betaFeedback

WorksMathematics

Averaging for multi-scale McKean-Vlasov equation

Abstract

We consider a system of Stochastic Differential Equations with two time scales, so called slow-fast system of SDEs. As customary when studying slow-fast dynamics, we seek a coarse grained description of the system, obtained, in the limit of scale separation, by 'averaging out' the fast component of the dynamics (so-called averaging limit). We study slow-fast systems where the fast variable is described by a process which is non-linear in the sense of McKean i.e. by an SDE whose coefficients depend on the law of the process itself, while the slow variable is a linear SDE (linear in the sense of McKean). We show that, when the fast variable is a McKean-Vlasov process and, additionally, the coefficients of the SDE depend on the position of the slow process, the limit of scale separation is not provided by the usual ansatz which one would expect by applying classical averaging paradigms. We identify the limit and provide complete proof.

Introduction

This paper is concerned with the study of the averaging problem for two-scale, coupled SDEs, in the case in which the fast degrees of freedom are described by a McKean-Vlasov process. More precisely, we consider the following system of SDEs, for the variables Xtε∈RdX_t^\varepsilon\in\mathbb{R}^d, Ytε∈RmY_t^\varepsilon\in\mathbb{R}^m:

dXtε=v(Xtε,Ytε) dt+2κ dBt,dYtε=1εb(Xtε,Ytε,νtε) dt+2σε dWt,νtε=L(Ytε),(1)\begin{aligned} \mathrm{d}X_t^\varepsilon&= v(X_t^\varepsilon,Y_t^\varepsilon)\,\mathrm{d}t+\sqrt{2\kappa}\,\mathrm{d}B_t,\\ \mathrm{d}Y_t^\varepsilon&= \frac{1}{\varepsilon}b(X_t^\varepsilon,Y_t^\varepsilon,\nu_t^\varepsilon)\,\mathrm{d}t+\sqrt{\frac{2\sigma}{\varepsilon}}\,\mathrm{d}W_t,\\ \nu_t^\varepsilon&= \mathcal{L}(Y_t^\varepsilon), \tag*{(1)} \end{aligned}

where νtε\nu_t^\varepsilon is the law of YtεY_t^\varepsilon, κ≥0\kappa\ge0 is a non-negative constant, (Bt)t≥0(B_t)_{t\ge0} and (Wt)t≥0(W_t)_{t\ge0} are two independent Brownian motions in Rd\mathbb{R}^d and Rm\mathbb{R}^m, respectively, and 0<ε≪10 < \varepsilon\ll1 is a parameter describing the scale separation between the slow variable XtεX_t^\varepsilon and the fast variable YtεY_t^\varepsilon. The initial conditions are taken to be deterministic:

X0ε=x0∈Rd,Y0ε=y0∈Rm,X_0^\varepsilon=x_0\in\mathbb{R}^d,\qquad Y_0^\varepsilon=y_0\in\mathbb{R}^m,

and the drift functions are smooth functions with v:Rd×Rm→Rdv:\mathbb{R}^d\times\mathbb{R}^m\to\mathbb{R}^d, b:Rd×Rm×P2(Rm)→Rmb:\mathbb{R}^d\times\mathbb{R}^m\times\mathcal{P}_2(\mathbb{R}^m)\to\mathbb{R}^m, where P2(Rm)\mathcal{P}_2(\mathbb{R}^m) denotes the space of probability measures on Rm\mathbb{R}^m with finite second moment. For the purposes of this paper, the important features of system (1.1) are: the fast process is a McKean-Vlasov process (i.e. the evolution of YtεY_t^\varepsilon is non-linear in the sense that it depends on the law of the process YtεY_t^\varepsilon itself);1 the drift coefficient bb depends on the position of the slow process; and the slow process is described by an SDE (as opposed to an ODE), i.e. κ>0\kappa> 0 is a strictly positive constant. To be more precise we treat both the case κ=0\kappa= 0 and the case κ>0\kappa> 0, see Section 5, but the most relevant one to us is the latter. As customary, we wish to identify the limit of the slow variable XtεX^\varepsilon_t as ε\varepsilon goes to zero. This limit is routinely studied using the framework of stochastic averaging. The purpose of this paper is to show that, in the setting of system (1), the limit which one would expect to obtain by using the usual averaging ansatz is not the correct one. In particular, if one was to use the classical averaging ansatz, then the limit process would be described by a linear SDE. We prove that, as long as κ>0\kappa> 0, the limit dynamics is described by a McKean–Vlasov process. More precisely, in Theorem 3.7 we show that, as ε\varepsilon tends to zero, the slow process (Xtε)t≥0(X^\varepsilon_t)_{t\geq0} converges to the process (Xˉt)t≥0(\bar{X}_t)_{t\geq0}, solution of the following McKean–Vlasov SDE:

