Finite Dimensional Projections of HJB Equations in the Wasserstein Space

TL;DR

This paper extends the analysis of controlled particle systems with common noise to multiplicative noise, proving convergence of finite particle value functions to an infinite-dimensional limit and establishing a projection property linking finite and infinite-dimensional control problems.

Contribution

It generalizes previous results to multiplicative noise, proves regularity of the infinite-dimensional value function, and establishes a projection property connecting finite and infinite-dimensional control solutions.

Findings

Convergence of finite particle value functions to the infinite-dimensional value function.

Regularity of the value function in the Wasserstein space.

Projection property linking finite and infinite-dimensional control problems.

Abstract

This paper continues the study of controlled interacting particle systems with common noise started in [W. Gangbo, S. Mayorga and A. \'{S}wi\k{e}ch, SIAM J. Math. Anal. 53 (2021), no. 2, 1320--1356] and [S. Mayorga and A. \'{S}wi\k{e}ch, SIAM J. Control Optim. 61 (2023), no. 2, 820--851]. First, we extend the following results of the previously mentioned works to the case of multiplicative noise: (i) We generalize the convergence of the value functions $u_n$ corresponding to control problems of $n$ particles to the value function $V$ corresponding to an appropriately defined infinite dimensional control problem; (ii) we prove, under certain additional assumptions, $C^{1,1}$ regularity of $V$ in the spatial variable. The second main contribution of the present work is the proof that if $DV$ is continuous (which, in particular, includes the previously proven case of $C^{1,1}$ regularity in the spatial variable), the value function $V$ projects precisely onto the value functions $u_n$. Using this projection property, we show that optimal controls of the finite dimensional problem correspond to optimal controls of the infinite dimensional problem and vice versa. In the case of a linear state equation, we are able to prove that $V$ projects precisely onto the value functions $u_n$ under relaxed assumptions on the coefficients of the cost functional by using approximation techniques in the Wasserstein space, thus covering cases where $V$ may not be differentiable.