OctoInformBeta

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

Section 5: conditional negativity and reciprocal inertia: Finite quadratic forms, kernels of negative type, and the distance kernel obtained from full boundary evaluation and interaction

Conditional negativity and a boundary distance 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 disk Dirichlet Green kernel and Poisson kernel enter interior evaluation and boundary interaction. Their resolvent corrections define full boundary kernels, from which the library forms a boundary distance kernel. The accompanying algebraic conditions concern real matrices and finitely supported coefficients on arbitrary sets.

Identify the complex plane with the Euclidean plane ℝ2\mathbb R^2 by the coordinate isometry z↦(Re⁡z,Im⁡z)z\mapsto(\operatorname{Re}z,\operatorname{Im}z).✦ The unit disk DD and unit circle S1S^1 are the sets in (1): D={x∈ℝ2:|x|<1},S1={s∈ℝ2:|s|=1}.D=\{x\in\mathbb R^2:|x|<1\},\qquad S^1=\{s\in\mathbb R^2:|s|=1\}.(1) The complex disk Green and Poisson kernels give the planar kernels GG and PP through this identification. Their formulas are (2), with x,y∈Dx,y\in D and s∈S1s\in S^1: G(x,y)=14πlog⁡(1+(1−|x|2)(1−|y|2)|x−y|2),P(s,x)=1−|x|22π|s−x|2.\begin{aligned} G(x,y)&=\frac1{4\pi} \log\!\left(1+\frac{(1-|x|^2)(1-|y|^2)}{|x-y|^2}\right),\\ P(s,x)&=\frac{1-|x|^2}{2\pi|s-x|^2}. \end{aligned}(2) The boundary interaction RR is the real kernel on S1×S1S^1\times S^1 given by (3), also obtained from the complex boundary interaction: R(s,t)=1π|s−t|2.R(s,t)=\frac1{\pi|s-t|^2}.(3)✦

Write dAdA for area measure restricted to DD, and let ν\nu be a measure dominated by CdAC\,dA for a finite bound CC. The Green integral operator 𝒢\mathcal G acts on L2(dA)L^2(dA). The weighted inclusion JJ from L2(dA)L^2(dA) to L2(ν)L^2(\nu) is the continuous change-of-measure map in (4).✦ The weighted Green operator TT acts on L2(ν)L^2(\nu) and is defined using the Hilbert adjoint J*J^*, as in (4): (𝒢f)(x)=∫DG(x,y)f(y)dA(y),Jf=f,T=J𝒢J*.(\mathcal Gf)(x)=\int_D G(x,y)f(y)\,dA(y),\qquad Jf=f,\qquad T=J\mathcal GJ^*.(4) For s∈S1s\in S^1, the smoothed Poisson kernel P1(s,⋅)P_1(s,\cdot) and its L2(ν)L^2(\nu) class rsr_s are given by (5). The Green row gpg_p at p∈Dp\in D is the L2(ν)L^2(\nu) class of the Green column shown there: P1(s,x)=∫DG(x,y)P(s,y)dν(y),rs=[P1(s,⋅)],gp=[G(⋅,p)].P_1(s,x)=\int_D G(x,y)P(s,y)\,d\nu(y),\qquad r_s=[P_1(s,\cdot)],\qquad g_p=[G(\cdot,p)].(5) For a continuous real linear operator AA on a real Banach space, its Neumann series resolvent 𝒮(A)\mathcal S(A) is the operator series in (6): 𝒮(A)=∑n=0∞An.\mathcal S(A)=\sum_{n=0}^{\infty}A^n.(6)✦

The full boundary evaluation kernel KK is the function on D×S1D\times S^1 defined by (7), where the inner product is that of L2(ν)L^2(\nu): K(p,s)=P(s,p)+P1(s,p)+⟨gp,𝒮(T)rs⟩L2(ν).K(p,s)=P(s,p)+P_1(s,p) +\langle g_p,\mathcal S(T)r_s\rangle_{L^2(\nu)}.(7)✦ The full boundary kernel NN is the function on S1×S1S^1\times S^1 in (8): N(s,t)=R(s,t)+∫DP(s,x)P(t,x)dν(x)+∫DP(s,x)P1(t,x)dν(x)+⟨rs,𝒮(T)rt⟩L2(ν).\begin{aligned} N(s,t)={}&R(s,t)+\int_D P(s,x)P(t,x)\,d\nu(x)\\ &+\int_D P(s,x)P_1(t,x)\,d\nu(x) +\langle r_s,\mathcal S(T)r_t\rangle_{L^2(\nu)}. \end{aligned}(8)✦

