Factorization of the Partition Function

A mole of gas has something like $10^{23}$ particles, each with three momentum components and three positions — a state space so vast that writing down its partition function directly, state by state, is not merely hard but physically impossible. And yet the ideal gas is a textbook problem, solved in a paragraph. The bridge between the impossible and the elementary is one algebraic observation: when the total energy is a sum of independent pieces, the partition function is a product of their partition functions. A sum in the exponent becomes a product of exponentials, and the astronomically large sum collapses into a handful of single-mode calculations. This is factorization, and it is the single most powerful computational move in equilibrium statistical mechanics. If $H=\sum_a H_a$ with the pieces acting on independent degrees of freedom, then $Z=\prod_a Z_a$ , and — taking the logarithm — $\ln Z=\sum_a\ln Z_a$ . That logarithm is the whole point: because thermodynamic potentials are logarithms of $Z$ , factorization is the microscopic origin of extensivity. Double the number of independent modes and you double $\ln Z$ , hence double the free energy, the energy, and the entropy. The three statements "no interaction," "energies add," and "the free energy is additive" are one statement viewed from three angles. The reach is enormous. The independent pieces can be separate subsystems, separate particles (an ideal gas, with the Gibbs $1/N!$ for indistinguishability), or the separate motions of a single molecule — translation, rotation, vibration, electronic. This article's owned result is that factorization is equivalent to the additivity of every extensive quantity, and its punchline is the diatomic heat-capacity staircase: $C_V$ climbs from $\tfrac32R$ to $\tfrac52R$ to $\tfrac72R$ as rotation and then vibration unfreeze — three plateaus that fall straight out of $Z=Z_{\text{tr}}Z_{\text{rot}}Z_{\text{vib}}$ . Every number here is python-verified.

Key takeaways

When the energy is a sum of independent contributions, $H=\sum_a H_a$ , the partition function factorizes: $Z=\prod_a Z_a$ , because $e^{-\beta\sum_a H_a}=\prod_a e^{-\beta H_a}$ .
Taking the log makes it additive: $\ln Z=\sum_a\ln Z_a$ , so $F$ , $U$ , $S$ , and $C_V$ are all sums over the independent pieces — factorization is extensivity.
For $N$ independent identical particles, $Z=Z_1^N/N!$ ; the Gibbs $1/N!$ restores an extensive entropy and resolves the Gibbs paradox.
A single molecule factorizes over its motions: $Z_1=Z_{\text{tr}}Z_{\text{rot}}Z_{\text{vib}}Z_{\text{el}}$ , so its heat capacity is the sum of the mode contributions.
The payoff is the diatomic $C_V$ staircase: $\tfrac32R\to\tfrac52R\to\tfrac72R$ as rotation then vibration unfreeze with rising $T$ .
Key idea: a sum in the exponent becomes a product of partition functions, so $\ln Z$ — and therefore all of thermodynamics — is additive over independent degrees of freedom. Factorization fails exactly when interactions couple the pieces, and that failure is where all the hard physics begins.

The core insight

When the Hamiltonian splits into a sum of independent pieces $H=\sum_a H_a$ , the partition function factorizes $Z=\prod_a Z_a$ and $\ln Z=\sum_a\ln Z_a$ . Every extensive quantity ( $U$ , $F$ , $S$ , $C_V$ ) then adds over the pieces — factorization is the microscopic origin of extensivity. It collapses a many-body sum to single-mode calculations, and it fails precisely when interactions couple the degrees of freedom.

