infty} is a natural measure in the inverse Ornstein–Uhlenbeck setting is given by the following lemma.

Lemma 2.1. The semigroup etAQ,Be^{tA_{Q,B}} is symmetric on L2(γ−∞)L^{2}(\gamma_{-\infty}) if and only if QB∗=BQQB^{*}=BQ.

Proof. It has been observed in [6 Remark 2.3] that the semigroup etAQ,Be^{tA_{Q,B}} is symmetric on L2(γ−∞)L^{2}(\gamma_{-\infty}) if and only if etLQ,Be^{t\mathcal{L}^{Q,B}} is symmetric on L2(γ∞)L^{2}(\gamma_{\infty}). Since the latter property holds if and only if QB∗=BQQB^{*}=BQ by [15 Theorem 2.4] (see also [25 Lemma 2.1]), the lemma follows. □

We recall from [6] the following facts. To simplify the notation, let us first set

cn=(2π)−n2(det⁡Q∞)−12.c_{n}=(2\pi)^{-\frac{n}{2}}(\det Q_{\infty})^{-\frac{1}{2}}.

Denoting by R(x)R(x) the quadratic form

R(x)=12⟨Q∞−1x,x⟩,x∈Rn,R(x)=\frac{1}{2}\langle Q_{\infty}^{-1}x,x\rangle,\qquad x\in\mathbb{R}^{n},

we have dγ∞(x)=cne−R(x) dxd\gamma_{\infty}(x)=c_{n}e^{-R(x)}\,dx, as well as dγ−∞(x)=eR(x) dxd\gamma_{-\infty}(x)=e^{R(x)}\,dx, and thus

dγ∞(x)=cne−2R(x)dγ−∞(x).(4)d\gamma_{\infty}(x)=c_{n}e^{-2R(x)}d\gamma_{-\infty}(x). \tag*{(4)}

(1) For each f∈L1(γ−∞)f \in L^1(\gamma_{-\infty}) and all t>0t>0 one has

etAQ,Bf(x)=∫KtUO(x,u)f(u) dγ−∞(u),x∈Rn,e^{tA^{Q,B}}f(x)=\int K_t^{UO}(x,u)f(u)\,d\gamma_{-\infty}(u),\qquad x\in\mathbb{R}^n,

where for x,u∈Rnx,u\in\mathbb{R}^n and t>0t>0 the inverse Mehler kernel KtUOK_t^{UO} (with respect to dγ−∞d\gamma_{-\infty}) is given by

KtUO(x,u)=(2π)−n2(det⁡Qt)−12K_t^{UO}(x,u)=(2\pi)^{-\frac n2}(\det Q_t)^{-\frac12}
×ettr⁡Be−R(x)exp⁡[−12⟨(Qt−1−Q∞−1)(x−Dtu),x−Dtu⟩].{}\times e^{t\operatorname{tr}B}e^{-R(x)}\exp\left[-\frac12\left\langle(Q_t^{-1}-Q_\infty^{-1})(x-D_tu),x-D_tu\right\rangle\right].

(see [6 Proposition 2.2] for a proof). Here DtD_t denotes a one-parameter group of matrices, given by

Dt:=Q∞e−tB∗Q∞−1,t∈R.D_t:=Q_\infty e^{-tB^*}Q_\infty^{-1},\qquad t\in\mathbb{R}.

We also recall [8 formula (4.3)], that is,

∂∂tR(Dtx)≃∣Dtx∣2,(5)\frac{\partial}{\partial t}R(D_tx)\simeq|D_tx|^2, \tag*{(5)}

for all xx in Rn\mathbb{R}^n and t∈Rt\in\mathbb{R}. This implies that for ∣t∣<1|t|<1

∣Dtx−x∣≃∣t∣∣x∣.(6)|D_tx-x|\simeq|t||x|. \tag*{(6)}

(2) Turning back to the Ornstein–Uhlenbeck setting, one has for each f∈L1(γ∞)f\in L^1(\gamma_\infty) and all t>0t>0

etLQ,Bf(x)=∫KtOU(x,u)f(u) dγ∞(u),x∈Rn,e^{tL^{Q,B}}f(x)=\int K_t^{OU}(x,u)f(u)\,d\gamma_\infty(u),\qquad x\in\mathbb{R}^n,

where for t>0t>0 the Mehler kernel KtOUK_t^{OU} (with respect to dγ∞d\gamma_\infty) is given by

KtOU(x,u)=(det⁡Q∞det⁡Qt)12eR(x)exp⁡[−12⟨(Qt−1−Q∞−1)(u−Dtx),u−Dtx⟩](7)K_t^{OU}(x,u)=\left(\frac{\det Q_\infty}{\det Q_t}\right)^{\frac12}e^{R(x)} \exp\left[-\frac12\left\langle(Q_t^{-1}-Q_\infty^{-1})(u-D_tx),u-D_tx\right\rangle\right] \tag*{(7)}

for x,u∈Rnx,u\in\mathbb{R}^n (see [8 (2.6)]).

(3) The relationship between the integral kernels KtOUK_t^{OU} and KtUOK_t^{UO} is given by

KtUO(x,u)=cne−R(x)e−R(u)ettr⁡BKtOU(u,x),(8)K_t^{UO}(x,u)=c_ne^{-R(x)}e^{-R(u)}e^{t\operatorname{tr}B}K_t^{OU}(u,x), \tag*{(8)}

holding for all t>0t>0 and x,u∈Rnx,u\in\mathbb{R}^n (see [6 (2.5)]).

(4) In [6 Proposition 2.5] we proved the following Kolmogorov-type formula in the inverse Gaussian setting. For all f∈Cb(Rn)f\in C_b(\mathbb{R}^n) and x∈Rnx\in\mathbb{R}^n one has

HtUOf(x)=(2π)−n2(det⁡Qt)−12∫Rnf(e−tB(u+x))e−12⟨Qt−1u,u⟩ du.(9)\mathcal{H}_t^{UO}f(x)=(2\pi)^{-\frac n2}(\det Q_t)^{-\frac12}\int_{\mathbb{R}^n}f(e^{-tB}(u+x))e^{-\frac12\langle Q_t^{-1}u,u\rangle}\,du. \tag*{(9)}

(5) The semigroup (HtUO)t>0(\mathcal{H}_t^{UO})_{t>0} is conservative, that is,

HtUO1=1for all t>0,(10)\mathcal{H}_t^{UO}1=1\qquad\text{for all }t>0, \tag*{(10)}

where 11 denotes the constant function.

If 0<t≤10<t\le1, one has

e−R(x)tn/2exp⁡(−C∣D−tx−u∣2t)≲KtUO(x,u)≲e−R(x)tn/2exp⁡(−c∣D−tx−u∣2t)(11)\frac{e^{-R(x)}}{t^{n/2}}\exp\left(-C\frac{\lvert D_{-t}x-u\rvert^2}{t}\right)\lesssim K_t^{UO}(x,u)\lesssim\frac{e^{-R(x)}}{t^{n/2}}\exp\left(-c\frac{\lvert D_{-t}x-u\rvert^2}{t}\right) \tag*{(11)}

for all (x,u)∈Rn×Rn(x,u)\in\mathbb{R}^n\times\mathbb{R}^n (see [6 Proposition 2.4]).

On the continuity and contractivity of (HtUO)t>0(\mathcal{H}^{UO}_t)_{t>0}

We start by considering some boundedness properties of (HtUO)t>0(\mathcal{H}^{UO}_t)_{t>0}.

Proposition 3.1. The measure γ−∞\gamma_{-\infty} is an invariant (infinite) measure for HtUO\mathcal{H}^{UO}_t, in the sense that

∫RnHtUOf dγ−∞=∫Rnf dγ−∞\int_{\mathbb{R}^n}\mathcal{H}^{UO}_t f\,d\gamma_{-\infty}=\int_{\mathbb{R}^n}f\,d\gamma_{-\infty}

for all f∈L1(γ−∞)f\in L^1(\gamma_{-\infty}) and t>0t>0. Moreover, for each t>0t>0 and for all 1≤p≤∞1\le p\le\infty the operator HtUO\mathcal{H}^{UO}_t is a contraction on Lp(γ−∞)L^p(\gamma_{-\infty}).

Proof. Let f∈L1(γ−∞)f\in L^1(\gamma_{-\infty}). We may assume, without loss of generality, that ff is nonnegative. Then

∫RnHtUOf(x) dγ−∞(x)=∫Rn∫RnKtUO(x,u)f(u) dγ−∞(u) dγ−∞(x)=∫Rn∫RnKtUO(x,u) dγ−∞(x)f(u) dγ−∞(u).\begin{aligned} \int_{\mathbb{R}^n}\mathcal{H}^{UO}_t f(x)\,d\gamma_{-\infty}(x)=\int_{\mathbb{R}^n}\int_{\mathbb{R}^n}K^{UO}_t(x,u)f(u)\,d\gamma_{-\infty}(u)\,d\gamma_{-\infty}(x) \\ &=\int_{\mathbb{R}^n}\int_{\mathbb{R}^n}K^{UO}_t(x,u)\,d\gamma_{-\infty}(x)f(u)\,d\gamma_{-\infty}(u). \end{aligned}

Setting

It(u):=∫RnKtUO(x,u) dγ−∞(x),I_t(u):=\int_{\mathbb{R}^n}K^{UO}_t(x,u)\,d\gamma_{-\infty}(x),

it will be enough to verify that

It(u)=1(12)I_t(u)=1 \tag*{(12)}

for all u∈Rnu\in\mathbb{R}^n.

