Boltzmann Entropy and the Density of States

Carved on Ludwig Boltzmann's tombstone in Vienna is a single equation: $S = k\log W$ . It is the deepest statement in all of thermal physics — that the macroscopic entropy of a body is nothing more than the logarithm of the number of microscopic arrangements consistent with its macroscopic state. This article is about turning that slogan into a working, calculable tool, and about a subtle and beautiful fact that makes it robust: the entropy you compute does not depend on the fussy details of how you count. The counting object is the density of states $g(E)$ — the number of microstates per unit energy interval. To get a definite count you seem to need a definite energy window, a shell of thickness $\delta E$ : the number of accessible states is $g(E)\,\delta E$ , and the entropy is $S = k_B\ln[g(E)\,\delta E]$ . But which $\delta E$ ? Should you count the thin surface shell at energy $E$ , or the entire enclosed phase volume $\Phi(E)$ of all states with energy up to $E$ ? These sound like different counts — and they are, by astronomical factors. Yet, remarkably, they give the same entropy to leading order. The reason is the tyranny of large numbers. For a macroscopic system the density of states rises so violently with energy — like $E^{3N/2}$ for an ideal gas, with $N \sim 10^{23}$ — that essentially all the enclosed phase volume sits in an infinitesimally thin skin at the surface. The volume of the ball, the area of its surface times any reasonable thickness, and the cumulative count up to $E$ are all dominated by the same exponentially large factor. Their logarithms differ only by terms of order $\ln N$ , which is utterly negligible against the leading term of order $N$ . The choice of $\delta E$ , and even the choice of surface-versus-volume counting, washes out. This is not a mere technical convenience; it is why the microcanonical entropy is well-defined at all. It means the Boltzmann relation gives an unambiguous entropy despite the apparent arbitrariness in setting up the count, and it is the same reason the microcanonical, canonical, and grand canonical ensembles all agree in the thermodynamic limit. Understanding the density of states, and the surface-equals-volume equivalence, is understanding why $S = k_B\ln W$ is a law and not just a definition.

Key takeaways

Boltzmann's relation $S = k_B\ln\Omega$ makes entropy the logarithm of the number of accessible microstates; the count comes from the density of states $g(E)$ .
The count can be done three ways — the surface shell $g(E)\,\delta E$ , the enclosed volume $\Phi(E)$ , or the cumulative count — that differ by astronomical factors.
Remarkably, all three give the same entropy to leading order in $N$ : they differ only by $O(\ln N)$ , negligible against the leading term $O(N)$ .
The reason is that the density of states rises so steeply ( $\propto E^{3N/2-1}$ for an ideal gas) that nearly all the phase volume lies in a thin surface skin.
Consequently the entropy is independent of the shell thickness $\delta E$ and of the surface-vs-volume choice — which is why $S = k_B\ln W$ is well-defined.
Key idea: because the density of states grows like $E^{3N/2}$ , the surface, volume, and cumulative counts all give the same entropy to $O(\ln N/N)$ , making the Boltzmann entropy unambiguous.

The core insight

The Boltzmann entropy $S = k_B\ln\Omega$ is computed from the density of states $g(E)$ . Because $g(E) \propto E^{3N/2-1}$ rises so steeply, the surface count $g(E)\,\delta E$ , the enclosed volume $\Phi(E)$ , and the cumulative count all coincide to $O(\ln N/N)$ — so the entropy is independent of the shell thickness $\delta E$ and of the counting convention, which is what makes it well-defined.

Where this lives today

The density of states is arguably the single most-used quantity in condensed-matter physics and materials science. In solid-state physics, the electronic density of states $g(\epsilon)$ — computed from band structure — determines a material's electrical, thermal, magnetic, and optical properties: the electronic heat capacity, the Pauli paramagnetic susceptibility, the tunneling conductance (which directly images $g(\epsilon)$ in scanning tunneling spectroscopy), and whether a material is a metal, insulator, or semiconductor (set by whether $g$ is nonzero at the Fermi level). Density-functional-theory codes, the workhorse of computational materials discovery, output the density of states as their central result. In photonics and nanotechnology, the local density of states governs how fast an atom emits light (the Purcell effect) and is engineered in photonic crystals and optical cavities to speed up or suppress spontaneous emission. In astrophysics, the density of states of degenerate matter sets the equation of state of white dwarfs and neutron stars, and hence the Chandrasekhar mass limit. And in the foundations of the subject, the surface-versus-volume entropy equivalence proven here — the fact that essentially all of a high-dimensional volume lives in its skin — is the geometric reason the microcanonical, canonical, and grand canonical ensembles all agree in the thermodynamic limit, the consistency that makes statistical mechanics a single coherent theory rather than three competing ones.

