{"id":"arxiv/2605.30055","url":"https://arxiv.org/abs/2605.30055","tags":["math.PR","math.FA","math.ST"],"score":0,"title":"The Wasserstein cost of Importance Sampling","domain":"math.PR","source":"arxiv","authors":["Simon Coste","Michael Goldman"],"reviews":0,"subject":"math.PR","summary":"Importance sampling (IS) consists in biasing samples from a distribution $f$ towards another distribution $g$. Concretely, given samples $X_i$ from $f$, the IS measure is $$\\hat{g}_n = \\frac{1}{Z_n}\\sum_{i=1}^n \\frac{g(X_i)}{f(X_i)} δ_{X_i},$$ with $Z_n = \\sum_{i=1}^n \\frac{g(X_i)}{f(X_i)}$. The random measure $\\hat{g}_n$ approximates $g$, and is used in many contexts ranging from Monte Carlo integration to Bayesian inference. We show that, in high dimension ($d \\geqslant 3$), the Wasserstein cost $W_p^p(\\hat{g}_n, g)$ has order $n^{-p/d}$ in expectation, i.e. $$β^{\\mathrm{low}}_{p,d}\\int…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-05-28","reactions":{}}