Machine-Learning-enabled ab initio study of quantum phase transitions in SrTiO$_3$

TL;DR

This study employs machine learning-enhanced ab initio methods to investigate quantum phase transitions in SrTiO$_3$, successfully reproducing isotope effects and ferroelectric behavior with quantum and anharmonic effects included.

Contribution

It introduces a machine learning-based approach within the self-consistent harmonic approximation to accurately simulate quantum phase transitions in SrTiO$_3$.

Findings

Reproduces isotope-induced ferroelectric transition in SrTiO$_3$

Demonstrates narrow phase space between quantum paraelectric and ferroelectric states

Shows machine learning potentials enable temperature-dependent quantum anharmonic phonon simulations

Abstract

We use the self-consistent harmonic approximation (SSCHA) with machine learning interatomic potentials to calculate the effect of $^{18}$O substitution on the properties of quantum paraelectric SrTiO$_3$ (STO). We find that calculations including both quantum and anharmonic effects are able to reproduce the experimentally observed isotope effect, in which replacement of $^{16}$O by $^{18}$O induces the ferroelectric state, and demonstrate that the ferroelectric phase transition in ST$^{18}$O can be reproduced in a purely displacive manner. We calculate the ferroelectric soft mode frequency as a function of volume, lattice parameters and temperature for ST$^{16}$O and ST$^{18}$O, and find that the phase space in which ST$^{16}$O shows quantum paraelectric behaviour, while ST$^{18}$O becomes ferroelectric is narrow. Our study shows that machine learning interatomic potentials enable temperature-dependent simulations that include quantum and anharmonic phonon effects, however quantitative prediction of phase diagrams remains challenging due to a lack of universally accurate electronic structure methods.