Convergence analysis for Lasserre's measure--based hierarchy of upper bounds for polynomial optimization

Abstract

We consider the problem of minimizing a continuous function f over a compact set K. We analyze a hierarchy of upper bounds proposed by Lasserre in [SIAM J. Optim. 21(3) (2011), pp. 864--885], obtained by searching for an optimal probability density function h on K which is a sum of squares of polynomials, so that the expectation $\int_{K}f(x)h(x)dx$ is minimized. We show that the rate of convergence is no worse than $O(1/\sqrt{r})$, where 2r is the degree bound on the density function. This analysis applies to the case when f is Lipschitz continuous and K is a full-dimensional compact set satisfying some boundary condition (which is satisfied, e.g., for convex bodies). The r-th upper bound in the hierarchy may be computed using semidefinite programming if f is a polynomial of degree d, and if all moments of order up to 2r+d of the Lebesgue measure on K are known, which holds for example if K is a simplex, hypercube, or a Euclidean ball.