Skip to content

Finite-temperature bulk moduli from an equation-of-state-based Grüneisen function

FieldValue
JournalPhysical review. B./Physical review. B
Published2026-08-17
AuthorsAnonymous
DOI10.1103/t5cg-vs3s

An EOS-based Grüneisen construction reproduced the leading thermal softening of bulk moduli for diamond, MgO, silicon, and NaCl without fitting to thermal data. At each model’s fitted characteristic temperature, the diamond calculation reproduced 61% to 79% of the experimental softening magnitude. Absolute modulus errors still reflected the static equation of state supplied by the interatomic potential.

The adiabatic and isothermal bulk moduli decreased with increasing temperature for all four solids. The room-temperature softening rates also reproduced the experimental relative ordering. Agreement was generally reasonable, but its quantitative level changed with material, potential, and analytic EOS.

For diamond, d-UMA shifted the absolute (K_S(T)) curve closest to experiment. The fitted characteristic temperatures were 965 K for UMA, 963 K for d-UMA, 937 K for UPET with Vinet, and 947 K for UPET with AP2. At those temperatures, the four models reproduced 76%, 79%, 61%, and 61% of the experimental softening magnitude, respectively.

MgO results improved with d-UMA. UPET with Vinet gave a rate comparable to d-UMA, while AP2 predicted substantially stronger softening.

For silicon, UPET with Vinet matched the experimental slope closely. AP2 gave somewhat weaker softening, although both UPET results reduced the absolute modulus offset.

For NaCl, d-UMA improved the absolute modulus. UPET improved it further, while AP2 produced slightly stronger softening than Vinet.

The new construction replaces constant (t) with (t(V)) in the generalized Burakovsky-Preston EOS form. The lowest-order expansion in powers of (V^{1/3}) contains (t_0) and (t_1). Fixing (t_0=5/2) imposes the infinite-compression limit (\gamma_\infty=1/2). The remaining coefficient (t_1) is fixed by equilibrium anchoring, (\gamma(V_0)=\gamma_0).

Each EOS has its own asymptotic Grüneisen value. Vinet gives (\gamma_\infty^{EOS}=1/6), requiring a constant shift (s=1/3). AP2 gives (\gamma_\infty^{EOS}=1/2), requiring no shift. The shift enforces the target limit but cannot correct an EOS extrapolation beyond its reliable compression range.

Static energy-volume data determine the fitted (V_0), (K_0), and (K_0’). The construction was performed separately with Vinet and AP2 fits, so EOS selection remained an explicit modeling choice.

Elastic anchoring and finite-temperature calculation

Section titled “Elastic anchoring and finite-temperature calculation”

Five near-equilibrium volumes anchor the Grüneisen function. They comprise (V_0), two slightly compressed volumes, and two slightly expanded volumes. At each volume, symmetry-adapted strains were applied to an internally relaxed structure. Opposite-strain pairs were averaged before weighted quadratic fits produced the elastic stiffness tensor.

The Debye temperature came from Christoffel acoustic velocities and a Monte Carlo average over 20,000 propagation directions. Increasing the sample to 32,000 directions changed every Debye temperature by less than 0.02 K. A constrained quadratic fit of (\ln\Theta_D) against volume gave (\gamma_0) at equilibrium. Linear fits gave virtually the same (\gamma_0) values, confirming that the narrow sampling interval was adequate.

The finite-temperature calculation used a Mie-Grüneisen-Debye equation of state with zero-point vibrational energy. At each temperature, the zero-pressure equilibrium volume (V_e) was found by solving (P(V_e,T)=0). The adiabatic and isothermal moduli were then evaluated at that volume.

A Wachtman form was fitted after calculation to interpolate each modulus curve and obtain a smooth derivative. It did not enter the Grüneisen construction. At high temperature, (dK/dT) approaches (-b), where (b) is the fitted high-temperature slope.

The zero-point term shifts the zero-pressure equilibrium volume relative to static (V_0). Thus, (K_S(0)=K_T(0)) does not imply (K_S(0)=K_0).

UMA reproduced the qualitative PBE-level static error pattern, including overestimated equilibrium volumes and underestimated static bulk moduli. UPET used r2SCAN reference data and generally moved the errors toward the origin without reproducing the tabulated r2SCAN trend exactly. The d-UMA variant added DFTD3 with BJ damping and PBE damping parameters.

Relaxations required a maximum atomic force below (5 \times 10^{-4}) eV/Å and a maximum unit-cell stress below (5 \times 10^{-3}) GPa. Fixed-volume relaxations supplied the EOS data, while small elastic strains avoided mechanical instability. The EOS sampling excluded the most extreme compression and dilation regimes where the MLIPs could extrapolate beyond their reliable domains.

