NOTE XXIII  ·  AVENUE C: PHASON-STRAIN MINIMISATION

The Fibonacci Family

Which projection depths n support the 4π lock?   n = 10 → 50 sweep at L=8

Key Finding

The 4π holonomy at the canonical ring (r=5.30, N=32) is not exclusive to n=31. It appears for a discrete family of Fibonacci depths: n∈{11, 13, 15, 17, 18, 23, 31, 34}. The governing variable is σint (mean perpendicular-space norm of ring sites, r=−0.88 with gap), not the Diophantine residual |ε(n)|. Note XXI’s “n=31 unique” result was correct within the tested set {n=31, 36, 41, 46} — but 31 is one member of a larger compatible family. The absolute best lock (gap=0.0021) occurs at n=20 and n=30 at r=5.45.

8/29
n values with gap < 0.05 at r=5.30
0.0021
Best gap (n=20, n=30 at r=5.45)
−0.88
r(σint, gap) — phason governs
n≥39
Frozen: φapprox→φ at float64

Setup

The projection parameter n controls the Fibonacci approximant φapprox = F(n)/F(n−1) used to build the QR-decomposed projection basis. The signed residual ε(n) = φapprox − φ alternates in sign (odd n: positive overestimate; even n: negative underestimate) and shrinks exponentially: |ε(n)| ≈ 1/(√5·F(n−1)²).

From Note XXI (thermodynamic limit), a targeted comparison of n∈{31, 36, 41, 46} at L=8 found n=31 uniquely near −2.0 while n=36, 41, 46 all gave ΔΘ/2π = −2.235. This Note tests the full range n=10→50 to determine whether n=31 is genuinely singular or one member of a family.

Base frames: compute_site_frames(pts_phys, pts_int, k=8)   [phys → internal]
Rotated frames: compute_site_frames(pts_phys, pts_rot, k=8)   [phys → helix-rotated]
Phason strain: σint(n) = mean ‖pts_int‖ for ring sites at best r

Machine-Precision Collapse

The first structural finding is a hard boundary at n ≈ 39. For n≥39, the ratio F(n)/F(n−1) equals φ to all 15 significant digits available in IEEE 754 double precision. All n≥39 produce identical projections, identical 521-site physical point clouds, and identical holonomies (ΔΘ/2π = −1.967 at r=5.55, gap=0.033).

Precision ceiling

The effective dynamic range of the projection is n=10…38 (29 distinct values). Any n≥39 is a repeated experiment. Claims about convergence of the holonomy as n→∞ cannot be tested with float64 arithmetic beyond this threshold.

The r=5.30 Family

At the canonical ring r=5.30 (N=32, Iaz=1.96), only a subset of n values produce near-integer holonomy. The full sweep reveals two sharply distinct populations:

Population A — The Family (gap < 0.06 at r=5.30)

  • n=11 ΔΘ/2π = −1.9738, gap = 0.026
  • n=13 ΔΘ/2π = −1.9439, gap = 0.056
  • n=15 ΔΘ/2π = −1.9593, gap = 0.041
  • n=17 ΔΘ/2π = −1.9790, gap = 0.021
  • n=18 ΔΘ/2π = −1.9865, gap = 0.014
  • n=23 ΔΘ/2π = −1.9775, gap = 0.023
  • n=31 ΔΘ/2π = −1.9865, gap = 0.014
  • n=34 ΔΘ/2π = −1.9825, gap = 0.018

Population B — Incompatible (gap > 0.4 at r=5.30)

  • n=12 gap = 0.417
  • n=14 gap = 0.449
  • n=16 gap = 0.475
  • n=19 gap = 0.459
  • n=21 gap = 0.475
  • n=22 gap = 0.486
  • n=36 gap = 0.477
  • n=41+ gap = 0.460 (frozen)

The gap is bimodal: values cluster near 0.02 (family) or near 0.46 (incompatible). There is almost no middle ground. This is a discrete topological switch, not a gradual convergence.

n=31 and n=18 Are Identical

A sharp result: n=31 and n=18 give exactly the same holonomy at r=5.30 (−1.9865, gap=0.0135), the same ring count (N=32), and the same azimuthal imbalance (I=1.96). Their Diophantine residuals differ by nine orders of magnitude (ε(18)=−1.75×10−7 vs ε(31)=+6.46×10−13), yet the holonomy is identical.

Diophantine residual does not govern holonomy

If the quality of the Fibonacci approximation determined the holonomy, n=31 (much better approximant) should differ from n=18. It does not. The holonomy depends on a property that saturates early in the Fibonacci sequence.

Phason Strain Is the Governing Variable

The phason strain σint(n) is the mean Euclidean norm of the perpendicular-space coordinates (pts_int = lat @ perp.T) for ring sites at the best radius. It measures how deeply the ring sites are embedded in the acceptance window — a direct proxy for their phason displacement from the ideal quasiperiodic tiling.

+0.81
Pearson r(|ε|, best_gap)
Larger Diophantine residual correlates with larger gap — but this is almost entirely driven by the small-n, large-|ε| regime (n=10…14). The signal reverses at n=18–31.
−0.88
Pearson r(σint, best_gap)
Higher phason strain (ring sites deeper in acceptance window) strongly predicts smaller holonomy gap. This is the most predictive single variable found.
+0.81
Pearson r(|ε|, canon530_gap)
Same correlation at fixed r=5.30 — also dominated by small-n regime. Within n=15–38, essentially no correlation.
−0.59
Pearson r(σint, canon530_gap)
Moderate correlation at fixed r=5.30 — the family/non-family split is better captured by the ring population (best-r) than a fixed probe.

