New interaction potentials for multicomponent oxide glasses

Abstract

Atomistic simulation methods like molecular dynamics (MD) can give great insight into the structure-property correlations in materials. The ability of such methods to predict various properties accurately requires a reliable interaction potential that describes the forces between atoms. However, oxide glass has posed an especial challenge for MD simulations, due to its mixed ionic-covalent bonding nature, the importance of both short- and medium-range order, and the different local environments that can be adopted by similar species, depending on composition and thermodynamic conditions. There is hence a dearth of reliable interaction potentials for MD simulations of these systems. Moreover, many of the potentials in the literature are limited to specific compositions and narrow thermodynamic conditions under which they can be used. The goal of my thesis work was to overcome this long-standing challenge, by developing a new optimization scheme to parameterize computationally efficient interaction potentials that can reliably predict various properties of multicomponent oxide glasses over a wide range of compositions and thermodynamic conditions. This was achieved by maintaining a simple pair-wise functional form and using the composition dependent anion charge scheme suggested in the literature to partially take into account the polarizability effect.

These computationally efficient force fields are essential for large scale virtual mechanical tests of glass under complex loading conditions. Furthermore, using a simple pair-wise functional form and reference data produced from ab-initio methods allows the optimization scheme to be easily extended to developing potentials for new systems with no or limited experimental data so far.

The boron anomaly where the coordination of boron depends on the composition of the system and the thermodynamic conditions is another phenomenon which has proved to be challenging even for more complicated interaction potentials in the literature. Most recent approaches to resolving this issue have been to include a system dependent parameter that was fit to the composition dependent boron coordination trends observed in experiments. The major issue with such parameter sets is that they work well for the system where they have been optimized towards but cannot give reasonable results if the compositions are changed. The interaction potential parameters that I have developed have no such system specific parameters and can still reproduce this phenomenon, which gives them good transferability to other systems.

Mechanical and vibrational properties were especially improved while maintaining the ability of the potentials to reproduce an excellent structure. These improved potentials for oxide glasses can not only predict these properties reliably at ambient conditions, but also their response to external stimuli like high pressure. The compressibility anomaly of silica where the bulk modulus of silica glass first decreases with pressure, after reaching a minimum at a certain pressure, then increases as expected for the normal response of a solid to pressure. Even though some of potentials in the literature were able to show this effect, the location of the minimum was usually predicted at much higher pressures. The new optimized parameters for silica with high pressure density data in the cost function were able to reproduce this phenomenon and even predict the minimum in buck modulus at the correct pressure.

I developed a new fitting scheme interfaced with the open-source Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) MD software package to optimize interaction potentials for oxide glasses. My approach was to fit to results from both accurate first principles calculations and experiments under various thermodynamics conditions, and explicitly incorporate the radial distribution function (RDF) of the liquid at high temperature and the vibration density of states (VDOS), density, coordination and elastic modulus of glass at room temperature into the cost function. Using this approach, I parameterized efficient pair-wise potentials for various systems containing multiple network formers (e.g., Si, B and Al) and modifiers (e.g., Li, Na, K, Ca). These new interaction potentials can predict very well the melt structure as compared to ab-initio data and various structural and elastic properties of glasses as compared to experimental data. These potentials are transferable and can be mixed to reproduce the structure and properties of various multi-component boroaluminosilicate glasses over a large range of compositions and thermodynamic conditions.