Comparison with finite-temperature experiments

Section titled “Comparison with finite-temperature experiments”

Normalized room-temperature isothermal softening exposed fractional changes without allowing static baseline errors to dominate. For NaCl, the experimental room-temperature fit gave (K_0=28.02) GPa and (K_T(\mathrm{RT})=25.89) GPa. The resulting ratio was 0.924.

All four calculated 300 K ratios fell below the experimental NaCl value. A PW91 AIMD result of 0.992 implied much weaker softening than the experimental compressibility and phonon data. The two UPET results were closer to experiment, with Vinet giving the better agreement of the two EOS choices.

For silicon, all four 300 K ratios stayed close to an LDA-QHA calculation. For MgO, the values lay within or very near the range reported by molecular dynamics. Diamond ratios remained close to unity. UMA and d-UMA lay near the PBE-AIMD estimate. The UPET ratios fell between PBE-AIMD and REBO-MD, closer to the former.

The calculated (\gamma) values exceeded representative experimental estimates throughout the sampled zero-pressure thermal paths. UPET came closest for diamond and NaCl, while d-UMA showed the largest deviation for silicon.

The (\gamma) and Debye temperature trends were comparatively robust. The logarithmic derivative (q=d\ln\gamma/d\ln V) was more sensitive to EOS choice. Compared with Vinet, AP2 produced nearly unchanged diamond softening, weaker silicon softening, and stronger MgO and NaCl softening.

The effective equilibrium index was nearly unchanged between Vinet and AP2 for the same UPET input. The coefficient (t_1) changed by about one unit because of the Vinet asymptotic shift. Neither AP2 nor a constant-(t) prescription gave uniformly better agreement. The calculated diamond effective indices lay near the Barton-Stacey value of 2.35. Silicon indices exceeded the conventional 0 to 2.35 range, while MgO gave negative effective indices.

The approach omits explicit anharmonic effects beyond the quasiharmonic Mie-Grüneisen-Debye level. The authors identify intrinsic anharmonicity as a main source of the remaining diamond discrepancy at high temperature. The Debye treatment also reduces direction-dependent acoustic velocities to one average scale. That reduction may lose accuracy for lower-symmetry or strongly anisotropic materials.

The calculated low-temperature and room-temperature reductions in Debye temperature were smaller than several experimental estimates. All four constructions overestimated (\gamma), and no finite-temperature MD benchmark using the same MLIPs was included. A direct comparison would help separate EOS errors from potential errors.

Only bulk moduli were assessed. Extending the treatment to shear modulus, Young’s modulus, and other elastic constants requires a tensorial thermoelastic approach. The method also cannot repair an analytic EOS at extreme compression through a constant asymptotic shift.

The comparison can diagnose deficiencies in thermoelastic transferability and inform universal MLIP training and validation. For engineers assessing diamond components whose stiffness changes with temperature, it offers a screening route before AIMD, quasiharmonic lattice dynamics, or thermodynamic integration. It predicts monotonic softening and weak fractional room-temperature change, but it does not establish a component’s temperature-dependent stiffness by itself. Wide-range stiffness designs still require a finite-temperature calculation with explicit anharmonic physics.

View Original Abstract

An equation-of-state (EOS)-based construction of the Grüneisen function is developed and assessed through predictions of finite-temperature bulk moduli. In this approach, the volume dependence of the Grüneisen function is expressed analytically in terms of static EOS information, anchored by Debye temperatures obtained from elastic data sampled near equilibrium, and constrained by the infinite-compression limit. The resulting form requires as material-specific input only static energy-volume data and near-equilibrium elastic properties, with no parameters adjusted to thermal data. Within a Mie—Grüneisen—Debye framework, the approach is examined for diamond, magnesium oxide, silicon, and sodium chloride, chosen to span a broad range of stiffness, using machine-learning interatomic potentials from the Universal Models for Atoms (UMA) and Universal Point Edge Transformer (UPET) families, together with a dispersion-corrected variant of UMA. The calculated bulk moduli reproduce the expected experimental softening trends and capture the overall scale of the bulk-modulus temperature derivatives, although the level of quantitative agreement depends on the material, the underlying interatomic potential, and the selected analytic EOS form. These results show that an EOS-based construction of the Grüneisen function can capture the leading thermal-softening behavior of bulk moduli without fitting to thermal data. The observed sensitivity to the underlying static description further suggests that the framework may help identify deficiencies relevant to thermoelastic transferability and thereby inform future training and validation strategies for universal machine-learning interatomic potentials.