Machine learning photodynamics beyond the Frenkel exciton model: Intermolecular vibrations trigger ultrafast singlet fission in pentacene crystal

Crystalline pentacene is a model solid-state light-harvesting material because its quantum efficiencies exceed 100% via ultrafast singlet fission (SF). The SF mechanism in pentacene crystals is unresolved due to insufficient electronic information in time-resolved experiments and intractable quantum mechanical calculations for simulating realistic crystal dynamics. We combine multiscale multiconfigurational approach and machine learning (ML) photodynamics to understand SF in pentacene crystals. Our simulations identify two competing SF channels in the herringbone and parallel dimers, underpinning the high efficiency of SF in pentacene crystal. The predicted SF time constants (61 and 33 fs) are in excellent agreement with experiments (78 and 35 fs). The trajectories highlight the essential role of intermolecular stretching between monomers in generating the multi-exciton state. The ML-photodynamics resolved the elusive interplay between electronic structure and vibrational relations in the SF dynamics, enabling fully atomistic excited-state dynamics for the solid state with multiconfigurational quantum mechanical calculations.