Conditioning of linear-quadratic two-stage stochastic optimization problems (Q484135)

Consider the following convex stochastic optimization model of the form \[ \min \left\{ \int_{\mathbb R^s} g(x,\xi) P(d\xi) :x\in X\right\}, \tag{1} \] where \(X\) is a nonempty closed convex subset of \(\mathbb{R}^{m}\), \(P\) a probability distribution on \(\mathbb{R}^{s}\) and \(g\) is an extended real-valued measurable function on \(\mathbb{R}^{m}\times \mathbb{R} ^{s}\) such that \(g(\cdot ,\xi)\) is convex for all \(\xi\) in the support of \(P\). Particular cases of (1) are two-stage linear or linear-quadratic stochastic programs. In this paper, the authors study linear-quadratic two-stage stochastic optimization problems for which a condition number is introduced as a Lipschitz modulus of the multifunction assigning to a (discrete) probability distribution the solution set of the problem. In particular, the authors apply the results to the special setting of so-called simple recourse.