Functional limit theorems for a time-changed multidimensional Wiener process

Yuliya Mishura, René L. Schilling · 2025

Read Functional limit theorems for a time-changed multidimensional Wiener process by Yuliya Mishura, René L. Schilling (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-10820.txt.

Introduction

Yuliya Mishura and René Schilling proved limit theorems in two thousand twenty-five for a Wiener process whose clock runs at a speed set by the process itself. Such time changes arise when the generator of Brownian motion is multiplied by an intensity that varies across space, possibly fading in places. They assume the intensity settles to a limit in each octant of space, though the limits may differ by direction.

Applying a limit theorem for stacked processes, they show what the rescaled time-changed process converges to. Among the possible limits is a multidimensional cousin of skew Brownian motion, which behaves differently on each side of a boundary. The work extends classical diffusion limits.

Catalog record

CREATORYuliya Mishura, René L. Schilling
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE41 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-10820

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Functional limit theorems for a time-changed multidimensional Wiener process Thanks: The authors thank Michael Röckner for fruitful discussions on this topic. The first-named author was supported by the Swedish Foundation for Strategic Research (Grant No. UKR24-0004), the Japan Science and Technology Agency CREST project (Grant No. 811JPMJCR2115) and the ToppForsk project of the Research Council of Norway with the title STORM: Stochastics for Time-Space Risk Models (Grant No. 274410). The second-named author was supported by the 6G-life and ScaDS.AI Dresden/Leipzig projects. Yuliya Mishura, René L. Schilling Abstract . We study the asymptotic behaviour of a properly normalized time-changed multidimensional Wiener process; the time change is given by an additive functional of the Wiener process itself. At the level of generators, the time change means that we consider the Laplace operator – which generates a multidimensional Wiener process – and multiply it by a (possibly degenerate) state-space dependent intensity. We assume that the intensity admits limits at infinity in each octant of the state space, but the values of these limits may be different. Applying a functional limit theorem for the superposition of stochastic processes, we prove functional limit theorems for the normalized time-changed multidimensional Wiener process. Among the possible limits there is a multidimensional analogue of skew Brownian motion. 2020 Mathematics Subject Classification Primary: 60J65. Secondary: 60J60; 60J55; 60F05; 60F17. 1. Introduction In this paper we study the asymptotic behaviour of a normalized and time-changed multidimensional Wiener process. Our motivation comes from the study of solutions to the following type of parabolic Cauchy problems (1) \displaystyle\begin{aligned} \frac{\partial}{\partial t}u(t,x)&=\lambda(x)\Delta u(t,x),&\quad&t\geq 0,\;x\in\real^{d},\\ u(0,x)&=f(x),&\quad&t=0,\;x\in\real^{d},\end{aligned} where the coefficient \lambda can be irregular and even degenerate. Following the suggestions of Yuri Kondratiev, we studied in the joint paper [ 9 ] the Cauchy problem ( 1 ) in dimension one, d=1 . Here we are going to generalize the problem ( 1 ) to higher dimensions. Quite often it is assumed in such problems that the diffusion coefficient is uniformly elliptic. In the situation considered here this reduces to the assumption that \lambda(x) is bounded away from zero; degenerate problems are often treated with the method of vanishing viscosity, see e.g. Oleĭnik & Radkevič [ 12 , § III.2] and Bogachev et al. [ 2 , §6.7.(ii)] , and the literature mentioned in these references. An alternative method to study such problems is to use a probabilistic approach. Assume that we can construct a Markov process X , starting from any point x\in\real^{d} , and with infinitesimal generator \lambda(x)\Delta . In this situation, we have the following stochastic representation of the parabolic Cauchy problem ( 1 ): \displaystyle u(t,x)=\mathds{E}^{x}[f(X_{t})]. Yet, a direct construction of this process may be difficult (if it is possible at all), if one wants to use stochastic differential equations (SDEs) in the case when the diffusion coefficient \sqrt{\lambda(x)} is irregular and degenerate. Typically, diffusions in random media lead to random coefficients \lambda=\lambda(\omega,x) , which have an additive form with respect to the points in the medium. This structure corresponds to the energy of an diffusing particle