In the light of (2.5), (1.1) and (2.4), we observe that

It(u)=cn∫Rne−R(x)e−R(u)ettr⁡BKtUO(u,x) dγ−∞(x)=cn∫Rne−R(u)ettr⁡BKtUO(u,x) dx(13)\begin{aligned} I_t(u)&=c_n\int_{\mathbb{R}^n}e^{-R(x)}e^{-R(u)}e^{t\operatorname{tr}B}K^{UO}_t(u,x)\,d\gamma_{-\infty}(x)\\ &=c_n\int_{\mathbb{R}^n}e^{-R(u)}e^{t\operatorname{tr}B}K^{UO}_t(u,x)\,dx \tag*{(13)} \end{aligned}
=(2π)−n2ettr⁡B(det⁡Qt)−12∫Rnexp⁡[−12⟨(Qt−1−Q∞−1)(x−Dtu),x−Dtu⟩] dx=(2π)−n2ettr⁡B(det⁡Qt)−12∫Rnexp⁡[−12⟨(Qt−1−Q∞−1)x,x⟩] dx=(2π)−n2ettr⁡B(det⁡Qt)−12det⁡(Qt−1−Q∞−1)−12∫Rnexp⁡[−12⟨x′,x′⟩] dx′.(14)\begin{aligned} &=(2\pi)^{-\frac n2}e^{t\operatorname{tr}B}(\det Q_t)^{-\frac12}\int_{\mathbb{R}^n}\exp\left[-\frac12\left\langle\left(Q_t^{-1}-Q_\infty^{-1}\right)(x-D_tu),x-D_tu\right\rangle\right]\,dx\\ &=(2\pi)^{-\frac n2}e^{t\operatorname{tr}B}(\det Q_t)^{-\frac12}\int_{\mathbb{R}^n}\exp\left[-\frac12\left\langle\left(Q_t^{-1}-Q_\infty^{-1}\right)x,x\right\rangle\right]\,dx\\ &=(2\pi)^{-\frac n2}e^{t\operatorname{tr}B}(\det Q_t)^{-\frac12}\det\left(Q_t^{-1}-Q_\infty^{-1}\right)^{-\frac12}\int_{\mathbb{R}^n}\exp\left[-\frac12\langle x',x'\rangle\right]\,dx'. \tag*{(14)} \end{aligned}

We know from [8 formulae (2.5) and (2.2)] that

Q∞−Qt=etBQ∞etB∗Q_\infty-Q_t=e^{tB}Q_\infty e^{tB^*}

(see also [25 formula (2.1)]), and

Qt−1−Q∞−1=Qt−1(Q∞−Qt)Q∞−1.Q_t^{-1}-Q_\infty^{-1}=Q_t^{-1}(Q_\infty-Q_t)Q_\infty^{-1}.

It follows that

Qt−1−Q∞−1=Qt−1etBQ∞etB∗Q∞−1,Q_t^{-1}-Q_\infty^{-1}=Q_t^{-1}e^{tB}Q_\infty e^{tB^*}Q_\infty^{-1},

whence

det⁡(Qt−1−Q∞−1)=(det⁡Qt)−1e2ttr⁡B.\det\left(Q_t^{-1}-Q_\infty^{-1}\right)=(\det Q_t)^{-1}e^{2t\operatorname{tr}B}.

Inserting this in (3.3), we finally have

It(u)=(2π)−n2∫Rnexp⁡[−12⟨x′,x′⟩] dx′=1.I_t(u)=(2\pi)^{-\frac{n}{2}}\int_{\mathbb{R}^n}\exp\left[-\frac{1}{2}\langle x',x'\rangle\right]\,dx'=1.

This proves (3.1) and also that γ−∞\gamma_{-\infty} is an invariant measure for HtUO\mathcal{H}^{UO}_t. Since ∣HtUOf∣≤HtUO∣f∣|\mathcal{H}^{UO}_t f|\leq\mathcal{H}^{UO}_t|f|, it also follows that each HtUO\mathcal{H}^{UO}_t is a contraction on L1(γ−∞)L^1(\gamma_{-\infty}). From (2.7) we see that HtUO\mathcal{H}^{UO}_t is a contraction also on L∞(γ−∞)L^\infty(\gamma_{-\infty}), and then by interpolation on Lp(γ−∞)L^p(\gamma_{-\infty}) for all 1≤p≤∞1\leq p\leq\infty. □

From the proof of Proposition 3.1 we can also deduce the following formula, which may be of independent interest.

Corollary 3.2. For all t>0t>0 and x∈Rnx\in\mathbb{R}^n one has

HtUO(eR(⋅))(x)=e−ttr⁡BeR(x).\mathcal{H}^{UO}_t\left(e^{R(\cdot)}\right)(x)=e^{-t\operatorname{tr}B}e^{R(x)}.

Proof. From (3.1) and (3.2) we get

1=It(u)=cnettr⁡Be−R(u)∫RnKtUO(u,x) dx=ettr⁡Be−R(u)∫RnKtUO(u,x)eR(x) dγ∞(x)=ettr⁡Be−R(u)HtUO(eR)(u).\begin{aligned} 1=I_t(u)=c_n e^{t\operatorname{tr}B}e^{-R(u)}\int_{\mathbb{R}^n}K_t^{UO}(u,x)\,dx \\ &=e^{t\operatorname{tr}B}e^{-R(u)}\int_{\mathbb{R}^n}K_t^{UO}(u,x)e^{R(x)}\,d\gamma_\infty(x) \\ &=e^{t\operatorname{tr}B}e^{-R(u)}\mathcal{H}^{UO}_t(e^R)(u). \end{aligned}

The corollary is proved. □

Proposition 3.3. Suppose p∈[1,∞)p\in[1,\infty). Then (HtUO)t>0(\mathcal{H}^{UO}_t)_{t>0} is strongly continuous on Lp(γ−∞)L^p(\gamma_{-\infty}).

Proof. It is known from Proposition 3.1 that HtUO\mathcal{H}^{UO}_t is bounded on Lp(γ−∞)L^p(\gamma_{-\infty}). Suppose now f∈C∞(Rn)f\in C^\infty(\mathbb{R}^n). To prove that ∥HtUOf−f∥Lp(γ−∞)→0\|\mathcal{H}^{UO}_t f-f\|_{L^p(\gamma_{-\infty})}\to0 as t→0+t\to0^+, we shall apply the Dominated Convergence Theorem. Since ff is continuous and bounded, HtUOf(x)\mathcal{H}^{UO}_t f(x) tends to f(x)f(x) pointwise as t→0+t\to0^+. Therefore, it suffices to verify that there exists δ∈(0,1)\delta\in(0,1) such that the dominating function

sup⁡0<t<δ∣HtUOf(x)−f(x)∣\sup_{0<t<\delta}\left|\mathcal{H}^{UO}_t f(x)-f(x)\right|

belongs to Lp(γ−∞)L^p(\gamma_{-\infty}) for 1≤p<∞1\leq p<\infty. If p>1p>1, we can bound this supremum by

sup⁡t>0∣HtUOf(x)∣+∣f(x)∣=H∗UOf(x)+∣f(x)∣,\sup_{t>0}|\mathcal{H}^{UO}_t f(x)|+|f(x)|=\mathcal{H}^{UO}_*f(x)+|f(x)|,

where H∗UO\mathcal{H}^{\mathrm{UO}}_{*} denotes the inverse Ornstein–Uhlenbeck maximal operator. Since H∗UO\mathcal{H}^{\mathrm{UO}}_{*} is bounded on Lp(γ−∞)L^{p}(\gamma_{-\infty}) for all 1<p≤∞1 < p \leq\infty (see [6 Theorem 3.1]), and f∈Cc∞(Rn)⊂Lp(γ−∞)f \in C_{c}^{\infty}(\mathbb{R}^{n}) \subset L^{p}(\gamma_{-\infty}), the dominating function belongs to Lp(γ−∞)L^{p}(\gamma_{-\infty}).

Assume now that p=1p = 1. Using (11), (10) and the fact that ∣D−tx−u∣≃∣x−Dtu∣\lvert D_{-t}x-u\rvert\simeq\lvert x-D_tu\rvert, one has

∣HtUOf(x)−f(x)∣≲e−R(x)t−n/2∫Rnexp⁡(−c∣x−Dtu∣2t)∣f(u)−f(x)∣eR(u) du.\left|\mathcal{H}^{\mathrm{UO}}_t f(x)-f(x)\right| \lesssim e^{-R(x)}t^{-n/2}\int_{\mathbb{R}^n}\exp\left(-c\frac{\lvert x-D_tu\rvert^2}{t}\right)\lvert f(u)-f(x)\rvert e^{R(u)}\,du.

for 0<t<10 < t < 1. Here we can estimate, applying (6) for the last step,

∣x−Dtu∣2=∣x−u+u−Dtu∣2≥∣x−u∣2−2∣x−u∣∣u−Dtu∣≥∣x−u∣2−C∣x−u∣∣u∣t.\lvert x-D_tu\rvert^2=\lvert x-u+u-D_tu\rvert^2\geq\lvert x-u\rvert^2-2\lvert x-u\rvert\lvert u-D_tu\rvert\geq\lvert x-u\rvert^2-C\lvert x-u\rvert\lvert u\rvert t.

