OctoInformBeta

OpenAI preprint, September 2026 · A variational principle for tangent vector fields

Proposition 3.1: A strict Dirichlet gap gives the first Neumann bound, with equality only for gradients of mean-zero Rayleigh minimizers

A spectral bound for weakly tangent fields.

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

Let Ω⊂ℝ2\Omega\subset\mathbb R^2 be a nonempty bounded simply connected open set. All L2L^2 spaces and inner products below use area measure on Ω\Omega; ∥⋅∥2\|\cdot\|_2 denotes the scalar or vector L2L^2 norm.

A test function φ∈Cc∞(Ω)\varphi\in C_c^\infty(\Omega) is a smooth real function on ℝ2\mathbb R^2 with compact closed support contained in Ω\Omega.✦ For a test function φ∈Cc∞(Ω)\varphi\in C_c^\infty(\Omega) and a constant vector e∈ℝ2e\in\mathbb R^2, the directional derivative ∂eφ=Dφ[e]\partial_e\varphi=D\varphi[e] is the derivative of φ\varphi in the direction ee✦, and φe\varphi e is the associated test vector field.✦ A function u∈L2(Ω)u\in L^2(\Omega) has weak gradient g∈L2(Ω;ℝ2)g\in L^2(\Omega;\mathbb R^2) when ∫Ωu∂eφ=−∫Ω(g⋅e)φ(φ∈Cc∞(Ω),e∈ℝ2).\int_\Omega u\,\partial_e\varphi =-\int_\Omega (g\cdot e)\varphi \qquad \bigl(\varphi\in C_c^\infty(\Omega),\ e\in\mathbb R^2\bigr).(1)✦ The Sobolev space H1(Ω)H^1(\Omega) consists of functions u∈L2(Ω)u\in L^2(\Omega) having weak gradients g∈L2(Ω;ℝ2)g\in L^2(\Omega;\mathbb R^2) satisfying (1).✦ Its scalar value and gradient maps are continuous linear maps into the respective L2L^2 spaces. The Sobolev class of a smooth test function has its classical gradient as weak gradient. The map Cc∞(Ω)→H1(Ω)C_c^\infty(\Omega)\to H^1(\Omega) sending each test function to its Sobolev class is linear.✦

The Dirichlet Sobolev space H01(Ω)H_0^1(\Omega) is the closure of Cc∞(Ω)C_c^\infty(\Omega) in H1(Ω)H^1(\Omega).✦ The mean-zero Sobolev space 𝒩1(Ω)\mathcal N^1(\Omega) is the subspace of H1(Ω)H^1(\Omega) with zero integral.✦ The value and gradient maps on these subspaces are the restrictions of those on H1(Ω)H^1(\Omega). The first positive Neumann value μ\mu of Ω\Omega is the infimum of the mean-zero Rayleigh quotients in (2): μ=inf⁡{∥∇u∥22∥u∥22:u∈𝒩1(Ω),∥u∥22>0}.\mu=\inf\left\{ \frac{\|\nabla u\|_2^2}{\|u\|_2^2}: u\in\mathcal N^1(\Omega),\ \|u\|_2^2>0 \right\}.(2)✦ The Sobolev mean m:H1(Ω)→ℝm:H^1(\Omega)\to\mathbb R is the continuous real linear functional assigning to each function its mean on Ω\Omega.✦

Weak Vector Data and Energy

Write e1,e2e_1,e_2 for the standard orthonormal basis of ℝ2\mathbb R^2 and ∂iu=(∇u)⋅ei\partial_i u=(\nabla u)\cdot e_i for a weak coordinate derivative. The quarter-turn JJ sends (x1,x2)(x_1,x_2) to (−x2,x1)(-x_2,x_1) and acts pointwise on vector L2L^2 fields. A field X∈L2(Ω;ℝ2)X\in L^2(\Omega;\mathbb R^2) has divergence f∈L2(Ω)f\in L^2(\Omega) with zero weak normal trace, and weak curl c∈L2(Ω)c\in L^2(\Omega), when the respective identities in (3) hold: ⟨X,∇w⟩L2=−⟨f,w⟩L2(w∈H1(Ω)),⟨X,J∇j⟩L2=−⟨c,j⟩L2(j∈H01(Ω)).\begin{aligned} \langle X,\nabla w\rangle_{L^2} &=-\langle f,w\rangle_{L^2} &&\bigl(w\in H^1(\Omega)\bigr),\\ \langle X,J\nabla j\rangle_{L^2} &=-\langle c,j\rangle_{L^2} &&\bigl(j\in H_0^1(\Omega)\bigr). \end{aligned}(3)

