Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

§8.3.4, pp. 70–71 — the feature-split SVM z̄-update is shifted soft thresholding of v = Ax̄^{k+1} + u^k

Proved
BoydADMM.ModelFit.svm_zbar_shifted_soft_threshold

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

admmhinge-lossp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-bookp2o-v1support-vector-machine

Let N≥1N\ge1N≥1, ρ>0\rho>0ρ>0, and Ax‾,u∈Rm\overline{Ax},u\in\mathbb R^mAx,u∈Rm (the averaged partial predictor Ax‾k+1\overline{Ax}^{k+1}Axk+1 and the scaled dual variable uku^kuk); write v=Ax‾+uv=\overline{Ax}+uv=Ax+u. The zˉ\bar zzˉ-update of the feature-split support vector machine minimizes

1T(Nzˉ+1)++ρ2∥zˉ−Ax‾−u∥22,\mathbf 1^T(N\bar z+\mathbf 1)_++\frac{\rho}{2}\|\bar z-\overline{Ax}-u\|_2^2,1T(Nzˉ+1)+​+2ρ​∥zˉ−Ax−u∥22​,

where (w)+(w)_+(w)+​ is the componentwise positive part. A point zˉ∈Rm\bar z\in\mathbb R^mzˉ∈Rm minimizes this function if and only if, for every component jjj,

zˉj={vj−N/ρvj>−1/N+N/ρ−1/Nvj∈[−1/N, −1/N+N/ρ]vjvj<−1/N.\bar z_j=\begin{cases}v_j-N/\rho & v_j>-1/N+N/\rho\\ -1/N & v_j\in[-1/N,\,-1/N+N/\rho]\\ v_j & v_j<-1/N.\end{cases}zˉj​=⎩⎨⎧​vj​−N/ρ−1/Nvj​​vj​>−1/N+N/ρvj​∈[−1/N,−1/N+N/ρ]vj​<−1/N.​

In particular the minimizer is unique.

The zˉ\bar zzˉ-update thus splits into mmm scalar problems, each solved by this shifted soft thresholding.

Formalization Note The objective is exactly the one printed on p. 70, with coefficient ρ/2\rho/2ρ/2 on the quadratic; the general feature-split zˉ\bar zzˉ-update of p. 68 carries Nρ/2N\rho/2Nρ/2 instead, and the printed thresholds N/ρN/\rhoN/ρ are the ones that match the printed ρ/2\rho/2ρ/2. The book writes v=Ax‾k+1+uˉkv=\overline{Ax}^{k+1}+\bar u^kv=Axk+1+uˉk; after the reduction of §8.3 all duals are equal, so uˉk=uk\bar u^k=u^kuˉk=uk.

Preamble
import Mathlib
import Definitions.Def_BoydADMM_ModelFit_Basic

open Matrix
Formal statement
namespace BoydADMM.ModelFit

/-- §8.3.4, pp. 70–71: the feature-split SVM `z̄`-update
`argmin_{z̄} (1ᵀ(N z̄ + 1)₊ + (ρ/2)‖z̄ − Ax̄ − u‖₂²)` is the shifted soft thresholding of
`v = Ax̄ + u`, componentwise. -/
theorem svm_zbar_shifted_soft_threshold {N m : ℕ} (hN : 0 < N) (ρ : ℝ) (hρ : 0 < ρ)
    (Axbar u zb : EuclideanSpace ℝ (Fin m)) :
    IsMinOn (fun z : EuclideanSpace ℝ (Fin m) =>
        (∑ j, max ((N : ℝ) * z j + 1) 0) + ρ / 2 * ‖z - Axbar - u‖ ^ 2) Set.univ zb ↔
      ∀ j, zb j = shiftedSoftThreshold N ρ (Axbar j + u j) := by sorry

end BoydADMM.ModelFit
Source
Boyd, Parikh, Chu, Peleato, Eckstein, Distributed Optimization and Statistical Learning via the Alternating Direction Method of Multipliers, Found. Trends Mach. Learn. 3(1) (2011), pp. 70–71, §8.3.4
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