CSI High Performance Computing Seminar: Thibault Charpentier, CEA Paris-Saclay

Wednesday 20 Dec 2023, 11:00 → 15:00 US/Eastern

Title: Modelling NMR Properties of Oxide Glasses with Machine Learning

Speaker: Dr. Thibault Charpentier, Research Director at CEA Paris-Saclay

Abstract: 

Solid-State NMR has become an essential spectroscopy for the elucidation of the glass structure. With recent advances in DFT computations, NMR can now be combined with Molecular Dynamics (MD) simulations to help interpret the experimental data [1] because the spectral broadening associated with the structural disorder of glass that often limits their detailed interpretation of NMR spectra. However, such calculations are severely limited in system size by the high- computational cost of DFT computations.

In recent years, machine learning (ML) approaches have emerged as a powerful method for accelerating MD and computing materials properties with an accuracy close to that of DFT methods. [2] In solid state NMR, few approaches have been recently proposed that can be applied to oxide glasses [3] which are complex materials with structural features that are still debated. We describe here an approach based on the concept of atomic- centered descriptors[4] (such as SOAP) combined with Kernel Ridge Regression (KRR) or Artificial Neural Network (ANN) techniques. This enables the prediction of NMR properties for structural models of thousands of atoms in a few seconds. Its combination with Monte Carlo simulations methodologies combining MD, NMR and neutron data will be illustrated through several applications to simple borate and silicate glasses.

Speaker Bio: Dr. Thibault Charpentier is Research Director at CEA Paris-Saclay since 2008. He received his degree of ‘‘Ingénieur’’ in 1994 from the ESPCI ParisTech, and earned his PhD in Solid State Physics in 1998 from the University of Orsay in the field of NMR of quadrupolar nuclei. Since he was hired by the CEA in 1998, he worked on the development of NMR methodologies for nuclear waste materials such as oxide glasses (structure, impact of irradiation and chemical durability), ceramics and cements. In parallel, he worked on theoretical aspects of Spin Dynamics in NMR (Floquet-Magnus Theory). His current research focuses on developing computational methodologies based on machine learning for modelling NMR experiments of disordered materials from first-principles.