A thin-wall instanton that gave exactly one

The third correction is the archive’s most theatrical failure: a suppression that was supposed to be ten to the minus sixty-four, a wall that leaks essentially never, computed by a naive estimate as exactly one third. The estimate is called the thin-wall instanton, the error is three factors cancelling, and the repair is a good story about why dimension counts and action integrals live in different worlds.

The estimate and its cancellation

The boundary between branches was modelled as a domain wall: a thin sheet with surface tension sigma and thickness l. An instanton is the tunnelling event that flips the wall, and its action, the thing whose exponential suppresses the rate, is, in the thin-wall approximation, the tension times the volume of the flipped region. For a wall of thickness l, flipping a patch of size R: sigma times R squared times l.

The naive draft plugged in the framework’s own numbers. The tension sigma is about the square of the mass scale M, and the thickness l is about one over M, where M is the mass gap from the substrate’s condensation, about 10 to the 12 GeV. Multiply: M squared, times one over M, times R squared, and the M factors cancel, leaving an action of R squared, and with R set to the scale where the calculation closes, the action came out to one. An instanton action of one gives a suppression of e to the minus one: about 0.37. The wall, supposedly near-perfect, leaked a third of the time. The printed draft said the estimate fails by two orders of magnitude in the exponent, and understated the case: an action of one against the required 147 is a failure by a factor of 147 in the exponent, which is 64 orders of magnitude in the probability.

The disease has a name: dimensional cancellation. Every quantity in the estimate was built from the same scale, so the scale cancelled, and the answer lost all dependence on the physics that was supposed to protect the wall. The number one was not a calculation of anything. It was the sound of a formula with no physics left in it.

The two repairs

The first repair kept the error-correction frame and changed the geometry: leakage is not a wall flipping locally, it is a defect string traversing the code distance, so the action is the distance, 64, times a logarithmic factor, about 147. The repair fixed the size of the action and, importantly, fixed its dependence: the action now scales with the boundary’s information structure instead of cancelling it. This version survived for two paper generations before its own audit, which found the logarithmic ratio reverse-engineered from the target. It survives today only as a labelled proxy.

The second repair replaced the mechanism: leakage as theta-vacuum mixing, the standard Yang-Mills instanton with the coupling pinned by the boundary category’s central charge. That action is 4 pi squared times 7, about 276, and its match to the independently required suppression is the subject of its own post in the walkthroughs. The progression, one to 147 to 276, is the archive in miniature: a cancellation bug, a proxy that fixed the size but not the provenance, and a mechanism whose inputs predate the target.

Why the failure was useful

The thin-wall disaster is the programme’s clearest demonstration that a dimensionally consistent estimate can be physically empty. The formula was not wrong; it was vacuous, and vacuity is harder to catch than falsehood because every check short of evaluating the physics signs off. The audit question that catches this species is always the same: does the answer depend on the quantities it claims to be about? Here the answer was no, the action did not care about the protection, and any formula with that property is decoration regardless of how respectable its derivation looks. The next correction is the arithmetic cousin of this one: a probability that left its interval.

The next entry is the same species of error in a probability rather than a suppression: a leakage probability that exceeded one.