Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 4.1, p. 13 — a Newton step of length α scales the three residuals and the complementarity gap by 1 − α(1 − γ)

Proved
ConicQuadIPM.NewtonStep.lemma_4_1

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

conic-optimizationhomogeneous-modelinterior-pointnewton-directionp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1second-order-cone

Let K=K1×⋯×KkK = K^1\times\cdots\times K^kK=K1×⋯×Kk be a product of cones, each the half-line R+\mathbb R_+R+​, a quadratic cone or a rotated quadratic cone, with ni=1n^i = 1ni=1 for R+\mathbb R_+R+​, ni≥1n^i\ge1ni≥1 for a quadratic cone and ni≥2n^i\ge2ni≥2 for a rotated one. Let A∈Rm×nA\in\mathbb R^{m\times n}A∈Rm×n, b∈Rmb\in\mathbb R^mb∈Rm, c∈Rnc\in\mathbb R^nc∈Rn, let (x(0),τ(0),y(0),s(0),κ(0))(x^{(0)},\tau^{(0)},y^{(0)},s^{(0)},\kappa^{(0)})(x(0),τ(0),y(0),s(0),κ(0)) be any point and γ∈[0,1]\gamma\in[0,1]γ∈[0,1], and let (dx,dτ,dy,ds,dκ)(d_x,d_\tau,d_y,d_s,d_\kappa)(dx​,dτ​,dy​,ds​,dκ​) solve the Newton system (22):

Adx−bdτ=(γ−1)(Ax(0)−bτ(0)),ATdy+ds−cdτ=(γ−1)(ATy(0)+s(0)−cτ(0)),−cTdx+bTdy−dκ=(γ−1)(−cTx(0)+bTy(0)−κ(0)),X(0)Tds+S(0)Tdx=−X(0)S(0)e+γμ(0)e,τ(0)dκ+κ(0)dτ=−τ(0)κ(0)+γμ(0),\begin{aligned} A d_x - b d_\tau &= (\gamma-1)(Ax^{(0)} - b\tau^{(0)}),\\ A^Td_y + d_s - c d_\tau &= (\gamma-1)(A^Ty^{(0)} + s^{(0)} - c\tau^{(0)}),\\ -c^Td_x + b^Td_y - d_\kappa &= (\gamma-1)(-c^Tx^{(0)} + b^Ty^{(0)} - \kappa^{(0)}),\\ X^{(0)}Td_s + S^{(0)}Td_x &= -X^{(0)}S^{(0)}e + \gamma\mu^{(0)}e,\\ \tau^{(0)}d_\kappa + \kappa^{(0)}d_\tau &= -\tau^{(0)}\kappa^{(0)} + \gamma\mu^{(0)}, \end{aligned}Adx​−bdτ​ATdy​+ds​−cdτ​−cTdx​+bTdy​−dκ​X(0)Tds​+S(0)Tdx​τ(0)dκ​+κ(0)dτ​​=(γ−1)(Ax(0)−bτ(0)),=(γ−1)(ATy(0)+s(0)−cτ(0)),=(γ−1)(−cTx(0)+bTy(0)−κ(0)),=−X(0)S(0)e+γμ(0)e,=−τ(0)κ(0)+γμ(0),​

where X(0)=diag⁡(mat⁡(Tix(0)i))X^{(0)} = \operatorname{diag}(\operatorname{mat}(T^ix^{(0)i}))X(0)=diag(mat(Tix(0)i)), S(0)=diag⁡(mat⁡(Tis(0)i))S^{(0)} = \operatorname{diag}(\operatorname{mat}(T^is^{(0)i}))S(0)=diag(mat(Tis(0)i)), T=diag⁡(Ti)T = \operatorname{diag}(T^i)T=diag(Ti), eee stacks the first unit vectors and μ(0)=((x(0))Ts(0)+τ(0)κ(0))/(k+1)\mu^{(0)} = ((x^{(0)})^Ts^{(0)} + \tau^{(0)}\kappa^{(0)})/(k+1)μ(0)=((x(0))Ts(0)+τ(0)κ(0))/(k+1). For a step size α∈[0,1]\alpha\in[0,1]α∈[0,1] let (x(1),τ(1),y(1),s(1),κ(1))=(x(0),τ(0),y(0),s(0),κ(0))+α(dx,dτ,dy,ds,dκ)(x^{(1)},\tau^{(1)},y^{(1)},s^{(1)},\kappa^{(1)}) = (x^{(0)},\tau^{(0)},y^{(0)},s^{(0)},\kappa^{(0)}) + \alpha(d_x,d_\tau,d_y,d_s,d_\kappa)(x(1),τ(1),y(1),s(1),κ(1))=(x(0),τ(0),y(0),s(0),κ(0))+α(dx​,dτ​,dy​,ds​,dκ​) as in (23). Then

