Alphabeta Math
DefinitionDefinition: AI-adaptedProof: AI-generatedPipeline-generatedprecheck passjudge pass (gpt-6-luna)verified 2026-11-04 (gpt-6.1-sol)
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

Gaussian free field on the Riemann sphere in the Liouville convention

Definition

Let S=C^=C∪{∞} with its usual two complex charts. In the finite chart write z=x+iy, put ∣z∣+=max⁡{1,∣z∣}, and use the finite positive background measure dA0(z)=∣z∣+−4 dx dy. It has total mass π+2π∫1∞r−3dr=2π. Under w=1/z, the Jacobian dxzdyz=∣w∣−4dxwdyw gives the same coordinate expression dA0(w)=∣w∣+−4 dℜ(w) dℑ(w), which is continuous and positive but not smooth across ∣w∣=1. Independently, use the smooth round metric grd=4(dx2+dy2)/(1+∣z∣2)2 in the finite chart; in the w chart it has the same expression. Write drd for its geodesic distance and dArd for its area measure. The covariance kernel below is interpreted globally in the two charts as a locally integrable function off its logarithmic diagonal singularity.

A centered spherical Gaussian free field in the Liouville convention is a probability space (Ω,F,P) and a real linear map X:C∞(S;R)→L2(Ω,P) such that every finite vector (X(f1),…,X(fn)) is centered Gaussian and, for all real smooth test functions f,h,

E[X(f)X(h)]=∫S∫Sf(z)h(w)K(z,w) dArd(z) dArd(w),K(z,w)=log⁡1∣z−w∣+log⁡∣z∣++log⁡∣w∣+.

This finite-chart formula is global: when ∣z∣,∣w∣>1, it becomes K=log⁡(1/∣z−1−w−1∣) in the w=1/z chart; when ∣z∣>1 and ∣w∣≤1, it becomes K=log⁡(1/∣1−w/z∣), and the other mixed case follows by symmetry. These formulas agree on overlaps and have only the stated diagonal logarithmic singularity.

Require that, outside one null set, f↦X(f) is a continuous distribution on C∞(S) with its usual test-function topology. The point notation X(z) is only shorthand for this random distribution and its covariance kernel; it does not assert a pointwise value. Fix 0<ε0<1 and a finite constant M large enough for the constructions below. A smoothing family is a family of nonnegative smooth local averaging kernels ρε(z,w), for 0<ε<ε0, supported where drd(z,w)≤ε, satisfying ρε(z,w)≤Mε−2, with unit integral in w against dArd(w) and satisfying ∫Sρε(z,w)f(w)dArd(w)→f(z) for every smooth f. For every two such families ρ,σ, the mixed regularized covariance is Kε,δρ,σ(z,w):=E[X(ρε(z,⋅))X(σδ(w,⋅))] and converges to K(z,w) off the diagonal, with Kε,δρ,σ(z,w)≤C+log⁡+ ⁣1drd(z,w)+ε+δ uniformly in z,w,ε,δ,ρ,σ for one finite C depending only on M and the fixed sphere geometry, as verified below. Set Xερ(z)=X(ρε(z,⋅)); this is a smooth Gaussian field for each ε>0. Thus the local L2 lemma has its mixed-family covariance hypothesis. The test pairing uses the smooth measure dArd; the distinct measure dA0 remains the interaction/background measure in GMC and is not inserted into the smoothing pairing. No division by the nonsmooth density of dA0 is used.

This convention fixes the background measure and additive normalization used by the probabilistic Liouville sphere construction. The constant mode is not part of X and is introduced separately in that construction.

Proof

Given: A random distribution with the covariance and almost-sure continuity in the Definition, the round sphere, and the fixed scale bound M.

1.1givenconstructalgebra

Choose a nonnegative smooth function χ on [0,∞), positive at zero and identically zero for arguments at least 1, with all derivatives zero at 1. For 0<ε<ε0<1, set Nε(z)=∫Sχ(drd(z,u)2/ε2)dArd(u) and ρε(z,u)=χ(drd(z,u)2/ε2)/Nε(z). The squared distance is smooth below the injectivity radius; outside that region the numerator is identically zero, so the kernel is globally smooth. Round-ball area bounds give cε2≤Nε(z)≤Cε2. This constructs a nonempty class with a fixed finite scale bound M. The unit integral and shrinking support imply approximation of each continuous test function uniformly.

1.2givenalgebra

The kernel satisfies K(u,v)=log⁡+(1/drd(u,v))+H(u,v) with H bounded on the complement of the diagonal and extending boundedly across it. Indeed the chordal-distance identity rewrites the finite-chart singularity using the bounded continuous function log⁡∣u∣+−12log⁡(1+∣u∣2) in each variable; round distance and chordal distance have bounded ratio. Continuity off the diagonal includes points on the equator: log⁡∣u∣+ is continuous and locally Lipschitz there, although not differentiable. Probability averaging on shrinking balls therefore gives the asserted off-diagonal covariance limit.

2.1givenstep 1.2algebra

Write a=max⁡(ε,δ) and d=drd(z,w). If d>4a, all pairs in the two supports have distance between d/2 and 3d/2, so the logarithmic bound follows immediately. If d≤4a, average first in the variable whose smoothing radius is a. Its probability density is at most Ma−2. For every fixed other point v, the round polar-area bound gives ∫drd(u,v)<alog⁡(a/drd(u,v))dArd(u)≤Ca2∫01tlog⁡(1/t)dt≤C′a2. Since a<1, the inequality log⁡+(1/r)≤log⁡(1/a)+log⁡+(a/r) gives the averaged upper bound log⁡(1/a)+CM. In this case a≤d+ε+δ≤6a; together with the bounded remainder from step 1.2 this proves the displayed uniform mixed bound.

3.1givenstep 1.1step 1.2step 2.1constructalgebra∎

For every sample distribution in the one full-probability event in the Definition, z↦X(ρε(z,⋅)) is smooth: derivatives in z pass through the continuous linear functional on smooth tests. This assertion differentiates the smooth test kernel, not K or dA0. In a planar chart with dArd(u)=a(u)d2u, a fixed nonnegative compact planar convolution is represented exactly by the smooth test ηε(z−u)/a(u). A cutoff and a smooth partition of unity extend such kernels to global averaging kernels; local equivalence of the round and planar distances gives their scale bound after a fixed rescaling of the parameter. Any fixed finite selection of these profiles and the kernels of step 1.1 fits one common M. In particular planar radial convolutions remain available across ∣z∣=1, since a is smooth there. Their covariance averages the original K and no density multiplier survives. This proves the smoothing claims without any smoothness assumption on the equatorial background measure.

Dependency tree · 0 levels

Nothing. This result depends on no other item in the library.

Sources