Xue-Mei Li

Xue-Mei Li’s work lies at the vibrant intersection of probability theory, stochastic analysis, differential geometry (both finite and infinite dimensional), and multiscale dynamics, including fractional and non-Markovian systems. Combining deep geometric insights with rigorous analysis, she explores how noise “feels” curvature and topology, revealing how stochastic flows encode geometric and topological features and deepening our understanding of how randomness interacts with geometry within complex structures. She has forged remarkable collaborations, advancing the field through innovative applications of rough path theory to extend classical results to multiscale SDEs driven by rough or long-range dependent noise such as fractional Brownian motion and Volterra processes. By combining rough path theory with Malliavin calculus, she has extended large-scale dynamics and fluctuation theory for SPDEs to spatially long-range systems.

Among her many contributions are solving the longstanding open problem of strong completeness for SDEs by establishing fundamental criteria guaranteeing global smooth flows on non-compact manifolds; proving the derivative (BEL) formula for diffusion semigroups using an innovative martingale method; and coining and pioneering the study of strict local martingales. Her work in infinite-dimensional analysis and Malliavin calculus on manifolds includes profound contributions to the analysis of degenerate diffusion measures, proving Poincaré inequalities for loop measures, and obtaining sharp logarithmic estimates for heat kernels.

While on the topic of rough path, a recent progress is:

Rough Differential Equations

  • Strong Completeness of SDEs and Non-Explosion for RDEs with Coefficients Having Unbounded Derivatives , by Xue-Mei Li and Kexing Ying (2025). In this work, they establish a non-explosion result for rough differential equations (RDEs) where both the noise and drift coefficients, together with their derivatives, are allowed to grow at infinity. This extends beyond the standard assumptions in RDE theory, which typically require the driving vector fields and their derivatives up to a certain order to be bounded to ensure global well-posedness. Additionally, they prove the existence of a bi-continuous solution flow for stochastic differential equations (SDEs). In the case of RDEs with additive noise, they show that their result is optimal by providing a counterexample.
  • Coarse Curvatures

  • Coarse Ricci Curvature of Weighted Riemannian Manifolds (2023) , in Transactions of the AMS (2025). In this work, M. Arnaudon, Xue-Mei Li, and Benedikt Petico show how to recover the generalized Ricci curvature of a weighted Riemannian manifold using optimal transport techniques, inspired by Ollivier's notion of coarse Ricci curvature. Specifically, they prove that the generalized Ricci tensor, in the sense of Bakry and Emery, can be obtained asymptotically by examining how Wasserstein-1 distances between locally supported, normalized weighted volume measures change under small shifts in the manifold. As an application, they study random geometric graphs sampled from Poisson point processes with non-uniform intensity, demonstrating that their limiting coarse Ricci curvature converges to the manifold's generalized Ricci tensor. This bridges discrete curvature notions and smooth geometry via optimal transport and random sampling.

    Coarse Extrinsic Curvature of Riemannian Submanifolds , in European Journal of Mathematics (2025). Following the previous work, in this article, they introduce a novel concept of coarse extrinsic curvature for Riemannian submanifolds. This curvature is derived from Wasserstein-1 distances between probability measures supported in tubular neighbourhoods of a submanifold, providing new insights into the extrinsic geometry of isometrically embedded manifolds in Euclidean spaces. Their framework also offers methods to approximate mean curvature from statistical data, such as point clouds generated by Poisson point processes. This approach has potential applications in manifold learning and metric embedding theory, enabling geometric information to be inferred directly from empirical data.

    Together, these works establish a framework for understanding both intrinsic and extrinsic geometric quantities through coarse curvature concepts, optimal transport, and random sampling, offering new tools for geometric analysis, data science, and metric measure geometry.
  • Fluctuations of SHE and KPZ Equations with Long Range Dependent Noise

  • In these works, Xue_Mei Li et al study the large-scale behaviour of nonlinear stochastic heat equations (SHE) and the KPZ equation in dimensions \( d \ge 3 \) driven by multiplicative Gaussian noise that is white in time and spatially coloured with non‑integrable covariance decaying as \( |x|^{-\kappa} \) for \( \kappa \in (2,d) \). They show that, unlike the case with compactly supported spatial correlations, the long-range correlations persist in the scaling limit.

