Moderate Deviation Principles for Stochastic Differential Equations in Fast-Varying Markovian Environment
Hongjiang Qian · 2025
Read Moderate Deviation Principles for Stochastic Differential Equations in Fast-Varying Markovian Environment by Hongjiang Qian (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-2512-00236.txt.
Introduction
Many systems mix a slow, noisy process with a fast, jumpy environment. Hongjiang Qian proves moderate deviation principles for such stochastic differential equations in 2025. The slow process carries small noise, while the fast process is a rapidly changing pure jump process on finitely many states.
Moderate deviation principles describe the probability of rare fluctuations that are larger than typical noise but smaller than extreme events. They refine both central limit behavior and large deviation theory. Results of this kind support risk estimates in physics, finance, and engineering wherever two time scales collide. Two time scales appear in climate, biology, and markets, so the principles travel widely.
Catalog record
| CREATOR | Hongjiang Qian |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 192 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-00236
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Moderate Deviation Principles for Stochastic Differential Equations in Fast-Varying Markovian Environment Hongjiang Qian Abstract In this paper, we proved moderate deviation principles for a fully coupled two-time-scale stochastic systems, where the slow process is given by stochastic differential equations with small noise, while the fast process is a rapidly changing purely jump process on finite state space. The system is fully coupled in that the drift and diffusion coefficients of the slow process, as well as the jump distribution of the fast process, depend on states of both processes. Moreover, the diffusion component in the slow process can be degenerate. Our approach is based on the combination of the weak convergence method from [A. Budhiraja, P. Dupuis, and A. Ganguly, Electron. J. Probab. 23 (2018), pp. 1-33; Ann. Probab. 44 (2016), pp. 1723-1775] with Poisson equation for the fast-varying purely jump process. Keywords . Moderate deviations; switching diffusion; jump diffusion; Markov chain; weak convergence; (nonlocal) Poisson equations. AMS Subject Classification (2020): . 60F10, 60J10, 60K37, 60H10. 1 Introduction In this paper, we study moderate deviation principles (MDPs) for a fully coupled two-time-scale stochastic system (X^{\varepsilon}(\cdot),Y^{\varepsilon}(\cdot)) . The X^{\varepsilon} is a d -dimensional stochastic process given as the solution of the following stochastic differential equations: dX^{\varepsilon}(t)=b(X^{\varepsilon}(t),Y^{\varepsilon}(t))dt+\sqrt{\varepsilon}\sigma(X^{\varepsilon}(t),Y^{\varepsilon}(t))dW(t),\ X^{\varepsilon}(0)=x_{0},Y^{\varepsilon}(0)=y_{0}, (1.1) where \varepsilon is the small parameter, W(\cdot) represents a d -dimensional Brownian motion. The functions b:{\mathbb{R}}^{d}\times{\mathbb{L}}\rightarrow{\mathbb{R}}^{d} and \sigma:\mathbb{R}^{d}\times{\mathbb{L}}\rightarrow{\mathbb{R}}^{d\times d} are continuous functions satisfying suitable conditions, which will be specified later. The Y^{\varepsilon} is a fast-varying jump process defined over a finite state space {\mathbb{L}}=\{1,\dots,|{\mathbb{L}}|\} , and its transition probabilities are determined by q:{\mathbb{L}}\times{\mathbb{L}}\times{\mathbb{R}}^{d}\to[0,1] , which depends on both X^{\varepsilon} and Y^{\varepsilon} . Specifically, the transition rates of Y^{\varepsilon} are defined as: \begin{array}[]{rl}{\mathbb{P}}(Y^{\varepsilon}(t+\Delta)=j|Y^{\varepsilon}(t)=i,X^{\varepsilon}(t)=x)=\left\{\begin{array}[]{ll}\varepsilon^{-1}q_{ij}(x)\Delta+o(\Delta),&\!\displaystyle\text{ if }j\neq i,\\ 1+\varepsilon^{-1}q_{ii}(x)\Delta+o(\Delta),&\!\displaystyle\text{ otherwise}.\end{array}\right.