Averaging principles for time-inhomogeneous multi-scale SDEs with partially dissipative coefficients

Xiaobin Sun, Jian Wang, Yingchao Xie · 2025

Read Averaging principles for time-inhomogeneous multi-scale SDEs with partially dissipative coefficients by Xiaobin Sun, Jian Wang, Yingchao Xie (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-2506-18558.txt.

Introduction

Xiaobin Sun, Jian Wang, and Yingchao Xie prove averaging rules for noisy systems that run on two clocks at once, where the fast clock's rules themselves change over time. Many real systems mix a slow trend with fast jitter; averaging replaces the jitter by its mean effect. Their 2025 paper handles the harder case where the fast drift depends on time and only partly pulls back. When the slow part's noise ignores the fast part, the slow path converges strongly to the averaged equation. Otherwise they prove weak convergence and name the limit's law.

Such theorems justify the shortcuts modelers already take in climate, finance, and engineering. The authors work under stated assumptions on the coefficients. Their result extends averaging to systems whose fast layer ages as it spins. Rigor here protects every simulation built on the shortcut.

Catalog record

CREATORXiaobin Sun, Jian Wang, Yingchao Xie
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE102 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-18558

Browse the full library · About the Tower of Babel

Full text

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

Averaging principles for time-inhomogeneous multi-scale SDEs with partially dissipative coefficients Xiaobin Sun, Jian Wang, Yingchao Xie Abstract . In this paper, we study averaging principles for a class of time-inhomogeneous stochastic differential equations (SDEs) with slow and fast time-scales, where the drift term in the fast component is time-dependent and only partially dissipative. Under asymptotic assumptions on the coefficients, we prove that the slow component ( X t ε ) t ≥ 0 (X^{\varepsilon}_{t})_{t\geq 0} converges strongly to the unique solution ( X ¯ t ) t ≥ 0 (\bar{X}_{t})_{t\geq 0} to an averaged SDE, when the diffusion coefficient in the slow component is independent of the fast component; on the other hand, we establish the weak convergence of ( X t ε ) t ≥ 0 (X_{t}^{\varepsilon})_{t\geq 0} in the space C ⁡ ( [ 0 , T ] , ℝ n ) C([0,T];\mathbb{R}^{n}) and identify the limiting process by the martingale problem approach, when the diffusion coefficient of the slow component depends on the fast component. The proofs of strong and weak averaging principles are partly based on the study of the existence and uniqueness of an evolution system of measures for time-inhomogeneous SDEs with partially dissipative drift. 