The physical picture

The Boltzmann relation $S = k_B\ln\Omega$ demands a count: how many microstates $\Omega$ are accessible to the system at energy $E$ ? For a classical system, "counting" microstates means measuring a phase-space volume (in units of $h^{3N}$ ). But an isolated system has a precise energy, which lives on a surface of zero thickness — and a surface has zero volume. So we count the states in a thin shell $[E, E+\delta E]$ : $\Omega = g(E)\,\delta E$ , where $g(E)$ is the density of states. This immediately raises a worry. The answer seems to depend on $\delta E$ , which we chose arbitrarily. And there is a second ambiguity: instead of the thin surface shell, why not count the entire enclosed phase volume $\Phi(E)$ — all states with energy up to $E$ ? For a $3N$ -dimensional system these are wildly different numbers. The surface and the volume of a high-dimensional ball differ by huge factors. How can entropy be well-defined?

Figure

Three ways to count: cumulative volume Φ(E), surface shell g(E)δE, and the density g(E)

The resolution is a striking feature of high-dimensional geometry: for a $3N$ -dimensional ball, essentially all the volume is concentrated in a vanishingly thin skin at the surface. In three dimensions the outer $1\%$ of the radius holds about $3\%$ of the volume; in $30$ dimensions it holds $26\%$ ; in $10^{23}$ dimensions any finite-thickness skin holds essentially $100\%$ . So the enclosed volume $\Phi(E)$ , the surface shell $g(E)\,\delta E$ , and the cumulative count are all dominated by the same enormous factor. Their logarithms — the entropies — differ only by terms of order $\ln N$ , negligible against the leading $\sim N$ .

Figure

High-dimensional skin: nearly all the phase volume sits at the surface as N grows

The core equations

The Boltzmann relation defines entropy as the logarithm of the accessible count:

S = k_B\ln\Omega, \qquad \Omega = g(E)\,\delta E \qquad (1)

where: $g(E)$ — the density of states (states per unit energy); $\delta E$ — the shell thickness; $k_B$ — Boltzmann's constant. The count is measured in units of $h^{3N}$ to be dimensionless (with a $1/N!$ Gibbs correction for indistinguishable particles). The density of states is the energy derivative of the cumulative phase volume $\Phi(E)$ (the number of states with energy up to $E$ ):

g(E) = \frac{\mathrm{d}\Phi}{\mathrm{d}E} \qquad (2)

where: $\Phi(E)$ — the enclosed phase volume; $g(E)$ — its surface density. For an ideal gas $\Phi(E) \propto V^N E^{3N/2}$ , so $g(E) \propto E^{3N/2 - 1}$ . The three entropies — from the volume, the surface shell, and the cumulative count — agree to leading order:

\ln\Phi(E) \;\approx\; \ln[g(E)\,\delta E] \;\approx\; \frac{3N}{2}\ln E \qquad \big(\text{differ by } O(\ln N)\big) \qquad (3)

where: the leading term is $\sim N$ ; the differences are of order $\ln E$ , $\ln N$ , $\ln\delta E$ — all negligible in the thermodynamic limit. This is the surface–volume equivalence. For a single free particle in 3D, the density of states has a clean, exact prefactor — the building block for quantum gases:

g(\epsilon) = \frac{V}{4\pi^2}\left(\frac{2m}{\hbar^2}\right)^{3/2}\sqrt{\epsilon} \qquad (4)

where: $\epsilon$ — single-particle energy; $V$ — volume; $m$ — particle mass. The $\sqrt{\epsilon}$ dependence (and this prefactor) governs the free-electron gas, the photon gas, and Bose–Einstein condensation. The invariance to shell thickness is quantitative: changing $\delta E$ shifts the entropy only by an additive constant $k_B\ln(\delta E'/\delta E)$ , so

\frac{\Delta S}{S} \sim \frac{\ln N}{N} \to 0 \qquad (5)

where: the relative change in entropy from any reasonable choice of $\delta E$ or of counting convention vanishes as $N \to \infty$ — which is why the microcanonical entropy is well-defined.

