Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Corollary 1 — subgradient rate L~2(24κ2(N+1)+1−1)/(N+1)\widetilde L^2(2\sqrt{4\kappa^2(N+1)+1}-1)/(N+1)L2(24κ2(N+1)+1​−1)/(N+1) on weakly convex functions

Proved
BnBPEP.WeakCvx.corollary_1

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

convergence-ratemoreau-envelopep2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1subgradient-methodweakly-convex

Let N∈NN\in\mathbb NN∈N, L>0L>0L>0, and let f∈W1,Lf\in\mathcal W_{1,L}f∈W1,L​ (f+12∥⋅∥2f+\frac12\|\cdot\|^2f+21​∥⋅∥2 convex, all subgradients of norm at most LLL) have a global minimizer x⋆x_\starx⋆​. Let R>0R>0R>0 and κ=R/L\kappa=R/Lκ=R/L. Run the subgradient method

xk+1=xk−h f′(xk),k∈[0:N],x_{k+1}=x_k-h\,f'(x_k),\qquad k\in[0:N],xk+1​=xk​−hf′(xk​),k∈[0:N],

with arbitrary subgradients f′(xk)∈∂f(xk)f'(x_k)\in\partial f(x_k)f′(xk​)∈∂f(xk​) and the stepsize

h=4κ2(N+1)+12(N+1),h=\frac{\sqrt{4\kappa^2(N+1)+1}}{2(N+1)},h=2(N+1)4κ2(N+1)+1​​,

and assume, as in the setup of Theorem 1, that h≤12h\le\tfrac12h≤21​. Let yk=prox(1/2)f(xk)y_k=\mathrm{prox}_{(1/2)f}(x_k)yk​=prox(1/2)f​(xk​) for k∈[0:N+1]k\in[0:N+1]k∈[0:N+1] and assume the initial condition f(y0)−f(x⋆)+∥x0−y0∥2≤R2f(y_0)-f(x_\star)+\|x_0-y_0\|^2\le R^2f(y0​)−f(x⋆​)+∥x0​−y0​∥2≤R2. Then

1N+1∑i=0N∥∇f(1/2)(xi)∥2 ≤ L2(24κ2(N+1)+1−1)N+1,\frac{1}{N+1}\sum_{i=0}^{N}\|\nabla f_{(1/2)}(x_i)\|^2\ \le\ \frac{L^2\big(2\sqrt{4\kappa^2(N+1)+1}-1\big)}{N+1},N+11​i=0∑N​∥∇f(1/2)​(xi​)∥2 ≤ N+1L2(24κ2(N+1)+1​−1)​,

where f(1/2)(x)=min⁡y{f(y)+∥y−x∥2}f_{(1/2)}(x)=\min_y\{f(y)+\|y-x\|^2\}f(1/2)​(x)=miny​{f(y)+∥y−x∥2} is the Moreau envelope of fff.

This is the explicit rate of the subgradient method on nonsmooth weakly convex functions obtained by Das Gupta, Van Parys and Ryu with the BnB-PEP methodology; it improves on the rate L2 4κ/N+1L^2\,4\kappa/\sqrt{N+1}L24κ/N+1​ of Davis and Drusvyatskiy for all large enough NNN. Since the minimum of the summands is at most their average, it also bounds min⁡i∈[0:N]∥∇f(1/2)(xi)∥2\min_{i\in[0:N]}\|\nabla f_{(1/2)}(x_i)\|^2mini∈[0:N]​∥∇f(1/2)​(xi​)∥2.

Formalization Note "In the setup of Theorem 1" includes Theorem 1's requirement h∈(0,12]h\in(0,\frac12]h∈(0,21​]; for this stepsize it reads 4κ2(N+1)+1≤(N+1)24\kappa^2(N+1)+1\le(N+1)^24κ2(N+1)+1≤(N+1)2 and is stated as the explicit hypothesis h≤12h\le\frac12h≤21​ (it fails, e.g., for N=0N=0N=0). LLL is the paper's L~\widetilde LL (normalized weak-convexity modulus ρ=1\rho=1ρ=1), and κ=R/L\kappa=R/Lκ=R/L is written inline. The prox points are hypotheses IsProxPoint f (1/2) (x k) (y k), the subdifferential is the weak-convexity inequality with modulus 111, and ∇f(1/2)\nabla f_{(1/2)}∇f(1/2)​ is the gradient of the real-valued Moreau envelope moreauEnv 2 f.

Preamble
import Mathlib
import Definitions.Def_SAGA_Convex_IsProxPoint
import Definitions.Def_BnBPEP_WeakCvx_IsWeakSubgrad
import Definitions.Def_BnBPEP_WeakCvx_InWeakClass
import Definitions.Def_BnBPEP_WeakCvx_moreauEnv
Formal statement
namespace BnBPEP.WeakCvx

/-- Corollary 1, p. 625: in the setup of Theorem 1 (which requires `h ∈ (0, 1/2]`), the stepsize
`h = √(4κ²(N+1) + 1) / (2(N+1))` with `κ = R/L` yields
`(1/(N+1)) Σ_{i=0}^N ‖∇f_{(1/2)}(x_i)‖² ≤ L² (2√(4κ²(N+1) + 1) - 1)/(N+1)`.
Here `L` is the paper's `L̃` (subgradient bound after normalizing `ρ = 1`). -/
theorem corollary_1 {d : ℕ} (f : EuclideanSpace ℝ (Fin d) → ℝ) (L : ℝ)
    (hL : 0 < L) (hf : InWeakClass 1 L f)
    (xstar : EuclideanSpace ℝ (Fin d)) (hxstar : ∀ z, f xstar ≤ f z)
    (R : ℝ) (hR : 0 < R) (N : ℕ) (h : ℝ)
    (hhdef : h = √(4 * (R / L) ^ 2 * ((N : ℝ) + 1) + 1) / (2 * ((N : ℝ) + 1)))
    (hh : h ≤ 1 / 2)
    (x g y : ℕ → EuclideanSpace ℝ (Fin d))
    (hg : ∀ k ≤ N, IsWeakSubgrad 1 f (x k) (g k))
    (hstep : ∀ k ≤ N, x (k + 1) = x k - h • g k)
    (hy : ∀ k ≤ N + 1, SAGA.Convex.IsProxPoint f (1 / 2) (x k) (y k))
    (hinit : f (y 0) - f xstar + ‖x 0 - y 0‖ ^ 2 ≤ R ^ 2) :
    1 / ((N : ℝ) + 1) * ∑ i ∈ Finset.range (N + 1), ‖gradient (moreauEnv 2 f) (x i)‖ ^ 2
      ≤ L ^ 2 * (2 * √(4 * (R / L) ^ 2 * ((N : ℝ) + 1) + 1) - 1) / ((N : ℝ) + 1) := by sorry

end BnBPEP.WeakCvx
Source
Das Gupta, Van Parys, Ryu, Branch-and-bound performance estimation programming, Math. Program. 204 (2024), §6.3.3, p. 625, Corollary 1 (κ = R/L defined on p. 624, L the normalized bound L̃)
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