\end{array} Here q_{ij}(x) is the shorthand of q(i,j,x) for x\in{\mathbb{R}}^{d} , and Q(x):=(q_{ij}(x))_{i,j\in{\mathbb{L}}} is the generator of a Markov chain. The process Y^{\varepsilon} is also called state-dependent regime-switching process and is widely used to model the discrete random events in systems such as manufacturing, product planing, queuing networks, Monte Carlo simulations, and wireless communications, and so on. We refer to the work [ 40 ] for its comprehensive study on switching process. A key feature of system ( 1.1 ) is its combination of continuous dynamics (via X^{\varepsilon} ) and discrete events (via Y^{\varepsilon} ). The process X^{\varepsilon} evolves on a slower timescale compared to Y^{\varepsilon} , which oscillates rapidly due to the timescale factor \varepsilon^{-1} . Under suitable conditions, the fast process Y^{\varepsilon} , for fixed x\in{\mathbb{R}}^{d} , admits a unique invariant measure \mu^{x}=\{\mu_{i}^{x}\}_{i\in{\mathbb{L}}} , leading to the well-known averaging principle. That is, X^{\varepsilon} converges to the solution of an averaged equation: d\bar{X}(t)=\bar{b}(\bar{X}(t))dt,\quad\bar{X}(t)=x_{0}, (1.2) in various sense, where \bar{b}(x):=\sum_{i\in{\mathbb{L}}}b(x,i)\mu_{i}^{x} ; see recent work [ 27 ] on averaging principle for two time-scale regime-switching diffusions. For x\in{\mathbb{R}}^{d} and i,j\in{\mathbb{L}} , we sometimes write b(x,i),\sigma(x,i),q(i,j,x) as b_{i}(x),\sigma_{i}(x),q_{ij}(x) for notation convenience. For each (x,i,j)\in{\mathbb{R}}^{d}\times{\mathbb{L}}\times{\mathbb{L}} , we let E_{ij}(x) be the interval [0,q_{ij}(x)] . Define \begin{array}[]{rl}&\!\displaystyle\zeta:=\sup_{(x,i,j)\in{\mathbb{R}}^{d}\times{\mathbb{L}}\times{\mathbb{L}}}|q_{ij}(x)|+1<\infty.\end{array} We denote by \lambda=\lambda_{\zeta} the Lebesgue measure on the space ([0,\zeta],\mathcal{B}([0,\zeta])) . In this paper, we focus on deviations of X^{\varepsilon}(t) from the averaged solution \bar{X}(t) as \varepsilon\to 0 , specifically on analyzing the asymptotic behavior of trajectory: \eta^{\varepsilon}(t)=\frac{X^{\varepsilon}(t)-\bar{X}(t)}{\sqrt{\varepsilon}h(\varepsilon)},\quad t\in[0,T], (1.3) as \varepsilon\to 0 , where h(\varepsilon) is the deviation scale satisfying h(\varepsilon)\to+\infty,\quad\sqrt{\varepsilon}h(\varepsilon)\to 0,\text{ as }\varepsilon\to 0. (1.4) The result concerning to decay rate of probabilities for \eta^{\varepsilon} is called moderate deviation principles. It bridges the gap between central limit theorems (CLTs) and large deviation principles (LDPs). If h(\varepsilon)=1 , the limiting behavior of \eta^{\varepsilon}(t) falls in the domain of CLTs; see [ 37 ] . For h(\varepsilon)=\varepsilon^{-1/2} , the limiting behavior of \eta^{\varepsilon}(t) leads to large deviation principles in [ 6 ] , where the authors established the large deviation principle for {X^{\varepsilon}} in C([0,T];{\mathbb{R}}^{d}) as \varepsilon\rightarrow 0 , using the weak convergence method [ 9 ] . Define the function \ell:[0,\infty)\to[0,\infty) by \ell(x)=x\log x-x+1 . Let \begin{array}[]{rl}{\mathbb{T}}&\!\!\!:=\{(i,j)\in{\mathbb{L}}\times{\mathbb{L}}:q_{ij}(x)>0\text{ for some }x\in{\mathbb{R}}^{d}\},\\ \mathcal{R}&\!\!\!:=\{\varphi=(\varphi_{ij})_{(i,j)\in{\mathbb{T}}}:\varphi_{ij}:[0,T]\times[0,\zeta]\to{\mathbb{R}}_{+}\text{ is a measurable map},(i,j)\in{\mathbb{T}}\}.\end{array} They demonstrated {X^{\varepsilon}} adheres to LDPs with speed \varepsilon^{-1} and rate function J(\cdot) given by \begin{array}[]{ll}&\!\displaystyle\!\!\!\!\!\!\!\!J(\xi)=\!\!\!\!\inf_{(u,\varphi,\pi)\in\mathcal{V}(\xi)}\bigg\{\sum_{i\in{\mathbb{L}}}\frac{1}{2}\int_{0}^{T}\|u_{i}(s)\|^{2}\pi_{i}(s)ds+\sum_{(i,j)\in{\mathbb{T}}}\int_{[0,T]\times[0,\zeta]}\ell(\varphi_{ij}(s,z))\pi_{i}(s)\lambda_{\zeta}(dz)ds\bigg\},\end{array} (1.5) where \mathcal{V}(\xi) is the collection of all \{u=(u_{i}),\varphi=(\varphi_{ij}),\pi=(\pi_{i})\}\in({\mathbb{M}}([0,T];{\mathbb{R}}^{d}))^{|{\mathbb{L}}|}\times\mathcal{R}\times{\mathbb{M}}([0,T];\mathcal{P}({\mathbb{L}})) , such that \int_{0}^{T}\|u_{i}(s)\|^{2}\pi_{i}(s)ds<\infty , and for each i\in{\mathbb{L}} , \begin{array}[]{rl}&\!