Local times and sample path properties of the Rosenblatt process
Introduction
The Rosenblatt process is a self-similar stochastic process with long-range dependence and heavier tails than those of the normal distribution. It depends on a parameter which is fixed in what follows. The process belongs to the family of Hermite processes that naturally arise as limits of normalized sums of long-range dependent random variables [8]. The first Hermite process is the fractional Brownian motion, which is Gaussian and thus satisfies many desirable properties. The Rosenblatt process is the second Hermite process: it is no longer Gaussian and was introduced by Rosenblatt in [20]. Although less popular than fractional Brownian motion, the Rosenblatt process is attracting increasing interest in the literature, mainly due to its self-similarity, stationarity of increments, and long-range dependence properties. See for example [1], [6], [15], [18] for recent studies on different aspects of this process.
The Rosenblatt process admits the following stochastic representation, also known as the spectral representation: In (1.1), the double Wiener–Itô integral is taken over and is a complex-valued random white noise with control measure satisfying . The kernel is given by and is a complex valued Hilbert–Schmidt kernel satisfying and . In particular, the spectral theorem applies, see [8], and allows one to write where is a sequence of independent standard Gaussian random variables and are the eigenvalues1 (repeated according to the possible multiplicity) of the self-adjoint operator , given by Furthermore, and thus (1.3) converges.
The goal of the present paper is to study the occupation density (also known as the local time) of , where and is a finite closed interval. Recall that for a deterministic function , the occupation measure of is defined by where and are Borel sets and is the Lebesgue measure on . Observe that represents the amount of time during a period where takes value in . Then, when is absolutely continuous with respect to , the occupation density (or local time) is given by the Radon–Nikodym derivative We study the local time via the analytic approach initiated by Berman [4]. The idea is to relate properties of with the integrability properties of the Fourier transform of . Recall the following key result [10]:
Proposition 1.1 The function has an occupation density for which is square integrable in for every fixed iff Moreover, in this case, the occupation density can be represented as
The deterministic function can be chosen to be the single path of a stochastic process . To show almost sure existence and square integrability of in that case, it will be enough to show that or equivalently If is Gaussian, then one can evaluate explicitly to establish (1.4). It leads to the well-known Gaussian criterion:
Proposition 1.2 Lemma 19 in [10] Suppose that is Gaussian and centered, and satisfies where . Then has an occupation density which, for any , is square integrable, in particular (1.4) holds.
In our setting is the Rosenblatt process which is not Gaussian. Nevertheless, a careful analysis of via (1.3) yields (1.4). This is the approach in [21] where existence of the local time of the Rosenblatt process was first established. In this paper we show a considerably more involved bound on the Fourier transform which in turn will yield deeper results regarding the irregularity properties of the sample paths. The following is the key result of our paper:
Proposition 1.3 Let and . Then, for any times , the Rosenblatt process satisfies where the constant depends only on and .
Proposition 1.3 can be applied to obtain the following Hölder conditions on .
Theorem 1.4 Let be a Rosenblatt process with . The local time is almost surely jointly continuous and has finite moments. For a finite closed interval , let . There exist deterministic constants and such that, almost surely, for any and
In particular, the local time is well defined for any fixed and interval . Explicit estimates for the moments of the local time are provided in Theorem 3.1. As a direct corollary we obtain:
Corollary 1.5 For any finite closed interval there exist deterministic constants and , independent of and , such that, almost surely, for every and every
and for every
Moreover, we get the following for a particular Hausdorff measure:
Corollary 1.6 Let be a finite closed interval. There exists a deterministic constant such that for every we have where denotes -Hausdorff measure with .
Furthermore, we can quantify the behavior of the trajectories of . We also point the reader to the recent paper [2] for a similar result obtained with a different approach.
Corollary 1.7 Let be a finite closed interval. There exists a deterministic constant such that for every we have, almost surely, and In particular, is almost surely nowhere differentiable.
