Improving atom-scale models of clay minerals using machine learning

Abstract

Bentonite clay is a geomaterial with numerous interesting physicochemical properties. Notably, it shows promise as a buffer material for underground long-term storage of radioactive waste. The material is predominantly composed of smectite montmorillonite, as well as considerable quantities of other minerals such as quartz, illite, feldspars, and others. Understanding this material at the atomic scale can help understand how minute changes in the environment or the potential release of radioisotopes would affect a spent fuel long-term repository. This paper presents a model to describe the interatomic interactions within the main components of bentonite clay via machine learning. Specifically, the paper introduces the moment tensor potential (MTP), how it will be adapted to describe clay minerals, and present the first set of results, involving an interatomic model of silicon and silica, that constitute the backbone of clay minerals. Preliminary results show great promise for the description of interatomic forces within clays systems.