First, let us take ∣x∣>1+2ρ\lvert x\rvert>1+2\rho, where ρ=sup⁡{∣u∣:u∈supp⁡f}\rho=\sup\{\lvert u\rvert:u\in\operatorname{supp}f\}. Then f(x)=0f(x)=0, it suffices to integrate over u∈supp⁡fu\in\operatorname{supp}f, and we have 1<∣x−u∣≃∣x∣1<\lvert x-u\rvert\simeq\lvert x\rvert. If we choose δ\delta and thus tt small enough, we obtain

t−n/2exp⁡(−c∣x−Dtu∣2t)≤t−n/2exp⁡(−c∣x−u∣2t+C∣x−u∣∣u∣)≲exp⁡(−c∣x∣2),t^{-n/2}\exp\left(-c\frac{\lvert x-D_tu\rvert^2}{t}\right)\leq t^{-n/2}\exp\left(-c\frac{\lvert x-u\rvert^2}{t}+C\lvert x-u\rvert\lvert u\rvert\right)\lesssim\exp(-c\lvert x\rvert^2),

and therefore

sup⁡0<t<δ∣HtUOf(x)−f(x)∣≲e−R(x)exp⁡(−c∣x∣2)∫Rn∣f(u)∣eR(u) du,\sup_{0<t<\delta}\left|\mathcal{H}^{\mathrm{UO}}_t f(x)-f(x)\right|\lesssim e^{-R(x)}\exp(-c\lvert x\rvert^2)\int_{\mathbb{R}^n}\lvert f(u)\rvert e^{R(u)}\,du,

which is clearly in L1(γ−∞)L^1(\gamma_{-\infty}).

Next, let ∣x∣<1+2ρ\lvert x\rvert<1+2\rho and consider the integral restricted to ∣u∣<2+4ρ\lvert u\rvert<2+4\rho. Then this part of ∣HtUOf(x)−f(x)∣\left|\mathcal{H}^{\mathrm{UO}}_t f(x)-f(x)\right| is at most a constant times

t−n/2∫∣u∣<2+4ρexp⁡(−c∣x−u∣2t+C)∣f(u)−f(x)∣ du≲∫∣u∣<2+4ρ∣x−u∣1−n du≤C,t^{-n/2}\int_{\lvert u\rvert<2+4\rho}\exp\left(-c\frac{\lvert x-u\rvert^2}{t}+C\right)\lvert f(u)-f(x)\rvert\,du\lesssim\int_{\lvert u\rvert<2+4\rho}\lvert x-u\rvert^{1-n}\,du\leq C,

and the supremum in tt is thus integrable over ∣x∣<1+2ρ\lvert x\rvert<1+2\rho.

Finally, consider the case where ∣x∣<1+2ρ\lvert x\rvert<1+2\rho and ∣u∣>2+4ρ\lvert u\rvert>2+4\rho, so that ∣x−u∣≃∣u∣\lvert x-u\rvert\simeq\lvert u\rvert. Then

t−n/2exp⁡(−c∣x−Dtu∣2t)≤t−n/2exp⁡(−c∣x−u∣2t+C∣x−u∣∣u∣).t^{-n/2}\exp\left(-c\frac{\lvert x-D_tu\rvert^2}{t}\right)\leq t^{-n/2}\exp\left(-c\frac{\lvert x-u\rvert^2}{t}+C\lvert x-u\rvert\lvert u\rvert\right).

With δ\delta small enough, this will be at most Cexp⁡(−c∣u∣2/t)C\exp(-c\lvert u\rvert^2/t). Moreover, since ff is bounded, we can achieve

∫∣u∣>2+4ρexp⁡(−c∣u∣2t)∣f(x)∣eR(u) du≤Csup⁡∣f∣.\int_{\lvert u\rvert>2+4\rho}\exp\left(-c\frac{\lvert u\rvert^2}{t}\right)\lvert f(x)\rvert e^{R(u)}\,du\leq C\sup\lvert f\rvert.

The last quantity is thus a uniform bound for the corresponding part of ∣HtUOf(x)−f(x)∣\left|\mathcal{H}^{\mathrm{UO}}_t f(x)-f(x)\right|, which ends the argument for p=1p=1.

By the Dominated Convergence Theorem, ∥HtUOf−f∥Lp(γ−∞)→0\|\mathcal{H}^{\mathrm{UO}}_t f-f\|_{L^p(\gamma_{-\infty})}\to0 for all 1≤p<∞1\leq p<\infty and f∈Cc∞(Rn)f\in C_c^\infty(\mathbb{R}^n). The proposition now follows from this dense convergence, together with the Lp(γ−∞)L^p(\gamma_{-\infty}) contractivity of HUO\mathcal{H}^{\mathrm{UO}}, via a standard density argument. □

The spectrum of AQ,BA^{Q,B}

By Proposition 3.3, for 1≤p<∞1 \le p < \infty the semigroup (HtUO)t>0(\mathcal{H}^{UO}_t)_{t>0} has an infinitesimal generator ApQ,BA^{Q,B}_p on Lp(γ−∞)L^p(\gamma_{-\infty}). We write σp(AQ,B)\sigma_p(A^{Q,B}) for its spectrum on Lp(γ−∞)L^p(\gamma_{-\infty}), and generally omit the subscript pp when the underlying space is clear.

We start by proving a lemma.

Lemma 4.1. The eigenvalues of the matrix M=−(B+QQ∞−1)M=-(B+QQ_\infty^{-1}) have negative real parts, tr M=tr⁡BM=\operatorname{tr} B, and

∫0∞esMQesM∗ ds=Q∞.(15)\int_0^\infty e^{sM}Qe^{sM^*}\,ds=Q_\infty. \tag*{(15)}

Notice that this means that QQ and MM satisfy the assumptions (H1) and (H2). Thus we can associate an Ornstein–Uhlenbeck theory with the operator LQ,M\mathcal{L}^{Q,M}, and (4.1) implies that the corresponding invariant measure γ∞\gamma_\infty is the same as that for LQ,B\mathcal{L}^{Q,B}.

Proof. Let us recall from [25 Lemma 2.1] that

BQ∞+Q=−Q∞B∗,(16)BQ_\infty+Q=-Q_\infty B^*, \tag*{(16)}

which yields

M=−(B+QQ∞−1)=Q∞B∗Q∞−1.(17)M=-(B+QQ_\infty^{-1})=Q_\infty B^*Q_\infty^{-1}. \tag*{(17)}

Equation (4.3) says that MM is similar to B∗B^*. Therefore, the eigenvalues of MM are the complex conjugates of the eigenvalues of BB, and thus the same as those of BB, with negative real parts. Moreover, since tr⁡(XY)=tr⁡(YX)\operatorname{tr}(XY)=\operatorname{tr}(YX) and BB is a real matrix, it immediately follows that tr⁡M=tr⁡B∗=tr⁡B\operatorname{tr} M=\operatorname{tr} B^*=\operatorname{tr} B.

By (4.3) moreover

M∗=Q∞−1BQ∞,M^*=Q_\infty^{-1}BQ_\infty,

and

esM=Q∞esB∗Q∞−1andesM∗=Q∞−1esBQ∞.e^{sM}=Q_\infty e^{sB^*}Q_\infty^{-1} \qquad\text{and}\qquad e^{sM^*}=Q_\infty^{-1}e^{sB}Q_\infty.

Therefore,

∫0∞esMQesM∗ ds=∫0∞(Q∞esB∗Q∞−1)Q(Q∞−1esBQ∞) ds=Q∞(∫0∞esB∗(Q∞−1QQ∞−1)esB ds)Q∞.\begin{aligned} \int_0^\infty e^{sM}Qe^{sM^*}\,ds = \int_0^\infty \left(Q_\infty e^{sB^*}Q_\infty^{-1}\right) Q \left(Q_\infty^{-1}e^{sB}Q_\infty\right)\,ds \\ &= Q_\infty \left( \int_0^\infty e^{sB^*} \left(Q_\infty^{-1}QQ_\infty^{-1}\right) e^{sB}\,ds \right) Q_\infty. \end{aligned}

By (4.2) we have

Q∞−1QQ∞−1=Q∞−1(−BQ∞−Q∞B∗)Q∞−1=−Q∞−1B−B∗Q∞−1.Q_\infty^{-1}QQ_\infty^{-1} = Q_\infty^{-1}(-BQ_\infty-Q_\infty B^*)Q_\infty^{-1} = -Q_\infty^{-1}B-B^*Q_\infty^{-1}.

Thus,

∫0∞esMQesM∗ ds=Q∞(∫0∞esB∗(−Q∞−1B−B∗Q∞−1)esB ds)Q∞.\int_0^\infty e^{sM}Qe^{sM^*}\,ds = Q_\infty \left( \int_0^\infty e^{sB^*} \left(-Q_\infty^{-1}B-B^*Q_\infty^{-1}\right) e^{sB}\,ds \right) Q_\infty.

It remains to observe that the integrand here is the derivative of the function s↦−esB∗Q∞−1esBs\mapsto-e^{sB^*}Q_\infty^{-1}e^{sB}, so that

∫0∞esMQesM∗ ds=Q∞[−esB∗Q∞−1esB]0∞Q∞=Q∞,\int_0^\infty e^{sM}Qe^{sM^*}\,ds = Q_\infty \left[-e^{sB^*}Q_\infty^{-1}e^{sB}\right]_0^\infty Q_\infty = Q_\infty,

and the proof is complete. □

Lemma 4.2. Let U:L2(γ∞)→L2(γ−∞)U:L^2(\gamma_{\infty})\to L^2(\gamma_{-\infty}) be the isometric isomorphism given by multiplication by the factor cn1/2e−Rc_n^{1/2}e^{-R}. For f∈L2(γ−∞)f\in L^2(\gamma_{-\infty}), one has

HtUO=ettr⁡BU(HtOU)∗U−1,t>0.\mathcal{H}^{\mathrm{UO}}_t=e^{t\operatorname{tr}B}U\left(\mathcal{H}^{\mathrm{OU}}_t\right)^*U^{-1},\qquad t>0.

Here (HtOU)∗\left(\mathcal{H}^{\mathrm{OU}}_t\right)^* denotes the adjoint of HtOU\mathcal{H}^{\mathrm{OU}}_t in L2(γ∞)L^2(\gamma_{\infty}).

