Parallel optimization of a reactive force field for polycondensation of alkoxysilanes

We have optimized a reactive force field (ReaxFF) in order to model the gelation of alkoxysilanes in bulk precursor solutions. The force field parameter set was refined using a parallelized local search algorithm. Using this approach, each processor is assigned a small list of parameters. At the end of every iteration, all parameters are updated simultaneously after being independently evaluated. In comparison to the serial evaluation of parameters, this results in faster parametrization of ReaxFF, as well as helps to prevent entrapment in local minima. The resulting model is found to reproduce hydrolysis and condensation reaction energies well. By applying the model to the condensation of silicic acid monomers at several temperatures, the activation energy of silane condensation is determined. The expected behavior, a gradual depletion of hydrolyzed silicon and growth of condensed silica clusters is observed over timescales of a few nanoseconds. The new model is also verified by modeling the early stages of clusterization in an alkoxysilane precursor solution. Both hydrolysis and condensation reactions are observed in a system containing a mixture of tetramethoxysilane, methanol, and water.