A solid expands when it is heated. That sounds far too obvious, but as long as the springs are ideal springs, a solid never expands at all. To expand, the force between atoms must be “asymmetric between stretching and compression” — that is, anharmonic. Here we squeeze that anharmonicity into a single number, the Grüneisen parameter, and derive the Mie–Grüneisen equation of state from a thermodynamic relation and the free energy of phonons. Then, once we impose the quasiharmonic approximation (QHA), the volumetric thermal expansion coefficient becomes αV = γCV/(BTV) — a surprisingly simple expression. Finally, we look at which factor in this expression actually governs thermal expansion, and why there are materials that shrink when heated (negative thermal expansion).
The volumetric thermal expansion coefficient is how much the volume grows when the temperature is raised while the pressure is held constant.
Here V is the mean atomic volume, that is, the volume of the lattice divided by the number of atoms in it — this way, materials of different composition can be compared on an equal footing.
This definition is awkward to handle as it stands, so we rewrite it using the triple-product relation () as follows.
The first factor is the compressibility κ, which is the reciprocal of the bulk modulus BT. In other words, to evaluate Eq. (1) we only need to calculate how much the internal pressure of the crystal rises when the temperature is raised at fixed volume — nothing more.
Take a pair of atoms and let x be the displacement from the equilibrium position. If the potential is a perfect parabola then the Boltzmann distribution is symmetric in x. No matter how high the temperature is raised,
— the atoms merely vibrate violently, and their mean position does not move by even a millimeter. However many harmonic oscillators you line up, thermal expansion never comes out.
So let us add a cubic term. The side where the atom stretches (x > 0) is easier to climb — conversely, the compressed side rises steeply. This is the asymmetry. However, if we add only the cubic term, then as x is made large the potential falls away and has no bottom (U → −∞). In that case the integral of the Boltzmann factor (the partition function) diverges, and the average itself cannot even be defined. So we add a quartic term that makes a bottom.
The mean position shifts in proportion to the temperature. This is the most elementary picture of thermal expansion. Check it in Fig. 1 — set g = 0, and the mean position does not move however high you raise the temperature. Note, though, that Eq. (3) is a classical approximation that takes the average with the Boltzmann distribution, so it is valid only at high temperatures — quantum mechanically, the energy of an oscillator including its zero-point motion enters in place of kBT, and as T → 0 the shift of the mean position no longer depends on temperature (this corresponds to αV → 0 in Section 7).
Fig. 1. In an anharmonic potential, the mean position shifts. The light shading is the Boltzmann distribution , and the vertical line is its mean . We fix k = 15 eV/Ų and take the stabilizing quartic term to be λ = g²/k (for a reduced mass of 30 u, this spring constant corresponds to about 11 THz — the frequency is not determined by the spring constant alone; the mass is needed). Note also that the shift is very small — thermal expansion is, by nature, a phenomenon of this size.
For a phonon with wave vector q and band ν, we define the mode Grüneisen parameter as follows.
The form on the right is easier to read — it is by what percentage the frequency drops when the volume is increased by 1 %. The minus sign is there so that “it drops when expanded” comes out positive.
The Helmholtz free energy of a crystal with volume V and temperature T can be written as the sum of the static energy at 0 K, E(V), and the phonon part.
The first term is the zero-point motion, and the second term is the part excited by heat. This expression itself follows from the partition function of the harmonic oscillator — I wrote it up in the derivation of the harmonic oscillator and the derivation of the specific heat. The internal energy of one mode is
The quantity in parentheses is “the number of phonons in that mode + 1/2.” Even at absolute zero, the 1/2 remains — the zero-point motion cannot be removed.
The pressure is obtained by differentiating the free energy with respect to the volume: . The second term of Eq. (5) does not contain V explicitly and depends on the volume only through ωqν, so by the chain rule we have
Here is the satisfying part — when we differentiate, exactly Uqν/ωqν from Eq. (6) appears. All that remains is to substitute from Eq. (4).
