Contraction: on when the maximal column sum of is
ProvedHarrisonReimanRBM.Orthant.contractionLet and let be a nonnegative matrix with zeros on the diagonal whose maximal column sum satisfies
Fix , , and let be the map (componentwise). Equip paths on with the norm . Then for all paths continuous on ,
Hence is a contraction of the complete metric space , which yields the unique fixed point of (13)–(14).
Formalization Note The paper writes with the maximal row sum. Under the paper's row-vector convention , the Lipschitz constant of in the max-norm is the maximal column sum, and with the printed reading the inequality is false: for , and all other entries (spectral radius , row sums ), , , on gives while . The statement therefore assumes the maximal column sum is ; the Veinott step applied to provides such a rescaling. The paper restricts to ; the inequality holds for all continuous on , and that stronger form is stated.
import Mathlib import Definitions.Def_Reiman84_QueueLength_Paths import Definitions.Def_HarrisonReimanRBM_Orthant_Basic
namespace HarrisonReimanRBM.Orthant
open Reiman84.QueueLength
/-- Proof of Theorem 1, p. 304: if `Q ≥ 0` has zero diagonal and maximal column sum
`α < 1` (the norm for which `y ↦ yQ` is a contraction under the row-vector convention), then
for `x ∈ C_S` and `y, y'` continuous on `[0, T]`,
`‖π(y) − π(y')‖ ≤ α ‖y − y'‖` in the norm `max_j sup_{0 ≤ t ≤ T} |·_j(t)|`. -/
theorem contraction {K : ℕ} (hK : 0 < K) (Q : Matrix (Fin K) (Fin K) ℝ)
(hQ0 : ∀ i j, 0 ≤ Q i j) (hQd : ∀ j, Q j j = 0) (α : ℝ) (hα : maxColSum Q = α)
(hα1 : α < 1) (x : ℝ → Fin K → ℝ) (hx : IsCPlus x) (T : ℝ) (hT : 0 < T)
(y y' : ℝ → Fin K → ℝ) (hy : ContinuousOn y (Set.Icc 0 T))
(hy' : ContinuousOn y' (Set.Icc 0 T)) :
supNormOn T (piMap Q x y - piMap Q x y') ≤ α * supNormOn T (y - y') := by sorry
end HarrisonReimanRBM.Orthant
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.