Common-face dimension bound for nonvertex checkpoints
ProvedHirsch.common_face_checkpoint_localizationThe common-face dimension of two points, not necessarily vertices, is controlled by the number of inequalities together with the two endpoint nullities and the all-neutral rank defect of the displacement.
Let be cut out by linear inequalities. For points write for the dimension of the common-direction space of rows tight at both, write (resp. ) for the corresponding nullity at (resp. ) alone, and write for
Then
No feasibility, boundedness, irredundancy, or circuit hypothesis is used. When and are vertices of a circuit step, the already proved vertex localization is the special case up to the usual tight-row spanning. The extra terms are the correction needed for nonvertex checkpoints.
Formalization Note The Lean statement writes in the public HirschCommonFace vocabulary. Natural subtraction is truncated.
import Definitions.Def_Hirsch_common_face_geometry set_option autoImplicit false open scoped RealInnerProductSpace open Hirsch
namespace Hirsch
theorem common_face_checkpoint_localization
{d n : ℕ}
(a : Fin n → EuclideanSpace ℝ (Fin d)) (b : Fin n → ℝ)
(x y : EuclideanSpace ℝ (Fin d)) :
2 * HirschCommonFace.commonFaceDim a b x y + d ≤
n + HirschCommonFace.commonFaceDim a b x x +
HirschCommonFace.commonFaceDim a b y y +
((d - 1) -
Module.finrank ℝ
(HirschCommonFace.rowEvalMap a
(Finset.univ.filter (fun i =>
a i ≠ 0 ∧ ⟪a i, y - x⟫ = 0))).range) + 1 := by sorry
end Hirsch