Derivation ASurface, volume, and cumulative entropies all agree to O(ln N)

The goal is to prove the article's central structural result: that the three different ways of counting microstates at energy $E$ — the enclosed phase volume $\Phi(E)$ , the surface shell $g(E)\,\delta E$ , and the cumulative count — give entropies that differ only by terms of order $\ln N$ , negligible against the leading term of order $N$ . This is what makes the Boltzmann entropy independent of the arbitrary shell thickness $\delta E$ and of the surface-versus-volume choice, and hence well-defined. We work it out for the ideal gas and verify the cancellation symbolically.

Step 1. The enclosed phase volume is $\Phi(E) \propto E^{3N/2}$ , so the "volume entropy" is $S_\Phi/k_B = \ln\Phi = \tfrac{3N}{2}\ln E + \text{const}$ . This is the leading, order- $N$ term. Why: the cumulative count of all states up to energy $E$ scales as the volume of a $3N$ -dimensional momentum ball, $\propto E^{3N/2}$ .

Step 2. The density of states is $g(E) = \mathrm{d}\Phi/\mathrm{d}E \propto \tfrac{3N}{2}E^{3N/2 - 1}$ , so the "surface entropy" is $S_g/k_B = \ln[g(E)\,\delta E] = (\tfrac{3N}{2} - 1)\ln E + \ln(\tfrac{3N}{2}) + \ln\delta E + \text{const}$ . Why: differentiating $\Phi$ drops the exponent by one and brings down the factor $\tfrac{3N}{2}$ ; multiplying by the shell width $\delta E$ adds $\ln\delta E$ .

Step 3. Subtract: $\ln\Phi - \ln[g(E)\,\delta E] = \ln E - \ln(\tfrac{3N}{2}) - \ln\delta E$ . Every term here is of order $\ln E$ or $\ln N$ — there is no term of order $N$ . The leading order- $N$ pieces ( $\tfrac{3N}{2}\ln E$ ) are identical in both. Why: the difference between surface and volume counting is entirely subleading; the dominant contribution is shared.

Step 4. Therefore $S_\Phi = S_g + O(\ln N)$ , and the relative difference is $O(\ln N/N) \to 0$ . Numerically, at $N = 10^{23}$ the two entropies agree to more than twenty significant figures. The same argument shows the entropy is independent of $\delta E$ : changing $\delta E$ by even a factor of $10^6$ shifts $S/k_B$ by $\ln(10^6) \approx 14$ , negligible against $\tfrac{3N}{2}\ln E \sim 10^{24}$ . Why: this astronomical dominance of the leading term is exactly why $S = k_B\ln W$ is unambiguous despite the apparent arbitrariness of the counting setup. *Verified: $g(E) = \tfrac{3N}{2}E^{3N/2-1}$ , the exact difference $\ln\Phi - \ln[g\,\delta E] = \ln E - \ln(3N/2) - \ln\delta E$ (with no $O(N)$ term), the relative agreement at $N = 10^6, 10^{12}, 10^{23}$ , and the $\delta E$ -independence are all checked in verify_aq-sm-boltzmann-entropy-density-of-states.py.*

Go deeper · Derivation B · 1 minThe single-particle density of states and its √E form

Derivation BThe single-particle density of states and its √E form

The goal is to derive the free-particle density of states $g(\epsilon) = \tfrac{V}{4\pi^2}(2m/\hbar^2)^{3/2}\sqrt\epsilon$ , the fundamental building block for every ideal quantum gas — the free-electron gas in metals, the photon gas, the phonon gas, and Bose–Einstein condensation. The $\sqrt\epsilon$ dependence controls the low-temperature behavior of all of them, so getting the exponent and the prefactor exactly right matters. We count momentum states in a box and differentiate.

Step 1. A particle in a cubic box of side $L$ (volume $V = L^3$ ) has quantized momenta $p = \hbar k$ with $k$ -components spaced by $\pi/L$ ; the number of states with momentum magnitude up to $p$ is the volume of a $k$ -space octant divided by the cell size. The number of states with energy up to $\epsilon = p^2/2m$ is $\mathcal{N}(\epsilon) = \dfrac{V}{6\pi^2}\left(\dfrac{2m\epsilon}{\hbar^2}\right)^{3/2}$ . Why: counting $k$ -space modes in a box gives a factor $V/(2\pi)^3$ per unit $k$ -volume, and the energy constraint is a sphere of radius $k = \sqrt{2m\epsilon}/\hbar$ .

