betaFeedback

WorksMathematics

The equilateral triangle maximises the first Steklov eigenvalue among triangles

Abstract

Let σ1(T)\sigma_1(T) be the first nonzero Steklov eigenvalue of a triangle TT and L(T)L(T) its perimeter. Among triangles the equilateral triangle EE uniquely maximises σ1L\sigma_1 L. If the shortest and longest sides of TT have lengths a≤ba\le b, then σ1(E)L(E)−σ1(T)L(T)≥(b−a)/(2b)\sigma_1(E)L(E)-\sigma_1(T)L(T)\ge (b-a)/(2b). Since ∣T∣/L(T)2|T|/L(T)^2 is also maximal at EE, the equilateral triangle uniquely maximises σ1∣T∣\sigma_1\sqrt{|T|} as well. This settles the case n=3n=3 of the Steklov analogue of the Pólya–Szegő conjecture for nn-gons, for both normalisations. The extremal value satisfies 3.8724615957722≤σ1(E)L(E)≤3.87246159577343.8724615957722\le \sigma_1(E)L(E)\le 3.8724615957734. For side length one, σ1(E)=σ2(E)<σ3(E)=23<σ4(E)\sigma_1(E)=\sigma_2(E)<\sigma_3(E)=2\sqrt3<\sigma_4(E). Under displacement of the apex over a fixed unit base, the first two perimeter-normalised eigenvalues split linearly with equal and opposite slopes independent of direction. The proof is computer-assisted. Near EE we transplant the first eigenspace of EE affinely; away from EE the trial function cos⁡(πx)\cos(\pi x) suffices. We verify the finite matrix and interval inequalities in ball arithmetic.

AI transparency statement. The main result, its computer-assisted proof, the partial Lean formalisation and the manuscript were produced by artificial intelligence. The human author initiated the broader research and directed the AI systems, but did not write the text and has not checked the arguments. Details are given at the end of the paper.

Introduction

For a bounded Lipschitz domain Ω⊂R2\Omega\subset\mathbb{R}^2, the Steklov problem Δu=0\Delta u=0 in Ω\Omega, ∂νu=σu\partial_\nu u=\sigma u on ∂Ω\partial\Omega has eigenvalues 0=σ0<σ1(Ω)≤σ2(Ω)≤⋯0=\sigma_0<\sigma_1(\Omega)\le\sigma_2(\Omega)\le\cdots. They satisfy

σ1(Ω)=min⁡u∈H1(Ω)∖RRΩ(u),RΩ(u)=∫Ω∣∇u∣2∫∂Ωu2−1∣∂Ω∣(∫∂Ωu)2.(1)\sigma_1(\Omega)=\min_{u\in H^1(\Omega)\setminus\mathbb{R}} \mathcal{R}_\Omega(u), \qquad\mathcal{R}_\Omega(u)=\frac{\int_\Omega|\nabla u|^2}{\int_{\partial\Omega}u^2-\frac{1}{|\partial\Omega|}\left(\int_{\partial\Omega}u\right)^2}. \tag*{(1)}

The scale-invariant quantity is

Σ(Ω):=σ1(Ω)L(Ω),L(Ω)=∣∂Ω∣.\Sigma(\Omega):=\sigma_1(\Omega)L(\Omega), \qquad L(\Omega)=|\partial\Omega|.

Weinstock’s inequality [16] states that Σ(Ω)≤2π\Sigma(\Omega)\le2\pi for simply connected planar domains, with equality only for the disk; see the survey [10]. For polygons with a prescribed number of sides, the natural conjecture is the analogue of Pólya–Szegő’s conjecture for the Dirichlet Laplacian:

Problem (Oudet; AIM problem list Steklov eigenproblems, Problem 4.6 [1]). Do regular nn-gons maximise σ1\sigma_1 among nn-gons of given area, or of given perimeter?

