Prescription of Gauss curvature using optimal mass transport (Q303990)

If \(\Omega\subset\mathbb R^{m+1}\) is a convex body, i.e., a closed bounded convex set whose interior is nonempty, with the origin within \(\Omega\), then the map \(\vec\rho:\mathbb S^m\to\partial\Omega\), \(x\mapsto\rho(x)x\) is a homeomorphism, where the radial function \(\rho\) is defined by \(\rho(x)=\sup\{s;\;sx\in \Omega\}\). \(\rho\) is a Lipschitz function bounded away from \(0\) and \(\infty\). The Gauss curvature measure is the Borel probability measure \(\mu(x)=\sigma({\mathcal G}(\vec\rho(x)))\), where \(\sigma\) is the uniform Borel probability measure on \(\mathbb S^m\) and \({\mathcal G}:\partial\Omega\rightrightarrows\mathbb S^m\) is the Gauss multivalued map. In [C. R. (Dokl.) Acad. Sci. URSS, n. Ser. 35, 131--134 (1942; Zbl 0061.37604)], \textit{A. D Aleksandrov} proved that if \(\sigma\) is the uniform probability measure on \(\mathbb S^m\) and \(\mu\) is a Borel probability measure on \(\mathbb S^m\) satisfying \(\mu(\omega)<\sigma(\omega_{\pi/2})\) for any non-empty convex subset \(\omega\subset\mathbb S^m\), where \(\omega_{\pi/2}=\{x;\;d(x,\omega)<\pi/2\}\), then there exists a unique convex body in \(\mathbb R^{m+1}\) containing \(0\) in its interior and whose \(\mu\) is the Gauss curvature measure. If \(\mu\) and \(\sigma\) are two probability measures on \(\mathbb S^m\), \(c:\mathbb S^m\times\mathbb S^m\to\mathbb R^+\) is a cost function, then the set \(\Gamma(\mu,\sigma)\) of transport plans is the set of probability measures \(\Pi\) such that for any Borel set \(A\subset\mathbb S^m\) \(\sigma(A)=\Pi(A\times\mathbb S^m)\) and \(\mu(A)=\Pi(\mathbb S^m\times A)\). The mass transport problem \((*)\) studies \[ \inf\limits_{\Pi\in\Gamma(\sigma,\mu)}\int\limits_{\mathbb S^m\times\mathbb S^m}c(n,x\,d\,\Pi(n,x). \] The compactness of \(\Gamma(\sigma,\mu)\) combined with the continuity of \(c\) yields the existence of minimizers for this problem. These minimizers are called optimal transport plans. To study properties of optimal plans, \textit{L. Kantorovitch} [C. R. (Dokl.) Acad. Sci. URSS, n. Ser. 37, 199--201 (1942; Zbl 0061.09705)] introduced a dual problem. If \({\mathcal A}=\{(\varphi,\psi)\}\) is the set of pairs of Lipschitz functions defined on \(\mathbb S^m\) satisfying \(\varphi(n)+\psi(x)\leq c(n,x)\) for all \(x, n\in\mathbb S^m\), then the Kantorovitch variational problem \((**)\) is \[ \sup\limits_{(\varphi,\psi)\in{\mathcal A}}\left\{\int\limits_{\mathbb S^m}\varphi(n)\,d\,\sigma(n)+\int\limits_{\mathbb S^m}\psi(x)\,d\,\mu(x)\right\}. \] This problem is called Kantorovitch's dual problem if the cost function is continuous and nonnegative. Then \((*)=(**)\). In [Adv. Math. 213, No. 2, 600--620 (2007; Zbl 1233.49024], \textit{V. Oliker} proved the strong Kantorovich duality theorem. It was shown that \[ \max\limits_{(\varphi,\psi)\in{\mathcal A}}\left\{\int\limits_{\mathbb S^m}\varphi(n)\,d\,\sigma(n)+\int\limits_{\mathbb S^m}\psi(x)\,d\,\mu(x)\right\}=\min\limits_{\Pi\in\Gamma(\sigma,\mu)}\int\limits_{\mathbb S^m\times\mathbb S^m}c(n,x\,d\,\Pi(n,x)<\infty. \] The goal of this paper is to present a new proof of a theorem by Alexandrov on the Gauss curvature prescription of Euclidean convex sets. This proof is based on the strong Kantorovich duality theorem. A novelty of this proof is that it does not rely on the theory of convex polyhedra nor on partial differential equation methods, which appeared in all the previous proofs of this result.