Accessible Phase Space and Energy Shells

An isolated gas of a trillion trillion molecules would seem to have an unthinkable freedom: each molecule could be anywhere, moving at any speed. Yet the moment you fix the total energy, almost all of that freedom evaporates. Energy conservation pins the system to a single razor-thin surface inside its vast phase space — the constant-energy hypersurface $H(q,p)=E$ — and the entire predictive power of equilibrium statistical mechanics comes from the geometry of that one surface, not from the roomy space around it. This is the accessible phase space: the only region the system can visit, and the only region we ever need to weight. A surprise waits inside that geometry. The energy surface has zero volume in the full space — it is infinitely thin — so we fatten it into a shell of small width $\Delta$ and count the phase-space volume there. You might fear that the arbitrary choice of $\Delta$ contaminates every answer. It does not, and the reason is one of the most beautiful facts in the subject: in high dimensions almost all of a ball's volume hides in an infinitesimally thin skin at its surface. For $N\sim10^{23}$ the shell, the enclosed ball, and the bare surface all have logarithms that agree to one part in $N$ . The thickness of the shell is a formal convenience with no thermodynamic consequences. Master this and the microcanonical ensemble stops being a formula to memorize and becomes a picture you can see. The fundamental postulate spreads uniform probability over this shell; the Boltzmann entropy is $k_B$ times the log of its volume; Liouville's theorem keeps that uniform density stationary; and the ergodic hypothesis is the conjecture that a single trajectory samples this surface fairly. Every downstream result — temperature, pressure, the equivalence of ensembles — is the geometry of a thin shell in a space of enormous dimension.

Key takeaways

The accessible phase space is the set of microstates a system can occupy given its conserved quantities — above all the fixed energy, which confines it to the hypersurface $H(q,p)=E$ .
Since that surface has zero $6N$ -volume and perfect energy resolution is unphysical, one works with a thin shell $E\le H\le E+\Delta$ ; its volume defines the density of states $g(E)=\partial\Sigma/\partial E$ .
For an ideal gas the accessible volume scales as $\Sigma(E)\propto V^N E^{3N/2}$ : the momentum part is the volume of a $3N$ -dimensional ball of radius $\sqrt{2mE}$ .
In high dimensions almost all of a ball's volume sits in a thin skin at the surface — the fraction within a fractional depth $\delta$ is $1-(1-\delta)^n\to1$ — so $\ln\Sigma\approx\ln(\Gamma\Delta)\approx\ln\Omega$ to order $1/N$ .
That equivalence makes the choice of $\Delta$ (and even shell-vs-ball-vs-surface) irrelevant to thermodynamics, and underlies the equivalence of the microcanonical, canonical, and grand-canonical ensembles.
Key idea: an isolated system explores only the constant-energy hypersurface; thickening it into a shell of width $\Delta$ gives a finite accessible volume whose logarithm — independent of $\Delta$ to exponential accuracy — is the Boltzmann entropy and the geometric basis of the microcanonical ensemble.

The core insight

An isolated system is confined by energy conservation to the constant-energy hypersurface $H(q,p)=E$ . Thickening this zero-volume surface into a shell of width $\Delta$ gives a finite accessible phase-space volume whose logarithm — independent of $\Delta$ to exponential accuracy in the thermodynamic limit — is the Boltzmann entropy and the foundation of the microcanonical ensemble.

