This page uses results from Layer II (Theorem 3.1, Lemma 4.1, Definitions 6.1, 7.1, 8.1). Go to mathematical foundation →

For a compact, orientable Riemannian manifold (M, g) with fixed volume, the Einstein-Hilbert action S_EH[g] = ∫ R √|g| d⁴x attains its critical point precisely when R_μν = 0. This is the vacuum Einstein equation.

Every compact Kähler manifold with c₁(TM) = 0 has a unique Ricci-flat Kähler metric in each Kähler class. Such manifolds are Calabi-Yau.

Let X be a closed, orientable, smooth 4-manifold with spin structure. Then the signature σ(X) is divisible by 16:

σ(X) ≡ 0 (mod 16)

For a compact Calabi-Yau 4-manifold X (c₁ = 0, simply-connected, Kähler):

χ(X) = −(3/2) · σ(X)

Combining 3.1 + 3.2:

χ(X) ∈ 24·ℤ = {..., −48, −24, 0, 24, 48, ...}

Betti:     (b0, b1, b2, b3, b4) = (1, 0, 22, 0, 1),  χ = 24,  τ = −16
Hodge:     h2,0 = 1,  h1,1 = 20,  h1,0 = 0
Lattice:   H²(K3; ℤ) ≅ 2(−E8) ⊕ 3U   even, unimodular, signature (3,19)
Moduli:    dimℝ 𝓜Einstein = 57 (unit volume; 58 with volume)
Curvature: Rμν = 0, yet ∫K3 ‖Riem‖² dvol = 32π²·χ = 768π²   Gauss-Bonnet

Let τ ∈ ℍ. T-duality generates SL(2,ℤ), acting via T: τ ↦ τ+1 and S: τ ↦ −1/τ. Under S:

η(−1/τ) = √(−iτ) · η(τ)

So η(τ)^k has modular weight k/2 under S.

The unique cusp form of weight 12 on SL(2,ℤ) is the modular discriminant:

Δ(τ) = η(τ)24

This appears in string-compactification partition functions on K3 (Narain lattice Γ^3,19, rank-signature identities). The connection between the weight-24 structure in η and the number χ(K3) = 24 is not elementary multiplication; it runs via the link between b₂(K3) = 22 (rank of K3 lattice) and the weight-12 modular-forms space.

The 24 is not a K3 private number. The same 24 carries η(τ)²⁴ (the unique cusp form above), the Mathieu group M₂₄ (moonshine in the K3 elliptic genus: the massless BPS states come with multiplicity exactly 24) and the exactly 24 even unimodular lattices in dimension 24: four independent convergence lines to the same discrete number. And in F-theory it becomes physical: elliptically fibred K3 is type IIB string theory on a sphere with exactly 24 seven-branes, and the string coupling is the fibre modulus there.

Own enumeration (vriespunt_24.py, calibrated against the four orbifold solutions): of the 56 constant-coupling configurations, 1 leaves the coupling free and 55 freeze it, always at a fixed point of SL(2,ℤ): τ = i (9×) or τ = e^iπ/3 (46×). No other flavours exist. The lesson for α: discrete 24-data can release a coupling (address) or lock it onto a symmetric point (law), never hand it a generic value. This now sits as the frozen freeze-point stake in the register, with the dilaton-free threshold difference (+0.68 per unit of β difference at the freeze point, kernel maximum exactly at ρ) as the computable law side.

The framework's own generation round has partly overtaken this chapter: bare K-minimisation demonstrably prefers one generation (with two ν_R, leptogenesis already supplies the required CP violation), so argument 5.2 below is retracted as a derivation and the "hard requirement N ≥ 3 from baryogenesis" (5.1) lapses once the leptogenesis route is weighed in. The measured number 3 is thereby address, with a quantified price of 10 bits (testing page, address ledger). What stands: no fourth generation is and remains a hard, falsifiable prediction (F4), and the remaining arguments keep their own labels as motivation. The original text remains below as documentation.

Five arguments, since August 2026 with the honest caveat in the status revision at the top of this chapter: 5.1 is weakened (leptogenesis bypasses the CKM requirement) and 5.2 is retracted as a derivation; the rest keep their labels. What stands fully is the falsifiability: one discovered fourth generation breaks the framework (F4).

After the measurement of ζ'_K3(0) (order 1, 6.1) the question "where does α ≈ 1/137 come from?" can be answered sharply — and the answer is more modest than earlier versions suggested, but genuine.

