angle of view: physicist

Loss of covariance, τ-geometry, four forces

For people who enjoy physical motivation, not just the rhetoric around it.

The starting tension

Relativity insists that no observer frame is privileged. Quantum mechanics, meanwhile, treats time as a parameter — not an operator — and thereby silently selects preferred frames when it comes to collapse, measurement, energy. That's not just a philosophical stumble; it's the point where the two formalisms cannot be physically merged without loss of information.

τ-geometry takes this tension seriously. Instead of a compromise it proposes a shift: time is complex. The real part is what clocks measure. The imaginary part is where quantum dynamics lives. Together the two components form a time plane that is itself hyperbolic — curved, not flat.

Why PSL(2,ℤ)

Identify a photon topologically with an elliptic curve (S¹ × S¹ with complex structure τ). The moduli space of elliptic curves over ℂ is then exactly ℍ²/PSL(2,ℤ) — a standard theorem of modular forms. M0 (observables factor through isomorphism classes) enforces this factorisation. Modular invariance is not an axiom; it is a theorem (Theorem 1).

The fundamental domain has three fixed points (i, ρ, i∞) and a global orientation of Im(τ). The classification of Fuchsian groups (Katok 1992) says that no other finite stabiliser orders in PSL(2,ℤ) are possible — for M ∈ SL(2,ℤ) with |tr(M)| ∈ {0, 1} only orders 2 or 3 arise in PSL(2,ℤ). Three fixed points + orientation = exactly four modes (proved 2026-04-18).

These four topological data are candidates for the four fundamental interactions: gravity ↔ cusp (parabolic, long-range), EM ↔ i (Z₂, spin ½), strong ↔ ρ (Z₃, three colours), weak ↔ orientation (chiral, flavour-changing). The attribution is physically natural; formal derivation is pending.

Testable prediction

Theorem 1 fixes the shape of the metric (Poincaré metric on ℍ²); its scale (K = −1, i.e. λ = 1 in Planck units) is not fixed by invariance — that is set by the normalization postulate N. Given N, in the α-attractor class (Kallosh-Linde 2013) α = 2λ/3 = 2/3 (= 2·Area(𝒻)/π). Consistency relation (the only continuous input is N):

r = 2(1 − nₛ)² ≈ 0.0025 for nₛ = 0.9649.

Numerics for T-model V = Λ⁴ tanh²(φ/2) confirm: r = 0.00239 at N = 57. For the canonical modular-invariant potential V_A = Λ⁴·|j(τ)|, Theorem 4 (closed 2026-04-21) additionally forces Nₑ = [PSL(2,ℤ):Γ(5)] = |A₅| = 60 topologically — the maximum-modulus principle fixes Configuration A with no Hessian, and A₅ combinatorics (20 order-3 elements → 10 axes → ×6 triangles) gives 60 with no open computation. The result is r = 2/900 ≈ 0.00222, fully parameter-free, with no slow-roll fit. LiteBIRD ~2032 (σ(r) ≈ 0.001) decides between branch A (T-model) and branch B (strictly modular-invariant, r ~ 10⁻⁵). Starobinsky (r ≈ 0.0037; ratio r_A/r_Star = 2/3, a difference of ≈ 0.7–1.2 σ) is distinguishable from branch A only with CMB-S4-class sensitivity (σ(r) ≈ (3–5)×10⁻⁴).

Side results

What to read next