Where this lives today

The density of states $g(E)$ that this article builds from the accessible shell is one of the most measured quantities in all of physics — far beyond classical gases. In a solid, the electronic density of states at the Fermi energy controls the heat capacity, the electrical and thermal conductivity, and whether the material is a metal, insulator, or superconductor; angle-resolved photoemission and scanning tunneling spectroscopy map it directly. In statistical computation, the Wang–Landau algorithm (2001) is a celebrated Monte Carlo method whose entire purpose is to estimate $g(E)$ across the full energy range of a system, precisely because — as this article shows — the microcanonical entropy $S(E)=k_B\ln g(E)$ contains all the equilibrium thermodynamics. The high-dimensional geometry at the heart of the article — almost all the volume in a thin surface skin — has escaped statistical mechanics entirely and become foundational elsewhere. It is the "concentration of measure" phenomenon that makes high-dimensional statistics and machine learning behave the way they do: random vectors in high dimension are nearly orthogonal, distances concentrate, and a Gaussian in $n$ dimensions lives essentially on a thin sphere of radius $\sqrt n$ . Every time a data scientist is surprised that intuition from three dimensions fails in three hundred, they are meeting the same fact that makes the energy shell equal to the energy ball for a mole of gas.

The physical picture

Picture the full phase space as an enormous room and energy conservation as a single thin, curved wall running through it. The system is forbidden from the empty air on either side of the wall; it can only slide along the wall's surface, forever. All of equilibrium statistical mechanics is the geometry of that one surface — its area, how fast that area grows with energy — and none of it is the geometry of the room. The first mental shift is to stop thinking about the whole $6N$ -dimensional space and start thinking about a single $(6N-1)$ -dimensional sheet. The second shift is stranger and is the heart of the article. In everyday three dimensions, a ball's volume is spread through its interior; the outer skin holds only a little. But volume grows like $r^n$ , so the fraction of an $n$ -ball lying within a fractional depth $\delta$ of the surface is $1-(1-\delta)^n$ , and as $n\to\infty$ this races to $1$ . For a mole, $n=3N\sim10^{24}$ , and even a skin one part in $10^{20}$ deep contains essentially the entire ball. Almost all of a high-dimensional ball is surface. That is why the enclosed volume $\Sigma(E)$ , the thin shell $\Gamma(E)\Delta$ , and the bare surface area $\Omega(E)$ become interchangeable — they differ by factors that vanish against the exponential size of each.

Figure

Almost all of a high-dimensional ball hides in a thin surface skin

The three shaded rings tell the whole story: as the dimension climbs from 3 to 300, the white interior a naive intuition expects to matter shrinks to nothing, and the colored surface skin swallows the volume. This is not a curiosity — it is the reason the microcanonical entropy is robust. The width $\Delta$ of the shell is a formal device, not a physical parameter: because $S=k_B\ln\Gamma$ and $\Gamma\sim E^{3N/2}$ , changing $\Delta$ by any reasonable factor shifts $S$ by at most $k_B\ln(\text{order }N)$ , utterly negligible against $S\sim Nk_B$ .

Figure

Energy fixes the system to a thin shell in phase space

The trajectory in red never leaves the blue shell: energy conservation is a wall it cannot cross. Everything the system does — every microstate it visits — lives in that thin band. Statistical mechanics weights the band uniformly (the fundamental postulate) and asks only one geometric question of it: how does its volume grow with energy?

The core equations

Formally, the accessible set at sharp energy is the level surface of the Hamiltonian — a $(6N-1)$ -dimensional hypersurface carved out of the full phase space by the single equation $H=E$ .

\mathcal{A}(E) = \{(q,p)\in\Gamma : H(q,p)=E\} \qquad (1)

where: $\mathcal{A}(E)$ — the accessible phase space at energy $E$ ; $\Gamma$ — the full $6N$ -dimensional phase space; $H(q,p)$ — the Hamiltonian; $(q,p)$ — a phase point, $3N$ coordinates and $3N$ momenta. This surface has zero $6N$ -volume, which is why we cannot count microstates on it directly and must thicken it into a shell. The workhorse quantity is the phase-space volume enclosed below energy $E$ — the number of microstates with energy at most $E$ , made dimensionless and indistinguishability-corrected.

