Extensivity and Additivity of Entropy

Cut a cup of coffee in half and you have two half-cups, each with half the entropy of the whole. That sounds trivial, yet it hides one of the deepest structural facts in physics: entropy is extensive, a homogeneous function of degree one in energy, volume, and particle number, $S(\lambda E,\lambda V,\lambda N)=\lambda S(E,V,N)$ . Everything in equilibrium thermodynamics leans on this one property. It is what makes temperature, pressure, and chemical potential well-defined as ratios rather than as size-dependent quantities; it is what makes the Euler relation $E=TS-pV+\mu N$ hold; it is what lets the free energies of independent systems simply add. Statistical mechanics has to earn extensivity from a microscopic count of states, and doing so honestly exposes hidden physics. The primitive notion is additivity: when two subsystems barely interact, their microstate counts multiply, $\Omega=\Omega_1\Omega_2$ , and Boltzmann's $S=k_{\mathrm B}\ln\Omega$ turns that product into a sum. Multiplicativity of counts is equivalent to additivity of entropy — and the logarithm is the only function that performs the conversion. The physical precondition is that forces are short-ranged, so the energy of interaction across a dividing surface scales as area, not volume, and vanishes relative to the bulk as $N\to\infty$ . The historically decisive test is the classical ideal gas. Count distinguishable particles and you get an entropy with a $Nk_{\mathrm B}\ln V$ term that refuses to be extensive — the notorious Gibbs paradox, in which removing a partition between two samples of the same gas spuriously raises the entropy. The cure is to admit that identical particles are indistinguishable: divide phase-space volume by $N!$ and cell size by $h^{3N}$ , and the extensive Sackur–Tetrode entropy emerges. Both fixes are quantum in origin, yet classical thermodynamics forces them on us. Extensivity is not universal, though: gravity and other long-range forces break it outright, and that failure signals genuinely new physics — negative specific heat, ensemble inequivalence, gravothermal collapse.

Key takeaways

Entropy is extensive when $S(\lambda E,\lambda V,\lambda N)=\lambda S(E,V,N)$ (it scales linearly with system size) and additive when a composite's entropy is the sum of its parts, $S=S_1+S_2$ .
Additivity is equivalent to the multiplicativity of microstate counts, $\Omega=\Omega_1\Omega_2$ , which holds when subsystems are statistically independent — guaranteed by short-range forces whose cross-surface coupling is subextensive ( $\sim N^{-1/d}$ of the bulk).
The distinguishable ideal-gas entropy from $\Omega\propto V^N$ is not extensive and gives the Gibbs paradox: naive counting predicts a spurious $\Delta S=2Nk_{\mathrm B}\ln 2$ when two identical $N$ -gas halves are combined.
Dividing by $N!$ (correct indistinguishability counting) turns $\ln V$ into $\ln(V/N)$ , cancels the paradox exactly, and yields the extensive Sackur–Tetrode entropy, verified against argon to the third figure.
Extensivity fails for long-range interactions (gravity $\sim N^{5/3}$ , unscreened Coulomb), for mesoscopic systems where surface $\sim$ bulk, and near critical points — and thermodynamics built on extensivity must then be rebuilt.
Key idea: Additivity of entropy ( $S=S_1+S_2$ ) is equivalent to multiplicativity of microstate counts ( $\Omega=\Omega_1\Omega_2$ ), which requires statistically independent subsystems — guaranteed by short-range interactions. Extensivity additionally requires correct indistinguishability counting (the $1/N!$ factor).

The core insight

Additivity of entropy ( $S=S_1+S_2$ ) is equivalent to multiplicativity of microstate counts ( $\Omega=\Omega_1\Omega_2$ ), which requires statistically independent subsystems — guaranteed by short-range interactions whose cross-surface coupling is subextensive. Extensivity additionally requires correct indistinguishability counting (the $1/N!$ factor).

