Source-linked AI summary
Phase transitions of hybrid perovskites simulated by machine-learning force fields trained on-the-fly with Bayesian inference
Ryosuke Jinnouchi, Jonathan Lahnsteiner, Ferenc Karsai, Georg Kresse, Menno Bokdam
TL;DR
First-principles finite-temperature simulations are limited by unattainable system sizes and simulation times, while suitable machine-learning training structures are difficult to construct. The paper introduces on-the-fly machine-learning force fields generated during molecular dynamics and demonstrates them on hybrid-perovskite phase transitions, reproducing phases and transition behavior while revealing microscopic mechanisms.
Problem
First-principles simulations of finite-temperature materials require system sizes and simulation times that standard techniques cannot attain, and conventional machine-learning approaches require manually selected reference structures.
Method
The paper combines first-principles calculations with Bayesian on-the-fly machine learning to generate force fields automatically during molecular-dynamics simulations.
Results
The generated force fields predict energies, forces, and stress tensors with near-first-principles quality and reproduce hybrid-perovskite structures and phase transitions, including a 353 K tetragonal-to-cubic transition in MAPbI3.
Takeaways & Limitations
On-the-fly machine learning enables efficient large-scale finite-temperature sampling with little human intervention and provides atomic-scale insight into entropy-driven hybrid-perovskite transitions.
Abstract
from arXiv · showhide
Realistic finite temperature simulations of matter are a formidable challenge for first principles methods. Long simulation times and large length scales are required, demanding years of compute time. Here we present an on-the-fly machine learning scheme that generates force fields automatically during molecular dynamics simulations. This opens up the required time and length scales, while retaining the distinctive chemical precision of first principles methods and minimizing the need for human intervention. The method is widely applicable to multi-element complex systems. We demonstrate its predictive power on the entropy driven phase transitions of hybrid perovskites, which have never been accurately described in simulations. Using machine learned potentials, isothermal-isobaric simulations give direct insight into the underlying microscopic mechanisms. Finally, we relate the phase transition temperatures of different perovskites to the radii of the involved species, and we determine the order of the transitions in Landau theory.