Click here to join.

Deep Learning for Excited States of Molecules: Efficient Prediction of Orbital Energies and Photoemission Spectra

WCPM: Julia Westermayr (Warwick)

Abstract:

Theoretical simulations of the excited states of molecules can help to decipher fundamental mechanisms that underly light-matter interaction. However, many accurate first-principles methods are computationally not efficient enough to allow for the simulation of large and complex molecules or for a high-throughput screening, which would be necessary for a targeted search of novel functional materials.(1,2) In this talk, we will show how physically-inspired machine learning models can overcome the limitations of more expensive theoretical methods and enable computationally efficient simulations of electronic properties of excited molecules across chemical compound space. We have developed a model that can predict orbital energies and molecular resonances associated with electron addition and removal at the accuracy of many body perturbation theory. The applicability of the developed model is demonstrated by predicting orbital and quasiparticle energies for a wide range of organic molecules with potential relevance in molecular electronic devices.(3) Furthermore, we show the ability of our method to predict photoemission spectra of unseen molecules in close agreement with experiment.(4)

[1] J. Westermayr, P. Marquetand, Chem. Rev., in press, doi:10.1021/acs.chemrev.0c00749 (2020).

[2] J. Westermayr, M. Gastegger, K. T. Schütt, R. J. Maurer, arXiv:2102.08435 (2021).

[3] Stuke A., C. Kunkel, D. Golze, M. Todorović, J. T. Margraf, K. Reuter, P. Rinke, H. Oberhofer, Sci. Data. 7, 58 (2020).

[4] J. Westermayr, R. J. Maurer, arXiv:2103.09948 (2021)