Extraction of material properties through multi-fidelity deep learning from molecular dynamics simulation

Simulation of reasonable timescales for any long physical process using molecular dynamics (MD) is a major challenge in computational physics. In this study, we have implemented an approach based on multi-fidelity physics informed neural network (MPINN) to achieve long-range MD simulation results over large sample space with significantly less cost. The fidelity our present study the integration timestep size simulations. While simulations larger produce lower level accuracy, it can provide enough computationally cheap training data MPINN learn accurate relationship between these low-fidelity and high-fidelity obtained smaller timestep. We performed two benchmark studies, involving one component LJ systems, determine optimum percentage required high saving. show that important system properties such as energy per atom, pressure diffusion coefficients be determined accuracy while saving 68% costs. Finally, demonstration applicability methodology practical studied viscosity argon-copper nanofluid its variation temperature volume fraction by MPINN. Then compared them numerous previous studies theoretical models. Our indicate predict at wide range small number first implementation conjunction predicting nanoscale properties. This pave pathways investigate more complex engineering problems demand