We introduce the Geodesic Walk for sampling Riemannian manifolds and apply it to the problem of generating uniform random points from the interior of a convex polytope in n-dimensional Euclidean space. The walk is a simulation of a stochastic differential equation defined using a convex barrier function; each step is the solution of an ordinary differential equation. It improves the time complexity of sampling polytopes and is the first walk to achieve sub-quadratic complexity. Part of the talk will be spent introducing relevant aspects of manifolds, stochastic processes, efficient solution of differential equations, and how this walk overcomes the bottlenecks of random walks in Euclidean space. No background in string theory (or Riemannian geometry) will be assumed.
Joint work with Santosh Vempala.