A computational investigation of nucleation mechanism and polymorph selection in crystallization

This thesis demonstrates the use of molecular simulations and specifically multidimensional free energy calculations as a powerful tool to understand non-classical nucleation mechanisms. With a motivation to develop a computational methodology that can predict nucleation rates and identify the kinetically favored polymorph without relying on CNT, two studies have been conducted: 

First,  NaCl nucleation from aqueous solution was studied using molecular simulations. By calculating the free energy of nucleation as a function of structure specific nucleus size coordinates, it was shown that at high concentrations, NaCl nucleation occurs through the formation of composite clusters that have a crystalline core surrounded by a layer of amorphous particles. The effect of supersaturation on the mechanism was investigated by rerunning the calculations at a higher concentration and a shift was observed in the nucleation mechanism from single step to 2-step nucleation. The shift was shown to be resulting from a change in the relative stability of the intermediate amorphous solid phase.

Second, nucleation of the Lennard-Jones (LJ) fluid from the melt was investigated with a focus on the competition between two possible crystal structures. The free energy surface of nucleation, plotted against polymorph-specific nucleus size variables, revealed that polymorph selection in the LJ fluid does not happen during nucleation, but when the emerging clusters are much larger than the critical cluster size, in contrast with the classical nucleation theory assumption. It was observed that the post-nucleation events like crystal growth and polymorphic transformations played a significant role in determining the final structure of the grown crystals. Since a CNT based crystallization model is not able to track the structural changes within the nucleated clusters, a novel population balance modeling framework was proposed. The proposed model connects the structure-specific nucleation rates obtained from atomic-scale simulations with the post-nucleation events and is able to predict the correct polymorphic distribution obtained from 500 nucleation trajectories.