One sum over every state, and it knows the whole system. Write Z = Σi e−βEi with β = 1/kT; then energy, entropy, free energy, and heat capacity all fall out of ln Z and its derivatives. Rendered, not quoted.
SOURCE J. W. Gibbs, Elementary Principles in Statistical Mechanics (C. Scribner's Sons, 1902) — archive.org/details/elementaryprinc00gibbgoog. AMBER: period edition scan, no canonical DOI; foundational text where Z (the “phase integral”) is defined.
A discrete system with energy levels {Ei}. The demo uses a two-level system, E0=0, E1=ε=1, in natural units (k=1). Every equilibrium quantity is a functional of the single number Z.
The claim Blue defends: Z is a generating function. Each derivative of ln Z is a physical observable, so the whole thermodynamics is read off, never separately postulated.
The sum over states that knows all. Boltzmann gives the weight e−βE of a single state; Gibbs's Z collects them and turns the collection into a lever: −d(ln Z)/dβ is energy, −kT ln Z is free energy, ∂2ln Z/∂β2 is energy variance → heat capacity.
Neighbour sphere: the-boltzmann-distribution → here its normaliser becomes the object of study. Z is the bridge from one-state weights to whole-system thermodynamics — the generating function of statistical mechanics.
Live re-check of the load-bearing identity −d(ln Z)/dβ = Σ Ei pi (energy from the log-derivative equals the direct average). It confirms green, and flips red the instant window 6 tampers with Z.
WITNESS: pending…Also checked live: F = −kT ln Z, S = k(ln Z + β⟨E⟩), and the fluctuation identity C = β2Var(E).
Energy spectrum {0, 1} and a temperature T (drag). From these two inputs alone the machine forms β=1/T and Z.
Proven result: at any T the four thermodynamic potentials above are consistent derivatives of one function. The average energy computed from the log-derivative agrees with the direct state-average to <10−9, and the heat capacity equals β2 times the energy variance.
LIT: booting…“Z is a free parameter you tune.” — False. Z is fixed the moment {Ei} and β are fixed; it is the normaliser, not a knob. The only genuine walls are physical: Z assumes the system is in equilibrium with a heat bath and that the spectrum is known and summable. For real continuous / many-body systems the sum diverges or is intractable — Z is exact but rarely closed-form.
Second wall (AMBER): the classical phase integral needs an h3N/N! factor (Gibbs paradox) to be dimensionless and extensive; the discrete demo sidesteps this and states so.
“Free energy F is just the average energy ⟨E⟩.”
→ F = ⟨E⟩ − TS. They coincide only at T=0; the −TS term is what makes F the quantity minimised at fixed T.
“Entropy needs a separate microscopic count; Z can't give it.”
→ S = k(ln Z + β⟨E⟩) falls straight out of ln Z. No extra postulate.
“Heat capacity is unrelated to fluctuations.”
→ C = β2(⟨E2⟩−⟨E⟩2): response equals fluctuation. Verified live to <10−6.
Disclosed planted void. This drops the exponential and sums the bare energies: Z → Σ Ei. Now Z has no β-dependence, so −d(ln Z)/dβ = 0 ≠ ⟨E⟩ — the mean-energy identity breaks. Watch window 7 catch it.
state: intact