Dominguez, Nigam and Shahriari [7] gave numerical evidence for triangles using finite elements and Bayesian optimisation. Nigam also posed the perimeter question as Open Problem 8 at the 2020 miniconference on sharp eigenvalue estimates [13]. Among rectangles of given perimeter or area, the square maximises σ1\sigma_1: this is the planar case of the cuboid inequality of Girouard, Lagacé, Polterovich and Savo [9 Thm. 1.6]. Cheng, Gui, Hu, Li and Yao [4] state the conjecture for convex polygons with at most NN sides at fixed perimeter (their Conjecture 1.1), and leave it open for every N≥3N \ge3. They prove that σ1\sigma_1 of the perimeter-normalised regular NN-gon increases strictly with NN, and that its first eigenspace is two-dimensional. Dryden, Gordon, Moreno, Rowlett and Villegas-Blas [8] prove σk(T)L(T)≤833π2k2ϑ(T)\sigma_k(T)L(T) \le\frac{8\sqrt{3}}{3}\pi^2 k^2\vartheta(T) for triangles, where ϑ(T)\vartheta(T) is the smallest angle. This shows that degenerate triangles are far from optimal, but it does not compare triangles with EE.

For the Neumann problem, Laugesen and Siudeja [11] proved that the equilateral triangle maximises the first nonzero Neumann eigenvalue times L2L^2 among triangles, and hence times the area. Near the equilateral triangle their trial functions are the eigenfunctions of the equilateral triangle, transplanted by the linear map onto the given triangle. This affine transplantation is the precedent for our local argument. The Steklov problem adds two difficulties: the first eigenvalue of EE is double, so the maximum is a cone and the comparison degenerates at EE; and a rigorous lower bound for σ1(E)\sigma_1(E) is needed. We are not aware of a proof of the Steklov case for any n≥3n \ge3.

Theorem 1.1. Let EE be an equilateral triangle. For every triangle TT,

Σ(T)≤Σ(E),\Sigma(T) \le\Sigma(E),

with equality if and only if TT is equilateral. More precisely, normalise a longest side to [0,1][0,1] and write the apex as (p,q)(p,q) with p≤1/2p \le1/2 and q>0q > 0. If ρ=∣(p,q)−(1/2,3/2)∣\rho= \left|(p,q) - (1/2,\sqrt{3}/2)\right|, then

Σ(E)−Σ(T)≥ρ2≥ℓmax⁡−ℓmin⁡2ℓmax⁡,\Sigma(E) - \Sigma(T) \ge\frac{\rho}{2} \ge\frac{\ell_{\max} - \ell_{\min}}{2\ell_{\max}},

where ℓmax⁡\ell_{\max} and ℓmin⁡\ell_{\min} are the extreme side lengths of TT. Moreover

3.8724615957722≤Σ(E)≤3.8724615957734.3.8724615957722 \le\Sigma(E) \le3.8724615957734.

Girouard and Polterovich computed the first sixty normalised eigenvalues of the equilateral triangle numerically [10 Example 3.2.1]. For side length one we obtain σ1(E)=σ2(E)<σ3(E)=23<σ4(E)\sigma_1(E) = \sigma_2(E) < \sigma_3(E) = 2\sqrt{3} < \sigma_4(E) (Proposition 3.4). The double eigenvalue agrees with [4 Thm. 1.2] for N=3N = 3.

Corollary 1.2. For every triangle TT,

σ1(T)∣T∣≤σ1(E)∣E∣∈[0.8494076787253,0.8494076787257],\sigma_1(T)\sqrt{|T|} \le\sigma_1(E)\sqrt{|E|} \in[0.8494076787253, 0.8494076787257],

with equality if and only if TT is equilateral.

Proof. σ1∣T∣=Σ(T)(∣T∣/L2)1/2\sigma_1\sqrt{|T|} = \Sigma(T)(|T|/L^2)^{1/2}. By Heron’s formula and the AM–GM inequality, ∣T∣/L2≤3/36|T|/L^2 \le\sqrt{3}/36, with equality only for equilateral triangles. Now apply Theorem 1.1. □\square

Method

Upper bounds for σ1(T)\sigma_1(T) come directly from (1); a lower bound is needed only for the single domain EE. We parametrise triangles by the apex (p,q)(p,q) over the base [0,1][0,1] (Lemma 2.3), and let ρ\rho be the distance of (p,q)(p,q) from the equilateral apex.

