Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

CPP: Fachverband Chemische Physik und Polymerphysik

CPP 107: Emerging Topics in Chemical and Polymer Physics, New Instruments and Methods

CPP 107.1: Vortrag

Freitag, 20. März 2020, 09:45–10:00, ZEU 255

Machine-Learned Polarizabilities: from Molecules to Bulk Raman Spectra — •David Wilkins, Andrea Grisafi, and Michele Ceriotti — Laboratory of Computational Science and Modelling, Institute of Materials, EPFL, Switzerland

The polarizability α of a chemical system is key in theoretically determining the results of spectroscopic experiments. However, α requires a high level of theory as well as extra computational effort on top of the ground-state electronic structure, meaning that to calculate an accurate polarizability can take several days for even a modest-sized molecule. Since an evaluation of spectra requires α to be known for an entire molecular simulation, the expense required would seem to stymie efforts to calculate an accurate spectrum.

The recently developed symmetry-adapted Gaussian process regression (SA-GPR) approach allows the prediction of tensor properties for general systems. I begin by showing that a model based on SA-GPR predicts the molecular α with the accuracy of coupled-cluster calculations, in a matter of seconds. This model can be applied to complex molecules with very high accuracy, suggesting that machine-learning should be the method of first resort for α predictions, rather than density functional theory, which is generally less accurate.

I finally show that SA-GPR is applicable also to the condensed phase, facilitating a prediction of spectroscopic experiments. In particular, I develop a model that allows a fast and accurate calculation of Raman spectra of bulk aqueous systems, as well as paving the way for the simulation of more complex nonlinear spectra.