Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Appendix D.1 — f(K0)≥f(K0)−f(KT)=∫0T∥∇f(Kt)∥2dt≥Tmin⁡0≤t≤T∥∇f(Kt)∥2f(K_0)\ge f(K_0)-f(K_T)=\int_0^T\|\nabla f(K_t)\|^2dt\ge T\min_{0\le t\le T}\|\nabla f(K_t)\|^2f(K0​)≥f(K0​)−f(KT​)=∫0T​∥∇f(Kt​)∥2dt≥Tmin0≤t≤T​∥∇f(Kt​)∥2

Proved
FatkhullinPolyak.Flow.flow_integral_bound

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

gradient-flowlqrp2o-batch-p100bp2o-gran-per-chapterp2o-plan-paperp2o-v1

Let Q,R,Σ≻0Q, R, \Sigma\succ0Q,R,Σ≻0 and let KtK_tKt​ be a solution on [0,∞)[0,\infty)[0,∞) of the gradient flow (4.1). Then for every T≥0T\ge0T≥0:

  1. f(K0)≥f(K0)−f(KT)f(K_0) \ge f(K_0) - f(K_T)f(K0​)≥f(K0​)−f(KT​);
  2. f(K0)−f(KT)=∫0T∥∇f(Kt)∥F2 dtf(K_0) - f(K_T) = \displaystyle\int_0^T \|\nabla f(K_t)\|_F^2\,dtf(K0​)−f(KT​)=∫0T​∥∇f(Kt​)∥F2​dt;
  3. there is t∈[0,T]t\in[0,T]t∈[0,T] with T ∥∇f(Kt)∥F2≤∫0T∥∇f(Ks)∥F2 dsT\,\|\nabla f(K_t)\|_F^2 \le \displaystyle\int_0^T\|\nabla f(K_s)\|_F^2\,dsT∥∇f(Kt​)∥F2​≤∫0T​∥∇f(Ks​)∥F2​ds.

Together,

f(K0)  ≥  f(K0)−f(KT)  =  ∫0T∥∇f(Kt)∥F2 dt  ≥  Tmin⁡0≤t≤T∥∇f(Kt)∥F2,f(K_0) \;\ge\; f(K_0)-f(K_T) \;=\; \int_0^T\|\nabla f(K_t)\|_F^2\,dt \;\ge\; T\min_{0\le t\le T}\|\nabla f(K_t)\|_F^2 ,f(K0​)≥f(K0​)−f(KT​)=∫0T​∥∇f(Kt​)∥F2​dt≥T0≤t≤Tmin​∥∇f(Kt​)∥F2​,

which is the chain from which the estimate (4.2) follows.

Formalization Note The minimum over [0,T][0,T][0,T] is rendered as the existence of a point t∈[0,T]t\in[0,T]t∈[0,T] at which the bound holds (the minimum of a continuous function on a compact interval is attained). The integral is the interval integral from 000 to TTT.

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

open Matrix

theorem flow_integral_bound {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)
    (K₀ : Matrix (Fin m) (Fin r) ℝ) (K : ℝ → Matrix (Fin m) (Fin r) ℝ)
    (hK : IsGradFlow A B C Q R Sig K₀ K) (T : ℝ) (hT : 0 ≤ T) :
    cost A B C Q R Sig K₀ - cost A B C Q R Sig (K T) ≤ cost A B C Q R Sig K₀ ∧
      cost A B C Q R Sig K₀ - cost A B C Q R Sig (K T) =
        ∫ t in (0 : ℝ)..T, frobNorm (grad A B C Q R Sig (K t)) ^ 2 ∧
      ∃ t ∈ Set.Icc (0 : ℝ) T,
        T * frobNorm (grad A B C Q R Sig (K t)) ^ 2 ≤
          ∫ s in (0 : ℝ)..T, frobNorm (grad A B C Q R Sig (K s)) ^ 2 := by sorry

end FatkhullinPolyak.Flow
Source
Fatkhullin, Polyak, Optimizing Static Linear Feedback: Gradient Method, arXiv:2004.09875v2, p. 18, Appendix D.1, display
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.

Hypotheses.

  • QQQ, RRR, Σ\SigmaΣ are positive definite (symmetric, with x⊤Mx>0x^\top Mx>0x⊤Mx>0 for x≠0x\ne0x=0).
  • K:R→Rm×rK:\mathbb R\to\mathbb R^{m\times r}K:R→Rm×r is a gradient-flow solution from K0K_0K0​. This means:
    • K(0)=K0K(0)=K_0K(0)=K0​;
    • K(t)∈SK(t)\in\mathcal SK(t)∈S for all t≥0t\ge0t≥0;
    • each entry of KKK has derivative −∇f(K(t))ij-\nabla f(K(t))_{ij}−∇f(K(t))ij​ at every t≥0t\ge0t≥0 relative to [0,∞)[0,\infty)[0,∞).
  • T≥0T\ge 0T≥0 is a real number.
  • No rank condition on CCC and no condition on BBB is imposed.

Conclusion. All three of the following hold:

  1. f(K0)−f(K(T))≤f(K0)f(K_0)-f(K(T))\le f(K_0)f(K0​)−f(K(T))≤f(K0​), which is the same as f(K(T))≥0f(K(T))\ge 0f(K(T))≥0.
  2. f(K0)−f(K(T))=∫0T∥∇f(K(t))∥F2 dt\displaystyle f(K_0)-f(K(T))=\int_0^T\|\nabla f(K(t))\|_F^2\,dtf(K0​)−f(K(T))=∫0T​∥∇f(K(t))∥F2​dt.
  3. There is some t∈[0,T]t\in[0,T]t∈[0,T] with
T ∥∇f(K(t))∥F2≤∫0T∥∇f(K(s))∥F2 ds.T\,\|\nabla f(K(t))\|_F^2\le\int_0^T\|\nabla f(K(s))\|_F^2\,ds.T∥∇f(K(t))∥F2​≤∫0T​∥∇f(K(s))∥F2​ds.

Degenerate cases.

  • The integral is the Lebesgue interval integral, which is 000 by convention when the integrand is not integrable on [0,T][0,T][0,T]. If that happened, item 2 would assert f(K(T))=f(K0)f(K(T))=f(K_0)f(K(T))=f(K0​).
  • For T=0T=0T=0: item 2 reads 0=00=00=0 (since K(0)=K0K(0)=K_0K(0)=K0​), item 3 holds with t=0t=0t=0 as 0≤00\le00≤0, and item 1 reads f(K0)≥0f(K_0)\ge0f(K0​)≥0.
  • If no gradient-flow solution exists for the data, the statement is vacuous.
  • If n=0n=0n=0, all costs and gradients are 000.
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