Quadratic Forms

For a real matrix AA indexed by a finite set II, its associated bilinear form bAb_A is the function of x,y∈ℝIx,y\in\mathbb R^I defined in (9): bA(x,y)=x𝖳Ay=∑i,j∈IxiAijyj.b_A(x,y)=x^{\mathsf T}Ay=\sum_{i,j\in I}x_iA_{ij}y_j.(9)✦

The Definition

Definition 1. A real matrix AA indexed by a finite set II has at most one positive square, or property (P)(P), if it satisfies (10) for all x,y∈ℝIx,y\in\mathbb R^I: A𝖳=A,bA(x,x)>0 and bA(x,y)=0⇒bA(y,y)≤0.A^{\mathsf T}=A,\qquad b_A(x,x)>0\ \text{ and }\ b_A(x,y)=0 \quad\Longrightarrow\quad b_A(y,y)\leq0.(10)✦ A real matrix BB indexed by II is conditionally negative if it satisfies (11) for every c∈ℝIc\in\mathbb R^I: B𝖳=B,∑i∈Ici=0⇒bB(c,c)≤0.B^{\mathsf T}=B,\qquad \sum_{i\in I}c_i=0\quad\Longrightarrow\quad b_B(c,c)\leq0.(11)✦ For an arbitrary set XX, a real kernel d:X×X→ℝd:X\times X\to\mathbb R is conditionally negative, or of negative type, if (12) holds for all x,y∈Xx,y\in X and every finitely supported real function cc on XX: d(x,y)=d(y,x),d(x,x)=0,∑x∈supp⁡cc(x)=0⇒∑x,y∈supp⁡cc(x)d(x,y)c(y)≤0.\begin{gathered} d(x,y)=d(y,x),\qquad d(x,x)=0,\\ \sum_{x\in\operatorname{supp}c}c(x)=0 \quad\Longrightarrow\quad \sum_{x,y\in\operatorname{supp}c}c(x)d(x,y)c(y)\leq0. \end{gathered}(12)✦ For p∈Dp\in D, the boundary distance kernel dpd_p is the real kernel on S1×S1S^1\times S^1 defined by (13): dp(s,t)={0,s=t,K(p,s)K(p,t)N(s,t),s≠t.d_p(s,t)= \begin{cases} 0,&s=t,\\[2pt] \displaystyle\frac{K(p,s)K(p,t)}{N(s,t)},&s\ne t. \end{cases}(13)✦

What Is Checked

The matrix definitions are stated over finite sets and real coefficients. The negative type definition allows an arbitrary underlying set, with finitely supported coefficients. The boundary distance construction uses p∈Dp\in D and a measure ν\nu dominated by a finite multiple of disk area measure.

The Green rows belong to L2(dA)L^2(dA) almost everywhere✦, and their norms have a common nonnegative bound for every point of DD.✦ Each Green column G(⋅,p)G(\cdot,p) belongs to L2(ν)L^2(\nu).✦ For finite ν\nu, each smoothed Poisson kernel P1(s,⋅)P_1(s,\cdot) also belongs to L2(ν)L^2(\nu), giving the classes used in (5).✦

For the underlying L2L^2 kernel construction with kernel HH, source measure β\beta and target measure α\alpha, the kernel integral f↦∫H(⋅,y)f(y)dβ(y)f\mapsto\int H(\cdot,y)f(y)\,d\beta(y) belongs to L2(α)L^2(\alpha) for every f∈L2(β)f\in L^2(\beta).✦ Its associated kernel operator and the Green integral operator in (4) are continuous and linear. The weighted inclusion is linear.