The authors of the study set out to determine the nature of the phase transition of hydrogen fluid from the molecular phase to the conducting (metallic) phase. Earlier, theoretical and computational studies showed that at high pressure hydrogen experiences an abnormal increase in diffusion (the ability of molecules of a substance to penetrate the molecules of another substance). However, it was problematic to determine the coefficients of diffusion and viscosity (resistance to penetration) from experimental data, whereas direct calculations (ab initio) were too cost-intensive. This is why the researchers decided to use a combination of machine learning methods and classical molecular dynamics. This approach has made it possible to study the dynamic properties of hydrogen fluids in large models.
“In order to build interatomic potential, we collected data from ab initio calculations: energies and forces for system configurations at different temperatures and densities. Our co-author Nikolai Chtchelkatchev from the Institute for High Pressure Physics RAS selected the configurations with the largest prediction error in active learning mode and performed additional calculations to improve the accuracy of the model. As a result, we obtained DeepMD potential, system energy function from the coordinates of all atoms. It reproduces the results of ab initio calculations, but much faster,” Vyacheslav Lukyanchuk, one of the authors of the study and assistant at the Department of Computational Physics, is quoted as saying by the MIPT.
The aforementioned DeepMD potential provides data on vibration spectra and coefficients of diffusion and viscosity in various temperature and density ranges. With its help, the scientists managed to calculate the viscosity of dense heated hydrogen fluid, which was previously impossible due to high computational costs. They found that viscosity increases significantly during the phase transition and later decreases as density grows. This is consonant with trends that can be observed in alkali metals, including lithium.
“We are exploring the idea that the viscosity of hydrogen fluid at high pressures can behave similarly to that of alkaline melts. We will test this hypothesis in our future studies,” Nikolai Kondratyuk, executive director of the Center for Computational Physics, is quoted as saying by the MIPT.
The calculations carried out during the study confirmed the existence of a first-order phase transition in liquid hydrogen (during the transition from one aggregate state to another), accompanied by a pronounced change in density, diffusion and viscosity. A significant increase in the diffusion coefficient occurs at temperatures of 700 K, 800 K and 900 K (from 427 to 627 degrees Celsius), including due to the dissociation of hydrogen molecules and heightened mobility of atoms.
