Dresden 2020 – wissenschaftliches Programm
Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
TT: Fachverband Tiefe Temperaturen
TT 16: Poster Session Superconductivity, Cryogenic Particle Detectors, Cryotechnique
TT 16.42: Poster
Montag, 16. März 2020, 15:00–19:00, P2/EG
Machine learning for quantum chemistry with quantum computers — •Tomislav Piskor1,2, Sebastian Zanker1, Thilo Mast1, Peter Schmitteckert1, Frank Wilhelm-Mauch2, and Michael Marthaler1 — 1HQS Quantum Simulations GmbH, Haid-und-Neu-Straße 7, 76131 Karlsruhe — 2Theoretical Physics, Saarland University, 66123 Saarbrücken, Germany
Simulating chemical systems is a major field of interest not only for the pharma and chemistry, but also for the automotive industry. One such example is the simulation of functional groups of a large molecule or proteine, which can be useful for QM/MM methods. In order to get the exact ground state of the functional group, we can use quantum computers in the future. However, every call to a quantum computer will be relatively expensive, making high-throughput simulations with quantum computers unfeasible.
To bypass this, we propose the following scheme: a few single point calculations are determined with the quantum computer and then extended to more conformations with machine learning methods. We use sGDML (symmetric gradient domain machine learning), where the atomic coordinates of the molecules are given as the input and the corresponding forces as the output. A modified Gaussian kernel is then used in order to obtain the trained force fields and by performing an integration with respect to the atomic coordinates one gets the potential energy surface. This routine was tested on molecules such as enediyne and malonaldehyde, where we observe very good results not only for the force field, but also energy predictions.