Assessing Solution Quality in Stochastic Programs

Assessing whether a solution is of high quality (optimal or near optimal) is a fundamental question in optimization. We develop Monte Carlo sampling-based procedures for assessing solution quality in stochastic programs. Quality is defined via the optimality gap and our procedures' output is a confidence interval on this gap. We review a multiple-replications procedure and then present a result that justifies a computationally simplified single-replication procedure. Even though the single replication procedure is computationally significantly less demanding, the resulting confidence interval may have low coverage for small sample sizes on some problems. We provide variants of this procedure and provide preliminary guidelines for selecting a candidate solution. Both are designed to improve the basic procedure's performance.