Eq. (2.4) — the projected path is a descent path
ProvedCalamaiMore.Convergence.projPath_descentdescentgradient-projectionp2o-batch-p200ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be a nonempty closed convex subset of a finite-dimensional real inner product space , the projection into , and continuously differentiable on . For let . Then for every ,
In particular the step from to is a descent direction whenever ; the estimate is used with throughout the convergence analysis.
Formalization Note "Continuously differentiable on " is differentiable at every point of with continuous on (the paper's standing assumption for §2).
Preamble
import Mathlib import Definitions.Def_CalamaiMore_Convergence_IsGradientProjectionRun
Formal statement
namespace CalamaiMore.Convergence
/-- Calamai–Moré, Eq. (2.4) (p. 98): for `x ∈ Ω` and `α > 0`,
`⟨∇f(x), x - x(α)⟩ ≥ ‖x(α) - x‖² / α` where `x(α) = P(x - α ∇f(x))`. -/
theorem projPath_descent {E : Type*} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E]
(Ω : Set E) (hΩne : Ω.Nonempty) (hΩc : IsClosed Ω) (hΩcv : Convex ℝ Ω)
(f : E → ℝ) (hfd : ∀ x ∈ Ω, DifferentiableAt ℝ f x) (hfc : ContinuousOn (gradient f) Ω)
(x : E) (hx : x ∈ Ω) (α : ℝ) (hα : 0 < α) :
‖projPath f Ω x α - x‖ ^ 2 / α ≤ inner ℝ (gradient f x) (x - projPath f Ω x α) := by sorry
end CalamaiMore.Convergence
Source
Calamai & Moré, Projected gradient methods for linearly constrained problems, Math. Programming 39 (1987), p. 98, Eq. (2.4)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.