in the medium. Such coefficients cannot be uniformly positive, in general. A standard reference for stochastic differential equations with singular coefficients is Cherny & Engelbert [ 4 ] . A possible way out is offered by the theory of Dirichlet forms, see e.g. Fukushima et al. [ 6 ] , but the resulting process may be defined only up to an exceptional set. Just as in the one-dimensional case, see [ 9 ] , we will use a different approach, namely additive functionals and random time changes. Recall the following statement which is proved differently and for different purposes in Böttcher et al. [ 3 ] and Kurtz [ 10 ] . Proposition 1.1 . Let \left\{X_{t},\,t\geq 0\right\} be a d -dimensional Feller process with natural filtration \mathcal{F}^{X} , \mathcal{F}_{t}^{X}=\sigma(X_{s},s\leq t) , and generator (A,\mathcal{D}(A)) . Assume that the function \lambda\in C_{b}(\real^{d}) is real-valued and strictly positive. Set \displaystyle S_{X}(t,\omega):=\int_{0}^{t}\frac{ds}{\lambda(X_{s}(\omega))} and denote by \tau_{t}(\omega)=\inf\{u>0:S_{X}(u,\omega)>t\} (\inf\emptyset:=+\infty) the generalized right-continuous inverse. Then the time-changed process \left\{X_{\tau_{t}},\,t\geq 0\right\} with filtration \left\{\mathcal{F}_{\tau_{t}}^{X},\,t\geq 0\right\} is again a Feller process, and its generator is the closure of (\lambda(\cdot)A,\mathcal{D}(A)) . Due to the different recurrence and transience behaviour of one- and multi-dimensional diffusion processes, we restricted ourselves in the paper [ 9 ] to the case d=1 ; we will now consider the case where X_{t}=B_{t} is a multidimensional Wiener process. Throughout, we assume that \lambda:\real^{d}\to[0,\infty) is a positive measurable function, which is Lebesgue a.e. strictly positive, i.e. \mathrm{Leb}\{x:\lambda(x)=0\}=0 . In our main result, Theorem 3.4 , we do not require that \lambda is bounded or bounded away from zero. However, in fact we assume that \lambda(x) is bounded away from zero for |x|\gg 1 , since the limit exists and is strictly positive, see assumption (vi) . Since B_{t} has a transition density w.r.t. Lebesgue measure, the a.e. strict positivity of \lambda guarantees that S_{B}(t,\omega) is a.s. strictly increasing and continuous; thus, \tau_{t} is also strictly increasing and continuous. Finally, we assume that 1/\lambda is locally integrable, but not necessarily of class L^{1}(\real^{d},dx) . Note that under these assumptions, the time-changed process B_{\tau_{t}} is both a diffusion process and a local martingale w.r.t. the filtration \mathcal{F}_{\tau_{t}} ; in fact, writing Y_{t}=B_{\tau_{t}} , we have (2) \displaystyle Y_{t}=Y_{0}+\int_{0}^{t}\sqrt{\lambda(Y_{s})}\,d\tilde{B}_{s}, with some d -dimensional Wiener process \tilde{B} . Our aim is to establish functional limit theorems for the process B_{\tau_{t}} without resorting to the martingale representation ( 2 ) or the Feller property and the form of its generator; instead, we want to focus on the pathwise representation as superposition of two stochastic processes. Using this approach, we need neither that the time change \tau and the original process X are independent (e.g. as in Bochner’s subordination) nor that \lambda is regular (as in the Feller case) or non-degenerate (as in the SDE approach). There exists a substantial literature on random-time changes, using various perspectives. Let us mention a few of them: The monograph by Fukushima et al. [ 6 ] treats time changes of symmetric Markov processes, and Harlamov [ 7 ] describes the representation of semi-Markov processes in the form of a Markov process, transformed by a time change. The paper [ 11 ] discusses the asymptotic behaviour of the standard one-dimensional Brownian motion time changed by the (generalized) inverse of an (independent) subordinator. A detailed bibliography on the properties of time-changed processes can be found in the paper [ 1 ] . Our paper is organized as follows. Since it is important to have a properly defined time change for any t>0 , we study in Section 2 the asymptotic behaviour of some additive functionals of integral type for a multi-dimensional Brownian motion, and we introduce conditions that ensure the existence of the time change \tau_{t} for any t>0 . Section 3 contains the main results, namely functional limit theorems for a time-changed multidimensional Wiener process. In Section 3.1 we briefly recall (a modification of) a theorem on weak convergence of superpositions of stochastic processes from Silvestrov [ 13 ] . This is then applied to our setting, first (in Section 3.2 ) with the simplifying assumption that \lambda is separated from zero. Then we remove this restriction, assuming instead that

Read the complete text