Thema der Dissertation:
Machine learning and reaction dynamics: From spectroscopic constants of diatomic molecules to buffer gas chemistry
Machine learning and reaction dynamics: From spectroscopic constants of diatomic molecules to buffer gas chemistry
Abstract: This research delves into the spectroscopic properties and chemistry of diatomic molecules, which have potential applications in quantum information and ultracold chemistry.
In the first part of this research, the Diatomic Molecular Spectroscopy Database has been implemented, accessible through a dynamic website, which consolidates spectroscopic information and facilitates computation and visualization of Franck-Condon factors. A comprehensive dataset of experimental electric dipole moments has also been created. Machine learning models based on these datasets reveal relationships among spectroscopic constants. The results showcase the machine learning model's ability to predict them accurately using spectroscopic constants, emphasizing their intricate relationship with chemical bonding. The implementation of the datasets also enables a thorough assessment of advanced quantum chemistry methods. The accuracy of the coupled-cluster with single, double, and perturbative triple excitations [CCSD(T)] in predicting electric dipole moments is explored, along with the computation and comparison of hyperfine constants for the a$^3\Pi$ state of AlF to experimental values. The study highlights the importance of evaluating both experimental and theoretical methodologies.
In the second part, metal monofluorides that are relevant to laser cooling and trapping, including AlF and CaF, have been investigated via ab initio molecular dynamics simulations. A comparison of different fluorine-donor molecules in producing AlF and CaF through metal atom ablation in a buffer gas cell is presented. Additionally, an efficient machine learning method for fitting the potential energy surface of the AlF-AlF system is introduced, trained on relevant configurations from molecular dynamics simulations at the CCSD(T) level.
In the first part of this research, the Diatomic Molecular Spectroscopy Database has been implemented, accessible through a dynamic website, which consolidates spectroscopic information and facilitates computation and visualization of Franck-Condon factors. A comprehensive dataset of experimental electric dipole moments has also been created. Machine learning models based on these datasets reveal relationships among spectroscopic constants. The results showcase the machine learning model's ability to predict them accurately using spectroscopic constants, emphasizing their intricate relationship with chemical bonding. The implementation of the datasets also enables a thorough assessment of advanced quantum chemistry methods. The accuracy of the coupled-cluster with single, double, and perturbative triple excitations [CCSD(T)] in predicting electric dipole moments is explored, along with the computation and comparison of hyperfine constants for the a$^3\Pi$ state of AlF to experimental values. The study highlights the importance of evaluating both experimental and theoretical methodologies.
In the second part, metal monofluorides that are relevant to laser cooling and trapping, including AlF and CaF, have been investigated via ab initio molecular dynamics simulations. A comparison of different fluorine-donor molecules in producing AlF and CaF through metal atom ablation in a buffer gas cell is presented. Additionally, an efficient machine learning method for fitting the potential energy surface of the AlF-AlF system is introduced, trained on relevant configurations from molecular dynamics simulations at the CCSD(T) level.
Zeit & Ort
18.01.2024 | 16:00
Hörsaal A (1.3.14)
(Fachbereich Physik, Arnimallee 14, 14195 Berlin)