Proof. By (2.5) we have that

KtUO(x,u)=cne−R(u)e−R(x)ettr⁡BKtOU(u,x).K^{\mathrm{UO}}_t(x,u)=c_n e^{-R(u)}e^{-R(x)}e^{t\operatorname{tr}B}K^{\mathrm{OU}}_t(u,x).

Thus

HtUOf(x)=∫RnKtUO(x,u)f(u) dγ−∞(u)=cnettr⁡Be−R(x)∫Rne−R(u)KtOU(u,x)f(u) dγ−∞(u)=ettr⁡Be−R(x)∫RnKtOU(u,x)f(u)eR(u) dγ∞(u),\begin{aligned} \mathcal{H}^{\mathrm{UO}}_t f(x)=\int_{\mathbb{R}^n}K^{\mathrm{UO}}_t(x,u)f(u)\,d\gamma_{-\infty}(u) \\ &=c_n e^{t\operatorname{tr}B}e^{-R(x)}\int_{\mathbb{R}^n}e^{-R(u)}K^{\mathrm{OU}}_t(u,x)f(u)\,d\gamma_{-\infty}(u) \\ &=e^{t\operatorname{tr}B}e^{-R(x)}\int_{\mathbb{R}^n}K^{\mathrm{OU}}_t(u,x)f(u)e^{R(u)}\,d\gamma_{\infty}(u), \end{aligned}

where we used (2.1). Since the kernel of the adjoint of the semigroup operator HtOU\mathcal{H}^{\mathrm{OU}}_t with respect to the measure γ∞\gamma_{\infty} is (KtOU)∗(x,u)=KtOU(u,x)\left(K^{\mathrm{OU}}_t\right)^*(x,u)=K^{\mathrm{OU}}_t(u,x), we conclude that

HtUOf(x)=ettr⁡Be−R(x)∫Rn(KtOU)∗(x,u)f(u)eR(u) dγ∞(u)=ettr⁡BU(HtOU)∗U−1f(x),\begin{aligned} \mathcal{H}^{\mathrm{UO}}_t f(x)=e^{t\operatorname{tr}B}e^{-R(x)}\int_{\mathbb{R}^n}\left(K^{\mathrm{OU}}_t\right)^*(x,u)f(u)e^{R(u)}\,d\gamma_{\infty}(u) \\ &=e^{t\operatorname{tr}B}U\left(\mathcal{H}^{\mathrm{OU}}_t\right)^*U^{-1}f(x), \end{aligned}

which completes the proof. □

Remark 4.3. It follows from (4.3) that

tr⁡(QQ∞−1)=−(tr⁡M+tr⁡B)=−2tr⁡B.\operatorname{tr}(QQ_{\infty}^{-1})=-(\operatorname{tr}M+\operatorname{tr}B)=-2\operatorname{tr}B.

The main result of this section is the following.

Theorem 4.4. (1) The spectrum σ2(AQ,B)\sigma_2(A^{Q,B}) of AQ,BA^{Q,B} in L2(γ−∞)L^2(\gamma_{-\infty}) consists only of eigenvalues given by

μ=tr⁡B−∑j=1nmjρj,\mu=\operatorname{tr}B-\sum_{j=1}^{n}m_j\rho_j,

where ρ1,…,ρn\rho_1,\ldots,\rho_n are the eigenvalues of the matrix B+QQ∞−1B+QQ_{\infty}^{-1} (counted with algebraic multiplicity) and the numbers mjm_j are any non-negative integers. In particular, AQ,BA^{Q,B} has a spectral gap. Each of the corresponding eigenfunctions is of the form e−R(x)P(x)e^{-R(x)}P(x), where PP is a polynomial.

(2) μ∈σ2(AQ,B)\mu\in\sigma_2(A^{Q,B}) if and only if there exists a nonzero homogeneous polynomial PP such that

⟨(B+QQ∞−1)x,∇P⟩=(tr⁡B−μ)P.\left\langle(B+QQ_{\infty}^{-1})x,\nabla P\right\rangle=(\operatorname{tr}B-\mu)P.

The proof of this theorem will be reduced to the study of an appropriate Ornstein–Uhlenbeck operator, whose spectrum is known. We shall use the following statement, which summarizes Proposition 3.1 and Lemma 3.2 in [24].

Proposition 4.5. Let MM be the matrix of Lemma 4.1, and consider LQ,M\mathcal{L}^{Q,M} on L2(γ∞)L^2(\gamma_\infty). Let μ1,…,μn\mu_1,\ldots,\mu_n be the eigenvalues of MM, counted with multiplicity. Then

(1) LQ,M\mathcal{L}^{Q,M} has a pure point spectrum. Its eigenfunctions are polynomials; the eigenvalues are given by the numbers ∑j=1nmjμj\sum_{j=1}^n m_j\mu_j, where the mjm_j are nonnegative integers.

(2) A complex number λ\lambda is an eigenvalue of LQ,M\mathcal{L}^{Q,M} if then and only if there is a nonzero homogeneous polynomial PP satisfying

⟨Mx,∇P⟩=λP.\langle Mx,\nabla P\rangle=\lambda P.

Before the proof of Theorem 4.4, we make some computations. We start by noticing that [8 (4.1)] implies

2⟨Q∞−1x,Bx⟩=2⟨B∗Q∞−1x,x⟩=−∣Q1/2Q∞−1x∣2,(18)2\langle Q_{\infty}^{-1}x,Bx\rangle=2\langle B^{*}Q_{\infty}^{-1}x,x\rangle=-\lvert Q^{1/2}Q_{\infty}^{-1}x\rvert^2, \tag*{(18)}

Moreover, we have

∇γ∞(x)=cn∇e−12⟨Q∞−1x,x⟩=−cnQ∞−1xe−12⟨Q∞−1x,x⟩=−γ∞(x)Q∞−1x(19)\nabla\gamma_\infty(x)=c_n\nabla e^{-\frac{1}{2}\langle Q_{\infty}^{-1}x,x\rangle}=-c_nQ_{\infty}^{-1}xe^{-\frac{1}{2}\langle Q_{\infty}^{-1}x,x\rangle}=-\gamma_\infty(x)Q_{\infty}^{-1}x \tag*{(19)}

and because of (4.1) and (4.4)

tr⁡(Q∇2γ∞)(x)=−tr⁡(QQ∞−1)γ∞(x)+⟨QQ∞−1x,Q∞−1x⟩γ∞(x)=(−tr⁡(QQ∞−1)+∣Q1/2Q∞−1x∣2)γ∞(x).\begin{aligned} \operatorname{tr}(Q\nabla^2\gamma_\infty)(x)=-\operatorname{tr}(QQ_{\infty}^{-1})\gamma_\infty(x)+\langle QQ_{\infty}^{-1}x,Q_{\infty}^{-1}x\rangle\gamma_\infty(x) \\ &=\left(-\operatorname{tr}(QQ_{\infty}^{-1})+\lvert Q^{1/2}Q_{\infty}^{-1}x\rvert^2\right)\gamma_\infty(x). \end{aligned}

Combining the last two identities and applying again (4.4), we can compute the action of AQ,B\mathcal{A}^{Q,B} on γ∞\gamma_\infty, that is,

AQ,Bγ∞(x)=12tr⁡(Q∇2γ∞)(x)−⟨Bx,∇γ∞(x)⟩=12(∣Q1/2Q∞−1x∣2−tr⁡(QQ∞−1))γ∞(x)+⟨Bx,Q∞−1x⟩γ∞(x)=−12tr⁡(QQ∞−1)γ∞(x).\begin{aligned} \mathcal{A}^{Q,B}\gamma_\infty(x)=\frac{1}{2}\operatorname{tr}(Q\nabla^2\gamma_\infty)(x)-\langle Bx,\nabla\gamma_\infty(x)\rangle\\ &=\frac{1}{2}\left(\lvert Q^{1/2}Q_{\infty}^{-1}x\rvert^2-\operatorname{tr}(QQ_{\infty}^{-1})\right)\gamma_\infty(x)+\langle Bx,Q_{\infty}^{-1}x\rangle\gamma_\infty(x) \\ &=-\frac{1}{2}\operatorname{tr}(QQ_{\infty}^{-1})\gamma_\infty(x). \end{aligned}

Recalling that tr⁡(QQ∞−1)=−2tr⁡B\operatorname{tr}(QQ_{\infty}^{-1})=-2\operatorname{tr}B (see Remark 4.3), we immediately conclude that

AQ,Bγ∞=tr⁡B γ∞.(20)\mathcal{A}^{Q,B}\gamma_\infty=\operatorname{tr}B\,\gamma_\infty. \tag*{(20)}

Proof of Theorem 4.4. The formulae (4.6) and (4.5) yield

AQ,B(γ∞f)=γ∞AQ,Bf+fAQ,Bγ∞+⟨Q∇γ∞,∇f⟩=γ∞AQ,Bf+tr⁡Bγ∞f−γ∞⟨QQ∞−1x,∇f⟩=12γ∞tr⁡(Q∇2f)+tr⁡Bγ∞f−γ∞⟨(B+QQ∞−1)x,∇f⟩,\begin{aligned} \mathcal{A}^{Q,B}(\gamma_\infty f)=\gamma_\infty\mathcal{A}^{Q,B}f+f\mathcal{A}^{Q,B}\gamma_\infty+\langle Q\nabla\gamma_\infty,\nabla f\rangle\\ &=\gamma_\infty\mathcal{A}^{Q,B}f+\operatorname{tr}B\gamma_\infty f-\gamma_\infty\langle QQ_{\infty}^{-1}x,\nabla f\rangle\\ &=\frac{1}{2}\gamma_\infty\operatorname{tr}(Q\nabla^2f)+\operatorname{tr}B\gamma_\infty f-\gamma_\infty\langle(B+QQ_{\infty}^{-1})x,\nabla f\rangle, \end{aligned}

and hence

γ∞−1AQ,B(γ∞f)=12tr⁡(Q∇2f)−⟨(B+QQ∞−1)x,∇f⟩+tr⁡Bf.(21)\gamma_\infty^{-1}\mathcal{A}^{Q,B}(\gamma_\infty f)=\frac{1}{2}\operatorname{tr}(Q\nabla^2f)-\langle(B+QQ_{\infty}^{-1})x,\nabla f\rangle+\operatorname{tr}Bf. \tag*{(21)}

