Theorem 3.2 — MPEC-MFCQ implies standard MFCQ for Scholtes' relaxed programs near x*
ProvedMPECRelax.ScholtesMFCQ.theorem_3_2Let the data of the MPEC (1) be continuously differentiable, let be feasible for (1), and suppose MPEC-MFCQ holds at . Then there exist a neighbourhood of and such that, for every , standard MFCQ for Scholtes' relaxed program
is satisfied at all , where is the feasible set of .
The theorem guarantees that KKT multipliers exist at local minimizers of the relaxed programs near , which is the hypothesis of the convergence result for the method (Theorem 3.1).
Formalization Note. The paper introduces but never uses it: is free in the statement, and the proof works for every (the relaxed program is defined for , p. 8). We keep to mirror the statement and quantify over all ; the neighbourhood is chosen before and does not depend on it. "Standard MFCQ for " is MFCQ of the NLP at , with its own active set, including , and the product constraint when active. The C¹ standing assumption of p. 1 is the hypothesis IsC1.
import Mathlib import Definitions.Def_MPECRelax_ScholtesMFCQ_Basic open Filter Topology open scoped RealInnerProductSpace
namespace MPECRelax.ScholtesMFCQ
/-- Theorem 3.2 (p. 11): if `xs` is feasible for the MPEC (1) and MPEC-MFCQ holds at
`xs`, there are a neighbourhood `N` of `xs` and `t̄ > 0` such that standard MFCQ for
the relaxed program R^S(t) holds at every `x ∈ N ∩ X^S(t)`. The paper's statement never
uses `t̄`; the parameter `t > 0` is free, as in its proof, and `N` does not depend on `t`. -/
theorem theorem_3_2 {n m p l : ℕ} (P : MPEC n m p l) (hP : P.IsC1)
(xs : MPECRelax.ScholtesConv.E n) (hfeas : P.Feasible xs) (hCQ : P.MPEC_MFCQ xs) :
∃ N ∈ 𝓝 xs, ∃ tbar : ℝ, 0 < tbar ∧
∀ t : ℝ, 0 < t → ∀ x ∈ N, (P.RS t).Feasible x → (P.RS t).IsMFCQ x := by sorry
end MPECRelax.ScholtesMFCQ
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.