Alphabeta Math
TheoremStatement: Literature-sourcedProof: AI-adaptedPipeline-generatedjudge pass (gpt-6-luna)
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.

Borell–Sudakov–Tsirelson expansion for a Banach Gaussian law

Statement

Assume AC and countable choice. Let μ be a centered tight Borel Gaussian law on a real separable Banach space B, with covariance RKHS H and unit ball H1. For every Borel set A⊆B and r≥0, μ∗(A+rH1)≥Φ ⁣(Φ−1(μ(A))+r), where μ∗(S)=sup⁡{μ(K):K⊆S, K compact} is the inner measure and Φ−1(0)=−∞, Φ−1(1)=+∞. When A+rH1 is Borel, its inner measure equals its ordinary μ-measure.

Facts & Assumptions

[A1]

AC and countable choice are assumed by the finite-dimensional Gaussian and countable-cylinder interfaces (The Axiom of Choice, The Axiom of Countable Choice (ACω)).

[A2]

Under AC, the real dominated-extension theorem supplies HB for the Hahn–Banach dual-norming interface (Hahn-Banach dominated extension theorem for real vector spaces, The real dominated-extension principle as an additional hypothesis over ZF).

[F1]

For Borel E⊆Rd and r≥0, Gaussian inner measure satisfies γd,∗(E+rB2d)≥Φ(Φ−1(γd(E))+r) (Finite-dimensional Gaussian isoperimetric expansion). This supplier proves the sharp finite-dimensional expansion locally.

[F2]

The covariance RKHS H is a Hilbert space continuously and injectively embedded in B, with ℓ(ih)=⟨h,[ℓ]⟩H (Covariance operator and reproducing-kernel Hilbert space of a Banach Gaussian, The Banach Gaussian covariance and RKHS construction is well-defined).

[F3]

The Hilbert unit ball H1 is compact in the Banach norm (The Cameron–Martin unit ball is compact in the Banach norm).

[F4]
[F5]

A separable metric space has a countable base, and countable choice selects one cover member for each relevant basis element (Separability: the existence of an at most countable dense subset, The Axiom of Countable Choice (ACω)).

[F6]

Probability is continuous from above on decreasing measurable sets (Continuity from above when one set has finite measure).

[F7]

Under HB, distinct points in a normed space are separated by a continuous linear functional (Relative dual norming, point separation, and recovery of the norm).

Proof

Given: AC, countable choice, a centered tight Borel Gaussian law μ on a real separable Banach space B, its covariance RKHS H, and H1={h∈H:∥h∥H≤1}.

1.1F2givenchoose

First let F=T−1(M) be a closed finite-coordinate cylinder, with T=(ℓ1,…,ℓk):B→Rk continuous and M closed. Let Q=span⁡{[ℓi]:1≤i≤k}⊂H, choose an orthonormal basis e1,…,em, and choose representatives φj∈B∗ for ej. Set G=∑j=1mφj(W)ej and S=T∘i∣Q. The vector of coefficients is standard Gaussian on Q because its covariance matrix is the identity. For each i, [ℓi]=∑j⟨[ℓi],ej⟩Hej; by the covariance quotient and [F2], the residual ℓi(W)−ℓi(iG) has variance ∥[ℓi]−∑j⟨[ℓi],ej⟩Hej∥H2=0. Hence T(W)=S(G) almost surely. The map S is injective: if S(q)=0, then ⟨q,[ℓi]⟩H=0 for every i, and these classes span Q, so q=0.

1.2A1A2F4F5F7givenchoose

Now let A⊆B be Borel with 0<μ(A)<1, and fix β<Φ−1(μ(A)). By [F4] choose compact K⊆A with μ(K)>Φ(β). For each x∉K and y∈K, [F7] gives ℓy∈B∗ with ℓy(x)≠ℓy(y). The open sets Uy={z:∣ℓy(z)−ℓy(y)∣<∣ℓy(x)−ℓy(y)∣/3} cover K; compactness gives a finite subcover Uy1,…,Uyq. For every z∈K, some i has z∈Uyi, which forces ℓyi(z)≠ℓyi(x). Thus the finite coordinate map Tx=(ℓy1,…,ℓyq) satisfies Tx(x)∉Tx(K). The cylinder Fx=Tx−1(Tx(K)) is closed, contains K, and excludes x, since Tx(K) is compact. The open complements B∖Fx cover B∖K. By [F5] this cover has a countable subcover: for every basis element meeting B∖K and contained in a cover member, countable choice selects one such member; every point is covered by one selected member. If the subcover is finite, repeat one member to index it by j≥1.

2.1A1F1F2F4step 1.1algebra

If m=0, then T(W)=0 almost surely, so μ(F) is zero or one; the endpoint convention, 0∈H1, and inner regularity [F4] give the claimed inequality. If m>0, define E=S−1(M)⊂Q. It is closed and μ(F)=γQ(E). For r≥0, Er=E+r(Q∩H1) is closed because its second summand is compact. Thus [F4] identifies its inner and ordinary Gaussian measures, and [F1] gives γQ(Er)≥Φ(Φ−1(γQ(E))+r). Let Cr=T−1(S(Er)), a closed finite-coordinate cylinder because S is a linear isomorphism from finite-dimensional Q onto the closed subspace S(Q); injectivity of S and T(W)=S(G) give μ(Cr)=γQ(Er). If y∈Cr, write T(y)=S(u+v) with u∈E and v∈r(Q∩H1). Then T(y−i(v))=S(u)∈M, so y−i(v)∈F and y∈F+rH1. Hence Cr⊆F+rH1, and inner regularity [F4] gives μ∗(F+rH1)≥μ(Cr)≥Φ(Φ−1(μ(F))+r).

3.1F3F4step 2.1step 1.2

Let Gn=⋂j≤nFxj for the countable subcover in step 1.2. Each Gn is a closed finite-coordinate cylinder because it is a finite intersection of such cylinders; the sequence decreases, ⋂nGn=K, and μ(Gn)≥μ(K)>Φ(β). Applying step 2.1 yields μ∗(Gn+rH1)≥Φ(β+r). The set Gn+rH1 is closed: if aj+hj converges with aj∈Gn and hj∈rH1, compactness [F3] gives a convergent subsequence of hj, after which closedness of Gn gives a limit representation. Thus [F4] gives equality of its inner and ordinary measures.

4.1F3F4F6step 3.1algebra∎

The closed sets Gn+rH1 decrease and their intersection is K+rH1. Indeed, if x=an+hn with an∈Gn and hn∈rH1 for every n, compactness [F3] gives a subsequence hnj→h∈rH1; then anj→x−h, and for each fixed N the tail belongs to the closed set GN, so x−h∈K. The reverse inclusion is immediate. Continuity from above [F6] gives μ(K+rH1)≥Φ(β+r). Since K⊆A, μ∗(A+rH1)≥Φ(β+r). Letting β↑Φ−1(μ(A)) proves the interior-probability case. If μ(A)=0 the right side is zero; if μ(A)=1, inner regularity [F4] and 0∈H1 give μ∗(A+rH1)=1.

Depends on

Used by

Dependency tree · two levels

75 results within two dependency steps of this one, each drawn at its shortest distance from it. An arrow runs from a result to what uses it, so the chart reads left to right and ends at this result, which carries a heavier outline. Every node is a link to that result. Click elsewhere on the chart to enlarge it.

Sources