Failure of Equipartition and the Classical-Quantum Boundary

Classical equipartition makes a bold, testable claim: every quadratic degree of freedom carries $\tfrac12 k_BT$ at every temperature, so heat capacities should be constant. They are not. The heat capacity of hydrogen falls from $\tfrac72 R$ toward $\tfrac52 R$ and then $\tfrac32 R$ as it is cooled; the heat capacity of every solid falls toward zero as $T\to0$ rather than sitting at the Dulong-Petit value $3R$ . This systematic, temperature-dependent failure of equipartition — the "specific-heat problem" — was one of the sharpest empirical signals that classical statistical mechanics was incomplete, and its resolution launched the quantum theory of matter. The cause is energy quantization. A classical mode can absorb an arbitrarily small amount of energy, so it is always "on." A quantum mode has a minimum excitation quantum $\hbar\omega$ (or a rotational spacing $\hbar^2/2I$ ), and if $k_BT$ is smaller than that quantum, the Boltzmann factor $e^{-\hbar\omega/k_BT}$ makes excitation exponentially unlikely: the mode sits in its ground state, stores essentially nothing, and contributes nothing to the heat capacity. It is frozen out. As temperature rises past the characteristic temperature $\Theta=\hbar\omega/k_B$ , the mode thaws and equipartition is recovered. A single quantum oscillator makes this quantitative. Its heat capacity, $C(T)=k_B(\Theta/T)^2 e^{\Theta/T}/(e^{\Theta/T}-1)^2$ , rises smoothly from zero (frozen, $C\to0$ exponentially) to $k_B$ (classical equipartition, two quadratic terms) as $T$ crosses $\Theta$ . This is the Einstein model of a solid; the more accurate Debye model, which treats the solid's collective vibrations with a spectrum of frequencies, sharpens the low-temperature behavior to the famous $C_V\propto T^3$ law. Both reduce to Dulong-Petit at high temperature and both vanish at absolute zero, in obedience to the third law that classical equipartition flatly violates.

⚡ Key Takeaways

Equipartition predicts temperature-independent heat capacities; experiments show them falling as $T$ decreases — the mode freeze-out.
A quantum mode of quantum $\hbar\omega$ is frozen when $k_BT<\hbar\omega$ : the Boltzmann factor $e^{-\hbar\omega/k_BT}$ suppresses excitation, and it contributes nothing.
The Einstein oscillator heat capacity $C(T)=k_B(\Theta/T)^2 e^{\Theta/T}/(e^{\Theta/T}-1)^2$ interpolates from $0$ (frozen) to $k_B$ (classical) at $\Theta=\hbar\omega/k_B$ .
Rotation freezes below $\Theta_{\rm rot}$ (light $\mathrm H_2$ : $\approx85\,\mathrm K$ ); vibration below $\Theta_{\rm vib}$ (thousands of K).
The Debye model gives the low- $T$ solid heat capacity $C_V\propto T^3$ ; equipartition (Dulong-Petit $3R$ ) is only the high- $T$ limit.
Key idea: classical equipartition is the high-temperature limit of quantum statistics; a mode contributes its $\tfrac12 k_BT$ only when $k_BT$ exceeds its level spacing, otherwise it freezes out — which is why heat capacities fall to zero at $T\to0$ and satisfy the third law.

💡 The Core Insight

Equipartition holds only when $k_BT\gg\hbar\omega$ for every mode. When $k_BT\lesssim\hbar\omega$ , the quantum of energy is too large to excite thermally, the mode freezes, and its contribution to the heat capacity vanishes exponentially. The characteristic freeze-out temperature is $\Theta=\hbar\omega/k_B$ ; below it the classical prediction fails.

