OctoInformBeta

OpenAI preprint, September 2026 · Hilbert multipliers and the absence of critical points

Lemma 6.1: An injective Lipschitz map for weighted disk kernels, together with a general criterion for kernels of negative type

A Hilbert realization of the boundary 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

A weighted Dirichlet problem on the unit disk gives a boundary kernel formed from Poisson evaluations and boundary interactions. The geometric question is whether this kernel is the squared distance of an injective Lipschitz map into a real Hilbert space.

Identify the Euclidean plane ℝ2\mathbb R^2 with ℂ\mathbb C by the real-linear isometry z↦(Re⁡z,Im⁡z)z\mapsto(\operatorname{Re}z,\operatorname{Im}z).✦ Let D={x∈ℝ2:‖x‖<1}D=\{x\in\mathbb R^2:\lVert x\rVert<1\} be the open unit disk and S1={s∈ℝ2:‖s‖=1}S^1=\{s\in\mathbb R^2:\lVert s\rVert=1\} its boundary. Fix a measure ν\nu on ℝ2\mathbb R^2 with a finite density bound M≥0M\geq0 as in (1), where dx|Ddx|_D denotes area measure restricted to DD: ν≤Mdx|D.\nu\leq M\,dx|_D.(1)

For an open planar set UU, a test function φ∈Cc∞(U)\varphi\in C_c^\infty(U) is a smooth real function on ℝ2\mathbb R^2 whose closed support is compact and lies in UU.✦ The test derivative ∂eφ\partial_e\varphi of a test function φ\varphi in a direction e∈ℝ2e\in\mathbb R^2 is its directional derivative, and the test vector field φe\varphi e is the vector field obtained by multiplying ee by φ\varphi. A vector field g∈L2(U;ℝ2)g\in L^2(U;\mathbb R^2) is the weak gradient of u∈L2(U)u\in L^2(U) when (2) holds for every such φ\varphi and every ee: ∫Uu∂eφdx=−∫U(g⋅e)φdx.\int_U u\,\partial_e\varphi\,dx =-\int_U (g\cdot e)\varphi\,dx.(2)✦ The weak gradient graph H1(U)H^1(U) consists of scalar-vector pairs (u,g)(u,g) in which gg is the weak gradient of uu, with squared norm ‖u‖L2(U)2+‖g‖L2(U)2\lVert u\rVert_{L^2(U)}^2+\lVert g\rVert_{L^2(U)}^2. The Sobolev class of a function uu with weak gradient gg is the pair of the L2L^2 classes of uu and gg.✦ The Sobolev value and gradient maps on H1(U)H^1(U) are the continuous linear projections (u,g)↦u(u,g)\mapsto u and (u,g)↦g(u,g)\mapsto g, with gg denoted by ∇u\nabla u.

The test-function embedding from Cc∞(U)C_c^\infty(U) into H1(U)H^1(U) sends φ\varphi to (φ,∇φ)(\varphi,\nabla\varphi).✦ The Dirichlet Sobolev space H01(U)H_0^1(U) is the closure in H1(U)H^1(U) of the image of the test-function embedding. The Dirichlet value and gradient maps on H01(U)H_0^1(U) are the restrictions of the Sobolev value and gradient projections on H1(U)H^1(U). On DD, the weighted Sobolev value map sends u∈H01(D)u\in H_0^1(D) to its value class in L2(ν)L^2(\nu).✦

The disk Dirichlet Green kernel GG, the Poisson kernel PP, and the boundary interaction RR are the real kernels in (3), with x,y∈Dx,y\in D and s,t∈S1s,t\in S^1. These are the disk formulas in complex coordinates under the preceding isometry. G(x,y)=14πlog⁡(1+(1−‖x‖2)(1−‖y‖2)‖x−y‖2)(x≠y),G(x,x)=0,P(s,x)=1−‖x‖22π‖s−x‖2,R(s,t)=1π‖s−t‖2(s≠t),R(s,s)=0.\begin{aligned} G(x,y)&=\frac{1}{4\pi}\log\!\left( 1+\frac{(1-\lVert x\rVert^2)(1-\lVert y\rVert^2)} {\lVert x-y\rVert^2}\right) &&(x\ne y), &G(x,x)&=0,\\ P(s,x)&=\frac{1-\lVert x\rVert^2}{2\pi\lVert s-x\rVert^2},\\ R(s,t)&=\frac{1}{\pi\lVert s-t\rVert^2} &&(s\ne t), &R(s,s)&=0. \end{aligned}(3)

The weighted L2L^2 inclusion J:L2(D,dx)→L2(ν)J:L^2(D,dx)\to L^2(\nu) sends a value class to the same function for the measure ν\nu.✦ The Green integral operator AA on L2(D,dx)L^2(D,dx) is the continuous kernel operator given in (4).✦ With J*J^* denoting the Hilbert adjoint of JJ, the weighted Green operator TT acts on L2(ν)L^2(\nu), and 𝒮\mathcal S denotes its Neumann series resolvent: (Af)(x)=∫DG(x,y)f(y)dy,T=JAJ*,𝒮=∑n=0∞Tn.(Af)(x)=\int_D G(x,y)f(y)\,dy, \qquad T=JAJ^*, \qquad \mathcal S=\sum_{n=0}^{\infty}T^n.(4)