Where this lives today

Extensivity is quietly assumed every time an engineer sizes a chemical reactor or a materials scientist computes the free energy of an alloy — but the frontier is where it breaks. Self-gravitating systems (globular clusters, galaxies) have non-additive entropy, negative microcanonical specific heat, and inequivalent ensembles; the "gravothermal catastrophe" that drives core collapse in star clusters is a direct consequence. The same non-additivity appears in trapped long-range dipolar gases and in mean-field spin models, now studied intensively under the banner of "long-range interacting systems". Sackur–Tetrode remains a live experimental target: modern measurements of the absolute entropy of trapped noble gases, and precision determinations of the Boltzmann constant via acoustic and dielectric gas thermometry, test the same $h^3$ coarse-graining that Sackur and Tetrode introduced in 1912. Every high-entropy alloy — a class of structural materials stabilized by their large configurational mixing entropy $-Nk_{\mathrm B}\sum x_i\ln x_i$ — is an application of the distinct-species mixing formula this article derives.

The physical picture

Picture a large box of gas mentally divided into two halves by an imaginary plane. Almost every molecule lives in the bulk and only feels neighbours within a short interaction range; the molecules straddling the plane feel forces across it, but they form a vanishing surface fraction $\sim N^{2/3}/N=N^{-1/3}\to0$ . So to overwhelming accuracy the two halves are statistically independent, their state counts multiply, and the logarithm — the entropy — adds. Extensivity is simply the macroscopic shadow of short-range microscopic forces. The Gibbs paradox is a sanity check the universe insists on passing. If you imagine labelling the two halves of an identical gas and then "mixing" them, nothing physically changes — yet naive labelled counting says the entropy rose by $2Nk_{\mathrm B}\ln 2$ . The fix is to recognise that swapping two identical atoms produces no new microstate: nature counts configurations, not labelled permutations. The $1/N!$ is not a fudge factor but the imprint of quantum indistinguishability on classical bookkeeping.

Figure

Short-range forces make entropy additive — long-range forces do not

Why the extra factor of $h^{3N}$ ? Classically the phase-space volume is dimensionful, so $\ln\Omega$ would carry an arbitrary additive constant per particle. Quantum mechanics quantizes phase space into cells of size $h^{3N}$ — one cell per state, by the uncertainty principle — making $\Omega$ a pure number. Both quantum corrections are required together: $h$ for the cell size makes entropy absolute (the third law), and $N!$ for indistinguishability makes it extensive. Neither is optional accounting.

Figure

The Gibbs paradox and its cure: mixing two identical N-gas halves

The core equations

Extensivity is the precise mathematical statement that entropy is a first-order homogeneous function. If you double every extensive input at fixed composition, you double the entropy — no more, no less. This is the single equation that the rest of the article must reproduce from microscopics and that occasionally fails in nature.

S(\lambda E,\lambda V,\lambda N)=\lambda\,S(E,V,N) \qquad (1)

where: $S$ — entropy as a function of the extensive variables; $E,V,N$ — internal energy, volume, particle number; $\lambda>0$ — an arbitrary scaling factor. Reading it physically: any intensive quantity, being a ratio of two extensive ones, is invariant under this scaling — so temperature and pressure do not care how big the sample is. The microscopic engine of additivity is that independent subsystems have multiplicative state counts. This is where the logarithm in Boltzmann's entropy is not a choice but a necessity — it is the unique operation that converts a product of counts into a sum of entropies.

\Omega = \Omega_1\,\Omega_2 \;\Longrightarrow\; S = k_{\mathrm B}\ln\Omega = S_1 + S_2 \qquad (2)

where: $\Omega$ — accessible microstate count of the composite; $\Omega_{1,2}$ — counts of the independent subsystems; $k_{\mathrm B}=1.380649\times10^{-23}\,\mathrm{J\,K^{-1}}$ (exact, SI). Reading it physically: statistical independence ( $\Omega$ factorizes) is the same statement as additivity of entropy — the two are logically equivalent. Now the classical ideal gas, counted as if the particles carried name tags. The result contains $Nk_{\mathrm B}\ln V$ , and the bare $\ln V$ (rather than $\ln(V/N)$ ) is exactly what wrecks extensivity — scaling $V$ and $N$ together leaves a stubborn $\lambda\ln\lambda$ term behind.

