Dependence of the structure and dynamics of liquid silicon on the choice of density functional approximation

Abstract

In addition to its technological relevance, silicon poses a challenge for first principles simulations because it undergoes a semiconductor-to-metal transition upon melting. Moreover, the resulting metallic liquid contains a mixture of metallic and covalent bonding. This coexistence of fundamentally different interactions is difficult to describe within approximate density functional methods, which oftentimes cannot accurately describe these two extremes simultaneously. We report an investigation of the structure, dynamics, and thermodynamics of liquid silicon using ab initio molecular dynamics simulations with three density functional approximations: the local density approximation (LDA), the Perdew-Burke-Ernerhof (PBE) generalized gradient approximation, and the strongly-constrained and appropriately normed (SCAN) meta-generalized gradient approximation. We demonstrate that SCAN describes this liquid with better accuracy than the other often-used functionals because it can simultaneously capture covalent and metallic bonding with similar high accuracy.

Accurate critical pressures for structural phase transitions of group IV, III-V, and II-VI compounds from the SCAN density functional

“…makes SCAN more predictive than GGAs or meta-GGAs that satisfy fewer exact constraints. At a computational cost at most only a few times greater than that of PBE, SCAN meets many condensed-matter challenges: ferroelectrics 12,13 , metal surfaces 14 , formation energies 15 , structure prediction 15 , liquid water 16 , and liquid silicon 17 .…”

Refined description of liquid and supercooled silicon from ab initio simulations

“…In this work, we move toward a quantitative description of l-Si using the strongly-constrained and appropriately-normed (SCAN) meta-generalized gradient approximation (meta-GGA) [20]. SCAN has recently been shown to provide a quantitative representation of diversely bonded systems, including metallic, covalent, and even intermediate-range many-body van der Waals interactions [21][22][23]. Using SCAN, we provide a refined description of the behavior of l-Si at low temperatures, including an improved estimate of the melting point and the onset of features reminiscent of a liquid-liquid phase transition (LLPT).…”

“…We simulated l-Si at a range of temperatures us-ing Born-Oppenheimer molecular dynamics simulations within the Vienna ab initio simulation package (VASP) [24], following previous work [21,22]. Electronic structure calculations were performed using DFT within the framework of the projector augmented wave method [25], employing the SCAN meta-GGA [20].…”

Reliable lattice dynamics from an efficient density functional

“…While efficient and reasonably accuracy, one problem with these conventional density functionals is their transferability in the materials space where different compounds can have very different chemical bonds. Recent progress has shown that semi-local meta-GGAs can maintain this efficiency while being accurate for a wide variety of materials [9,[11][12][13][14][15][16][17][18][19][20][21][22], exemplified by the strongly constrained and appropriately normed (SCAN) meta-GGA [11,12]. Unfortunately, extensive use has shown that SCAN suffers numerical problems that are exaggerated in phonon calculations, making reliably obtaining accurate phonon spectra from SCAN calculations a challenging task.…”