\displaystyle\xi(t)=x_{0}+\sum_{j\in{\mathbb{L}}}\int_{0}^{t}b_{j}(\xi(s))\pi_{j}(s)ds+\sum_{j\in{\mathbb{L}}}\int_{0}^{t}\sigma_{j}(\xi(s))u_{j}(s)\pi_{j}(s)ds,\;t\in[0,T],\end{array} and \sum_{i\in{\mathbb{L}}}\pi_{i}(s)\Gamma_{ij}^{\varphi_{i,\cdot}(s,\cdot)}(\xi(s))=0,\quad\text{ for a.e. }s\in[0,T]\text{ and }j\in{\mathbb{L}}, (1.6) where, for \psi=\{\psi_{j}\}_{j\in{\mathbb{L}}} with each \psi_{j}:[0,\zeta]\to{\mathbb{R}}_{+} measurable, the \Gamma_{ij}^{\psi} is defined as \begin{array}[]{ll}\Gamma_{ij}^{\psi}(x)=\left\{\begin{array}[]{ll}\int_{E_{ij}(x)}\psi_{j}(z)\lambda_{\zeta}(dz),&\!\displaystyle\quad i\neq j,\\ -\sum_{y:y\neq j}\Gamma_{jy}^{\psi}(x),&\!\displaystyle\quad i=j.\end{array}\right.\end{array} (1.7) Here {\mathbb{M}}([0,T];{\mathbb{R}}^{d}),{\mathbb{M}}([0,T];\mathcal{P}({\mathbb{L}})) denote the space of measurable maps from [0,T] to {\mathbb{R}}^{d} and from [0,T] to \mathcal{P}({\mathbb{L}}) , respectively. The primary objective of this paper is to investigate moderate deviation principles (MDPs) for the process X^{\varepsilon} (or say large deviation principle for \eta^{\varepsilon} ), focusing on the intermediate regime between CLT and LDP. To the best of our knowledge, no prior results have been established for the MDP of X^{\varepsilon} in a fast-varying purely-jump random environment on a finite state space. We address this gap in current paper. For a general theory of large deviations, we refer to the works of [ 8 , 9 , 11 ] . The study of large and moderate deviation principles for two-time-scale (or multi-time-scale) stochastic dynamics has attracted considerable interest in the research community recent years. Early work on moderate deviations for diffusion processes can be traced back to [ 1 , 12 ] , which considered cases where the slow motion has no diffusion coefficients. In [ 14 ] , Guillin explored the MDPs for the diffusion process X^{\varepsilon} of the form ( 1.1 ), where the fast-varying environment Y^{\varepsilon} is modeled as an exponentially ergodic Markov process, independent of the Brownian motion W(\cdot) . Such work builds on [ 15 ] , which addresses MDPs for inhomogeneous functionals of exponentially ergodic Markov processes. When the fast-varying process is modeled by a finite-state homogeneous or non-homogeneous Markov chain, MDPs were developed in [ 17 ] for diffusion processes and in [ 34 ] for Langevin dynamics. The literature mentioned above generally assume that the fast-varying process is independent of the slow process. In contrast, coupled systems have also been investigated a lot. In [ 28 ] , Morse and Spiliopulos examined MDPs for fully coupled fast-slow diffusions using Poisson equations and weak convergence method. The Poisson equation approach can also be extended to study infinite dimensional fast-slow dynamics, where an elliptic Kolmogorov equation in Hilbert space replaces the role of the Poisson equation. Interested reader are referred to the work of [ 13 ] which studied MDPs for fast-slow stochastic reaction-diffusion equations, where the slow process has diffusion coefficients independent of the fast dynamics. The MDPs of fully coupled fast-slow stochastic reaction-diffusion equations still remain an open problem due to the lack of regularity results for elliptic Kolmogorov equations in infinite dimensional space. For further insights on LDPs and MDPs in two- or multi-time-scale stochastic dynamics, readers are referred to [ 38 , 39 , 25 , 18 , 16 , 20 , 32 , 24 , 19 ] and references therein. The MDPs for fully coupled stochastic differential equations where the fast process is given by a state-dependent regime-switching process is still unknown. Though we still rely on Poisson equation technique, the discrete feature of fast process makes analysis more difficult and results are different compared to that of fast-slow diffusions in [ 28 ]