ICTP-East African Institute for Fundamental Research
KIST2 Building CST
University of Rwanda
CMSP Webinar - Julia Westermayr
Prof Julia Westermayr (University of Leipzig) will present a seminar on machine learning for excited states of molecules
- Speaker: Dr Julia Westermayr (University of Leipzig, Germany)
- Date: Wednesday 18 January, 2023
- Time: 16:00 – 17:00 HRS (GMT+2)
- Venue: Online
Title: From light excitation to nonradiative decay: Machine Learning for excited states of molecules
Abstract: Machine learning algorithms have had an enormous impact on almost every part of the theoretical natural sciences, but the photochemistry of molecules has remained widely untouched due to the high complexity and computational effort involved in a photochemical study . In this talk, we will introduce the SchNarc approach  to enable an efficient and accurate computation of the excited states, especially the nonadiabatic dynamics, via a machine-learned description of excited-state energies, their derivatives, and related vectorial properties, such as dipole moments or coupling vectors [2,3]. Three limitations are addressed and solved within the SchNarc framework: (i) The high costs of reference computations are minimized by applying an efficient training set generation for the excited states . (ii) The excited-state properties are arbitrary with respect to their sign, which greatly complicates a training with conventional machine learning algorithms, but can be solved by applying a "phase-free" training algorithm . (iii) Vectorial excited-state properties are described with high accuracy via latent machine-learned properties, which are based on physical relations [2,3]. The developed algorithms are successfully applied to study the excited states of the methyleneimmonium cation: A UV/visible absorption spectrum and the atomic charge distribution due to light excitation can be studied along with the mixed quantum-classical, nonadiabatic dynamics. These can be investigated computationally efficiently, accurately, and on long time scales, which would not be feasible without the help of machine learning [2-4]. The method will further be demonstrated for the study of the excited state dynamics of the amino acid tyrosine, unravelling unconventional dynamics after light excitation .
 J. Westermayr, P. Marquetand, “Machine learning and excited-state molecular dynamics” Mach. Learn.: Sci. Technol. 1(2), 025009 (2020).
 J. Westermayr, M. Gastegger, P. Marquetand, „Combining SchNet and SHARC: The SchNarc machine learning approach for Excited-State Dynamics”, J. Phys. Chem. Lett. 11(10), 3828-3834 (2020).
 J. Westermayr, P. Marquetand, “Deep learning for UV absorption spectra with SchNarc: First steps towards transferability in chemical compound space”, accepted in J. Chem. Phys. (2020).
 J. Westermayr, M. Gastegger, M. Menger, S. Mai, L. González, P. Marquetand, “Machine learning enables long time scale molecular photodynamics simulations”, Chem. Sci. 10, 8100-8107 (2019).
 J. Westermayr, M. Gastegger, D. Vörös, L. Panzenboeck, F. Joerg, L. González, P. Marquetand, “Deep learning study of tyrosine reveals that roaming can lead to photodamage”, Nat. Chem. 14, 914-919 (2022).
Zoom Meeting ID: 822 9614 7302; Password: 053152
Short bio: Julia Westermayr studied Chemistry at the University of Vienna (Austria), where she completed her PhD under the supervision of Prof. Leticia González and Dr. Philipp Marquetand in 2020. For her PhD, she was awarded the uni:docs fellowship of the University and received the Sigrid-Peyerimhoff prize, a PhD prize that is yearly awarded to outstanding PhD thesis in German-speaking countries. After her PhD, she received a fellowship of the Austrian Science Fund for her PostDoctoral research in the group of Prof. Reinhard Maurer at the University of Warwick, which she terminated early to start a Juniorprofessor position at the University of Leipzig in October this year. In her research, she uses artificial intelligence to develop new methods and theories to better understand light-matter interactions and to leverage this knowledge to accelerate molecular and materials discovery.