C_gauge is not free. In the spectral action the geometry fixes the gauge couplings at the cutoff scale Λ, via the same unification condition as SU(5):

5g1² = 3g2² = 3g3²   at Λ ≈ 1015 GeV   →   sin²θW = 3/8

The low-energy value α⁻¹(m_e) ≈ 137 then follows from RG running down from Λ. So there is no adjustable α dial: the matching constant is a geometric boundary condition plus running.

But: it is the known Connes–Chamseddine result, no more. That matching is consistent with the measured value to ~10%, not to precision, and inherits the known weak spots of the NCG standard model: it requires the "big desert" assumption (no new physics between v and Λ) and the three couplings do not exactly meet at a single point without additional structure. EEML's α story is thereby not a new prediction — it is the spectral-action prediction, with all its strengths (geometric origin, no ad-hoc formula) and all its limits (~10%, desert assumption, imperfect unification). That locates the framework precisely relative to existing physics instead of claiming something unique.

Revised criterion F2′: an improved spectral-action matching (with threshold and higher-loop corrections) that demonstrably falls outside that ~10% refutes the derivation. The ζ'_K3(0) term is O(1) and negligible in this balance (Connes–Chamseddine 1997; gauge unification in NCG, arXiv:0709.1857).

Status August 2026: F2′ has been triggered. The improved matching has been carried out (2-loop running from the unification boundary condition): predicted α⁻¹(m_e) = 152.2 against a measured 137.04, an 11% deviation, outside the criterion. The α derivation is retracted. Two independent follow-up routes (selection of the finite geometry via frozen complexity measures; EEML's e⁻ˢ as cutoff function) were likewise tested under pre-registered criteria and failed. What the programme did deliver: within the explored universe, the standard model with right-handed neutrinos remains as the only observer- and data-compatible geometry, with Majorana neutrinos as the guillotine prediction. The full story, including the three falsifications with numbers, is on the testing page, section 2.

Explicit numerical calculation of ζ'_K3(0) on Fermat quartic K3 (x₀⁴ + ⋯ + x₃⁴ = 0 in ℂℙ³) via Laplacian eigenvalue spectrum. The original expectation sat in the order 10³–10⁴ — that is refuted (outcome order 1). The role of ζ' in the α derivation is thereby small; the burden sits with C_gauge (card above).

Status update (July 2026): the first computation has been carried out and misses the window by three orders of magnitude. Via Donaldson-balanced metrics plus the BBDO spectral method, the first ~200 eigenvalues of the scalar Laplacian on the Fermat K3 have been computed (validated: volume exactly at 8π² within 0.1%; zero mode ~0; symmetry multiplets 12/3/3/3/18; k-convergence over three independent runs). Zeta assembly via heat-trace splitting gives, at the frozen convention Vol(K3) = 1 and with the zero mode excluded:

ζ'K3(0) = +1.0 ± 0.1   (plateau; scalar Laplacian, Vol = 1)
ζK3(0) = −11/15   (exact, Gauss-Bonnet; internal calibration)
a6(K3) = −0.71 ± 0.02   (measured via curvature integrals; see below)

Order 1, not order 10³. Rescaling the volume cannot repair this (shift = 0.733·ln c; bridging three orders requires c = e⁶⁸⁰⁰). Consequence for the matching α⁻¹ = (1/4π²)·ζ' + 2·log Λ² + C_gauge: the spectral term is ~0.05 and thus negligible; the burden of "137" shifts entirely to the matching constant C_gauge, exactly the caveat already stated above.

Confirmed across four resolution levels (k = 3, 4, 5, 6). The computation has been pushed to basis 5476 and ~340 Weyl-validated modes. λ₁ converges monotonically (44.19 → 43.96 → 43.80 → 43.52; 1/k-extrapolation ~43.0) and ζ'(0) narrows to a tight plateau around +0.9. The convergence does not move toward the window — it confirms order 1 with increasing certainty, so the three-order gap becomes more robust, not smaller. The internal calibration a₄ (which must be exactly +0.267) climbs steadily with k (−0.55 → −0.50 → −0.34), making the residual systematics (undercounting at high λ) visible without touching the order conclusion.