\Sigma(E) = \frac{1}{h^{3N}N!}\int_{H(q,p)\le E}\mathrm{d}^{3N}q\,\mathrm{d}^{3N}p \qquad (2)

where: $\Sigma(E)$ — the dimensionless count of states with energy $\le E$ ; $h^{3N}$ — Planck's constant per degree of freedom, setting the phase-space cell size; $N!$ — the Gibbs factor for indistinguishable particles. Differentiating $\Sigma$ with respect to energy gives the density of states — the number of microstates per unit energy — from which the thin-shell count follows.

g(E)=\frac{\partial \Sigma}{\partial E},\qquad \Omega(E)\,\Delta = g(E)\,\Delta = \Sigma(E+\Delta)-\Sigma(E) \qquad (3)

where: $g(E)$ — the density of states; $\Omega(E)\Delta$ — the number of microstates in the shell of width $\Delta$ . Read physically: $g(E)$ measures how fast new microstates open up as you add energy, and it is the quantity the Boltzmann entropy is built from. For the concrete case that anchors everything — the classical ideal gas — the momentum integral is the volume of a high-dimensional ball, giving the characteristic $E^{3N/2}$ growth that makes the shell so sharp.

\Sigma(E) = \frac{V^N}{h^{3N}N!}\,\frac{\pi^{3N/2}\,(2mE)^{3N/2}}{\Gamma\!\left(\tfrac{3N}{2}+1\right)} \;\propto\; V^N E^{3N/2} \qquad (4)

where: the configurational integral over $N$ particles in volume $V$ gives $V^N$ ; the momentum integral over $\sum p_i^2\le 2mE$ is the volume of a $3N$ -ball of radius $\sqrt{2mE}$ , namely $\pi^{3N/2}R^{3N}/\Gamma(\tfrac{3N}{2}+1)$ ; and $\Gamma$ is the gamma function. The energy exponent $3N/2$ — one power of $E^{1/2}$ per momentum dimension — is what makes $\Sigma$ grow astronomically and the shell equivalence work. The volume of an $n$ -dimensional ball, the geometric fact behind (4) and behind the shell-concentration phenomenon, is worth stating on its own.

V_n(R)=\frac{\pi^{n/2}\,R^{\,n}}{\Gamma\!\left(\tfrac n2+1\right)} \qquad (5)

where: $V_n(R)$ — the volume of a radius- $R$ ball in $n$ dimensions; the $\Gamma$ function in the denominator is what eventually overwhelms the $\pi^{n/2}$ growth, so the unit ball's volume actually shrinks to zero as $n\to\infty$ . The competition between $R^n$ and this $\Gamma$ -suppression is exactly what concentrates a fixed ball's volume near its surface. Finally, the payoff identity: for a system with many degrees of freedom, the logarithms of the enclosed volume, the shell volume, and the surface area all coincide to relative order $1/N$ .

\ln\Sigma(E)\;\approx\;\ln\big(g(E)\,\Delta\big)\;\approx\;\ln\Omega(E) \qquad \big(\text{to relative order } 1/N\big) \qquad (6)

where: the three quantities differ by factors of order $N$ or $\Delta$ , whose logarithms are $\mathcal{O}(\ln N)$ — negligible against $\ln\Sigma=\mathcal{O}(N)$ . This is why the entropy is unambiguous: it does not matter whether you count the shell, the surface, or the whole enclosed ball, nor how thick you draw the shell.

The consequences, derived

Derivation AThe ideal-gas energy shell grows as E to the 3N over 2

The goal is to compute the accessible phase-space volume of a classical ideal gas and read off why its energy dependence is $E^{3N/2}$ — the single most important scaling in elementary statistical mechanics, from which the equation of state and the equipartition of energy both follow. The strategy is to split the phase-space integral into a trivial configurational part and a momentum part that is exactly the volume of a high-dimensional ball, then invoke the $n$ -ball formula. The subtle point to watch is that each of the $3N$ momentum components contributes one power of $\sqrt E$ , so the half-integer exponent $3N/2$ is not a fudge — it is literally counting momentum dimensions.

