Adaptive Optimization of Noisy Black-Box Functions Inherent In Microscopic Models

For stochastic systems where exact constitutive relations are unknown, a microscopic or molecular level description can be alternatively used. As microscopic simulations are computationally very expensive, there is a need for the development of robust algorithms capable of economically optimizing these noisy processes. In this paper, two approaches – an adaptive strategy using a gradient-based NLP method and a local response surface method are applied to a stochastic reaction system. The effectiveness of these methods is evaluated in terms of the number of microscale function calls and computational time.