WorksMathematics
Certified Algorithms for Gaussian Location Mixture NPMLEs via Discretized Gradient Flows
Abstract
We study the computation of approximate nonparametric maximum likelihood estimators (NPMLEs) for Gaussian location mixtures in arbitrary dimension via the Fisher-Rao gradient flow. We provide a certified algorithm based on an entropic proximal discretization of the flow which, given a sufficiently small , returns an absolutely continuous mixing measure admitting an exact finite-dimensional representation and satisfying . For almost every dataset satisfying , where is the number of support points of the NPMLE, the algorithm requires time. En route, we also establish results on the well-posedness and convergence properties of the continuous-time Fisher–Rao flow. In particular, initializing the flow at the uniform probability measure on a ball containing the convex hull of the observations, we show that . We also establish that, for almost every dataset satisfying the same condition on as above, the set of global maxima of the optimal score is the same as the support of the NPMLE, and the Hessian of the optimal score is nondegenerate at these points, extending a result of (Polyanskiy-Sellke, 2025) to higher dimensions.