Solving stochastic mathematical programs with equilibrium constraints via approximation and smoothing implicit programming with penalization (Q959946): Difference between revisions

The authors consider the stochastic mathematical programs with linear complementarity constraints, which include two kinds of models: the old one, so-called lower-level wait-and-see model, and the new model called here-and-now. They study mainly the following here-and-now model: \[ \begin{aligned} &\underset{x,y,z}{\text{minimize}}\;E_\omega[f(x,y,\omega)+d^Tz(\omega)]\\ &\text{subject to}\quad x\in X,\quad y\geq 0,\quad F(x,y,\omega)+z(\omega)\geq 0,\\ &y^T(F(x,y,\omega)+z(\omega))=0,\quad z(\omega)\geq 0,\quad \omega\in\Omega\;\text{a.e.} \end{aligned} \] Here the mapping \(F\) is affine, \(d\) is a vector with positive elements, \(\omega\) is a discrete or continuous random variable; and the rest of the notations are conventional ones. The continuous problem is discretized by a quasi-Monte Carlo method. The authors present a combined smoothing implicit programming and penalty method with appropriate convergence results. The numerical results (for a picnic vender decision problem) are also present.