On twelve sites, the only non-constant harmonic compatible with six-fold rotational symmetry is the folding frequency itself. If the operator chain is a sampling operation — and this paper argues the correspondence is exact rather than analogical — then the hexagon does not merely appear near the fold. It is the fold, in both senses of the word at once.
Four sections were rewritten. The reasons are kept here rather than applied silently, and each is now marked with what was done.
§3 overstated and did not survive its own objection. — APPLIED.
It claimed aliasing is "the normal form of a Whitney A₁ fold". It is not: Whitney
A₁ is x ↦ x², a smooth map with vanishing derivative at an
isolated point, while k ↦ N−k on ℤ/12 is a reflection on a finite
set with no smooth structure and no normal form to be in. §3 now claims the weaker and
true thing — a shared quotient by an involution with isolated fixed points, with
x ↦ x² as the smooth model of ℝ/±. The conflation of the two
sides of the transform is answered in §2 instead.
§4 understated. — APPLIED. The coincidence is sharp, not a
structural fact: at six-fold symmetry the non-DC fixed modes are {6, 12, …} ∩ [1, N/2],
and that set equals {N/2} if and only if N = 12. Verified algebraically and numerically
for N = 6, 12, 18, 24, 30, 36, 42, 48, 60, and since generalised — the same argument at any
k gives N = 2k, kernel-checked in §4's closing box. Fin 12 is the k = 6 instance
of that theorem rather than an arbitrary choice, which is a stronger result than the one
§4 originally stated.
§2 relocated the assumption rather than resolving it. — APPLIED, and the
relocation is now tagged. C as Poincaré sampling of a flow is lossless, so C
cannot be where aliasing happens; F is where the flow ceases to be a diffeomorphism and aliasing
is F's dual-side signature. But that reading assumes PhaseVector is the
twelve-section representation of the LAW3M flow, which nothing in AXLE states. That assumption
now carries its own box and its own tag in §2.
§6 gained a consequence. — APPLIED.
Twelve sections per revolution resolve transverse frequencies to 6 per revolution. The
μmax = −2 contraction is non-oscillatory and cannot alias, but a
Neimark–Sacker bifurcation introduces a second incommensurate frequency — and if it
exceeds 6 per revolution, Fin 12 folds a real frequency onto a false one and the
discrete model fails outright. WP-79 already tags NS as the falsifiable branch. §5 and its
falsifier stand unchanged.
This did not arrive as mathematics. It arrived as a physical observation and a night of bad sleep, and the honest record of that belongs in the paper rather than outside it.
The observation: attempting to arrange neodymium spheres into a spiral, they settle instead into a hexagon. The author reported it in conversation on 2026-08-29, with the note that dynamic analysis and the Fourier transform were what the night had left owing. The direction — that Fourier belongs underneath the operator chain rather than beside it at some rung of an external ladder — is the author's. The correspondence table in §2 and the computation in §4 were produced in response, and the computation is the only part of this paper that has been checked. OPEN
If the chain is a Fourier operation, its four operators should be signal operations, not resemble them:
| Operator | Signal operation | Acting on |
|---|---|---|
| C | sampling — a Poincaré section of the flow | continuous θ onto N transversals |
| K | the gate: a filter | multiplication by a multiplier in the dual |
| F | the fold: loss of injectivity of the return map | a two-to-one identification on the section |
| U | the unfolding: reconstruction | inverse transform from samples |
Two of these are not interpretations. In signal processing a gate is a filter and a fold is what spectra do about the Nyquist frequency; the vocabulary the series has used for three years is the vocabulary of the field it may have been doing all along.
An earlier statement of this table read C as sampling a signal, and that reading fails immediately: Volume I’s Assumption 3 requires C to be bi-Lipschitz with a non-collapse condition — distinct trajectories stay distinguishable — and sampling a signal discards everything above Nyquist. The two are incompatible.
The repair is that the object being sampled is not a signal. With θ̇ = 1, taking twelve equally spaced sections per revolution is stroboscopic sampling of a flow, and the section-to-section map is the time-π/6 flow map restricted to a transversal — a diffeomorphism. Nothing is lost, because what happens between piercings is not arbitrary: it is determined by the vector field, and can be integrated back. C survives Assumption 3 precisely because the object is a flow rather than a signal, which is the reading Volume XIII chapter 2 reaches independently from the categorical side.