Ax(1)−bτ(1)=(1−α(1−γ))(Ax(0)−bτ(0)),ATy(1)+s(1)−cτ(1)=(1−α(1−γ))(ATy(0)+s(0)−cτ(0)),−cTx(1)+bTy(1)−κ(1)=(1−α(1−γ))(−cTx(0)+bTy(0)−κ(0)),dxTds+dτdκ=0,(x(1))Ts(1)+τ(1)κ(1)=(1−α(1−γ))((x(0))Ts(0)+τ(0)κ(0)).\begin{aligned} Ax^{(1)} - b\tau^{(1)} &= (1-\alpha(1-\gamma))(Ax^{(0)} - b\tau^{(0)}),\\ A^Ty^{(1)} + s^{(1)} - c\tau^{(1)} &= (1-\alpha(1-\gamma))(A^Ty^{(0)} + s^{(0)} - c\tau^{(0)}),\\ -c^Tx^{(1)} + b^Ty^{(1)} - \kappa^{(1)} &= (1-\alpha(1-\gamma))(-c^Tx^{(0)} + b^Ty^{(0)} - \kappa^{(0)}),\\ d_x^Td_s + d_\tau d_\kappa &= 0,\\ (x^{(1)})^Ts^{(1)} + \tau^{(1)}\kappa^{(1)} &= (1-\alpha(1-\gamma))\big((x^{(0)})^Ts^{(0)} + \tau^{(0)}\kappa^{(0)}\big). \end{aligned}Ax(1)−bτ(1)ATy(1)+s(1)−cτ(1)−cTx(1)+bTy(1)−κ(1)dxT​ds​+dτ​dκ​(x(1))Ts(1)+τ(1)κ(1)​=(1−α(1−γ))(Ax(0)−bτ(0)),=(1−α(1−γ))(ATy(0)+s(0)−cτ(0)),=(1−α(1−γ))(−cTx(0)+bTy(0)−κ(0)),=0,=(1−α(1−γ))((x(0))Ts(0)+τ(0)κ(0)).​

The lemma says that one Newton step moves the homogeneous model's infeasibility and its complementarity gap toward zero at exactly the same rate 1−α(1−γ)1-\alpha(1-\gamma)1−α(1−γ). This is the basis of the convergence analysis of the homogeneous algorithm and of its stopping criteria.

Formalization Note Indices are 0-based and vectors are stored block by block; Ax=∑iAixiAx = \sum_iA^ix^iAx=∑i​Aixi, (ATy)i=(Ai)Ty(A^Ty)^i = (A^i)^Ty(ATy)i=(Ai)Ty, cTx=∑i(ci)Txic^Tx = \sum_i(c^i)^Tx^icTx=∑i​(ci)Txi, and the second identity of (22) and of the conclusion is stated block by block, as is the fourth line of (22). Three printed slips are corrected: the third line of (22) prints κ\kappaκ for κ(0)\kappa^{(0)}κ(0) (with a free κ\kappaκ the third identity is false); the fourth identity of (24) prints dxTdsTd_x^Td_s^TdxT​dsT​ for dxTdsd_x^Td_sdxT​ds​; and the proof's constant (γ−1)μ(0)k(\gamma-1)\mu^{(0)}k(γ−1)μ(0)k on p. 14 should be (γ−1)μ(0)(k+1)(\gamma-1)\mu^{(0)}(k+1)(γ−1)μ(0)(k+1) (this affects only the proof). The dimension conventions are the paper's implicit ones and are needed: without them a block of dimension 000 breaks eTe=ke^Te = keTe=k. The hypotheses α∈[0,1]\alpha\in[0,1]α∈[0,1] (from (23)) and γ∈[0,1]\gamma\in[0,1]γ∈[0,1] (from (21)) are kept as on the page, although the identities hold for all real α\alphaα, γ\gammaγ. No interiority of the current point is assumed, as in the lemma.