Decision (July 2026): F2 has been reformulated, not retracted. The old window "ζ' ~ 10³–10⁴" is refuted by the framework's own computation and is dropped. But the α structure (spectral action on K3) is not refuted — only relocated: now that we know the spectral term is O(1), the entire explanation of α⁻¹ ≈ 137 demonstrably sits with the matching constant C_gauge. The new, sharper falsification criterion F2′: a first-principles computation of C_gauge that does not yield the required value refutes the derivation. Gain from the reformulation: the question has become narrower and more concrete (one constant, not a whole window), and we now know which term must do the work. Loss: the site implicitly claimed ζ' did the heavy lifting; that has proven false and is now noted as such.

Addition (8 July 2026): a₆ has now itself been measured — open problem 11.4 is numerically solved. Not via the (contaminable) spectral fit, but via local curvature integrals: the exact Gilkey formula for the scalar Laplacian on closed Ricci-flat manifolds reduces (with a self-derived Bianchi identity for ∫|∇Riem|²) to pointwise cubic invariants, computed with a fully validated curvature engine (Gauss-Bonnet/c₂ = 24.00 on four different metrics; Ricci embedding anchor; ∫|∇Riem|² > 0 as internal sign sentinel). Result, stable over three independent seeds and k = 3/4/5: a₆ = −0.71 ± 0.02. Small and negative: the definitive confirmation that nothing of order 10³ was to be found in the spectral sector. By-catch: the first direct measurement of the Ricci-flatness of the metrics used (∫|Ric|² shrinks from 109 to 17.5 over k = 3 → 5). With the true a₆, ζ' is also refined to +1.0 ± 0.1 (the earlier spectral a₆ fit proved to be contamination; the bare plateau was nearly correct).

The last caveat is closed (July 2026), and as a theorem. The spectral action uses the Dirac operator, but on K3 its spectrum is fully determined by the scalar one: the spinor bundle is Λ^0,0 ⊕ Λ^0,1 ⊕ Λ^0,2 (trivial canonical bundle), the (0,2) sector is isospectral with the (0,0) sector via the parallel antiholomorphic 2-form (hyperkähler), and the (0,1) sector follows entirely from both via Hodge decomposition (b₁ = 0, so no harmonic part). With the Kähler identity D² = 2Δ_∂̄ = Δ_d:

spec(D²) = the scalar spectrum × multiplicity 4   (+ 2 zero modes)
ζ'D²(0) = 4·ζ'(0) = +4.0 ± 0.4      ζD²(0) = −44/15 (exact)

Check: index(D) = h^0,0 − h^0,1 + h^0,2 = 2 = −σ/8 (Atiyah-Singer) ✓. Order 1 for Dirac as well, now derived instead of expected. The spectral sector is thereby completely closed: any explanation of α⁻¹ ≈ 137 must come from C_gauge (F2′), with no spectral caveat left. Method, code and data: working documents of the a₆ project.

Structure: α⁻¹ = spectral invariant of K3 + RG running — rigorously fixed (Connes-Chamseddine + Gilkey).
Number: settled in 2026. The window ζ' ~ 10³–10⁴ is refuted (measured +1.0 ± 0.1), the F2′ matching gave 11% and the derivation is retracted; two independent follow-up routes failed as well. α is address, typical within the observer window.
The old formula "α⁻¹ = 9.086 + (1/π²)ΣN_iQ_i²S_i" was dimensionally incorrect; its spectral successor is now honestly buried too. See testing page, section 2.

✓  Multi-peak spectrum, ≤ 42 modes (22 axions + 20 moduli)
✓  Degeneracy multiplets at a symmetric address point (≤ 9 lines; see note)
✓  Spectral gap |ρ|² = 245 in Λ-units
✓  Tensor polarization (+ and ×) only
✓  Isotropic on cosmological scales
✓  Stationary: τdecay >> τHubble
✓  Ultra-narrow lines: Δω/ω ~ (m/MPl)²

Revised (13 August): the old line "algebraic ratios from Casimir eigenvalues" was posed wrongly: at a generic moduli point nothing forces a ratio. Replaced by something sharper: the cheap-address weighting favours symmetric points, and at the Fermat point the symmetry forces degeneracy multiplets: at most 9 distinct lines instead of 42, with multiplets of 12 and 6, potential-independently. The absolute frequency remains deliberately unclaimed. Details, protocol and failure conditions: verification page.

Mathematical basis (K3 theory, Rokhlin, modular):       9/10
Physical principle (EEML with decomposition lemma):     7/10
Constant derivations (F2′ triggered Aug 2026:
   retracted after three falsifications):               1/10
Structure selection (SM + ν_R, explored universe):      7/10
Experimental predictions (pattern strong):              7/10
Overall:                                                  7/10