Closed Loop Adaptive Refinement of Dynamical Models for Complex Chemical Reactions

Abstract

Abstract : This research is concerned with the development of a systematic method for efficiently performing molecular dynamics (MD) simulations of complex chemical reactions and optimizing the underlying potential energy surfaces (PESs), ultimately using suitable laboratory data in a closed loop fashion. Two main objectives of the research are to (a) identify key parameters of each PES based on the global non-linear input-output Random Sampling High Dimensional Model Representation (RS-HDMR) mapping technique [1-7] and (b) use the RS-HDMR maps to efficiently capture the PES observable relationships [8-10]. The RS-HDMR analysis in turn provides essential information for subsequent full implementation of PES optimization within the proposed adaptive closed-loop learning algorithm in conjunction with laboratory feedback. In this project we have (1) formulated a fully equivalent operational model (FEOM) based on RS-HDMR, in place of the time-consuming Newton equations of motion for performing multi-dimensional MD simulations, and (2)performed detailed studies on intermolecular energy transfer for the model systems.