The physical picture

Classical physics lets a mode soak up energy in infinitesimal sips, so the thermal bath always keeps it filled to $\tfrac12 k_BT$ per quadratic slot. Quantum mechanics forbids sips smaller than one quantum $\hbar\omega$ . If the thermal energy $k_BT$ is much less than $\hbar\omega$ , then a full quantum is a rare, expensive purchase — the probability of being excited is $\sim e^{-\hbar\omega/k_BT}$ , exponentially small — and the mode stays in its ground state doing nothing. The bath cannot deposit energy in a slot that will not accept less than a whole quantum. Raise the temperature. As $k_BT$ approaches and then exceeds $\hbar\omega$ , quanta become cheap, the mode fills up, and it starts absorbing energy in proportion to $\mathrm dT$ again — its heat capacity climbs from zero back to the classical value. The transition is smooth and centered on the characteristic temperature $\Theta=\hbar\omega/k_B$ . So the heat capacity of a real substance is not a set of permanent plateaus but a series of risers and treads: each mode contributes a smooth step from $0$ to its classical value as the temperature sweeps up through its $\Theta$ .

Figure

The Einstein oscillator heat capacity C(T)/k_B rises from zero (frozen: k_B T ≪ ħω, exponential suppression) through the freeze-out temperature Θ = ħω/k_B to the classical plateau C = k_B (equipartition). Classical physics sees only the plateau.

The payoff is the low-temperature behavior of a solid. Dulong-Petit says every solid has $C_V=3R$ ; equipartition insists this holds forever. But cool any crystal and $C_V$ collapses toward zero — and it does so in a very specific way, as $T^3$ , captured by Peter Debye's model of the solid's collective vibrations. Plot Dulong-Petit as a flat line at $3R$ and the real (Debye) curve peeling away below it as temperature drops, hitting zero at $T=0$ . That downturn is equipartition failing, and it is exactly what the third law of thermodynamics requires — a law classical equipartition cannot honor.

Figure

Payoff: molar heat capacity of a solid. Dulong-Petit (equipartition) predicts a flat 3R at all T; the real Debye curve falls away below the Debye temperature and vanishes as C_V ∝ T³ at low T — obeying the third law that equipartition violates.

The core equations

The single-mode quantum oscillator is the prototype. Its mean energy is the Planck form, and its heat capacity — the Einstein function — is the temperature derivative. This one expression contains both the classical limit and the freeze-out.

\langle E\rangle=\frac{\hbar\omega}{e^{\hbar\omega/k_BT}-1},\qquad C(T)=k_B\left(\frac{\Theta}{T}\right)^2\frac{e^{\Theta/T}}{(e^{\Theta/T}-1)^2},\quad \Theta=\frac{\hbar\omega}{k_B} \qquad (1)

where: $\langle E\rangle$ — mean energy of a quantum oscillator (excluding zero-point); $\Theta$ — Einstein/characteristic temperature; $C(T)$ — its heat-capacity contribution. Reading: $C\to k_B$ when $T\gg\Theta$ (equipartition of the two quadratic terms), $C\to0$ when $T\ll\Theta$ . The two limits are what matter. At high temperature the mode behaves classically; at low temperature it is exponentially frozen.

T\gg\Theta:\ C\to k_B;\qquad T\ll\Theta:\ C\to k_B\left(\frac{\Theta}{T}\right)^2 e^{-\Theta/T}\to0 \qquad (2)

where: the high- $T$ limit is equipartition (a Taylor expansion of the Einstein function); the low- $T$ limit is dominated by the exponential $e^{-\Theta/T}$ , the freeze-out. Reading: the crossover happens around $T\sim\Theta$ , so $\Theta$ is the single number that sets where a mode's classical description fails. For a solid, the Debye model sums over the phonon spectrum and yields a sharper low-temperature law than the single-oscillator exponential.

C_V\xrightarrow{T\ll\Theta_D}\frac{12\pi^4}{5}Nk_B\left(\frac{T}{\Theta_D}\right)^3 \qquad (3)

where: $\Theta_D$ — Debye temperature; the $T^3$ law replaces the single-oscillator exponential because the solid has a continuum of low-frequency acoustic modes, some always thawed at any $T>0$ . Reading: real solids follow $T^3$ , not the Einstein exponential, at low temperature — but both vanish at $T=0$ .

