The ability to perform accurate calculations efficiently is crucial for computational materials design. Molecular dynamics (MD) is one such popular technique for materials design and provides information on structure and dynamical evolution of materials. The accuracy and efficiency of MD, however, hinges strongly on the quality of the force fields that describe the various  inter/intramolecular interactions. Historically, force field (FF) development for MD can take months to several years while being limited to specific material systems. In this talk, we will discuss how our streamlined approach to FF development using first principles training data and machine learning algorithms is dramatically bringing down this timeframe to weeks to days. The procedure involves several steps including (a) generation and manipulation of extensive fitting data sets through electronic structure calculations and/or experiments, (b) defining functional forms, (c) formulating novel highly optimized training procedures, (d) coarse-graining to improve efficiency while retaining accuracy, and (e) subsequently coding and implementing these algorithms on high performance computers (HPCs). We will also discuss the validation of this approach on several diverse material systems ranging from precious metal/oxide nanocatalysts to newly discovered two dimensional materials such as stanene to more common yet complex system such as water.

Machine learnt models for accurate yet efficient materials design

254th National Meeting (2017)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.