  1. Near EE (ρ≤ρ1≈0.6\rho\le\rho_1 \approx0.6). The trial space is the exact first eigenspace of EE, transplanted by the affine map E→TE \to T. The comparison involves only a few moments of the eigenspace and two exact identities. These identities let us divide out the factor ρ\rho (Lemma 4.2), which turns the degenerate comparison at ρ=0\rho= 0 into a sign condition. We check the sign condition on boxes in ball arithmetic. The moments are enclosed using a harmonic-polynomial approximation, a residual bound for the Dirichlet-to-Neumann map, and spectral isolation of a cubic eigenfunction by a Crouzeix–Raviart bound for σ4(E)\sigma_4(E).

  1. Away from EE (ρ≥ρ1\rho\ge\rho_1). The single trial function cos⁡(πx)\cos(\pi x) gives Σ(T)≤3.114382798607\Sigma(T) \le3.114382798607, well below Σ(E)>3.8724\Sigma(E) > 3.8724. A one-dimensional check on 16 intervals suffices.

Preliminaries

The Dirichlet-to-Neumann map

Let Ω\Omega be a bounded Lipschitz domain. The Dirichlet-to-Neumann map Λ\Lambda is the self-adjoint operator on L2(∂Ω)L^2(\partial\Omega) associated with the closed form f↦∫Ω∣∇Hf∣2f \mapsto\int_\Omega\lvert\nabla Hf\rvert^2 on H1/2(∂Ω)H^{1/2}(\partial\Omega), where HfHf is the harmonic extension. It has compact resolvent and eigenvalues σk\sigma_k, with an orthonormal basis of eigenvectors χk\chi_k of L2(∂Ω)L^2(\partial\Omega); χ0\chi_0 is constant. If uu is a harmonic polynomial with trace ff, Green’s formula gives f∈dom⁡Λf \in\operatorname{dom}\Lambda and Λf=∂νu\Lambda f = \partial_\nu u. For σ~∈R\widetilde{\sigma} \in\mathbb{R} and η>0\eta> 0 write J=[σ~−η,σ~+η]J = [\widetilde{\sigma} - \eta,\widetilde{\sigma} + \eta].

Lemma 2.1 (Inclusion). Let W⊂dom⁡ΛW \subset\operatorname{dom}\Lambda have dimension kk, and suppose ∥(Λ−σ~)f∥≤η∥f∥\lVert(\Lambda- \widetilde{\sigma})f\rVert\le\eta\lVert f\rVert for all f∈Wf \in W. Then Λ\Lambda has at least kk eigenvalues, counted with multiplicity, in JJ.

Proof. Suppose not. Then some f∈W∖0f \in W \setminus0 is orthogonal to the spectral subspace of JJ, and

∥(Λ−σ~)f∥2=∑∣σj−σ~∣>η∣σj−σ~∣2∣⟨f,χj⟩∣2>η2∥f∥2,\lVert(\Lambda-\widetilde{\sigma})f\rVert^2=\sum_{\lvert\sigma_j-\widetilde{\sigma}\rvert>\eta}\lvert\sigma_j-\widetilde{\sigma}\rvert^2\lvert\langle f,\chi_j\rangle\rvert^2>\eta^2\lVert f\rVert^2,

a contradiction.

Lemma 2.2 (Eigenvector enclosure; cf. [6]). Let PP be the spectral projection of Λ\Lambda onto JJ. Suppose every eigenvalue of Λ\Lambda outside JJ is either 00 or at least σlo>σ~≥0\sigma^{\mathrm{lo}} > \widetilde{\sigma} \ge0, and put γ=σlo−σ~\gamma= \sigma^{\mathrm{lo}} - \widetilde{\sigma}. Let f∈dom⁡Λf \in\operatorname{dom}\Lambda be orthogonal to the constants, and r=(Λ−σ~)fr = (\Lambda- \widetilde{\sigma})f. Then

∥(I−P)f∥L2(∂Ω)≤∥r∥γ,∫Ω∣∇H(I−P)f∣2≤σloγ2∥r∥2.\lVert(I-P)f\rVert_{L^2(\partial\Omega)}\le\frac{\lVert r\rVert}{\gamma},\qquad\int_\Omega\lvert\nabla H(I-P)f\rvert^2\le\frac{\sigma^{\mathrm{lo}}}{\gamma^2}\lVert r\rVert^2.

