Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

(3.4)–(3.5), p. 11 — on (3.1), x₁^{k+1} is given by (3.4) and L(x₂, x₃, μ)^{k+1} = R(x₂, x₃, μ)^k with μ = λ/β

Proved
ExtADMM.Diverge.eq_3_4_3_5

by mikedeng1 · Oct 7, 2026 · Mathlib 0df444a (Lean v4.33.1)

admmlinear-recursionp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1splitting-methods

Let a1,a2,a3∈R3a_1,a_2,a_3\in\mathbb R^3a1​,a2​,a3​∈R3 be column vectors such that the 3×33\times33×3 matrix [A1,A2,A3]=[a1,a2,a3][A_1,A_2,A_3]=[a_1,a_2,a_3][A1​,A2​,A3​]=[a1​,a2​,a3​] is nonsingular, and consider the linear system (3.1), A1x1+A2x2+A3x3=0A_1x_1+A_2x_2+A_3x_3=0A1​x1​+A2​x2​+A3​x3​=0 with xi∈Rx_i\in\mathbb Rxi​∈R, as an instance of problem (1.1). Let β>0\beta>0β>0 and let (x1k,x2k,x3k,λk)k≥0(x_1^k,x_2^k,x_3^k,\lambda^k)_{k\ge0}(x1k​,x2k​,x3k​,λk)k≥0​ be any run of the direct extension of ADMM (1.5) on it. Put μk=λk/β\mu^k=\lambda^k/\betaμk=λk/β. Then for every k≥0k\ge0k≥0,

x1k+1=1A1TA1(−A1TA2x2k−A1TA3x3k+A1Tμk)(3.4)x_1^{k+1}=\frac{1}{A_1^TA_1}\bigl(-A_1^TA_2x_2^k-A_1^TA_3x_3^k+A_1^T\mu^k\bigr)\qquad(3.4)x1k+1​=A1T​A1​1​(−A1T​A2​x2k​−A1T​A3​x3k​+A1T​μk)(3.4)

and

L(x2k+1x3k+1μk+1)=R(x2kx3kμk)(3.5)L\begin{pmatrix}x_2^{k+1}\\x_3^{k+1}\\\mu^{k+1}\end{pmatrix}=R\begin{pmatrix}x_2^{k}\\x_3^{k}\\\mu^{k}\end{pmatrix}\qquad(3.5)L​x2k+1​x3k+1​μk+1​​=R​x2k​x3k​μk​​(3.5)

with the 5×55\times55×5 matrices LLL of (3.6) and RRR of (3.7).

The point is that neither LLL nor RRR involves β\betaβ: in the scaled variables μ=λ/β\mu=\lambda/\betaμ=λ/β the iteration is the same linear map for every penalty parameter. This is what makes the divergence of the example hold for every β>0\beta>0β>0 at once.

Formalization Note. The run is any sequence satisfying the minimiser conditions of (1.5); no particular minimiser is chosen. x10x_1^0x10​ is never read. μk\mu^kμk has components λik/β\lambda^k_i/\betaλik​/β, indexed 0,1,20,1,20,1,2 in Lean.

Preamble
import Mathlib
import Definitions.Def_ExtADMM_Diverge_Setting

open Matrix Filter Topology
Formal statement
namespace ExtADMM.Diverge

/-- (3.4)–(3.5), p. 11. For the linear system (3.1) with columns `a₁, a₂, a₃ ∈ ℝ³` such that
`[A₁, A₂, A₃]` is nonsingular, and any `β > 0`, every run of the direct extension of ADMM (1.5)
satisfies, for every `k`, the closed form (3.4) of `x₁^{k+1}` and the recursion (3.5)
`L (x₂^{k+1}, x₃^{k+1}, μ^{k+1}) = R (x₂ᵏ, x₃ᵏ, μᵏ)` in the variables `μ = λ/β`. -/
theorem eq_3_4_3_5 (a1 a2 a3 : Fin 3 → ℝ) (hA : (colMat a1 a2 a3).det ≠ 0)
    (β : ℝ) (hβ : 0 < β)
    (x1 x2 x3 : ℕ → Fin 1 → ℝ) (lam : ℕ → Fin 3 → ℝ)
    (hrun : (instance31 a1 a2 a3).IsRun15 β x1 x2 x3 lam) (k : ℕ) :
    x1 (k+1) 0 = (1 / (a1 ⬝ᵥ a1)) *
        (-(a1 ⬝ᵥ a2) * x2 k 0 - (a1 ⬝ᵥ a3) * x3 k 0 + a1 ⬝ᵥ (β⁻¹ • lam k)) ∧
      Lmat a2 a3 *ᵥ stateVec β (x2 (k+1)) (x3 (k+1)) (lam (k+1)) =
        Rmat a1 a2 a3 *ᵥ stateVec β (x2 k) (x3 k) (lam k) := by sorry

end ExtADMM.Diverge
Source
Chen, He, Ye & Yuan, The direct extension of ADMM for multi-block convex minimization problems is not necessarily convergent, Math. Program., DOI 10.1007/s10107-014-0826-5 (authors' version of January 22, 2014), p. 11, (3.3)–(3.5)
Human review
  • Endorsed by Shuze Chen · Oct 8, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 8, 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