Derivation AEquipartition as the high-temperature limit of the quantum oscillator, and the freeze-out below

The goal is to derive the Einstein heat-capacity function from the quantized oscillator, show that it reduces to the classical equipartition value $k_B$ when $k_BT\gg\hbar\omega$ , and that it collapses exponentially to zero when $k_BT\ll\hbar\omega$ . This is the calculation that explains why equipartition works at high temperature and fails at low temperature — the entire content of the specific-heat problem in one function.

Step 1. A quantum harmonic oscillator has energy levels $E_n=(n+\tfrac12)\hbar\omega$ . Its partition function is a geometric series, $Z=\sum_{n=0}^\infty e^{-\beta(n+1/2)\hbar\omega}=\dfrac{e^{-\beta\hbar\omega/2}}{1-e^{-\beta\hbar\omega}}$ . The mean energy (dropping the constant zero-point) is $\langle E\rangle=-\partial_\beta\ln Z=\dfrac{\hbar\omega}{e^{\beta\hbar\omega}-1}$ .

Step 2. Differentiate to get the heat capacity. With $x=\beta\hbar\omega=\Theta/T$ , $C=\dfrac{\partial\langle E\rangle}{\partial T}=k_B x^2\dfrac{e^{x}}{(e^{x}-1)^2}$ , equation (1). Why this form: the numerator $e^x$ and denominator $(e^x-1)^2$ come from differentiating the Bose-Einstein-like occupation $1/(e^x-1)$ .

Step 3. High-temperature limit, $x=\Theta/T\to0$ . Expand $e^x\approx1+x+\tfrac{x^2}{2}$ , so $e^x-1\approx x(1+\tfrac x2)$ and $(e^x-1)^2\approx x^2(1+x)$ . Then $C\approx k_B x^2\dfrac{1+x}{x^2(1+x)}=k_B$ . Why equipartition emerges: at high $T$ the discrete levels blur into a continuum and the oscillator's two quadratic terms each earn $\tfrac12 k_BT$ , giving the classical $k_B$ .

Step 4. Low-temperature limit, $x=\Theta/T\to\infty$ . Here $e^x-1\approx e^x$ , so $C\approx k_B x^2\dfrac{e^x}{e^{2x}}=k_B x^2 e^{-x}=k_B(\Theta/T)^2 e^{-\Theta/T}\to0$ . Why the mode freezes: the exponential $e^{-\Theta/T}$ is the Boltzmann probability of exciting even the first quantum; when $k_BT\ll\hbar\omega$ this is vanishingly small and the mode stores nothing. Verified: the Einstein function $C=k_B x^2 e^x/(e^x-1)^2$ , its high- $T$ limit $\to k_B$ , and its low- $T$ limit $\to k_B(\Theta/T)^2 e^{-\Theta/T}$ are confirmed symbolically in verify_aq-sm-failure-of-equipartition.py.

Derivation BWhy hydrogen's room-temperature heat capacity is anomalously low, and the Debye T³ law

The goal is to apply the freeze-out to two celebrated cases: the rotational freeze-out that makes hydrogen's $C_V$ fall below $\tfrac52 R$ on cooling, and the Debye $T^3$ law for the low-temperature heat capacity of a solid. Both show equipartition failing in a concrete, measurable way, and both reveal the third law at work.

Step 1. Hydrogen rotation. The rotational quantum is $\hbar^2/2I$ , giving a characteristic temperature $\Theta_{\rm rot}=\hbar^2/2Ik_B$ . For $\mathrm H_2$ the moment of inertia is tiny (light atoms, short bond), so $\Theta_{\rm rot}\approx85\,\mathrm K$ — unusually high. Why hydrogen is special: for heavier diatomics like $\mathrm N_2$ , $\Theta_{\rm rot}\approx2.9\,\mathrm K$ , so rotation is active far below room temperature; only $\mathrm H_2$ has a rotational freeze-out reaching into an accessible range.

