XMCQDPT2-fidelity transfer-learning potentials and a wavepacket oscillation model for ultrafast photodynamics
The Journal of Chemical Physics · 14.07.2026 · 165 (2)
DOI: 10.1063/5.0315698 ↗Kurz gefasst
Fachartikel in der Zeitschrift The Journal of Chemical Physics, online erschienen am 14.07.2026. Verfasst von Ivan V Dudakov und 7 weiteren Autor:innen. Beteiligte Einrichtung: Department of Chemistry, Lomonosov Moscow State University, Leninskie Gory 1/3, Moscow 119991, Russia. Die Arbeit nennt 52 Literaturangaben. Gefördert von: Russian Science Foundation.
Abstract (Originalsprache)
The accurate simulation of photochemical reactions requires methods that capture nonadiabatic transitions through conical intersections between different electronic states. While machine-learning interatomic potentials (MLIPs) offer a promising route to efficient nonadiabatic molecular dynamics, their training for excited states is often limited by the prohibitive cost of generating extensive datasets at a sufficiently high level of quantum chemistry theory. Here, we explore strategies for developing MLIPs that achieve multistate multireference perturbation theory accuracy, systematically comparing single-state, multi-output, multi-state, transfer learning, and delta-learning architectures. We show that transfer learning (TL) from CASSCF to XMCQDPT2 provides the best balance of accuracy and computational efficiency, dramatically reducing the amount of expensive reference data required. This methodology is validated on the methaniminium cation, whose complete photodissociation landscape, including S1 branching into photoisomerization and a direct H2-loss pathway mediated by a recently discovered conical intersection, is captured at the XMCQDPT2/SA(3)-CASSCF(12,12) level. We show that final product yields are largely independent of training strategy, whereas the TL model produces distinct ultrafast population dynamics compared to the randomly initialized model. Finally, we develop a wavepacket oscillation model for fitting excited-state population dynamics, which quantitatively reproduces the ultrafast non-exponential decay and extracts state- and channel-specific lifetimes, directly linking quantum transition probabilities to classical rate constants.
Autor:innen und Institutionen
- Department of Chemistry, Lomonosov Moscow State University, Leninskie Gory 1/3, Moscow 119991, Russia.
- MSU Institute for Artificial Intelligence, Lomonosov Moscow State University, Moscow 119192, Russia.
- Institute of Quantum Physics, Irkutsk National Research Technical University, 83 Lermontov Street, Irkutsk 664074, Russia.
ORCID-Profile: Ivan V Dudakov, Pavel M Radzikovitsky, Dmitry S Popov, Denis A Firsov, Vadim V Korolev, Daniil N Chistikov, Vladimir E Bochenkov, Anastasia V Bochenkova
Förderung
- Russian Science Foundation
Zitieren
APA
Dudakov, I. V., Radzikovitsky, P. M., Popov, D. S., Firsov, D. A., Korolev, V. V., Chistikov, D. N., Bochenkov, V. E., & Bochenkova, A. V. (2026). XMCQDPT2-fidelity transfer-learning potentials and a wavepacket oscillation model for ultrafast photodynamics. The Journal of Chemical Physics, 165(2). https://doi.org/10.1063/5.0315698
Vancouver
Dudakov IV, Radzikovitsky PM, Popov DS, Firsov DA, Korolev VV, Chistikov DN, et al. XMCQDPT2-fidelity transfer-learning potentials and a wavepacket oscillation model for ultrafast photodynamics. The Journal of Chemical Physics. 2026;165(2). doi:10.1063/5.0315698
BibTeX
@article{Dudakov2026,
author = {Dudakov, Ivan V and Radzikovitsky, Pavel M and Popov, Dmitry S and Firsov, Denis A and Korolev, Vadim V and Chistikov, Daniil N and Bochenkov, Vladimir E and Bochenkova, Anastasia V},
title = {XMCQDPT2-fidelity transfer-learning potentials and a wavepacket oscillation model for ultrafast photodynamics},
journal = {The Journal of Chemical Physics},
year = {2026},
volume = {165},
number = {2},
doi = {10.1063/5.0315698}
}
EUVIMED