For i=1,2i=1,2, the coordinate embedding f↦feif\mapsto fe_i is the continuous real linear map from L2(Ω)L^2(\Omega) to L2(Ω;ℝ2)L^2(\Omega;\mathbb R^2) that places ff in coordinate ii and sets the other coordinate to zero.✦ For i=1,2i=1,2, the coordinate projection X↦X⋅eiX\mapsto X\cdot e_i is the continuous real linear map from L2(Ω;ℝ2)L^2(\Omega;\mathbb R^2) to L2(Ω)L^2(\Omega) that extracts coordinate ii.✦

For a Sobolev vector field X=(X1,X2)∈H1(Ω;ℝ2)X=(X_1,X_2)\in H^1(\Omega;\mathbb R^2), the associated square-integrable field is X1e1+X2e2X_1e_1+X_2e_2✦, and its divergence and curl are the L2L^2 functions in (4): div⁡X=∂1X1+∂2X2,curl⁡X=∂1X2−∂2X1.\operatorname{div}X=\partial_1X_1+\partial_2X_2, \qquad \operatorname{curl}X=\partial_1X_2-\partial_2X_1.(4) The space 𝒯(Ω)\mathcal T(\Omega) of weakly tangent Sobolev vector fields is the subspace of H1(Ω;ℝ2)H^1(\Omega;\mathbb R^2) consisting of fields XX satisfying the first identity of (3) with f=div⁡Xf=\operatorname{div}X.✦ For Sobolev vector fields X,Y∈H1(Ω;ℝ2)X,Y\in H^1(\Omega;\mathbb R^2) and λ∈ℝ\lambda\in\mathbb R, the divergence-curl energy QQ and its shifted bilinear form ℬλ\mathcal B_\lambda are defined by (5): Q(X)=∥div⁡X∥22+∥curl⁡X∥22,ℬλ(X,Y)=⟨div⁡X,div⁡Y⟩L2+⟨curl⁡X,curl⁡Y⟩L2−λ⟨X,Y⟩L2.\begin{aligned} Q(X)&=\|\operatorname{div}X\|_2^2+\|\operatorname{curl}X\|_2^2,\\ \mathcal B_\lambda(X,Y) &=\langle\operatorname{div}X,\operatorname{div}Y\rangle_{L^2} +\langle\operatorname{curl}X,\operatorname{curl}Y\rangle_{L^2} -\lambda\langle X,Y\rangle_{L^2}. \end{aligned}(5)

A weak Neumann eigenfunction u∈H1(Ω)u\in H^1(\Omega) at a real value λ\lambda is a function satisfying (6); the real first Neumann eigenspace VV consists of the solutions at λ=μ\lambda=\mu: ⟨∇u,∇w⟩L2=λ⟨u,w⟩L2(w∈H1(Ω)).\langle\nabla u,\nabla w\rangle_{L^2} =\lambda\langle u,w\rangle_{L^2} \qquad\bigl(w\in H^1(\Omega)\bigr).(6)✦

The Claim

Theorem 1. Assume μ>0\mu>0 and let L>μL>\mu be a real number satisfying the Dirichlet bound (7): L∥j∥22≤∥∇j∥22(j∈H01(Ω)).L\|j\|_2^2\leq\|\nabla j\|_2^2 \qquad\bigl(j\in H_0^1(\Omega)\bigr).(7) Let X∈L2(Ω;ℝ2)X\in L^2(\Omega;\mathbb R^2) have divergence ff with zero weak normal trace and weak curl cc, in the sense of (3). Then μ∥X∥22≤∥f∥22+∥c∥22.\mu\|X\|_2^2\leq\|f\|_2^2+\|c\|_2^2.(8)✦ If equality holds in (8), there is a function v∈𝒩1(Ω)v\in\mathcal N^1(\Omega) with X=∇v,c=0,μ∥v∥22=∥∇v∥22.X=\nabla v,\qquad c=0,\qquad \mu\|v\|_2^2=\|\nabla v\|_2^2.(9)✦

On any open planar set Ω\Omega, let λ∈ℝ\lambda\in\mathbb R and let u∈H1(Ω)u\in H^1(\Omega) satisfy (6). If X∈H1(Ω;ℝ2)X\in H^1(\Omega;\mathbb R^2) equals ∇u\nabla u almost everywhere, then X∈𝒯(Ω),ℬλ(X,X)=0.X\in\mathcal T(\Omega),\qquad \mathcal B_\lambda(X,X)=0.(10)✦

In the paper, the result reads: for Ω\Omega as above and every tangent X∈H1(Ω;ℝ2)X\in H^1(\Omega;\mathbb R^2), Q(X)≥μ∫Ω|X|2Q(X)\geq\mu\int_\Omega|X|^2, with equality exactly when X=∇vX=\nabla v for some v∈Vv\in V.

What Is Checked

The assertions (8) and (9) concern L2L^2 vector fields with the weak divergence and curl relations (3), while (10) concerns weak Neumann gradients whose components belong to H1(Ω)H^1(\Omega).