Regensburg 2019 – wissenschaftliches Programm

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

DY: Fachverband Dynamik und Statistische Physik

DY 31: Nonlinear Dynamics, Synchronization and Chaos

DY 31.4: Vortrag

Mittwoch, 3. April 2019, 10:45–11:00, H20

Application of Machine Learning Methods to Problems in Transition State Theory — •Tobias Mielich and Jörg Main — Institut für Theoretische Physik 1, Universität Stuttgart, Germany

Machine learning algorithms are getting increasingly more popular across various fields of science. Their usage in solving physics problems has already shown them to be an effective tool to assist classical algorithms and solutions [1]. This talk aims to show different approaches of using artificial neural networks to aid in rate calculations of chemical reactions in driven systems in the realm of transition state theory. The networks can be used to approximate functions like the time-dependent dividing surface, the times when trajectories cross that surface, or even the potential parameter dependent reaction rate itself. It is important to use the correct approach during training to get optimal results. Techniques like cyclic learning rates and network ensemble predictions shall be discussed.

P. Schraft et al., Phys. Rev. E 97, 042309 (2018)