Hit-and-run mixes fast

Abstract.

It is shown that the “hit-and-run” algorithm for sampling from a convex body K (introduced by R.L. Smith) mixes in time O ^*(n ² R ²/r ²), where R and r are the radii of the inscribed and circumscribed balls of K. Thus after appropriate preprocessing, hit-and-run produces an approximately uniformly distributed sample point in time O ^*(n ³), which matches the best known bound for other sampling algorithms. We show that the bound is best possible in terms of R,r and n.