Where this lives today

Factorization is the workhorse of computational thermochemistry. Software that predicts reaction rates, equilibrium constants, and thermodynamic tables constructs each molecule's partition function as the product $Z_1=Z_{\rm tr}Z_{\rm rot}Z_{\rm vib}Z_{\rm el}$ , using rotational constants and vibrational frequencies obtained from quantum-chemistry calculations or spectroscopy. Transition-state theory — the backbone of chemical kinetics — computes a rate constant as a ratio of such factorized partition functions. The entire discipline of statistical thermodynamics in chemistry is, operationally, the art of factoring $Z$ correctly and knowing when the factorization breaks (anharmonicity, internal rotation, mode coupling). In condensed-matter physics and quantum field theory, factorization provides the zeroth-order starting point for nearly every calculation. A crystal's phonons and a cavity's photons are treated as independent normal modes, so their partition functions are products over $k$ , giving the Debye and Planck laws. When interactions are present, one begins with the factorized non-interacting partition function $Z_0$ and treats the coupling perturbatively. The cluster expansion, virial expansion, and diagrammatic many-body theory are systematic corrections to this factorized starting point. Even in machine learning, "mean-field" approximations to intractable probabilistic models work by forcing a factorized (product) form on a distribution that does not truly factorize, then correcting — the same tension between independence and interaction that runs through all of statistical physics.

The physical picture

Roll several dice at once. Because the dice are independent, the number of ways to obtain a particular combination of faces is the product of the numbers of ways for the individual dice—the choices multiply. The partition function is exactly a weighted count of microstates, so independent degrees of freedom multiply their $Z$ 's. That is the entire mechanism: independence of the pieces $\Rightarrow$ additivity of energies in the exponent $\Rightarrow$ product of Boltzmann factors $\Rightarrow$ product of partition functions.

Figure

Independent pieces multiply: a joint microstate is one choice from each subsystem, so the total count Z factorizes into the product of the pieces' counts Z₁ × Z₂ × Z₃.

The deep consequence lies in the logarithm. Because $F=-k_B T\ln Z$ , $U=-\partial_\beta\ln Z$ , and $S=-\partial_TF$ are all derived from $\ln Z$ , and because $\ln$ converts the product $Z=\prod_a Z_a$ into the sum $\ln Z=\sum_a\ln Z_a$ , every extensive quantity inherits additivity from factorization. This is not a coincidence to be memorized; it is why extensivity exists at all. A macroscopic system is extensive precisely to the extent that it is built from weakly-coupled independent pieces — and the moment genuine long-range or strong interactions couple those pieces, both factorization and simple extensivity break down together (self-gravitating systems, which are non-extensive, are the classic warning). The most beautiful payoff is a single molecule factorizing over its own motions. A diatomic molecule stores energy in translation (3 directions), rotation (2 axes), and vibration (1 stretch mode), and to an excellent approximation these are independent, so $Z_1=Z_{\text{tr}}Z_{\text{rot}}Z_{\text{vib}}$ . Each motion has its own characteristic temperature below which it is quantum-mechanically "frozen out." As you heat the gas, the modes switch on one at a time, and the heat capacity climbs in discrete steps — the famous staircase below, computed directly from the factorized $Z$ .

Figure

The diatomic heat-capacity staircase from Z = Z_tr Z_rot Z_vib: C_V/R climbs 3/2 → 5/2 → 7/2 as rotation (near θ_rot) then vibration (near θ_vib) unfreeze.

The core equations

A system's degrees of freedom are independent when its Hamiltonian is a sum of pieces that act on separate state spaces (they commute, and no term couples them). This is the precondition for everything that follows — the definition of "non-interacting":

H=\sum_{a=1}^{M} H_a,\qquad [H_a,H_b]=0 \qquad (1)

where: $H$ — total Hamiltonian; $H_a$ — the Hamiltonian of the $a$ -th independent piece (a subsystem, a particle, or a single mode); $M$ — number of independent pieces; $[H_a,H_b]=0$ means the pieces live in independent state spaces. There are no cross terms $V_{ab}$ ; the instant one appears, this equation and its consequences fail. Because the exponential of a sum of commuting operators factorizes into the product of exponentials, and because the trace over a product Hilbert space also factorizes, the partition function factorizes:

Z=\mathrm{Tr}\,e^{-\beta H}=\prod_{a=1}^{M}\mathrm{Tr}_a\,e^{-\beta H_a}=\prod_{a=1}^{M}Z_a \qquad (2)

where: $\beta=1/k_BT$ ; $\mathrm{Tr}_a$ — trace over the state space of piece $a$ alone; $Z_a$ — its partition function. The equilibrium probability distribution factorizes into independent marginals, $P=\prod_a P_a$ , where $P_a=e^{-\beta H_a}/Z_a$ . This factorization is precisely the statistical signature of independent degrees of freedom. Taking the logarithm is where the physics lives, because it converts the product into the additive quantity that all thermodynamic potentials are built from:

\ln Z=\sum_{a=1}^{M}\ln Z_a,\qquad F=-k_BT\ln Z=\sum_{a=1}^{M}F_a \qquad (3)

where: $F_a=-k_BT\ln Z_a$ — the free energy of piece $a$ . Every extensive quantity now adds: $U=\sum_a U_a$ , $S=\sum_a S_a$ , $C_V=\sum_a C_{V,a}$ . Reach for this whenever a system is a collection of weakly-coupled parts — it turns one hard problem into $M$ easy ones. For $N$ identical particles the naive product $Z_1^N$ overcounts, because relabeling identical particles does not give a new physical state; the Gibbs factor $1/N!$ corrects this in the dilute (classical) regime:

Z_N=\frac{Z_1^N}{N!},\qquad Z_1=\frac{V}{\lambda_T^3},\qquad \lambda_T=\frac{h}{\sqrt{2\pi m k_BT}} \qquad (4)

where: $Z_1$ — single-particle partition function; $\lambda_T$ — thermal de Broglie wavelength; $N!$ — Gibbs indistinguishability factor (essential for extensive $S$ ). Without the $1/N!$ , the ideal-gas entropy is not extensive and you resurrect the Gibbs paradox of mixing identical gases. Inside a single molecule the same logic applies to its separable motions, and their partition functions multiply:

Z_1=Z_{\text{tr}}\,Z_{\text{rot}}\,Z_{\text{vib}}\,Z_{\text{el}},\qquad \varepsilon=\varepsilon_{\text{tr}}+\varepsilon_{\text{rot}}+\varepsilon_{\text{vib}}+\varepsilon_{\text{el}} \qquad (5)

where: each factor is the partition function of one kind of motion. The vibrational factor for a single mode of frequency $\omega$ is $Z_{\text{vib}}=e^{-\beta\hbar\omega/2}/(1-e^{-\beta\hbar\omega})=1/[2\sinh(\beta\hbar\omega/2)]$ , and a crystal's phonons factorize further into a product over normal modes $k$ .

Derivation AFactorization is equivalent to additivity of every extensive quantity

This article's central structural result is that factorization of $Z$ is not merely a computational trick — it is mathematically equivalent to the additivity of every extensive thermodynamic quantity. The goal is to show that $Z=\prod_a Z_a$ implies $U=\sum_a U_a$ , $F=\sum_a F_a$ , $S=\sum_a S_a$ , and $C_V=\sum_a C_{V,a}$ , so that "the pieces do not interact" and "the thermodynamics adds up" are equivalent statements. The strategy is to take the logarithm once, then differentiate it in the four ways that generate the four potentials, and watch each derivative distribute over the sum.

Step 1. Take the logarithm of the factorized partition function. Since $Z=\prod_a Z_a$ , it follows immediately that $\ln Z=\sum_a\ln Z_a$ . Why: the logarithm of a product is the sum of the logarithms — this single identity is the seed of all the additivity below.

Step 2. Free energy is additive. $F=-k_BT\ln Z=-k_BT\sum_a\ln Z_a=\sum_a F_a$ , with $F_a=-k_BT\ln Z_a$ . Why: $F$ is linear in $\ln Z$ , so it distributes over the sum immediately — the free energy of the whole is the sum of the free energies of the independent parts.

Step 3. Internal energy is additive. $U=-\partial_\beta\ln Z=-\partial_\beta\sum_a\ln Z_a=\sum_a(-\partial_\beta\ln Z_a)=\sum_a U_a$ . Why: differentiation is linear, so $\partial_\beta$ passes through the sum; each term is exactly the energy of one piece, so the total energy is additive.

