Researchers from the State Key Laboratory of Theoretical Physics at the Institute of Theoretical Physics, Chinese Academy of Sciences, have developed a new method to simulate photochemical processes more efficiently. The team, led by Zhenxing Zhu and Wei-Hai Fang, has enhanced the fewest-switches surface hopping (FSSH) technique, a popular method for simulating how molecules behave when exposed to light. Their work was recently published in the Journal of Chemical Physics.
The researchers have integrated long short-term memory (LSTM) networks, a type of artificial intelligence model, with FSSH to create a more efficient simulation framework. LSTM networks are particularly good at learning from sequential data, making them well-suited for modeling the complex dynamics of molecular systems. The team has redesigned the input features and training procedures of the LSTM networks to better represent the high-dimensional nuclear degrees of freedom in molecules.
To build accurate potential energy surfaces for both ground and excited states, the researchers have incorporated equivariant neural networks into their LSTM-FSSH framework. These networks help to ensure that the simulated molecular geometries and energies are physically realistic. The team has demonstrated the effectiveness of their new method by simulating the photoisomerizations of CH2NH and azobenzene, two molecules that undergo changes in structure when exposed to light.
The simulations showed that the new LSTM-FSSH method can accurately reproduce excited-state lifetimes and product yields, compared to conventional FSSH simulations. Notably, the researchers found that only 10 reference trajectories are needed to train the LSTM networks. Once trained, the LSTM-FSSH dynamics simulations can generate a large ensemble of trajectories very efficiently, providing collective results that are in good agreement with experimental data.
This new method has significant implications for the energy industry, particularly in the field of photovoltaics and photocatalysis. By providing a more efficient and accurate way to simulate photochemical processes, the LSTM-FSSH method could help researchers design and optimize new materials for solar energy conversion and storage. Additionally, the method could be used to study the degradation of materials exposed to light, helping to improve the durability and lifespan of energy-related technologies.
This article is based on research available at arXiv.