Scalable nonadiabatic dynamics with correct decoherence and branching beyond surface hopping

• Workhorse. Fewest-switches surface hopping (Tully 1990) with ad hoc decoherence corrections is used in most excited-state dynamics studies. Subotnik et al. (Annu. Rev. Phys. Chem. 2016) review its known failures: overcoherence, inconsistent internal electronic-nuclear state, frustrated hops and sensitivity to the representation.
• Exact references. MCTDH and multilayer MCTDH (Beck et al. 2000; Wang and Thoss 2003) give numerically exact quantum dynamics but on model vibronic Hamiltonians of up to about a hundred modes, not with on-the-fly ab initio electronic structure in solution.
• Trajectory alternatives. Ab initio multiple spawning, multiconfigurational Ehrenfest, and exact-factorization-based surface hopping (Vindel-Zandbergen et al. 2021–2022; Dupuy et al. 2024; Ha et al. 2025) are improving decoherence treatments, but have not yet been shown in broad blind tests to beat well-parametrised surface hopping on realistic molecules. Benchmark frameworks are emerging (Martinka et al., arXiv:2512.19152, 2025).
• Experiments. Trapped-ion analogue simulators have observed geometric-phase interference around a conical intersection in model systems (Valahu et al., Nat. Chem. 2023), and Quantinuum has reported phase-estimation runs on an ethylene conical-intersection effective Hamiltonian (company white paper, 2026; a vendor claim, not a refereed or independently reproduced result). These are not general-molecule dynamics.
• Gap. No method meets all the listed requirements; reference quantum dynamics is unavailable for the chemically realistic cases that matter.

Photochemistry (vision, DNA photodamage, photoswitches, photocatalysts), OLED and solar-cell materials, charge-transfer and singlet-fission dynamics, and ultrafast spectroscopy interpretation depend on correct branching; a reliable method would turn excited-state simulation into a predictive design tool.

A published method with open implementation whose results match numerically exact or experimentally anchored reference dynamics within the stated tolerances on a pre-registered blind benchmark of at least 10 molecules including at least two in explicit solvent, validated by at least one independent group.