Genetic Algorithms and Machine Learning for Predicting Surface Composition, Structure, and Chemistry: A Historical Perspective and Assessment

Abstract

Genetic algorithms (GA) and machine learning (ML) have a long history of development and use in chemistry. Recent algorithmic and computational advances, however, have brought these methods to the forefront of chemical research, and chemistry is experiencing a transformation in the way that machines and humans interact to pursue scientific advances. The field of materials chemistry, in particular, has witnessed a considerable expansion in the maturity of GA and ML approaches, as machine-based materials design ushers in a new era of materials development, discovery, and deployment. In addition to predicting new compositions and properties of bulk materials, GA and ML have also guided new insights into the structure, composition, and chemistry of materials surfaces. In this review, we focus on how GA and ML have been used in conjunction with chemical simulation techniques to advance understanding of surface chemistry, examining the history, recent work, and overall success of these applications.

Machine Learned Interatomic Potentials for Ternary Carbides trained on the AFLOW Database

“…The convex hulls discussed and presented above (Figure 3) were optimized with DFT, and the conclusions regarding thermodynamic stability of particular phases were made based upon these hulls. This procedure is common-place [63] but it might make one wonder about the limits of the utility of ML-IAPs in CSP. Part of the answer lies above where we show that ML can significantly reduce the number of required DFT calculations.…”

Global Optimization of Cation Ordering in Perovskites by Recommendation-Based Basin-Hopping

“…A common aspect of all these methods is the inclusion of local geometry relaxation of the structure, which is key to the success of global optimization. , Traditionally, these methods require expensive density functional theory (DFT) calculations to obtain the energy and forces, which face challenges in high-dimensional systems. Recent advances in artificial intelligence have enabled the acceleration of structural search by using accurate machine learning (ML) potentials trained on large data sets. − Further, on-the-fly ML potentials were recently proposed to avoid the demand of a large amount of training data and reduce the efforts in parameter tuning during model training …”

Ringing in the New Year with Gratitude and Editorial Team Updates

“…His group develops theoretical chemistry approaches to understand and design organic semiconductors, catalysts, and more. His group recently provided an excellent review for the journal entitled “Genetic Algorithms and Machine Learning for Predicting Surface Composition, Structure, and Chemistry: A Historical Perspective and Assessment”, and in 2019 he co-organized the Festschrift in honor of Jean-Luc Brédas. , …”

“…His group recently provided an excellent review for the journal entitled "Genetic Algorithms and Machine Learning for Predicting Surface Composition, Structure, and Chemistry: A Historical Perspective and Assessment", and in 2019 he coorganized the Festschrift in honor of Jean-Luc Bredas. 8,9 CM: What scientific questions or unsolved challenges in your field do you find most interesting?…”