Step 1. Separate configuration and momentum. For $H=\sum_i p_i^2/2m$ (no potential), the enclosed integral factorizes:

\Sigma(E)=\frac{1}{h^{3N}N!}\Big(\!\int_V\mathrm{d}^{3N}q\Big)\Big(\!\int_{\sum p_i^2\le 2mE}\mathrm{d}^{3N}p\Big)=\frac{V^N}{h^{3N}N!}\,I_p.

Why: the free-particle Hamiltonian does not couple positions and momenta, so each particle contributes a factor $V$ and the momentum integral stands alone.

Step 2. Recognize the momentum integral as a ball. The condition $\sum_{i=1}^{3N}p_i^2\le 2mE$ is a $3N$ -dimensional ball of radius $R=\sqrt{2mE}$ , so $I_p=V_{3N}(R)$ . Why: fixed total kinetic energy is a sphere in momentum space; "energy $\le E$ " is the solid ball inside it.

Step 3. Apply the $n$ -ball formula. With (5),

I_p=\frac{\pi^{3N/2}\,(2mE)^{3N/2}}{\Gamma\!\left(\tfrac{3N}{2}+1\right)}\;\propto\;E^{3N/2}.

Why: $R^{3N}=(2mE)^{3N/2}$ , and each of the $3N$ momentum axes has contributed one factor $\sqrt{2mE}$ , producing the half-integer power.

Step 4. Read off the density of states. Differentiating, $g(E)=\partial\Sigma/\partial E\propto E^{3N/2-1}$ . So the shell count grows as $E^{3N/2-1}$ — for $N=3$ that is $E^{7/2}$ ; for a mole it is $E^{9\times10^{23}}$ , the astronomical growth that makes the shell effectively the whole ball. Verified: the exponent $3N/2$ and $g\propto E^{3N/2-1}$ are checked symbolically in the companion script, along with $V_n$ for $n=1,2,3$ (giving $2R$ , $\pi R^2$ , $\tfrac43\pi R^3$ ).

Go deeper · Derivation B · 2 minAlmost all the volume is in the shell: surface equals volume in high dimension

Derivation BAlmost all the volume is in the shell: surface equals volume in high dimension

The claim to establish is the geometric heart of the microcanonical ensemble: in high dimension a ball's volume is concentrated in an arbitrarily thin skin at its surface, so the enclosed volume, the shell, and the surface area have logarithms that agree to relative order $1/N$ . This is what makes the arbitrary shell width $\Delta$ physically irrelevant and licenses using whichever of the three is most convenient. The plan is to compute the fraction of an $n$ -ball's volume within a fractional depth $\delta$ of the surface directly from the $R^n$ scaling, then take $n=3N$ enormous — being careful to work in log space, because the naive fraction underflows the moment $\delta$ is tiny and $n$ is a mole.

Step 1. Fraction outside the inner ball. The volume within fractional depth $\delta$ of the surface of a radius- $R$ ball is $V_n(R)-V_n((1-\delta)R)$ , so the fraction there is

f=1-\frac{V_n((1-\delta)R)}{V_n(R)}=1-(1-\delta)^n.

Why: the $\pi^{n/2}/\Gamma$ prefactor cancels in the ratio, leaving only the $R^n$ scaling — all the dimension-dependence is in the exponent $n$ .

Step 2. High dimension drives the skin to hold everything. For fixed $\delta$ , $(1-\delta)^n\to0$ as $n\to\infty$ , so $f\to1$ . Concretely at $\delta=0.1$ : $n=3$ gives $f=0.27$ , $n=30$ gives $0.96$ , $n=300$ gives $0.99997$ — the numbers in the first figure. Why: raising a number less than 1 to a large power sends it to zero; the interior ball becomes a vanishing fraction of the whole.

