(524c) Optimizing Sampling Efficiency in the Einstein Crystal Free Energy Method

The Einstein Crystal Method (ECM) is an atomistic simulation approach for computing the free energy of molecular solids that has received significant interest in recent years. The ECM has been used in the last decade to compute the melting temperature of ice, assess relative polymorph stabilities at ambient temperature, and estimate the absolute solubility of pharmaceutical compounds in solution among other properties. ECM calculations can be quite computationally expensive, and therefore a key to successfully using the ECM for property estimation is achieving efficient sampling along the entire thermodynamic path from the physical crystal to the ideal reference state. In particular, the efficiency of the sampling in the ECM is extremely sensitive to the lambda scheduling function used for the various alchemical steps in the method.

In this work, we assess a variety of different lambda scheduling functions for the various steps in the ECM including adding all-atom harmonic restraints as well as decoupling the various interactions in the system. We present the overlap matrices corresponding to different scheduling functions for each step to show the most efficient lambda spacing. Finally, we present an iterative lambda schedule optimization approach that can adaptively adjust lambda scheduling from a short simulation to optimize the overlap across adjacent lambda windows for a subsequent longer production run. These improved lambda scheduling functions have the potential to significantly improve the efficiency of all future ECM calculations.