Berlin 2024 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

CPP: Fachverband Chemische Physik und Polymerphysik

CPP 23: Modeling and Simulation of Soft Matter III

CPP 23.2: Talk

Wednesday, March 20, 2024, 09:45–10:00, H 0107

Symmetry-adapted polarization learning for vibrational spectroscopy — •David Wilkins — Centre for Quantum Materials and Technologies, School of Mathematics and Physics, Queen’s University Belfast, United Kingdom

The polarization of a material system is vital in modelling its response to light, and thus in predicting the results of spectroscopic experiments. However, there are several problems in these kinds of calculations: chief among them the requirement for electronic structure theory calculations and the fact that bulk polarizations are only defined modulo the so-called "quantum of polarization".

The solution of the first of these problems using machine-learning methods, some of which I have developed in the past few years, is by now in the mainstream of computational chemistry, but the second problem is still extant: the polarization is not a continuous function of the atomic positions. In this talk, I will compare and contrast two methods for overcoming this problem, based on localized dipole moments or on localized charges.

While the former method performs best for systems in which long-ranged electrostatics are not important, learning local charges is by far the best method when these effects are important (e.g. in electrolyte solutions). I show how both methods, where they are applicable, lead to very high-quality modelling of infrared and sum-frequency generation spectroscopy.

Keywords: Machine learning; Polarization; Sum-frequency generation; Molecular dynamics; Path integrals