Step 3. Work in log space for a mole. For $n=3N\sim1.8\times10^{24}$ and a tiny $\delta=10^{-20}$ , the complement is $(1-\delta)^n=\exp[n\ln(1-\delta)]\approx\exp[-n\delta]=\exp[-1.8\times10^4]$ , so $f\approx1-e^{-18000}\approx1$ . Why: float arithmetic rounds $1-10^{-20}$ to exactly $1$ ; the honest computation keeps the exponent $n\delta\sim10^4$ , which is what makes the fraction indistinguishable from 1.

Step 4. Conclude the log-equivalence. Since $\Sigma\propto E^{3N/2}$ and the shell holds essentially all of $\Sigma$ , $\ln(g\,\Delta)-\ln\Sigma=\ln(g\Delta/\Sigma)=\mathcal{O}(\ln N)$ , negligible against $\ln\Sigma=\mathcal{O}(N)$ . Hence $\ln\Sigma\approx\ln(g\Delta)\approx\ln\Omega$ to relative order $1/N$ — the entropy is the same however you count. Verified: the fraction $1-(1-\delta)^n$ for $n=3N$ at several $N$ , the mole case in log space ( $n\ln(1-\delta)\approx-1.8\times10^4$ ), and $|\ln(g\Delta)-\ln\Sigma|/\ln\Sigma\to0$ are all checked in the companion script.

Limiting cases

How the accessible region and its volume behave in the regimes that matter:

Case	Result	Meaning
Thermodynamic limit $N\to\infty$	$\ln\Sigma\approx\ln(g\Delta)\approx\ln\Omega$	Shell, ball, and surface counts coincide in log; $\Delta$ irrelevant. Entropy is unambiguous.
Sharp energy $\Delta\to0$	$\rho\to\dfrac{\delta(H-E)}{g(E)}$	The shell collapses to the surface; the density of states $g(E)$ replaces the shell volume as normalizer.
Extra conserved quantities	$\mathcal{A}=\{H=E\}\cap\{I_2=i_2\}\cap\cdots$	Momentum, angular momentum, or integrable invariants further restrict the accessible set to a lower-dimensional intersection.
Ideal gas	$\Sigma\propto V^N E^{3N/2}$ , $g\propto V^N E^{3N/2-1}$	Explicit $E$ -scaling; gives Sackur–Tetrode entropy and $pV=Nk_BT$ on differentiating.
Low dimension ( $N$ small)	shell $\ne$ ball	The concentration argument fails; one must count the shell honestly, and $\Delta$ genuinely matters.

Common trap“The shell width Δ contaminates the entropy, so the answer is ambiguous.”

It does not, and seeing why is the whole point of the article. Because $\Sigma(E)\sim E^{3N/2}$ grows astronomically, the enclosed ball $\Sigma$ , the shell $g\Delta$ , and the surface $\Omega$ have logarithms that agree to relative order $1/N$ : multiplying $\Delta$ by a million shifts $\ln\Omega$ by $\ln 10^6\approx14$ , nothing against $S/k_B\sim N\sim10^{23}$ . So the microcanonical entropy is robust to how thick you draw the shell and even to whether you count the shell, the surface, or the whole ball. Two related cautions: (1) this equivalence is a thermodynamic-limit statement — for a small number of particles the shell and the ball genuinely differ and $\Delta$ matters. (2) When computing the shell fraction for a macroscopic system, never evaluate $1-(1-\delta)^n$ directly in floating point: $1-\delta$ rounds to $1$ and you get a spurious $0$ ; work with the exponent $n\ln(1-\delta)$ instead.

Worked examples

1 · The energy shell of a three-atom gas (confidence builder)

Three monatomic gas atoms are isolated in a box. Find the dimension of the momentum space, the exponent with which the enclosed phase volume $\Sigma(E)$ grows with energy, the density-of-states exponent, and the radius of the momentum ball.

Step 1. Count momentum dimensions. Three atoms in three-dimensional space have $3N=3\times3=\mathbf{9}$ momentum components. The accessible set at energy $E$ is an $8$ -dimensional surface in the $9$ -dimensional momentum space (times the configuration volume).