{dXˉt=vˉ(Xˉt,ρˉt) dt+2κ dBt,ρˉt=L(Xˉt),Xˉ0=x0a.s.,(2)\begin{cases} \mathrm{d}\bar{X}_t = \bar{v}(\bar{X}_t,\bar{\rho}_t)\,\mathrm{d}t + \sqrt{2\kappa}\,\mathrm{d}B_t,\\ \bar{\rho}_t = \mathcal{L}(\bar{X}_t),\\ \bar{X}_0 = x_0 \quad\text{a.s.}, \tag*{(2)} \end{cases}

where the averaged drift vˉ:Rd×P(Rd)→Rd\bar{v}:\mathbb{R}^d \times\mathcal{P}(\mathbb{R}^d) \to\mathbb{R}^d is given by

vˉ(x,ρ):=∫Rmv(x,y)Πx,ρ(dy),(3)\bar{v}(x,\rho) := \int_{\mathbb{R}^m} v(x,y)\Pi^{x,\rho}(\mathrm{d}y), \tag*{(3)}

and, for each x∈Rdx \in\mathbb{R}^d and ρ∈P(Rd)\rho\in\mathcal{P}(\mathbb{R}^d), Πx,ρ\Pi^{x,\rho} is a probability measure on Rm\mathbb{R}^m, found by solving equation (15) below. We will discuss below and throughout the paper the precise meaning of the measure Πx,ρ\Pi^{x,\rho}, and compare with the limit found when κ=0\kappa= 0. For now we note that if κ=0\kappa= 0 the dynamics (2) still makes sense (it is just the case that ρˉt\bar{\rho}_t will be a dirac); moreover, we will show at the start of § 2 that if κ=0\kappa= 0 then (1.2)–(1.3) reduces to the the averaged equation that one would expect (namely to (1.9)–(1.11) below). Hence (1.2)–(1.3) is truly only different from the expected averaging limit when κ>0\kappa> 0.

In order to better explain the result of this paper, let us start by recalling the averaging paradigm, in the more standard context of linear SDEs. To this end, consider the system

{dXtε=v(Xtε,Ytε) dt+2κ dBt,dYtε=1εb(Xtε,Ytε) dt+2ε dWt,(4)\begin{cases} \mathrm{d}X^\varepsilon_t = v(X^\varepsilon_t,Y^\varepsilon_t)\,\mathrm{d}t + \sqrt{2\kappa}\,\mathrm{d}B_t,\\ \mathrm{d}Y^\varepsilon_t = \frac{1}{\varepsilon}b(X^\varepsilon_t,Y^\varepsilon_t)\,\mathrm{d}t + \sqrt{\frac{2}{\varepsilon}}\,\mathrm{d}W_t, \tag*{(4)} \end{cases}

which (with some abuse of notation) can be viewed as a special case of (1), when the drift coefficient of the fast process does not depend on the law of the process YεY^\varepsilon itself (and we have fixed σ=1\sigma= 1, which will be the case throughout the paper, as σ\sigma does nos not play any role to our purposes). The averaging method can be intuitively understood by observing that, as ε→0\varepsilon\to0, the process YεY^\varepsilon will evolve much faster than the process XεX^\varepsilon;3 in particular, the former will have reached equilibrium (assuming it has one) while the slow process has remained substantially unchanged. To formalise this idea one introduces an auxiliary process, Y(ε,x)Y^{(\varepsilon,x)}, defined as the solution of the following SDE:

dYt(ε,x)=1εb(x,Yt(ε,x)) dt+2ε dWt.(5)\mathrm{d}Y^{(\varepsilon,x)}_t = \frac{1}{\varepsilon}b(x,Y^{(\varepsilon,x)}_t)\,\mathrm{d}t + \sqrt{\frac{2}{\varepsilon}}\,\mathrm{d}W_t. \tag*{(5)}

