The Magnetic Forest: a two-channel instrument for the integer orchestra, the gas it turns out to be, and the price of a coincidence

We build a two-channel instrument that reads the multiplicative structure of the integers the way an interferometer reads a light source, find that the object it reads is a gas, and use both facts to measure what a near-coincidence between two prime towers costs to see.

The channels are a magnetic graph Laplacian on the tower network of smooth numbers and an open-beam interference screen indexed by the same towers. A character on the tower graph is pure gauge, so the spectrum must not move when characters are swept — a non-result that serves as the instrument's built-in correctness receipt, measured at 2.04e-14. Structure enters only as flux through the graph's faces, where the spectrum moves by 0.725 and shows band splitting at rational flux. On the open channel each prime tower is a comb of resonances, and the Pythagorean comma appears as two teeth 0.111810 apart which a 142-node aperture cannot separate.

Weighting a tower node by n^-sigma is a Boltzmann factor with energy log n, so the forest is the two-species primon gas; its mean energy is Bose-Einstein occupation per species and its partition function is exactly solvable. It has no Hagedorn wall of its own — a two-species gas cannot — while the full integer gas does, appearing as a heat-capacity knee that climbs through 5.99, 9.38, 13.62 and 18.73 as the patch grows through four decades. That same weight is an apodization of the aperture, which makes temperature a resolution parameter.

We derive a closed form for the beam near any near-coincidence costing three exponentials rather than one term per lattice node, validate it against the honest node sum wherever that still runs, and reach a coincidence whose aperture would require summing 1.3e14 nodes. In scaled coordinates every near-coincidence in every prime pair collapses onto one universal curve, and the barrel at which one peak becomes two — Sparrow's 1916 criterion, which has no free parameter — is the first root of one scalar function, solved here to 1.20871040964352282. A census of 993 prime pairs and 14,098 rungs measures 13,978 of them; of the 8,295 deep rungs, not one lands above the solved value. A falsification test at primes eleven digits long, with apertures of ten billion columns, agrees to 6.46e-05.

That constant is the infinite-temperature limit of a curve, and the curve ends: beyond a critical damping no barrel of any length separates the two teeth. The threshold is one half of a turn. So a coincidence is visible only if the gas is hotter than the barrel it demands divided by pi, and every finite temperature cuts the ladder of coincidences at a finite depth — a limit on peak-counting, not an information-theoretic impossibility, and the paper says so.

The census's most approachable find makes the point without apparatus: 13^7 = 62,748,517 and 89^4 = 62,742,241 differ by one part in ten thousand, findable with a pocket calculator, and resolving them into two would take a telescope 115,100 logarithmic units long.

We claim the instrument, the census, and the constant. The criterion is Sparrow's, the floor is Baker's, the bargain distribution is Gauss and Kuzmin's, the gas is Julia's, and the comma's identification with a Baker-floor row is Barbour's. Every figure and number regenerates from a receipt in the accompanying crate, two rejected tests are reported beside their corrections rather than deleted, and the paper ends with four specific ways to break the result.