For s∈S1s\in S^1 and x∈Dx\in D, the smoothed Poisson kernel P1(s,x)P_1(s,x) is the integral in (5). Write gp∈L2(ν)g_p\in L^2(\nu) for the Green row at p∈Dp\in D, and qs∈L2(ν)q_s\in L^2(\nu) for the smoothed Poisson class at s∈S1s\in S^1: P1(s,x)=∫G(x,y)P(s,y)dν(y),gp=G(⋅,p),qs=P1(s,⋅).P_1(s,x)=\int G(x,y)P(s,y)\,d\nu(y), \qquad g_p=G(\cdot,p), \qquad q_s=P_1(s,\cdot).(5)

For p∈Dp\in D and s,t∈S1s,t\in S^1, the full boundary evaluation kernel K(p,s)K(p,s) and full boundary kernel N(s,t)N(s,t) are the functions in (6), with inner products in L2(ν)L^2(\nu). K(p,s)=P(s,p)+P1(s,p)+⟨gp,𝒮qs⟩,N(s,t)=R(s,t)+∫P(s,x)P(t,x)dν(x)+∫P(s,x)P1(t,x)dν(x)+⟨qs,𝒮qt⟩.\begin{aligned} K(p,s)&=P(s,p)+P_1(s,p)+\langle g_p,\mathcal S q_s\rangle,\\ N(s,t)&=R(s,t)+\int P(s,x)P(t,x)\,d\nu(x)\\ &\quad+\int P(s,x)P_1(t,x)\,d\nu(x) +\langle q_s,\mathcal S q_t\rangle. \end{aligned}(6) For p∈Dp\in D, the boundary distance kernel Dp:S1×S1→ℝD_p:S^1\times S^1\to\mathbb R is defined by (7). 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}(7)✦

Kernels of Negative Type

For a set XX, a real kernel δ:X×X→ℝ\delta:X\times X\to\mathbb R is of negative type, or conditionally negative, when it is symmetric, vanishes on the diagonal, and satisfies (8) for every finitely supported real family (cx)x∈X(c_x)_{x\in X} with ∑xcx=0\sum_x c_x=0: ∑x,y∈Xcxδ(x,y)cy≤0.\sum_{x,y\in X}c_x\delta(x,y)c_y\leq0.(8)✦

The Claim

Theorem 1. Assume the measure bound (1). Let dd be a real number with 0≤d<10\leq d<1, and suppose that ∫|u|2dν≤d∫D|∇u|2dxfor every u∈H01(D).\int |u|^2\,d\nu \leq d\int_D |\nabla u|^2\,dx \qquad\text{for every }u\in H_0^1(D).(9) For every p∈Dp\in D, there exist a complete separable real Hilbert space HH, an injective map b:S1→Hb:S^1\to H, and a constant L≥0L\geq0 satisfying (10): ‖b(s)−b(t)‖2=Dp(s,t),‖b(s)−b(t)‖≤L‖s−t‖(s,t∈S1).\lVert b(s)-b(t)\rVert^2=D_p(s,t), \qquad \lVert b(s)-b(t)\rVert\leq L\lVert s-t\rVert \qquad(s,t\in S^1).(10)✦

Let XX be a nonempty separable pseudometric space, with distance dXd_X, and let δ\delta be a kernel of negative type on XX. Suppose that δ(x,y)>0\delta(x,y)>0 for distinct x,yx,y and that a constant C≥0C\geq0 satisfies δ(x,y)≤C2dX(x,y)2\delta(x,y)\leq C^2d_X(x,y)^2 for all x,y∈Xx,y\in X. There exist a complete separable real Hilbert space HH and an injective map b:X→Hb:X\to H satisfying (11): ‖b(x)−b(y)‖2=δ(x,y),‖b(x)−b(y)‖≤CdX(x,y)(x,y∈X).\lVert b(x)-b(y)\rVert^2=\delta(x,y), \qquad \lVert b(x)-b(y)\rVert\leq C\,d_X(x,y) \qquad(x,y\in X).(11)✦

In the paper, the result reads: for the associated disk kernels and every p∈Dp\in D, there is an injective Lipschitz map b:S1→Hb:S^1\to H into a separable real Hilbert space such that ‖b(s)−b(t)‖2=Dp(s,t)\lVert b(s)-b(t)\rVert^2=D_p(s,t) for all s,t∈S1s,t\in S^1.

What Is Checked

The boundary statement (10) is stated over the explicit disk kernels and the weighted Dirichlet space under (1) and (9). The criterion (11) is stated over nonempty separable pseudometric spaces and strictly positive off-diagonal kernels of negative type with a quadratic distance bound.

The Green kernel has square-integrable area rows for almost every first argument✦ and a uniform row norm bound on DD.✦ The integrals defining the L2L^2 kernel operators belong to their target L2L^2 spaces for every L2L^2 input.✦ Under (1), every Green column G(⋅,p)G(\cdot,p) with p∈Dp\in D is square integrable with respect to ν\nu. For finite ν\nu, the smoothed Poisson kernel P1(s,⋅)P_1(s,\cdot) is square integrable with respect to ν\nu for every s∈S1s\in S^1.✦