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 32: Condensed-matter simulations augmented by advanced statistical methodologies (joint session DY/CPP)
CPP 32.3: Vortrag
Montag, 16. März 2020, 16:00–16:15, HÜL 186
Exploring Chemical Reaction Space with Machine Learning — •Sina Stocker1, Gábor Csányi2, Karsten Reuter1, and Johannes T. Margraf1 — 1Chair of Theoretical Chemistry, Technical University Munich, Germany — 2Department of Engineering, University of Cambridge, United Kingdom
Reaction networks are essential tools for the analysis, visualization and understanding of chemical processes in such diverse fields as catalysis, combustion and the origin of life. For complex processes, the number of individual reaction steps in such a network is so large that an exhaustive first-principles calculation of all reaction energies and rates becomes prohibitively expensive. In this contribution, we use machine learning (ML) to accelerate the exploration of chemical reaction space, in analogy to the more established ML-based exploration of chemical space. To this end, we generated a new reactive reference database of open- and closed-shell organic molecules. This allows us to apply "chemical space" ML methods to predict the thermochemistry of reaction networks. We also develop explicitly "reaction space" based ML approaches to directly predict reaction properties. The performance of these methods confirms the potential of ML for the high-throughput screening of large reaction networks.