Differentiation and regularity of semi-discrete optimal transport with respect to the parameters of the discrete measure (Q1719564)

The paper focuses on the second-order differentiability of the optimal transport problem of the cost function \(c(y,x)\) between the measures \(\nu\) and \(\mu=\sum_{i=1}^n m^{i}\delta_{z^i}\), with respect to the parameters of the latter discrete measure. Taking advantage of the dual formulation of this problem, the main goal of the article is to show that it is possible to differentiate with respect to \(z=(z_i)_{i=1}^n\), \(z_i\in \mathbb{R}^d\), under rather general assumptions on the cost function \(c\) and the measure \(\nu\). In particular, the result applies when \(c(y,x)=\frac{1}{2}\| y-x\|_2^2\) and \(\nu\) admits a \(C^0(\Omega)\cap W^{1,1}(\Omega)\) density with respect to the Lebesgue measure on a bounded, convex, Lipschitz set \(\Omega\subset \mathbb{R}^d\). In the final part of the paper, the authors show an application of their theoretical results providing numerical illustrations in the blue noise and the stippling problems. A similar study, under slightly different assumptions, is performed in [\textit{J. Kitagawa} et al., J. Eur. Math. Soc. (JEMS) 21, No. 9, 2603--2651 (2019; Zbl 1439.49053)], which kind of complement the present paper.