Machine learning for excitation energy transfer dynamics

A wellknown approach to describe the dynamics of an open quantum system is compute master equation evolving reduced density matrix system. This plays important role in describing excitation transfer through photosynthetic light harvesting complexes (LHCs). The hierarchical equations motion (HEOM) was adapted by Ishizaki and Fleming [J. Chem. Phys.130, 234111 (2009)] simulate biological regime. We generate a set time-dependent observables that depict coherent propagation electronic excitations LHCs solving HEOM. computationally intractable problem here addressed using classical machine learning (ML). ML architecture constructed model character it used solve inverse for systems within HEOM approach. objective determine whether trained can perform Hamiltonian tomography time dependence as inputs. demonstrate capability convolutional neural networks tackle this research problem. models developed predict parameters such excited state energies inter-site couplings up 99.28% accuracy.