2000 Mathematics Subject Classification Primary 34D08, 34D25; Secondary 60H20 1. Background and main results 1.1. Background Multi-scale systems encompass diverse applications, where the associated processes evolve across varying temporal and spatial scales. They are prevalent in various fields such as climate modeling, material science, biology, and nonlinear oscillations, see e.g. [ 10 ] . Multi-scale systems frequently exhibit unconventional characteristics that defy common sense, to which using traditional, singular theoretical frameworks inadequate. Consequently, the exploration of multi-scale modeling systems becomes extremely important and indispensable. The mathematical model of multi-scale stochastic differential equations (SDEs) usually can be described as follows: \left\{\begin{array}[]{l}\displaystyle dX^{\varepsilon}_{t}=b(X^{\varepsilon}_{t},Y^{\varepsilon}_{t})\,dt+\sigma(X^{\varepsilon}_{t},Y^{\varepsilon}_{t})\,dB^{1}_{t},\quad X^{\varepsilon}_{0}=x\in\mathbb{R}^{n},\\ dY^{\varepsilon}_{t}={\varepsilon}^{-1}f(X^{\varepsilon}_{t},Y^{\varepsilon}_{t})\,dt+\varepsilon^{-1/2}g(X^{\varepsilon}_{t},Y^{\varepsilon}_{t})\,dB^{2}_{t},\quad Y^{\varepsilon}_{0}=y\in\mathbb{R}^{m},\\ \end{array}\right. where (B^{1}_{t})_{t\geq 0} and (B^{2}_{t})_{t\geq 0} are two independent d_{1} -dimensional and d_{2} -dimensional standard Brownian motions respectively, \varepsilon is a small and positive parameter describing the ratio of the time scales between the slow component (X^{\varepsilon}_{t})_{t\geq 0} and the fast component (Y^{\varepsilon}_{t})_{t\geq 0} , and the coefficients b\!:\mathbb{R}^{n}\times\mathbb{R}^{m}\to\mathbb{R}^{n} , \sigma\!:\mathbb{R}^{n}\times\mathbb{R}^{m}\to\mathbb{R}^{n\times d_{1}} , f\!:\mathbb{R}^{n}\times\mathbb{R}^{m}\to\mathbb{R}^{m} and g\!:\mathbb{R}^{n}\times\mathbb{R}^{m}\to\mathbb{R}^{m\times d_{2}} are Borel measurable. The techniques employed in this context are commonly known as averaging and homogenization methods, see e.g. [ 12 , 17 , 23 ] and the references therein. For example, as discussed in Khasminskii’s pioneer work [ 17 ] , it is assumed that there exist \bar{b}\!:\mathbb{R}^{n}\rightarrow\mathbb{R}^{n} and \bar{A}\!:\mathbb{R}^{n}\rightarrow\mathbb{R}^{n\times n} such that for all T\geq 0 , x\in\mathbb{R}^{n} and y\in\mathbb{R}^{m} , (1.1) \displaystyle\sup_{t\geq 0}\left|\frac{1}{T}\int^{t+T}_{t}\mathbb{E}b(x,Y^{x,y}_{s})\,ds-\bar{b}(x)\right|\leq\alpha(T)(1+|x|^{2}) and (1.2) \displaystyle\sup_{t\geq 0}\left|\frac{1}{T}\int^{t+T}_{t}\mathbb{E}(\sigma\sigma^{\ast})(x,Y^{x,y}_{s})\,ds-\bar{A}(x)\right|\leq\alpha(T)(1+|x|^{2}), where \alpha:\mathbb{R}_{+}:=[0,\infty)\to\mathbb{R}_{+} satisfies that \alpha(T)\to 0 as T\rightarrow\infty , and (Y^{x,y}_{t})_{t\geq 0} is the unique solution to the following frozen SDE (1.3) dY^{x,y}_{t}=f(x,Y^{x,y}_{t})\,dt+g(x,Y^{x,y}_{t})\,dB^{2}_{t},\quad Y^{x,y}_{0}=y. Then, the (weak) averaging principle claims that, for any T>0 , the slow component (X^{\varepsilon}_{t})_{t\in[0,T]} converges weakly to the solution (\bar{X}_{t})_{t\in[0,T]} of the following averaged SDE (1.4) d\bar{X}_{t}=\bar{b}(\bar{X}_{t})\,dt+\bar{\sigma}(\bar{X}_{t})\,d\bar{B}_{t},\quad X_{0}=x as \varepsilon\to 0 , where \bar{\sigma}(x):=(\bar{A}(x))^{1/2} , i.e., \bar{\sigma}^{2}(x)=\bar{A}(x) for all x\in\mathbb{R}^{n} , and (\bar{B}_{t})_{t\geq 0} is a standard n -dimensional Brownian motion. Undoubtedly, a key step for understanding the averaging principle lies in comprehending the corresponding averaged SDE ( 1.4 ), and the most important thing is how to accurately describe the averaged coefficients