深度学习加速分子动力学模拟研究岩浆洋的传输性质

Understanding the physical and chemical properties of magma oceans is of crucial significance for understanding the habitability and internal constituents of the Earth and planetary bodies. Compared with high-temperature and high-pressure experiments, the molecular dynamics simulation has the advantage of simulating under more extreme conditions. However, the high computational cost of the molecular dynamics (MD) simulation had significantly constrained the time- and length- scales of the simulation. With the development of deep learning, many efforts have been devoted to tackling the computational accuracy and efficiency issues that molecular dynamics simulations had faced. Learning potential energy surfaces or forces among atoms are trained in a deep neural network based on the data set produced by using quantum mechanical methods including the density function theory (DFT). The trained deep learning model enables the MD simulation to have achieved good accuracy and efficiency at the same time. For calculations that are computationally intensive and require lengthy simulation times, research efforts such as building deep learning surrogate models of molecular dynamics or using machine learning to mine MD simulation trajectories for direct prediction have also shown an improved efficiency when compared to the original molecular dynamics simulations. The above research works have greatly improved efficiencies of the MD simulations. As an illustration of the acceleration in molecular dynamics simulation of the physical parameters of the magma ocean by using the deep learning, the calculation of thermal conductivity has been introduced. The studies show that with the increase of pressure, thermal conductivities of the silicate melt (MgSiO₃ ) are increased, but they are lower than the minerals under the corresponding conditions. The viscosities of the melt are firstly decreased and then increased with the increase of pressure. The studies suggest that the low cooling efficiency of silicate melt, as a product of magma ocean evolution, can be preserved for a long time in geological history in the core-mantle boundary.