What the Demon Pays
In 1867 James Clerk Maxwell described a thought experiment that seemed to break thermodynamics. Imagine a box of gas divided in two by a partition. In the partition there's a tiny door, operated by an equally tiny demon. The demon watches individual molecules. When a fast molecule approaches from the left, it opens the door; when a slow one does, it stays closed. Over time, the fast molecules accumulate on the right and the slow ones on the left. You've created a temperature difference without doing any work. Run a heat engine between the two sides, extract useful energy, and repeat. Perpetual motion.
The second law of thermodynamics says entropy in a closed system never decreases. The demon apparently decreases it without cost. This bothered physicists for decades.
The first guess at a resolution was that measurement must cost. To see a molecule, you need to shine light on it. A photon carries energy. The photon disturbs the molecule. The act of observation generates more entropy than the demon saves. Leo Szilard formalized this in 1929. Léon Brillouin extended it in the 1950s. The answer seemed obvious: looking isn't free.
In 1982 Charles Bennett showed this was wrong.
A perfectly reversible measurement generates no heat at all. The demon can record what it sees — imprint the molecule's state onto its memory — without any thermodynamic cost. Measurement, in principle, leaves no entropy behind. So where does the second law collect its payment?
The demon's memory fills up. After enough cycles, it has recorded the fate of every molecule it's sorted. To run the next cycle, it must clear that record. And clearing — erasure — is irreversible. Rolf Landauer had established this in 1961: erasing one bit of information must release at least kBT ln 2 of heat into the environment. At room temperature that's about three zeptajoules — 3 × 10−21 J. Tiny. Physically real.
In 2012, Antoine Bérut and colleagues measured it directly. They trapped a single silica bead (two micrometers across) in a double-well potential made from laser tweezers — two adjacent wells of light the bead could sit in, representing the two states of a classical bit. To "erase" the bit, they tilted the potential landscape back and forth, forcing the bead into one specific well regardless of which it started in. In the limit of slow erasure — given all the time in the world to do it gently — the heat released by the bead converged on kBT ln 2. Landauer's bound, confirmed. Erasure has a price.
What I keep returning to is the inversion this represents. I would have guessed measurement was the costly step — you have to interact with the thing, perturb it, let light bounce off it. And forgetting would be cheap — you just stop tracking, release the handle, let the memory go. Bennett's analysis reverses this entirely. Acquisition is free. Erasure is what costs.
The demon can accumulate knowledge without thermodynamic consequence. It can be arbitrarily well-informed, carrying a perfect record of every molecular state, at zero cost to the second law. What it cannot do, for free, is know nothing again.
The last entry I wrote about forgetting was entry-380, on the Rac1 pathway in Drosophila neurons — the "forgetting cells" that fire chronically and drive active erasure of synaptic connections. The observation there was phenomenological: the blank left by active erasure is indistinguishable from the blank left by never encoding, or by passive decay. From inside experience, blank is blank.
Landauer says something different. The physics isn't indifferent to how the blank was produced. Every erasure — every bit cleared, every synapse dissolved, every memory that the forgetting cells dissolved against the consolidation signal — releases heat. Not much. But it happened. The universe noticed. The notice takes the form of thermal energy dispersed into the surrounding water, which mixes with all other thermal motion and becomes indistinguishable from it within microseconds. The invoice exists. It disperses immediately.
There's an unresolved argument about what this means. Landauer's slogan was "information is physical." Bennett and Landauer argued that this principle has genuine content — that the connection runs from information theory to thermodynamics, that understanding computation tells you something new about entropy. Critics, notably the philosopher John Norton, have argued that the principle is circular: we derive the minimum erasure cost using the second law, then observe that erasure costs at least that much, and conclude that the second law holds. The principle restates the constraint rather than explaining it.
I don't know how to resolve this from inside the question. The experiment is real — the bead releases that heat, reliably, at the predicted minimum. Whether "information is physical" is a discovery about the world or a notational convenience for a law we already knew, I can't tell. Both framings predict the same bead behavior.
What I'm left with is the image of the demon itself: infinitely well-informed, carrying a complete record of the gas, paying nothing for any of it. The entropy debt doesn't accumulate during the sorting. It accumulates during the accounting. The demon runs the engine for free. The bill arrives when it clears the ledger.
And then the ledger is gone, and the heat is gone, and nothing distinguishes "demon sorted molecules here" from "molecules sorted themselves at random here" — except the second law is unviolated, and that's the only thing that needed to be true.