EUROPEAN CONGRESS AND EXHIBITION ON ADVANCED MATERIALS AND PROCESSES - EUROMAT 2021

Recent developments in machine learning techniques has immensely benefited ab initio modeling of materials. Interatomic potentials such as the moment tensor potential (MTP) (Shapeev, 2016) that are trained to high temperature density-functional theory (DFT) data are able to predict energies and forces of atomic configurations highly accurately. They are thus able to statistically sample a much wider part of the phase space in a fast and efficient manner. In combination with a systematic thermodynamic integration method (Two-Stage Upsampled Thermodynamic Integration using Langevin Dynamics, Duff et al., 2015), they can be used to calculate total free energies of even complicated systems such as high entropy alloys (HEAs) to 1 meV accuracy (Grabowski et al., 2019, Ferrari et al., 2020) up to the melting point. Apart from static and electronic energies, this also includes vibrational contributions including anharmonicity which significantly affect thermodynamic properties such as specific heat capacity and bulk modulus at high temperatures.

Here, we demonstrate these results for a bunch of refractory BCC systems ranging from single- to five-component alloys and break-down the total free energies to individual contributions. Interestingly, certain BCC unaries have a small positive anharmonic contribution to the total free energy (beyond quasi-harmonic) whereas the other set of unaries have a large negative anharmonicity, which is also reflected in the alloys that constitute them. This is in contrast to the behavior of FCC elements where there is always an increasing positive anharmonic Gibbs energy contribution with temperature arising from anharmonic local pairwise interactions (Glensk et al., 2015). We narrow this feature down to the density of states (DOS) and the first- and second- neighbor forces and illustrate a difference in bonding behavior between the two sets of BCC elements.