Why temperature is defined as a derivative of energy with respect to entropy — and why it is not simply proportional to the energy per particle except when the energy-level spacing is small.
Put two systems — say two Einstein solids — in thermal contact and isolate the pair, so the total energy is fixed while energy flows freely between them:
Each system's number of microstates depends on how much energy it holds, Ω1(E1) and Ω2(E2). Because the systems are independent, microstates multiply, and so entropies add:
The fundamental postulate is that every microstate is equally probable. The pair therefore spends almost all its time at the energy split with the most microstates — the split that maximizes Ωtotal, equivalently maximizes Stotal. That maximum is thermal equilibrium. Setting the derivative to zero:
At equilibrium the quantity dS/dE is equal for the two bodies — which is precisely the everyday behaviour of temperature (the zeroth law). The quantity decreases as energy rises (adding energy to a hot body raises its entropy less than adding it to a cold one), so identifying it with 1/T rather than T makes heat flow from high to low temperature, as observed:
This is a definition, chosen so the statistical quantity coincides with the thermodynamic T in dS = δQ/T and reproduces the ideal-gas temperature.
With S/k = ln Ω = ln C(q+p−1, q) and energy E = q·ħω (so dE = ħω dq):
using d(ln Γ)/dx = ψ (the digamma function), since ln Ω = lnΓ(q+p) − lnΓ(q+1) − lnΓ(p). With Stirling's approximation this reduces to the familiar closed form:
This is the T_calc in micro_movr3 and the temperature readout in the
1-D and microstate models; the digamma form is just the exact version that skips Stirling.
It is tempting to read temperature as "average energy per particle." That is only true in a limit. The average number of quanta on one oscillator is the Planck/Bose form
The whole behaviour is governed by the dimensionless ratio of the level spacing ħω to the thermal energy kT:
| Regime | Condition | Behaviour of T vs. energy/particle |
|---|---|---|
| spacing "small" (high T) |
ħω ≪ kT ⟨n⟩ ≫ 1, i.e. q ≫ p |
⟨E⟩ ≈ kT: energy per oscillator is proportional to T (equipartition / Dulong–Petit). Here "T ∝ E/N" holds. |
| spacing "large" (low T) |
ħω ≫ kT ⟨n⟩ → 0, i.e. q ≪ p |
⟨E⟩ ≈ ħω·e−ħω/kT: energy is exponentially related to T, not proportional. The relationship is strongly nonlinear and the heat capacity falls toward zero. |
"Small" is never absolute — it is small compared to kT. The proportionality T ∝ (energy per particle) emerges only once the thermal energy kT dwarfs the gap between levels, so that many quanta sit on each oscillator and the discrete ladder looks effectively continuous. The same ħω is "large" at low temperature and "small" at high temperature.
Translational energy levels (a particle in a box) are spaced so finely that ħω ≪ kT at virtually any temperature — so for translation, temperature tracks the average kinetic energy per particle down to very low T. Vibrational levels of a stiff bond have a large ħω, so the proportionality only sets in at high temperature; below that the modes "freeze out." This is exactly why the Einstein solid's heat capacity drops at low T — and why it captures that drop qualitatively, even though it decays faster than the Debye T3 law that real crystals obey.
T = dE/dS is the definition of thermodynamic temperature — the right definition because dS/dE is exactly the quantity that equalizes between two bodies when entropy (microstate count) is maximized. Reading it as energy per particle is a high-temperature convenience that holds only when the level spacing is small relative to kT.