Documentation

LeanPool.OSforGFF.OS.NonTrivial

Nontriviality of the Gaussian Free Field #

The OS axiom verification in OS.Master would be trivially satisfied by the Dirac measure at ω = 0 (the "zero field"). This file closes that loophole by proving the GFF measure is strictly non-degenerate:

  1. The square-root propagator embedding T : S(ℝ⁴) → L² is injective.
  2. The smeared covariance C(f,f) > 0 for every nonzero test function f.
  3. Every field pairing ⟨ω,f⟩ has strictly positive variance under muGFF.
  4. The pointwise kernel C(x,y) → +∞ as xy (UV divergence).

Proof strategy #

Injectivity of T follows from:

Main results #

References #

Injectivity of the real-to-complex embedding #

The embedding toComplex : S(ℝ⁴,ℝ) → S(ℝ⁴,ℂ) is injective. Follows from injectivity of ℝ → ℂ applied pointwise.

Injectivity of the Fourier transform on Schwartz space #

The Fourier transform is injective on complex Schwartz space. Proof: FourierPair gives 𝓕⁻(𝓕 f) = f, so 𝓕 has a left inverse.

Continuous functions that vanish a.e. vanish everywhere #

Injectivity of the square-root propagator embedding #

theorem OSforGFF.sqrtPropagatorMap_eq_zero_iff (m : ) [Fact (0 < m)] (f : TestFunction) :
(∀ (k : SpaceTime), QFT.sqrtPropagatorMap m f k = 0) f = 0

The square-root propagator map is zero pointwise only if f = 0.

sqrtPropagatorMap m f k = 𝓕(toComplex f)(k) · w(k) where w(k) > 0, so vanishing of the product forces 𝓕(toComplex f) = 0, hence f = 0 by Fourier injectivity.

The embedding T : S(ℝ⁴,ℝ) → L²(ℝ⁴,ℂ) is injective.

If T f = T g then ‖T(f−g)‖ = 0, so ∫ |sqrtPropagatorMap m (f−g)|² = 0. The integrand is continuous and nonneg, so it vanishes a.e., hence everywhere (volume is IsOpenPosMeasure). Since the momentum weight is positive, the Fourier transform of f−g vanishes, giving f = g.

Strict positivity of the covariance #

Strict positive definiteness: the smeared covariance C(f,f) > 0 for any nonzero test function f. This rules out the Dirac-at-zero measure as a model satisfying the OS axioms.

Proof: C(f,f) = ‖T f‖² where T is injective, so f ≠ 0 ⟹ T f ≠ 0 ⟹ ‖T f‖ > 0 ⟹ ‖T f‖² > 0.

Nontriviality of the GFF measure #

The variance of ⟨ω,f⟩ under the GFF is strictly positive for f ≠ 0. Equivalently, the pushforward by the pairing is a non-degenerate Gaussian.

The GFF is not a Dirac measure: there exists a test function whose pairing with ω has nonzero variance. This is the formal statement that the OS axiom verification in Master.lean is nontrivial.

Any nonzero Schwartz function witnesses this. We use a standard bump function on ℝ⁴, which exists by ContDiff.exists_eq_one_of_isOpen.

UV divergence: pointwise covariance diverges at coincident points #

The pointwise regularization C(x,x) = 0 in freeCovarianceBessel is a convention for the smeared (distribution) theory. The actual limit diverges, confirming that the free field has genuine UV singularity.

The smeared covariance C(f,f) = ∫∫ f(x) C(x,y) f(y) dx dy remains finite for all Schwartz functions because the 1/r² singularity of K₁(mr)/r is integrable in 4 spatial dimensions (surface area r³ compensates the kernel 1/r²).

K₁(z) → +∞ as z → 0⁺.

For any T > 0, K₁(z) = ∫₀^∞ e^{-z cosh t} cosh t dt ≥ T · e^{-z cosh T} since cosh t ≥ 1 on [0,T]. As z → 0⁺ the RHS → T, so K₁(z) eventually exceeds any bound.

Formal proof uses monotone convergence: the integrand e^{-z cosh t} cosh t increases monotonically to cosh t as z ↓ 0, and ∫₀^∞ cosh t dt = +∞.

The free covariance C(x,y) → +∞ as xy (UV divergence).

C(x,y) = (m/(4π²r)) · K₁(mr) where r = ‖x-y‖. As r → 0⁺, K₁(mr) ≥ K₁(1) > 0 for mr ≤ 1 and m/(4π²r) → +∞, so the product diverges.