involved in. In particular, the averaging coefficients depend on the invariant measure of the frozen SDE ( 1.3 ). For example, if the frozen SDE ( 1.3 ) admits a unique invariant measure \mu^{x} , then \bar{b}(x)=\int_{\mathbb{R}^{m}}b(x,y)\,\mu^{x}(dy),\quad\bar{A}(x)=\int_{\mathbb{R}^{m}}(\sigma\sigma^{*})(x,y)\,\mu^{x}(dy). Currently most research results focus on the case where the frozen SDEs are time-homogeneous, because of the rich theoretical research on the existence and uniqueness of invariant measures in this case, see e.g. [ 2 , 3 , 6 , 14 , 15 , 16 , 18 , 20 , 24 , 26 , 27 , 28 , 30 , 32 , 35 ] . The solution to a time-nonhomogeneous frozen SDE exhibits richer dynamic characteristics, leading to the understanding of the so-called equilibrium point extremely complex. More precisely, consider the following time-dependent SDE: dY^{x,y}_{t}=f(t,x,Y^{x,y}_{t})\,dt+g(t,x,Y^{x,y}_{t})\,dB^{2}_{t}, which corresponds to the following multi-scale nonautonomous SDE: (1.5) \left\{\begin{array}[]{l}\displaystyle dX^{\varepsilon}_{t}=b(X^{\varepsilon}_{t},Y^{\varepsilon}_{t})\,dt+\sigma(X^{\varepsilon}_{t},Y^{\varepsilon}_{t})\,dB^{1}_{t},\quad X^{\varepsilon}_{0}=x\in\mathbb{R}^{n},\\ \displaystyle dY^{\varepsilon}_{t}=\varepsilon^{-1}f(t/\varepsilon,X^{\varepsilon}_{t},Y^{\varepsilon}_{t})\,dt+\varepsilon^{-1/2}g(t/\varepsilon,X^{\varepsilon}_{t},Y^{\varepsilon}_{t})\,dB^{2}_{t},\quad Y^{\varepsilon}_{0}=y\in\mathbb{R}^{m}.\end{array}\right. As far as we know, there are very limited existing works on the time-inhomogeneous stochastic system such like ( 1.5 ) . If the coefficients f and g are time periodic, Wainrib [ 33 ] and Uda [ 31 ] studied the strong averaging principle for the stochastic system ( 1.5 ) when \sigma\equiv 0 . Cerrai and Lunardi [ 4 ] explored the weak averaging principle for time-inhomogeneous slow-fast stochastic reaction diffusion equations, when the coefficients of the fast component satisfy the almost periodic condition. For uniformly dissipative coefficients f and g , recently we used the technique of nonautonomous Poisson equations to study the strong and weak averaging principle for ( 1.5 ) in [ 29 ] . 1.2. Main results In this paper, we are interested in the following nonautonomous stochastic system (1.6) \left\{\begin{array}[]{l}\displaystyle dX^{\varepsilon}_{t}=b(t/\varepsilon,X^{\varepsilon}_{t},Y^{\varepsilon}_{t})\,dt+\sigma(t/\varepsilon,X^{\varepsilon}_{t},Y^{\varepsilon}_{t})\,dW^{1}_{t},\quad X^{\varepsilon}_{0}=x\in\mathbb{R}^{n},\\ \displaystyle dY^{\varepsilon}_{t}=\varepsilon^{-1}f(t/\varepsilon,X^{\varepsilon}_{t},Y^{\varepsilon}_{t})\,dt+\varepsilon^{-1/2}\,dW^{2}_{t},\quad Y^{\varepsilon}_{0}=y\in\mathbb{R}^{m}.\end{array}\right. Here, (W^{1}_{t})_{t\geq 0} and (W^{2}_{t})_{t\geq 0} are mutually independent d -dimensional and m -dimensional standard Brownian motions respectively, on a complete probability space (\Omega,\mathscr{F},\mathbb{P}) with the natural filtration \{\mathscr{F}_{t},t\geq 0\} generated by (W^{1}_{t})_{t\geq 0} and (W^{2}_{t})_{t\geq 0} . The coefficients b:\mathbb{R}_{+}\times\mathbb{R}^{n}\times\mathbb{R}^{m}\rightarrow\mathbb{R}^{n} , \sigma:\mathbb{R}_{+}\times\mathbb{R}^{n}\times\mathbb{R}^{m}\rightarrow\mathbb{R}^{n\times