Step 2. The density of states is the energy derivative: $g(\epsilon) = \dfrac{\mathrm{d}\mathcal{N}}{\mathrm{d}\epsilon} = \dfrac{V}{6\pi^2}\left(\dfrac{2m}{\hbar^2}\right)^{3/2}\cdot\dfrac32\sqrt\epsilon = \dfrac{V}{4\pi^2}\left(\dfrac{2m}{\hbar^2}\right)^{3/2}\sqrt\epsilon$ . Why: differentiating $\epsilon^{3/2}$ gives $\tfrac32\sqrt\epsilon$ , and $\tfrac32/(6\pi^2) = 1/(4\pi^2)$ fixes the prefactor.

Step 3. The $\sqrt\epsilon$ form is universal for nonrelativistic particles in 3D and is the reason the electronic heat capacity of a metal is linear in $T$ , the reason the photon gas gives the $T^3$ Debye/Stefan–Boltzmann law (with a modified power for the linear photon dispersion), and the reason BEC occurs in 3D but not in lower dimensions. Why: the low-energy behavior of any ideal quantum gas is set by how $g(\epsilon)$ vanishes as $\epsilon \to 0$ , which the exponent $\tfrac12$ dictates.

Step 4. For $N$ such particles the many-body density of states is a convolution that, for the ideal gas, reproduces $g(E) \propto E^{3N/2 - 1}$ — consistent with the shell count used above. The single-particle $g(\epsilon)$ and the many-body $g(E)$ are the same physics at two scales. Why: the many-body count is built from the single-particle spectrum, so the two must agree, tying this article's two derivations together. *Verified: the single-particle $g(\epsilon) = \tfrac{V}{4\pi^2}(2m/\hbar^2)^{3/2}\sqrt\epsilon$ (from differentiating $\mathcal{N}(\epsilon)$ ) is checked symbolically in the verify script, along with the many-body $g(E) \propto E^{3N/2-1}$ .*

Limiting cases

How the density of states and entropy behave:

Case	Result	Meaning
Ideal gas	$g(E) \propto E^{3N/2-1}$ , $S = k_B\ln[g\,\delta E]$	Steeply rising; regenerates equipartition and $PV = Nk_BT$ .
Single free particle	$g(\epsilon) \propto \sqrt\epsilon$	Building block for quantum gases; sets low- $T$ behavior.
Surface vs. volume	$S_\Phi = S_g + O(\ln N)$	Agree to leading order; entropy well-defined.
Shell thickness $\delta E$	$S$ shifts by $k_B\ln\delta E$ only	Thermodynamically irrelevant ( $\ll N$ ).
Bounded spectrum	$g(E)$ peaks, then falls	Negative temperature possible ( $\partial S/\partial E < 0$ ).
Small $N$	$\ln N$ terms matter	Surface, volume counts genuinely differ; care needed.

Common trap“The entropy depends on the arbitrary shell thickness δE, so it isn't well-defined”

This worry is natural but resolved by the size of the numbers. Yes, the microcanonical count $\Omega = g(E)\,\delta E$ formally contains the arbitrary shell thickness $\delta E$ , and yes, the entropy $S = k_B\ln[g(E)\,\delta E] = k_B\ln g(E) + k_B\ln\delta E$ picks up a $k_B\ln\delta E$ term. But look at the magnitudes: $k_B\ln g(E) \sim k_BN$ (of order $10^{23}k_B$ for a mole), while $k_B\ln\delta E$ is at most of order $k_B\ln E$ — some tens of $k_B$ . Changing $\delta E$ by a factor of a million shifts the entropy by $\sim 14\,k_B$ , an unmeasurably tiny fraction of $10^{23}\,k_B$ . Every derivative of the entropy ( $T$ , $P$ , $\mu$ ) is completely independent of $\delta E$ . So while the entropy has an additive ambiguity of order $\ln N$ , this is thermodynamically invisible, and the Boltzmann entropy is well-defined for all practical and theoretical purposes. The apparent arbitrariness is a feature of the leading-order dominance, not a flaw.

Worked examples

1 · Surface versus volume entropy of an ideal gas (confidence builder)

For an ideal gas with $\Phi(E) \propto E^{3N/2}$ , compute the surface entropy $S_g = k_B\ln[g(E)\,\delta E]$ and the volume entropy $S_\Phi = k_B\ln\Phi(E)$ , and show they agree to $O(\ln N)$ for $N = 10^{6}$ .

