Mathematics > Optimization and Control
[Submitted on 2 Nov 2022 (this version), latest version 4 Oct 2023 (v5)]
Title:Wasserstein Steepest Descent Flows of Discrepancies with Riesz Kernels
View PDFAbstract:The aim of this paper is twofold. Based on the geometric Wasserstein tangent space, we first introduce Wasserstein steepest descent flows. These are locally absolutely continuous curves in the Wasserstein space whose tangent vectors point into a steepest descent direction of a given functional. This allows the use of Euler forward schemes instead of the minimizing movement scheme (MMS) introduced by Jordan, Kinderlehrer, and Otto. The MMS finds Wasserstein gradient flows by successively computing Wasserstein proxies. For locally Lipschitz continuous functionals which are $\lambda$-convex along generalized geodesics, we show that there exists a unique Wasserstein steepest descent flow coinciding with the Wasserstein gradient flow.
The second aim is to study Wasserstein flows of the (maximum mean) discrepancy with respect to Riesz kernels. The crucial part is hereby the treatment of the interaction energy. Although it is not $\lambda$-convex along generalized geodesics, we give analytic expressions for Wasserstein steepest descent flows of the interaction energy starting at Dirac measures. In contrast to smooth kernels, the particle may explode, i.e., the particle goes over to non-Dirac measures. The computation of steepest descent flows amounts to finding equilibrium measures with external fields, which nicely links Wasserstein flows of interaction energies with potential theory. Furthermore, we prove convergence of MMS to our Wasserstein steepest descent flows. Finally, we provide analytic Wasserstein steepest descent flows of discrepancies in one dimension and numerical simulations in two and three dimensions showing relations to interaction energy flows.
Submission history
From: Johannes Hertrich [view email][v1] Wed, 2 Nov 2022 16:10:28 UTC (6,036 KB)
[v2] Fri, 11 Nov 2022 10:02:52 UTC (6,037 KB)
[v3] Mon, 16 Jan 2023 09:39:10 UTC (6,014 KB)
[v4] Wed, 22 Mar 2023 08:47:29 UTC (6,019 KB)
[v5] Wed, 4 Oct 2023 07:45:35 UTC (4,324 KB)
Current browse context:
math.OC
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.