Applying Virtual Reality Technology and Neural Network Potentials for Molecular Dynamics Simulations in Industrial R&D

Neural Network Potentials (NNP) is a cutting-edge technique in computational materials science. This method employs atomic-level potentials learned from the ab-initio method to efficiently perform reactive molecular dynamics (MD) simulations with many atoms. As a result, NNP-MD can consider complex chemical systems such as organic-inorganic interfaces with water solvents.<br/><br/>However, analyzing the results of calculations with the software on a simple computer display, as in classical MD simulations, is difficult due to the large number of atoms involved in NNP-MD simulations. This difficulty has limited the analysis of MD calculations to the statistical level, even though atomistic-level observations are essential for understanding the origin of phenomena. Overcoming this limitation is essential for efficient academic research and industrial research and development (R&D).<br/><br/>To address this problem, we have employed virtual reality (VR) technology with a head-mounted display to analyze classical MD trajectories at an atomic scale, such as the motion of molecules at the interface of an inorganic substrate. VR technology not only assists computational science experts in analyzing calculation results but also dramatically supports phenomenological understanding by non-experts. Additionally, we found that VR technology is an effective communication tool between experts and non-experts, which is crucial for accelerating material R&D.<br/><br/>In this study, we apply VR technology to analyze complex catalytic reactions calculated by NNP-MD simulations and study the impact of the combination of VR and NNP-MD on industrial R&D. We employ Suzuki-Miyaura Cross-Coupling in water solvent with an atomically rough Rd substrate surface as an example of a complex chemical system.<br/><br/>By replaying the NNP-MD trajectory in VR, we could identify sites on the Pd substrate where chemical surface reactions occur, approach those reaction points, and observe the reaction step-by-step. As a result, VR proved to be a powerful tool for understanding the multi-step interfacial reactions obtained with NNP-MD. However, we realized that while NNP-MD has the advantage of observing the atomic behavior of chemical reactions, it is difficult to understand the physical origin of the reactions. Therefore, it is necessary to employ ab-initio calculations to gain a deeper understanding of physics.<br/><br/>From an industrial perspective, the combination of NNP-MD and VR should effectively share intuitive images of chemical reactions and accelerate materials R&D, as we have found in the case of classical MD.