Automated Framework for the Inversion of Experimental Data to Atomistic Structure Using Computer Vision and Multi-Objective Evolutionary Algorithms

We can achieve significant progress in our efforts to improve existing materials and discover new materials when we combine theory and experiments. We develop software tools to utilize state-of-the-art computational methods and computing resources to solve complex structure inversion problems from characterization data.<br/><br/>First, we developed the software package Ingrained [1] to obtain an approximate initial structure from TEM images of grain boundaries and STM images of surface structures. Ingrained can generate atomic structures of grain boundaries using only a few specifications such as the system, orientation, and angle. We use this ability to obtain the atomistic model of grain boundaries in battery and photovoltaic materials and further characterize the model to understand underlying mechanisms. The Ingrained software can also produce the best fit STM simulation of a surface structure to an experimental counterpart. We use this tool to successfully determine the atomistic structure of the newly synthesized “2D borophane” [2]. The Ingrained software successfully captures the subtle differences in the simulated STM images of three candidate structures and helps to identify the exact structure of the experimentally synthesized 2D material.<br/><br/>Secondly, we developed FANTASTX (Fully Automated Nanoscale To Atomistic Structure from Theory and eXperiments), a multi-objective genetic algorithm to find low energy (metastable) structures that provide the best fit experimental data. This modular software can generate simulated experimental data of different methods such as pairwise distribution function (PDF), structure factor (S(Q)), X-ray absorption spectra (XAS), X-ray reflectivity (XRR) and, TEM and STM images (using Ingrained). FANTASTX is a steady-state genetic algorithm that creates new structures using basin-hopping and mating operations. It calculates the energy using density functional theory or force fields and then simulates the required experimental data of each candidate structure. Thus, FANTASTX helps to find the experimentally observed and metastable atomistic structures. The software supports different geometries such as bulk materials, surfaces, grain boundaries, nanoclusters, and polymers and is easily extendable to other geometries and experimental simulation methods. We show an example where we combine the utility of Ingrained and FANTASTX software packages. We obtain an initial structure of a grain boundary from the TEM image using Ingrained. We then perform a local structure search of the interface region of this grain boundary structure to find a lower energy local configuration with a better fit to the experimental TEM image.<br/><br/><b>References</b><br/>[1] Eric Schwenker, Venkata Surya Chaitanya Kolluru, Jinglong Guo, Xiaobing Hu, Qiucheng Li, Mark C Hersam, Vinayak P Dravid, Robert F Klie, Jeffrey R Guest, Maria KY Chan, “Ingrained--An automated framework for fusing atomic-scale image simulations into experiments,” arXiv preprint arXiv:2105.10532 (2021).<br/>[2] Qiucheng Li, Venkata Surya Chaitanya Kolluru, Matthew S Rahn, Eric Schwenker, Shaowei Li, Richard G Hennig, Pierre Darancet, Maria KY Chan, Mark C Hersam, “Synthesis of borophane polymorphs through hydrogenation of borophene,” Science 371, 1143-1148 (2021).<br/><br/><b>Funding Acknowledgements</b><br/>This work is supported by the U.S. Department of Energy (DOE) Office of Science Scientific User Facilities AI/ML project titled, “A Digital Twin for Spatiotemporally Resolved Experiments. We acknowledge the support from the BES SUFD Early Career award. Use of the Center for Nanoscale Materials, an Office of Science user facility, was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Contract No. DE-AC02-06CH11357.