OctoInformBeta

OpenAI preprint, September 2026 · §6. Exact transmission on the regular harmonic subspaces

Lemma 6.1: Integrated and pointwise Frobenius bounds under fixed quadratic-form lower bounds

Sylvester damping controls matrix error growth.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

The estimates concern matrix errors in harmonic transmission, with damping acting from the left and the right. Fix a finite matrix size NN. The Frobenius norm ∥C∥F\|C\|_F of a real N×NN\times N matrix CC is the quantity in (1): ∥C∥F=(∑i,j=1NCij2)1/2.\|C\|_F=\left(\sum_{i,j=1}^N C_{ij}^2\right)^{1/2}.(1)

Let u≤vu\le v, and let X,F,L,RX,F,L,R be functions from ℝ\mathbb R to the real N×NN\times N matrices. The forced Sylvester equation for XX on [u,v][u,v] is (2), with the derivative required at every point of that interval: X′(t)=F(t)−L(t)X(t)−X(t)R(t),t∈[u,v].X'(t)=F(t)-L(t)X(t)-X(t)R(t),\qquad t\in[u,v].(2) The uniform quadratic-form lower bounds (3) have fixed real constants c,dc,d and hold for every vector z∈ℝNz\in\mathbb R^N: z𝖳L(t)z≥c|z|2,z𝖳R(t)z≥d|z|2,t∈[u,v].z^{\mathsf T}L(t)z\ge c\,|z|^2, \qquad z^{\mathsf T}R(t)z\ge d\,|z|^2, \qquad t\in[u,v].(3)

The Claim

Theorem 1 (Sylvester estimates). Suppose XX satisfies (2) and L,RL,R satisfy (3). Put γ=c+d\gamma=c+d. If FF is continuous on [u,v][u,v], the integrated Sylvester estimate (4) holds: γ∫uv∥X(t)∥Fdt≤∥X(u)∥F−∥X(v)∥F+∫uv∥F(t)∥Fdt.\gamma\int_u^v\|X(t)\|_F\,dt \le \|X(u)\|_F-\|X(v)\|_F+\int_u^v\|F(t)\|_F\,dt.(4)✦ Let MM be a real constant such that ∥F(t)∥F≤M\|F(t)\|_F\le M for every t∈[u,v]t\in[u,v]; then the pointwise Sylvester estimate (5) holds for every t∈[u,v]t\in[u,v]: ∥X(t)∥F≤{e−γ(t−u)∥X(u)∥F+Mγ(1−e−γ(t−u)),γ≠0,∥X(u)∥F+M(t−u),γ=0.\|X(t)\|_F\le \begin{cases} e^{-\gamma(t-u)}\|X(u)\|_F +\dfrac{M}{\gamma}\bigl(1-e^{-\gamma(t-u)}\bigr), &\gamma\ne0,\\[6pt] \|X(u)\|_F+M(t-u),&\gamma=0. \end{cases}(5)✦

In the paper, the result reads: if X:[u,v]→ℝN×NX:[u,v]\to\mathbb R^{N\times N} is C1C^1, the functions F,L,RF,L,R are continuous on [u,v][u,v], XX satisfies (2), and L(t),R(t)L(t),R(t) are symmetric with λmin(L(t))+λmin(R(t))≥γ>0\lambda_{\min}(L(t))+\lambda_{\min}(R(t))\ge\gamma>0 throughout the interval, then the estimates (6) and (7) hold: ∥X(t)∥F≤e−γ(t−u)∥X(u)∥F+∫ute−γ(t−w)∥F(w)∥Fdw,u≤t≤v,\|X(t)\|_F\le e^{-\gamma(t-u)}\|X(u)\|_F +\int_u^t e^{-\gamma(t-w)}\|F(w)\|_F\,dw, \qquad u\le t\le v,(6) ∫uv∥X(t)∥Fdt≤γ−1(∥X(u)∥F+∫uv∥F(t)∥Fdt).\int_u^v\|X(t)\|_F\,dt \le \gamma^{-1}\left(\|X(u)\|_F+\int_u^v\|F(t)\|_F\,dt\right).(7)

What Is Checked

The statements concern differentiable curves of real square matrices satisfying (2) and (3) on a closed interval, with continuous forcing in (4) and uniformly bounded forcing in (5).