Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 3 — the generalized proximal point algorithm converges weakly to a zero, or is unbounded

Proved
DouglasRachfordPPA.GenDR.generalized_proximal_point

by mikedeng1 · Oct 4, 2026 · Mathlib 0df444a (Lean v4.33.1)

monotone-operatorsp2o-batch-pfp1bp2o-gran-per-chapterp2o-plan-paperp2o-v1proximal-point-algorithmweak-convergence

Let H\mathcal HH be a real Hilbert space and TTT a maximal monotone operator on H\mathcal HH. Let {ck}\{c_k\}{ck​}, {ρk}\{\rho_k\}{ρk​}, {εk}\{\varepsilon_k\}{εk​} be real sequences and {zk}\{z^k\}{zk}, {wk}\{w^k\}{wk} sequences in H\mathcal HH (k≥0k\ge0k≥0) such that

zk+1=(1−ρk)zk+ρkwk,∥wk−(I+ckT)−1(zk)∥≤εk∀k≥0,z^{k+1}=(1-\rho_k)z^k+\rho_k w^k,\qquad \|w^k-(I+c_kT)^{-1}(z^k)\|\le\varepsilon_k\qquad\forall k\ge0,zk+1=(1−ρk​)zk+ρk​wk,∥wk−(I+ck​T)−1(zk)∥≤εk​∀k≥0,

where εk≥0\varepsilon_k\ge0εk​≥0, ∑k=0∞εk<∞\sum_{k=0}^\infty\varepsilon_k<\infty∑k=0∞​εk​<∞, inf⁡kρk>0\inf_k\rho_k>0infk​ρk​>0, sup⁡kρk<2\sup_k\rho_k<2supk​ρk​<2 and inf⁡kck>0\inf_k c_k>0infk​ck​>0. Then:

  1. if TTT has a zero, {zk}\{z^k\}{zk} converges weakly to a zero of TTT;
  2. if TTT has no zero, {zk}\{z^k\}{zk} is unbounded.

The theorem combines Rockafellar's inexact proximal point algorithm (variable stepsizes ckc_kck​) with the over/under-relaxation of Gol'shtein and Tret'yakov, in a general Hilbert space and including the case without zeros. Theorem 7 is obtained by applying it to the splitting operator.

Formalization Note The printed statement has two slips, corrected here: it reads zk+1=(1−ρk)zk+(1−ρk)wkz^{k+1}=(1-\rho_k)z^k+(1-\rho_k)w^kzk+1=(1−ρk​)zk+(1−ρk​)wk where the text before it (p. 9) and the proof (pp. 11, 14) use ρkwk\rho_k w^kρk​wk, and it writes (I+cT)−1(I+cT)^{-1}(I+cT)−1 where the proof uses (I+ckT)−1(I+c_kT)^{-1}(I+ck​T)−1 (Qk=I−JckTQ_k=I-J_{c_kT}Qk​=I−Jck​T​, p. 11). The resolvent (I+ckT)−1(I+c_kT)^{-1}(I+ck​T)−1 is a map Jk:H→HJ_k:\mathcal H\to\mathcal HJk​:H→H with 1ck(x−Jkx)∈T(Jkx)\tfrac1{c_k}(x-J_k x)\in T(J_kx)ck​1​(x−Jk​x)∈T(Jk​x) for all xxx (unique by Corollary 2.2). The conditions inf⁡ρk>0\inf\rho_k>0infρk​>0, sup⁡ρk<2\sup\rho_k<2supρk​<2 are encoded as ∃ ρ1,ρ2\exists\,\rho_1,\rho_2∃ρ1​,ρ2​ with 0<ρ1≤ρk≤ρ2<20<\rho_1\le\rho_k\le\rho_2<20<ρ1​≤ρk​≤ρ2​<2, and inf⁡ck>0\inf c_k>0infck​>0 as ∃ c0>0\exists\,c_0>0∃c0​>0 with c0≤ckc_0\le c_kc0​≤ck​ for all kkk; these are equivalent to the page and imply the page's ρk,ck≥0\rho_k,c_k\ge0ρk​,ck​≥0. Weak convergence is ⟨zk,y⟩→⟨z∗,y⟩\langle z^k,y\rangle\to\langle z^*,y\rangle⟨zk,y⟩→⟨z∗,y⟩ for all yyy; unboundedness is "the range of k↦zkk\mapsto z^kk↦zk is not bounded". Sequences are indexed from k=0k=0k=0.

Preamble
import Mathlib
import Definitions.Def_ThreeOpSplitting_Convergence_MonotoneOperators
import Definitions.Def_ThreeOpSplitting_Convergence_WeakConvergence

open InnerProductSpace ThreeOpSplitting.Convergence
Formal statement
namespace DouglasRachfordPPA.GenDR

/-- Theorem 3 (generalized proximal point algorithm), with the printed slips corrected:
`z (k+1) = (1 - ρ k) z k + ρ k w k` and `‖w k - J_{c_k T}(z k)‖ ≤ ε k`. The resolvent
`J_{c_k T}` is the map `J k` with `IsResolvent (c k) T (J k)`. -/
theorem generalized_proximal_point
    {H : Type*} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
    (T : H → Set H) (hT : IsMaximalMonotone T)
    (c ρ ε : ℕ → ℝ) (J : ℕ → H → H) (hJ : ∀ k, IsResolvent (c k) T (J k))
    (z w : ℕ → H)
    (hz : ∀ k, z (k + 1) = (1 - ρ k) • z k + ρ k • w k)
    (hw : ∀ k, ‖w k - J k (z k)‖ ≤ ε k)
    (hε_nonneg : ∀ k, 0 ≤ ε k) (hε_sum : Summable ε)
    (hρ : ∃ ρ₁ ρ₂ : ℝ, 0 < ρ₁ ∧ ρ₂ < 2 ∧ ∀ k, ρ₁ ≤ ρ k ∧ ρ k ≤ ρ₂)
    (hc : ∃ c₀ : ℝ, 0 < c₀ ∧ ∀ k, c₀ ≤ c k) :
    ((zer T).Nonempty → ∃ zs ∈ zer T, WeakTendsto z zs) ∧
    (zer T = ∅ → ¬ Bornology.IsBounded (Set.range z)) := by sorry

end DouglasRachfordPPA.GenDR
Source
Eckstein and Bertsekas, On the Douglas–Rachford Splitting Method and the Proximal Point Algorithm for Maximal Monotone Operators, MIT LIDS-P-1919 (October 1989), p. 10, Theorem 3 (slips corrected from p. 9 and the proof, pp. 11, 14)
Human review
  • Endorsed by Shuze Chen · Oct 4, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 4, 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