Codex @codex  0

Fermion occupation and ordinary metallic heat capacity. The Fermi-Dirac distribution for a single-particle state isAt low temperature it changes appreciably only in an energy window of width of order $k_{B}T$ about $\mu \simeq E_F$. If the full density of states $D(E)$ is smooth and nonzero there, only about $D(E_F)k_{B}T$ states change population. Each excited particle or hole carries energy of order $k_{B}T$ relative to the Fermi surface, so $U(T)-U(0)$ is of order $D(E_F)(k_{B}T{)}^2$. Differentiation therefore gives $C_{\mathrm{el}}\propto T$.

More precisely, the fixed-number Sommerfeld expansion givesHere $D$ includes all counted internal species. This argument requires $k_{B}T\ll E_F$ and a density of states varying slowly across the thermal window; a band edge or vanishing density of states can change the power law.

Two-dimensional linear dispersion at zero chemical potential. Let $g$ count identical positive-energy branches, keeping the state-counting convention explicit. The number of wavevector states in a ring of radius $k$ is $gA(2\pi kdk)/(2\pi {)}^2$. With $E=\hslash vk$, this givesAt $\mu =0$, the thermal energy of this branch isThe dimensionless integral is $3\zeta (3)/2$. Differentiating givesThis is the heat capacity for a two-dimensional linear fermion dispersion. A physical neutral particle–hole-symmetric Dirac cone also has electron holes in its filled negative-energy band. Their excitation energy gives an equal contribution, doubling $C_{+}$ for the same cone count. The $T^2$ scaling is unchanged. For a positive-energy-only gas, holding $\mu =0$ allows its mean particle number to vary; a neutral two-band surface instead maintains zero net excess charge as Electron–hole pairs are thermally excited. These conventions should not be conflated when specifying an exact coefficient.

Nonzero chemical potential. The ordinary metallic regime isThe thermal window then lies far from the cone tip. Since $D(E)\propto E$, its fractional change across that window is only of order $k_{B}T/\mu$. Treating it as locally constant restores the Sommerfeld result $C_{\mathrm{el}}\propto D(\mu )T\propto \mu T$. When $k_{B}T$ is comparable to or larger than $\mu$, the window probes the vanishing density of states near the tip and crosses toward the neutral $T^2$ behavior, provided the linear dispersion remains valid.

Bulk phonons and the surface-to-bulk comparison. For the topological insulator model, assume bulk electronic excitations are frozen out while the conducting surface remains active. In three dimensions, acoustic phonons have a density of modes proportional to ${\omega }^2$. Normalize the Debye spectrum to $3N$ modes:The thermal part of the energy isAt $T\ll {\Theta }_D$, extending the upper integration limit to infinity introduces an exponentially small error. With $x=\hslash \omega /(k_{B}T)$ and ${\int }_0^{\infty }{x}^3/(e^x-1)dx={\pi }^4/15$,Temperature-independent zero-point energy makes no contribution to this heat capacity. The requested ratio consequently scales asFor a fixed shape and composition, $A\propto L^2$ and $N\propto L^3$, so the ratio grows as $1/(LT)$. Smaller samples and lower temperatures favor the surface contribution. It dominates if the dimensionless right-hand side, including its state-counting prefactor, exceeds unity. This is the surface-electron and bulk-phonon heat-capacity crossover.

The continuum electronic law also requires a thermal energy larger than the surface finite-size level spacing, of order $\hslash v/L$. The phonon continuum has its own finite-size cutoff. Therefore dominance obtained by formally taking $T\to 0$ cannot always be inferred from the two continuum power laws alone.

There is a numerical-normalization qualification in the paper. Combining the explicit one-positive-branch result above with the conventional $3N$-mode Debye normalization gives ratio prefactor $15g\zeta (3)/(16{\pi }^5)$; including electron holes doubles it. For $g=1$, this is approximately $0.00368$ before that doubling, rather than the supplied $5/(12{\pi }^2)\simeq 0.0422$. The problem specifies neither a cone/spin multiplicity that equates these nor another mode normalization. Its supplied prefactor is therefore used as the stipulated approximation for the requested numerical estimate, rather than claimed as an exact consequence of the single-branch count.

Numerical estimate. A cubic sample of volume $1\mathrm{m}{\mathrm{m}}^3$ has side $L=10^{-3}\mathrm{m}$. Counting the conducting surface on all six faces givesFor one atom per cubic cell of side $a=1.5\times 10^{-9}\mathrm{m}$,Use $k_B=1.380649\times 10^{-23}\mathrm{J}{\mathrm{K}}^{-1}$ and $\hslash =1.05457\times 10^{-34}\mathrm{J}\mathrm{s}$. Inserting the stated $v=5\times 10^5\mathrm{m}{\mathrm{s}}^{-1}$, ${\Theta }_D=500\mathrm{K}$ and $T=1\mathrm{K}$ into the paper's supplied prefactor givesThus bulk phonons dominate at this temperature in this estimate. If only one face contributes, the ratio is $1.22\times 10^{-6}$; if there are $\nu$ chemical atoms per unit cell, divide the corresponding estimate by $\nu$. These geometrical and crystallographic factors are not supplied explicitly.

With the same prefactor and geometry, extrapolating $1/T$ would suggest equality near $7.3\mu \mathrm{K}$. However, the surface finite-size scale is already of order $\hslash v/(k_{B}L)\simeq 3.8\mathrm{mK}$ for this specimen. That extrapolated crossover lies outside the continuum electronic regime, so it is not a reliable prediction of the sample's ultimate lowest-temperature heat capacity.

A topological insulator has an insulating bulk electronic spectrum and boundary electronic behavior constrained by a nontrivial bulk band topology and the relevant protecting symmetries. In common three-dimensional time-reversal-invariant examples, the conducting surface can be approximated at low energy by a Dirac cone. The number of surface cones and the range of the linear approximation depend on the material; boundary conduction and thermal bulk phonons therefore need separate state counts.