This is the Mie–Grüneisen equation of state (G. Mie, 1903 / E. Grüneisen, 1912). This is to a solid what PV = NkBT is to an ideal gas.
If we leave the crystal without external pressure (P = 0), it settles at the volume where the right-hand and left-hand sides of Eq. (7) balance. If we approximate the elastic energy at 0 K by the parabola then the amount of expansion itself can be written down.
Since Uqν is always positive, whether the crystal expands or shrinks is determined only by the sign of γqν — but what decides it is the sum. A single mode with negative γqν is not enough; must hold.
Eq. (7) is beautiful, but it cannot be used as it stands. This is because both ωqν and γqν really depend on both temperature and volume. The assumption we make is the quasiharmonic approximation (QHA).
An actual first-principles calculation proceeds in the following steps under this assumption.
Look at how “the valley moves” in Fig. 2. Relative to the valley at 0 K (V0), the valley bottom shifts as the temperature is raised. If you make γ negative, the valley bottom moves to the left — it shrinks even though it was heated; this is negative thermal expansion.
Fig. 2. In the quasiharmonic approximation, the bottom of the free-energy valley moves with temperature. This is Eq. (5) itself for an Einstein model (3 modes per atom, ΘE = 500 K), with . V0 = 12 ų/atom. The two curves are drawn so that each has its own valley bottom at 0 — because what we want to see is not the depth of the valley but the sideways shift of the position of its bottom. We also show that the αV obtained by searching numerically for the valley bottom agrees with the αV obtained from Eq. (11). Note that the “0 K valley” is the valley bottom of the free energy including the zero-point motion, F(V, 0) — it differs from the valley of the static E(V), V0 = 12 ų, by an amount due to the zero-point pressure whenever γ ≠ 0.
Recall that what Eq. (2) requires is — so we differentiate Eq. (7) with respect to temperature at constant volume. Since E(V) does not contain the temperature, it drops out, and we get
In the second line we used the quasiharmonic approximation — at constant volume the frequencies do not depend on temperature, so γqν does not depend on temperature either, and the second term drops out. The remaining first term is nothing other than the specific heat at constant volume of that mode, CV,qν. Finally, we define the total specific heat and the specific-heat-weighted average of the Grüneisen parameter as
Then Eqs. (2) and (9) give the result in a single line.
The thermal expansion of a solid can be written with just four quantities. Moreover, CV, BT, and V are always positive, so it is only γ that determines the sign of αV.
Four quantities appear on the right-hand side of Eq. (11). Which of them matters most? Lining up how much each of them varies among real materials makes the answer clear.
| Factor | Range at room temperature | Fold difference | Role |
|---|---|---|---|
| CV | 0.7–3 kB/atom | about 4× | Levels off at high temperatures (Dulong–Petit). A material with a high ΘD, such as diamond, has not yet leveled off at room temperature |
| γ | −0.06 to 3.3 | Changes sign 37 positive materials: 0.38–3.34, about 9× |
Determines the sign. Set by the crystal structure and the coordination number |
| V | 5.7–71 ų | 12× | Anticorrelated with BT, so it has little effect on its own |
| BT | 2.9–438 GPa | 150× | This one dominates |
The ranges were taken from the 38 cubic materials in Tables S2/S3 of the paper (you can see the same table in tab ② of the simulator). Only BT differs by more than two orders of magnitude. This is why, when αV is plotted against BT as a scatter plot, an inverse-proportion curve αV ∝ 1/BT appears cleanly. To be precise, among these 38 cubic crystals, the factor that most strongly affects the spread in the “magnitude” of positive thermal expansion is BT — the sign is still determined by γ, and, as in ferroelectrics, magnetic transitions, and electronic-state transitions, there are also materials in which something other than lattice vibrations governs thermal expansion.
The αV of diamond (BT = 438 GPa) is 3.3 ppm/K, while that of potassium (BT = 2.9 GPa) is 245 ppm/K — 75 times larger. Hard things do not expand. Soft things expand a lot. This agrees with everyday intuition, and the denominator of Eq. (11) says the same.
