Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 4.1 — approximate minimizers of Lr(⋅,yk)L_r(\cdot, y^k)Lr​(⋅,yk) along a bounded maximizing dual sequence are asymptotically minimizing

Proved
PenaltyLag.Asymptotic.theorem_4_1

by mikedeng1 · Oct 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

augmented-lagrangiandualitynonlinear-programmingp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1

Throughout, XXX is a nonempty convex subset of a real vector space EEE and f0,f1,…,fm:X→Rf_0, f_1, \dots, f_m : X \to \mathbb Rf0​,f1​,…,fm​:X→R are convex (the paper's standing assumption of §3, p. 358). Suppose the asymptotic optimal value in (P) is finite. Let r>0r > 0r>0, let {yk}\{y^k\}{yk} be a bounded maximizing sequence for (Dr)(D_r)(Dr​), i.e. gr(yk)→sup⁡y∈Rmgr(y)g_r(y^k) \to \sup_{y \in \mathbb R^m} g_r(y)gr​(yk)→supy∈Rm​gr​(y), and for each kkk let xk∈Xx^k \in Xxk∈X satisfy

Lr(xk,yk)−inf⁡x∈XLr(x,yk)=Lr(xk,yk)−gr(yk)≤αk,(4.7)L_r(x^k, y^k) - \inf_{x \in X} L_r(x, y^k) = L_r(x^k, y^k) - g_r(y^k) \le \alpha_k, \tag{4.7}Lr​(xk,yk)−x∈Xinf​Lr​(x,yk)=Lr​(xk,yk)−gr​(yk)≤αk​,(4.7)

where αk→0\alpha_k \to 0αk​→0. Then {xk}\{x^k\}{xk} is an asymptotically minimizing sequence for (P): lim sup⁡kfi(xk)≤0\limsup_k f_i(x^k) \le 0limsupk​fi​(xk)≤0 for every iii, and lim sup⁡kf0(xk)\limsup_k f_0(x^k)limsupk​f0​(xk) equals the asymptotic optimal value in (P).

This is the paper's headline result: any method that maximizes the smooth, unconstrained dual grg_rgr​ over all of Rm\mathbb R^mRm, using only approximate minimizations of Lr(⋅,yk)L_r(\cdot, y^k)Lr​(⋅,yk), produces an asymptotically optimal primal sequence, with no constraint qualification and no assumption that (P) has feasible or optimal solutions.

Formalization Note (4.7) is written Lr(xk,yk)≤gr(yk)+αkL_r(x^k, y^k) \le g_r(y^k) + \alpha_kLr​(xk,yk)≤gr​(yk)+αk​ in EReal, equivalent to the printed difference because Lr(xk,yk)L_r(x^k, y^k)Lr​(xk,yk) is real and gr(yk)<+∞g_r(y^k) < +\inftygr​(yk)<+∞; it avoids EReal subtraction. Only αk→0\alpha_k \to 0αk​→0 is assumed (nonnegativity follows from (4.7)). "Maximizing" is convergence in EReal to the supremum over all of Rm\mathbb R^mRm, with no sign restriction on yyy; "bounded" is boundedness of {yk}\{y^k\}{yk} in Rm\mathbb R^mRm; "finite" is ≠±∞\ne \pm\infty=±∞ in EReal. The paper's functions fi:X→Rf_i : X \to \mathbb Rfi​:X→R are total functions E→RE \to \mathbb RE→R convex on XXX, and only their values on XXX enter; constraint indices are Fin m (0-based, the paper's 1,…,m1, \dots, m1,…,m), and f0f_0f0​ is a separate argument; the standing assumption (p. 358: XXX nonempty convex, fif_ifi​ convex) is a hypothesis even where the statement does not repeat it.

Preamble
import Mathlib
import Definitions.Def_PenaltyLag_Asymptotic_Basic

open Filter Topology
Formal statement
namespace PenaltyLag.Asymptotic

/-- Theorem 4.1 (p. 365): if the asymptotic optimal value in (P) is finite, {yᵏ} is a bounded
maximizing sequence for (D_r), r > 0, and xᵏ ∈ X satisfies L_r(xᵏ, yᵏ) − g_r(yᵏ) ≤ αₖ (4.7) with
αₖ → 0, then {xᵏ} is an asymptotically minimizing sequence for (P). -/
theorem theorem_4_1 {E : Type*} [AddCommGroup E] [Module ℝ E] {m : ℕ}
    (X : Set E) (hX : Convex ℝ X) (hXne : X.Nonempty)
    (f₀ : E → ℝ) (f : Fin m → E → ℝ) (hf₀ : ConvexOn ℝ X f₀) (hf : ∀ i, ConvexOn ℝ X (f i))
    (hfin : asympValue X f₀ f ≠ ⊥ ∧ asympValue X f₀ f ≠ ⊤)
    (r : ℝ) (hr : 0 < r)
    (y : ℕ → Mult m) (hyb : Bornology.IsBounded (Set.range y)) (hymax : IsMaximizing X f₀ f r y)
    (x : ℕ → E) (hxX : ∀ k, x k ∈ X)
    (α : ℕ → ℝ) (hα : Tendsto α atTop (𝓝 0))
    (h47 : ∀ k, (Lr f₀ f r (x k) (y k) : EReal) ≤ gr X f₀ f r (y k) + (α k : EReal)) :
    IsAsympMinimizing X f₀ f x := by sorry

end PenaltyLag.Asymptotic
Source
Rockafellar, A Dual Approach to Solving Nonlinear Programming Problems by Unconstrained Optimization, Math. Programming 5 (1973), p. 365, Theorem 4.1, (4.7)
Human review
  • Endorsed by Shuze Chen · Oct 6, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 6, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me