Step 2. Energy exponent of $\Sigma$ . From $\Sigma\propto E^{3N/2}$ with $N=3$ , the exponent is $3N/2=\mathbf{9/2}=4.5$ . The enclosed volume grows as $E^{4.5}$ — already steep, and this is only three atoms.

Step 3. Density-of-states exponent. $g(E)=\partial\Sigma/\partial E\propto E^{3N/2-1}=E^{\mathbf{7/2}}=E^{3.5}$ . So the number of microstates per unit energy climbs as $E^{3.5}$ .

Step 4. Momentum-ball radius. The energy shell is the surface $\sum_{i=1}^{9}p_i^2=2mE$ , a sphere of radius $R=\sqrt{2mE}$ in the $9$ -dimensional momentum space. For a helium atom ( $m=6.6\times10^{-27}\,\mathrm{kg}$ ) at $E=10^{-20}\,\mathrm J$ , $R=\sqrt{2(6.6\times10^{-27})(10^{-20})}=\sqrt{1.32\times10^{-46}}=\mathbf{1.15\times10^{-23}\ \mathrm{kg\,m\,s^{-1}}}$ . Every accessible microstate has its nine momentum components landing on this one sphere — the geometry the whole ensemble lives on.

2 · Why the shell width doesn't matter — for a mole (exam level)

A mole of gas ( $N=6.0\times10^{23}$ ) has $\Sigma(E)\propto E^{3N/2}$ . Show quantitatively that (a) essentially all the enclosed volume lies within a fractional energy depth of $10^{-20}$ of the surface, and (b) changing the shell width $\Delta$ by a factor of $10^6$ shifts the entropy negligibly.

Step 1. Translate energy depth to radius depth. Since $\Sigma\propto E^{3N/2}$ and the momentum ball radius is $R\propto\sqrt E$ , a fractional energy depth $\delta_E$ corresponds to a fractional radius depth $\delta\approx\delta_E/2$ . Take $\delta=10^{-20}$ for the estimate; the momentum-space dimension is $n=3N=1.8\times10^{24}$ .

Step 2. Fraction in the skin, done in log space. The fraction of volume outside the inner ball is $1-(1-\delta)^n$ . Do not evaluate directly — compute the exponent: $n\ln(1-\delta)\approx-n\delta=-(1.8\times10^{24})(10^{-20})=\mathbf{-1.8\times10^{4}}$ . So the inner ball holds a fraction $e^{-18000}\approx0$ , and the skin holds $1-e^{-18000}\approx\mathbf{1}$ . Essentially the entire enclosed volume is a one-part-in- $10^{20}$ skin.

Step 3. Entropy shift from changing $\Delta$ . The shell entropy is $S=k_B\ln(g(E)\Delta)$ . Multiplying $\Delta$ by $10^6$ adds $k_B\ln10^6=k_B(13.8)$ to $S$ . Compare to the total: $S\sim Nk_B\sim6\times10^{23}k_B$ .

Step 4. The ratio is the punchline. The fractional change is

\frac{\Delta S}{S}\sim\frac{13.8}{6\times10^{23}}\approx\mathbf{2\times10^{-23}}.

A million-fold change in the arbitrary shell width moves the entropy by two parts in $10^{23}$ — undetectable. This is why the microcanonical ensemble gives a definite entropy despite the apparently arbitrary $\Delta$ , and why one may count the shell, the surface, or the whole ball interchangeably.