S_{\text{dist}} = N k_{\mathrm B}\left[\ln\!\left(V\left(\tfrac{4\pi mE}{3Nh^2}\right)^{3/2}\right)+\tfrac{3}{2}\right] \qquad (3)

where: $N$ — particle number; $m$ — particle mass; $h$ — Planck constant (phase-space cell size); $V,E$ — volume and total kinetic energy. Reading it physically: this is a legitimate count of labelled configurations — but labelled configurations are not distinct physical states for identical atoms, so this entropy over-counts. The repair is a single division. Permuting $N$ identical particles yields the same physical microstate, so the labelled count over-counts by exactly $N!$ .

\Omega_{\text{indist}}=\frac{1}{N!}\,\Omega_{\text{dist}} \qquad (4)

where: $N!$ — the number of particle permutations, each producing the identical physical state. Reading it physically: this factor is the entire content of indistinguishability in the classical limit; it is forced not by any measured number but by the demand of extensivity. With the $1/N!$ inserted and Stirling applied, $\ln V$ becomes $\ln(V/N)$ and every term now depends on intensive ratios times the prefactor $N$ . This is the Sackur–Tetrode equation — the correct, extensive, absolute entropy of a monatomic ideal gas.

S = N k_{\mathrm B}\left[\ln\!\left(\frac{V}{N}\left(\frac{4\pi mE}{3Nh^2}\right)^{3/2}\right)+\frac{5}{2}\right] \qquad (5)

where: $V/N$ — volume per particle, the intensive combination that restores extensivity; $\tfrac{5}{2}=\tfrac{3}{2}+1$ , the $+1$ coming from Stirling's $\ln N!\approx N\ln N-N$ . Reading it physically: because $h$ makes $\Omega$ a pure number, this predicts absolute entropies, not just differences — a prediction confirmed for the noble gases. Extensivity of $E(S,V,N)$ has an immediate structural payoff. Euler's theorem for homogeneous functions collapses the fundamental relation into a finite, closed form.

E = TS - pV + \mu N \qquad (6)

where: $T=(\partial E/\partial S)_{V,N}$ ; $p=-(\partial E/\partial V)_{S,N}$ ; $\mu=(\partial E/\partial N)_{S,V}$ . Reading it physically: this Euler relation exists only because $E$ is extensive; it is the backbone of the whole apparatus of thermodynamic potentials, and it fails whenever extensivity fails. Finally, when the species are genuinely distinct, mixing does raise the entropy, and the amount is a universal combinatorial formula.

\Delta S_{\text{mix}} = -N k_{\mathrm B}\big(x\ln x + (1-x)\ln(1-x)\big) \qquad (7)

where: $x$ — mole fraction of species A; the result is strictly positive for distinct gases and identically zero for identical gases once $1/N!$ is included. Reading it physically: the discontinuity between "identical" (zero) and "distinct" (positive) is the sharp experimental signature of indistinguishability.

Derivation AResolving the Gibbs paradox: mixing identical gases

We show that labelled ("distinguishable") counting predicts a spurious entropy increase on removing a partition between two samples of the same gas, and that the indistinguishability factor $1/N!$ cancels it exactly. The setup is two volumes of one ideal gas at the same density and temperature, separated by a partition that is then removed; the physical macrostate is unchanged, so the correct answer must be $\Delta S=0$ , and the naive answer's failure to deliver that is precisely the paradox we must dissolve.

Step 1. Take the concrete symmetric case: two halves each holding $N$ particles in volume $V_h$ , at equal density $n=N/V_h$ . Labelled counting gives each half an entropy containing $Nk_{\mathrm B}\ln V_h$ (equation 3).

Step 2. After the partition is removed, all $2N$ particles share $2V_h$ , so the combined labelled entropy contains $2Nk_{\mathrm B}\ln(2V_h)$ . Why this over-counts: labelled counting treats a swap of two identical atoms as a new configuration, which it is not.

