The position of the critical point determines the top of the liquid-vapor coexistence dome, and it is a physical parameter of fundamental importance in the study of high-energy shocks, including those associated with large planetary impacts. For most major planetary materials, such as oxides and silicates, the estimated position of the critical point is below 1 g/cm3 at temperatures above 5000 K. Here we compute the position of the critical point of one of the most ubiquitous materials: MgO. For this we perform first-principles Molecular Dynamics (MD) simulations. We find the critical density to be in the 0.45–0.6g/cm3 range and the critical temperature in the 6500–7000 K range. We investigate in detail the behavior of MgO in the subcritical and supercritical regimes, and we provide insight into the structure and chemical speciation. We see a change in Mg-O speciation toward lower degrees of coordination as the temperature is increased from 4000 to 10 000 K. This change in speciation is less pronounced at higher densities. We observe the liquid-gas separation in nucleating nanobubbles at densities below the liquid spinodal. The majority of the chemical species forming the incipient gas phase consists of isolated Mg and O atoms and some MgO and O2 molecules. We find that the ionization state of the atoms in the liquid phase is close to the nominal charge, but it almost vanishes close to the liquid-gas boundary and in the gas phase, which is consequently largely atomic. We perform the same analysis on CaO. As both MgO and CaO are important geological materials, and they share many physical similarities, we are interested in their degree of similarity under extreme conditions. We find a slightly lower critical temperature of between 6000 and 6500 K, with a critical density between 0.5 and 0.7 g/cm3. Structurally we find that the MgOx is slightly more skewed to higher degrees of coordination than what we see for CaOx. This difference is likely caused by the smaller Mg-O distance compared to Ca-O. The biggest difference in the vapor species between the two systems is the amount of O, which is three times more common in CaO compared to MgO.
As artificial intelligence and machine learning (ML) have become the new buzz-words in most fields of research and industry, so has molecular dynamics started applying these methodologies to increase its performance. Recent developments in machine learning have exponentially increased the interest and new ML schemes seem to be published every year. In this work we apply one of the various ML potential to our previous work on MgO. Our choice of ML potential is the Gaussian Approximation Potential (GAP), which is based on Gaussian Process Regression (GPR). Training data is produced by performing static calculations on chosen MD configurations. Our new ML potential allows us to run simulations of three orders of magnitude larger within a fraction of the time it cost with firstprinciples MD, while keeping nearly the same level of accuracy. This enables us to study the vaporization on a more realistic scale and perform longer time-scale simulations to determine the viscosity of the liquid in the system.