In Section 2 we establish our main result Proposition 1.3. As we are dealing with a second order Hermite process, we are forced to control from below singular values of somewhat unwieldy operators (see Remark 5.3). For this purpose we need to introduce several technical lemmas exhibiting tools from operator theory and harmonic analysis, including the theory of weighted integrals. Their proofs are postponed into Section 5. Section 3 is dedicated to some results regarding the existence and joint continuity of the local time. In particular, bounds on the moments of are obtained. Finally, in Section 4, Theorem 1.4 and Corollary 1.5, Corollary 1.6, Corollary 1.7 are established.
Section snippets
Integrability of the Fourier transform
The purpose of this section is to provide a proof of Proposition 1.3. We first outline the main steps as Lemma 2.1, Lemma 2.2, Lemma 2.3, Lemma 2.4 and then we establish (1.5). The proofs of the lemmas are carried out in Section 5.
We use the following normalization for the Fourier transform for and . The norm in the weighted space is defined as .
First, we obtain a representation of the left-hand side of (1.5) using the eigenvalues of an
Joint continuity of the local times and moment estimates
In the present section we apply Proposition 1.3 to produce moment estimates for the local time, that are of some independent interest.
Let and . We recall from [21] that the local time for the Rosenblatt process exists and admits the representation3
Proofs of main theorems
In this section we present proofs to our main results, namely Theorem 1.4 and Corollary 1.5, Corollary 1.6, Corollary 1.7. We start with two auxiliary lemmas.
Lemma 4.1 There exists such that
Proof This follows by observing that by Lemma 2.1 the characteristic function of has a bounded analytic extension to a strip for some . □
Proposition 4.2 Let be the Rosenblatt process and set , where and . Then, where and are constants that depend
Spectral estimates
Proof of Lemma 2.1 Note that by (1.1), (1.2), where the integral is taken over , and is a Hilbert–Schmidt kernel. Then the operator satisfies Similarly to (1.3), for all , where is a sequence of independent Gaussian random variables and are the eigenvalues of the operator . Then, the characteristic
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
The authors are grateful to two referees for their valuable comments and suggestions which helped to improve the presentation of the paper. G. Kerchev and I. Nourdin are supported by the FNR, Luxembourg OPEN grant APOGee at Luxembourg University.
References (22)
Lower bound for local oscillations of Hermite processes
Stochastic Process. Appl.
(2020)- et al.
Hitchhiker’s guide to the fractional Sobolev spaces
Bull. Sci. Math.
(2012) - et al.
Random homogenization and convergence to integrals with respect to the Rosenblatt process
J. Differential Equations
(2012) - et al.
Statistical inference for Vasicek-type model driven by Hermite processes
Stochastic Process. Appl.
(2019) - et al.
Persistence probabilities and a decorrelation inequality for the Rosenblatt process and Hermite processes
Teor. Veroyatn. Primen.
(2018) - et al.
Joint continuity of the local times of fractional Brownian sheets
Ann. Inst. H. Poincaré Probab. Statist.
(2008) Local times and sample function properties of stationary Gaussian processes
Trans. Amer. Math. Soc.
(1969)- et al.
Variations and hurst index estimation for a rosenblatt process using longer filters
Electron. J. Stat.
(2009) - et al.
Non-central limit theorems for nonlinear functionals of Gaussian fields
Z. Wahrsch. Verw. Gebiete.
(1979)
Spectral properties of the operator of Riesz potential type
Proc. Amer. Math. Soc.
Cited by (10)
Ergodicity of increments of the Rosenblatt process and some consequences
2024, Czechoslovak Mathematical JournalBesov-Orlicz Path Regularity of Non-Gaussian Processes
2024, Potential AnalysisSelf-intersection local time derivative for systems of non-linear stochastic heat equations
2023, Journal of Mathematical PhysicsLocal times for systems of non-linear stochastic heat equations
2023, Stochastics and Partial Differential Equations: Analysis and ComputationsOn Non-degenerate Chaos Processes
2022, arXiv