Thermoelastic properties of bridgmanite using Deep Potential Molecular Dynamics

TL;DR

This paper employs deep learning-enhanced molecular dynamics to accurately predict the thermoelastic properties of MgSiO3-perovskite under Earth's lower mantle conditions, bridging the gap between computational efficiency and precision.

Contribution

It introduces a hybrid deep learning and DFT approach with new potentials for reliable, high-temperature elastic property predictions of MgPv, outperforming traditional methods.

Findings

DP-SCAN and DP-LDA accurately predict high-temperature elastic properties.

The method closely matches experimental measurements.

It offers a computationally efficient way to study Earth's interior materials.

Abstract

MgSiO_3-perovskite (MgPv) plays a crucial role in the Earth's lower mantle. This study combines deep-learning potential (DP) with density functional theory (DFT) to investigate the structural and elastic properties of MgPv under lower mantle conditions. To simulate complex systems, we developed a series of potentials capable of faithfully reproducing DFT calculations using different functionals, such as LDA, PBE, PBEsol, and SCAN meta-GGA functionals. The obtained predictions exhibit remarkable reliability and consistency, closely resembling experimental measurements. Our results highlight the superior performance of the DP-SCAN and DP-LDA in accurately predicting high-temperature equations of states and elastic properties. This hybrid computational approach offers a solution to the accuracy-efficiency dilemma in obtaining precise elastic properties at high pressure and temperature conditions for minerals like MgPv, which opens a new way to study the Earth's interior state and related processes.