Note that the above equation is obtained by "freezing" the value of the slow variable in the equation (1) for the fast process. Since Y(ε,x)Y^{(\varepsilon,x)} is decoupled from the evolution of the slow process, due to the time-scaling properties of Brownian motion, when looking at process (5), letting t→∞t \to\infty for ε\varepsilon fixed is the same as letting ε→0\varepsilon\to0, for each tt fixed. Hence it is typical to fix the value of ε\varepsilon and consider the so-called frozen process, namely:

dYt(x)=b(x,Yt(x)) dt+2 dWt.(6)\mathrm{d}Y^{(x)}_t = b(x,Y^{(x)}_t)\,\mathrm{d}t + \sqrt{2}\,\mathrm{d}W_t. \tag*{(6)}

Notice that the above is a family of processes parametrised by x∈Rdx \in\mathbb{R}^d, and xx here is fixed deterministically.

Assuming that, for each xx fixed, the process Y(x)Y^{(x)} admits a unique invariant measure πx\pi^x – which, we recall, is the solution of the following partial differential equation:

−∇y⋅(b(x,y)πx)+Δyπx=0;(7)-\nabla_y \cdot\left(b(x,y)\pi^x\right) + \Delta_y\pi^x = 0; \tag*{(7)}

and further more assuming that Y(x)Y^{(x)} is ergodic with respect to such a measure, in the sense that

lim⁡t→∞1t∫0tf(Ys(x)) ds=∫Rmf(y)πx(dy),\lim_{t\to\infty}\frac{1}{t}\int_0^t f(Y_s^{(x)})\,\mathrm{d}s = \int_{\mathbb{R}^m} f(y)\pi^x(\mathrm{d}y),

