Mixed spherical integral
Definitionrybin2026_p01_cross_integralMixed spherical double integral for the CUHK-Shenzhen AI Math Problem 1 model.
For every dimension and arbitrary real matrices , let be the unit sphere carrying the surface measure induced from Lebesgue measure by polar decomposition (no probability normalisation). Define
The numerator is the absolute bilinear form of the difference , exactly as in the Problem 1 integrand, but the two quadratic forms in the denominator are taken from independent matrices: is paired with the outer variable and with the inner variable . Setting and recovers the Problem 1 distance .
This object is the natural interpolation device when the diagonal denominator pair is replaced by a larger pair such as : it separates numerator effects (the difference being integrated) from denominator effects (the regularising quadratic forms), and it is what makes the additive Problem 1 inequality decompose into a numerator-subadditivity step and a denominator-normalisation step.
Formalization Note Real division is total: at a pair where a denominator factor is the integrand is , and a non-integrable inner or outer integrand makes the corresponding Bochner integral . For the sphere is empty and the value is . The definition reuses bilinear and surfaceMeasure from the Problem 1 definition module Def_rybin2026_p01_matrix_integral.
import Definitions.Def_rybin2026_p01_matrix_integral
open Matrix MeasureTheory Metric
open scoped BigOperators
namespace RybinAI2026.P01
/-- Mixed spherical double integral. The numerator measures the bilinear form of the
difference `X - Y` on the pair of unit vectors, while the two quadratic forms in the
denominator are taken from *independent* matrices `P` (paired with `u`) and `Q` (paired
with `v`). Taking `P = X` and `Q = Y` recovers `distance X Y`. Real division is total:
where a denominator factor vanishes the integrand at that pair is `0`, and a non-integrable
integrand contributes `0`. -/
noncomputable def crossIntegral {n : ℕ}
(X Y P Q : Matrix (Fin n) (Fin n) ℝ) : ℝ :=
∫ u, ∫ v,
|bilinear (X - Y) u.1 v.1| /
(bilinear P u.1 u.1 * bilinear Q v.1 v.1)
∂surfaceMeasure n ∂surfaceMeasure n
/-- `distance` is the diagonal case `P = X`, `Q = Y` of `crossIntegral`. -/
theorem distance_eq_crossIntegral {n : ℕ} (X Y : Matrix (Fin n) (Fin n) ℝ) :
distance X Y = crossIntegral X Y X Y := rfl
end RybinAI2026.P01