Obstructions to regularity in the classical Monge problem (Q481133)

If \(c:\mathbb R^n\times\mathbb R^n\to[0,\infty)\) is a cost function, \(f\) and \(g\) are probability densities on domains \(\Omega\), \(\Omega^*\subset\mathbb R^n\), respectively, then the classical Monge problem is to minimize the functional \(T\mapsto\int_\Omega c(x,T(x))\,f(x)\,dx\) among maps \(T:\Omega\to\Omega^*\) satisfying \(\int_Ag(y)\,dy=\int_{T^{-1}(A)}f(x)\,dx\) for all Borel sets \(A\subset\Omega^*\). The minimizer is called the optimal transport map for \(c\). The existence of optimal maps was studied by many authors. In general, minimizers are not unique. However, in [Calc. Var. Partial Differ. Equ. 15, No. 1, 81--113 (2002; Zbl 1003.49031)], \textit{M.~Feldman} and \textit{R.~J.~McCann} showed that there is a unique optimal transport which is monotone on transfer rays. The regularity of optimal maps is still widely open. \textit{I. Fragalà} et al. [J. Math. Pures Appl. (9) 84, No. 9, 1261--1294 (2005; Zbl 1075.49018)] proved the existence of a continuous optimal transport, under the assumptions that the densities are continuous and strictly positive in the interior part of their supports, and that such supports are convex, compact, and disjoint. In [J. Math. Pures Appl. (9) 102, No. 6, 1015--1040 (2014; Zbl 1304.49094)], \textit{Q.-R. Li} et al. showed that the monotone optimal transport between positive \(C^\infty\) densities supported on the same bounded, convex domain fails to be Lipschitz continuous at interior points, and they constructed an example in which the monotone optimal map does not belong to \(C^{\frac23+\varepsilon}\) for any \(\varepsilon>0\). This paper focuses on examples relating the regularity of the initial data, which are assumed to be strictly positive and bounded on a convex set, with the regularity of the monotone optimal transport. The authors construct a family of examples where the modulus of continuity of the monotone optimal transport is worse than the modulus of continuity of the densities. For example, they show that if the densities \(f\) and \(g\) are in \(C^{2,1}(\Omega)\), then the monotone optimal transport is not \(C^{0,\frac12+\varepsilon}(\Omega)\).