Combining molecular simulations with artificial neural networks to determine the electrochemical properties of solvated species

Molecular simulations play an important role in the development of new technologies for energy production and storage. There are, of course, various approaches to molecular simulation, each of which offers its own advantages and disadvantages. Classical molecular dynamics simulations, for example, are able to be applied to large, complex systems, yet electron transfer and changes of oxidation state during the simulations are not captured. Density functional theory molecular dynamics (DFTMD) simulations compute the electronic structure but are limited to smaller and less complex systems. Artificial neural networks (ANN) have surged in recent years as a means to relate various features of a system to a measurable quantity.

In this work, we propose an approach using an ANN trained to DFT calculations following classical molecular dynamics simulations to predict the so-called vertical energy gap, which can be used to quantify electrochemical properties. We investigate the electrochemical properties of solvated organic molecules proposed for use in supercapacitor devices. This approach reduces computational costs associated with predicting the electrochemical properties of complex systems and may also be used to improve classical force fields used for electrochemically active species.