Ion diffusion in ceramics is a fundamental process in many technologically important materials. This includes transport of ions through solid electrolytes in solid-state batteries, fuel cells and sensors. It is of interest to gain a better understanding of this process to help design new materials for applications with improved functionality.
Efficient computational predictions of solid-state diffusion from atomistic models can be a resource-efficient aid towards this goal, by permitting large numbers of structures to be modeled to find novel candidate materials or elucidate structure-property relationships. However, traditional methods such as molecular dynamics become inefficient for this purpose due to the long timescales associated with diffusion in most solids. Lattice models and kinetic Monte Carlo present an alternative that can bridge the gap to sufficient timescales, but are challenging to automate and apply when atomic structures become complex and varying.
This thesis addresses these computational modeling challenges by building on the Tunnel and Transition State algorithm, by which a lattice model can be automatically constructed from a three-dimensional potential energy field felt by a mobile particle. First an efficient mean-field sampling scheme was introduced to take into account interactions between the mobile ions. Then, accuracy and transferability were improved by employing density functional theory to sample the ion-framework interactions, which was made possible by reducing the computational cost of the sampling by leveraging symmetry, exclusion of blocked regions and interpolation.Through this work an efficient and transferable method was established for predicting ion diffusion coefficients in desirable agreement with molecular dynamics at significantly reduced computational cost.
The developed method was subsequently applied alongside other models to investigate Li+ diffusion in inorganic electrolytes and entropic contributions from the distribution of ions, their interactions, and framework dynamics were elucidated, shedding light on their importance for fast ion conduction and consequences of modeling approximations. Furthermore, the compositional dependence of Li+-ion diffusivity in lithium aluminum titanium phosphate was studied, where effects of Li+ concentration, framework interactions and structure were disentangled across a wide compositional interval. Here, the developed method in conjunction with machine-learning potentials enabled efficient simulation of larger, more disordered systems and varying mobile ion concentration.
Hannes Gustafsson successfully defended his thesis on 11.06.2026. The supervisor team consisted of Amber Mace, Senja Barthel, Christer Elvingson, and Carl Nettelblad. Brandon Wood was the opponent. The corresponding publications can be found here.
Picture: The candidate Hannes Gustafsson, with his supervisor team consisting of Amber Mace, Senja Barthel, Christer Elvingson, and Carl Nettelblad, and opponent Brandon Wood.