Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 8 — min⁡0≤k≤NF(xk)≤F∗+(αβmc2+ϵ)/2\min_{0\le k\le N}F(x_k)\le F^*+(\alpha\beta mc^2+\epsilon)/2min0≤k≤N​F(xk​)≤F∗+(αβmc2+ϵ)/2 a.s., with E{N}≤m dist(x0;X∗)2/(αϵ)E\{N\}\le m\,\mathrm{dist}(x_0;X^*)^2/(\alpha\epsilon)E{N}≤mdist(x0​;X∗)2/(αϵ)

Proved
IncProx.Randomized.proposition8

by mikedeng1 · Oct 8, 2026 · Mathlib 0df444a (Lean v4.33.1)

convex-optimizationincremental-methodsp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1stochastic-optimization

Let xkx_kxk​ be generated as in Proposition 7 (one of the randomized methods (42)–(44), constant stepsize α>0\alpha>0α>0, Assumption 3 or 4 with constant ccc), assume that X∗X^*X∗ is nonempty, and let ϵ>0\epsilon>0ϵ>0. Let

L={x∈X ∣ F(x)<F∗+αβmc2+ϵ2},β=5,L=\Bigl\{x\in X\ \Bigm|\ F(x)<F^*+\frac{\alpha\beta mc^2+\epsilon}{2}\Bigr\},\qquad\beta=5,L={x∈X ​ F(x)<F∗+2αβmc2+ϵ​},β=5,

and let N=min⁡{k≥0∣xk∈L}N=\min\{k\ge0\mid x_k\in L\}N=min{k≥0∣xk​∈L} be the first iteration at which the run enters LLL (N=∞N=\inftyN=∞ if it never does). Then NNN is a random variable, with probability 1

min⁡0≤k≤NF(xk)≤F∗+αβmc2+ϵ2,(54)\min_{0\le k\le N}F(x_k)\le F^*+\frac{\alpha\beta mc^2+\epsilon}{2},\qquad(54)0≤k≤Nmin​F(xk​)≤F∗+2αβmc2+ϵ​,(54)

and

E{N}≤m dist(x0;X∗)2αϵ.(55)E\{N\}\le m\,\frac{\mathrm{dist}(x_0;X^*)^2}{\alpha\epsilon}.\qquad(55)E{N}≤mαϵdist(x0​;X∗)2​.(55)

The bound (55) is on the expected number of iterations needed to reach the error tolerance, to be compared with the deterministic count of Proposition 5 for the cyclic order.

Formalization Note The paper writes "where NNN is a random variable with (55)"; an arbitrary NNN would make the statement nearly empty (N=0N=0N=0 satisfies (55) whenever (54) holds at k=0k=0k=0), so NNN is the paper's own random variable from the proof of Proposition 8 (p. 19): the first index kkk with xk∈Lx_k\in Lxk​∈L, valued in N∪{∞}\mathbb N\cup\{\infty\}N∪{∞}. With probability 1 it is finite, and (54) is stated as the existence of k≤Nk\le Nk≤N with F(xk)F(x_k)F(xk​) below the bound. Added hypothesis: the starting point lies in XXX. The proof applies Eq. (50) at k=0k=0k=0, which for (42) and (44) uses x0=PX(x0)x_0=P_X(x_0)x0​=PX​(x0​); the paper does not state this. dist(x0;X∗)\mathrm{dist}(x_0;X^*)dist(x0​;X∗) is Metric.infDist, F∗F^*F∗ is the extended-real optimal value (finite since X∗≠∅X^*\neq\emptysetX∗=∅), and E{N}E\{N\}E{N} is the lower Lebesgue integral of NNN in [0,∞][0,\infty][0,∞].

Preamble
import Mathlib
import Definitions.Def_IncProx_Randomized_Basic
Formal statement
namespace IncProx.Randomized

open Filter Topology MeasureTheory

/-- Proposition 8 (p. 19), for a run as in Proposition 7 started inside X: assume X* ≠ ∅ and
ε > 0, and let N be the first k with x_k in the level set
L = {x ∈ X | F(x) < F* + (αβmc² + ε)/2}, β = 5 (proof of Prop. 8, p. 19). Then N is a random
variable; with probability 1, N < ∞ and min_{0≤k≤N} F(x_k) ≤ F* + (αβmc² + ε)/2 (54); and
E{N} ≤ m dist(x₀; X*)²/(αε) (55). The hypothesis x₀ ∈ X is added. -/
theorem proposition8 {Ω : Type*} [MeasurableSpace Ω] {n m : ℕ} [NeZero m]
    (μ : Measure Ω) [IsProbabilityMeasure μ] (P : Problem n m) (hP : P.Standing) (A : Alg)
    (α : ℝ) (hα : 0 < α) (ω : ℕ → Ω → Fin m) (x : ℕ → Ω → Vec n)
    (zc gFc gHc xc : ℕ → Fin m → Ω → Vec n) (x₀ : Vec n) (c : ℝ)
    (hR : RandHyp μ P A (fun _ => α) ω x zc gFc gHc xc x₀ c)
    (hx₀ : x₀ ∈ P.X) (hXstar : P.Xstar.Nonempty) (ε : ℝ) (hε : 0 < ε) :
    let L : Set (Vec n) :=
      {y | y ∈ P.X ∧ ((P.F y : ℝ) : EReal) < P.Fstar + (((α * 5 * m * c ^ 2 + ε) / 2 : ℝ) : EReal)}
    let N : Ω → ℕ∞ := firstHit x L
    Measurable N ∧
      (∀ᵐ s ∂μ, N s ≠ ⊤ ∧ ∃ k : ℕ, (k : ℕ∞) ≤ N s ∧
        ((P.F (x k s) : ℝ) : EReal) ≤ P.Fstar + (((α * 5 * m * c ^ 2 + ε) / 2 : ℝ) : EReal)) ∧
      ∫⁻ s, ENat.toENNReal (N s) ∂μ ≤
        ENNReal.ofReal (m * Metric.infDist x₀ P.Xstar ^ 2 / (α * ε)) := by sorry

end IncProx.Randomized
Source
Bertsekas, Incremental Proximal Methods for Large Scale Convex Optimization, LIDS-P-2847 (rev. March 2011), p. 19, Proposition 8 (L and N from its proof, p. 19)
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