Proof. (σj−σ~)⟨f,χj⟩=⟨r,χj⟩(\sigma_j-\widetilde{\sigma})\langle f,\chi_j\rangle=\langle r,\chi_j\rangle for every jj, and the component along χ0\chi_0 vanishes. The remaining jj with σj∉J\sigma_j \notin J have σj≥σlo\sigma_j\ge\sigma^{\mathrm{lo}}, so ∥(I−P)f∥2≤∥r∥2/γ2\lVert(I-P)f\rVert^2\le\lVert r\rVert^2/\gamma^2, and

∫Ω∣∇H(I−P)f∣2=∑jσj∣⟨r,χj⟩∣2(σj−σ~)2≤σlo∥r∥2γ2,\int_\Omega\lvert\nabla H(I-P)f\rvert^2=\sum_j\frac{\sigma_j\lvert\langle r,\chi_j\rangle\rvert^2}{(\sigma_j-\widetilde{\sigma})^2}\le\frac{\sigma^{\mathrm{lo}}\lVert r\rVert^2}{\gamma^2},

because σ/(σ−σ~)2\sigma/(\sigma-\widetilde{\sigma})^2 is decreasing for σ>σ~\sigma>\widetilde{\sigma} when σ~≥0\widetilde{\sigma}\ge0 (its derivative is −(σ+σ~)/(σ−σ~)3-(\sigma+\widetilde{\sigma})/(\sigma-\widetilde{\sigma})^3).

Parametrisation

For (p,q)(p,q) with q>0q>0, write

T(p,q)=conv⁡{(0,0),(1,0),(p,q)},T(p,q)=\operatorname{conv}\{(0,0),(1,0),(p,q)\},

and set

F={(p,q):0<p≤12, q>0, (1−p)2+q2≤1}.F=\{(p,q):0<p\le\tfrac{1}{2},\ q>0,\ (1-p)^2+q^2\le1\}.

Since Σ\Sigma is invariant under similarities, it suffices to treat the triangles T(p,q)T(p,q), (p,q)∈F(p,q)\in F.

Lemma 2.3. Every triangle is similar to T(p,q)T(p,q) for some (p,q)∈F(p,q)\in F, and T(p,q)T(p,q) is equilateral if and only if (p,q)=(12,32)(p,q)=(\tfrac{1}{2},\tfrac{\sqrt{3}}{2}).

Proof. Rescale so that a longest side has length 1. Move it to [0,1]×{0}[0,1] \times\{0\} with the third vertex (p,q)(p,q) in the upper half-plane, and reflect in x=12x=\frac{1}{2} if necessary so that p≤12p \le\frac{1}{2}. The side from (1,0)(1,0) to the apex is not longer than the base, so (1−p)2+q2≤1(1-p)^2+q^2 \le1; this forces p>0p>0. The other side then satisfies p2+q2≤(1−p)2+q2≤1p^2+q^2 \le(1-p)^2+q^2 \le1. The equilateral case is clear.

We write p=12+ρcos⁡ψp=\frac{1}{2}+\rho\cos\psi, q=32+ρsin⁡ψq=\frac{\sqrt{3}}{2}+\rho\sin\psi for polar coordinates about the equilateral apex; ρ\rho is the Euclidean distance in the (p,q)(p,q)-plane.

Affine transplantation

Let A∈GL2+(R)A \in GL_2^+(\mathbb{R}), u∈H1(Ω)u \in H^1(\Omega) and uA=u∘A−1∈H1(AΩ)u_A=u\circ A^{-1}\in H^1(A\Omega). Write G=A−1A−TG=A^{-1}A^{-T}. For an edge ee of a polygon Ω\Omega with unit tangent τe\tau_e, put se=∣Aτe∣s_e=|A\tau_e|. Changing variables gives

