Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma A4 — a subdifferential chain rule for f₀ ∘ (f₁, …, f_m) with f₀ nondecreasing

Proved
BregmanPPA.ProxMult.lemmaA4

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

chain-ruleconvex-analysisp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1subdifferential

Let f1,…,fmf_1,\dots,f_mf1​,…,fm​ be proper convex functions on Rn\mathbb R^nRn with ⋂i=1mri⁡(dom⁡fi)≠∅\bigcap_{i=1}^m\operatorname{ri}(\operatorname{dom}f_i)\ne\emptyset⋂i=1m​ri(domfi​)=∅, and write f(x)=(f1(x),…,fm(x))∈(−∞,+∞]mf(x)=(f_1(x),\dots,f_m(x))\in(-\infty,+\infty]^mf(x)=(f1​(x),…,fm​(x))∈(−∞,+∞]m. Let f0f_0f0​ be a proper convex function on Rm\mathbb R^mRm which is nondecreasing (u≤vu\le vu≤v componentwise implies f0(u)≤f0(v)f_0(u)\le f_0(v)f0​(u)≤f0​(v)). Define G:Rn→(−∞,+∞]G:\mathbb R^n\to(-\infty,+\infty]G:Rn→(−∞,+∞] by G(x)=f0(f(x))G(x)=f_0(f(x))G(x)=f0​(f(x)) when f(x)∈Rmf(x)\in\mathbb R^mf(x)∈Rm and G(x)=+∞G(x)=+\inftyG(x)=+∞ when fi(x)=+∞f_i(x)=+\inftyfi​(x)=+∞ for some iii. Then GGG is proper and convex, and if f0f_0f0​ is differentiable at f(x0)∈Rmf(x^0)\in\mathbb R^mf(x0)∈Rm, with y=∇f0(f(x0))y=\nabla f_0(f(x^0))y=∇f0​(f(x0)),

∂G(x0)=∑i=1myi ∂fi(x0)+Ndom⁡G(x0).\partial G(x^0)=\sum_{i=1}^m y_i\,\partial f_i(x^0)+N_{\operatorname{dom}G}(x^0).∂G(x0)=i=1∑m​yi​∂fi​(x0)+NdomG​(x0).

The lemma computes ∂Gk\partial G_k∂Gk​ for Gk(x)=1ckhp∗+(∇hp(pk)+ckg(x))G_k(x)=\frac1{c_k}h_p^{*+}(\nabla h_p(p^k)+c_kg(x))Gk​(x)=ck​1​hp∗+​(∇hp​(pk)+ck​g(x)) in the proof of (14).

Formalization Note The paper says "nondecreasing in the sense of Lemma 5" and, in the definition of GGG, "f(x)∈Rnf(x)\in\mathbb R^nf(x)∈Rn"; the paper has no Lemma 5 and f(x)f(x)f(x) has mmm components, so the intended readings are Lemma 2 and Rm\mathbb R^mRm. "f0f_0f0​ differentiable at f(x0)∈Rmf(x^0)\in\mathbb R^mf(x0)∈Rm" is encoded as: each fi(x0)f_i(x^0)fi​(x0) finite, f0f_0f0​ finite on a neighbourhood of f(x0)f(x^0)f(x0), and the real function z↦f0(z)z\mapsto f_0(z)z↦f0​(z) differentiable there. The sum is a Minkowski sum, and a term with yi=0y_i=0yi​=0 is {0}\{0\}{0}, following the proof's convention (p. 225, where the terms with yi=0y_i=0yi​=0 are absorbed into the normal cones of dom⁡fi\operatorname{dom}f_idomfi​): v∈v\inv∈ RHS iff v=ν+∑iyiξiv=\nu+\sum_i y_i\xi_iv=ν+∑i​yi​ξi​ with ν∈Ndom⁡G(x0)\nu\in N_{\operatorname{dom}G}(x^0)ν∈NdomG​(x0) and ξi∈∂fi(x0)\xi_i\in\partial f_i(x^0)ξi​∈∂fi​(x0) whenever yi≠0y_i\ne0yi​=0. ri⁡\operatorname{ri}ri is Mathlib's intrinsicInterior.

Preamble
import Mathlib
import Definitions.Def_BregmanPPA_Convergence_Model
import Definitions.Def_BregmanPPA_IneqMult_Program

open Filter Topology InertialFB.IFB
Formal statement
namespace BregmanPPA.ProxMult

/-- Lemma A4 (a subdifferential chain rule), p. 224. `F i` are the paper's `fᵢ`, `F₀` is `f₀`
(nondecreasing in the sense of Lemma 2), `compositeFn F F₀` is `G`. "`f₀` differentiable at
`f(x⁰) ∈ ℝᵐ`" is: every `fᵢ(x⁰)` finite, `f₀` finite near `f(x⁰)`, and its real version
differentiable there. In the sum, a term with coefficient `0` is `{0}` (the paper's convention,
p. 225). -/
theorem lemmaA4 {n m : ℕ} (F : Fin m → BregmanPPA.IneqMult.E n → EReal) (F₀ : BregmanPPA.IneqMult.E m → EReal) (x₀ : BregmanPPA.IneqMult.E n)
    (hF : ∀ i, IsProperFn (F i) ∧ IsConvexFn (F i))
    (hri : ∃ x : BregmanPPA.IneqMult.E n, ∀ i, x ∈ intrinsicInterior ℝ {y : BregmanPPA.IneqMult.E n | F i y ≠ ⊤})
    (hF₀ : IsProperFn F₀ ∧ IsConvexFn F₀)
    (hmono : ∀ u v : BregmanPPA.IneqMult.E m, (∀ i, u i ≤ v i) → F₀ u ≤ F₀ v)
    (hfinite : ∀ i, F i x₀ ≠ ⊤)
    (hF₀near : ∀ᶠ z in 𝓝 (BregmanPPA.IneqMult.gvec F x₀), F₀ z ≠ ⊤)
    (hdiff : DifferentiableAt ℝ (fun z : BregmanPPA.IneqMult.E m => (F₀ z).toReal) (BregmanPPA.IneqMult.gvec F x₀)) :
    IsProperFn (BregmanPPA.IneqMult.compositeFn F F₀) ∧
    IsConvexFn (BregmanPPA.IneqMult.compositeFn F F₀) ∧
    BregmanPPA.Convergence.subdiffOp (BregmanPPA.IneqMult.compositeFn F F₀) x₀ =
      {v | ∃ ν : BregmanPPA.IneqMult.E n, ν ∈ BregmanPPA.IneqMult.normalCone {x | BregmanPPA.IneqMult.compositeFn F F₀ x ≠ ⊤} x₀ ∧
        ∃ ξ : Fin m → BregmanPPA.IneqMult.E n,
          (∀ i, (gradient (fun z : BregmanPPA.IneqMult.E m => (F₀ z).toReal) (BregmanPPA.IneqMult.gvec F x₀)) i ≠ 0 →
            IsSubgradient (F i) x₀ (ξ i)) ∧
          v = ν + ∑ i, (gradient (fun z : BregmanPPA.IneqMult.E m => (F₀ z).toReal) (BregmanPPA.IneqMult.gvec F x₀)) i • ξ i} := by sorry

end BregmanPPA.ProxMult
Source
Eckstein, Nonlinear proximal point algorithms using Bregman functions, with applications to convex programming, Math. Oper. Res. 18(1) (1993), p. 224, Lemma A4
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