Excited-state intramolecular proton transfer dynamics with neural network potentials and nuclear quantum effects

Excited-state intramolecular proton transfer is a prototypical ultrafast process that forms the foundation for many chemical and biological phenomena and is highly sensitive to changes in the underlying potential energy landscape. Here, we develop neural network potentials for 10-hydroxybenzo[h]quinoline that describe ground- and excited-state dynamics with near ab initio accuracy. These neural network potentials enable efficient sampling of the proton transfer coordinate and incorporation of nuclear quantum effects through path-integral molecular dynamics. The free energy profiles show that excited-state intramolecular proton transfer proceeds on a largely barrierless excited-state surface, with nuclear quantum effects reducing and smoothing the residual barrier. Non-equilibrium simulations yield an average proton transfer time of ~11.5 fs for quantum hydrogen, consistent with experiment, and reveal a clear isotope effect, with slower dynamics for deuterium and classical nuclei. Mechanistic analysis identifies transient proton donor–acceptor contraction as the key enabling coordinate, while the proton crosses the midpoint prior to reaching the minimum of this coordinate, indicating a dynamically driven transfer. Subtle changes in the conjugated backbone precede proton transfer, reflecting early electronic reorganization and coherent coupling between proton motion and backbone vibrations. Bond-length alternation and vibrational coherence analysis reveal isotope-dependent coherence loss, with deuteration producing stronger damping and more localized proton/backbone dynamics relative to hydrogen. These results establish neural network potential-based approaches as an effective framework for simulating excited-state proton transfer with explicit nuclear quantum effects.