d} and f:\mathbb{R}_{+}\times\mathbb{R}^{n}\times\mathbb{R}^{m}\rightarrow\mathbb{R}^{m} are measurable. Meanwhile, the coefficient f(t,x,y) enjoys the partial dissipative property with respect to y ; see 1.9 below. Such property seems to be novel within the context of multi-scale SDEs and even in the time-independent framework. We next provide the assumptions on the coefficients of the system ( 1.6 ), and show the main results of the paper, including the strong and weak averaging principles for ( 1.6 ). First of all, the following three conditions are always assumed throughout the paper. A1 . There exists a constant C_{0}>0 such that for any t\geq 0 , x_{1},x_{2}\in\mathbb{R}^{n} and y_{1},y_{2}\in\mathbb{R}^{m} , (1.7) \displaystyle|b(t,x_{1},y_{1})-b(t,x_{2},y_{2})|+\|\sigma(t,x_{1},y_{1})-\sigma(t,x_{2},y_{2})\|\leq C_{0}(|x_{1}-x_{2}|+|y_{1}-y_{2}|),\vskip 8.53581pt (1.8) \displaystyle|b(t,x_{1},y_{1})|\leq C_{0}(1+|x_{1}|+|y_{1}|),\quad\|\sigma(t,x_{1},y_{1})\|\leq C_{0}(1+|x_{1}|). A2 . There exist constants C,K and r_{0}>0 such that for all t\geq 0 , x_{1},x_{2}\in\mathbb{R}^{n} and y_{1},y_{2}\in\mathbb{R}^{m} , (1.9) \begin{split}&\langle f(t,x_{1},y_{1})-f(t,x_{2},y_{2}),y_{1}-y_{2}\rangle\\ &\leq\begin{cases}C|y_{1}-y_{2}|^{2}+C|x_{1}-x_{2}||y_{1}-y_{2}|,\quad&|y_{1}-y_{2}|\leq r_{0},\\ -K|y_{1}-y_{2}|^{2}+C|x_{1}-x_{2}||y_{1}-y_{2}|,\quad&|y_{1}-y_{2}|>r_{0}.\end{cases}\end{split} Moreover, there exists a constant C_{*}>0 such that for all t\geq 0 and x\in\mathbb{R}^{n} , (1.10) \displaystyle|f(t,x,{\mathbf{0}})|\leq C_{*}(1+|x|). A3 . There exists \hat{b}:\mathbb{R}^{n}\times\mathbb{R}^{m}\rightarrow\mathbb{R}^{n} such that for any T\geq 0 , x\in\mathbb{R}^{n} and y\in\mathbb{R}^{m} , (1.11) \displaystyle\sup_{t\geq 0}\frac{1}{T}\left|\int^{t+T}_{t}(b(s,x,y)-\hat{b}(x,y))\,ds\right|\leq\phi_{1}(T)(1+|x|+|y|). Moreover, there exist k\geq 0 and \bar{f}:\mathbb{R}^{n}\times\mathbb{R}^{m}\rightarrow\mathbb{R}^{m} such that for all T\geq 0 , x\in\mathbb{R}^{n} and y\in\mathbb{R}^{m} , (1.12) \displaystyle|f(T,x,y)-\bar{f}(x,y)|\leq\phi_{2}(T)(1+|x|+|y|^{k}). Here, for all i=1,2 , \phi_{i}\geq 0 defined on \mathbb{R}_{+} satisfies \phi_{i}(T)\rightarrow 0 as T\to\infty . It is well known that under Assumptions 1 and 2 , the SDE ( 1.6 ) admits a unique strong solution (X_{t}^{\varepsilon},Y_{t}^{\varepsilon})_{t\geq 0} , e.g. see [ 19 , Theorem 3.1.1] . On the other hand, ( 1.9 ) is called a partially dissipative condition in the literature (see [ 11 , 22 ] ). It is obvious that the condition ( 1.9 ) is weaker than the classical uniformly dissipative condition. By applying the elementary inequality ab\leq\delta a^{2}+C_{\delta}b^{2} for all a,b\geq 0 and \delta>0 , we can see from ( 1.9 ) and ( 1.10 ) that for all t\geq 0 , x\in\mathbb{R}^{n} and y\in\mathbb{R}^{m} , (1.13) \begin{split}\langle f(t,x,y),y\rangle&=\langle f(t,x,y)-f(t,x,{\mathbf{0}}),y\rangle+\langle f(t,x,{\mathbf{0}}),y\rangle\\ &\leq-K|y|^{2}+(C+K)r_{0}^{2}+|f(t,x,{\mathbf{0}})||y|\\ &\leq-K|y|^{2}+(C+K)r_{0}^{2}+C_{*}(1+|x|)|y|\\ &\leq-K_{1}|y|^{2}+K_{2}(1+|x|^{2}),\end{split} where 0<K_{1}<K and

Read the complete text