If U:L2(γ∞)→L2(γ−∞)U:L^2(\gamma_\infty)\to L^2(\gamma_{-\infty}) is the isometry of Lemma 4.2, setting as before

M=−(B+QQ∞−1),M=-(B+QQ_{\infty}^{-1}),

then we may rewrite (4.7) as

AQ,Bf=U(LQ,M+tr⁡B)U−1f.(22)\mathcal{A}^{Q,B}f=U(\mathcal{L}^{Q,M}+\operatorname{tr}B)U^{-1}f. \tag*{(22)}

Since Cc∞(Rn)C^{\infty}_{c}(\mathbb{R}^{n}) is a core for LQ,M\mathcal{L}^{Q,M} on L2(γ∞)L^{2}(\gamma_{\infty}) and UU maps Cc∞(Rn)C^{\infty}_{c}(\mathbb{R}^{n}) onto itself, the above is an equality also of domains. Thus, the operator AQ,B\mathcal{A}^{Q,B} in L2(γ−∞)L^{2}(\gamma_{-\infty}) is unitarily equivalent to LQ,M+tr⁡B\mathcal{L}^{Q,M}+\operatorname{tr} B in L2(γ∞)L^{2}(\gamma_{\infty}). This implies that

σ2(AQ,B)=tr⁡B+σ2(LQ,M),\sigma_{2}(\mathcal{A}^{Q,B})=\operatorname{tr} B+\sigma_{2}(\mathcal{L}^{Q,M}),

where σ2(LQ,M)\sigma_{2}(\mathcal{L}^{Q,M}) denotes the spectrum of LQ,M\mathcal{L}^{Q,M} on L2(γ∞)L^{2}(\gamma_{\infty}). The theorem follows by Proposition 4.5. □

Remark 4.6. As a consequence of the unitary equivalence between AQ,B\mathcal{A}^{Q,B} and LQ,M+tr⁡B\mathcal{L}^{Q,M}+\operatorname{tr} B, we can deduce some orthogonality properties for the generalized eigenspaces of AQ,B\mathcal{A}^{Q,B} from the Ornstein–Uhlenbeck setting. By [10 Proposition 3.1], if the matrix MM has only one eigenvalue, then the generalized eigenspaces of LQ,M\mathcal{L}^{Q,M} are mutually orthogonal in L2(γ∞)L^{2}(\gamma_{\infty}). Since MM is similar to B∗B^{*} (see Lemma 4.1), MM has a unique eigenvalue if and only if BB has. Therefore, if BB has only one eigenvalue, any two distinct generalized eigenspaces of AQ,B\mathcal{A}^{Q,B} are orthogonal in L2(γ−∞)L^{2}(\gamma_{-\infty}). But if BB has distinct eigenvalues, orthogonality may fail, as can be seen from an example given in [10 Section 5].

Proposition 4.7. (1) If p∈(1,∞)p\in(1,\infty), then σp(AQ,B)=σ2(AQ,B)\sigma_{p}(\mathcal{A}^{Q,B})=\sigma_{2}(\mathcal{A}^{Q,B}), and each point of σp(AQ,B)\sigma_{p}(\mathcal{A}^{Q,B}) is an eigenvalue.

(2) If p=1p=1, then σ1(AQ,B)={z∈C:Re⁡z≤0}\sigma_{1}(\mathcal{A}^{Q,B})=\{z\in\mathbb{C}:\operatorname{Re}z\leq0\}.

Proof. Let us first prove (1), thus assuming 1<p<∞1<p<\infty.

We start with the Ornstein–Uhlenbeck semigroup HtOU\mathcal{H}^{OU}_{t} in L2(γ∞)L^{2}(\gamma_{\infty}). From [20 Section 3(c)] we have that HtOU\mathcal{H}^{OU}_{t} is a compact operator on L2(γ∞)L^{2}(\gamma_{\infty}) for each t>0t>0. The resolvent of LQ,B\mathcal{L}^{Q,B} satisfies

(λ−LQ,B)−1=∫0∞e−tλHtOU dt(\lambda-\mathcal{L}^{Q,B})^{-1}=\int_{0}^{\infty}e^{-t\lambda}\mathcal{H}^{OU}_{t}\,\mathrm{d}t

for λ>0\lambda>0. Here, and similarly in the sequel, we write simply λ\lambda for λI\lambda I. This resolvent is also compact for λ>0\lambda>0 (see [24 p. 41] or [20 p. 12]), and by Lemma 4.1 the same property holds for (λ−LQ,M)−1(\lambda-\mathcal{L}^{Q,M})^{-1}, where M=−(B+QQ∞−1)M=-(B+QQ_{\infty}^{-1}). Further, the nonzero part of the spectrum of (λ−LQ,B)−1(\lambda-\mathcal{L}^{Q,B})^{-1} consists only of eigenvalues.

We then apply [18 Ch. II, Theorem 1.10, p. 55], with w=0w=0, M=1M=1 and λ\lambda replaced by λ+tr⁡B\lambda+\operatorname{tr}B, to the contractive semigroup (HtOU)t>0(\mathcal{H}^{OU}_{t})_{t>0} in the Banach space Lp(γ−∞)L^{p}(\gamma_{-\infty}), to get that for Re⁡λ+tr⁡B>0\operatorname{Re}\lambda+\operatorname{tr}B>0 the resolvent (λ−AQ,B+tr⁡B)−1(\lambda-\mathcal{A}^{Q,B}+\operatorname{tr}B)^{-1} is a bounded operator on Lp(γ−∞)L^{p}(\gamma_{-\infty}), given by

(λ−AQ,B+tr⁡B)−1f=∫0∞e−(λ+tr⁡B)tHtOUf dt,f∈Lp(γ−∞).(\lambda-\mathcal{A}^{Q,B}+\operatorname{tr}B)^{-1}f=\int_{0}^{\infty}e^{-(\lambda+\operatorname{tr}B)t}\mathcal{H}^{OU}_{t}f\,\mathrm{d}t,\qquad f\in L^{p}(\gamma_{-\infty}).

We now fix a real λ>−tr⁡B>0\lambda>-\operatorname{tr}B>0. With the isometric isomorphism U:L2(γ∞)→L2(γ−∞)U:L^{2}(\gamma_{\infty})\to L^{2}(\gamma_{-\infty}) defined above, (4.8) says that AQ,B−tr⁡B=ULQ,MU−1\mathcal{A}^{Q,B}-\operatorname{tr}B=U\mathcal{L}^{Q,M}U^{-1}, and it follows that

(λ−AQ,B+tr⁡B)−1=U(λ−LQ,M)−1U−1.(\lambda-\mathcal{A}^{Q,B}+\operatorname{tr}B)^{-1}=U(\lambda-\mathcal{L}^{Q,M})^{-1}U^{-1}.

Since (λ−LQ,M)−1(\lambda-\mathcal{L}^{Q,M})^{-1} is compact on L2(γ∞)L^{2}(\gamma_{\infty}), it follows that (λ−AQ,B+tr⁡B)−1(\lambda-\mathcal{A}^{Q,B}+\operatorname{tr}B)^{-1} is compact on L2(γ−∞)L^{2}(\gamma_{-\infty}), and its nonzero spectrum contains only eigenvalues. By [17 Theorem 1.6.1 and Corollary 1.6.2], the compactness of (λ−AQ,B+tr⁡B)−1(\lambda-\mathcal{A}^{Q,B}+\operatorname{tr}B)^{-1} on L2(γ−∞)L^{2}(\gamma_{-\infty}), combined with its boundedness on Lp(γ−∞)L^{p}(\gamma_{-\infty}) for all 1<p<∞1<p<\infty, ensures that (λ−AQ,B+tr⁡B)−1(\lambda-\mathcal{A}^{Q,B}+\operatorname{tr}B)^{-1} is compact on Lp(γ−∞)L^{p}(\gamma_{-\infty}) and that its spectrum σp((λ−AQ,B+tr⁡B)−1)\sigma_{p}((\lambda-\mathcal{A}^{Q,B}+\operatorname{tr}B)^{-1}) is independent of p∈(1,∞)p\in(1,\infty). Notice that Davies’ arguments work also under the modified hypothesis 1<p1<p<p0<∞1<p_{1}<p<p_{0}<\infty.

