First-principles machine-learning study of infrared spectra of methane under extreme pressure and temperature conditions

Abstract

Methane’s role in the Earth’s mantle environment highlights the need for studies under extreme conditions. Traditional methods like ab initio molecular dynamics (AIMD) are limited by time and system size, but machine learning offers a new approach. This study uses machine learning to create a force field for bulk methane, simulating conditions from 1445 K to 2000 K and pressures from 14.4 to 48 GPa. We generate molecular dynamics trajectories, compare them with AIMD, and develop a neural network model to predict dipoles for infrared (IR) spectra calculation. Our methodology advances efficient exploration of hydrocarbons under extreme conditions.