210112 Rik Versendaal

January 12, 2021

Title: Brownian motion in Riemannian manifolds

Speaker: Rik Versendaal

When studying stochastic processes, and diffusions in particular, one of the most important processes is Brownian motion. Brownian motion is, in a way, the natural analogue of the normal distribution for processes. Furthermore, it is intimately related to the heat equation, since the Laplacian describes the infinitesimal evolution of Brownian motion. We will look at how to define and construct Brownian motion in a Riemannian manifold, the so-called Riemannian Brownian motion. There are various ways to do this, both geometric and probabilistic in nature.

First of all, as mentioned above, we can consider the process generated by the Laplacian of the Riemannian manifold, i.e., the Laplace-Beltrami operator. Second, we can use an invariance principle. In Euclidean space, this states that the paths of suitably scaled random walks converge to Brownian motion. One can define an analogue of random walks in manifolds, so-called geodesic random walks, and use these to obtain Riemannian Brownian motion in the limit.

Finally, Riemannian Brownian motion can also be obtained in a geometric way from a Euclidean Brownian motion. The idea is that we can transfer curves in Euclidean space to a manifold by suitably rolling the manifold along the curve. By Malliavin's transfer principle, it turns out that this also makes sense for stochastic processes. In particular, if we roll the manifold along a Euclidean Brownian motion, we will obtain a Brownian motion in the manifold.

If time permits, we will look into some results regarding large deviations for Riemannian Brownian motion. These large deviations are concerned with quantifying exponentially small probabilities of atypical trajectories of Brownian motion with vanishing variance. In particular, the action of a trajectory determines the exponential rate of decay of the probability. This can be shown to hold even in time-evolving Riemannian manifolds, i.e., manifolds where the metric depends on time.

Bookmark the permalink.

Comments are closed