A notion of BSDE on the Wasserstein space and its applications to control problems and PDEs

TL;DR

This paper develops a new class of backward stochastic differential equations on the Wasserstein space, linking mean-field control problems and PDEs in a probabilistic framework, extending classical BSDE theory to measure-dependent settings.

Contribution

It introduces a novel BSDE formulation on the Wasserstein space that connects mean-field control problems with PDEs, including existence, uniqueness, and comparison principles.

Findings

Established a correspondence between measure-dependent BSDEs and mean-field control problems.

Proved a comparison principle ensuring uniqueness of solutions.

Obtained existence results for linear and quadratic generators.

Abstract

We introduce a class of backward stochastic differential equations (BSDEs) on the Wasserstein space of probability measures. This formulation extends the classical correspondence between BSDEs, stochastic control, and partial differential equations (PDEs) to the mean--field (McKean--Vlasov) setting, where the dynamics depend on the law of the state process. The standard BSDE framework becomes inadequate in this context, motivating a new definition in terms of measure--dependent solutions. Under suitable assumptions, we demonstrate that this formulation is in correspondence with both mean--field control problems and partial differential equations defined on the Wasserstein space. A comparison principle is established to ensure uniqueness, and existence results are obtained for generators that are linear or quadratic in the $z$--variable. This framework provides a probabilistic approach to control and analysis on the space of probability measures.