Pressure and stress tensor of complex anharmonic crystals within the stochastic self-consistent harmonic approximation

TL;DR

This paper extends the stochastic self-consistent harmonic approximation (SCHA) to complex anharmonic crystals, enabling accurate computation of pressure, stress, and free energy minimization, demonstrated on water ice with excellent agreement to experiments.

Contribution

The work introduces a new formalism for evaluating pressure and stress tensors within stochastic SCHA and improves the free energy minimization algorithm for complex systems.

Findings

Accurately reproduces water ice thermal expansion from 0 K to 270 K.

Outperforms quasi-harmonic approximation by capturing anharmonic effects.

Efficiently handles systems with many degrees of freedom and broad phonon spectra.

Abstract

The self-consisted harmonic approximation (SCHA) allows the computation of free energy of anharmonic crystals considering both quantum and thermal fluctuations. Recently, a stochastic implementation of the SCHA has been developed, tailored for applications that use total energy and forces computed from first principles. In this work, we extend the applicability of the stochastic SCHA to complex crystals with many degrees of freedom, with the optimisation of both the lattice vectors and the atomic positions. To this goal, we provide an expression for the evaluation of the pressure and stress tensor within the stochastic SCHA formalism. Moreover, we develop a more robust free energy minimisation algorithm, which allows us to perform the SCHA variational minimisation very efficiently in systems having a broad spectrum of phonon frequencies and many degrees of freedom. We test and illustrate the new approach with an application to the phase XI of water ice using density-functional theory. We find that the SCHA reproduces extremely well the experimental thermal expansion of ice in the whole temperature range between 0 K and 270 K, in contrast with the results obtained within the quasi-harmonic approximation, that underestimates the effect by about 25 %.