Origins · 1 mindensity of states and how this was discovered

density of states

the phase-volume bookkeeping of the energy shell · from Boltzmann and Gibbs to Ehrenfest

The density of states $g(E)=\partial\Sigma/\partial E$ is the quantity that converts the geometry of the energy shell into thermodynamics, and its lineage runs through the founders of the field. Boltzmann, counting microstates on the energy surface in the 1870s, established that entropy is the logarithm of an accessible count; Gibbs, in his 1902 Elementary Principles in Statistical Mechanics, gave the modern definitions of the microcanonical ensemble and the phase-space volume $\Sigma(E)$ , and analyzed the "extension-in-phase" that $g(E)$ measures. Paul Ehrenfest, with Tatyana Ehrenfest, clarified in their 1911 encyclopaedia article how the thin-shell and enclosed-volume counts become equivalent in the thermodynamic limit — the concentration-of-measure fact this article proves. The result that $\ln\Sigma\approx\ln\Omega$ for large $N$ is a cornerstone of that analysis, and the steep $E^{3N/2}$ growth of $g(E)$ for the ideal gas is the calculation from which the Sackur–Tetrode entropy and the ideal-gas law both descend.

History

1870

s — Boltzmann counts microstates on the constant-energy surface, defining entropy as a logarithmic count.

1902

— Gibbs's Elementary Principles introduces the phase-space volume

\Sigma(E)

and the microcanonical ensemble.

1911

— Paul and Tatyana Ehrenfest's encyclopaedia article clarifies the equivalence of shell, surface, and enclosed-volume counts.

1912

— Sackur and Tetrode independently obtain the absolute entropy of a monatomic ideal gas from the phase-space count.

1927

— the concentration-of-measure viewpoint matures in probability theory, later illuminating why high-dimensional volume lives at the surface.

Going further

Texts. Pathria & Beale, Statistical Mechanics, 3rd ed., §1.4, §2.4 · Kardar, Statistical Physics of Particles, ch. 4 · Reif, Fundamentals of Statistical and Thermal Physics, ch. 2 · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, ch. 3 · Landau & Lifshitz, Statistical Physics Part 1, §7.

Provenance. Standard results: the accessible surface $H=E$ , the enclosed volume $\Sigma(E)$ with the $h^{3N}N!$ factor, the density of states $g=\partial\Sigma/\partial E$ , the $n$ -ball volume $V_n(R)=\pi^{n/2}R^n/\Gamma(n/2+1)$ , the ideal-gas scaling $\Sigma\propto V^N E^{3N/2}$ , and the high-dimensional concentration $1-(1-\delta)^n\to1$ . A verifier should confirm: (1) $V_n$ for $n=1,2,3$ gives $2R$ , $\pi R^2$ , $\tfrac43\pi R^3$ ; (2) the ideal-gas momentum ball gives energy exponent $3N/2$ and $g\propto E^{3N/2-1}$ ; (3) the skin fraction $1-(1-\delta)^n$ at $n=3$ , $30$ , $300$ and (in log space) for a mole; (4) $|\ln(g\Delta)-\ln\Sigma|/\ln\Sigma\to0$ as $N$ grows. The helium-atom momentum radius $R=1.15\times10^{-23}\,\mathrm{kg\,m\,s^{-1}}$ and the $\Delta S/S\sim2\times10^{-23}$ shift are recomputed in the companion script. Everything is reproduced there.

Prerequisites. Phase space · microstates and macrostates · Liouville's theorem. Leads to. The fundamental postulate of equal a priori probabilities · Boltzmann entropy · statistical ensembles · the ergodic hypothesis.

Why can't we just use the exact energy surface instead of a shell of finite width?

The exact surface $H=E$ is a $(6N-1)$ -dimensional set with zero volume in the full $6N$ -dimensional phase space, so you cannot straightforwardly count microstates "in" it — a uniform measure on a zero-volume set needs extra machinery (a surface measure $\mathrm dS/|\nabla H|$ ). Thickening it into a shell $E\le H\le E+\Delta$ gives a genuinely finite volume you can count directly. There is also a physical reason: no real isolation fixes the energy to infinite precision, and quantum mechanics forbids arbitrarily sharp energy anyway. The reassuring result is that the answer doesn't depend on $\Delta$ : in the thermodynamic limit the shell, the surface, and the whole enclosed ball all give the same entropy to relative order $1/N$ .

