Smooth sample average approximation of stationary points in nonsmooth stochastic optimization and applications (Q1013981): Difference between revisions

The authors propose a smoothing SAA (sample average approximation) method for solving the following nonsmooth stochastic minimization problem \[ \min \mathbb{E}\left[ f\left( x,\xi \left( \omega \right) \right) \right] \text{ s.t. } x\in \mathcal{X}, \tag{1} \] where \(f:\mathbb{R}^{m}\times \mathbb{R}^{k}\rightarrow \mathbb{R}\) is locally Lipschitz continuous but not necessarily continuously differentiable, \(\xi :\Omega \rightarrow \Xi \subset \mathbb{R}^{k}\) is a random vector defined on the probability space \(\left( \Omega ,\mathcal{F} ,P\right)\), \(\mathbb{E}\) denotes the mathematical expectation, \(x\in \mathcal{X}\) is a decision vector with \(\mathcal{X}\) being a nonempty subset of \(\mathbb{R}^{m}\). It is supposed that \(\mathbb{E}\left[ f\left( x,\xi \left( \omega \right) \right) \right] \) is well defined for every \(x\in \mathcal{X}\). The authors generalize a convergence theorem established by \textit{A. Shapiro} [in: Rusczyński, A., Shapiro A. (eds.), Stochastic Programming, Handbooks in OR \& MS, vol. 10. North-Holland, Amsterdam (2003)] on a SSA method for a generalized stochastic equation and use it to show that under moderate conditions w.p.1 the stationary points of the smoothed sample average approximation problem converge to the weak stationary points of problem (1) and, when the underlying functions are convex, to optimal solutions. When the smoothing parameter is fixed, the authors obtain an error bound for the SAA stationary points under some metric regularity condition. Finally, they apply the convergence results to a CVaR problem and an inventory control problem in a supply chain.