Electron correlation and the phase diagram of Si

TL;DR

This paper investigates the inaccuracies in first-principles calculations of silicon's phase diagram caused by exchange-correlation errors in LDA, and shows that GGA improves the predictions by addressing these errors.

Contribution

It demonstrates that exchange-correlation errors in LDA lead to underestimation of silicon's melting temperature and phase transition pressures, and that GGA reduces these errors.

Findings

LDA under-predicts silicon's melting temperature by ~20%

GGA significantly improves phase diagram predictions

Errors are mainly due to exchange-correlation, not system size or sampling

Abstract

Previous first-principles calculations of the melting properties of Si, based on the local-density approximation (LDA) for electronic exchange-correlation energy, under-predict the melting temperature by ~ 20%. We present new first-principles results demonstrating that this problem is due to non-cancellation of exchange-correlation errors between the semiconducting solid and the metallic liquid. It is shown that other sources of error, particularly those due to system size and Brillouin-zone sampling, can be made negligible. The same LDA errors cause an underprediction of the pressure of the diamond-Si --> beta-tin-Si transition. The generalized-gradient approximation largely corrects both features of the Si phase diagram.