A stochastic algorithm for deterministic multistage optimization problems

Abstract

Several attempts to dampen the curse of dimensionality problem of the Dynamic Programming approach for solving multistage optimization problems have been investigated. One popular way to address this issue is the Stochastic Dual Dynamic Programming method (Sddp) introduced by Pereira and Pinto (Math Program 52(1–3):359–375. https://doi.org/10.1007/BF01582895). Assuming that the value function is convex (for a minimization problem), one builds a non-decreasing sequence of lower (or outer) convex approximations of the value function. Those convex approximations are constructed as a supremum of affine cuts. On continuous time deterministic optimal control problems, assuming that the value function is semiconvex, Zheng Qu, inspired by the work of McEneaney, introduced in 2013 a stochastic max-plus scheme that builds a non-increasing sequence of upper (or inner) approximations of the value function. In this note, we build a common framework for both the Sddp and a discrete time version of Zheng Qu’s algorithm to solve deterministic multistage optimization problems. Our algorithm generates a monotone sequence of approximations of the value function as a pointwise supremum, or infimum, of basic (affine or quadratic for example) functions which are randomly selected. We give sufficient conditions on the way basic functions are selected in order to ensure almost sure convergence of the approximations to the value function on a set of interest.