An imaginary crowd of copies
There is a tension we have been quietly ignoring. We want the *average* behaviour of a system, but any real system is just one system, doing one thing at a time. How do you average over something there is only one of? Gibbs' beautiful trick was to imagine not one system but an enormous crowd of identical copies, all prepared under the same conditions, each frozen in its own microstate at one instant. This imagined crowd is an ensemble. Instead of watching one system over a long time, you take a snapshot across the whole crowd and average that. It is a thought experiment, but a wonderfully clarifying one.
The canonical ensemble: a system in a warm bath
Which crowd you imagine depends on what your system is allowed to do. The most useful choice for chemistry is the canonical ensemble: a system held at a fixed temperature by sitting in a vast warm bath, free to trade energy back and forth across the wall but with a fixed number of particles. This is exactly the situation of a beaker on a bench in a temperature-controlled room. Because energy can flow in and out, the system's energy is not pinned to a single value — it wanders, settling into the Boltzmann spread we have been building toward all along.
There is also a related quantity worth meeting: the density of states, which simply asks 'how thickly are the energy rungs packed near a given height?' In a big system the rungs become so closely spaced that they smear into a near-continuous band, and knowing how densely they crowd at each energy is enough to do all the counting. Different ensembles are different lenses on the same reality; for a large system they all agree on the answer, so you simply pick whichever makes the arithmetic kindest.
The master recipe, start to finish
Now we can lay out the whole machine of deriving thermodynamics from statistics as a single workflow. Every result in classical thermodynamics, which earlier generations had to wring out of careful experiments, drops out of these few steps once you know what the molecules are like:
- List the energy ladder. Work out what energies one molecule is allowed to have — its translation, rotation, vibration, and electronic rungs — from quantum theory or measurement.
- Build the partition function. Sum each rung's reachability against kT to get the molecular partition function, then combine the molecules into one for the whole system.
- Turn the crank. From the partition function and how it changes with temperature and volume, read off the internal energy, the entropy, the pressure, and the Helmholtz free energy.
- Compare with the bench. Check the predicted heat capacities, equilibrium constants, and entropies against real measurements — they match, often to several decimal places.
The bridge to free energy is especially elegant: the Helmholtz free energy is essentially just the logarithm of the partition function, scaled by kT. Since free energy is the quantity that decides which way a reaction runs at constant temperature, this one link means that knowing the partition function lets you predict the position of a chemical equilibrium from the molecules up. That is the entire promise of statistical thermodynamics, delivered.
Equilibrium trembles, and that is not a flaw
We end with a confession the averages hid from you. Because energy keeps trading across the wall of the bath, the system's energy never sits perfectly still — it jitters a hair above and a hair below its average, forever. These tiny, ceaseless wobbles are fluctuations. In a teaspoon of water they are so vanishingly small next to the average that no ordinary instrument could ever notice them, which is precisely why the macrostate looks rock-steady. Equilibrium is not a frozen stillness; it is a furious, balanced churn whose average happens to hold dead level.