Toss counters into boxes at random ── no rule imposed, and yet an exponential distribution rises by itself
Episode 1 said temperature is energy. That isn't enough ── a bathtub and a cup of water at 40 °C are at the same temperature but hold wildly different energies. Temperature is not a total. So what is it? The slope of the sharing ── and this is not an armchair claim; you can watch it stand up in front of you. Line up a lot of boxes and let counters (the smallest units of energy) be passed around at random. No rule, no exponential, no Boltzmann is put in ── and within seconds an exponential distribution appears. Its slope is the temperature. Following that experiment through also shows why the true identity of temperature is \(\beta=1/k_BT\), not \(T\).
What you need:
That's all. Energy is conserved (the number of counters never changes). Nothing else is imposed. Not "temperature," not "equilibrium," not "entropy" ── none of them are written anywhere yet.
The figure below is that experiment. Above: the contents of the boxes (bar height = number of counters). Below: the histogram (how many boxes hold \(n\) counters).
Start with "put them all in one box." The histogram is a mess at first, but as counters get passed around ── it settles into a straight exponential (switch the vertical axis to logarithmic and it becomes a line). Nobody specified an exponential. And when you move the mean-energy slider, only the slope of that line changes.
This is the moment temperature is born. The individual exchanges are random, and yet the whole settles into a definite shape ── a shape with exactly one parameter: the slope.
The reveal is remarkably simple. Just count the number of ways of distributing the counters.
Split the system in two ── a small part you care about (one box) and everything else (the bath). The total energy \(E_{\text{tot}}\) is fixed. If the small part holds energy \(E\), the probability of that is proportional to the number of ways for the rest:
$$P(E)\ \propto\ \Omega_{\text{bath}}(E_{\text{tot}}-E)$$"The more energy the small part takes, the fewer counters remain for the rest, and the fewer ways the rest has" ── so large \(E\) is unlikely. Obvious so far. But how does it fall off?
\(E\) is small compared with the bath, so expand \(\ln\Omega_{\text{bath}}\) to first order in \(E\):
$$\ln\Omega_{\text{bath}}(E_{\text{tot}}-E)\approx \ln\Omega_{\text{bath}}(E_{\text{tot}})-E\cdot\underbrace{\frac{d\ln\Omega_{\text{bath}}}{dE}}_{\textstyle \equiv\ \beta}$$Exponentiate again:
$$P(E)\ \propto\ e^{-\beta E}$$That's it. The exponential appeared because we truncated a Taylor series at first order ── that is, because the bath is large. The Boltzmann factor is the logarithm of a count of ways, linearised. Nothing more, nothing less.
And the \(\beta\) defined along the way is the true identity of temperature.
In words ── "if you gain one unit of energy, by what factor does the number of ways multiply?"
A system whose count of ways explodes when it gains energy (i.e. one that wants more energy) has large \(\beta\) = it is cold. One that barely gains has small \(\beta\) = it is hot. Put two in contact and energy flows toward the larger \(\beta\) (the colder one), stopping when the two \(\beta\)s are equal ── that is thermal equilibrium.
Because \(\beta=1/k_BT\) is "one over temperature," it usually gets treated as a supporting player. But look at the definition: \(\beta\) was born first. \(T\) is merely defined afterwards as its reciprocal.
Make \(\beta\) the lead and a lot of things become straightforward.
| Situation | β | T | Seen through \(\beta\) |
|---|---|---|---|
| near absolute zero | +∞ | 0 | an endpoint; infinite eagerness for energy |
| ordinarily cold | large positive | small | continuous |
| ordinarily hot | small positive | large | continuous |
| infinite temperature | 0 | ∞ | just zero. Nothing special; a waypoint |
| negative temperature (Ep. 5) | negative | negative | simply past zero. Continuously connected |
Written in \(T\) there is an awkward "infinity" sitting in the middle; written in \(\beta\) that place is just \(0\). So Episode 5's negative temperatures hold no mystery from \(\beta\)'s point of view ── the sign of a slope changed. The apparent \(\pm\infty\) wall in temperature is an artifact of writing the reciprocal.
The ratio of occupancies of the upper and lower level is the Boltzmann factor itself:
$$\frac{n_{\text{upper}}}{n_{\text{lower}}}=e^{-\Delta E/k_BT}$$Episode 1's "only \(E/k_BT\) matters" plugs straight in ── if \(\Delta E\ll k_BT\) the ratio is 1 (both equally filled); if \(\Delta E\gg k_BT\) it is essentially 0 (only the lower is filled).
For instance at room temperature (25 meV) with \(\Delta E=1\) eV, \(e^{-40}\approx4\times10^{-18}\) ── a few in ten quintillion. Which is why ordinary matter sits almost entirely in its ground state. What happens as that ratio approaches 1, and what happens when it exceeds 1, is Episode 5.
By now we can say what a thermometer is. It does not measure "hotness" ── it is a device that touches something and waits until the two \(\beta\)s agree. Read your own state afterwards and you know the other's \(\beta\).
So a thermometer needs two things. (1) To be much smaller than what it measures (measuring must not change it). (2) To wait until \(\beta\) equalises. When the second fails ── rapid changes, or systems far from equilibrium ── the very quantity "temperature" ceases to be definable. Temperature is a word for equilibrium.
Established: that the Boltzmann factor \(P(E)\propto e^{-\beta E}\) follows from expanding the logarithm of the bath's state count to first order; that \(\beta=d\ln\Omega/dE=1/k_BT\), i.e. \(1/T=\partial S/\partial E\) with \(S=k_B\ln\Omega\), is the statistical-mechanical definition of temperature; that thermal equilibrium is the equality of \(\beta\) between systems in contact; the two-level occupancy ratio \(e^{-\Delta E/k_BT}\); and that the model in the figure (indistinguishable quanta distributed among distinguishable boxes) has a geometric equilibrium distribution \(p(n)\propto(q/(1+q))^n\) as \(N\to\infty\) ── all standard statistical mechanics.
What is assumed: (1) Truncating at first order is valid only when the bath is far larger than the system of interest. For small baths the higher-order terms matter and the distribution departs from exponential (this is the subject of Episode 4). (2) "Random sharing must reach equilibrium" holds in this model but is in general the assumption of ergodicity. Systems where it fails do exist (integrable systems, glasses, many-body localisation). (3) The counter-passing in the figure does not model real molecular collisions; it is a minimal model to show that a distribution is fixed by nothing but a conservation law plus random exchange. (4) Temperature is defined only in equilibrium (or local equilibrium); far from equilibrium, "the temperature of that system" generally does not exist.
Toss counters into boxes at random and an exponential distribution rises by itself, though nobody specified an exponential. The reveal is a count of ways: expand the bath's \(\ln\Omega\) to first order and \(P(E)\propto e^{-\beta E}\) follows ── the Boltzmann factor is the first term of a Taylor series.
The \(\beta=d\ln\Omega/dE=1/k_BT\) defined there is the true identity of temperature. In words: "gain one unit of energy, and by what factor does the count of ways multiply?" ── the energy derivative of the \(S=k_B\ln\Omega\) from Episode 1 of the sister series "Renormalization That Clicks." So Episode 1's "temperature = the exchange rate between energy and bits" was literal.
Make \(\beta\) the lead and infinite temperature becomes merely \(\beta=0\), with negative temperature naturally continuing on the far side. The apparent wall in temperature is an artifact of writing a reciprocal.
Print / make a PDF: ⌘+P (Ctrl+P on Windows). On screen the counters keep moving and the histogram settles into an exponential. "Put them all in one box" resets; switching to a logarithmic axis turns it into a straight line. "See the answer" opens each solution.