Where does the exponent 3N/2 actually come from?

From counting momentum dimensions. For a monatomic ideal gas the energy is purely kinetic, $E=\sum_{i=1}^{3N}p_i^2/2m$ , so the set of momenta with energy $\le E$ is a ball of radius $\sqrt{2mE}$ in $3N$ -dimensional momentum space. The volume of an $n$ -ball scales as $R^n$ , and here $n=3N$ with $R\propto\sqrt E$ , so the volume scales as $(\sqrt E)^{3N}=E^{3N/2}$ . Each of the $3N$ momentum components contributes one factor of $\sqrt E$ ; that is the entire origin of the half-integer exponent. Differentiating gives the density of states $g\propto E^{3N/2-1}$ , and it is this steep growth that both makes the shell equal to the ball and, on differentiating the entropy, yields the equipartition result $E=\tfrac32Nk_BT$ .

How can "almost all the volume is at the surface" possibly be true? It contradicts my 3D intuition.

Your 3D intuition is correct for 3D — there, the outer 10% skin of a ball holds only about 27% of the volume. The surprise is how fast this changes with dimension. The fraction in that skin is $1-(0.9)^n$ : at $n=30$ it is 96%, at $n=300$ it is 99.997%, and for a mole ( $n=3N\sim10^{24}$ ) it is 1 minus an astronomically small number. The mechanism is simply that volume scales as $R^n$ , so shrinking the radius by even 10% removes a factor $(0.9)^n$ of the volume, which is nearly everything when $n$ is huge. This "concentration of measure" is not a statistical-mechanics quirk; it is a general fact of high-dimensional geometry that also governs high-dimensional statistics and machine learning.

Is Σ(E) (the enclosed ball), Ω(E) (the surface), or gΔ (the shell) the "right" one to use?

In the thermodynamic limit it genuinely does not matter — their logarithms agree to relative order $1/N$ , so all three give the same entropy. Which one is most convenient depends on the problem: the enclosed volume $\Sigma(E)$ is often easiest to integrate (a solid ball rather than a surface); the shell $g(E)\Delta$ is closest to the physical picture of a system near energy $E$ ; and the surface count $\Omega(E)=\int\mathrm dS/|\nabla H|$ appears when you take the sharp-energy $\Delta\to0$ limit. For pedagogy and most calculations people use $\Sigma$ or $g$ . The freedom to switch is itself a theorem — the shell-equals-ball result — and for a small number of particles that freedom disappears and you must be careful about which you mean.

What does "accessible" add beyond "constant energy"?

Energy is usually the dominant constraint, but it is not the only conserved quantity. An isolated system also conserves total linear momentum and total angular momentum, and an integrable system conserves a whole tower of action variables. Each conservation law is another equation the trajectory must satisfy, cutting the accessible set down to the intersection of all the constraint surfaces — a lower-dimensional submanifold than the energy surface alone. For a gas in a fixed box the walls break momentum conservation, so energy is effectively the only constraint and "accessible" ≈ "constant energy." But for a freely floating system, or an integrable one, "accessible" is strictly smaller than the energy shell, and using the full shell would over-count — the same caveat that appears in the fundamental postulate.

Does this connect to the density of states I've seen in solid-state physics?

Yes — it is the same object, $g(E)=\partial\Sigma/\partial E$ , the number of microstates (or single-particle states) per unit energy. In this classical foundations article $g(E)$ counts the growth of accessible phase-space volume with energy; in solid-state physics the single-particle electronic density of states counts quantum states per unit energy near the Fermi level. They share the name and the meaning because both answer "how many states open up as I add a little energy?" The solid-state version controls heat capacity, conductivity, and the metal/insulator distinction, and it is measured directly by photoemission and tunneling spectroscopy. The classical shell picture here is the conceptual ancestor: entropy $=k_B\ln$ (accessible states), with $g(E)$ doing the counting.