Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 3.15 (qualitative form) — ∇f\nabla f∇f is Lipschitz on the sublevel set S0\mathcal S_0S0​

Proved
FatkhullinPolyak.Flow.grad_lipschitz_on_sublevel

by mikedeng1 · Sep 27, 2026 · Mathlib 0df444a (Lean v4.33.1)

lqrp2o-batch-p100bp2o-gran-per-chapterp2o-plan-paperp2o-v1smoothness

Assume the standing assumptions: B≠0B\ne0B=0, rank⁡C=r\operatorname{rank}C=rrankC=r, Q,R,Σ≻0Q, R, \Sigma\succ0Q,R,Σ≻0, and K0∈SK_0\in\mathcal SK0​∈S. Then fff is LLL-smooth on the sublevel set S0={K∈S:f(K)≤f(K0)}\mathcal S_0=\{K\in\mathcal S: f(K)\le f(K_0)\}S0​={K∈S:f(K)≤f(K0​)} for some constant LLL: there is L>0L>0L>0 with

∥∇f(K)−∇f(K′)∥F≤L ∥K−K′∥Ffor all K,K′∈S0.\|\nabla f(K) - \nabla f(K')\|_F \le L\,\|K-K'\|_F\qquad\text{for all } K, K'\in\mathcal S_0 .∥∇f(K)−∇f(K′)∥F​≤L∥K−K′∥F​for all K,K′∈S0​.

This is the property the paper calls LLL-smoothness (§3.6, p. 9: "a function is called LLL-smooth, if its gradient satisfies Lipschitz condition with constant LLL"). It shows that the Lipschitz hypothesis of the main theorem can always be met.

Formalization Note The paper's Theorem 3.15 also gives an explicit value (3.8) for LLL. That value is not claimed here, because it is false as printed: for n=m=r=1n=m=r=1n=m=r=1, A=0A=0A=0, B=12B=\tfrac12B=21​, C=10C=10C=10, Q=R=Σ=1Q=R=\Sigma=1Q=R=Σ=1, K0=0.2K_0=0.2K0​=0.2 one has f′′=1600f''=1600f′′=1600 on S0\mathcal S_0S0​ while (3.8) gives L≈678.2L\approx678.2L≈678.2; for state feedback with A=0A=0A=0, B=100B=100B=100, Q=100Q=100Q=100, R=10−3R=10^{-3}R=10−3, Σ=0.1\Sigma=0.1Σ=0.1, K0=10−6K_0=10^{-6}K0​=10−6, f′′(K0)f''(K_0)f′′(K0​) is twice the value of (3.8). Only the existence of a Lipschitz constant is stated.

Preamble
import Mathlib
import Definitions.Def_FatkhullinPolyak_Flow_LQR
Formal statement
namespace FatkhullinPolyak.Flow

open Matrix

theorem grad_lipschitz_on_sublevel {n m r : ℕ}
    (A : Matrix (Fin n) (Fin n) ℝ) (B : Matrix (Fin n) (Fin m) ℝ) (C : Matrix (Fin r) (Fin n) ℝ)
    (Q : Matrix (Fin n) (Fin n) ℝ) (R : Matrix (Fin m) (Fin m) ℝ) (Sig : Matrix (Fin n) (Fin n) ℝ)
    (hQ : Q.PosDef) (hR : R.PosDef) (hSig : Sig.PosDef) (hC : C.rank = r) (hB : B ≠ 0)
    (K₀ : Matrix (Fin m) (Fin r) ℝ) (hK₀ : K₀ ∈ stabSet A B C) :
    ∃ L : ℝ, 0 < L ∧
      ∀ K ∈ sublevel A B C Q R Sig K₀, ∀ K' ∈ sublevel A B C Q R Sig K₀,
        frobNorm (grad A B C Q R Sig K - grad A B C Q R Sig K') ≤ L * frobNorm (K - K') := by sorry

end FatkhullinPolyak.Flow
Source
Fatkhullin, Polyak, Optimizing Static Linear Feedback: Gradient Method, arXiv:2004.09875v2, p. 9, §3.6 and Theorem 3.15 (qualitative form; the explicit constant (3.8) is not stated)
Read-back