Step 2. Consequently $\mathrm H_2$ 's heat capacity falls from $\tfrac52 R$ (rotation active, $T\gg\Theta_{\rm rot}$ ) toward $\tfrac32 R$ (rotation frozen, $T\ll85\,\mathrm K$ ) as it is cooled below $\sim100\,\mathrm K$ — a direct, historic observation of equipartition failing. Vibration ( $\Theta_{\rm vib}\approx6300\,\mathrm K$ for $\mathrm H_2$ ) is frozen at all ordinary temperatures.

Step 3. Debye solid. Instead of one frequency (Einstein), a solid has a spectrum of acoustic modes with density $g(\omega)\propto\omega^2$ up to a cutoff $\omega_D$ . At low $T$ only modes with $\hbar\omega\lesssim k_BT$ are thawed; counting them, $C_V=\dfrac{12\pi^4}{5}Nk_B(T/\Theta_D)^3$ , equation (3). Why $T^3$ and not exponential: the solid always has some arbitrarily-low-frequency acoustic modes that never fully freeze, so the heat capacity dies as a power law, not an exponential — but it still vanishes at $T=0$ .

Step 4. The Debye constant follows from the integral $\int_0^\infty x^4 e^x/(e^x-1)^2\,\mathrm dx=4\pi^4/15$ ; with the density of states this gives the coefficient $12\pi^4/5\approx233.8$ . Why this matters: both Einstein and Debye send $C_V\to0$ as $T\to0$ , satisfying the third law, whereas classical equipartition's constant $3R$ violates it. Verified: the freeze-out temperatures ( $\Theta_{\rm rot}^{\mathrm H_2}\approx85\,\mathrm K$ order of magnitude), the Debye constant $12\pi^4/5$ from $\int_0^\infty x^4 e^x/(e^x-1)^2\mathrm dx=4\pi^4/15$ , and the $T^3$ scaling are confirmed in the verify script.

Limiting cases

Case	Result	Meaning
$T\gg\Theta$ (all modes)	equipartition, $C=\tfrac f2 R$	Classical limit recovered; Dulong-Petit $3R$ for solids.
$T\ll\Theta_{\rm vib}$	vibration frozen	Diatomic gas is $\tfrac52 R$ , not $\tfrac72 R$ , at room $T$ .
$T\ll\Theta_{\rm rot}$ ( $\mathrm H_2$ )	rotation frozen	$\mathrm H_2$ falls to $\tfrac32 R$ below $\sim85\,\mathrm K$ .
Solid, $T\ll\Theta_D$	$C_V\propto T^3$ (Debye)	Power-law collapse; obeys third law.
Single oscillator, $T\ll\Theta$	$C\propto(\Theta/T)^2 e^{-\Theta/T}$	Exponential freeze-out (Einstein).

⚠️ Common pitfall

Equipartition is a high-temperature limit, not a universal law — the mistake is to apply it at all temperatures. Three specific traps: (1) Do not assume a mode contributes $\tfrac12 k_BT$ if $k_BT$ is below its level spacing; it is frozen and contributes nothing, so the heat capacity is less than the equipartition value at low $T$ . (2) The heat capacity of a solid is not permanently $3R$ (Dulong-Petit); it falls as $T^3$ below the Debye temperature and vanishes at $T=0$ . (3) Classical equipartition violates the third law of thermodynamics ( $S,C\to0$ as $T\to0$ ) — a constant nonzero $C$ would give a divergent entropy $\int C\,\mathrm dT/T$ at low $T$ . The failure of equipartition is not an approximation error; it is a genuine physical effect and a signature of quantization.