To pass from the spectrum of (λ−AQ,B+tr⁡B)−1(\lambda-\mathcal{A}^{Q,B}+\operatorname{tr}B)^{-1} to that of AQ,B\mathcal{A}^{Q,B}, we first observe that, by [18 Ch. II, Theorem 1.4, p. 51], the generator AQ,B\mathcal{A}^{Q,B} of the semigroup (HtUO)t>0(\mathcal{H}^{UO}_{t})_{t>0} is a closed operator in Lp(γ−∞)L^{p}(\gamma_{-\infty}). This allows us to apply the Spectral Mapping Theorem, see [18 Ch. IV, Theorem 1.13(i), p. 243]. As a result,

σp((λ−AQ,B+tr⁡B)−1)∖{0}={1λ+tr⁡B−μ:μ∈σp(AQ,B)}.(23)\sigma_{p}\left((\lambda-\mathcal{A}^{Q,B}+\operatorname{tr}B)^{-1}\right)\setminus\{0\}=\left\{\frac{1}{\lambda+\operatorname{tr}B-\mu}:\mu\in\sigma_{p}(\mathcal{A}^{Q,B})\right\}. \tag*{(23)}

or equivalently,

σp(AQ,B)={λ+tr⁡B−1ν:ν∈σp((λ−AQ,B+tr⁡B)−1)∖{0}}.(24)\sigma_{p}(\mathcal{A}^{Q,B})=\left\{\lambda+\operatorname{tr}B-\frac{1}{\nu}:\nu\in\sigma_{p}\left((\lambda-\mathcal{A}^{Q,B}+\operatorname{tr}B)^{-1}\right)\setminus\{0\}\right\}. \tag*{(24)}

Since the right-hand side in (24) is independent of p∈(1,∞)p\in(1,\infty), so is σp(AQ,B)\sigma_{p}(\mathcal{A}^{Q,B}). Part (ii) of the same theorem [18 Ch. IV, Theorem 1.13, p. 243] says that (23) and thus (24) hold also for the point spectra. Since (λ−AQ,B+tr⁡B)−1(\lambda-\mathcal{A}^{Q,B}+\operatorname{tr}B)^{-1} is compact on Lp(γ−∞)L^{p}(\gamma_{-\infty}), the set σp((λ−AQ,B+tr⁡B)−1)∖{0}\sigma_{p}((\lambda-\mathcal{A}^{Q,B}+\operatorname{tr}B)^{-1})\setminus\{0\} coincides with the point spectrum of this resolvent, and hence the same holds for AQ,B\mathcal{A}^{Q,B}. This ends the proof of (1).

Let us now prove (2). As in (17), we write M=−(B+QQ∞−1)M=-(B+QQ_{\infty}^{-1}). The map

U~=cn−1/2U:L1(Rn,dx)→L1(γ−∞),U~f:=e−R(x)f,\widetilde{U}=c_{n}^{-1/2}U:L^{1}(\mathbb{R}^{n},dx)\to L^{1}(\gamma_{-\infty}),\qquad\widetilde{U}f:=e^{-R(x)}f,

is an isometric isomorphism. Since Cc∞(Rn)C_{c}^{\infty}(\mathbb{R}^{n}) is a core for the Ornstein–Uhlenbeck generator on L1(dx)L^{1}(dx) by [22 Proposition 3.2], and U~\widetilde{U} maps Cc∞(Rn)C_{c}^{\infty}(\mathbb{R}^{n}) onto itself, the spectrum of AQ,B\mathcal{A}^{Q,B} on L1(γ−∞)L^{1}(\gamma_{-\infty}) coincides with that of LQ,M+tr⁡B\mathcal{L}^{Q,M}+\operatorname{tr}B on L1(Rn,dx)L^{1}(\mathbb{R}^{n},dx). By [22 Theorem 4.12], the spectrum of LQ,M\mathcal{L}^{Q,M} on L1(Rn,dx)L^{1}(\mathbb{R}^{n},dx) is the half-plane {z∈C:Re⁡z≤−tr⁡M}\{z\in\mathbb{C}:\operatorname{Re}z\leq-\operatorname{tr}M\}. Recalling from Lemma 4.1 that tr⁡M=tr⁡B\operatorname{tr}M=\operatorname{tr}B, we conclude that the spectrum of AQ,B\mathcal{A}^{Q,B} on L1(γ−∞)L^{1}(\gamma_{-\infty}) is the half-plane {z∈C:Re⁡z≤0}\{z\in\mathbb{C}:\operatorname{Re}z\leq0\}. □

Remark 4.8. The domain D(AQ,B)D(\mathcal{A}^{Q,B}) of the infinitesimal generator in L2(γ−∞)L^{2}(\gamma_{-\infty}) can be characterized by means of the isometry UU introduced above. More precisely, for p=2p=2, the unitary equivalence (22) implies that

D(AQ,B)L2(γ−∞)={f∈L2(γ−∞):U−1f∈W2,2(γ∞)},D(\mathcal{A}^{Q,B})_{L^{2}(\gamma_{-\infty})}=\{f\in L^{2}(\gamma_{-\infty}):U^{-1}f\in W^{2,2}(\gamma_{\infty})\},

where W2,2(γ∞)W^{2,2}(\gamma_{\infty}) is the domain of the classic Ornstein–Uhlenbeck operator LQ,M\mathcal{L}^{Q,M} in L2(γ∞)L^{2}(\gamma_{\infty}) [25 Section 3].

Ultracontractivity of (HtUO)t>0(\mathcal{H}^{UO}_{t})_{t>0}

We recall that a semigroup (Tt)t>0(T_{t})_{t>0} is called ultracontractive if for every p≥1p\geq1 and t>0t>0 there exists some positive constant cp(t)c_{p}(t) such that ∥Ttf∥∞≤cp(t)∥f∥p\|T_{t}f\|_{\infty}\leq c_{p}(t)\|f\|_{p}.

It follows from earlier results by Nelson [23] that the Ornstein–Uhlenbeck semigroup (HtOU)t>0(\mathcal{H}^{OU}_{t})_{t>0} is not ultracontractive in Lp(γ∞)L^{p}(\gamma_{\infty}). But (HtOU)t>0(\mathcal{H}^{OU}_{t})_{t>0} is ultracontractive in LpL^{p} spaces endowed with Lebesgue measure [20 p. 13]. Unexpectedly, in the inverse Gaussian setting it is easy to prove ultracontractivity for (HtUO)t>0(\mathcal{H}^{UO}_{t})_{t>0} in LpL^{p} spaces both with respect to Lebesgue measure and with respect to γ−∞\gamma_{-\infty}.

We shall need an elementary lemma.

Lemma 5.1. For 1≤r≤∞1 \le r \le\infty we have

∥e−12⟨Qt−1(etBu−x),etBu−x⟩∥Lr(du)=(2πr)n2r(det⁡Qt)12re−1rttr⁡B,x∈Rn,\left\|e^{-\frac{1}{2}\langle Q_t^{-1}(e^{tB}u-x),e^{tB}u-x\rangle}\right\|_{L^r(du)}=\left(\frac{2\pi}{r}\right)^{\frac{n}{2r}}(\det Q_t)^{\frac{1}{2r}}e^{-\frac{1}{r}t\operatorname{tr}B},\qquad x\in\mathbb{R}^n,

where the right-hand side is 1 for r=∞r=\infty.

Proof. When r=∞r=\infty this is trivial, and for 1≤r<∞1\le r<\infty one has

∫Rne−r2⟨Qt−1/2(etBu−x),Qt−1/2(etBu−x)⟩ du=(det⁡Qt)1/2e−ttr⁡B∫Rne−r2∣z∣2 dz,\int_{\mathbb{R}^n}e^{-\frac{r}{2}\langle Q_t^{-1/2}(e^{tB}u-x),Q_t^{-1/2}(e^{tB}u-x)\rangle}\,du=(\det Q_t)^{1/2}e^{-t\operatorname{tr}B}\int_{\mathbb{R}^n}e^{-\frac{r}{2}|z|^2}\,dz,

which implies the lemma. □

Proposition 5.2. (1) The semigroup (HtUO)t>0(\mathcal{H}^{UO}_t)_{t>0} is ultracontractive with respect to Lebesgue measure with constant

cp(t)=(2π)−n2pettr⁡B1pn2p(p′)n2p′(det⁡Qt)−12p,1<p<∞;c_p(t)=(2\pi)^{-\frac{n}{2p}}e^{t\operatorname{tr}B}\frac{1}{p^{\frac{n}{2p}}(p')^{\frac{n}{2p'}}}(\det Q_t)^{-\frac{1}{2p}},\qquad1<p<\infty;
c1(t)=(2π)−n2ettr⁡B(det⁡Qt)−12.(25)c_1(t)=(2\pi)^{-\frac{n}{2}}e^{t\operatorname{tr}B}(\det Q_t)^{-\frac{1}{2}}. \tag*{(25)}

(2) The semigroup (HtUO)t>0(\mathcal{H}^{UO}_t)_{t>0} is ultracontractive with respect to γ−∞\gamma_{-\infty} with the same constant cp(t)c_p(t) given by (25).

Proof. We may assume f∈Cc∞(Rn)f\in C^\infty_c(\mathbb{R}^n) since this space is dense in Lp(du)L^p(du), for 1≤p<∞1\le p<\infty. Then it follows from (9) that

HtUOf(x)=(2π)−n2(det⁡Qt)−12ettr⁡B∫Rnf(u)e−12⟨Qt−1(etBu−x),etBu−x⟩ du,x∈Rn,\mathcal{H}^{UO}_t f(x)=(2\pi)^{-\frac{n}{2}}(\det Q_t)^{-\frac{1}{2}}e^{t\operatorname{tr}B}\int_{\mathbb{R}^n}f(u)e^{-\frac{1}{2}\langle Q_t^{-1}(e^{tB}u-x),e^{tB}u-x\rangle}\,du,\qquad x\in\mathbb{R}^n,

and Lemma 5.1 implies the assertion (1). To prove the assertion (2), it is enough to observe that

∥f∥Lp(du)≤∥f∥Lp(γ−∞),\|f\|_{L^p(du)}\le\|f\|_{L^p(\gamma_{-\infty})},

which completes the proof. □

Corollary 5.3. For any 1≤p<q1\le p<q and t>0t>0

∥HtUOf∥Lq(γ−∞)≤cp(t)1−p/q∥f∥Lp(γ−∞).(26)\|\mathcal{H}^{UO}_t f\|_{L^q(\gamma_{-\infty})}\le c_p(t)^{\frac{1-p/q}{}} \|f\|_{L^p(\gamma_{-\infty})}. \tag*{(26)}

Proof. This is immediate by interpolation between the LpL^p-contractivity in Theorem 3.1 and the ultracontractivity bound. □

Remark 5.4. Thanks to the explicit expression of the ultracontractivity constant cp(t)c_p(t) in (25), we notice what follows. For small tt one has

Qt=∫0tesBQesB∗ ds≃tQQ_t=\int_0^t e^{sB}Qe^{sB^*}\,ds\simeq tQ

and consequently

det⁡Qt≃tndet⁡Q,\det Q_t\simeq t^n\det Q,

which implies that the constant cp(t)c_p(t) blows up like t−n2pt^{-\frac{n}{2p}} as t→0+t\to0^+. Notice that the blow-up rate t−n2pt^{-\frac{n}{2p}} is analogous to that of the standard heat semigroup (etΔ)t>0(e^{t\Delta})_{t>0} in Rn\mathbb{R}^n from Lp(dx)L^p(dx) to

L∞(dx)L^\infty(dx). This is a local effect. As t→+∞t \to+\infty, the matrix QtQ_t converges to the positive definite matrix Q∞Q_\infty. Since tr⁡B<0\operatorname{tr}B < 0, the factor etptr⁡Be^{\frac{t}{p}\operatorname{tr}B} makes cp(t)c_p(t) decay exponentially as t→+∞t \to+\infty.