Step 3. The naive change is $\Delta S = k_{\mathrm B}\big[2N\ln(2V_h) - N\ln V_h - N\ln V_h\big] = 2Nk_{\mathrm B}\ln 2 > 0$ . A strictly positive entropy jump for a process in which nothing physical happened — the Gibbs paradox.

Step 4. Now insert the Gibbs factor: dividing $\Omega$ by $N!$ replaces $\ln V$ by $\ln(V/N)$ everywhere (Stirling: $\ln V^N/N!\to N\ln(V/N)+N$ ). Entropy now depends on the density $V/N$ , not on $V$ alone.

Step 5. At equal density $V_h/N=(2V_h)/(2N)=1/n$ , so $\Delta S = k_{\mathrm B}\big[2N\ln\tfrac{2V_h}{2N} - N\ln\tfrac{V_h}{N} - N\ln\tfrac{V_h}{N}\big]=0$ . The same algebra, applied to distinct species whose individual $V/N_i$ genuinely change, reproduces $\Delta S_{\text{mix}}=-Nk_{\mathrm B}\sum_i x_i\ln x_i>0$ . Verified: symbolic check confirms naive $\Delta S=2Nk_{\mathrm B}\ln2$ and, with $1/N!$ , $\Delta S=0$ at equal density; the distinct-species formula peaks at $\ln2$ per particle at $x=\tfrac12$ (Pathria & Beale §1.5–1.6; Reif §7.3).

Go deeper · Derivation B · 1 minExtensivity from short-range interactions: subextensivity of the surface energy

Derivation BExtensivity from short-range interactions: subextensivity of the surface energy

We demonstrate why short-range forces make the cross-surface interaction energy negligible, so that $\Omega$ factorizes and entropy becomes additive. The strategy is to partition a homogeneous $d$ -dimensional system of $N$ particles into two halves and estimate the interaction energy that couples them across the dividing surface, then compare it to the bulk energy; the whole question of additivity reduces to a scaling comparison of these two numbers as $N\to\infty$ .

Step 1. For short-range forces (interaction range $\ll$ system size $L$ ), the bulk energy scales as the number of particles, $E_{\text{bulk}}\sim N$ , while the cross-surface coupling scales as the interface area, $E_{\text{surface}}\sim L^{d-1}\sim N^{(d-1)/d}$ .

Step 2. Write the joint partition function $Z=\mathrm{Tr}\,e^{-\beta(H_1+H_2+H_{12})}$ , where $H_{12}$ couples only particles within the interaction range of the surface. Why $H_{12}$ can be dropped: $\beta\langle H_{12}\rangle\sim N^{(d-1)/d}$ grows slower than the extensive $\beta\langle H\rangle\sim N$ , so it is a vanishing fraction of the whole.

Step 3. The ratio $E_{\text{surface}}/E_{\text{bulk}}\sim N^{(d-1)/d}/N = N^{-1/d}\to0$ . In three dimensions this is $N^{-1/3}$ : for $N\sim10^{24}$ the surface contributes one part in $10^{8}$ .

Step 4. Hence $\ln Z=\ln Z_1+\ln Z_2+O(N^{(d-1)/d})$ , so the free energy and the entropy are additive to leading order, $S=S_1+S_2$ . For long-range forces where $H_{12}\sim N$ (every pair couples), this factorization fails and additivity is lost. Verified: numeric check confirms $E_{\text{surface}}/E_{\text{bulk}}=N^{-1/d}$ for $d=1,2,3$ across $N=10^6$ – $10^{24}$ ; existence-of-thermodynamic-limit theorems (Ruelle, Lieb) require exactly such tempered short-range potentials (Kardar SPoP ch. 4).

Limiting cases

When does $S$ scale linearly with $N$ , and when does it break?