Worked examples

Example 1 — When does a vibrational mode thaw? The freeze-out condition for nitrogen (compute the threshold, then the fraction excited). $\mathrm N_2$ has a vibrational wavenumber $\tilde\nu\approx2330\,\mathrm{cm^{-1}}$ , giving $\hbar\omega=hc\tilde\nu$ .

Step 1. Compute $\Theta_{\rm vib}=\hbar\omega/k_B=hc\tilde\nu/k_B$ . With $hc=1.986\times10^{-25}\,\mathrm{J\,m}$ and $\tilde\nu=2.33\times10^5\,\mathrm{m^{-1}}$ : $\hbar\omega=4.63\times10^{-20}\,\mathrm J$ , so $\Theta_{\rm vib}=4.63\times10^{-20}/1.381\times10^{-23}=\mathbf{3350\,K}$ .

Step 2. At room temperature, $T=300\,\mathrm K$ , so $\Theta_{\rm vib}/T=11.2$ . The probability of vibrational excitation is $\sim e^{-\Theta_{\rm vib}/T}=e^{-11.2}=1.4\times10^{-5}$ — about one molecule in seventy thousand is vibrationally excited.

Step 3. Boundary number: the vibrational heat-capacity contribution is $C_{\rm vib}=k_B(\Theta/T)^2 e^{-\Theta/T}=k_B(11.2)^2(1.4\times10^{-5})=1.7\times10^{-3}k_B$ per molecule — utterly negligible. This is why $\mathrm N_2$ 's $C_V$ is $\tfrac52 R=20.8$ , not $\tfrac72 R=29.1$ , at room temperature: the vibration is frozen because $k_BT$ is eleven times smaller than the quantum.

Example 2 — Low-temperature heat capacity of copper: Debye T³ versus Dulong-Petit. Copper has Debye temperature $\Theta_D=343\,\mathrm K$ . Compare its heat capacity at $T=20\,\mathrm K$ (low) predicted by Debye $T^3$ against the classical Dulong-Petit value.

Step 1. Dulong-Petit (equipartition): $C_V=3R=24.94\,\mathrm{J\,mol^{-1}K^{-1}}$ , temperature-independent.

Step 2. Debye $T^3$ : $C_V=\dfrac{12\pi^4}{5}R(T/\Theta_D)^3=233.8\times8.314\times(20/343)^3=233.8\times8.314\times1.98\times10^{-4}=\mathbf{0.385\,J\,mol^{-1}K^{-1}}$ .

Step 3. Boundary check: the Debye prediction is $0.385\,\mathrm{J\,mol^{-1}K^{-1}}$ , about $65$ times smaller than Dulong-Petit — and it matches the measured low-temperature heat capacity of copper. Classical equipartition overpredicts by nearly two orders of magnitude at $20\,\mathrm K$ . Equipartition is not just slightly off at low temperature; it is catastrophically wrong, and only quantization gives the right answer.

peter debye

1884–1966 · 1912 T³ law, 1936 Nobel Prize in Chemistry

Peter Debye extended Einstein's quantum model of specific heat by replacing the single vibrational frequency with a spectrum of collective lattice waves (phonons) up to a cutoff frequency. His 1912 theory produced the celebrated low-temperature law $C_V\propto T^3$ , which fits the measured heat capacities of solids far better than the Einstein exponential and correctly sends $C_V\to0$ as $T\to0$ , in accord with the third law. The characteristic Debye temperature $\Theta_D$ became a standard material parameter. Debye's broader career spanned dipole moments (the debye unit), X-ray and electron diffraction of powders and gases, and electrolyte theory (Debye-Hückel), for which and related work he received the 1936 Nobel Prize in Chemistry. His specific-heat theory stands as one of the earliest triumphs of quantum statistics applied to condensed matter.

⏳ History

1819

— Dulong and Petit find solids have molar heat capacity near

3R

, independent of material.

1876

— Boltzmann's equipartition predicts constant heat capacities, contradicting the low-

T

data.

1907

— Einstein applies quantized oscillators to solids, explaining the fall of

C_V

toward zero.

