Paperlane

no. 17 · Wednesday 18 November 2026

Asymptotic analysis of sample average approximations for optimal control problems with uncertain dynamics

Maria Soledad Aronna, João Miguel Machado and Oscar Sierra Fonseca

math.OC · 35 pp

Abstract

We study sample average approximations (SAA) of optimal control problems with control-affine dynamics depending on an uncertain parameter. The objective is an expected terminal cost plus a Tikhonov regularization term, and the controls are square-integrable functions subject to pointwise convex constraints. We show that the reduced objective is twice continuously differentiable in the square-integrable topology, so that no two-norm discrepancy arises, and that its Hessian is a Hilbert–Schmidt perturbation of a multiple of the identity. This yields Lipschitz regularity of stationary controls and second-order conditions without gap for polyhedral constraints. For the SAA scheme with sample size $M$, we prove almost sure uniform consistency of the empirical minimizers, convergence rates of order $M^{-1/2}$ for the optimal values, independent of the dimension of the parameter space. We then derive the asymptotic distributions. The optimal values satisfy a central limit theorem, obtained from a functional central limit theorem for the SAA objective. The empirical minimizers are asymptotically Gaussian when the constraints are inactive; under polyhedral constraints, their limit solves a quadratic problem on the critical cone with a Gaussian linear term. We also derive confidence intervals for linear functionals of the optimal control and illustrate the results on two numerical examples.

comments
35 pages, 2 figures
fields
Optimization and Control
pdf ↗ page arXiv:2611.21831 ↗