School of Chemistry Colloquium: Dr Chris Collins (University of Liverpool) Developing machine learning, artificial intelligence, classical simulations and robotics towards automating materials discovery

In the grand challenge of discovering new inorganic solids numerous tools continue to emerge: for the prediction of compositions, chemical structures, physical properties, and high throughput synthesis. In this talk I will discuss our work from recent publications, where we are developing our own workflows to guide the exploration of new materials, including integrating the use of machine learning (ML), artificial intelligence (AI), crystal structure prediction (CSP) and high throughput synthesis in two parts:
1. Accelerating target identification: The integration of generative machine learning with heuristic crystal structure prediction. We present a new implementation of our crystal structure prediction method FUSE, where the method is adapted to use both generative machine learning models to generate trial crystal structures & machine learnt inter atomic potentials. The combination of the introduction of the two ML methods results in a speed up in calculations of up to ~ 8 times. FUSE is then coupled with a new automated reasoning tool COMGEN, which allows a user to specify sets of chemical constraints and then explore the valid compositions which are a consequence of them, thus providing explainable results. We integrate COMGEN into an automatable workflow, where we perform probe structure prediction with FUSE on compositions suggested by COMGEN and use composition only ML property prediction to identify several candidate compounds for investigation as electrolytes in all solid-state Li batteries.
2. Accelerating the high throughput synthesis of solid-state oxides: The versatile but labour-intensive sub-solidus reaction pathway is the backbone of many functional materials syntheses. However, it proves challenging to automate because of the use of solid-state reagents. We present a high-throughput solid state workflow that permits rapid screening of oxide chemical space. This workflow accelerates materials discovery by enabling simultaneous exploration of compositions and synthetic conditions. The throughput is increased by using manual steps where actions are undertaken on multiple samples which are combined with automated processes. We exemplify this by extending the BaYxSn1−xO3−x/2 solid solution beyond the reported limit to a previously unreported composition and by exploring the Nb–Al–P–O composition space showing the applicability of the workflow to polyanion-based compositions beyond pure oxides.

This talk is open to final year undergraduate students, MSc students, PhD students, PDRAs and academic staff.