The Absolute Best: n=20 and n=30

The smallest gap found in the entire sweep is gap=0.0021, occurring at both n=20 and n=30, at r=5.45 (Nring=34, Iaz=2.56). This is more than 6× smaller than n=31’s gap at r=5.30 (0.0135).

n=20 and n=30 fall at r=5.45 — a slightly different ring from n=31’s canonical r=5.30. The 34-site ring at r=5.45 has higher azimuthal imbalance (I=2.56 vs 1.96 for the 32-site ring). From Note XVII, I≥2.0 flags potential artefacts. These near-exact results at r=5.45 should be treated as candidates pending azimuthal-balance verification.

n=31 is not the globally optimal depth

At its own canonical ring (r=5.30, I=1.96), n=31 gives gap=0.0135 — a well-balanced, confirmed result. n=20 and n=30 achieve smaller gaps at r=5.45 but on a ring with I=2.56, borderline for artefact classification. n=31’s strength is its combination of small gap AND low azimuthal imbalance.

The Full Sweep Table

n ε(n) rbest ΔΘ/2π gap Nring Iaz σint r=5.30 gap
10−3.9e−44.6−1.5660.434161.910.6920.476
11+1.5e−45.3−1.9740.026321.960.7940.026
12−5.6e−55.55−2.0510.051382.730.7730.417
13+2.2e−55.25−1.9930.008362.210.7900.056
14−8.2e−65.55−2.1270.127382.730.7730.449
15+3.1e−65.25−2.0180.018362.210.7900.041
16−1.2e−65.55−1.9300.070382.730.7730.475
17+4.6e−75.5−2.0070.007322.410.7830.021
18−1.8e−75.3−1.9870.013321.960.7940.014
19+6.7e−85.55−2.0540.054382.730.7730.459
20−2.6e−85.45−1.9980.002 ✓342.560.7890.039
21+9.8e−95.55−1.9300.070382.730.7730.475
22−3.7e−95.55−2.0060.006382.730.7730.486
23+1.4e−95.3−1.9780.023321.960.7940.023
24−5.4e−105.55−2.0540.054382.730.7730.459
25–295.55>0.06382.730.773>0.4
30−1.7e−125.45−1.9980.002 ✓342.560.7890.039
31+6.5e−135.3−1.9870.014321.960.7940.014
32−2.5e−135.55−1.9330.067382.730.7730.487
33+9.4e−145.55−1.9690.031382.730.7730.501
34−3.6e−145.3−1.9820.018321.960.7940.018
35+1.4e−145.55−2.0600.060382.730.7730.450
36−5.0e−155.55−1.8570.143382.730.7730.477
37+2.0e−155.25−2.0170.017362.210.7900.120
38−1.0e−155.55−2.0540.054382.730.7730.459
39–50ε=05.55−1.9670.033382.730.7730.460

Blue rows: Population A (family, near-2.0 at r=5.30). Green rows: n=20,30 (absolute best gap=0.002 at r=5.45). Greyed rows: machine-precision frozen (n≥39 all identical).

Revision to Note XXI

Correction — “n=31 unique”

Note XXI tested n∈{31, 36, 41, 46}. n=36, 41, 46 all gave ΔΘ/2π=−2.235 at the r=5.30 probe. Conclusion: “n=31 uniquely near −2.0.”

This Note shows that conclusion was accidentally correct within the tested set, but incomplete. n=36 belongs to Population B (incompatible at r=5.30); n=41 and n=46 are both machine-frozen (identical to n=39). The correct statement is: n=31 belongs to a Fibonacci family {11, 13, 15, 17, 18, 23, 31, 34} that supports 4π holonomy at r=5.30 with N=32, I=1.96.

Open Questions

1. What determines family membership?

The family {11, 13, 15, 17, 18, 23, 31, 34} does not follow a simple arithmetic or parity rule. The non-family members that achieve near-2.0 at other radii (n=13, 17, 20, 22, 30, 37) complicate a clean geometric derivation. The σint correlation (−0.88) points to phason-space geometry as the mechanism, but the specific alignment condition remains to be derived.

2. Is the family structure Fibonacci-periodic?

The gaps in the sequence 11, 13, 15, 17, 18, 23, 31, 34 hint at a sub-Fibonacci recurrence. Note that 18−17=1, 23−18=5, 31−23=8, 34−31=3. The differences {1, 5, 8, 3} are themselves Fibonacci-adjacent numbers. A modular arithmetic analysis of F(n) mod some resonance period may reveal a cleaner pattern.

3. Best lock at r=5.45, n=20 and n=30

The 34-site ring at r=5.45 achieves gap=0.0021 at both n=20 and n=30, with Iaz=2.56. Since I>2.0 is the artefact threshold (Note XVII), this result requires dedicated azimuthal verification before being classified as a confirmed detection.

Method

for n in range(10, 51):
    par, perp, meta = build_projection_bases(n)   # φ_approx = F(n)/F(n-1)
    lat = generate_hyperlattice(L=8)
    pts_phys    = lat @ par.T                      # physical 2D coordinates
    pts_int     = lat @ perp.T                     # perpendicular-space (phason) coords
    acc = acceptance_mask(pts_int, golden_cantor)
    pts_rot, _  = apply_boundary_helix(pts_phys, pts_int, ..., α=0.18, λ=0.145)
    frames_base = compute_site_frames(pts_phys, pts_int, k=8)   # phys→internal
    frames_rot  = compute_site_frames(pts_phys, pts_rot,  k=8)  # phys→rotated
    for r in np.arange(4.5, 6.05, 0.05):
        ΔΘ/2π = ring_holonomy(frames_base, frames_rot, center, r, dr)
    # Select r with min |ΔΘ/2π + 2|