Optimizationtheoryandmethods Sunyuan 2006 -- Chapter 13 -- Section 13325
noncomputable sectionopen scoped Matrix Matrix.Norms.L2Operatorsectionvariable {m n : ℕ}
The Section 13.3 branch hypothesis (13.3.25) for the stage data
(c_k, A_k, ξ_k, Δ_k, b₂), written directly on the canonical Moore-Penrose pseudoinverse
A_k⁺ = (A_k)⁺.
def powellYuanSection13325
(ck : EuclideanSpace ℝ (Fin m))
(Ak : Matrix (Fin n) (Fin m) ℝ)
(trustRegionRadius ξk b2 : ℝ) : Prop :=
(b2 * trustRegionRadius ≥ ‖(Ak⁺).transpose.toEuclideanLin ck‖ → ξk = 0) ∧
(b2 * trustRegionRadius < ‖(Ak⁺).transpose.toEuclideanLin ck‖ →
ξk ≤ ‖ck - Ak.transpose.toEuclideanLin
(((b2 * trustRegionRadius) / ‖(Ak⁺).transpose.toEuclideanLin ck‖) •
(Ak⁺).transpose.toEuclideanLin ck)‖)
Unfolding powellYuanSection13325 ck Ak trustRegionRadius ξk b2 recovers the two source
branches from (13.3.25).
theorem powellYuanSection13325_iff
(ck : EuclideanSpace ℝ (Fin m))
(Ak : Matrix (Fin n) (Fin m) ℝ)
(trustRegionRadius ξk b2 : ℝ) :
powellYuanSection13325 ck Ak trustRegionRadius ξk b2 ↔
(b2 * trustRegionRadius ≥ ‖(Ak⁺).transpose.toEuclideanLin ck‖ → ξk = 0) ∧
(b2 * trustRegionRadius < ‖(Ak⁺).transpose.toEuclideanLin ck‖ →
ξk ≤ ‖ck - Ak.transpose.toEuclideanLin
(((b2 * trustRegionRadius) / ‖(Ak⁺).transpose.toEuclideanLin ck‖) •
(Ak⁺).transpose.toEuclideanLin ck)‖) :=
Iff.rflend