Theorem 2.1, p. 10 — for G = G_HB or G_NAG and 0 < α ≤ 2/√κ(A), the Gauss–Seidel scheme (41) is A-stable and ρ(E(α, G)) ≤ 1/√(1 + 2α)
ProvedNAGFlow.GSSpectrum.theorem_2_1Let be a real symmetric matrix whose eigenvalues satisfy , and let . Let be either of the transformations (39),
split blockwise as with the lower triangular part (diagonal blocks included), and let be the one-step matrix of the Gauss–Seidel splitting scheme of (41). If
then the scheme is A-stable, , and
With this gives the contraction factor per step, the accelerated rate, for an explicit scheme that never inverts .
Formalization Note. Eigenvalues of the non-symmetric real matrix are complex. "" and "" are stated for every eigenvalue : and . This avoids a junk supremum; is nonempty here anyway since . and are encoded as: both are eigenvalues of and bound all eigenvalues of . The case is the one of Section 2.3, where and are defined.
import Mathlib import Definitions.Def_NAGFlow_GSSpectrum_Setting
namespace NAGFlow.GSSpectrum
open Matrix
/-- Theorem 2.1, p. 10. In the quadratic model of Section 2 with μ = λ_min(A) > 0, for
G = G_HB or G_NAG of (39), if 0 < α ≤ 2/√κ(A), then the Gauss–Seidel splitting scheme (41) is
A-stable, i.e. ρ(E(α, G)) < 1, and ρ(E(α, G)) ≤ 1/√(1 + 2α). Both spectral-radius claims are
stated eigenvalue by eigenvalue. -/
theorem theorem_2_1 {d : ℕ} (A : Matrix (Fin d) (Fin d) ℝ) (μ L α : ℝ) (hA : IsQuadModel A μ L)
(hμ : 0 < μ) (hα : 0 < α) (h127 : α ≤ 2 / Real.sqrt (kappa μ L)) :
∀ G, (G = GHB A μ ∨ G = GNAG A μ) →
(∀ z ∈ cspec (EG α G), ‖z‖ < 1) ∧
∀ z ∈ cspec (EG α G), ‖z‖ ≤ 1 / Real.sqrt (1 + 2 * α) := by sorry
end NAGFlow.GSSpectrum
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.