What the Lean code literally says, in plain math · claude-opus-5-5

Setting. Let

A∈Rn×n,B∈Rn×m,C∈Rr×n,Q∈Rn×n,R∈Rm×m,Σ∈Rn×n.A\in\mathbb R^{n\times n},\quad B\in\mathbb R^{n\times m},\quad C\in\mathbb R^{r\times n},\quad Q\in\mathbb R^{n\times n},\quad R\in\mathbb R^{m\times m},\quad \Sigma\in\mathbb R^{n\times n}.A∈Rn×n,B∈Rn×m,C∈Rr×n,Q∈Rn×n,R∈Rm×m,Σ∈Rn×n.

Gains are m×rm\times rm×r real matrices. The notions used are:

  • S\mathcal SS is the set of gains with every complex eigenvalue of A−BKCA-BKCA−BKC having negative real part.
  • X(K)X(K)X(K) and Y(K)Y(K)Y(K) are the unique solutions (or 000 if the solution is not unique) of
(A−BKC)⊤X+X(A−BKC)+C⊤K⊤RKC+Q=0,(A-BKC)^\top X+X(A-BKC)+C^\top K^\top RKC+Q=0,(A−BKC)⊤X+X(A−BKC)+C⊤K⊤RKC+Q=0, (A−BKC)Y+Y(A−BKC)⊤+Σ=0.(A-BKC)Y+Y(A-BKC)^\top+\Sigma=0.(A−BKC)Y+Y(A−BKC)⊤+Σ=0.
  • The cost is f(K)=tr⁡(X(K)Σ)f(K)=\operatorname{tr}(X(K)\Sigma)f(K)=tr(X(K)Σ).
  • ∇f(K):=2(RKC−B⊤X(K))Y(K)C⊤\nabla f(K):=2\big(RKC-B^\top X(K)\big)Y(K)C^\top∇f(K):=2(RKC−B⊤X(K))Y(K)C⊤ is the explicit formula.
  • ∥⋅∥F\|\cdot\|_F∥⋅∥F​ is the Frobenius norm.
  • S0={K∈S:f(K)≤f(K0)}\mathcal S_0=\{K\in\mathcal S: f(K)\le f(K_0)\}S0​={K∈S:f(K)≤f(K0​)}.

Hypotheses.

  • QQQ, RRR, Σ\SigmaΣ are positive definite (symmetric, with x⊤Mx>0x^\top Mx>0x⊤Mx>0 for x≠0x\ne0x=0).
  • rank⁡C=r\operatorname{rank} C=rrankC=r.
  • B≠0B\neq 0B=0.
  • K0∈SK_0\in\mathcal SK0​∈S.

Conclusion. There exists a real L>0L>0L>0 such that for all K,K′∈S0K,K'\in\mathcal S_0K,K′∈S0​,

∥∇f(K)−∇f(K′)∥F≤L ∥K−K′∥F.\|\nabla f(K)-\nabla f(K')\|_F\le L\,\|K-K'\|_F.∥∇f(K)−∇f(K′)∥F​≤L∥K−K′∥F​.

LLL may depend on all the data, including K0K_0K0​. The inequality is asserted for every pair of points of S0\mathcal S_0S0​, whether or not the segment between them stays in S0\mathcal S_0S0​. Nothing is claimed outside S0\mathcal S_0S0​.

Degenerate cases.

  • B≠0B\ne 0B=0 forces n,m≥1n,m\ge 1n,m≥1.
  • If r=0r=0r=0, S0\mathcal S_0S0​ has at most one point and both sides are 000.
  • S0\mathcal S_0S0​ always contains K0K_0K0​, so it is nonempty. If it is a single point, the inequality reads 0≤00\le 00≤0.
Human review
  • Endorsed by Shuze Chen · Sep 27, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Sep 27, 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