Strong displacement convexity on Riemannian manifolds (Q2642269)

Ricci curvature bounds in Riemannian geometry are known to be equivalent to the weak convexity of certain functionals in the space of probability measures. The authors show that the weak convexity can be reinforced into strong convexity, thus solving a question left open in [Lott and Villani, Ann. Math., to appear]. In this paper, the authors first give a definition: Let \(\nu\) be a reference measure on an \(n\)-dimensional Riemannian manifold \((M,g)\), absolutely continuous with respect to the volume measure. Let \(U:\mathbb R_{+}\rightarrow\mathbb R\) be a continuous convex function with \(U(0)=0\); let \(U^{'}(\infty)\) be the limit of \(U(r)/r\) as \(r\rightarrow \infty\). Let \(\mu\) be a probability measure on \(M\) and let \(\mu =\rho\nu+\mu_s\) be its Lebesgue decomposition with respect to \(\nu\). (i) If \(U(\rho)\) is bounded below by a \(\nu\)-integrable function, then the quantity \(U_{\nu}(\mu)\) is defined by the formula \[ U_{\nu}(\mu)=\int_MU(\rho(x))\nu(dx)+U^{'}(\infty)\mu_s[M]. \] (ii) If \(\pi\) is a probability measure on \(M\times M\), admitting \(\mu\) as first martingale, \(\beta\) is a positive function on \(M\times M\), and \(\beta U(\rho/\beta)\) (as a function of \(x,y\)) is bounded below by a \(\nu\)-integrable function of \(x\), then the quantity \(U^{\beta}_{\pi,\nu}(\mu)\) is defined by the formula \[ U^{\beta}_{\pi,\nu}(\mu)= \int_{M\times M} U\biggl(\frac{\rho(x)}{\beta(x,y)}\biggr) \beta(x,y)\pi(dy| x)\nu(dx)+ U'(\infty)\mu_s[M], \] where \(\pi(dy| x)\) is the disintegration of \(\pi(dxdy)\) with respect to the \(x\) variable. Then they obtain the following main result: Theorem. Let \(U, \nu\) and \(\beta\) be as in the above definition. Assume that \(U\) is Lipschitz. For each \(a>0\), define \(U_a(r)=U(ar)/a\). Then (i) If \((U_a)_{\nu}\) is weakly \(\lambda\)-a.c.c.s. displacement convex for any \(a\in (0,1]\), \(U_{\nu}\) is \(\lambda\)-displacement convex; (ii) If \((U_a)_{\nu}\) is weakly \(\lambda\)-a.c.c.s. displacement convex with distortion \(\beta\) for any \(a\in (0,1]\), \(U_{\nu}\) is displacement convex with distortion \(\beta\). From this theorem, they obtain the following Corollary. Let \(M\) be a smooth complete Riemannian manifold with nonnegative Ricci curvature and dimension \(n\). Let \(U(r)=-r^{1-\frac{1}{n}}\), and let \(\nu\) be the volume measure on \(M\). Then \(U_{\nu}\) is displacement convex on the set of probability measures \(P(M)\), where \(p=2\) if \(n\geq 3\), and \(p\) is any real number greater than \(2\) if \(n=2\).