Case	Result	Meaning
Short-range, thermodynamic limit	$S(\lambda E,\lambda V,\lambda N)=\lambda\,S(E,V,N)$	Standard regime: potentials decaying faster than $r^{-d}$ in $d$ dimensions. Surface energy is subextensive ( $\sim N^{-1/d}$ of bulk); extensivity and additivity both hold.
Long-range interactions	$E_{\text{int}}\sim N^2,\ S\sim N^{5/3}\ \Rightarrow\ \Omega\neq\Omega_1\Omega_2$	Gravity ( $1/r$ ), unscreened Coulomb. Energy is non-additive; ensembles inequivalent; microcanonical negative specific heat appears. Kac rescaling $E_{\text{int}}/N$ restores a formal extensivity but with non-concave $S(E)$ .
Finite / mesoscopic systems	$S = aN + bN^{2/3} + \dots$	Surface terms $\sim N^{2/3}$ are not negligible for small $N$ ; strict extensivity is only the leading-order $N\to\infty$ behaviour.
Distinguishable localized modes	$S_{\text{crystal}}=3Nk_{\mathrm B}[1+\ln(k_{\mathrm B}T/\hbar\omega)]$ (no $1/N!$ )	Oscillators fixed at lattice sites are distinguishable by position; the entropy is already extensive and must not be divided by $N!$ .

Worked examples

Example 1 — The Gibbs paradox as a number. Compute the spurious mixing entropy that labelled counting assigns to combining two boxes of $N$ identical argon atoms at equal density, and show the $1/N!$ correction removes it.

Step 1. Labelled entropy of each half contains $Nk_{\mathrm B}\ln V_h$ ; combined system contains $2Nk_{\mathrm B}\ln(2V_h)$ (all $2N$ atoms share $2V_h$ ).

Step 2. Naive change: $\Delta S = k_{\mathrm B}[2N\ln(2V_h)-2N\ln V_h]=2Nk_{\mathrm B}\ln 2$ . For one mole, $N=N_A$ and $2N_A k_{\mathrm B}\ln2 = 2R\ln2 = 2(8.314)(0.6931)=11.5\ \mathrm{J\,K^{-1}}$ of entropy conjured from nothing.

Step 3. With $1/N!$ , entropy depends on $V/N$ only. At equal density $V_h/N=(2V_h)/(2N)$ , so $\Delta S = 2Nk_{\mathrm B}\ln\tfrac{2V_h}{2N} - 2Nk_{\mathrm B}\ln\tfrac{V_h}{N}=0$ . The naive answer is $2R\ln2\approx11.5\ \mathrm{J\,K^{-1}}$ per mole of spurious entropy; the correct answer with indistinguishability is $\Delta S=0$ . The gap is precisely the Gibbs paradox.

Example 2 — Absolute entropy of argon from Sackur–Tetrode. Evaluate the Sackur–Tetrode molar entropy of argon at standard conditions and compare with the measured value $154.8\ \mathrm{J\,mol^{-1}K^{-1}}$ . Argon: $m=6.63\times10^{-26}\,\mathrm{kg}$ , $T=298.15\,\mathrm{K}$ , $p=10^5\,\mathrm{Pa}$ .

Step 1. Thermal de Broglie wavelength $\lambda_T = h/\sqrt{2\pi m k_{\mathrm B}T} = (6.626\times10^{-34})/\sqrt{2\pi(6.63\times10^{-26})(1.381\times10^{-23})(298.15)} = 1.60\times10^{-11}\,\mathrm{m}$ .

Step 2. Volume per particle from the ideal-gas law: $V/N = k_{\mathrm B}T/p = (1.381\times10^{-23})(298.15)/10^5 = 4.12\times10^{-26}\,\mathrm{m^3}$ .

Step 3. The dimensionless log-argument is $\ln\!\big((V/N)\lambda_T^{-3}\big) = \ln\!\big(4.12\times10^{-26}/(1.60\times10^{-11})^3\big) = \ln(1.005\times10^{7}) = 16.12$ . Why $\lambda_T^{-3}$ : the momentum-space factor $(4\pi mE/3Nh^2)^{3/2}$ equals $\lambda_T^{-3}$ for an ideal gas with $E=\tfrac32 Nk_{\mathrm B}T$ .

