SAMOP 2021 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MO: Fachverband Molekülphysik
MO 9: Poster 2
MO 9.12: Poster
Freitag, 24. September 2021, 17:30–19:30, P
Spectral deep-learning for (ro-)vibrational calculations of weakly-bound molecules — •Jannik Eggers1,2, Yahya Saleh1,2, Vishnu Sanjay1,2,3, Andrey Yachmenev1,3, Armin Iske2, and Jochen Küpper1,3,4 — 1Center for Free-Electron Laser Science CFEL, Deutsches Elektronen-Synchrotron DESY, Hamburg — 2Department of Mathematics, Universität Hamburg — 3Center for Ultrafast Imaging CUI, Universität Hamburg — 4Department of Physics, Universität Hamburg
Planning and elucidating experiments on resonances in dissociation dynamics of molecules and molecular clusters requires accurate quantum mechanical calculations of (ro-)vibrational energies up to dissociation, which is a big challenge especially for larger molecules.
Standard approaches represent wavefunctions as linear combinations of some fixed basis set and the quality of the predictions highly depends on the choice of the basis set. Furthermore, the computational costs scale poorly with the dimension of the problem.
We present a nonlinear neural network-based variational framework to simultaneously compute several eigenstates and eigenfunctions of the Hamiltonian.
Unlike linear variational methods, neural network-based models seem to scale relatively well with the dimension of the problem.
While they were mainly used to successfully model ground states of quantum systems, our approach extends to excited states.
The key principle is to use neural networks as an adaptive basis and to optimize it, enabling us to use a much smaller basis set than in standard approaches without sacrificing accuracy.