A note on Ricci flow and optimal transportation (Q610663): Difference between revisions

This paper deals with an extension of the following classical interpolation inequality due to \textit{A.~Prekopa} [Acta Sci. Math. 32, 301--316 (1971; Zbl 0235.90044)] and \textit{L.~Leindler} [Proc. Conf. Oberwolfach 1971, ISNM 20, 182--184 (1972; Zbl 0256.26015)]. Theorem 1. Fix a real number \(0<\lambda<1\). Let \(u_1\), \(u_2\), \(v:\mathbb R^n\to \mathbb R\) be nonnegative measurable functions satisfying \[ v(\, (1-\lambda)x+\lambda y\,) \geq u_1(x)^{1-\lambda} u_2(y)^\lambda \] for all points \(x, y\in\mathbb R^n\). Then \[ \int_{{\mathbb{R}}^n}v \geq \left(\int_{{\mathbb{R}}^n}u_1\right)^{1-\lambda} \left(\int_{{\mathbb{R}}^n}u_2\right)^\lambda. \] Some generalizations in the context of Riemannian geometry were obtained in [\textit{D.~Cordero-Erausquin, R.~McCann} and \textit{M.~Schmuckenschläger}, Invent. Math. 146, No.~2, 219--257 (2001; Zbl 1026.58018); Ann. Fac. Sci. Toulouse, Math. (6) 15, No.~4, 613--635 (2006; Zbl 1125.58007)]. The aim of the paper is to obtain other generalizations of the interpolation inequality replacing the Riemannian distance by the \(\mathcal L\)-distance of Perelman. Namely, let \(M\) be a compact \(n\)-dimensional manifold, and let \(g(t)\), \(t\in [0,T]\) be a one-parameter family of metrics on \(M\), which evolve by the backward Ricci flow, i.e. \(\frac{\partial g(t)}{\partial t} = 2 \text{Ric}_{g(t)}\). The \(\mathcal{L}\)-length of a path \(\gamma: [\tau_1,\tau_2]\to M\) is defined by \[ {\mathcal L}(\gamma) = \int_{\tau_1}{\tau_2} \sqrt{t}\,\left( R_{g(t)}(\gamma(t))+|\gamma^\prime(t)|^2_{g(t)}\right) dt,\tag{*} \] where \(R_{g(t)}\) stands for the scalar curvature of \(g(t)\). The corresponding \(\mathcal{L}\)-distance is defined by \[ Q(x,\tau_1; y,\tau_2) = \inf\left\{{\mathcal L}(\gamma)| \gamma: [\tau_1,\tau_2]\to M, \gamma(\tau_1)=x, \gamma(\tau_2)=y \right\}.\tag{*} \] The main results of the paper is the following extension of the interpolation inequality in the setting of Ricci flow and \(\mathcal{L}\)-distance. Theorem 2. Fix real numbers \(\tau_1\), \(\tau_2\), \(\tau\) such that \(0<\tau_1<\tau<\tau_2<T\). Write \(\frac{1}{\sqrt{\tau}}=\frac{1-\lambda}{\sqrt{\tau_1}}+\frac{\lambda}{\sqrt{\tau_2}}\), where \(0<\lambda<1\). Let \(u_1\), \(u_2\), \(v: M\to\mathbb R\) be nonnegative measurable functions such that \[ \left(\frac{\tau}{\tau_1^{1-\lambda}\tau_2^\lambda}\right)^{n/2}v(\, \gamma(\tau)\,) \geq \exp\left(-\frac{1-\lambda}{2\sqrt{\tau_1}}Q(\gamma(\tau_1),\tau_1; \gamma(\tau),\tau)\right) u_1(\gamma(\tau_1))^{1-\lambda} \] \[ \cdot \exp\left(\frac{\lambda}{2\sqrt{\tau_2}}Q(\gamma(\tau),\tau; \gamma(\tau_2),\tau_2)\right) u_2(\gamma(\tau_2))^\lambda \] for every minimizing \(\mathcal{L}\)-geodesic \(\gamma: [\tau_1,\tau_2]\to M\). Then \[ \int_{M}v\, \text{dvol}_{g(\tau)} \geq \left(\int_{M}u_1\,\text{dvol}_{g(\tau_1)}\right)^{1-\lambda} \left(\int_{M}u_2\, \text{dvol}_{g(\tau_2)}\right)^\lambda. \] As consequence, by sending \(\tau_1\to 0\), one can recover the monotonicity of Perelman's reduced volume.