多粒子座標の不変性及び同変性を満足した深層学習モデルの提案と分子データ解析への応用

抄録

Molecular simulation is an effective method for investigating the behavior and properties of molecular systems. However, expert knowledge, experience are required to obtain useful information from the large amount of data produced from simulations of systems with multiple particles. In contrast, in recent years, researchers have increasingly used deep learning to obtain valuable predictions from such data. To use deep learning in this way, we must consider the invariance and the equivariance of the particle coordinates, depending on the physical quantities of interest. There are currently several deep learning models that satisfy the requirement of invariance in systems with multiple particles. However, there are, as yet, no deep learning models that meet the equivariance condition. In this study, we proposed a deep learning model with minor theoretical modifications that has invariance and equivariance for systems with multiple particles. To validate our model, we compared our method with simulation results for the binary classification of solids and liquids in a quasi-two-dimensional confined system. In doing so, we were able to confirm the effectiveness of the proposed model.