φ is exact at N = 4. Then the Sun gets a vote.
Take a chain of N sites where each site talks only to its neighbors, the simplest coupling
structure there is, and the same second-order recursion the RS cost equation produces on a discrete chain.
Its vibration eigenvalues are λk = 2cos(kπ/(N+1)). At N = 4, and only there, the four
eigenvalues are ±φ and ±1/φ, exactly. That half of this lab is exact linear
algebra, verified here to machine precision; it is not yet a Lean theorem (the proved neighbor is the φ
root lemma fibonacci_char_poly_unique_pos_root in Physics/ThermalFixedPoint.lean,
which is not the path-graph statement). The other half asks whether real stars know about it, tests the
idea against BiSON helioseismology and Kepler rotation periods, and reports what it finds, which is mostly
misses.
Why N = 4 is golden
Nothing golden is assumed. The characteristic polynomial of the 4-site chain factors through x² − 3x + 1, whose roots are φ² and φ−2. The dominant-eigenvalue ratios across N = 2, 3, 4 are then exact algebraic numbers: √2, φ, and φ/√2, with zero free parameters.
Do stars run on chains?
The model: a star with N coupled convective shells equilibrates like AN, and the slow-rotator period envelope scales as 1/λ₁. Assign M dwarfs N=2, K dwarfs N=3, G dwarfs N=4 and the period ratios are parameter-free predictions: √2, φ, φ/√2. The shell assignment has no derivation behind it. It is a guess with consequences, which is what makes it testable.
Mostly no, one maybe
Kepler envelope ratios: M/K ≈ 1.32 vs √2 = 1.414 (7% off), M/G ≈ 1.46 vs φ = 1.618 (10% off), K/G ≈ 1.10 vs φ/√2 = 1.144 (4% off). The BiSON p-mode tests are cleaner and harsher: small-separation ratios sit near 0.97, nowhere near 1/φ = 0.618, and the direct AN overlay on mode frequencies misfits at 68% RMS. The oscillation spectrum does not show the naive φ pattern, and this page says so.
The eigenvalue ladder
Every eigenvalue of the N-site chain for N = 1…8, from the run. Gold squares mark the N = 4 quartet where φ lives. Labeled values (√2 at N = 3, √3 at N = 5, integers) are exact closed forms identified by the script; N = 6 and the top eigenvalues of N ≥ 6 have no such simple form.
Predicted vs observed, no anchors hidden
| Ratio | Prediction (exact) | Observed (Kepler envelope) | Miss |
|---|---|---|---|
| P(M dwarf) / P(K dwarf) | √2 = 1.4142 | ≈ 37/28 = 1.32 | ~7% |
| P(M dwarf) / P(G dwarf) | φ = 1.6180 | ≈ 37/25.4 = 1.46 | ~10% |
| P(K dwarf) / P(G dwarf) | φ/√2 = 1.1441 | ≈ 28/25.4 = 1.10 | ~4% |
| BiSON d₀₂ consecutive ratios | 1/φ = 0.618 (naive) | 0.966–0.987 | fails |
| AN overlay on p-mode ν(n) | eigenvalue spacing | 68% relative RMS | fails |
The observed envelope numbers are read off published Kepler/M67 slow-rotator envelopes, not fitted by us; the BiSON rows come from the bundled Davies et al. 2014 frequency table, computed live by the second script.
Run it yourself
Two files, Python standard library only. The p-mode script reads the bundled BiSON frequency table
(davies2014.txt, 22-year dataset).
$ python3 path_graph_golden_eigenvalues.py N = 4: λ_1 = +1.6180339887 = φ λ_2 = +0.6180339887 = 1/φ λ_3 = -0.6180339887 = -1/φ λ_4 = -1.6180339887 = -φ $ python3 solar_pmode_phi_test.py d₀₂(n=10)/d₀₂(n=9) = 0.976531 (φ⁻¹ = 0.618034, diff = +0.358497)
Split tier, and the split is the point. The N = 4 identity (λ = ±φ,
±1/φ) and the ratio ladder √2, φ, φ/√2 are exact mathematics, verified here
numerically; they are not Lean theorems (no path-graph spectrum theorem exists in the library). The
underlying cost-equation chain (RCL → d'Alembert) is proved in Lean
(Cost/FunctionalEquation.lean, Cost/AczelTheorem.lean), while restricting it to an
N-site chain is a MODEL step. The stellar application is HYPOTHESIS at best: the shell-count assignment
(M=2, K=3, G=4) is underived, the comparisons were not preregistered, and the envelope numbers carry
multi-percent ambiguity. Measured today: the rotation ratios miss by 4–10%, and the BiSON p-mode
tests show no φ signature at all. Falsifier status: the naive p-mode formulation is dead as tested;
the rotation-envelope formulation survives only within its error bars and needs a preregistered test on
cluster data (M67, Praesepe) to be worth more than this page claims.