1911

— Nernst and Lindemann measure heat capacities down to low temperature, confirming the freeze-out.

1912

— Debye derives the

C_V\propto T^3

law from the phonon spectrum.

Where this lives today

The freeze-out of degrees of freedom is now a routine tool rather than a puzzle. Debye temperatures are tabulated for every crystalline solid and used to model lattice thermal properties, phonon transport, and the low-temperature specific heats that reveal electronic contributions (the electronic $\gamma T$ term is separated from the lattice $T^3$ by plotting $C/T$ versus $T^2$ ). The rotational and vibrational partition functions with their characteristic temperatures are the standard inputs to chemical thermodynamics codes that predict reaction enthalpies and equilibrium constants. In cryogenics and quantum-device engineering, the $T^3$ suppression of the lattice heat capacity is what makes it possible to cool systems to millikelvin temperatures. And conceptually, the specific-heat problem remains the cleanest textbook demonstration that classical statistical mechanics is a high-temperature limit of a deeper quantum theory — the very transition that the SM4 quantum-statistics chapter takes up next.

Going further

References. Pathria & Beale, Statistical Mechanics, 3rd ed., §3.7–3.8, §6.4 (heat capacity of gases and solids) · Kardar, Statistical Physics of Particles, Ch. 4 (equipartition) and Ch. 7 (quantum statistics) · Reif, Fundamentals of Statistical and Thermal Physics, §7.7–7.8 and Ch. 10 · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 7 · Landau & Lifshitz, Statistical Physics Part 1, §44–49 (rotation, vibration, Debye theory).

Provenance. All characteristic temperatures are standard textbook values: $\Theta_{\rm rot}^{\mathrm H_2}\approx85\,\mathrm K$ , $\Theta_{\rm vib}^{\mathrm H_2}\approx6300\,\mathrm K$ , $\Theta_{\rm rot}^{\mathrm N_2}\approx2.9\,\mathrm K$ , $\Theta_{\rm vib}^{\mathrm N_2}\approx3340\,\mathrm K$ , Debye temperatures Pb $105\,\mathrm K$ , Cu $343\,\mathrm K$ , diamond $2230\,\mathrm K$ (Kittel/Pathria/Ashcroft-Mermin). A verifier should: (1) re-derive $C(u)=k_B u^2 e^u/(e^u-1)^2$ and confirm its $u\to0$ ( $\to k_B$ ) and $u\to\infty$ ( $\to k_B u^2 e^{-u}$ ) limits; (2) check the Debye low- $T$ constant $12\pi^4/5$ via $\int_0^\infty x^4 e^x/(e^x-1)^2\,\mathrm dx=4\pi^4/15$ ; (3) verify $\tfrac52 R\approx20.8\,\mathrm{J\,mol^{-1}K^{-1}}$ using $R=8.314\,\mathrm{J\,mol^{-1}K^{-1}}$ . Sources: Pathria & Beale ch. 3,6,7; Kardar ch. 4,7; Reif ch. 7,10; Landau-Lifshitz §44–49,§64; Sethna ch. 7. Constants: $k_B=1.380649\times10^{-23}\,\mathrm{J/K}$ , $hc=1.986\times10^{-25}\,\mathrm{J\,m}$ . All checks are in verify_aq-sm-failure-of-equipartition.py.

Builds on: Equipartition Theorem · Generalized Equipartition Theorem · Classical Ideal Gas · Thermal de Broglie Wavelength Leads to: Sackur-Tetrode Equation · Virial Theorem in Statistical Mechanics · Correct Boltzmann Counting

Why does a frozen mode contribute nothing to the heat capacity instead of a little bit?