Retired 2026-10-27: a Lean-vs-paper fidelity audit found this statement weaker than, or narrower than, the cited result. A corrected statement will be resubmitted.

Preamble
import Mathlib
import Definitions.Def_ConicQuadIPM_NewtonStep_Setting
Formal statement
namespace ConicQuadIPM.NewtonStep

open Matrix

theorem lemma_4_1
    {m k : ℕ} (kind : Fin k → ConicQuadIPM.Complementarity.ConeKind) (n : Fin k → ℕ) (hwf : ConicQuadIPM.Complementarity.WellFormed kind n)
    (A : (i : Fin k) → Matrix (Fin m) (Fin (n i)) ℝ) (b : Fin m → ℝ)
    (c : (i : Fin k) → Fin (n i) → ℝ)
    (x0 : (i : Fin k) → Fin (n i) → ℝ) (τ0 : ℝ) (y0 : Fin m → ℝ)
    (s0 : (i : Fin k) → Fin (n i) → ℝ) (κ0 : ℝ) (γ : ℝ) (hγ : γ ∈ Set.Icc (0 : ℝ) 1)
    (dx : (i : Fin k) → Fin (n i) → ℝ) (dτ : ℝ) (dy : Fin m → ℝ)
    (ds : (i : Fin k) → Fin (n i) → ℝ) (dκ : ℝ)
    (h22 : NewtonSystem kind n A b c x0 τ0 y0 s0 κ0 γ dx dτ dy ds dκ)
    (α : ℝ) (hα : α ∈ Set.Icc (0 : ℝ) 1)
    (x1 : (i : Fin k) → Fin (n i) → ℝ) (τ1 : ℝ) (y1 : Fin m → ℝ)
    (s1 : (i : Fin k) → Fin (n i) → ℝ) (κ1 : ℝ)
    (hx1 : ∀ i, x1 i = x0 i + α • dx i) (hτ1 : τ1 = τ0 + α * dτ)
    (hy1 : y1 = y0 + α • dy) (hs1 : ∀ i, s1 i = s0 i + α • ds i)
    (hκ1 : κ1 = κ0 + α * dκ) :
    (∑ i, A i *ᵥ x1 i) - τ1 • b = (1 - α * (1 - γ)) • ((∑ i, A i *ᵥ x0 i) - τ0 • b) ∧
    (∀ i, (A i)ᵀ *ᵥ y1 + s1 i - τ1 • c i
        = (1 - α * (1 - γ)) • ((A i)ᵀ *ᵥ y0 + s0 i - τ0 • c i)) ∧
    -(∑ i, c i ⬝ᵥ x1 i) + b ⬝ᵥ y1 - κ1
        = (1 - α * (1 - γ)) * (-(∑ i, c i ⬝ᵥ x0 i) + b ⬝ᵥ y0 - κ0) ∧
    (∑ i, dx i ⬝ᵥ ds i) + dτ * dκ = 0 ∧
    (∑ i, x1 i ⬝ᵥ s1 i) + τ1 * κ1
        = (1 - α * (1 - γ)) * ((∑ i, x0 i ⬝ᵥ s0 i) + τ0 * κ0) := by sorry

end ConicQuadIPM.NewtonStep
Source
Andersen, Roos & Terlaky, On implementing a primal-dual interior-point method for conic quadratic optimization, Math. Program. (2003); authors' preprint of 18 Dec 2000, p. 13, Lemma 4.1, (24); system (22) p. 12, step (23) p. 13; proof pp. 13–14
Human review
  • Endorsed by Shuze Chen · Oct 7, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 7, 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