Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 1, p. 11 — for a pure SOCP, the HKM and NT directions are invariant under automorphic scaling of the cones

Proved
SDPT3.ScaleInv.proposition_1

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

interior-pointp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1scale-invariancesecond-order-cone

Consider a pure second-order cone program (no semidefinite or linear blocks) with blocks i=1,…,nqi = 1, \dots, n_qi=1,…,nq​ of dimensions qiq_iqi​, data Ai∈Rqi×mA_i \in \mathbb R^{q_i\times m}Ai​∈Rqi​×m, b∈Rmb \in \mathbb R^mb∈Rm, ci∈Rqic_i \in \mathbb R^{q_i}ci​∈Rqi​, an iterate (x,y,z)(x, y, z)(x,y,z) with every xix_ixi​ and ziz_izi​ in the interior of the second-order cone, and a centering parameter σ\sigmaσ. Fix the search direction to be either HKM or NT. For each block let FiF_iFi​ be an automorphism of the cone KqqiK_q^{q_i}Kqqi​​, and form the scaled problem and iterate

A^i=Fi−1Ai,b^=b,c^i=Fi−1ci,x^i=Fi⊤xi,y^=y,z^i=Fi−1zi.\hat A_i = F_i^{-1}A_i,\quad \hat b = b,\quad \hat c_i = F_i^{-1}c_i,\quad \hat x_i = F_i^\top x_i,\quad \hat y = y,\quad \hat z_i = F_i^{-1}z_i.A^i​=Fi−1​Ai​,b^=b,c^i​=Fi−1​ci​,x^i​=Fi⊤​xi​,y^​=y,z^i​=Fi−1​zi​.

Then, for every (Δx,Δy,Δz)(\Delta x, \Delta y, \Delta z)(Δx,Δy,Δz),

(Δx,Δy,Δz) solves (4) for the original data  ⟺  (F⊤Δx, Δy, F−1Δz) solves (4) for the scaled data,(\Delta x, \Delta y, \Delta z) \text{ solves (4) for the original data} \iff (F^\top\Delta x,\ \Delta y,\ F^{-1}\Delta z) \text{ solves (4) for the scaled data},(Δx,Δy,Δz) solves (4) for the original data⟺(F⊤Δx, Δy, F−1Δz) solves (4) for the scaled data,

where F⊤ΔxF^\top \Delta xF⊤Δx and F−1ΔzF^{-1}\Delta zF−1Δz act block by block.

In words: the HKM and NT search directions are scale-invariant — the direction computed for the scaled problem is exactly the image of the original direction under the scaling. This is Proposition 1 of the paper.

Formalization Note The automorphism group Gi\mathcal G_iGi​ is encoded algebraically: Fi⊤JˉFi=λ2JˉF_i^\top \bar J F_i = \lambda^2 \bar JFi⊤​JˉFi​=λ2Jˉ for some λ>0\lambda > 0λ>0 and (Fi)00>0(F_i)_{00} > 0(Fi​)00​>0. The theorem covers all of Gi\mathcal G_iGi​, including λ≠1\lambda \ne 1λ=1; the paper's (31) and its remark that ω\omegaω is unchanged hold only for λ=1\lambda = 1λ=1 (printed slip), but the proposition itself holds for every λ>0\lambda > 0λ>0. The direction is characterized by the raw Newton system (4) with E,F\mathcal E, \mathcal FE,F from (8) and GGG from (19)/(21), and the conclusion is stated for the solution sets, so no uniqueness is needed and the paper's assumption that AAA has full row rank (p. 5) is not used. Interiority of the scaled iterate is not assumed; it follows from Fi∈GiF_i \in \mathcal G_iFi​∈Gi​. Block vectors are Fin (d i + 1) → ℝ, with xi0x_i^0xi0​ at index 0 and xi1x_i^1xi1​ at the indices Fin.succ j; the interior is encoded by xi0>0x_i^0 > 0xi0​>0 and (xi0)2−⟨xi1,xi1⟩>0(x_i^0)^2 - \langle x_i^1, x_i^1\rangle > 0(xi0​)2−⟨xi1​,xi1​⟩>0.

Preamble
import Mathlib
import Definitions.Def_SDPT3_ScaleInv_Setting
Formal statement
namespace SDPT3.ScaleInv

open Matrix

theorem proposition_1 {nq m : ℕ} {d : Fin nq → ℕ} (dir : Dir)
    (A : (i : Fin nq) → Matrix (Fin (d i + 1)) (Fin m) ℝ) (b : Fin m → ℝ)
    (c x : (i : Fin nq) → Fin (d i + 1) → ℝ) (y : Fin m → ℝ)
    (z : (i : Fin nq) → Fin (d i + 1) → ℝ) (σ : ℝ)
    (hx : ∀ i, socInt (x i)) (hz : ∀ i, socInt (z i))
    (Fsc : (i : Fin nq) → Matrix (Fin (d i + 1)) (Fin (d i + 1)) ℝ) (hF : ∀ i, IsSOCAut (Fsc i))
    (Δx : (i : Fin nq) → Fin (d i + 1) → ℝ) (Δy : Fin m → ℝ)
    (Δz : (i : Fin nq) → Fin (d i + 1) → ℝ) :
    IsNewtonDir dir A b c x y z σ Δx Δy Δz ↔
      IsNewtonDir dir (fun i => (Fsc i)⁻¹ * A i) b (fun i => (Fsc i)⁻¹ *ᵥ c i)
        (fun i => (Fsc i)ᵀ *ᵥ x i) y (fun i => (Fsc i)⁻¹ *ᵥ z i) σ
        (fun i => (Fsc i)ᵀ *ᵥ Δx i) Δy (fun i => (Fsc i)⁻¹ *ᵥ Δz i) := by sorry

end SDPT3.ScaleInv
Source
Tütüncü, Toh & Todd, Solving semidefinite-quadratic-linear programs using SDPT3, Math. Program. 95 (2003) 189–217; authors' copy, p. 11, Proposition 1; proof p. 12
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