Interatomic potential parameterization using particle swarm optimization: Case study of glassy silica

Abstract

We present a new method based on particle swarm optimization (PSO) for parameterization of interatomic potentials. Using glassy silica as a case study, we parameterize two interatomic potentials based on structural features obtained from ab initio simulations and experimental neutron diffraction data.

Theoretical study of the thermal conductivity of silica glass–crystal composites

“…Figure 6 shows the vDOS of selected NT structures, silica glass and α-quartz from Deng and Du potential, and also vDOS of α-quartz from BKS potential and ab initio 52 are included for comparison purpose. It is found that the vDOS peaks for α-quartz from both Deng and Du and BKS tend to lose details between 10 and 20 THz, which has also been found in the previous work by Christensen et al 35 The vDOS for α-quartz shows low densities at low acoustic region around 0-5 THz, and at intermediate acoustic region around 10 THz and high acoustic region around 25 THz. As the radius of the NT decreases, these peaks become almost flattened by the broadening of nearby peaks.…”

“…The long‐range forces were calculated by using the particle–particle particle–mesh algorithm with an accuracy of 10 −5 , and periodic boundary conditions were applied along all directions in all simulations. The Deng and Du and BKS potentials were able to replicate the vibrational density of states (vDOS) from ab initio simulation 30,35,36 . For coordination and radial and angular distribution function (ADF) analysis, bond cutoffs of 2.4 and 3.65 Å were used for defining the Si–O pair and Si–Si pair, respectively, which correspond to the position of the valley between the first peak and second peak in the Si–O and Si–Si pair distribution functions.…”

“…The Deng and Du and BKS potentials were able to replicate the vibrational density of states (vDOS) from ab initio simulation. 30,35,36 For coordination and radial and angular distribution function (ADF) analysis, bond cutoffs of 2.4 and 3.65 Å were used for defining the Si-O pair and Si-Si pair, respectively, which correspond to the position of the valley between the first peak and second peak in the Si-O and Si-Si pair distribution functions. These cutoffs were appropriate to define the first shell for the Si-O and Si-Si pairs.…”

Challenges and opportunities in atomistic simulations of glasses: a review

“…Various optimization methods are available to "navigate" the cost function landscape, that is, to identify the optimal forcefield parameters that minimize the cost function [Carré et al, 2016], including MC search [Iype et al, 2013], Bayesian optimization [Liu et al, 2019c,e], particle swarm optimization [Christensen et al, 2021], or gradient descent search [Shewchuk, 1994]. Note that this optimization problem is typically ill-defined since several degenerate sets of forcefield parameters can often minimize the cost function, that is, several competitive minimum positions can be found in the cost function landscape [Liu et al, 2020a[Liu et al, , 2019c.…”

Thermal conductivity of glasses: A structural point of view

“…This often serves as the shortcoming for making reliable predictions of κ due to how phonon-properties are often overlooked in the literature and due to how classical potentials may greatly reproduce structure and mechanics, but utterly fail to properly reproduce the thermal properties, especially that of the vibrational density of states. For oxide glasses the most prominent potential which nicely mimics the vibrational density of states is that of Matsui, developed for calcium aluminosilicates [110,125] while some success has been obtained for replicating the VDOS of glassy SiO 2 using other potentials [126]. The development of potentials useful for replicating phonon dynamics is, however, of rising interest [127].…”