Scenario decomposition of risk‐averse multistage stochastic programming problems

For a risk‐averse multistage stochastic optimization problem with a finite scenario tree, we introduce a new scenario decomposition method and we prove its convergence. The main idea of the method is to construct a family of risk‐neutral approximations of the problem. The method is applied to a risk‐averse inventory and assembly problem. In addition, we develop a partially regularized bundle method for nonsmooth optimization.