Step 1. Volume entropy: $S_\Phi/k_B = \ln\Phi = \tfrac{3N}{2}\ln E + c$ . Take $E = 10^{5}$ (arbitrary units), $N = 10^{6}$ : $\tfrac{3N}{2}\ln E = 1.5\times10^{6}\times\ln(10^5) = 1.5\times10^{6}\times 11.51 = \mathbf{1.727\times10^{7}}$ .

Step 2. Surface entropy: $S_g/k_B = \ln[g\,\delta E] = (\tfrac{3N}{2}-1)\ln E + \ln(\tfrac{3N}{2}) + \ln\delta E + c$ . With $\delta E = 1$ : $= (1.5\times10^{6}-1)(11.51) + \ln(1.5\times10^{6}) = 1.727\times10^{7} - 11.51 + 14.2 = \mathbf{1.727\times10^{7}}$ .

Step 3. Difference: $S_\Phi - S_g = \ln E - \ln(\tfrac{3N}{2}) - \ln\delta E = 11.51 - 14.2 - 0 = -2.7$ (in units of $k_B$ ). Compare to the entropies themselves, $\sim 1.7\times10^{7}$ : the relative difference is $2.7/1.7\times10^{7} = \mathbf{1.6\times10^{-7}}$ .

Step 4. For a real mole ( $N = 10^{23}$ ) the relative difference would be $\sim 10^{-23}$ — the two entropies agree to twenty-three significant figures. Surface and volume counting give the same thermodynamic entropy, and neither the choice nor $\delta E$ matters. The Boltzmann entropy is unambiguous.

2 · The density of states of the free-electron gas (exam level)

A copper sample has free-electron density $n = 8.5\times10^{28}\ \mathrm{m^{-3}}$ . Using $g(\epsilon) = \tfrac{V}{4\pi^2}(2m/\hbar^2)^{3/2}\sqrt\epsilon$ (times 2 for spin), find the Fermi energy from $\int_0^{\epsilon_F} g\,\mathrm{d}\epsilon = N$ .

Step 1. Including the factor of $2$ for spin, the cumulative count is $\mathcal{N}(\epsilon_F) = 2\times\tfrac{V}{6\pi^2}(2m/\hbar^2)^{3/2}\epsilon_F^{3/2} = \tfrac{V}{3\pi^2}(2m/\hbar^2)^{3/2}\epsilon_F^{3/2} = N$ . Solve for $\epsilon_F$ : $\epsilon_F = \dfrac{\hbar^2}{2m}(3\pi^2 n)^{2/3}$ .

Step 2. Compute $3\pi^2 n = 3(9.87)(8.5\times10^{28}) = 2.52\times10^{30}\ \mathrm{m^{-3}}$ , so $(3\pi^2 n)^{2/3} = (2.52\times10^{30})^{2/3} = 1.85\times10^{20}\ \mathrm{m^{-2}}$ .

Step 3. Fermi energy: $\epsilon_F = \dfrac{(1.055\times10^{-34})^2}{2(9.11\times10^{-31})}(1.85\times10^{20}) = (6.11\times10^{-39})(1.85\times10^{20}) = 1.13\times10^{-18}\ \mathrm{J} = \mathbf{7.06\ eV}$ (here $\hbar^2/2m = 1.11\times10^{-68}/1.82\times10^{-30} = 6.11\times10^{-39}\ \mathrm{J\,m^2}$ ).

Step 4. This matches the measured Fermi energy of copper ( $\approx 7.0\ \mathrm{eV}$ ) essentially exactly — the free-electron model, resting entirely on the $\sqrt\epsilon$ density of states derived here, nails the Fermi energy of a real metal. The same $g(\epsilon)$ governs the electronic heat capacity ( $\propto T$ ) and the Pauli paramagnetism of the metal. The density of states is the workhorse of solid-state statistical mechanics.