Step 4. Entropy and heat capacity are additive. Differentiating $U=\sum_a U_a$ with respect to $T$ gives $C_V=\sum_a C_{V,a}$ , and $S=-\partial_TF=-\partial_T\sum_a F_a=\sum_a S_a$ . Why: both $C_V$ and $S$ are further derivatives of the already-additive $U$ and $F$ , so additivity propagates to every extensive quantity — this is why we can add the translational, rotational, and vibrational heat capacities of a molecule. *These three results — $\ln(Z_1Z_2)=\ln Z_1+\ln Z_2$ ; $U=-\partial_\beta\ln Z$ additive; and $C_V$ additive for two independent oscillators — all hold exactly and can be confirmed by direct symbolic computation.*

Go deeper · Derivation B · 2 minThe diatomic heat-capacity staircase from the factorized molecular Z

Derivation BThe diatomic heat-capacity staircase from the factorized molecular Z

The goal is to run the factorized molecular partition function all the way to the observable heat-capacity staircase: compute $Z_1=Z_{\text{tr}}Z_{\text{rot}}Z_{\text{vib}}$ , take the log, and read off $C_V$ in three temperature regimes, obtaining $\tfrac32R\to\tfrac52R\to\tfrac72R$ as rotation and then vibration unfreeze. The strategy is to use the additivity just proved — the total $C_V$ is the sum of the mode contributions — and to evaluate each mode's contribution in the limits set by its characteristic temperature.

Step 1. Split the heat capacity by factorization. Since $Z_1=Z_{\text{tr}}Z_{\text{rot}}Z_{\text{vib}}$ , the molar heat capacity is $C_V=C_{\text{tr}}+C_{\text{rot}}+C_{\text{vib}}$ . Translation always contributes $C_{\text{tr}}=\tfrac32R$ (three quadratic momentum terms, classical at any relevant $T$ ). Why: additivity from factorization lets us treat each motion in isolation and sum, exactly as in the previous derivation.

Step 2. Rotation switches on near its characteristic temperature $\theta_{\text{rot}}=\hbar^2/2Ik_B$ . For $T\ll\theta_{\text{rot}}$ the rotor is frozen ( $C_{\text{rot}}\to0$ ); for $T\gg\theta_{\text{rot}}$ a linear rotor reaches the classical value $C_{\text{rot}}\to R$ (two rotational degrees of freedom, $\tfrac12R$ each). Why: the level spacing $\sim k_B\theta_{\text{rot}}$ sets the temperature at which thermal energy can excite rotations; below it, rotation cannot absorb heat.

Step 3. Vibration switches on near $\theta_{\text{vib}}=\hbar\omega/k_B$ , which is much higher. Its Einstein contribution is $C_{\text{vib}}=R\,x^2e^x/(e^x-1)^2$ with $x=\theta_{\text{vib}}/T$ : this $\to0$ as $x\to\infty$ (frozen) and $\to R$ as $x\to0$ (fully active). Why: a vibration is a single quantum oscillator; because $\theta_{\text{vib}}\gg\theta_{\text{rot}}$ for typical molecules, vibration unfreezes only at much higher temperature, giving a second step.

Step 4. Combine the contributions from each degree of freedom. Cold ( $T\ll\theta_{\text{rot}}$ ): only translation, $C_V=\tfrac32R$ . Warm ( $\theta_{\text{rot}}\ll T\ll\theta_{\text{vib}}$ ): translation $+$ rotation, $C_V=\tfrac52R$ . Hot ( $T\gg\theta_{\text{vib}}$ ): all three, $C_V=\tfrac72R$ . Each step is exactly $+R$ , the classical contribution of the newly-thawed mode. Why: because the modes are independent, unfreezing one simply adds its full classical heat capacity on top of the others — the staircase is additivity made visible. *The three plateaus $\tfrac32R,\tfrac52R,\tfrac72R$ and the $+R$ steps, together with $N_2$ landing on the $\tfrac52R$ plateau at 300 K ( $C_V\approx20.8$ J/mol/K, matching experiment), follow directly from an exact rigid-rotor sum combined with the Einstein vibrational formula.*

Limiting cases

The behaviour of factorization across regimes:

Case	Result	Meaning
Perfectly independent	$Z=\prod_a Z_a$ , $\ln Z=\sum_a\ln Z_a$ exact	All extensive quantities add; the many-body problem decouples completely.
$N$ distinguishable copies	$Z=Z_1^N$	Fixed lattice sites (e.g. spins); no $1/N!$ because sites are labelled by position.
$N$ identical particles (dilute)	$Z=Z_1^N/N!$	Gibbs factor restores extensivity; valid when $n\lambda_T^3\ll1$ .
Single molecule	$Z_1=Z_{\text{tr}}Z_{\text{rot}}Z_{\text{vib}}Z_{\text{el}}$	Independent motions multiply; heat capacity is a staircase.
Weak interaction	$Z\approx Z_0\,\langle e^{-\beta V}\rangle_0$	Factorized $Z_0$ is the backbone; the coupling is a perturbation/cumulant expansion.
Strong / long-range interaction	factorization fails	Cross terms generate correlations between degrees of freedom, leading to genuinely many-body phenomena such as phase transitions, criticality, and collective behaviour.

Common trapJust multiply the partition functions — even when the particles interact

Factorization is exact only when the total energy is genuinely a sum of independent pieces with no cross terms. The instant an interaction potential $V_{ab}$ couples two degrees of freedom, $e^{-\beta(H_a+H_b+V_{ab})}\neq e^{-\beta H_a}e^{-\beta H_b}$ , and $Z$ does not factorize. Blindly writing $Z=\prod_a Z_a$ for an interacting gas or a system with coupled degrees of freedom is a fundamental error — it discards the correlations responsible for phenomena such as phase transitions and condensation. A second, quieter trap: for $N$ identical particles you must decide whether they are distinguishable (fixed lattice sites, $Z=Z_1^N$ ) or indistinguishable (a gas, $Z=Z_1^N/N!$ ). Using $Z_1^N$ for a gas gives a non-extensive entropy and the Gibbs paradox; using $Z_1^N/N!$ for lattice spins wrongly divides by $N!$ and ruins the counting. And the $1/N!$ itself is only the dilute-limit approximation to full quantum indistinguishability — when $n\lambda_T^3\sim1$ you must switch to Bose–Einstein or Fermi–Dirac statistics, where the factorization is over single-particle levels, not particles.

Worked examples

1 · The diatomic heat-capacity staircase, computed by factoring Z (a real, measurable number)

Take a model diatomic gas with rotational temperature $\theta_{\text{rot}}\approx85$ K and vibrational temperature $\theta_{\text{vib}}\approx3374$ K (H $_2$ -like values). Using $Z_1=Z_{\text{tr}}Z_{\text{rot}}Z_{\text{vib}}$ , find the molar $C_V$ at (a) $T=8$ K, (b) $T=300$ K, and (c) $T=40{,}000$ K, and identify the plateau in each case.

Step 1. By factorization, the molar heat capacity is the sum of the independent contributions, $C_V=C_{\text{tr}}+C_{\text{rot}}+C_{\text{vib}}$ , with $C_{\text{tr}}=\tfrac32R$ for all temperatures. Compute each mode from its own partition function (exact rigid-rotor sum for rotation, Einstein formula $C_{\text{vib}}=R\,x^2e^x/(e^x-1)^2$ , $x=\theta_{\text{vib}}/T$ for vibration).

Step 2. Low temperature, $T=8\,{\rm K}\ll\theta_{\rm rot}$ : rotational and vibrational degrees of freedom are frozen, so $C_{\rm rot}\approx0$ and $C_{\rm vib}\approx0$ . So $C_V=\tfrac32R=1.5\times8.314=\mathbf{12.5\ \text{J}\,\cdot\,\text{mol}^{-1}\text{K}^{-1}}$ — the $\tfrac32R$ (translation-only) plateau.

Step 3. Intermediate temperature, $T=300\,{\rm K}\gg\theta_{\rm rot}$ but $T\ll\theta_{\rm vib}$ : rotational modes are fully excited ( $C_{\rm rot}\approx R$ ), whereas vibrational modes remain essentially frozen ( $C_{\rm vib}\approx0.002R$ ). So $C_V\approx\tfrac52R=2.5\times8.314=\mathbf{20.8\ \text{J}\,\cdot\,\text{mol}^{-1}\text{K}^{-1}}$ — the $\tfrac52R$ plateau, exactly where real $N_2$ sits at room temperature.