References

  1. [1]Zachary William Bezemek and Konstantinos Spiliopoulos. Rate of homogenization for fully-coupled McKean–Vlasov SDEs. Stochastics and Dynamics, 23(2):2350013, 2023.arxiv.org/abs/2202.07753
  2. [2]René Carmona and Xiuneng Zhu. A probabilistic approach to mean field games with major and minor players. The Annals of Applied Probability, 26(3):1535 – 1580, 2016.arxiv.org/abs/1409.7141
  3. [3]Sandra Cerrai and Alessandra Lunardi. Averaging principle for nonautonomous slow-fast systems of stochastic reaction-diffusion equations: the almost periodic case. SIAM Journal on Mathematical Analysis, 49(4):2843–2884, 2017.DOI
  4. [4]Louis-Pierre Chaintron and Antoine Diez. Propagation of chaos: a review of models, methods and applications. i. models and methods. arXiv preprint arXiv:2203.00446, 2022.DOI
  5. [5]Dan Crisan, Paul Dobson, Benjamin D Goddard, Michela Ottobre, and Iain Souttar. Poisson equations with locally-lipschitz coefficients and uniform in time averaging for stochastic differential equations via strong exponential stability. Annales de l’Institut Henri Poincaré (B) Probabilités et statistiques, volume 62, pages 667–721. Institut Henri Poincaré, 2026.
  6. [6]Manh Hong Duong, Carsten Hartmann, and Michela Ottobre. Coarse graining of stochastic differential equations: averaging and projection method. arXiv preprint arXiv:2506.14939, 2025.arxiv.org/abs/2506.14939
  7. [7]Stewart N Ethier and Thomas G Kurtz. Markov processes: characterization and convergence. John Wiley & Sons, 2009.
  8. [8]M. I. Freidlin and A. D. Wentzell. Random perturbations of dynamical systems, volume 260 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, New York, 1984. Translated from the Russian by Joseph Szücs.
  9. [9]Wei Hong, Ge Li, and Shihu Li. Multi-scale McKean–Vlasov SDEs: Moderate deviation principle in different regimes. Annales de l’Institut Henri Poincaré, Probabilités et Statistiques, 62(2):1112–1163, 2026.arxiv.org/abs/2306.11569
  10. [10]Wei Hong, Shihu Li, and Wei Liu. Strong convergence rates in averaging principle for slow–fast McKean–Vlasov SPDEs. Journal of Differential Equations, 316:94–135, 2022.
  11. [11]Wei Hong, Wei Liu, Luhan Yang, and Shiyuan Yang. Averaging principle for multiscale stochastic pdes with fully local monotone coefficients. Discrete and Continuous Dynamical Systems-S, 18(10):2791–2809, 2025.
  12. [12]Yaru Jiang and Meiling Zhao. Averaging principle for the fast–slow McKean–Vlasov stochastic differential equations driven by mixed fractional brownian motion. Journal of Applied Probability, 62(4):1260–1279, 2025.
  13. [13]Rafail Z. Khas’minskii. The averaging principle for stochastic differential equations. Problemy Peredachi Informatsii, 4(2):86–87, 1968.
  14. [14]Frédéric Legoll and Tony Lelievre. Effective dynamics using conditional expectations. Nonlinearity, 23(9):2131–2163, 2010.DOI
  15. [15]Yun Li, Fuke Wu, and Longjie Xie. Poisson equation on wasserstein space and diffusion approximations for multiscale mckean–vlasov equation. SIAM Journal on Mathematical Analysis, 56(2):1495–1524, 2024.
  16. [16]Sean P Meyn and Richard L Tweedie. Stability of Markovian processes III: Foster-Lyapunov criteria for continuous-time processes. Advances in Applied Probability, 25(3):518–548, 1993.DOI
  17. [17]Kallenberg Olav. Foundations of modern probability. Probability Theory and Stochastic Modelling, 2021.
  18. [18]E Pardoux and A Yu Veretennikov. On poisson equation and diffusion approximation 2. The Annals of Probability, 31(3):1166–1192, 2003.DOI
  19. [19]Étienne Pardoux and A Yu Veretennikov. On the poisson equation and diffusion approximation. i. Annals of probability, pages 1061–1085, 2001.DOI
  20. [20]Etienne Pardoux and A Yu Veretennikov. On the poisson equation and diffusion approximation 3. 2005.DOI
  21. [21]Grigorios A Pavliotis and Andrea Zanoni. Linearization of ergodic mckean sdes and applications. Nonlinearity, 38(8):085008, 2025.arxiv.org/abs/2501.13655
  22. [22]Grigoris A Pavliotis and Andrew Stuart. Multiscale methods: averaging and homogenization. Springer Science & Business Media, 2008.DOI
  23. [23]Michael Röckner, Xiaobin Sun, and Yingchao Xie. Strong convergence order for slow–fast McKean–Vlasov stochastic differential equations. Annales de l’Institut Henri Poincaré, Probabilités et Statistiques, 57(1):547–576, 2021.arxiv.org/abs/1909.07665
  24. [24]Yangyang Shi, Hui Liu, and Hongjun Gao. Weak and strong averaging principle for non-autonomous slow–fast McKean–Vlasov SDEs with almost periodic coefficients. Mathematical Methods in the Applied Sciences, 48(16):15111–15131, 2025.
  25. [25]Cédric Villani et al. Optimal transport: old and new, volume 338. Springer, 2009.
  26. [26]E Wennian. Principles of multiscale modeling, volume 1534. Cambridge University Press, Cambridge, 2011.
  27. [27]Hao Wu, Junhao Hu, and Chenggui Yuan. Large deviation for slow–fast McKean–Vlasov stochastic differential equations driven by fractional brownian motions and brownian motions. Stochastics and Dynamics, 24(6):2450044, 2024.
  28. [28]Jie Xiang and Huijie Qiao. Averaging principles for nonautonomous multiscale mckean-vlasov stochastic systems. arXiv preprint arXiv:2606.12820, 2026.arxiv.org/abs/2606.12820
  29. [29]Longjie Xie and Xicheng Zhang. Uniform-in-time diffusion approximations for multiscale stochastic systems. arXiv preprint arXiv:2604.00692, 2026.arxiv.org/abs/2604.00692
  30. [30]Jie Xu, Juanfang Liu, Jicheng Liu, and Yu Miao. Strong Averaging Principle for Two-Time-Scale Stochastic McKean–Vlasov Equations. Applied Mathematics & Optimization, 84(Suppl 1):837–867, 2021.DOI

Paper details

vlnc:W_96at287a
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

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