Exciton Transfer Simulations in LH2 Accelerated by Machine-Learned Hamiltonians Reveal Transient Delocalization Mechanism

The astonishing efficiency of exciton transfer in light-harvesting (LH) complexes has prompted a multitude of experimental and theoretical studies over the past two decades. However, the revision of the long-held role of electronic coherences calls for alternative explanations. In this study, we investigate the exciton motion through the LH2 complex of Rhodospirillum molischianum using a mixed quantum-classical approach, leading to a fully ’classical’ explanation of the transport efficiency. Our multi-scale approach comprises a coupled quantum mechanical molecular mechanics (QM/MM) embedding, a fragmentation of the electronic structure in the QM region via a Frenkel Hamiltonian, neural networks to accelerate the quantum calculations, and non-adiabatic molecular dynamics (NAMD) to treat the combined electronic-nuclear dynamics. This enables us to sample hundreds of trajectories of this large quantum system, each lasting several picoseconds. Our simulations describe the transitions within the B800 and B850 rings in agreement with experimental findings, suggesting an incoherent hopping process in the B800, and a more delocalized transfer in the B850 subsystem. The reorganization energies and excitonic couplings are comparable in the B850 ring, promoting the transport model “transient delocalization”. This theory is characterized by a dynamic interdependence of moderately delocalized excitons that exhibit large delocalization events. To demonstrate the efficiency of this transport, we compare the exciton dynamics in LH2 with the most efficient electron and exciton transfer processes in organic semiconductors. We find that the Bacteriochlorophyll’s unusual electronic properties, leading to tiny inner and outer-sphere reorganization energies, are the reason for the astonishing efficiency.