Stochastic optimal transport with free end time

Abstract

We consider a stochastic transportation problem between two prescribed probability distributions (a source and a target) over processes with general drift dependence and with free end times. First, and in order to establish a dual principle, we associate two equivalent formulations of the primal problem in order to guarantee its convexity and lower semi-continuity with respect to the source and target distributions. We exhibit an equivalent Eulerian formulation, whose dual variational principle is given by Hamilton-Jacobi-Bellman type variational inequalities. In the case where the drift is bounded, regularity results on the minimizers of the Eulerian problem then enable us to prove attainment in the corresponding dual problem. We also address attainment when the drift component of the cost defining Lagrangian L is superlinear L ≈ |u| p with 1 < p < 2, in which case the setting is reminiscent of our approach -in a previous work-on deterministic controlled transport problems with free end time. We finally address criteria under which the optimal drift and stopping time are unique, namely strict convexity in the drift component and monotonicity in time of the Lagrangian.

The large time profile for Hamilton--Jacobi--Bellman equations

“…The connection between Aubry-Mather theory and optimal transport was explored in [5]. We also refer to [13] for recent work concerning stochastic optimal transport.…”