Origins · 1 minstatistical weight and how this was discovered

statistical weight

concept · the number of accessible microstates Ω

The statistical weight $\Omega$ — the number of microstates compatible with a macrostate — is the central quantity of the Boltzmann relation $S = k_B\ln\Omega$ , sometimes written $W$ (from the German Wahrscheinlichkeit, probability) as on Boltzmann's tombstone. The relation was Boltzmann's supreme achievement in the 1870s: identifying the thermodynamic entropy, previously a purely macroscopic quantity from Clausius, with a count of microscopic arrangements. The constant $k_B$ and the specific form $S = k\log W$ were actually first written by Max Planck in his 1900 blackbody work; Boltzmann expressed the proportionality but not in that notation with that constant. The statistical weight bridges the microscopic and macroscopic worlds: for the density of states $g(E)$ it is $\Omega = g(E)\,\delta E$ , and the astonishing insensitivity of $\ln\Omega$ to the details of the counting (surface versus volume, shell thickness) is what elevates the Boltzmann relation from a definition to a robust physical law. Boltzmann's statistical interpretation of entropy, fiercely resisted in his lifetime, is now the foundation of the entire subject.

History

1865

— Clausius defines the macroscopic entropy and coins the word "entropy."

1872

— Boltzmann's H-theorem connects a molecular quantity to entropy increase.

1877

— Boltzmann relates entropy to the number of microscopic complexions, S ∝ ln W.

1900

— Planck writes S = k log W with the constant k, the tombstone form, in his blackbody analysis.

1902

— Gibbs formalizes the density of states and the ensembles in statistical mechanics.

1911

— Sackur and Tetrode compute the absolute ideal-gas entropy using the density of states and h.

Going further

References. Pathria & Beale, Statistical Mechanics, 3rd ed., §1.4, §2.4 (density of states, surface vs. volume) · Reif, Fundamentals of Statistical and Thermal Physics, §2.5 · Kardar, Statistical Physics of Particles, ch. 4 · Ashcroft & Mermin, Solid State Physics, ch. 2 (electronic density of states) · Sethna, Statistical Mechanics, ch. 3 · Callen, Thermodynamics, ch. 15.

Provenance. The Boltzmann relation $S = k_B\ln\Omega$ , the density of states $g(E) = \mathrm{d}\Phi/\mathrm{d}E$ , and the surface–volume equivalence are standard (Pathria & Beale §1.4, §2.4; Reif §2.5). A verifier should confirm $g(E) \propto E^{3N/2-1}$ , that $\ln\Phi - \ln[g\,\delta E] = O(\ln N)$ (no $O(N)$ term), and the single-particle $g(\epsilon) = \tfrac{V}{4\pi^2}(2m/\hbar^2)^{3/2}\sqrt\epsilon$ . Numerical instances: surface/volume relative entropy difference $1.6\times10^{-7}$ at $N = 10^6$ (→ $0$ ); copper Fermi energy $\epsilon_F = \tfrac{\hbar^2}{2m}(3\pi^2 n)^{2/3} = 7.06\ \mathrm{eV}$ at $n = 8.5\times10^{28}\ \mathrm{m^{-3}}$ (measured $7.0\ \mathrm{eV}$ ). All in verify_aq-sm-boltzmann-entropy-density-of-states.py. $k_B = 1.381\times10^{-23}\ \mathrm{J/K}$ , $\hbar = 1.055\times10^{-34}\ \mathrm{J\,s}$ .

Builds on: The Microcanonical Ensemble and the Postulate of Equal a Priori Probabilities · Second Law of Thermodynamics and Entropy · Temperature, Pressure, and Chemical Potential as Entropy Derivatives · Density of States. Leads to: Phase-Space Volume and the Liouville Measure · Equilibrium Conditions Between Subsystems · Canonical Ensemble.

Why do we need a shell thickness δE at all — can't we just count states exactly at energy E?

Because for a classical system the states of exactly one energy form a surface of zero thickness in phase space, and a surface has zero volume — so counting "exactly at $E$ " gives zero classical microstates. (For a quantum system with a discrete spectrum you could in principle count the degeneracy at exactly $E$ , but for a macroscopic system the levels are so dense that you again need a window.) The resolution is to count states in a thin shell $[E, E+\delta E]$ , which has nonzero volume $g(E)\,\delta E$ . The physical justification is that no real system has perfectly sharp energy anyway — there is always some tiny uncertainty from imperfect isolation — and, as this article shows, the thermodynamic entropy is completely insensitive to the exact $\delta E$ chosen. So $\delta E$ is a mathematical necessity for classical counting that turns out to be thermodynamically invisible.

Is the density of states the same thing as the number of states?

