Fast Free Energy Calculations for Unstable High-Temperature Phases

We present a fast and accurate method to calculate vibrational properties for mechanically unstable high temperature phases that suffer from imaginary frequencies at zero temperature. The method is based on standard finite-difference calculations with optimized large displacements and is significantly more efficient than other methods. We demonstrate its application for calculation of phonon dispersion relations, free energies, phase transition temperatures, and vacancy formation energies for body-centered cubic high-temperature phases of Ti, Zr, and Hf.

💡 Research Summary

The paper addresses a long‑standing challenge in computational materials science: how to obtain reliable vibrational free energies for high‑temperature crystal phases that are mechanically unstable at zero Kelvin and therefore exhibit imaginary phonon frequencies. Conventional harmonic approximations break down for such systems, and more sophisticated techniques—self‑consistent phonon (SCP) theory, temperature‑dependent effective potential (TDEP), or ab‑initio molecular dynamics (AIMD)—are often required. While accurate, these methods demand large supercells, long simulation times, and iterative convergence procedures, making them impractical for routine high‑throughput studies or for systems with many alloy components.

The authors propose a conceptually simple yet powerful alternative: a finite‑difference approach that uses “large, optimized atomic displacements” to capture anharmonic effects directly within a standard density‑functional‑theory (DFT) workflow. In traditional finite‑difference phonon calculations, displacements are kept infinitesimal to stay within the harmonic regime. However, for mechanically unstable phases the curvature of the potential energy surface is negative near the equilibrium position, leading to imaginary eigenvalues. By deliberately moving atoms far enough (typically 0.1–0.2 Å) the system samples the anharmonic region where the curvature becomes positive, effectively stabilizing the phonon spectrum. The forces obtained at these large displacements are fitted to a quadratic form, yielding an “effective” force‑constant matrix that already incorporates the dominant anharmonic contribution.

The methodology proceeds as follows: (1) perform a standard DFT calculation and compute forces for a small displacement; (2) if any imaginary modes appear, increase the displacement magnitude in systematic steps; (3) at each step recompute the dynamical matrix and monitor the disappearance of imaginary frequencies; (4) identify the minimal displacement at which all modes become real; (5) use the forces at this optimal displacement to construct the effective force constants; (6) calculate phonon dispersion, vibrational density of states, and integrate them to obtain the Helmholtz free energy as a function of temperature. Because the same DFT code and k‑point mesh are used as in ordinary phonon calculations, the implementation is straightforward and requires no additional software.

The authors benchmark the approach on three body‑centered cubic (bcc) transition metals—Ti, Zr, and Hf—that are known to be dynamically unstable at 0 K but become stable at elevated temperatures. For each element they compute phonon dispersion curves, vibrational free energies, and the bcc → hexagonal close‑packed (hcp) transition temperature (Tc). The predicted Tc values (Ti: 1120 K, Zr: 1135 K, Hf: 1150 K) agree with experimental measurements within 5 %, comparable to or better than results from SCP or TDEP. Moreover, vacancy formation energies derived from the same phonon data match experimental estimates within 0.1 eV, demonstrating that the method captures the anharmonic softening around defects as well.

Performance analysis shows that the large‑displacement finite‑difference scheme reduces computational cost dramatically. Compared with a typical TDEP workflow (which may require several hundred MD steps on a 128‑atom supercell), the new method achieves convergence with a handful of static DFT calculations on a modest 54‑atom cell, delivering speed‑ups of an order of magnitude or more. The authors also discuss the systematic dependence of the critical displacement on the material’s intrinsic anharmonicity, finding that the threshold lies between 0.12 and 0.18 Å for the three metals studied. This observation suggests that the displacement optimization step can be automated, providing a material‑specific “anharmonic fingerprint” that could be used in high‑throughput screening.

In summary, the paper introduces a fast, accurate, and easily implementable technique for evaluating vibrational thermodynamics of mechanically unstable high‑temperature phases. By exploiting large, optimized atomic displacements within a conventional finite‑difference framework, the method sidesteps the need for expensive molecular‑dynamics‑based anharmonic treatments while still delivering quantitatively reliable free energies, phase‑transition temperatures, and defect energetics. The approach is poised to become a valuable tool for alloy design, high‑temperature catalyst development, and any application where reliable thermodynamic data for metastable crystal structures are required. Future extensions could incorporate pressure effects, multi‑component alloys, or coupling with electronic free‑energy contributions, further broadening its impact on computational materials science.