Atomistic behaviors of barocaloric layered materials investigated with machine-learned force fields
ORAL
Abstract
Layered materials leveraging the order-disorder phase transition of hydrocarbon chains have been shown to exhibit significant barocaloric effects with promise for applications in thermal energy storage and conversion [1, 2]. The class of barocaloric layered materials is a vast landscape for materials design, but a complete understanding of how layer structure, composition, and hydrocarbon chain length influence the barocaloric effect is currently lacking.
In this talk, we demonstrate how molecular dynamics simulations powered by E(3)-equivariant machine learning can accurately realize a fully atomistic description of barocaloric effects in multiple classes of barocaloric layered materials. We use an efficient data curation protocol based on FLARE [3] active learning and use the scalable machine-learned force field Allegro [4] within the NequIP framework [5] to enable large-scale molecular dynamics simulations.
The simulations provide new insights to the entropically driven phase-change, and are validated against experimental data including quasi-elastic neutron scattering. The methods employed here move towards enabling the rational design of new and more performant materials for thermal energy management.
[1] J. Seo et al. Colossal barocaloric effects with ultralow hysteresis in two-dimensional metal–halide perovskites. Nat. Commun. 13 2536 (2022).
[2] J. Seo et al. Barocaloric effects in dialkylammonium halide salts. J. Am. Chem. Soc. 146 2736 (2024).
[3] J. Vandermause et al. Active learning of reactive Bayesian force fields applied to heterogeneous catalysis dynamics of H/Pt. Nat. Commun. 13 5183 (2022).
[4] A. Musaelian et al. Learning local equivariant representations for large-scale atomistic dynamics. Nat. Commun. 14 579 (2023).
[5] C. W. Tan et al. High-performance training and inference for deep equivariant interatomic potentials. arXiv preprint arXiv:2504.16068 (2025).
In this talk, we demonstrate how molecular dynamics simulations powered by E(3)-equivariant machine learning can accurately realize a fully atomistic description of barocaloric effects in multiple classes of barocaloric layered materials. We use an efficient data curation protocol based on FLARE [3] active learning and use the scalable machine-learned force field Allegro [4] within the NequIP framework [5] to enable large-scale molecular dynamics simulations.
The simulations provide new insights to the entropically driven phase-change, and are validated against experimental data including quasi-elastic neutron scattering. The methods employed here move towards enabling the rational design of new and more performant materials for thermal energy management.
[1] J. Seo et al. Colossal barocaloric effects with ultralow hysteresis in two-dimensional metal–halide perovskites. Nat. Commun. 13 2536 (2022).
[2] J. Seo et al. Barocaloric effects in dialkylammonium halide salts. J. Am. Chem. Soc. 146 2736 (2024).
[3] J. Vandermause et al. Active learning of reactive Bayesian force fields applied to heterogeneous catalysis dynamics of H/Pt. Nat. Commun. 13 5183 (2022).
[4] A. Musaelian et al. Learning local equivariant representations for large-scale atomistic dynamics. Nat. Commun. 14 579 (2023).
[5] C. W. Tan et al. High-performance training and inference for deep equivariant interatomic potentials. arXiv preprint arXiv:2504.16068 (2025).
–
Presenters
-
Marc L Descoteaux
- Harvard University