OctoInformBeta

OpenAI preprint, September 2026 · Reciprocal Green functions and a squared-distance kernel

Lemma 5.3: Entrywise reciprocal matrices have at most one positive square throughout a monotone approximation to the singular kernel

Bounded regularizations of the disk Green kernel.

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 item concerns bounded continuous approximations to the singular Dirichlet Green kernel on the disk. The unit disk D⊂ℂD\subset\mathbb C is the open planar domain D={z∈ℂ:|z|<1}.D=\{z\in\mathbb C:|z|<1\}.(1)✦

The normalized squared distance h:D×D→ℝh:D\times D\to\mathbb R is the function h(x,y)=|x−y|2(1−|x|2)(1−|y|2).h(x,y)=\frac{|x-y|^2}{(1-|x|^2)(1-|y|^2)}.(2)✦ The real disk Green kernel G:D×D→ℝG:D\times D\to\mathbb R and its extended version G∞:D×D→[0,∞]G_\infty:D\times D\to[0,\infty] are defined using (2) by G(x,y)={14πlog⁡(1+1h(x,y)),x≠y,0,x=y,G∞(x,y)={max⁡{G(x,y),0},x≠y,+∞,x=y.\begin{aligned} G(x,y)&= \begin{cases} \dfrac{1}{4\pi}\log\!\left(1+\dfrac{1}{h(x,y)}\right),&x\ne y,\\ 0,&x=y, \end{cases}\\ G_\infty(x,y)&= \begin{cases} \max\{G(x,y),0\},&x\ne y,\\ +\infty,&x=y. \end{cases} \end{aligned}(3)

The Reciprocal Quadratic Form

For a finite index set II, a real matrix A∈ℝI×IA\in\mathbb R^{I\times I} determines the bilinear form BA(u,v)=u𝖳Av=∑i,j∈IuiAijvj,u,v∈ℝI.B_A(u,v)=u^{\mathsf T}Av =\sum_{i,j\in I}u_iA_{ij}v_j, \qquad u,v\in\mathbb R^I.(4)✦ A finite real matrix AA has at most one positive square, called property (P)(P), if it is symmetric and the form (4) satisfies BA(u,u)>0,BA(u,v)=0⇒BA(v,v)≤0(u,v∈ℝI).B_A(u,u)>0,\quad B_A(u,v)=0 \quad\Longrightarrow\quad B_A(v,v)\le0 \qquad(u,v\in\mathbb R^I).(5)✦

The Claim

Theorem 1. On the unit disk DD, there is a family of real kernels Cε:D×D→ℝC_\varepsilon:D\times D\to\mathbb R, indexed by ε∈ℝ\varepsilon\in\mathbb R, satisfying (6), where convergence is in [0,∞][0,\infty]: Cε∈C(D×D),0<Cε(x,y)≤12πε(0<ε≤1,x,y∈D),(Cε(xi,xj)−1)i,j∈I has property (P)(0<ε≤1,I finite,x∈DI),Cδ(x,y)≤Cε(x,y)(0<ε≤δ≤1,x,y∈D),limε↓0Cε(x,y)=G∞(x,y)(x,y∈D).\begin{aligned} &C_\varepsilon\in C(D\times D),\qquad 0<C_\varepsilon(x,y)\le\frac{1}{2\pi\varepsilon} &&(0<\varepsilon\le1,\ x,y\in D),\\ &\left(C_\varepsilon(x_i,x_j)^{-1}\right)_{i,j\in I} \text{ has property }(P) &&(0<\varepsilon\le1,\ I\text{ finite},\ x\in D^I),\\ &C_\delta(x,y)\le C_\varepsilon(x,y) &&(0<\varepsilon\le\delta\le1,\ x,y\in D),\\ &\lim_{\varepsilon\downarrow0}C_\varepsilon(x,y)=G_\infty(x,y) &&(x,y\in D). \end{aligned}(6)✦

What Is Checked

The claim (6) is stated over the explicit Green kernels (3) on the disk (1) and property (P)(P) in (5) for finite real matrices.