Step 4. High temperature, $T=40{,}000\,{\rm K}\gg\theta_{\rm vib}$ : vibrational modes are now fully excited, giving $C_{\rm vib}\approx R$ . So $C_V\approx\tfrac72R=3.5\times8.314=\mathbf{29.1\ \text{J}\,\cdot\,\text{mol}^{-1}\text{K}^{-1}}$ — the $\tfrac72R$ plateau. The heat capacity climbed in two steps of exactly $+R$ each — one when rotation thawed, one when vibration thawed — precisely because factorization makes the mode contributions add. The locations of these heat-capacity steps directly reveal the characteristic temperatures $\theta_{\rm rot}$ and $\theta_{\rm vib}$ , providing a thermodynamic measure of the molecular rotational and vibrational energy scales.

2 · Why the ideal-gas entropy needs the Gibbs 1/N! (extensivity level)

Show that factoring the ideal gas as $Z_1^N$ (without $1/N!$ ) gives a non-extensive entropy, while $Z_1^N/N!$ fixes it. Use $Z_1=V/\lambda_T^3$ .

Step 1. Without the Gibbs factor: $\ln Z=N\ln(V/\lambda_T^3)$ , so $F=-Nk_BT\ln(V/\lambda_T^3)$ and $S=-\partial_TF$ contains a term $Nk_B\ln V$ . Now double the system, $N\to2N$ , $V\to2V$ : the entropy per particle picks up $k_B\ln(2V)-k_B\ln V=k_B\ln2\neq0$ . The entropy is not extensive — it changes when you merely scale the system, which is physically absurd.

Step 2. With the Gibbs factor: $\ln Z=N\ln(V/\lambda_T^3)-\ln N!\approx N\ln(V/N\lambda_T^3)+N$ (Stirling). Now the $V$ appears only as the intensive combination $V/N$ . Doubling $N$ and $V$ together leaves $V/N$ unchanged, so the entropy per particle is unchanged — the entropy is extensive.

Step 3. The Gibbs factor $1/N!$ removes the overcounting of indistinguishable particle permutations. Using Stirling's approximation, $\ln N!\simeq N\ln N-N$ , it replaces the non-extensive dependence on $\ln V$ by the intensive combination $\ln(V/N)$ , thereby restoring extensivity. This is the Sackur–Tetrode entropy $S=Nk_B[\ln(V/N\lambda_T^3)+\tfrac52]$ , and it is what resolves the Gibbs paradox. Factorization got us to $Z_1^N$ in one line; indistinguishability is the one correction that factorization alone does not supply — you must put the $1/N!$ in by hand (or, more deeply, get it from quantum statistics).

Origins · 1 minfree energy and how this was discovered

free energy

thermodynamic potential (Helmholtz/Gibbs) · 1882

The Helmholtz free energy $F=U-TS=-k_BT\ln Z$ is the thermodynamic potential natural to a system at fixed temperature, volume, and particle number — and it is the quantity through which factorization delivers its power. Hermann von Helmholtz introduced the "free energy" in 1882 (in Zur Thermodynamik chemischer Vorgänge) as the portion of a system's internal energy available to do work at constant temperature, distinguishing it from the "bound" energy $TS$ that stays locked up as heat. Its statistical-mechanical meaning, given by Gibbs, is the bridge relation $F=-k_BT\ln Z$ . What makes $F$ the right variable for factorization is its additivity: because $\ln Z=\sum_a\ln Z_a$ for independent pieces, $F=\sum_a F_a$ , so the free energy of a composite non-interacting system is literally the sum of its parts' free energies. That additivity is the microscopic content of thermodynamic extensivity. Every equilibrium state minimizes $F$ at fixed $(T,V,N)$ , and its derivatives generate the entropy ( $S=-\partial_TF$ ), the pressure ( $p=-\partial_VF$ ), and the chemical potential ( $\mu=\partial_NF$ ). The free energy is the workhorse potential of chemistry, materials science, and soft matter, and — as the logarithm of a partition function — it reappears wherever a normalizing constant does, from field theory to machine learning.

History

1868

— Boltzmann establishes the exponential energy weighting for a system in a heat bath.

1877

— Boltzmann connects entropy to the multiplicity of microstates, S = k log W.

