Alberto Fabrizio¹, Sergi Vela¹, Ksenia R. Briling¹, Clemence Corminboeuf^1*

1 Institut des Sciences et Ingénierie Chimiques, École Polytechnique Fédérale de Lausanne (EPFL), CH-1015 Lausanne, Switzerland

* Corresponding authors emails: clemence.corminboeuf@epfl.ch

DOI10.24435/materialscloud:8n-50 [version v1]

Publication date: Jun 19, 2021

How to cite this record

Alberto Fabrizio, Sergi Vela, Ksenia R. Briling, Clemence Corminboeuf, Learning the exciton properties of azo-dyes, Materials Cloud Archive 2021.86 (2021), doi: 10.24435/materialscloud:8n-50.

Description

The ab initio determination of the character and properties of electronic excited states (ES) is the cornerstone of modern theoretical photochemistry. Yet, traditional ES methods become readily impractical when applied to fairly large molecules, or when used on thousands of different systems. In contrast, Machine Learning (ML) techniques have demonstrated their accuracy at retrieving ES properties of large molecular databases at a reduced computational cost. Especially for excited states applications, non-linear algorithms tend to be specialized and to target only individual properties. Learning fundamental quantum chemical objects potentially represents a more efficient, yet complex, alternative as a large number of molecular properties could be then extracted through post-processing. Herein, we report the general framework able to learn three fundamental objects of an ES: the hole and particle densities, as well as the transition density. We demonstrate the advantages of targeting those outputs of an ES computation, and address the additional complexity of learning 3-dimensional scalar fields. We apply our predictions to obtain a list of derived properties, including the state character and the topological descriptors of the exciton, for the two bands (nπ* and ππ*) of a collection of 3432 azoheteroarene photoswitches. In doing so, we prove that the proposed approach offers a way to access ES information of different kinds, bypassing the need of traditional electronic structure computations, while retaining ab initio accuracy.

Materials Cloud sections using this data

Files

File name	Size	Description
ML_TRANSITION.tar.gz MD5md5:9a5f04b18905ee80ad693cc7487d9463	995.3 MiB	Compressed tar ball containing ground state, transition, hole and particle densities, as well as geometries and additional data. See README.txt for a more detailed description of the content
README.txt MD5md5:7abd325219a0a8859ecd7c49faceebda	2.7 KiB	README

License

Files and data are licensed under the terms of the following license: Creative Commons Attribution 4.0 International.
Metadata, except for email addresses, are licensed under the Creative Commons Attribution Share-Alike 4.0 International license.

External references

Keywords

Machine Learning Photoswitches Azo-dyes Hole and Particle Densities EPFL MARVEL/DD1 SNSF ERC