We consider the problem of approximating a probability measure defined on a metric space by a measure supported on a finite number of points. More specifically we seek the asymptotic behavior of the minimal Wasserstein distance to an approximation when the number of points goes to infinity. The main result gives an equivalent when the space is a Riemannian manifold and the approximated measure is absolutely continuous and compactly supported.

Given the probability measure ν over the given region $\Omega\subset \mathbb{R}^n$ , we consider the optimal location of a setΣ composed by n points in Ω in order to minimize theaverage distance $\Sigma\mapsto \int_\Omega \mathrm{dist}\,(x,\Sigma)\,{\rm d}\nu$ (theclassical optimal facility location problem). The paper compares twostrategies to find optimal configurations: the long-term one whichconsists in placing all n points at once in an optimal position, and the short-term one which consists in placing the points one by one addingat each step at most one point and preserving the configurationbuilt at previous steps. We show that the respective optimizationproblems exhibit qualitatively different asymptotic behavior as $n\to\infty$ , although the optimization costs in both cases have the same asymptotic orders of vanishing.

Let

where Xi, 1 ≦ i ≦ n, are i.i.d. and uniformly distributed in [0, 1]2. It is proved that Mn ∽ cn1–p/2 a.s. for 1 ≦ p <2. This result is motivated by recent developments in the theory of algorithms and the theory of subadditive processes as well as by a well-known problem of H. Steinhaus.