1882

— Helmholtz introduces the free energy as the work-available part of the internal energy.

1902

— Gibbs's Elementary Principles gives the partition function and the bridge F = −kT ln Z their general form.

1911

— Sackur and Tetrode obtain the absolute entropy of a monatomic gas, forcing the Gibbs 1/N! and Planck's constant into the count.

1912

— Eucken measures the freezing-out of hydrogen's rotational heat capacity, an early confirmation of the mode staircase.

Going further

References. Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 3 · Kardar, Statistical Physics of Particles, Ch. 4 · Reif, Fundamentals of Statistical and Thermal Physics, Ch. 7 · McQuarrie, Statistical Mechanics, Ch. 6–8 (molecular partition functions) · Sethna, Statistical Mechanics, Ch. 6 · Gibbs, Elementary Principles in Statistical Mechanics (1902). Key results in this article: the additivity chain $U=\sum_aU_a$ , $C_V=\sum_aC_{V,a}$ (confirmed symbolically for two oscillators); the diatomic staircase $\tfrac32R\to\tfrac52R\to\tfrac72R$ with $+R$ steps (exact rigid-rotor sum $+$ Einstein vibration); $N_2$ at 300 K on the $\tfrac52R$ plateau ( $C_V\approx20.8$ J·mol $^{-1}$ K $^{-1}$ ); and $Z_N=Z_1^N/N!\Rightarrow$ extensive $F/N$ . The numerical instances 12.5, 20.8, and 29.1 J·mol $^{-1}$ K $^{-1}$ use CODATA values of $R$ , $h$ , and $k_B$ .

Builds on: Canonical Partition Function · Boltzmann Factor · Canonical Ensemble. Leads to: Helmholtz Free Energy from the Partition Function · Thermodynamic Quantities from the Partition Function · Grand Partition Function · Heat Capacity and Fluctuation-Response Relation.

Why does taking the logarithm matter so much — isn't Z the "real" quantity?

The logarithm is what turns factorization into physics. $Z$ itself grows exponentially with system size — for a mole of particles $Z$ is a number like $e^{10^{23}}$ , meaningless to write down. But $\ln Z\sim10^{23}$ is a sensible extensive quantity that scales linearly with the number of particles, and it is exactly what all the thermodynamic potentials are built from ( $F=-k_BT\ln Z$ , $U=-\partial_\beta\ln Z$ , $S=-\partial_TF$ ). When $Z=\prod_a Z_a$ factorizes, the logarithm converts that product into the sum $\ln Z=\sum_a\ln Z_a$ , which is additive — the defining property of extensive quantities. So the reason thermodynamic potentials are logarithms of partition functions is precisely to inherit additivity from the factorization of $Z$ . Working with $Z$ directly, you would be multiplying astronomically large numbers; working with $\ln Z$ , you are adding sensible extensive contributions.

What exactly is a "characteristic temperature" like θ_rot or θ_vib, and why does a mode freeze below it?

A characteristic temperature converts a mode's quantum energy-level spacing into a temperature: $\theta=\Delta\varepsilon/k_B$ , where $\Delta\varepsilon$ is roughly the gap between the mode's lowest energy levels. For rotation $\theta_{\text{rot}}=\hbar^2/2Ik_B$ (a few kelvin, since molecular moments of inertia $I$ are large); for vibration $\theta_{\text{vib}}=\hbar\omega/k_B$ (thousands of kelvin, since bond-stretch frequencies are high). A mode "freezes" when $T\ll\theta$ because the thermal energy $k_BT$ is then far smaller than the gap $\Delta\varepsilon$ needed to reach the first excited level. With no way to absorb energy in units of $\Delta\varepsilon$ , the mode is stuck in its ground state, contributes nothing to the heat capacity, and effectively disappears from the count. As $T$ rises through $\theta$ , thermal fluctuations become large enough to excite the mode, it "thaws," and it adds its classical heat capacity. Because $\theta_{\text{vib}}\gg\theta_{\text{rot}}$ , the modes thaw at well-separated temperatures — which is why you see a staircase rather than a single smooth rise.

Distinguishable versus indistinguishable — how do I know which one applies?

