Evaluation of the accuracy of the free-energy-minimization method

We have made a detailed comparison between three competing methods for determining the free energies of solids and their defects: the thermodynamic integration of Monte Carlo (TIMC) data, the quasiharmonic (QH) model, and the free-energy-minimization (FEM) method. The accuracy of these methods decreases from the TIMC to QH to FEM method, while the computational efficiency improves in that order. All three methods yield perfect crystal lattice parameters and free energies at finite temperatures which are in good agreement for three different Cu interatomic potentials [embedded atom method (EAM), Morse and Lennard-Jones]. The FEM error (relative to the TIMC) in the (001) surface free energy and in the vacancy formation energy were found to be much larger for the EAM potential than for the other two potentials. Part of the errors in the FEM determination of the free energies are associated with anharmonicities in the interatomic potentials, with the remainder attributed to decoupling of the atomic vibrations. The anharmonicity of the EAM potential was found to be unphysically large compared with experimental vacancy formation entropy determinations. Based upon these results, we show that the FEM method provides a reasonable compromise between accuracy and computational demands. However, the accuracy of this approach is sensitive to the choice of interatomic potential and the nature of the defect to which it is being applied. The accuracy of the FEM is best in high-symmetry environments (perfect crystal, high-symmetry defects, etc.) and when used to describe materials where the anharmonicity is not too large.