Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 5.1, pp. 18–19 — for Φ ∈ 𝚷_{n,T}, data are informative for quadratic stabilization iff (5.5) is feasible; under Φ₂₂ < 0 and full row rank, iff (5.6) is; K = LP⁻¹ stabilizes all of Σ

Proved
QuadMatIneq.Stabilization.theorem_5_1

by mikedeng1 · Oct 9, 2026 · Mathlib 0df444a (Lean v4.33.1)

data-driven-controlinformativitylinear-matrix-inequalityp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1quadratic-stabilization

Let Φ∈Πn,T\Phi\in\boldsymbol\Pi_{n,T}Φ∈Πn,T​ and consider data (U−,X)(U_-,X)(U−​,X) generated by x(t+1)=Asx(t)+Bsu(t)+w(t)x(t+1)=A_sx(t)+B_su(t)+w(t)x(t+1)=As​x(t)+Bs​u(t)+w(t) with noise satisfying (2.3). Write

Lβ(P,L)=[P−βI0000−P−L⊤00−L0L00L⊤P]−[IX+0−X−0−U−00][Φ11Φ12Φ21Φ22][IX+0−X−0−U−00] ⁣⊤.\mathcal L_\beta(P,L)=\begin{bmatrix}P-\beta I&0&0&0\\0&-P&-L^\top&0\\0&-L&0&L\\0&0&L^\top&P\end{bmatrix}-\begin{bmatrix}I&X_+\\0&-X_-\\0&-U_-\\0&0\end{bmatrix}\begin{bmatrix}\Phi_{11}&\Phi_{12}\\ \Phi_{21}&\Phi_{22}\end{bmatrix}\begin{bmatrix}I&X_+\\0&-X_-\\0&-U_-\\0&0\end{bmatrix}^{\!\top}.Lβ​(P,L)=​P−βI000​0−P−L0​0−L⊤0L⊤​00LP​​−​I000​X+​−X−​−U−​0​​[Φ11​Φ21​​Φ12​Φ22​​]​I000​X+​−X−​−U−​0​​⊤.
  1. (a) The data are informative for quadratic stabilization if and only if there exist P∈Rn×nP\in\mathbb{R}^{n\times n}P∈Rn×n, P>0P>0P>0, L∈Rm×nL\in\mathbb{R}^{m\times n}L∈Rm×n and β>0\beta>0β>0 with
Lβ(P,L)⩾0.(5.5)\mathcal L_\beta(P,L)\geqslant 0. \tag{5.5}Lβ​(P,L)⩾0.(5.5)
  1. (b) If in addition Φ22<0\Phi_{22}<0Φ22​<0 and rank⁡[X−U−]=n+m\operatorname{rank}\begin{bmatrix}X_-\\U_-\end{bmatrix}=n+mrank[X−​U−​​]=n+m, the data are informative for quadratic stabilization if and only if there exist P>0P>0P>0 and L∈Rm×nL\in\mathbb{R}^{m\times n}L∈Rm×n with
L0(P,L)>0.(5.6)\mathcal L_0(P,L)>0. \tag{5.6}L0​(P,L)>0.(5.6)
  1. Moreover, whenever (P,L,β)(P,L,\beta)(P,L,β) satisfies (5.5), or (P,L)(P,L)(P,L) satisfies (5.6), the gain K:=LP−1K:=LP^{-1}K:=LP−1 stabilizes every (A,B)∈Σ(A,B)\in\Sigma(A,B)∈Σ: P−(A+BK)P(A+BK)⊤>0P-(A+BK)P(A+BK)^\top>0P−(A+BK)P(A+BK)⊤>0 and A+BKA+BKA+BK is Schur stable.

This is the main data-driven control result of the paper: a necessary and sufficient LMI test for the existence of a single state-feedback controller, with a common quadratic Lyapunov function, that stabilizes every system consistent with noisy data, under any noise model in the class Πn,T\boldsymbol\Pi_{n,T}Πn,T​ and with no rank or Slater condition in part (a).

