Apr 22, 2024
4:00pm - 4:15pm
Room 320, Level 3, Summit
Stefano Falletta1,Andrea Cepellotti1,Albert Musaelian1,Anders Johansson1,Chuin Wei Tan1,Boris Kozinsky1,2
Harvard University1,Robert Bosch Research and Technology Center2
Stefano Falletta1,Andrea Cepellotti1,Albert Musaelian1,Anders Johansson1,Chuin Wei Tan1,Boris Kozinsky1,2
Harvard University1,Robert Bosch Research and Technology Center2
Polarization is an essential physical quantity for understanding the properties of dielectrics and ferroelectrics [1]. The modern theory of polarization [1] and further developments on electric enthalpy functionals [2] have enabled the study of polarization in periodic electronic structure calculations. Interestingly, predicting the polarization over a molecular dynamics allows for the determination of experimentally-relevant quantities such as infrared and Raman spectra from first principles. However, this remains a challenging problem in electronic structure calculations, because of their high computational cost.<br/><br/>Here, we introduce an equivariant neural network approach to efficiently learn and predict polarization for each atomic configuration, building on and extending state-of-the-art machine learning force field architectures. This allows one to determine the autocorrelation function of polarization over long-lasting molecular dynamics, thereby giving access to infrared spectrum, frequency-dependent dielectric constant, and Raman cross section from first principles. We implemented our scheme in the E(3)-equivariant NequIP/Allegro framework [3,4], and interfaced it with the LAMMPS code for performing molecular dynamics calculations.<br/><br/>[1] R. Resta, Macroscopic polarization in crystalline dielectrics: the geometric phase approach, Rev. Mod. Phys. 66, 899 (1994).<br/>[2] P. Umari and A. Pasquarello, Ab initio molecular dynamics in a finite homogeneous electric field, Phys. Rev. Lett. 89, 157602 (2002).<br/>[3] S. Batzner, A. Musaelian, L. Sun, M. Geiger, J. P. Mailoa, M. Kornbluth, N. Molinari, T. E. Smidt, and B. Kozinsky, E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials. Nat. Commun., 13, 2453 (2022)<br/>[4] A. Musaelian, S. Batzner, A. Johansson, L. Sun, C. Owen, M. Kornbluth, and B. Kozinsky, Learning local equivariant representations for large-scale atomistic dynamics Nat. Commun., 14, 579 (2023)