P-039

Gabrielius Mockus

gabrielius.mockus@ftmc.lt

Justinas Šlepavičius, Deimantas Šmigelskas,  Rokas Silkinis, Linas Vilčiauskas

Center for Physical Sciences and Technology (FTMC), Lithuania

Machine Learning Force Field Parametrization And Applications For Phosphoric Acid Based Systems

Machine learning methods are making their way into computational chemistry. Machine-learned force fields (MLFF) implemented in the Vienna Ab initio Simulation Package (VASP) offer a fast and easy implementation for a wide range of systems. Two major benefits of such force fields is the increased accuracy in comparison to traditional empirical force fields and significant reduction in computational resources and accelerated simulation speed when compared to methods such as Density Functional Theory (DFT).

The objective of this research was to construct a reactive MLFF for phosphoric acid and related systems with the goal of studying the mechanism of proton conductivity. Phosphoric acid is one of the best pure proton conductors therefore a better understanding of such a mechanism could aid in constructing fuel cells, as well as better explain the role of proton conductivity in biological systems.

Our results indicate that MLFFs are a viable option for modeling the molecular dynamics of such systems with a similar accuracy to DFT methods, however at a significantly lower computational cost. 

References:

[1] J. Hafner, Journal of Computational Chemistry, 2008, 29 (13), 2044-2078. 

[2] R. Jinnouchi, Physical Chemistry Chemical Physics, 2022, 24 (25), 15522-15531. 

[3] L. Vilčiauskas, M. Tuckerman, G. Bester, S. J. Paddison, K.-D. Kreuer, Nature Chemistry, 2012, 4 (6), 461–466.  

[4] A. Mikalčiūtė, L. Vilčiauskas, Physical Chemistry Chemical Physics, 2021, 23 (10), 6213-6224.