Infinite-dimensional spherical-radial decomposition for probabilistic functions, with application to constrained optimal control and Gaussian process regression
Published in arXiv, 2026
Recommended citation: Kewei Wang and Georg Stadler. "Infinite-dimensional spherical-radial decomposition for probabilistic functions, with application to constrained optimal control and Gaussian process regression." arXiv:2603.19907 (2026). https://arxiv.org/pdf/2603.19907
This preprint extends spherical-radial decomposition to infinite-dimensional stochastic settings by combining a subspace method with Monte Carlo sampling. The resulting approach provides unbiased, low-variance estimates and derivatives of probabilistic functions, with applications to chance-constrained stochastic PDE optimization and constrained Gaussian process hyperparameter learning.