Step 4. So $S/Nk_{\mathrm B} = 16.12 + \tfrac52 = 18.62$ , and per mole $S_m = 18.62\,R = 18.62(8.314) = 154.8\ \mathrm{J\,mol^{-1}K^{-1}}$ . The Sackur–Tetrode prediction is $154.8\ \mathrm{J\,mol^{-1}K^{-1}}$ , matching the experimental standard molar entropy of argon to the third figure — a quantitative confirmation that the $h^3$ cell and $1/N!$ factor are physically real, not bookkeeping conveniences.

History band

Origins · 1 minJosiah Willard Gibbs and how this was discovered

Josiah Willard Gibbs

1839–1903 · American mathematical physicist who founded chemical thermodynamics and statistical mechanics; his 1875–78 memoir "On the Equilibrium of Heterogeneous Substances" introduced the chemical potential and the mixing-entropy analysis now bearing his name, and his 1902 "Elementary Principles in Statistical Mechanics" coined the term "statistical mechanics" and the ensemble framework. The Gibbs paradox — that labelled counting spuriously raises entropy on mixing identical gases — sharpened the question of indistinguishability decades before quantum mechanics answered it.

History

1865

— Clausius names the state function "entropy" and states dS = dQ_rev/T

1875

— Gibbs begins "On the Equilibrium of Heterogeneous Substances", introducing the chemical potential

1876

— Gibbs analyzes the entropy of mixing and identifies the paradox for identical gases

1877

— Boltzmann relates entropy to microstate counting, S = k ln W (the constant k is Planck's later notation)

1902

— Gibbs publishes "Elementary Principles in Statistical Mechanics", the ensemble framework

1912

— Sackur and Tetrode independently derive the absolute ideal-gas entropy using Planck's h

1931

— theorems on the existence of the thermodynamic limit for short-range potentials (later Ruelle, Lieb) formalized

Going further

References. Pathria & Beale, Statistical Mechanics, 3rd ed., §1.5–1.6 (Gibbs paradox, Sackur–Tetrode) · Kardar, Statistical Physics of Particles, ch. 4 · Reif, Fundamentals of Statistical and Thermal Physics, §7.3 and §2.5 · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, ch. 3 · Landau & Lifshitz, Statistical Physics Part 1, §7, §24 · Campa, Dauxois, Ruffo, Physics Reports 480 (2009) 57 — long-range interactions and non-additivity.

Provenance. All equations are standard textbook results; Sackur–Tetrode and the Gibbs paradox follow Pathria & Beale §1.5–1.6 and Reif §7.3, and the short-range condition for the thermodynamic limit follows Kardar SPoP ch. 4 (Ruelle/Lieb). Numerical check: standard molar entropy of argon $=154.8\,\mathrm{J\,mol^{-1}K^{-1}}$ (NIST/CODATA); $k_{\mathrm B}=1.380649\times10^{-23}\,\mathrm{J\,K^{-1}}$ , $R=8.314\,\mathrm{J\,mol^{-1}K^{-1}}$ . A verifier should recompute $\lambda_T$ , $V/N$ at STP, the Sackur–Tetrode value, the mixing-entropy formula, and the $N^{-1/d}$ surface-to-bulk scaling. Base-migration note: the source draft mislabelled the argon intermediates ( $\ln$ -argument $18.6$ , $S/Nk_{\mathrm B}$ $21.1$ ); the correct chain is $\ln$ -argument $16.12$ , $S/Nk_{\mathrm B}=18.62$ , giving the same final $154.8\ \mathrm{J\,mol^{-1}K^{-1}}$ — corrected here and reflected in the worked example and cheat sheet.

Builds on: Boltzmann Entropy · Accessible Phase Space and Energy Shells · Microstates and Macrostates · Phase Space Leads to: Statistical Ensembles · Gibbs Entropy · Equal A Priori Probabilities

Additive and extensive sound like the same thing — are they?

