Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 6.3 (dual estimate), p. 28 — if η = ϵ₊/(1 − δ) < 1 then the dual step satisfies ‖q_s‖_s ≤ η

Open
SelfScaledIPM.ShortStep.lemma_6_3_qs_bound

by tianyipeng · Nov 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

interior-pointpath-followingself-scaled-coneshort-step

Let KKK be a self-scaled cone with ν\nuν-self-scaled barrier FFF and conjugate barrier F∗F_*F∗​, and let AAA be surjective. Let (x,y,s)∈N(β)(x,y,s)\in N(\beta)(x,y,s)∈N(β) with 0<β<10<\beta<10<β<1. Fix 0<κ<10<\kappa<10<κ<1, put μ+=(1−κ/ν) μ(x,s)\mu_+=(1-\kappa/\sqrt\nu)\,\mu(x,s)μ+​=(1−κ/ν​)μ(x,s) (6.4), let www be the scaling point of (x,s)(x,s)(x,s) (so F′′(w)x=sF''(w)x=sF′′(w)x=s), and let (qx,qy,qs)(q_x,q_y,q_s)(qx​,qy​,qs​) solve the short-step system (6.7):

F′′(w)qx+qs=s+μ+F′(x),Aqx=0,A∗qy+qs=0.F''(w)q_x+q_s=s+\mu_+F'(x),\qquad Aq_x=0,\qquad A^*q_y+q_s=0.F′′(w)qx​+qs​=s+μ+​F′(x),Aqx​=0,A∗qy​+qs​=0.

With δ=λ∞(x,s)\delta=\lambda_\infty(x,s)δ=λ∞​(x,s) and ϵ+=∥s/μ++F′(x)∥x\epsilon_+=\|s/\mu_++F'(x)\|_xϵ+​=∥s/μ+​+F′(x)∥x​, suppose η=ϵ+/(1−δ)<1\eta=\epsilon_+/(1-\delta)<1η=ϵ+​/(1−δ)<1 (6.11). Then the dual component of the step satisfies

∥qs∥s=⟨F∗′′(s)qs,qs⟩1/2≤η.\|q_s\|_s=\langle F_*''(s)q_s,q_s\rangle^{1/2}\le\eta .∥qs​∥s​=⟨F∗′′​(s)qs​,qs​⟩1/2≤η.

This is the second inequality of Lemma 6.3. The paper proves the first one, ∥qx∥x≤η\|q_x\|_x\le\eta∥qx​∥x​≤η, using Lemma 3.13 and the orthogonality ⟨qs,qx⟩=0\langle q_s,q_x\rangle=0⟨qs​,qx​⟩=0, and then says "similarly ∥qs∥s≤η\|q_s\|_s\le\eta∥qs​∥s​≤η". The dual argument uses the second half of (3.22), F∗′′(t)=[F′′(w)]−1F_*''(t)=[F''(w)]^{-1}F∗′′​(t)=[F′′(w)]−1 with t=−F′(w)t=-F'(w)t=−F′(w), the symmetric form qx+F∗′′(t)qs=x+μ+F∗′(s)q_x+F_*''(t)q_s=x+\mu_+F_*'(s)qx​+F∗′′​(t)qs​=x+μ+​F∗′​(s) of (6.7), and (3.3).

Formalization note. E∗E^*E∗ is identified with EEE, and ∥⋅∥s\|\cdot\|_s∥⋅∥s​ is lnorm (conj K F) s, the local norm at the dual point s∈int⁡K∗s\in\operatorname{int}K^*s∈intK∗ of the dual vector qsq_sqs​. The hypotheses are exactly those of the platform statement of Lemma 6.3.

Preamble
import Mathlib
import Definitions.Def_dualCone
import Definitions.Def_SelfScaledIPM_ShortStep_Setting
import Definitions.Def_SelfScaledIPM_ShortStep_Measures

open scoped InnerProductSpace
Formal statement
namespace SelfScaledIPM.ShortStep

/-- **Lemma 6.3, dual estimate** (p. 28): under the hypotheses of Lemma 6.3 —
`(x, y, s) ∈ N(β)`, `β, κ ∈ (0, 1)`, `µ₊ = (1 − κ/√ν)µ(x, s)` (6.4), `w` the scaling point of
`(x, s)`, `(q_x, q_y, q_s)` a solution of (6.7), and `η = ϵ₊/(1 − δ) < 1` (6.11) — the dual
step satisfies `‖q_s‖_s ≤ η`, the norm being the local norm of `q_s ∈ E*` at the dual point
`s ∈ int K*`, i.e. `⟨F*''(s) q_s, q_s⟩^{1/2}` (`lnorm (conj K F) s q_s`). -/
theorem lemma_6_3_qs_bound {n m : ℕ} (K : Set (EuclideanSpace ℝ (Fin n))) (hK : IsProperCone K)
    (F : EuclideanSpace ℝ (Fin n) → ℝ) (ν : ℝ) (hν : 1 ≤ ν) (hF : IsSelfScaledBarrier K F ν)
    (A : EuclideanSpace ℝ (Fin n) →L[ℝ] EuclideanSpace ℝ (Fin m)) (hA : Function.Surjective A)
    (b : EuclideanSpace ℝ (Fin m)) (c : EuclideanSpace ℝ (Fin n))
    (β κ : ℝ) (hβ0 : 0 < β) (hβ1 : β < 1) (hκ0 : 0 < κ) (hκ1 : κ < 1)
    (x : EuclideanSpace ℝ (Fin n)) (y : EuclideanSpace ℝ (Fin m)) (s : EuclideanSpace ℝ (Fin n))
    (hN : InN K F A b c ν β x y s)
    (w : EuclideanSpace ℝ (Fin n)) (hw : IsScalingPoint K F x s w)
    (qx : EuclideanSpace ℝ (Fin n)) (qy : EuclideanSpace ℝ (Fin m)) (qs : EuclideanSpace ℝ (Fin n))
    (hq : IsShortStepDir F A x s w (muPlus ν κ x s) qx qy qs)
    (hη : eta K F ν κ x s < 1) :
    lnorm (conj K F) s qs ≤ eta K F ν κ x s := by sorry

end SelfScaledIPM.ShortStep
Source
Nesterov & Todd, Primal-dual interior-point methods for self-scaled cones, SIAM J. Optim. 8 (1998) 324–364 (authors' copy, Cornell eCommons), p. 28, Lemma 6.3, second inequality ‖q_s‖_s ≤ η ("and similarly ‖q_s‖_s ≤ η" in the proof)

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