Proof of Theorem 9, p. 21 — if λ > 0 then ⟨ḡ, x − x̄⟩ ≧ 0 for all x with h(x) ≦ 0
ProvedMifflinSemismooth.Optimality.inner_nonneg_on_feasiblegeneralized-gradientnonsmooth-optimizationp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1semiconvex
Let be semiconvex on and let satisfy . Let , and with . Then
Combined with the semiconvexity of and , this inequality shows on the feasible set, i.e. that is optimal when .
Formalization Note The membership of the page's setting is not needed for this step and is not assumed; dropping it makes the statement stronger, not weaker.
Preamble
import Mathlib import Definitions.Def_MifflinSemismooth_Optimality_Setting
Formal statement
namespace MifflinSemismooth.Optimality
/-- Mifflin (1976), §5, proof of Theorem 9, p. 21: if `h` is semiconvex on `ℝⁿ`, `h(x̄) = 0`,
`0 < λ ≤ 1`, `ĝ ∈ ∂h(x̄)` and `λ ḡ + (1 - λ) ĝ = 0`, then `⟨ḡ, x - x̄⟩ ≥ 0` for every `x` with
`h(x) ≤ 0`. -/
theorem inner_nonneg_on_feasible {n : ℕ} (h : EuclideanSpace ℝ (Fin n) → ℝ)
(hh : MifflinSemismooth.Extremal.SemiconvexOn Set.univ h) (xbar : EuclideanSpace ℝ (Fin n)) (hzero : h xbar = 0)
(lam : ℝ) (gbar ghat : EuclideanSpace ℝ (Fin n)) (hlam : 0 < lam) (hlam1 : lam ≤ 1)
(hghat : ghat ∈ MifflinSemismooth.Extremal.genGrad h xbar) (hcomb : lam • gbar + (1 - lam) • ghat = 0) :
∀ x, h x ≤ 0 → 0 ≤ inner ℝ gbar (x - xbar) := by sorry
end MifflinSemismooth.Optimality
Source
Mifflin, Semismooth and semiconvex functions in constrained optimization, IIASA Research Report RR-76-21 (December 1976), p. 21, §5, proof of Theorem 9, case λ > 0
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.