Sparse Gaussian process based machine learning first principles potentials for materials simulations: Application to batteries, solar cells, catalysts, and macromolecular systems

To design new materials and understand their novel phenomena, it is imperative to predict the structure and properties of materials that often rely on first-principles theory. However, such methods are computationally demanding and limited to small systems. This topical review investigates machine learning (ML) approaches, specifically non-parametric sparse Gaussian process regression (SGPR), to model the potential energy surface (PES) of materials, while starting from the basics of ML methods for a comprehensive review. SGPR can efficiently represent PES with minimal ab initio data, significantly reducing the computational costs by bypassing the need for inverting massive covariance matrices. SGPR rank reduction accelerates density functional theory calculations by orders of magnitude, enabling accelerated simulations. An optimal adaptive sampling algorithm is utilized for on-the-fly regression with molecular dynamics, extending to interatomic potentials through scalable SGPR formalism. Through merging quantum mechanics with ML methods, the universal first-principles SGPR-based ML potential can create a digital-twin capable of predicting phenomena arising from static and dynamic changes as well as inherent and collective characteristics of materials. These techniques have been applied successfully to materials such as solid electrolytes, lithium-ion batteries, electrocatalysts, solar cells, and macromolecular systems, reproducing their structures, energetics, dynamics, properties, phase-changes, materials performance, and device efficiency. This review discusses the built-in library universal first-principles SGPR-based ML potential, showcasing its applications and successes, offering insights into the development of future ML potentials and their applications in advanced materials, catering to both educational and expert readers.