Because it takes a whole quantum $\hbar\omega$ to excite the mode at all, and when $k_BT\ll\hbar\omega$ the probability of even the first excitation is $\sim e^{-\hbar\omega/k_BT}$ , exponentially tiny. Almost every molecule sits in the ground state, storing no thermal energy, so raising the temperature by $\mathrm dT$ barely changes the mode's energy — its $\partial E/\partial T$ (its heat-capacity contribution) is exponentially suppressed. The mode is not contributing "a little"; its contribution falls faster than any power of $T$ , so for practical purposes it is genuinely off until $T$ approaches $\Theta=\hbar\omega/k_B$ .

Why is hydrogen the textbook example of rotational freeze-out?

Because hydrogen has by far the smallest moment of inertia of common diatomic molecules (two light protons, short bond), and the rotational temperature $\Theta_{\rm rot}=\hbar^2/2Ik_B$ scales inversely with the moment of inertia. For $\mathrm H_2$ , $\Theta_{\rm rot}\approx85\,\mathrm K$ — high enough to be reached by ordinary cooling. For heavier molecules like $\mathrm N_2$ or $\mathrm O_2$ , $\Theta_{\rm rot}$ is only a few kelvin, so rotation stays fully active down to near the liquefaction point and no freeze-out is observed. Hydrogen is thus the one gas where you can watch $C_V$ drop from $\tfrac52 R$ toward $\tfrac32 R$ as rotation switches off.

What is the difference between the Einstein and Debye models?

Both quantize the vibrations of a solid, but Einstein (1907) assumes every atom oscillates independently at a single frequency $\omega$ , while Debye (1912) treats the vibrations as collective sound waves with a whole spectrum of frequencies up to a cutoff $\omega_D$ . At high temperature both give Dulong-Petit $3R$ . At low temperature they differ sharply: Einstein predicts an exponential collapse $C\propto e^{-\Theta/T}$ , but Debye predicts the milder power law $C_V\propto T^3$ , which matches experiment. The reason is that a real solid always has some very-low-frequency acoustic modes that stay thawed at any nonzero $T$ ; the single Einstein frequency misses them.

Does the failure of equipartition mean equipartition is wrong?

No — it means equipartition is a limit, valid at high temperature. Whenever $k_BT$ exceeds the level spacing of every relevant mode, the quantized spectrum blurs into a continuum, each quadratic degree of freedom recovers its $\tfrac12 k_BT$ , and equipartition holds exactly. It is only when some mode's quantum is comparable to or larger than $k_BT$ that the classical result fails. So equipartition remains a correct and useful high-temperature rule; the "failure" is simply the breakdown of its assumption that modes have a continuous energy spectrum. The lesson is to check $k_BT$ against every mode's $\hbar\omega$ before applying it.

Why does classical physics violate the third law of thermodynamics?

The third law requires the entropy (and hence the heat capacity) to go to zero as $T\to0$ . Classical equipartition gives a constant, nonzero heat capacity $C$ at all temperatures. But the entropy is $S(T)=\int_0^T C\,\mathrm dT'/T'$ , and if $C$ is a nonzero constant this integral diverges logarithmically as the lower limit approaches zero — an infinite, unphysical entropy. Quantization saves the day: because every mode freezes out as $T\to0$ , the heat capacity vanishes (as $T^3$ for solids), the integral converges, and $S\to$ a finite constant. The third law is therefore an inherently quantum statement, and equipartition's inability to honor it is a direct signal of its classical limitations.

How does this connect to the ultraviolet catastrophe of blackbody radiation?

It is the same failure, applied to the electromagnetic field. Blackbody radiation is a collection of field oscillators (modes), and there are infinitely many high-frequency ones. Classical equipartition assigns $k_BT$ to each, giving an infinite total energy — the ultraviolet catastrophe. The cure is identical to the specific-heat cure: each mode is quantized with quantum $\hbar\omega$ , so high-frequency modes with $\hbar\omega\gg k_BT$ are frozen out and contribute negligibly. Planck's quantization of the radiation modes and Einstein's quantization of the solid's vibrational modes are the same idea, and both were early, decisive breaks with classical equipartition.