Geometric-mean affinity bound for the reciprocal quadratic form
DisprovedRybinAI2026.P01.one_sphere_geometric_affinityFor a positive-definite n by n matrix C and unit directions z, w, the spherical integral of sqrt(|<u,z>| |<u,w>|) divided by u^T C u is at least half the geometric mean of the two directional integrals of |<u,z>|/(u^T C u) and |<u,w>|/(u^T C u). Equivalently, the Hellinger affinity of the two measures with densities |<u,z>|/(u^T C u) and |<u,w>|/(u^T C u) (normalised to probability measures) is at least 1/2, i.e. their total-variation distance is at most 1/2. This is a single-integral comparison involving only one matrix; by Cauchy-Schwarz the ratio lies between 1/2 and 1, and the published constant is the lower end. Numerically the ratio is at least 0.83 in dimensions 2 to 5 over randomised anisotropic C and directions (dense quadrature in 2D, Monte-Carlo in 3, 4, 5D; for C = I and perpendicular z, w it equals 0.847 in every dimension).
import Definitions.Def_rybin2026_p01_cross_integral import Definitions.Def_rybin2026_p01_matrix_integral set_option autoImplicit false open Matrix MeasureTheory Metric RybinAI2026.P01 open scoped BigOperators
namespace RybinAI2026.P01
/-- Geometric-mean (affinity) lower bound for the reciprocal quadratic form: the spherical
integral of sqrt(|<u,z>| |<u,w>|) over u^T C u dominates half the geometric mean of the two
linear directional integrals of the same reciprocal quadratic form. -/
theorem one_sphere_geometric_affinity {n : ℕ} (C : Matrix (Fin n) (Fin n) ℝ) (hC : C.PosDef) :
∀ z w : Euclidean n,
(1 / 2) * Real.sqrt
((∫ u : Metric.sphere (0 : Euclidean n) 1,
|bilinear (1 : Matrix (Fin n) (Fin n) ℝ) u.1 z| / bilinear C u.1 u.1 ∂surfaceMeasure n) *
(∫ u : Metric.sphere (0 : Euclidean n) 1,
|bilinear (1 : Matrix (Fin n) (Fin n) ℝ) u.1 w| / bilinear C u.1 u.1 ∂surfaceMeasure n))
≤ ∫ u : Metric.sphere (0 : Euclidean n) 1,
Real.sqrt (|bilinear (1 : Matrix (Fin n) (Fin n) ℝ) u.1 z| *
|bilinear (1 : Matrix (Fin n) (Fin n) ℝ) u.1 w|) /
bilinear C u.1 u.1 ∂surfaceMeasure n := by sorry
end RybinAI2026.P01