A counterexample for q<pq<p

Clearly, (5.2) provides bounded operators Lp→LqL^p \to L^q for any tt as soon as q>pq > p. Let now instead 1<q<p<∞1 < q < p < \infty.

We claim that if tt is small enough, in a way that depends only on p,q,n,Qp,q,n,Q and BB, then HtUO\mathcal{H}^{\mathrm{UO}}_t is not bounded from Lp(γ−∞)L^p(\gamma_{-\infty}) to Lq(γ−∞)L^q(\gamma_{-\infty}).

Fix t∈(0,1)t \in(0,1) and let u0u_0 be a point in Rn\mathbb{R}^n with ∣u0∣>t−1/2|u_0| > t^{-1/2}. We let ff be the characteristic function of the ball B(u0,1/∣u0∣)B(u_0,1/|u_0|). Since the density eR(u)e^{R(u)} has constant order of magnitude in this ball, we have

∥f∥Lp(γ−∞)≃eR(u0)/p∣u0∣−n/p.\|f\|_{L^p(\gamma_{-\infty})} \simeq e^{R(u_0)/p}|u_0|^{-n/p}.

We will evaluate HtUOf(x)\mathcal{H}^{\mathrm{UO}}_t f(x) at points xx in the ball B(Dtu0,t)B(D_tu_0,\sqrt{t}), and actually in a thin, curved slice of this ball, given by

S={x∈B(Dtu0,t):∣R(x)−R(Dtu0)∣<1},S=\{x\in B(D_tu_0,\sqrt{t}): |R(x)-R(D_tu_0)|<1\},

thus containing the center of the ball. Since t<1t<1 we have R(Dtu0)≃R(u0)≃∣u0∣2R(D_tu_0)\simeq R(u_0)\simeq|u_0|^2. The width of SS is roughly 1/∣u0∣1/|u_0| as seen via R(x)R(x), and its Lebesgue measure will be

∣S∣≃t(n−1)/2/∣u0∣.|S|\simeq t^{(n-1)/2}/|u_0|.

For x∈Sx\in S and u∈B(u0,1/∣u0∣)u\in B(u_0,1/|u_0|) we have

∣D−tx−u∣≃∣x−Dtu∣≤∣x−Dtu0∣+∣Dt(u−u0)∣≲t+1/∣u0∣≲t,|D_{-t}x-u|\simeq|x-D_tu|\leq|x-D_tu_0|+|D_t(u-u_0)|\lesssim\sqrt{t}+1/|u_0|\lesssim\sqrt{t},

the last estimate by the choice of u0u_0, and (2.8) implies that KtUO(x,u)≳t−n/2e−R(x)≃t−n/2e−R(Dtu0)K^{\mathrm{UO}}_t(x,u)\gtrsim t^{-n/2}e^{-R(x)}\simeq t^{-n/2}e^{-R(D_tu_0)}. Thus

HtUOf(x)≳t−n/2e−R(Dtu0)eR(u0)∣u0∣−n,x∈S,\mathcal{H}^{\mathrm{UO}}_t f(x)\gtrsim t^{-n/2}e^{-R(D_tu_0)}e^{R(u_0)}|u_0|^{-n},\qquad x\in S,

and so

∥HtUOf∥Lq(γ−∞)≳t−n/2e−R(Dtu0)eR(u0)∣u0∣−nγ−∞(S)1/q≃t−n/2e−R(Dtu0)eR(u0)∣u0∣−n∣S∣1/qeR(Dtu0)/q≳eR(Dtu0)(−1+1/q)eR(u0)tC∣u0∣−C,\begin{aligned} \|\mathcal{H}^{\mathrm{UO}}_t f\|_{L^q(\gamma_{-\infty})} &\gtrsim t^{-n/2}e^{-R(D_tu_0)}e^{R(u_0)}|u_0|^{-n}\gamma_{-\infty}(S)^{1/q}\\ &\simeq t^{-n/2}e^{-R(D_tu_0)}e^{R(u_0)}|u_0|^{-n}|S|^{1/q}e^{R(D_tu_0)/q}\\ &\gtrsim e^{R(D_tu_0)(-1+1/q)}e^{R(u_0)}t^C|u_0|^{-C}, \end{aligned}

for some constants CC. For the quotient between the norms, we find

P:=∥HtUOf∥Lq(γ−∞)∥f∥Lp(γ−∞)≳eR(Dtu0)(−1+1/q)eR(u0)(1−1/p)tC∣u0∣−C.P:=\frac{\|\mathcal{H}^{\mathrm{UO}}_t f\|_{L^q(\gamma_{-\infty})}}{\|f\|_{L^p(\gamma_{-\infty})}} \gtrsim e^{R(D_tu_0)(-1+1/q)}e^{R(u_0)(1-1/p)}t^C|u_0|^{-C}.

In the light of (2.2), ∂sR(Dsu0)≃∣Dsu0∣2≃∣u0∣2\partial_sR(D_su_0)\simeq|D_su_0|^2\simeq|u_0|^2 for s∈(0,1)s\in(0,1), so that

R(Dtu0)−R(u0)≃∫0t∣u0∣2 ds=t∣u0∣2.R(D_tu_0)-R(u_0)\simeq\int_0^t |u_0|^2\,ds=t|u_0|^2.

Thus

P≳eR(u0)(1/q−1/p)−Ct∣u0∣2tC∣u0∣−C.(27)P\gtrsim e^{R(u_0)(1/q-1/p)-Ct|u_0|^2}t^C|u_0|^{-C}. \tag*{(27)}

Recall that R(u0)=12⟨Q∞−1u0,u0⟩≃∣u0∣2R(u_0)=\frac{1}{2}\langle Q_\infty^{-1}u_0,u_0\rangle\simeq|u_0|^2. By taking tt small in the way described in the claim, we can thus obtain the inequality Ct∣u0∣2<R(u0)(1/q−1/p)/2Ct|u_0|^2<R(u_0)(1/q-1/p)/2 between the first two exponents in (5.3). This leads to

P≳eR(u0)(1/q−1/p)/2tC∣u0∣−C,P\gtrsim e^{R(u_0)(1/q-1/p)/2}t^C|u_0|^{-C},

which tends to infinity with ∣u0∣|u_0|. The claim follows.

Hypercontractivity

We recall that a semigroup (Tt)t>0(T_t)_{t>0} is said to be hypercontractive if for every p>1p>1 and t>0t>0 there exists some q(t)>pq(t)>p such that

∥Ttf∥q(t)≤∥f∥p.\|T_tf\|_{q(t)}\leq\|f\|_p.

Theorem 6.1. The semigroup (HtUO)t>0(\mathcal{H}^{UO}_t)_{t>0} is hypercontractive.

Proof. The main hypercontractivity property of (HtOU)t>0(\mathcal{H}^{OU}_t)_{t>0} is the following. Given 1≤p<q<∞1\leq p<q<\infty and t>0t>0, if

q−1≤(p−1)∥Q∞−1/2etBQ∞1/2∥−2q-1\leq(p-1)\|Q_\infty^{-1/2}e^{tB}Q_\infty^{1/2}\|^{-2}

then HtOU\mathcal{H}^{OU}_t is a contraction from Lp(γ∞)L^p(\gamma_\infty) into Lq(γ∞)L^q(\gamma_\infty). For a proof of this result, holding also in the infinite-dimensional setting, we refer to [14] and [19].

It has been proved in [19 Proposition 4.2 and Proposition 4.1] that

∥Q∞−1/2etBQ∞1/2∥<1for all t>0.\|Q_\infty^{-1/2}e^{tB}Q_\infty^{1/2}\|<1\qquad\text{for all }t>0.

We fix t>0t>0. Choosing q=2q=2 above, we see that we can find a p∈(1,2)p\in(1,2) such that HtOU\mathcal{H}^{OU}_t is a contraction from Lp(γ∞)L^p(\gamma_\infty) to L2(γ∞)L^2(\gamma_\infty). For the dual operator we then conclude that (HtOU)∗(\mathcal{H}^{OU}_t)^* is a contraction from L2(γ∞)L^2(\gamma_\infty) to Lp′(γ∞)L^{p'}(\gamma_\infty), and thus from L2(γ∞)L^2(\gamma_\infty) to any Lϱ(γ∞)L^\varrho(\gamma_\infty) with ϱ∈(2,p′)\varrho\in(2,p').

