The structural dynamics of a molecule are determined by the underlying potential energy landscape. Conical intersections are funnels connecting otherwise separate energy surfaces. Posited almost a century ago 1 , conical intersections remain the subject of intense scientific investigation 2–4 . In biology, they play a pivotal role in vision, photosynthesis, and DNA stability 5,6 . In ultrafast radiationless de-excitation 1,7 , they are vital to ameliorating photon-induced damage. In chemistry, they tightly couple the normally separable nuclear and electronic degrees of freedom, precluding the Born-Oppenheimer approximation 8 . In physics, they manifest a Berry phase, giving rise to destructive interference between clockwise and anti-clockwise trajectories around the conical intersection 9 . Accurate theoretical methods for examining conical intersections are at present limited to small molecules. Experimental investigations are challenged by the required time resolution and sensitivity. Current structure-dynamical understanding of conical intersections is thus limited to simple molecules with around 10 atoms, on timescales of about 100 fs or longer 10 . Spectroscopy can achieve better time resolution, but provides only indirect structural information. Here, we present single-femtosecond, atomic-resolution movies of a 2,000-atom protein passing through a conical intersection. These movies, extracted from experimental data by geometric machine learning, reveal the dynamical trajectories of de-excitation via a conical intersection, yield the key parameters of the conical intersection controlling the de-excitation process, and elucidate the topography of the electronic potential energy surfaces involved.
