Grid-flux method for learning the solvent contribution to the mechanisms of reactions

We present a trajectory-based method with which to learn information about the qualitative role that solvent molecules play in dynamic simulations of reactions that occur in solution. A spatial grid is superimposed on the system being simulated, and average fluxes of solvent molecules between cells on the grid are calculated along reactive trajectories. The average solvent fluxes form a much smaller set of variables and provide a much simpler description of the reaction under study than the full list of atomic positions and momenta. The feasibility and utility of the method are demonstrated by producing reduced descriptions of the mechanisms of the dissociation reaction of NaCl in water and of the rearrangement of the alanine dipeptide in water. Comparisons between our method and other possible data reduction techniques are discussed.