It follows that (HtOU)∗U−1(\mathcal{H}^{OU}_t)^*U^{-1} is a contraction from L2(γ−∞)L^2(\gamma_{-\infty}) to Lϱ(γ∞)L^\varrho(\gamma_\infty), for ϱ∈(2,p′)\varrho\in(2,p'). Then we observe that UU is a bounded operator from Lϱ(γ∞)L^\varrho(\gamma_\infty) to Lϱ(γ−∞)L^\varrho(\gamma_{-\infty}), actually for any ϱ≥2\varrho\geq2; more precisely we have with 0≤g∈Lϱ(γ∞)0\leq g\in L^\varrho(\gamma_\infty)

∫Rn(Ug)ϱ dγ−∞=cnϱ/2∫Rne−ϱR(x)g(x)ϱeR(x) dx≤cnϱ/2∫Rng(x)ϱe−R(x) dx=cnϱ/2−1∫Rngϱ dγ∞.\begin{aligned} \int_{\mathbb{R}^n}(Ug)^\varrho\,d\gamma_{-\infty} =c_n^{\varrho/2}\int_{\mathbb{R}^n}e^{-\varrho R(x)}g(x)^\varrho e^{R(x)}\,dx \\ &\leq c_n^{\varrho/2}\int_{\mathbb{R}^n}g(x)^\varrho e^{-R(x)}\,dx =c_n^{\varrho/2-1}\int_{\mathbb{R}^n}g^\varrho\,d\gamma_\infty. \end{aligned}

Thus the operator norm is no larger than cn1/2−1/ϱc_n^{1/2-1/\varrho}.

Now Lemma 4.2 implies that HtUO\mathcal{H}^{UO}_t is bounded from L2(γ−∞)L^2(\gamma_{-\infty}) to Lϱ(γ−∞)L^\varrho(\gamma_{-\infty}) with operator norm at most ettr⁡Bcn1/2−1/ϱe^{t\operatorname{tr}B}c_n^{1/2-1/\varrho}, if 2<ϱ<p′2<\varrho<p'. To make it a contraction, it is enough to take ϱ∈(2,p′)\varrho\in(2,p') such that 1/2−1/ϱ≤t∣tr⁡B∣/log⁡+cn1/2-1/\varrho\leq t|\operatorname{tr}B|/\log_+c_n (if cn≤1c_n\leq1 any ϱ∈(2,p′)\varrho\in(2,p') will do). Interpolation with the contractive estimate on L1(γ−∞)L^1(\gamma_{-\infty}) yields a contractive estimate Lp(γ−∞)→Lq(γ−∞)L^p(\gamma_{-\infty})\to L^q(\gamma_{-\infty}) for some q>pq>p for every 1<p<21<p<2. Interpolation with the contractive estimate on L∞L^\infty similarly yields the stated contractivity for p>2.□p>2. \square

As a consequence of the above lemma, we also have the following.

Corollary 6.2. The semigroup (HtUO)t≥0(\mathcal{H}^{UO}_t)_{t\geq0} is exponentially stable on L2(γ−∞)L^2(\gamma_{-\infty}), in the sense that it satisfies a bound

∥HtUO∥L2(γ−∞)→L2(γ−∞)≤e−ωtfor all t≥0,\|\mathcal{H}^{UO}_t\|_{L^2(\gamma_{-\infty})\to L^2(\gamma_{-\infty})}\leq e^{-\omega t}\qquad\text{for all }t\geq0,

with ω=−tr⁡B>0\omega=-\operatorname{tr}B>0.

Dipartimento di Matematica, Università degli Studi di Genova, Via Dodecaneso 35, 16146 Genova, Italy

Email address: [email protected]

DTG, Università degli Studi di Padova, Stradella san Nicola 3, I-36100 Vicenza, Italy

Email address: [email protected]

Dipartimento di Matematica “Tullio Levi Civita”, Università degli Studi di Padova, Via Trieste, 63, 35131 Padova,, Italy

Email address: [email protected]

Mathematical Sciences, University of Gothenburg and Mathematical Sciences, Chalmers University of Technology, SE - 412 96 Göteborg, Sweden

Email address: [email protected]

References

  1. [1]V. Almeida and J. J. Betancor, Variation and oscillation for harmonic operators in the inverse Gaussian setting, Communications on Pure and Applied Analysis (2022), 21, 419–470.arxiv.org/abs/2012.11205
  2. [2]V. Almeida, J. J. Betancor, J. C. Fariña and L. Rodríguez-Mesa, Littlewood-Paley-Stein theory and Banach spaces in the inverse Gaussian setting, Potential Analysis (2023), 59, 1235–1284.arxiv.org/abs/2102.13381
  3. [3]J. J. Betancor, L. Rodríguez-Mesa, Higher-Order Riesz Transforms in the Inverse Gaussian Setting and UMD Banach Spaces, Journal of Function Spaces (2021), Article 6899603.arxiv.org/abs/2011.11285
  4. [4]J. J. Betancor, A. Castro, and M. de León-Contreras, The Hardy–Littlewood property and maximal operators associated with the inverse Gauss measure, Ann. Sc. Norm. Super. Pisa Cl. Sci. (2023), Vol. XXIV, 551–597.
  5. [5]T. Bruno, Singular integrals and Hardy type spaces for the inverse Gauss measure, J. Geom. Anal. (2021), 31, 6481–6528.arxiv.org/abs/1801.09000
  6. [6]T. Bruno, V. Casarino, P. Ciatti and P. Sjögren, Boundedness properties of the maximal operator in a nonsymmetric inverse Gaussian setting, Pot. Anal. 63, 2077–2098 (2025)arxiv.org/abs/2501.17517
  7. [7]T. Bruno and P. Sjögren, On the Riesz transform for the inverse Gauss measure, Ann. Fenn. Math., (2021), 46, 1–16.arxiv.org/abs/1906.03827
  8. [8]V. Casarino, P. Ciatti and P. Sjögren, On the maximal operator of a general Ornstein–Uhlenbeck semigroup, Math. Z. (2022) 301, 2393–2413.arxiv.org/abs/1901.04823
  9. [9]V. Casarino, P. Ciatti and P. Sjögren, Riesz transforms of a general Ornstein–Uhlenbeck semigroup, Calculus of Variations and Partial Differential Equations, 60, 135 (2021).DOI
  10. [10]V. Casarino, P. Ciatti and P. Sjögren, On the orthogonality of generalized eigenspaces for the Ornstein–Uhlenbeck operator, Archiv der Mathematik, 117 (2021), 547–556.DOI
  11. [11]V. Casarino, P. Ciatti and P. Sjögren, Spectral multipliers in a general Gaussian setting, to appear in Ann. Sc. Norm. Super. Pisa Cl. Sci., https://doi.org/10.2422/2036-2145.202311.014.
  12. [12]V. Casarino, P. Ciatti and P. Sjögren, Variational inequalities for the Ornstein–Uhlenbeck semigroup: the higher-dimensional case, J. Geom. Anal. (2025), 35, https://doi.org/10.1007/s12220-024-01859-4.
  13. [13]V. Casarino, P. Ciatti and P. Sjögren, Weak type (1, 1) jump inequalities in a nonsymmetric Gaussian setting, (2025), arXiv:2510.18755, to appear in Transactions of the American Math. Society.arxiv.org/abs/2510.18755
  14. [14]A. Chojnowska-Michalik and B. Goldys, Nonsymmetric Ornstein–Uhlenbeck semigroup as second quantized operator, J. Math. Kyoto Univ. (1996), 36, 481–498.
  15. [15]A. Chojnowska-Michalik and B. Goldys, Symmetric Ornstein–Uhlenbeck generators: Characterizations and identification of domains, Probab. Theory Relat. Fields, 124 (2002), 459–486.
  16. [16]G. Da Prato and J. Zabczyk, Stochastic equations in infinite dimensions, Cambridge University Press, Cambridge, 1992.
  17. [17]E. B., Davies, Heat Kernel and Spectral Theory. Cambridge University Press, Cambridge (1990)
  18. [18]K. J. Engel, R. Nagel, R, One-Parameter Semigroups for Linear Evolution Equations. Springer, (1999).
  19. [19]M. Fuhrman, Hypercontractivity properties of nonsymmetric Ornstein–Uhlenbeck semigroups in Hilbert spaces, Stoch. Anal. Appl. (1989), 16, 241–260.
  20. [20]A. Lunardi, G. Metafune and D. Pallara, The Ornstein–Uhlenbeck semigroup in finite dimension, Phil. Trans. R. Soc. A 378, (2020), 20200217.DOI
  21. [21]M. Mazzitelli, P. R. Stinga and J. L. Torrea, Fractional powers of first order differential operators and new families of polynomials associated to inverse measures, arXiv:2205.00050.arxiv.org/abs/2205.00050
  22. [22]G. Metafune, Lᵖ-spectrum of Ornstein–Uhlenbeck operators. Ann. Sc. Norm. Super. Pisa Cl. Sci. (4) 30(1), 97–124 (2001).
  23. [23]E. Nelson, The free Markoff field, J. Funct. Anal. (1973) 12, 211–227.DOI
  24. [24]G. Metafune, D. Pallara, E. Priola, Spectrum of Ornstein–Uhlenbeck operators in Lᵖ spaces with respect to invariant measures. J. Funct. Anal. 196 (2002), no. 1, 40–60.DOI
  25. [25]G. Metafune, J. Prüss, A. Rhandi, R. Schnaubelt, The domain of the Ornstein–Uhlenbeck operator on an Lᵖ-space with invariant measure. Ann. Sc. Norm. Super. Pisa Cl. Sci. (5) 1 (2002), no. 2, 471–485.
  26. [26]F. Salogni, Harmonic Bergman spaces, Hardy-type spaces and harmonic analysis of a symmetric diffusion semigroup on ℝⁿ. PhD Thesis, Università degli Studi di Milano-Bicocca, 2013. https://boa.unimib.it/retrieve/handle/10281/41814/62217/phd_unimib_058626.pdf

Paper details

vlnc:W_raujfayz
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 12

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents