Comprehensive calculations of Fe-Si system via deep learning force field: the core structure of Mercury

<title>Abstract</title> <p>Hybrid approaches combining deep learning and ab initio molecular dynamics (AIMD) provide comprehensive calculations abilities of large systems with full physical quantities coverage owing to the accelerated computation speed and ab initio accuracy advantages. Here we propose a unified framework to systematically and precisely extract the variation patterns of all relevant physical quantities as a function of state variables. A direct calculation, including the equation of states, thermoelastic properties, adiabatic curves, melting curves and transport properties, were conducted in Fe-Si (&lt;25 at. %) system to unveil the Mercury’s core structure. Except for solid phase at low temperatures, the viscosity and diffusion coefficients lie in the order of mPa s and 10- 9 m2 s- 1 26 respectively, validating the rationality of the predicted melting curves. The adiabatic and melting curves reveal a considerable solid inner core without top-down crystallization. The estimated total thermal conductivity of 48.0 W/m/K suggest the possibility of thermal stratification on top of Mercury's core. Analyzing six plausible structure mechanisms, the solid inner core with thermal stratified layer on top of the inner core were elucidated in the Mercury's core.</p>