    For the stochastic heat equation, presented in Fluctuations of stochastic PDEs with long-range correlations , Annals of Applied Probability (2025), they prove that fluctuations of the diffusively rescaled solution converge to those of an additive SHE driven by noise with Riesz-kernel covariance of degree \(-\kappa\), with convergence holding in optimal Hölder topologies as distribution-valued processes.

    For the KPZ equation (also in the same article), they demonstrate that its scaling limit is described by an additive SHE retaining the same spatial correlation. Remarkably, the limiting noise is the scaling limit of the original noise itself, leading to convergence in probability under suitable coupling. This contrasts with the integrable correlation case, where the limiting noise becomes spatially white and only weak convergence is obtained.

    These results highlight how non-integrable, long-range spatial correlations fundamentally alter the large-scale limits of stochastic PDEs, preserving noise correlations and yielding new types of universality classes.
  • Multi-Scale Analysis

  • Xue-Mei Li has made significant and original contributions to the theory of multi-scale stochastic systems, particularly in the context of geometric and manifold-valued models where classical methods do not apply directly. Her work combines rigorous stochastic analysis, differential geometry, and dynamical systems to address singular perturbation problems arising from physics and geometry.

    A distinctive feature of her approach is viewing singular perturbations as perturbations to conservation laws, even in settings where conventional scalar conserved quantities do not exist. Instead, the conserved quantities in her models are often manifold-valued, representing orbits or geometric structures. This perspective extends classical averaging and homogenisation theory to systems with geometric constraints, integrable structures, and non-trivial topology.

    Key contributions include: Overall, Xue-Mei Li's work on the multi-scale analysis of classical equations is characterised by its breadth across stochastic analysis, geometric mechanics, and manifold-valued systems, as well as its depth in developing rigorous mathematical frameworks for problems where standard PDE-based or Euclidean averaging techniques fail. Her research has opened pathways for understanding stochastic dynamics in geometric settings with multiple scales, with potential applications in applied mathematics.
  • Fractional Averaging (non-Markovian multi-scale)

    We study systems of non-Markovian stochastic evolutions with multi-scales in time. Our aim is to introduce auto-correlated noise instead of white noise in multi-time scale models. A protype for non-Markovian process with polynomial decay of correlation, stationary increments, self-similarity, and Gaussianaity is fractional Brownian motion. Such systems are prevalent. The `derivative' of a fractional Brownian motion is the simplest noise with correlation: It is Gaussian with stationary increment and $\E(B_{t+s}-B_s)^2=t^{2H}$, $H\in (0,1)$ is Hurst parameter, BM corresponds to the H=1/2 case. The sample paths of fBM is Holder with exponent up to H but not including H. The correlation delay of the increments is t^{2H-2}. WE prove the fractional averaging theorem for an SDE driven by a fractional B Brownian motion with vector fields depending on a fast evolving stochastic process y_{t/epsilon} where epsilon is a small parameter. We showed that

    (1) For H>1/2, the slow variables converges in probability to that of the solution drive by a fBM with vector fields simply averaged. Not this differes from the Markovisn case in two ways: the convergence for the latter is in general in a weak sense and one averages the generator of the slow variable, not the diffusion vetor fields. (2) Such a convergence is false for H<1/2. We show that with a scaling of the order epsilon^(1/2-H), the convergence holds to that of a Kunita type SDE whose generator is given by the fractional operator. Such a statement holds also if H>1/2 and the vector fields are centered. As a consequence we show the convergence of the stochastic flows (the n-point motion) which appears to be new even in the classical setting H=1/2.

