A Golden-Ratio Torus Model and a Falsifiable Candidate Closed Form for the Inverse Fine-Structure Constant

We present a geometric model based on a single axiom—a harmonic field on a flat two-torus
with a logarithmic golden-ratio “warp”—which reduces exactly to the critical almost-Mathieu
(Harper) operator at frequency 1/φ, a standard and well-studied object whose spectrum is a
multifractal Cantor set. From the model we extract a candidate closed form for the inverse
fine-structure constant,
α
−1 = ∆ · (R/r) · φ
4 −
π
2F9
+
ζ(3) ln φ
π F2
9
= 137.035998710,
where ∆ = 5 is the discriminant of Q(
√
5), φ the golden ratio, F9 = 34, and R/r the torus
aspect ratio. We state at the outset that this is a prediction under test, not a
derivation. The value is fixed (it cannot be tuned) and sits below all current measurements of
α
−1 by roughly 12σ to 45σ depending on the measurement (about 22σ against the CODATA
recommended value); it is therefore falsifiable and currently disfavored. We are explicit about
which ingredients are forced by the construction and which are inputs: in particular, the factor
R/r = 4 is a chosen boundary condition that is not forced by the golden structure, and it is
the factor to which the leading term is most sensitive. We document this weak point precisely
rather than conceal it. The portions of the work that do not depend on the α
−1 prediction—the
operator construction, the noncommutative-torus / real-multiplication correspondence, and a
forced module-determinant invariant equal to 5—stand independently and are presented first.