Advantages of multi-dimensional biasing in accelerated dynamics: application to the calculation of the acid $pK_a$ for acetic acid

Abstract

The use of accelerated sampling methods such as metadynamics has shown significant advantage in calculations that involve infrequent events, which would otherwise require sampling a prohibitive number of configurations to determine, e.g., difference in free energies between two or more chemically distinct states such as in the calculation of acid dissociation constants $K_a$. In this case, the most common method is to bias the system via a single collective variable (CV) representing the $coordination~number$ of the proton donor group, which yields results in reasonable agreement with experiments. Here we study the deprotonation of acetic acid using the reactive force field ReaxFF and observe a significant dependence of $K_a$ on the simulation box size when biasing only the coordination number CV, which is due to incomplete sampling of the deprotonated state for small simulation systems, and inefficient sampling for larger ones. Incorporating a second CV representing the distance between the H$_3$O$^+$ cation and the acetate anion results in a substantially more efficient sampling both accelerating the dynamics and virtually eliminating the computational box size dependence.