\{Euclidean, metric, and Wasserstein\} gradient flows: an overview

DOI10.1007/S13373-017-0101-1MaRDI QIDQ523629zbMATH OpenOpenAlexWikidataFDO

Publication date 21 April 2017

Published in Bulletin of Mathematical Sciences (Search for Journal in Brave)

Copyright license Creative Commons Attribution 4.0 International

Full work available at URL https://arxiv.org/abs/1609.03890

analysis in metric spaces Cauchy problem contractivity Fokker-Planck equation heat flow metric measure spaces numerical methods optimal transport subdifferential Wasserstein distances

Mathematics Subject Classification ID

Evolution inclusions (34G25) Heat equation (35K05) Methods involving semicontinuity and convergence; relaxation (49J45) Numerical methods involving duality (49M29) Variational problems in a geometric measure-theoretic setting (49Q20) Metric spaces, metrizability (54E35)

Abstract: This is an expository paper on the theory of gradient flows, and in particular of those PDEs which can be interpreted as gradient flows for the Wasserstein metric on the space of probability measures (a distance induced by optimal transport). The starting point is the Euclidean theory, and then its generalization to metric spaces, according to the work of Ambrosio, Gigli and Savar{'e}. Then comes an independent exposition of the Wasserstein theory, with a short introduction to the optimal transport tools that are needed and to the notion of geodesic convexity, followed by a precise desciption of the Jordan-Kinderleher-Otto scheme, with proof of convergence in the easiest case: the linear Fokker-Planck equation. A discussion of other gradient flows PDEs and of numerical methods based on these ideas is also provided. The paper ends with a new, theoretical, development, due to Ambrosio, Gigli, Savar{'e}, Kuwada and Ohta: the study of the heat flow in metric measure spaces.

Review text

This is a survey article on the theory of gradient flows in metric spaces with a big accent on the study of those flows that lead to evolution PDEs. For a given space \(X\), by the gradient flow of a given functional \(F:X\to\mathbb{R}\) we mean the solution of the ODE \[ x'(t)=-\nabla F(x(t)). \] It is pretty easy to study this equation in a finite dimensional setting, or even in the case of general infinite dimensional Hilbert spaces, where the gradient vector has to be understood as an element of \(X\). A very first interesting (infinite dimensional) example could be the case of the heat equation \[ \partial_t \rho -\Delta \rho =0\tag{1} \] in \(\mathbb{R}^d\), which -- by setting \(X=H^1(\mathbb{R}^d)\) -- can be seen formally as the \(L^2-\)gradient flow of the Dirichlet energy \(F(\rho)=\int_{\mathbb{R}^d}|\nabla \rho|^2 dx\). Gradient flows received a great attention in the late 90's and early 2000's, thanks to the works of \textit{R. Jordan} et al. [SIAM J. Math. Anal. 29, No. 1, 1--17 (1998; Zbl 0915.35120)] and \textit{F. Otto} [Commun. Partial Differ. Equations 26, No. 1--2, 101--174 (2001; Zbl 0984.35089)] on the one hand and \textit{L. Ambrosio} et al. [Gradient flows in metric spaces and in the space of probability measures. Basel: Birkhäuser (2005; Zbl 1090.35002)] on the other hand. These authors realized that many evolution equations of the form \[ \partial_t\rho-\nabla\cdot(v_t\rho)=0,\tag{2} \] where \[ v_t:=\nabla\frac{\delta F}{\delta\rho}(\rho_t) \] can be seen as gradient flow of functionals \(F\) defined on the space of probability measures over an ambient (typically finite dimensional) space equipped with the Wasserstein metric arising in the theory of optimal mass transportation. Here \(\frac{\delta F}{\delta\rho}\) stands for the first variation of the functional \(F\). In this setting, the heat equation (1) can be seen as the Wasserstein gradient flow of the entropy functional \(\rho\mapsto \int_{\mathbb{R}^d}\rho\log(\rho)dx\). The above mentioned monograph of Ambrosio-Gigli-Savaré provides certainly the first rigorous development of gradient flows in metric spaces and in particular in the space of probability measures endowed with the Wasserstein metric. The so-called displacement convexity of functional defined on probability measures introduced by \textit{R. J. McCann} [Adv. Math. 128, No. 1, 153--179 (1997; Zbl 0901.49012)] plays a key role in this rigorous analysis. It is also interesting to notice that the heat equation (1) turns out to be the gradient flow of some functionals in two completely different settings. Both frameworks can be extended to more general settings, to metric measure spaces (instead of the Euclidean ambient space), thus the question whether the two flows coincide, or under which assumptions they do, is natural. This question has been recently studied by Ambrosio, Gigli, Savaré and new collaborators (Kuwada and Ohta) in a series of papers and has been the starting point of recent deep researches on the differential structure of metric measure spaces. This paper provides an excellent overview of the current developments on gradient flows in the metric and more particularly in the Wassertein setting, as described briefly above. It is clearly written with a particular emphasis on applications of this theory. A large number of up to date contributions are listed in the reference list, thus the paper surely presents the whole picture on this theory to the reader.

Recommendations

Cites work

Cited in

(only showing first 100 items - show all)

This page was built for publication: \{Euclidean, metric, and Wasserstein\} gradient flows: an overview

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q523629)