In Eq. (11), CV, BT, and V were all positive. Therefore, the only way to get αV < 0 is γ < 0. This is the necessary condition for negative thermal expansion (NTE).
By Eq. (4), γqν < 0 means a mode whose frequency rises when the volume is expanded. This is counterintuitive, but it does happen. And having a few negative modes is not enough — it is only when the sum weighted by the specific heat satisfies that the whole crystal shrinks. The weights change with temperature, and this is why the temperature range of negative thermal expansion is limited even for the same material (Si, which shrinks only at low temperatures, is an example).
The paper systematically investigated 46 elemental solids and 45 binary oxides and showed the following five points.
Y. Mochizuki, H. Koiso, K. Nagamatsu, S. Bae, T. Isobe, A. Nakajima, J. Phys. Chem. C 128, 525–535 (2024) — summary of the conclusions.
In other words, the design guideline for negative thermal expansion is “voids, a small coordination number, and softness.” Voids give the atoms room to swing into, a small coordination number makes the bond angles easy to bend, and softness (a small BT) makes the denominator of Eq. (11) small and amplifies the effect.
| Material | Structure | γ | BT (GPa) | αV (ppm/K) | Remark |
|---|---|---|---|---|---|
| Cu2O | Cuprite type Pn3̄m | −0.061 | 117 | −1.40 | Cu has a coordination number of 2. O–Cu–O swings sideways |
| ReO3 | ReO3-type Pm3̄m | 0.384 | 244 | 3.47 | A perovskite with the A site vacant. The octahedra rotate |
| Si | Diamond type | 0.39 | 90.0 | 7.22 | At low temperatures αV becomes negative |
| MgO | Rock-salt type | 1.63 | 150 | 35.9 | Densely packed, with no voids |
| K | Body-centered cubic | 1.30 | 2.89 | 245 | Just plain soft |
It is especially important that the same composition can have a different γ when the structure changes. The paper compares the rock-salt and zinc-blende types of CaO: the rock-salt type has only positive γqν, whereas the zinc-blende type has negative γqν at the X and L points. γ is not a property of the element but a property of the arrangement.
Let us keep, rather than drop, the term that was dropped in the second line of Eq. (9). Define the thermal Grüneisen parameter δqν and the anharmonic parameter Δqν by
This δqν is a different quantity from the Anderson–Grüneisen parameter δT, which was used in the simulator to extrapolate BT above 595 K — the letter is the same, but the definition and the meaning are different.
With these two, the temperature derivative of γqν can be summarized neatly.
Putting this back into the first line of Eq. (9), the volumetric thermal expansion coefficient is extended as follows. Here we write the averages of the mode sums, weighted by Uqν, as
— δ contains γqν inside it. Setting , we obtain
The first term is the QHA, and the second term is the deviation from it.
For modes of lower frequency, when the temperature is raised, becomes relatively larger (δqν is a quantity divided by ω). Therefore soft materials with many low-frequency modes deviate more from the QHA.
You can work with the actual numbers in the simulator for the equation of state of an ideal solid — in ① (“Heat it and shake the crystal”) raise the temperature while rotating the crystal, in ② (“Stiffness controls expansion”) look at the αV–BT correlation of the 38 materials, and in ③ (“What changes with temperature”) follow the temperature dependence. The V(T), αV(T), BT(T), and CV(T) that appear in ①–③ are not just the 300 K cross-section but the output of the QHA calculation itself, over 0–595 K. The data come from Tables S2/S3 of Y. Mochizuki, H. Koiso, K. Nagamatsu, S. Bae, T. Isobe, A. Nakajima, “Thermal Properties of the Element and Binary Oxides toward Negative Thermal Expansion: A First-Principles Lattice-Dynamics Study,” J. Phys. Chem. C 128, 525–535 (2024) (open access, CC BY 4.0) and from the phonopy-qha output used for that calculation. Related pages: specific heat derivation, harmonic oscillator and second quantization, entropy and free energy.