Reading C as a Poincaré section removes the Assumption 3 contradiction, but it does not come free: it assumes that PhaseVector := Fin 12 → ℝ is the twelve-section representation of the LAW3M flow — that the twelve slots are transversals of θ̇ = 1 rather than an unrelated discretisation. ASSUME Nothing in AXLE states that. It is a new assumption, smaller than the one it replaces and in a place where it can be checked, but it is not a derivation. OPEN
If C is lossless and invertible then C does not alias — aliasing is the information loss of undersampling, and a diffeomorphism loses nothing. C and F cannot both be doing the work the first version of this table gave them.
The resolution is the one Volume XIII chapter 2 already predicted. C is the section: lossless, invertible, Assumption 3 satisfied — away from the fold. F is where the flow ceases to be a diffeomorphism: rank-1 loss, two-to-one, and this is where injectivity actually dies. Aliasing is not what C does. It is what F looks like in the dual. MODEL
That also answers an objection the first version could not. Whitney’s fold is a statement about the state space; aliasing is an identification on the dual. Calling them the same map conflates the two sides of the transform. Under the reading above they are not the same map: one is the image of the other under the transform, and §5’s computation is already the dual-side statement of the fold’s two-to-one identification — the fold identifies points related by an involution, observables become invariant under it, and the modes invariant under a reflection on ℤ/12 are exactly what §5 computes. ASSUME that the chain is a Fourier operation at all; everything after that is argued rather than posited.
The fold of catastrophe theory — Whitney A₁, the map x ↦ x² — and the fold of sampling theory — spectra reflecting about N/2 — are different technical terms that share an English word. A correspondence built on that coincidence would be wordplay, and this corpus has a standing record of claims that read correct and rest on nothing.
The answer is not that they are the same map, and an earlier draft of this paper claimed that they were. Whitney A₁ is x ↦ x²: a smooth map with vanishing derivative at an isolated point. Aliasing on ℤ/12 is a reflection k ↦ N−k on a finite set — there is no smooth structure on it, no derivative, and therefore no normal form for it to be in. Stated as an identity of normal forms the claim is a category error, and the gap box above is right to have predicted it.
What is true is weaker and sufficient. Both realise the quotient of a space by an involution with isolated fixed points: aliasing quotients ℤ/N by k ∼ N−k, fixing 0 and N/2; the Whitney fold quotients ℝ by x ∼ −x, fixing 0, and x ↦ x² is the smooth model of that quotient. Aliasing is the discrete shadow of the same quotient structure, not an instance of the same normal form. MODEL
The series' phase space is PhaseVector := Fin 12 → ℝ, so N = 12 and the folding
frequency is k = N/2 = 6. Six-fold rotational symmetry acts by rotation through two sites. A mode
k is fixed by that rotation exactly when 2k ≡ 0 (mod 12).
DC, and one other: k = 6, the folding frequency, and nothing else. Verified numerically by projecting a random real vector onto the invariant subspace and taking the real DFT; every mode but 0 and 6 vanishes to machine precision. DATA
Nothing above says the hexagonal mode and the folding frequency coincide in general. With the symmetry held at six-fold they coincide at N = 12 and nowhere else, and that is the sharper statement.
For six-fold symmetry on N sites the rotation is through N/6 sites, so the fixed modes are the k with (N/6)k ≡ 0 (mod N) — that is, the multiples of 6 in [0, N/2]. The non-constant ones are therefore always {6, 12, …}, and that set equals {N/2} exactly when N/2 = 6.
| sites N | 6-fold rotation by | fixed modes | Nyquist N/2 | hexagon = fold? |
|---|---|---|---|---|
| 12 | 2 | {0, 6} | 6 | yes |
| 18 | 3 | {0, 6} | 9 | no |
| 24 | 4 | {0, 6, 12} | 12 | no — 6 also survives |
| 30 | 5 | {0, 6, 12} | 15 | no |
| 36 | 6 | {0, 6, 12, 18} | 18 | no |
| 6 | 1 | {0} | 3 | no |
Verified by the same projection as above at each N, and below by theorem. DATA PhaseVector := Fin 12 → ℝ has looked like an arbitrary choice of resolution; at six-fold symmetry it is the one site count that carries no harmonic content except the fold itself. Any other count either separates the hexagonal mode from the folding frequency, or admits modes besides it.