They are closely related but not identical. Additive means the entropy of a composite equals the sum of its parts, $S=S_1+S_2$ ; it is a statement about combining independent systems. Extensive means $S$ is a homogeneous function of degree one, $S(\lambda E,\lambda V,\lambda N)=\lambda S$ ; it is additivity applied recursively to identical replicas. Additivity requires only short-range interactions (so counts factorize); extensivity requires additivity plus correct indistinguishability counting (the $1/N!$ ), because without it even a single homogeneous gas fails to scale linearly.

Where does the factor of 5/2 in Sackur–Tetrode actually come from?

Two pieces add up. The $\tfrac32$ is the momentum-space (kinetic) contribution: integrating over the $3N$ -dimensional momentum ball on the energy shell gives an $E^{3N/2}$ factor, hence $\tfrac32$ per particle after taking the log and dividing by $N$ . The extra $+1$ comes from Stirling's approximation to $N!$ : $\ln N!\approx N\ln N - N$ , and the $-N$ (i.e. $+1$ per particle after the sign flips through $-k_{\mathrm B}\ln N!$ ) lifts $\tfrac32$ to $\tfrac52$ . So the $\tfrac52$ literally encodes both the kinetic degrees of freedom and the indistinguishability correction.

If I mix two gases that differ only by isotope, is the mixing entropy zero or positive?

Positive. Different isotopes are genuinely distinct species — they have different masses and are, in principle, distinguishable — so each expands into the full volume and its own $V/N_i$ changes, giving $\Delta S_{\text{mix}}=-Nk_{\mathrm B}\sum_i x_i\ln x_i>0$ . This is not a technicality: it is why isotope separation costs free energy (you must pay back the mixing entropy to un-mix), and it is the thermodynamic basis of enrichment. Only truly identical particles — same isotope, same internal state — give zero mixing entropy.

Why does gravity break extensivity when electromagnetism does not?

Both are $1/r$ forces, but electromagnetic charges come in two signs and screen: in any neutral plasma the effective interaction is cut off beyond the Debye length, making it effectively short-range, so entropy is extensive. Gravity has only one sign (mass is always attractive), so there is no screening; the interaction energy of $N$ masses grows as $N^2$ and the entropy as $N^{5/3}$ . That superextensivity is why self-gravitating systems have non-additive entropy, negative specific heat, and inequivalent ensembles — genuinely different thermodynamics.

Does the

1/N!

mean classical particles are "secretly quantum"?

In a sense, yes — and that is the striking part. Purely classical statistical mechanics cannot justify the $1/N!$ on its own; it is imposed only because extensivity demands it. Quantum mechanics explains it: identical particles are described by (anti)symmetrized wavefunctions in which permutations are not distinct states, so indistinguishability is built in from the start. The classical $1/N!$ is the shadow this quantum fact casts on the classical count. Likewise the $h^{3N}$ cell size is the shadow of phase-space quantization. Both are quantum corrections that classical thermodynamics could not derive but could not do without.

Is the entropy of a small nanoparticle extensive?

Not strictly. For a finite system the entropy carries surface corrections, $S = aN + bN^{2/3} + \dots$ , where the $N^{2/3}$ term counts atoms on the surface. For $N\sim10^{24}$ this term is one part in $10^{8}$ and utterly negligible, but for a nanoparticle of a few thousand atoms it is a percent-level correction that matters for melting-point depression, catalytic activity, and thermodynamic stability. Extensivity is an asymptotic ( $N\to\infty$ ) statement; treat it with care in the mesoscopic world.

Could I define entropy without the logarithm and keep additivity?

No, and this is the deep reason the logarithm is not arbitrary. Additivity demands a function $f$ with $f(\Omega_1\Omega_2)=f(\Omega_1)+f(\Omega_2)$ — that is, $f$ must turn products into sums. The only continuous solution is $f(\Omega)=k\ln\Omega$ for some constant $k$ . Any other choice (say $\Omega$ itself, or $\sqrt\Omega$ ) fails to be additive for independent subsystems. So the moment you insist entropy add when you combine independent systems, Boltzmann's $S=k_{\mathrm B}\ln\Omega$ is forced on you up to the units-fixing constant $k_{\mathrm B}$ .