Global Hölder estimates for optimal transportation (Q650335)

Let \(\mu_1, \mu_2\) be two probability measures on \(\mathbb R^d\) of the form \(e^{-V(x)}\,dx\) and \(e^{-W(x)}\,dx\), respectively. Then there exists an (optimal transport) mapping \(T:\mathbb R^d \to \mathbb R^d\) such that \(\mu_2 = \mu_1\circ T^{-1}\) and it is known that \(T\) is of the form \(\nabla\phi\) for some convex function \(\phi\). If \(T\) is smooth, then \(e^{-W(\nabla\phi(x))}\text{det}D^2\phi(x) = e^{-V(x)}\). Now assume that \(V\) and \(W\) satisfy \[ V(x+y)+V(x-y)-2V(x)\leq A_p|y|^{p+1} \] and \[ W(x+y)+W(x-y)-2W(x)\geq A_q|y|^{q+1} \] for all \(x,y\in\mathbb R^d\), suitable constants \(A_p,A_q\) and some exponents \(0\leq p\leq 1\leq q\). The author shows that under these conditions the map \(T\) is globally Hölder continuous with exponent \((p+1)/(q+1)\). If \(\nabla\phi\) is the optimal transportation map of \(\mu\) to \(\nu=\lambda|_K/\lambda(K)\) (\(\lambda\) is Lebesgue measure and \(K\) a compact convex set in \(\mathbb R^d\)), it is shown that \(\|D^2\phi(x)\|\leq f(\nabla\phi(x))\) where \(f\) is some positive function.