The construction and application of Markov state models for colloidal self-assembly process control
Markov state models have been widely applied to study time sequential events in a variety of disciplines. Due to their versatility for representing system stochasticity, Markov state models hold the promise to simplify simulation and design control policies for colloidal self-assembly systems. In this manuscript, we investigate the effects of state discretization, transition time, sampling approach, and the number of samples on the accuracy of a Markov state model for a colloidal self-assembly process. The model accuracy is evaluated based on the performance of the optimal control policy, calculated with a Markov decision process-based optimization framework, for controlling a Brownian dynamics simulation to produce perfect crystals. The results suggest using a dynamic sampling, a transition time similar to the system characteristic time, a clustering-based state discretization, and an average of at least five samples per state, to efficiently build an accurate Markov state model.