Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 4.1, pp. 480–481 — on a compact level set where f is continuous, essentially cyclic BCD is defined and bounded; (a)–(c) on its cluster points

Proved
TsengBCD.Stationary.theorem_4_1

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

block-coordinate-descentcluster-pointconvergencep2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1pseudoconvex

Let f:Rn1+⋯+nN→R∪{∞}f:\mathbb R^{n_1+\cdots+n_N}\to\mathbb R\cup\{\infty\}f:Rn1​+⋯+nN​→R∪{∞} never take the value −∞-\infty−∞ and let x0∈dom⁡fx^0\in\operatorname{dom}fx0∈domf. Assume that the level set

X0={x:f(x)≤f(x0)}X^0=\{x: f(x)\le f(x^0)\}X0={x:f(x)≤f(x0)}

is compact and that fff is continuous on X0X^0X0. Then the sequence {xr}r=0,1,…\{x^r\}_{r=0,1,\dots}{xr}r=0,1,…​ generated by the BCD method is defined (for every choice of blocks, the minimizations (2) can be carried out), and every such sequence generated with the essentially cyclic rule is bounded. Moreover:

  1. (a) If f(x1,…,xN)f(x_1,\dots,x_N)f(x1​,…,xN​) is pseudoconvex in (xk,xi)(x_k,x_i)(xk​,xi​) for every i,k∈{1,…,N}i,k\in\{1,\dots,N\}i,k∈{1,…,N} and fff is regular at every x∈X0x\in X^0x∈X0, then every cluster point of {xr}\{x^r\}{xr} (essentially cyclic rule) is a stationary point of fff.
  2. (b) If f(x1,…,xN)f(x_1,\dots,x_N)f(x1​,…,xN​) is pseudoconvex in (xk,xi)(x_k,x_i)(xk​,xi​) for every i,k∈{1,…,N−1}i,k\in\{1,\dots,N-1\}i,k∈{1,…,N−1}, fff is regular at every x∈X0x\in X^0x∈X0, and the cyclic rule is used, then every cluster point of {xr}r≡(N−1) mod N\{x^r\}_{r\equiv(N-1)\bmod N}{xr}r≡(N−1)modN​ is a stationary point of fff.
  3. (c) If f(x1,…,xN)f(x_1,\dots,x_N)f(x1​,…,xN​) has at most one minimum in xkx_kxk​ for k=2,…,N−1k=2,\dots,N-1k=2,…,N−1 and the cyclic rule is used, then every cluster point zzz of {xr}r≡(N−1) mod N\{x^r\}_{r\equiv(N-1)\bmod N}{xr}r≡(N−1)modN​ is a coordinatewise minimum point of fff; if in addition fff is regular at zzz, then zzz is a stationary point of fff.

Here r≡(N−1) mod Nr\equiv(N-1)\bmod Nr≡(N−1)modN means r=N−1,2N−1,3N−1,…r=N-1,2N-1,3N-1,\dotsr=N−1,2N−1,3N−1,…. Theorem 4.1 unifies and extends results of Auslender for the convex nondifferentiable case and of Grippo–Sciandrone, Luenberger and Zadeh for the differentiable case.

Formalization Note. "Defined" is formalized as: for every index sequence a BCD run from x0x^0x0 exists (stronger than for essentially cyclic sequences only). "fff is continuous on X0X^0X0" is formalized as continuity of fff, as a map into R∪{∞}\mathbb R\cup\{\infty\}R∪{∞}, at every point of X0X^0X0; under the weaker reading (continuity of the restriction f∣X0f|_{X^0}f∣X0​) the printed statement is false, by an explicit two-block counterexample given in the mission description. Blocks are 0-based: paper block kkk is Lean index k−1k-1k−1, {1,…,N−1}\{1,\dots,N-1\}{1,…,N−1} is {k:k<N−1}\{k: k<N-1\}{k:k<N−1}, {2,…,N−1}\{2,\dots,N-1\}{2,…,N−1} is {k:1≤k<N−1}\{k: 1\le k<N-1\}{k:1≤k<N−1}. s r is the block chosen at iteration r+1r+1r+1; the cyclic rule is s r = r mod N; the subsequence r≡(N−1) mod Nr\equiv(N-1)\bmod Nr≡(N−1)modN is m↦xNm+(N−1)m\mapsto x^{Nm+(N-1)}m↦xNm+(N−1). Pseudoconvexity in (xk,xi)(x_k,x_i)(xk​,xi​) and "at most one minimum in xkx_kxk​" are read blockwise, as defined in the definitions item.

Preamble
import Mathlib
import Definitions.Def_TsengBCD_Stationary_Setting
Formal statement
namespace TsengBCD.Stationary

open Filter Topology

theorem theorem_4_1 {N : ℕ} [NeZero N] {n : Fin N → ℕ} (f : X n → EReal)
    (hbot : ∀ x, f x ≠ ⊥) (x0 : X n) (hx0 : f x0 ≠ ⊤) (hcpt : IsCompact (levelSet f x0))
    (hcont : ∀ y ∈ levelSet f x0, ContinuousAt f y) :
    -- defined: for every index sequence a BCD run from x⁰ exists
    (∀ s : ℕ → Fin N, ∃ x : ℕ → X n, x 0 = x0 ∧ IsBCDRun f s x) ∧
    -- bounded
    (∀ (s : ℕ → Fin N) (x : ℕ → X n), IsEssCyclic s → x 0 = x0 → IsBCDRun f s x →
      Bornology.IsBounded (Set.range x)) ∧
    -- (a)
    ((∀ i k : Fin N, PseudoconvexIn f {k, i}) → (∀ y ∈ levelSet f x0, IsRegularAt f y) →
      ∀ (s : ℕ → Fin N) (x : ℕ → X n), IsEssCyclic s → x 0 = x0 → IsBCDRun f s x →
        ∀ z, MapClusterPt z atTop x → IsStationary f z) ∧
    -- (b)
    ((∀ i k : Fin N, i.val < N - 1 → k.val < N - 1 → PseudoconvexIn f {k, i}) →
      (∀ y ∈ levelSet f x0, IsRegularAt f y) →
      ∀ (s : ℕ → Fin N) (x : ℕ → X n), IsCyclic s → x 0 = x0 → IsBCDRun f s x →
        ∀ z, MapClusterPt z atTop (fun m : ℕ => x (N * m + (N - 1))) → IsStationary f z) ∧
    -- (c)
    ((∀ k : Fin N, 1 ≤ k.val → k.val < N - 1 → AtMostOneMinIn f k) →
      ∀ (s : ℕ → Fin N) (x : ℕ → X n), IsCyclic s → x 0 = x0 → IsBCDRun f s x →
        ∀ z, MapClusterPt z atTop (fun m : ℕ => x (N * m + (N - 1))) →
          IsCoordMin f z ∧ (IsRegularAt f z → IsStationary f z)) := by sorry

end TsengBCD.Stationary
Source
Tseng, Convergence of a block coordinate descent method for nondifferentiable minimization, J. Optim. Theory Appl. 109 (2001), pp. 480–481, Theorem 4.1
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