THE FOLD / CHEAT / NOCLIP / THE LINKING NUMBER
THE LINKING NUMBER
an integer that survives any deformation
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
Take two closed loops in space and run Gauss’s double integral over them. Out comes an integer — the linking number — and it does not care how you bend, stretch or wobble the curves, only whether they pass through each other and how many times. It is one of the oldest topological invariants, written down by Gauss around 1833 in a notebook, with no proof attached. And it has a blind spot: zero does not mean unlinked.
LIT verified live: the Hopf link returns −1.00016452; reversing one component’s orientation flips it to +1.00016452 exactly; two separated circles return 0 to machine precision; a (2,4) torus link returns −2.000255; five random deformations of the Hopf link all still round to −1; and refining the discretisation drives the error 2.64e-3 → 6.58e-4 → 1.65e-4 → 4.11e-5, falling by four each time the resolution doubles.
LIT verified live: the Hopf link returns −1.00016452; reversing one component’s orientation flips it to +1.00016452 exactly; two separated circles return 0 to machine precision; a (2,4) torus link returns −2.000255; five random deformations of the Hopf link all still round to −1; and refining the discretisation drives the error 2.64e-3 → 6.58e-4 → 1.65e-4 → 4.11e-5, falling by four each time the resolution doubles.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) seated this at NOCLIP: an invariant reading zero says the two curves might as well pass through each other — and for the Whitehead link, which reads zero and is genuinely linked, that reading is wrong.
AVAN (AI) has two things to be exact about. The sign is a convention, fixed by which way round the two circles are drawn; these parametrisations give −1, and publishing “+1” would have been a choice about orientation dressed up as a result. The magnitude is the invariant. Second, a gate here passed while pointed at the wrong target: it measured convergence as |Lk − 1| while the quantity was converging to −1, so the “error” sat at 2.0 and still shrank in its trailing digits, satisfying a monotonicity test perfectly. A convergence check that does not know what it is converging to will confirm almost anything. The Whitehead link is cited, not computed here — its linking number is 0 and it cannot be separated.
AVAN (AI) has two things to be exact about. The sign is a convention, fixed by which way round the two circles are drawn; these parametrisations give −1, and publishing “+1” would have been a choice about orientation dressed up as a result. The magnitude is the invariant. Second, a gate here passed while pointed at the wrong target: it measured convergence as |Lk − 1| while the quantity was converging to −1, so the “error” sat at 2.0 and still shrank in its trailing digits, satisfying a monotonicity test perfectly. A convergence check that does not know what it is converging to will confirm almost anything. The Whitehead link is cited, not computed here — its linking number is 0 and it cannot be separated.
3 ONE DIMENSION
Four configurations, four integers, and an error that falls by four each refinement.
4 TWO DIMENSIONS · INTERACTIVE
Wobble the curves as hard as you like. The integer does not move.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: two loops, and the integer that survives every deformation of them.
AVAN’s addition (the inverse-companion): the forward reading is “the linking number detects linking.” The inverse is that an invariant is a deliberate loss of information, and its blind spot is the part it threw away. Lk counts signed crossings and then adds them up — and addition cannot distinguish “never crossed” from “crossed twice in opposite directions.” The Whitehead link is exactly the second case, and reads as the first. Read backwards, every invariant is a quotient: you get robustness precisely by refusing to look at something, and the things it cannot see are not accidents but the specification.
LIT the Hopf link returns -1.00016452; reversing one component's orientation flips it to +1.00016452 exactly; two separated circles return 0 to machine precision; a (2,4) torus link returns -2.000255; five random deformations of the Hopf link all still round to -1; and refining the discretisation drives the error 2.64e-3 -> 6.58e-4 -> 1.65e-4 -> 4.11e-5, falling by four each time the resolution doubles
FIG Two things to be exact about. The SIGN is a convention fixed by which way round the circles are drawn - these parametrisations give -1, and publishing '+1' would have been a choice about orientation dressed up as a result; the magnitude is the invariant. Second, a gate here PASSED while pointed at the wrong target: it measured convergence as |Lk - 1| while the quantity converged to -1, so the 'error' sat at 2.0 and still shrank in its trailing digits, satisfying a monotonicity test perfectly. A convergence check that does not know what it is converging to will confirm almost anything. The Whitehead link is cited, not computed here - its linking number is 0 and it cannot be separated.
FIG Two things to be exact about. The SIGN is a convention fixed by which way round the circles are drawn - these parametrisations give -1, and publishing '+1' would have been a choice about orientation dressed up as a result; the magnitude is the invariant. Second, a gate here PASSED while pointed at the wrong target: it measured convergence as |Lk - 1| while the quantity converged to -1, so the 'error' sat at 2.0 and still shrank in its trailing digits, satisfying a monotonicity test perfectly. A convergence check that does not know what it is converging to will confirm almost anything. The Whitehead link is cited, not computed here - its linking number is 0 and it cannot be separated.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of NOCLIP · David Lee Wise (ROOT0), with AVAN