On the regularity of the free boundary in the optimal partial transport problem for general cost functions

From MaRDI portal
Publication:2256012



Abstract: This paper concerns the regularity and geometry of the free boundary in the optimal partial transport problem for general cost functions. More specifically, we prove that a C1 cost implies a locally Lipschitz free boundary. As an application, we address a problem discussed by Caffarelli and McCann cite{CM} regarding cost functions satisfying the Ma-Trudinger-Wang condition (A3): if the non-negative source density is in some Lp(mathbbRn) space for pin(fracn+12,infty] and the positive target density is bounded away from zero, then the free boundary is a semiconvex Cloc1,alpha hypersurface. Furthermore, we show that a locally Lipschitz cost implies a rectifiable free boundary and initiate a corresponding regularity theory in the Riemannian setting.


For two compactly supported functions \(0 \leq f,g \in L^{1}({\mathbb R}^n)\) and a number \(0<m \leq \min \{\|f\|_{L_1}, \|g\|_{L_1}\}\), the objective in the optimal partial transport problem is to find an optimal transference plan between \(f\) and \(g\) with mass \(m\). A transference plan refers to a non-negative, finite Borel measure \(\gamma\) on \(\mathbb R^n\times\mathbb R^n\), whose first and second marginals are \(\gamma(A\times\mathbb R^n)\leq \int_Af(x)\,dx\), and \(\gamma(\mathbb R^n\times A)\leq \int_Ag(x)\,dx\) for any Borel set \(A\subset\mathbb R^n\). An optimal transference plan is a minimizer of the functional \(\gamma\mapsto\int c(x,y)\,d\gamma(x,y)\), where \(c\) is a non-negative cost function. In general, minimizers fail to be unique, however \textit{L. A. Caffarelli} and \textit{R. J. McCann} in [Ann. Math. (2) 171, No. 2, 673--730 (2010; Zbl 1196.35231)] and \textit{X.-N. Ma} et al. in [Arch. Ration. Mech. Anal. 177, No. 2, 151--183 (2005; Zbl 1072.49035)] proved that, under suitable assumptions on the quadratic cost function, minimizers are unique. In this paper, the authors consider the regularity and geometry of the free boundary in the optimal partial transport problem for non-quadratic cost functions, and they show that a \(C^1\) cost implies a locally Lipschitz free boundary. The authors prove that if \(f\) and \(g\) are non-negative integrable functions on a domain \(\Omega\), \(m\in(0,\min \{\|f\|_{L_1}, \|g\|_{L_1}\}]\), \(\Lambda\) is bounded and \(c\)-convex with respect to \(\Omega\), and a non-quadratic cost function \(c\in C^1(\mathbb R^n\times\mathbb R^n)\) satisfies certain conditions, then the free boundaries arising in the optimal partial transport problem are locally Lipschitz graphs inside \(\Omega\). As an application, they address a problem regarding cost functions, and they show that if the non-negative source density \(f\) is in some \(L^p(\mathbb R^n)\) space for \(p\in(\frac{n+1}2,\infty]\) and the positive target density \(g\) is bounded away from zero, then the free boundary is a semiconvex \(C^{1,\alpha}_{\text{loc}}\) hypersurface. Also, the authors show that if the cost function is locally Lipschitz, then the free boundary arising in the optimal partial transport problem is \((n-1)\)-rectifiable. Finally, they initiate a corresponding regularity theory in the Riemannian setting.











This page was built for publication: On the regularity of the free boundary in the optimal partial transport problem for general cost functions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2256012)