Theorem A.3: mixed Krivine preprocessing,
OpenGrothendieckLB.Supplement.theorem_A_3gaussian-analysisgrothendieck-constantkrivine-roundingp2o-gran-per-chapterp2o-plan-paperp2o-v1
Let and, for each , let be odd measurable functions with arcsine-normalized correlation function
Let with , and set . Suppose that has a local inverse at the origin, , holomorphic on a neighborhood of , and that
Then
Mixing finitely many schemes, possibly of different dimensions, is thus as good as a single scheme for the inverse-majorant bound.
Formalization Note The index set is ; each is a published KrivineScheme (k ℓ) with its correlationFunction. Holomorphy and are encoded as in Theorem A.1.
Differences from the source:
- encoding: γ D with γ > 0 implicit (a disk of radius γ) → explicit hypothesis 0 < γ; reason: as in Theorem A.1: the condition only excludes the undefined case γ = 0
- encoding: H_λ⁻¹ holomorphic on a neighborhood of γD → ∑ |a_r| R^r converges for some R > γ; reason: as in Theorem A.1; same instances
- encoding: ℓ ∈ {1, …, L} → ℓ : Fin L (0-based); reason: index shift only
Preamble
import Mathlib import Definitions.Def_GrothendieckConstantDefs import Definitions.Def_KrivineSchemeDefs import Definitions.Def_GrothendieckLB_IsLocalInverseBranch import Definitions.Def_GrothendieckLB_taylorCoeff
Formal statement
namespace GrothendieckLB
namespace Supplement
/-- Theorem A.3, mixed Krivine preprocessing and projection (Saha et al., *New Lower and Upper Bounds for the Grothendieck Constant*, arXiv:2608.11158 (held version, sha256 e7eb5d65…), Theorem A.3, p.49): let `L ∈ ℕ`
and, for `ℓ ∈ {1, …, L}`, let `f_ℓ, g_ℓ : ℝ^{k_ℓ} → {−1, 1}` be odd measurable functions with
arcsine-normalized correlation function `H_ℓ`. Let `λ₁, …, λ_L ≥ 0` with `∑ λ_ℓ = 1` and
`H_λ(t) := ∑_ℓ λ_ℓ H_ℓ(t)`. Suppose `H_λ` has a local inverse at the origin,
`H_λ⁻¹(ζ) = ∑_{r ≥ 1} a_r ζ^r`, holomorphic on a neighborhood of `γD`, and
`M_λ(γ) := ∑_{r ≥ 1} |a_r| γ^r ≤ 1`. Then `K_G ≤ π/(2γ)`.
**Formalization Note** The index set is `Fin L` (0-based); the dimensions `k ℓ` may differ.
Each `(f_ℓ, g_ℓ)` is a published `GrothendieckConstant.KrivineScheme (k ℓ)` and `H_ℓ` its
`correlationFunction` (see Theorem A.1). The weights are `lam`. Holomorphy on a neighborhood of
`γD` and `γ > 0` are encoded as in Theorem A.1. -/
theorem theorem_A_3 (L : ℕ) (k : Fin L → ℕ)
(S : (ℓ : Fin L) → GrothendieckConstant.KrivineScheme (k ℓ)) (lam : Fin L → ℝ)
(hlam_nonneg : ∀ ℓ, 0 ≤ lam ℓ) (hlam_sum : ∑ ℓ, lam ℓ = 1) (G : ℝ → ℝ)
(hG : IsLocalInverseBranch
(fun t => ∑ ℓ, lam ℓ * GrothendieckConstant.correlationFunction (S ℓ) t) G)
(γ : ℝ) (hγ : 0 < γ)
(hhol : ∃ R : ℝ, γ < R ∧ Summable (fun r => |taylorCoeff r G| * R ^ r))
(hM : ∑' r : ℕ, |taylorCoeff (r + 1) G| * γ ^ (r + 1) ≤ 1) :
GrothendieckConstant.grothendieckConst ≤ Real.pi / (2 * γ) := by sorry
end Supplement
end GrothendieckLB
Source
Saha et al., New Lower and Upper Bounds for the Grothendieck Constant, arXiv:2608.11158 (held version, sha256 e7eb5d65…), Theorem A.3, p.49