Title: Machine learning for first-principles simulations of electrons and nuclei

Abstract: Scientific progress in chemistry and material science relies on the accuracy and efficiency of first-principles simulations. Ideally, these simulations should incorporate the quantum nature of all electrons and nuclei, achieving predictive accuracy across materials science and chemistry. Traditional “full quantum” simulations are, however, computationally prohibitive. This presentation introduces a contemporary first principles framework that significantly reduces these costs. Our method leverages physics-based machine learning to estimate the system’s Born-Oppenheimer potential energy surface and other essential electronic properties, such as polarization and polarisability. We predict hitherto unfeasible first-principles phase diagrams of nanoscale systems [1] and relative stabilities of molecular crystal polymorphs [2]. Further, we address the challenge of modelling nuclear motion by mapping quantum dynamics to an effective classical correction akin to effective potentials by Feynman and Hibbs [3]. Our work translates quantum nuclear motion to simple classical molecular dynamics. To showcase our method’s capability, we predict vibrational spectra of bulk and interfacial aqueous phases, achieving quantitative agreement with experiments for the first time [4]. Our model offers a path for comprehensive quantum simulations, combining accuracy with the ease of prevalent classical methods.

References

1. V. Kapil, C. Schran, A. Zen, J. Chen, C. Pickard, and A. Michaelides. Nature 2022, 609, 7927

2. V. Kapil, and E. Engel. Proc. Nat. Acad. Sci. 2022, 119, 6

3. F. Musil, I. Zaporozhets, F. Noé, C. Clementi, and V. Kapil, J. Chem. Phys. 2022, 157, 18

4. V. Kapil, D. Kovács, G. Csányi, and A. Michaelides, Faraday Discuss., 2023

Bio: Venkat Kapil utilizes machine learning-driven techniques to model materials at finite temperatures based on first principles. His research focuses on understanding the thermodynamics, transport, and quantum mechanics of complex nanoscale systems.

Venkat received his undergraduate degree in Theoretical Chemistry from IIT Kanpur in 2015. He earned a PhD in Material Science in Michele Ceriotti's group from the Swiss Federal Institute of Technology Lausanne (EPFL). During his postdoctoral research at the University of Cambridge, Venkat worked with Angelos Michaelides' research group and others to develop new simulation methods at the intersection of machine learning and quantum statistical mechanics for full quantum first-principles simulations of materials. He received the Swiss National Science Foundation's "Mobility Fellowship," the "Early Career Oppenheimer Fellowship," and the "Sydney Harvey Junior Research Fellowship" from Churchill College, University of Cambridge. Since January 2024, Venkat has joined University College London and the London Centre of Nanotechnology as a Lecturer (Assistant Professor) in the Department of Physics and Astronomy.