No — they are related but distinct. The number of states (the statistical weight $\Omega$ ) is a pure count: how many microstates are accessible. The density of states $g(E)$ is a count per unit energy: how many states lie in a small energy window around $E$ , so $g(E) = \mathrm{d}\Phi/\mathrm{d}E$ where $\Phi(E)$ is the cumulative number of states up to $E$ . To get an actual number from $g(E)$ you multiply by an energy interval: $\Omega = g(E)\,\delta E$ . The distinction matters because $g(E)$ has units of (states/energy) and appears whenever you convert a sum over states into an integral over energy — $\sum_{\text{states}} f \to \int f(E)\,g(E)\,\mathrm{d}E$ — which is one of the most common operations in statistical mechanics. The single-particle $g(\epsilon) \propto \sqrt\epsilon$ is the version that governs quantum gases.

If the surface and volume entropies agree, why bother distinguishing them?

For macroscopic systems you often don't — they are interchangeable, and textbooks freely use whichever is more convenient (the cumulative $\Phi(E)$ is often easier to compute, the surface $g(E)$ more physical). But the distinction matters in three situations. First, for small systems (few particles, nanoscale, or a few spins), the $O(\ln N)$ differences are no longer negligible and you must be careful which you mean. Second, when deriving results rigorously you need to know the corrections are subleading, which requires understanding the difference. Third, the equivalence itself is a deep and useful fact — it is the geometric reason (all volume in the skin) that the different ensembles agree, and it explains why the arbitrary $\delta E$ drops out. So distinguishing them is partly bookkeeping for small systems and partly conceptual insight into why the whole framework is consistent.

How does the √ε density of states lead to the linear-in-T electronic heat capacity of metals?

It comes from combining $g(\epsilon) \propto \sqrt\epsilon$ with the Pauli exclusion principle. In a metal at low temperature, only electrons within about $k_BT$ of the Fermi energy $\epsilon_F$ can be thermally excited (the rest are Pauli-blocked). The number of such "active" electrons is roughly $g(\epsilon_F)\,k_BT$ , and each gains about $k_BT$ of energy, so the thermal energy is $\sim g(\epsilon_F)(k_BT)^2$ and the heat capacity $C = \partial U/\partial T \sim g(\epsilon_F)k_B^2 T$ — linear in $T$ . The whole result hinges on $g(\epsilon_F)$ being finite and nonzero, which the $\sqrt\epsilon$ form guarantees for a metal. This linear term dominates the specific heat at very low temperature (below the phonon $T^3$ contribution) and its measured slope directly gives the density of states at the Fermi level — one of the most important experimental probes in solid-state physics, and a direct application of the $g(\epsilon)$ derived here.

Does this surface-volume argument work for systems that aren't ideal gases?

Yes — the argument is general, requiring only that the density of states rise steeply with energy, which it does for essentially any macroscopic system with many degrees of freedom. The key ingredient is that $\ln\Omega \sim N$ is extensive (proportional to system size) while the corrections from surface-versus-volume counting or shell thickness are at most $O(\ln N)$ . This holds for interacting gases, liquids, solids, spin systems, and quantum systems alike, as long as $N$ is large and the interactions are short-range (so entropy is extensive). It can fail for systems where entropy is not extensive — self-gravitating systems, small clusters, systems at a phase transition where fluctuations become macroscopic — and there the different counts and ensembles can genuinely disagree. But for the vast majority of ordinary matter, the surface-volume equivalence and the well-definedness of the Boltzmann entropy are completely robust, not special to the ideal gas.

What is the 1/N! factor doing in the count, and does it affect the density of states?

The $1/N!$ is the Gibbs indistinguishability correction, and it fixes a genuine paradox rather than being a cosmetic factor. Classically, if you counted every permutation of identical particles as a distinct microstate, the entropy would come out non-extensive — mixing two samples of the same gas would appear to increase entropy (the Gibbs paradox), which is absurd. Dividing the count by $N!$ (the number of permutations of $N$ identical particles) restores extensivity and gives the correct Sackur–Tetrode entropy. As for the density of states: the $1/N!$ multiplies the whole count $\Omega$ , so it shifts $\ln\Omega$ by $-\ln N! \approx -N\ln N + N$ (Stirling) — this is an $O(N)$ term, so it genuinely matters for the absolute entropy and is what makes $S$ extensive. But it does not change the energy dependence $g(E) \propto E^{3N/2-1}$ , so it drops out of the temperature $1/T = \partial S/\partial E$ and every energy derivative. It matters for the volume/particle-number dependence (extensivity, chemical potential) but not for the shape of $g(E)$ in energy. This deep point is developed in the Gibbs-paradox and indistinguishability articles.