What the sweep cannot establish is that twelve is the distinguished quantity, because it holds k = 6 throughout. Vary both and the structure appears at every k.
On twelve sites, the hexagonal mode and the aliasing fold are the same mode. A six-fold symmetric arrangement of twelve phase slots carries no harmonic content except at the frequency where the sampling folds. Under §2's correspondence this says the hexagon is not a structure the chain produces at the fold — it is the fold, expressed in the sample domain.
For any k > 0 on an even number of sites, the k-fold symmetric ring keeps exactly one non-constant mode — the folding frequency — if and only if N = 2k.
At k = 6 it returns N = 12, which is what §4 found by sweeping. At k = 10 it returns N = 20. The hexagon on twelve sites and the decagon on twenty are the same fact at two values of one parameter, and the theorem covers every other value at once.
Kernel-checked. Six declarations in GTCT/book4/FoldingFrequency.lean, no sorry, every one resting on [propext, Classical.choice, Quot.sound] and nothing else. Report: tools/verify-audit/2026-09-09/FoldingFrequency.axioms.txt. VERIFIED
Stated exactly, because the gate and the file can count differently: the audit probes theorem and lemma, and here that is six of six — the file’s only other declarations are the definition visibleModes, which carries no proof obligation, and two anonymous examples that instantiate the theorem. Every proof in the file is therefore audited, which is not something the gate guarantees in general.
| declaration | statement |
|---|---|
| visibleModes_eq_nyquist_iff | the general theorem above |
| hexagon_at_twelve | visibleModes 12 6 = {6} |
| decagon_at_twenty | visibleModes 20 10 = {10} |
| eighteen_is_not_the_fold | visibleModes 18 6 = {6} ∧ 18/2 = 9 — one mode survives, but it is not the fold |
| twentyfour_keeps_two | visibleModes 24 6 = {6, 12} |
| forty_keeps_two | visibleModes 40 10 = {10, 20} — the same failure one k up |
A hypothesis dropped rather than carried. The physically motivated condition is k | N — the symmetry should rotate by a whole number of sites. The arithmetic never uses it: only 0 < k and 2 | N appear in the proof, so the stronger hypothesis is not assumed.
What prompted it. On 2 September 2026 a ten-sided wave was reported at Saturn’s south pole (Sánchez-Lavega et al., Science Advances, doi 10.1126/sciadv.aee4251), beside the long-known six-sided wave at the north. Two different k in the sky is a good reason to stop presenting one value of k as special.
An earlier version of §4 headed this result a characterisation of twelve. The arithmetic under that heading stands and the count is right; the word is withdrawn. Twelve is what the theorem returns at k = 6, and twenty is what it returns at k = 10.
What this does not claim, stated because the integers invite the leap. NOT CLAIMED
Nothing above bears on Saturn. The theorem is about which Fourier modes of a discretely sampled ring survive a cyclic symmetry. Saturn’s polar waves are jet-stream wavenumbers in a continuous fluid, selected by barotropic instability of a circumpolar jet — the width and shear of that jet fix k, and that is geophysical fluid dynamics with no established connection to the sampling statement here. The two share the integers 6 and 10 and, so far as this page is concerned, nothing else. The corpus’s reading of the polar waves is WP-100 · The Wavelength, Not the Count; the neighbouring over-generalisation is WP-97 · Thirty Was Doing the Work.
Under the full dihedral group the answer depends on the mirror axis, which the first computation did not anticipate:
| Symmetry imposed | Surviving modes |
|---|---|
| C₆ — rotation by two sites only | {0, 6} |
| D₆ — mirror through a site | {0, 6} |
| D₆ — mirror through a bond | {0} |
Reflect a hexagon about an axis through a vertex and the folding mode survives; reflect about an axis through an edge midpoint and it is annihilated. This is not a technicality of convention. It is the difference between a hexagon centred on a sphere and a hexagon centred on a gap, and it is decidable by hand with the magnets that generated the question. DATA