2017/07/19 by Florian Häse, Häse, Florian, Christoph Kreisbeck +3 · 2 citations
Computer Science · Materials Science · Physics and Astronomy · #Chemical Physics (physics.chem-ph) #FOS: Computer and information sciences #FOS: Physical sciences #Machine Learning (stat.ML) #Machine Learning in Materials Science #Neural Networks and Reservoir Computing #Spectroscopy and Quantum Chemical Studies
paper · pdf · doi:10.48550/arxiv.1707.06338
openalex publication_date 2017/07/19 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Understanding the relationship between the structure of light-harvesting\nsystems and their excitation energy transfer properties is of fundamental\nimportance in many applications including the development of next generation\nphotovoltaics. Natural light harvesting in photosynthesis shows remarkable\nexcitation energy transfer properties, which suggests that pigment-protein\ncomplexes could serve as blueprints for the design of nature inspired devices.\nMechanistic insights into energy transport dynamics can be gained by leveraging\nnumerically involved propagation schemes such as the hierarchical equations of\nmotion (HEOM). Solving these equations, however, is computationally costly due\nto the adverse scaling with the number of pigments. Therefore virtual\nhigh-throughput screening, which has become a powerful tool in material\ndiscovery, is less readily applicable for the search of novel excitonic\ndevices. We propose the use of artificial neural networks to bypass the\ncomputational limitations of established techniques for exploring the\nstructure-dynamics relation in excitonic systems. Once trained, our neural\nnetworks reduce computational costs by several orders of magnitudes. Our\npredicted transfer times and transfer efficiencies exhibit similar or even\nhigher accuracies than frequently used approximate methods such as secular\nRedfield theory\n