Abstract
Silicon undergoes a phase transition from the semiconducting diamond phase to the metallic $\beta$-Sn phase under pressure. We use quantum Monte Carlo (QMC) calculations to predict the transformation pressure and compare the results to density-functional theory (DFT) calculations employing various exchange-correlation functionals, including the local-density approximation (LDA), generalized-gradient approximations (PBE, PW91, WC, AM05, PBEsol), and the hybrid functional HSE06. Diffusion Monte Carlo (DMC) predicts a transition pressure of $14.0 \pm 1.0$ GPa, slightly above the experimental range of 11.3–12.6 GPa. The HSE06 hybrid functional predicts 12.4 GPa, in excellent agreement with experiments. Standard LDA and GGA functionals significantly underestimate the transition pressure, ranging from 3.5 to 10.0 GPa. We also demonstrate that stress anisotropy lowers the equilibrium transition pressure, potentially explaining experimental variations and the slight DMC overestimate.
Introduction
The pressure-induced transformation of Silicon from its four-fold coordinated semiconducting diamond structure to the six-fold coordinated metallic $\beta$-Sn structure is a fundamental benchmark in solid-state physics. Accurately predicting this transition is a rigorous test for computational materials science methods. Historically, standard DFT approaches have struggled to match experimental results, often underestimating the transition pressure by a wide margin. This study aims to resolve these discrepancies by employing high-level QMC methods and hybrid functionals, while also investigating the impact of non-hydrostatic stress conditions.
Methods
The research utilizes two primary computational frameworks:
- Density Functional Theory (DFT): A comprehensive suite of functionals was tested, including the Local Density Approximation (LDA), several Generalized Gradient Approximations (GGAs) such as PBE, PW91, WC, AM05, and PBEsol, and the hybrid functional HSE06. The hybrid functional is particularly notable for incorporating a portion of exact Hartree-Fock exchange, which often improves the description of electronic band gaps and structural properties.
- Quantum Monte Carlo (QMC): Specifically, Diffusion Monte Carlo (DMC) was used to solve the many-body Schrödinger equation more directly than DFT. DMC is often considered a "gold standard" for accuracy in electronic structure calculations, though it is significantly more computationally demanding.
- Stress Anisotropy Modeling: To account for real-world experimental conditions where pressure may not be perfectly hydrostatic, the authors modeled the energy landscape as a function of volume and the $c/a$ lattice ratio, allowing for the simulation of anisotropic stress.
Results and Discussion
Benchmarking Computational Methods
The study reveals a stark contrast between standard semilocal functionals and more advanced methods. LDA and various GGAs consistently predict transition pressures well below the experimental window of 11.3–12.6 GPa. For instance, LDA predicts a transition at 5.8 GPa, while GGAs range from 3.5 to 10.0 GPa.
In contrast, the HSE06 hybrid functional shows remarkable accuracy, predicting a transition pressure of 12.4 GPa. The DMC method predicts $14.0 \pm 1.0$ GPa, which, while slightly higher than the experimental range, represents a significant improvement over standard DFT.
The failure of semilocal functionals is attributed to their difficulty in describing the energy difference between the localized covalent bonds of the diamond phase and the more delocalized electron density of the metallic $\beta$-Sn phase. There is a clear correlation between a functional's ability to predict the electronic band gap and its accuracy in predicting this phase transition.
The Impact of Stress Anisotropy
A key insight of this work is the sensitivity of the transition pressure to stress anisotropy. In many high-pressure experiments, such as those using diamond anvil cells without a pressure-transmitting medium, the stress is not perfectly hydrostatic. The authors found that anisotropy along any of the cubic axes of the diamond phase significantly lowers the equilibrium transition pressure.
Quantitatively, the transition pressure decreases linearly with increasing deviatory stress $\Delta$ at a rate of approximately $2.4 \Delta$. This finding explains why experiments conducted under non-hydrostatic conditions often report lower transition pressures (around 8–9 GPa) and suggests that the slight overestimate by DMC might be partially due to the assumption of perfect hydrostaticity in the comparison.
Conclusion
This study provides a definitive benchmark for the diamond-to-$\beta$-Sn transition in Silicon. It highlights the superior accuracy of hybrid functionals like HSE06 and high-level QMC methods over standard LDA and GGA approaches. Furthermore, it identifies stress anisotropy as a critical factor in reconciling theoretical predictions with experimental observations, offering a more complete understanding of this classic phase transformation.
Summary updated