    Homogenisation with non-Markovian and Long Range Dependent Noise

    We begin with proving a Functional limit theorems in the Holder topology for dimension 1 and apply it for a homogenesation theorems with limit dynamics drivin by fBM' Rosenblatt, and Hermite processes. To our knowledgem our article is the frist one to give a FCL in the Holder topology. For higher dimensional cases we proved functional limit theorems in the rough topology allowing to obtain homogenisation theorems of random ODE's. This is published on the arxiv in the 70 page article ( arXiv:1911.12600). This article is superceed by late improvements, and therefore not in a journal. The published article is:
  • Functional limit theorems for the fractional Ornstein-Uhlenbeck process Journal of Theoretical Probability, Johann Gehringer and Xue-Mei Li link to article. Also, https://arxiv.org/pdf/2006.11540.pdf
  • Diffusive and rough homogenisation in fractional noise field, Johann Gehringer and Xue-Mei Li link to article. Also, https://arxiv.org/pdf/2006.11540.pdf
  • Homogenisation with weighted average noise

    We study a general random ODE with fast moving convolution noise. We treat this as an ODE in a function space driven by a rough path in the infinite dimensional spaces.
  • Functional Limit Theorems for Volterra Processes and Applications to Homogenization, Johann Gehringer, Xue-Mei Li, and Julian Sieber link to article. Also, https://arxiv.org/abs/2104.06364
  • Large time behaviour of fractional dynamics

    By the nature of non-Markovian dynamics, the ergodicity for a stochastic differential equation driven by a fractional Brownian motion is different. In the Markovian case the invariant measure solves an elliptic differential equations allowing to study the invariant measure, its tails, kernels, and smoothness. In our case this is no logner true. Furthermore one cannot prepare the initial condition to obtain convergence rate estimates. Even more puzzling is that as a fast dynamics in a two scale system, we appear need to know the large time behaviour of the dynamics conditioned on its history which could have have a slower rate of convergence. We prove nevertheless a fractional averaging theorem and obtain Gaussian bounds for the invariant measure.
  • On the (Non-)Stationary Density of Fractional-Driven Stochastic Differential Equations, Annals of Probability (To appear) 2023 Xue-Mei Li, Julian Sieber, and Fabien Panloup. link to article.
  • Slow-Fast Systems with Fractional Environment and Dynamics Xue-Mei Li and Julian Sieber. Annals of Applied Probability, 2022, Vol. 32, No. 5, 3964-4003. link to article on arxiv; on journal site ; and ICL open access site
  • Fractional averaging for stochastic partial differential equations

  • On the Mild Stochastic Sewing Lemma, SPDE in Random Environment, and Fractional Averaging by Xue-Mei Li and Julian Sieber is given the best paper award in Stochastics and Dynamics. link to article. They began with non-stationary fractional averaging for SPDEs establishing a quantitative stochastic sewing lemma.
  • Fredholm, Hypoelliptic Operators

    Hypoelliptic operators are differential operators that, although possibly degenerate (i.e. not elliptic), still regularise solutions: if the distributional solution is smooth where the data is smooth. In stochastic analysis, hypoelliptic operators often appear as generators of diffusion processes satisfying Hörmander’s bracket condition, leading to smooth densities and strong probabilistic properties. In the following article Li has involved Hypoelliptic operators : Homogenisation on Homogeneous Spaces (with an appendix by D. Rumynin), Limits of Random Differential Equations on Manifolds; and Perturbation of Conservation Laws and Averaging on Manifolds.

    Estimates for Fundamental Solutions of of Parabolic Equations

    Strictly Local Martingales, Martingales, Semi-martingales

    In the first two articles, the focus is on local martingales which are not true martingales (and name them as strictly local martingales). In the third, examples of strict local martingales are given. In the fourth manifold valued martingales are studied In the fifth, one studies hypoelliptic bridges, i.e. conditioned diffusions associated with hypoelliptic operators. Classical bridges are well-understood for elliptic diffusions (e.g. Brownian bridge), but conditioning hypoelliptic diffusions is subtle due to degeneracies in directions of noise. The main contribution is showing that a class of hypoelliptic bridge processes are semi-martingale and have the L1 integrability property.

    Brownian, Generalised and hypoelliptic Bridges

    Existence of Global Smooth Stochastic Flows (strong p-completeness)