Formalization Note "The data are generated by (2.1) with noise model (2.3)" is the hypothesis that some (As,Bs)(A_s,B_s)(As​,Bs​) lies in Σ\SigmaΣ; it is essential (if Σ=∅\Sigma=\emptysetΣ=∅, informativity holds vacuously). Lβ\mathcal L_\betaLβ​ and L0\mathcal L_0L0​ are lmi55 and lmi56; informativity is InformativeQS (Definition 2.1); "stabilizing feedback gain for all (A,B)∈Σ(A,B)\in\Sigma(A,B)∈Σ" is StabilizesAll, stated with the PPP of the feasible LMI: the Lyapunov inequality (2.4) together with (A+BK)k→0(A+BK)^k\to 0(A+BK)k→0. P−1P^{-1}P−1 is Mathlib's matrix inverse, the true inverse since P>0P>0P>0. The Moreover part is stated under the theorem's standing hypotheses only, without those of (b), as in the paper. No hypothesis n⩾1n\geqslant 1n⩾1 is added: at n=0n=0n=0 both sides of (a) and (b) hold.

Preamble
import Mathlib
import Definitions.Def_QuadMatIneq_Stabilization_QMI
import Definitions.Def_QuadMatIneq_Stabilization_Data

open Matrix Filter Topology
Formal statement
namespace QuadMatIneq.Stabilization

/-- Theorem 5.1, pp. 18–19. `hgen` is "the data `(U₋, X)` are generated by (2.1) with noise
model (2.3)": the true system `(A_s, B_s)` lies in `Σ`. -/
theorem theorem_5_1 {n m T : ℕ}
    (Φ : Matrix (Fin n ⊕ Fin T) (Fin n ⊕ Fin T) ℝ) (X : Matrix (Fin n) (Fin (T + 1)) ℝ)
    (Um : Matrix (Fin m) (Fin T) ℝ) (hΦ : InPi Φ)
    (hgen : ∃ As : Matrix (Fin n) (Fin n) ℝ, ∃ Bs : Matrix (Fin n) (Fin m) ℝ,
      (As, Bs) ∈ Sigma Φ X Um) :
    -- (a)
    (InformativeQS Φ X Um ↔
      ∃ P : Matrix (Fin n) (Fin n) ℝ, ∃ L : Matrix (Fin m) (Fin n) ℝ, ∃ β : ℝ,
        P.PosDef ∧ 0 < β ∧ (lmi55 Φ X Um P L β).PosSemidef) ∧
    -- (b)
    ((-Φ.toBlocks₂₂).PosDef → (Matrix.fromRows (Xminus X) Um).rank = n + m →
      (InformativeQS Φ X Um ↔
        ∃ P : Matrix (Fin n) (Fin n) ℝ, ∃ L : Matrix (Fin m) (Fin n) ℝ,
          P.PosDef ∧ (lmi56 Φ X Um P L).PosDef)) ∧
    -- Moreover, for (5.5)
    (∀ (P : Matrix (Fin n) (Fin n) ℝ) (L : Matrix (Fin m) (Fin n) ℝ) (β : ℝ),
      P.PosDef → 0 < β → (lmi55 Φ X Um P L β).PosSemidef → StabilizesAll Φ X Um P (L * P⁻¹)) ∧
    -- Moreover, for (5.6)
    (∀ (P : Matrix (Fin n) (Fin n) ℝ) (L : Matrix (Fin m) (Fin n) ℝ),
      P.PosDef → (lmi56 Φ X Um P L).PosDef → StabilizesAll Φ X Um P (L * P⁻¹)) := by sorry

end QuadMatIneq.Stabilization
Source
van Waarde–Camlibel–Eising–Trentelman, arXiv:2203.12959v3, Theorem 5.1, pp. 18–19, (5.5), (5.6)
Human review
  • Endorsed by Shuze Chen · Oct 10, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 10, 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