Ask whether the particles have fixed, labelled positions or whether they can wander and swap. If each "particle" sits at a fixed lattice site — spins in a magnet, atoms in a crystal, oscillators pinned to positions — then the sites are labelled by their location, swapping two of them gives a genuinely different (though equivalent-energy) configuration only if you track labels, but physically the particles are distinguishable by where they are, and you use $Z=Z_1^N$ with no $1/N!$ . If instead the particles form a gas, free to move and exchange places, then two configurations that differ only by relabeling identical particles are the same physical state, you have overcounted by $N!$ , and you must divide: $Z=Z_1^N/N!$ . The practical tell: a gas or liquid needs the Gibbs $1/N!$ ; a lattice of fixed sites does not. And remember the $1/N!$ is only the dilute-limit shadow of full quantum indistinguishability — when the gas becomes dense or cold enough that $n\lambda_T^3\sim1$ , even the $1/N!$ is not enough and you must use Bose–Einstein or Fermi–Dirac statistics.

If factorization needs no interactions, how is it ever useful for real (interacting) systems?

In two complementary ways. First, an astonishing number of important systems are genuinely (or very nearly) non-interacting: the ideal gas, the modes of the electromagnetic field (photons), the normal modes of a crystal lattice (phonons), and the internal motions of an isolated molecule are all exactly factorizable, and they cover blackbody radiation, heat capacities of solids and gases, and thermochemistry. Second — and this is the deeper point — even when interactions are present, the factorized non-interacting partition function $Z_0$ is the starting point for the calculation. You write $Z=Z_0\langle e^{-\beta V}\rangle_0$ and expand the interaction $V$ perturbatively; the cluster expansion, the virial expansion, and diagrammatic many-body theory are all systematic corrections to a factorized backbone. So factorization is not just for toy problems: it is the zeroth-order skeleton on which essentially every interacting-system calculation is built. The failure of exact factorization is not a dead end — it is precisely the doorway to the physics of correlations and phase transitions.

Does factorization work the same way in quantum mechanics, with non-commuting operators?

It works whenever the pieces of the Hamiltonian commute, which is exactly the condition for them to describe independent degrees of freedom. If $H=\sum_aH_a$ with $[H_a,H_b]=0$ , then $e^{-\beta\sum_aH_a}=\prod_ae^{-\beta H_a}$ (no Baker–Campbell–Hausdorff cross terms), and because independent degrees of freedom live in a tensor-product Hilbert space, the trace splits into a product of traces: $Z=\prod_a\mathrm{Tr}_a\,e^{-\beta H_a}=\prod_a Z_a$ . This is the quantum version, and it is what justifies factorizing over independent quantum numbers (translational, rotational, vibrational). The subtlety is that if two parts of the Hamiltonian do not commute — because a genuine interaction couples them — the exponential of the sum is no longer the product of exponentials, and factorization fails. So the commutator condition $[H_a,H_b]=0$ is the precise quantum statement of "non-interacting," and it is the exact hypothesis under which $Z$ factorizes.

Why is the free energy the right quantity to say "adds up," rather than the energy or entropy?

All three extensive quantities add for independent pieces — $U=\sum_aU_a$ , $S=\sum_aS_a$ , and $F=\sum_aF_a$ — but the free energy is the most direct consequence of factorization and the one that governs equilibrium. Directness: $F=-k_BT\ln Z$ is linear in $\ln Z$ , so the moment $\ln Z=\sum_a\ln Z_a$ you have $F=\sum_aF_a$ in one step, whereas $U$ and $S$ require differentiating first. Governance: at fixed temperature, volume, and particle number, a system minimizes its free energy, so the additive $F=\sum_aF_a$ tells you that independent subsystems reach equilibrium independently — each part minimizes its own $F_a$ . This is why chemists and physicists reach for the free energy when combining subsystems: it is additive, it is the quantity that decides which way a process runs, and its derivatives generate all the other thermodynamic quantities ( $S=-\partial_TF$ , $p=-\partial_VF$ , $\mu=\partial_NF$ ). Energy and entropy add too, but the free energy is where the additivity and the equilibrium principle meet.