Epi-Regularization of Risk Measures
Uncertainty pervades virtually every branch of science and engineering, and in many disciplines, the underlying phenomena can be modeled by partial differential equations (PDEs) with uncertain or random inputs. This work is motivated by risk-averse stochastic programming problems constrained by PDEs. These problems are posed in infinite dimensions, which leads to a significant increase in the scale of the (discretized) problem. In order to handle the inherent nonsmoothness of, for example, coherent risk measures and to exploit existing solution techniques for smooth, PDE-constrained optimization problems, we propose a variational smoothing technique called epigraphical (epi-)regularization. Here, we investigate the effects of epi-regularization on the axioms of coherency and prove differentiability of the smoothed risk measures. In addition, we demonstrate variational convergence of the epi-regularized risk measures and prove the consistency of minimizers and first-order stationary points for the approximate risk-averse optimization problem. We conclude with numerical experiments confirming our theoretical results.
- Research Organization:
- Sandia National Lab. (SNL-NM), Albuquerque, NM (United States)
- Sponsoring Organization:
- Defense Advanced Research Projects Agency (DARPA); USDOE
- Grant/Contract Number:
- AC04-94AL85000
- OSTI ID:
- 1559486
- Report Number(s):
- SAND-2019-6920J; 676576
- Journal Information:
- Mathematics of Operations Research, Vol. 45, Issue 2; ISSN 0364-765X
- Publisher:
- InformsCopyright Statement
- Country of Publication:
- United States
- Language:
- English
Web of Science
Similar Records
A proximal trust-region method for nonsmooth optimization with inexact function and gradient evaluations
Enhancing environmental restoration predictive modeling in undersampled environments