Atomistic structure and anomalous heat capacity of low-density liquid carbon: Molecular dynamics study with machine learning potential

Abstract

Liquid carbon remains the source of several unsolved questions related to its structure and region of thermodynamic stability. Experiments demonstrate a drastic decrease in the density for liquid carbon along the graphite melting line in the pressure range P = 1–3 GPa and the nature of this phenomenon is unclear. A recent experimental study [A.M. Kondratyev and A.D. Rakhel, PRL (2019)] revealed another peculiar and still unexplained feature of the liquid carbon – its excessive heat capacity. Using classical molecular dynamics with machine learning potential GAP-20, we study the structural properties of liquid carbon and demonstrate that at P < 1–2 GPa it resembles a net of sp-hybridized chains, rather than a typical covalent liquid, with nanoscale porosity of this phase being responsible for the density decrease. We also show that excessive heat capacity could be a direct manifestation of a smooth transition from a high-density sp2-hybridized phase into a low-density sp-hybridized.