Formulating Reduced Order Models for the Reaction Chemistry of High Explosives

The reaction chemistry of high explosives has been analyzed phenomenologically using models containing only a few reaction steps. This is the level of model desired for use in higher length scale simulations. For explicit models, one should nominally consider hundreds of reaction intermediates and thousands of possible reactions, but the reductions of these to several step models is difficult. An alternative approach is to perform and analyze condensed phase reactive molecular dynamics (MD) simulations on these systems. The typical method for quantifying these simulations relies on identifying explicit molecular species and tracking their concentrations. A different approach is demonstrated here which focuses on quantifying changes in the coordination chemistry / oxidation state of the atoms during the reaction processes. This classifies all atoms for every timestep of the simulation and produces a countable number of geometries that characterize the reaction progress. Initial reduced order models can then be formulated from these results using the non-negative matrix factorization method. Illustrations will be given for cook-off simulations using both the classical ReaxFF MD approach applied to HMX, and the DFT-TB LATTE code applied to nitromethane and PETN.