The Linear Programming Approach to Approximate Dynamic Programming

Abstract

The curse of dimensionality gives rise to prohibitive computational requirements that render infeasible the exact solution of large-scale stochastic control problems. We study an efficient method based on linear programming for approximating solutions to such problems. The approach “fits” a linear combination of pre-selected basis functions to the dynamic programming cost-to-go function. We develop error bounds that offer performance guarantees and also guide the selection of both basis functions and “state-relevance weights” that influence quality of the approximation. Experimental results in the domain of queueing network control provide empirical support for the methodology.