The Szilard Engine

In 1929 Leo Szilard did something wonderful to Maxwell's demon: he put it on a diet. Maxwell's original demon juggled countless molecules at a trapdoor; Szilard stripped the thought experiment down to the absolute minimum — one molecule, one box, one bit — and in doing so built the simplest machine in all of physics that turns knowledge into work. The exchange rate it reveals is precisely the one the previous lesson promised: one bit of information is worth kT \ln 2 of extracted work. And the trouble it causes — a machine that apparently turns heat straight back into work — is the crisis that Landauer's principle will be needed to resolve.

The cycle, move by move

The apparatus: a box of volume V containing a single gas molecule, kept at temperature T by contact with its surroundings (a heat bath). One full cycle has five moves — step through them below:

At the end of the cycle the box is exactly where it started, ready to go again. The engine can loop forever, and every lap it hands you kT \ln 2 of work. Keep an eye on the one thing that is not back to its starting state — the memory. That loose thread is the whole plot of the next two lessons.

Where kT ln 2 comes from

The work comes from move 4, a slow isothermal expansion. A single molecule bouncing off the walls exerts a pressure that obeys the one-molecule ideal gas law pV = kT (the usual NkT with N = 1). As the piston retreats and the molecule's territory grows from V/2 to V, the work done on the piston is

W \;=\; \int_{V/2}^{V} p \, \mathrm{d}V \;=\; \int_{V/2}^{V} \frac{kT}{V'} \, \mathrm{d}V' \;=\; kT \,\ln\!\frac{V}{V/2} \;=\; kT \ln 2.

At T = 300\ \mathrm{K} that is about 2.9 \times 10^{-21}\ \mathrm{J} \approx 2.9\ \mathrm{zJ} per cycle — the very number from the last lesson, now earned as honest mechanical work. Where does the energy come from? Not from the molecule: an isothermal process keeps its average kinetic energy fixed. Every joule delivered to the piston is replaced by heat flowing in from the surroundings (\Delta U = 0, so Q = W). The engine is a machine for converting the random jiggling of the environment into directed work — one bit's worth at a time. And note the general rule the integral hands us: expanding from a fraction 1/m of the box to the whole box yields kT \ln m — the work scales with the information, not the energy, of the confinement.

The scandal

Now compare this with every legitimate engine you know. A heat engine needs two temperatures: it takes heat from a hot reservoir, dumps some into a cold one, and keeps the difference — the second law caps its efficiency below 100%. Szilard's engine has one temperature. It draws heat kT \ln 2 from a single bath and converts all of it into work, cycle after cycle. Run it a billion times and you have a machine that powers itself by cooling the room — a perpetual motion machine of the second kind, exactly what the second law forbids.

Unless. The only unusual ingredient in the cycle is move 3: the measurement, where the experimenter gains one bit. If the second law is to survive, then somewhere in the life cycle of that bit — acquiring it, storing it, or clearing it away — there must be a compensating thermodynamic cost of at least kT \ln 2. Information itself must have a price. Where exactly the price is paid took physics fifty years to pin down, and the answer (it is not the measurement…) is the business of Landauer's principle and Bennett's exorcism of the demon — the next two lessons. For now, savour the cliffhanger: at the end of each cycle the box has forgotten everything, but the memory has not.

It sounds absurd — a piston is 10^{23} times heavier than a molecule. But pressure is just momentum delivered by collisions, and a molecule at room temperature moving at hundreds of metres per second hammers the piston millions of times per second in a small box. Each tap is minuscule; their average is the steady pressure p = kT/V. The catch is the word average: with one molecule the fluctuations are enormous — sometimes the piston lurches, sometimes it stalls, and kT \ln 2 is only the mean work per cycle over many runs. This is thermodynamics at the scale where its quantities flicker, which is exactly why Szilard's box became the favourite laboratory animal of modern stochastic thermodynamics — and why, since the 2010s, real experiments with colloidal beads and single electrons have run genuine Szilard cycles and measured the kT \ln 2. The engine is no longer a thought experiment. It has a lab bench. The man himself — Leo Szilard — went on to have arguably the most consequential "hold my coffee" career in physics; his page is worth the detour.

A common short-circuit: "work from nothing — that violates conservation of energy!" It does not. Energy is scrupulously conserved at every step: the work delivered by the piston is exactly balanced by heat Q = W flowing in from the surroundings; the first law is untouched. What the engine threatens is the second law: it converts disordered heat from a single reservoir entirely into ordered work, with no cold sink and no other change — the one transaction thermodynamics forbids. Keep the two laws separate in your head: the first law does bookkeeping on energy, the second on entropy, and Szilard's engine is a precision instrument for probing only the second. (You'll meet people online "refuting" Landauer by checking energy conservation. Now you know why that misses the point.)