Dresden 2020 – wissenschaftliches Programm

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 96: Organic Electronics and Photovoltaics IV

CPP 96.6: Vortrag

Donnerstag, 19. März 2020, 16:15–16:30, ZEU 260

Predicting the reorganization energies of organic semiconductors via machine learning — •Ke Chen, Christian Kunkel, Johannes Margraf, and Karsten Reuter — Technical University of Munich, Garching, Germany

The chemical space of possible organic semiconductors is enormous, and only few of them have been experimentally tested so far. It is therefore likely that many high performance organic semiconductors are still unknown. High-throughput computational screening can help accelerate their discovery, in particular when combined with highly efficient machine learning (ML) models. In this contribution, we focus on the so-called reorganization energy, which is one of the most critical molecular properties correlated with high charge carriers mobility in organic semiconductors.

The ML models presented herein use the smooth overlap of atomic positions (SOAP) for a local representation of atomic environments [1]. Based on this, two different global representations are studied to represent the molecular structures, namely the average global kernel and the 'auto-bag' method of Hammer et al. [2]. These representations are combined with linear and kernel ridge regression.

We find that these ML models can reliably identify the best candidates in a large chemical space of organic molecules. Furthermore, data-efficiency, prediction cost and the reliability of uncertainty estimates are compared.

[1] A. P. Bartók et al., Phys. Rev. B 87, 184115 (2013). [2] S. A. Meldgaard et al., J. Chem. Phys. 149, 134104 (2018).