arXiv · 2611.21092
Stochastic control representations for extremal solutions of HJB Dirichlet problems with measurable coefficients
Abstract
We study stochastic control problems for uniformly elliptic controlled diffusions on bounded domains whose coefficients are merely measurable in the state variable and whose drift is dominated by an $L_d$-function. The associated Hamilton-Jacobi-Bellman (HJB) Dirichlet problems may have more than one $L_p$-viscosity solution, so that the equation alone does not determine the value function. We show that the control problem selects the minimal solution: for all $p$ in a range $(d_0,d)$ determined by the spatial dimension, the ellipticity constant and the drift bound, the value function is the minimal $L_p$-viscosity supersolution of the HJB Dirichlet problem. In particular, every continuous $L_p$-viscosity supersolution bounds the optimal cost from above, which yields a one-sided verification theorem without regularity assumptions. We further prove the existence of optimal relaxed control rules for running costs dominated by an $L_p$-function, and the lower semicontinuity of the value function under perturbations of the coefficients. An example shows that upper semicontinuity may fail even when the coefficients are Lipschitz continuous in the control variable and the approximating coefficients are smooth, so that regularization may overestimate the optimal cost. The proof approximates arbitrary supersolutions from below by Sobolev supersolutions of HJB equations whose coefficients are modified only on sets of vanishing measure, and passes to the limit by means of Krylov estimates and martingale problem techniques.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
David Criens, Hyungbin Park, Jongjin Park. 2026-11-16. Stochastic control representations for extremal solutions of HJB Dirichlet problems with measurable coefficients. https://arxiv.science/abs/2611.21092
Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.