∫AΩ∣∇uA∣2=det⁡A∑i,jGij∫Ω∂iu ∂ju,∫AeuA2=se∫eu2,∫AeuA=se∫eu.(2)\begin{aligned} \int_{A\Omega}|\nabla u_A|^2 &= \det A \sum_{i,j}G_{ij}\int_\Omega\partial_i u\,\partial_j u,\\ \int_{Ae}u_A^2 &=s_e\int_e u^2,\qquad\int_{Ae}u_A=s_e\int_e u. \tag*{(2)} \end{aligned}

because the arc length on AeAe is ses_e times that on ee. For a finite-dimensional space V\mathcal{V} of trial functions, define the forms NA(u,u′)=∫∇uA⋅∇uA′\mathcal{N}_A(u,u')=\int\nabla u_A\cdot\nabla u'_A and DA(u,u′)=∑ese∫euu′−1L(AΩ)(∑ese∫eu)(∑ese∫eu′)\mathcal{D}_A(u,u')=\sum_e s_e\int_e uu'-\frac{1}{L(A\Omega)}\left(\sum_e s_e\int_e u\right)\left(\sum_e s_e\int_e u'\right) on V\mathcal{V}. If DA\mathcal{D}_A is positive definite on V\mathcal{V}, then (1) gives σ1(AΩ)≤min⁡V∖0NA/DA\sigma_1(A\Omega)\le\min_{\mathcal{V}\setminus0}\mathcal{N}_A/\mathcal{D}_A.

The equilateral triangle

Let E=T(12,32)E=T\left(\frac{1}{2},\frac{\sqrt{3}}{2}\right), with side 1 and L(E)=3L(E)=3. Let z0z_0 be its centroid and w=−i3(z−z0)w=-i\sqrt{3}(z-z_0), so that the circumradius is 1 in ww and the apex lies on the positive real ww-axis. The dihedral group D3D_3 acts by w↦e2πi/3ww\mapsto e^{2\pi i/3}w and w↦w‾w\mapsto\overline{w}.

An exactly equivariant trial pair

Given real coefficients ckc_k (k≤20k\le20, k≢0(mod3)k\not\equiv0\pmod{3}, fourteen in all), put f=∑k=120ckwk+∑k=220ckw‾kf=\sum_{k=1}^{20}c_kw^k+\sum_{k=2}^{20}c_k\overline{w}^k, u1=Re⁡fu_1=\operatorname{Re}f and u2=Im⁡fu_2=\operatorname{Im}f. Rotation by 2π/32\pi/3 multiplies ff by e2πi/3e^{2\pi i/3}, and reflection conjugates it. Hence (u1,u2)(u_1,u_2) transforms under D3D_3 exactly as the coordinates (Re⁡w,Im⁡w)(\operatorname{Re}w,\operatorname{Im}w) do, whatever the ckc_k are. In particular ∫∂Eua=0\int_{\partial E}u_a=0; each uau_a is the real part of a holomorphic polynomial faf_a (since Re⁡w‾k=Re⁡wk\operatorname{Re}\overline{w}^k=\operatorname{Re}w^k and Im⁡w‾k=−Im⁡wk\operatorname{Im}\overline{w}^k=-\operatorname{Im}w^k), and by Schur’s lemma all their Gram matrices are scalar. The ckc_k come from a floating-point Ritz computation and are then fixed as exact binary numbers.

Let σ~≈1.290820531924\widetilde{\sigma}\approx1.290820531924 be the Rayleigh quotient of u1u_1, rounded to binary64. All boundary integrals of products of uau_a and ∂νua\partial_\nu u_a are integrals of polynomials along segments, and we compute them exactly in ball arithmetic. This gives the relative residual

η:=max⁡f∈span⁡(u1,u2)∖0∥∂νf−σ~f∥L2(∂E)∥f∥L2(∂E)≤1.9559×10−13.\eta:=\max_{f\in\operatorname{span}(u_1,u_2)\setminus0}\frac{\|\partial_\nu f-\widetilde{\sigma}f\|_{L^2(\partial E)}}{\|f\|_{L^2(\partial E)}}\le1.9559\times10^{-13}.

By Lemma 2.1, ΛE\Lambda_E has at least two eigenvalues in J=[σ~−η,σ~+η]J=[\widetilde{\sigma}-\eta,\widetilde{\sigma}+\eta].

A Crouzeix–Raviart lower bound for σ4(E)\sigma_{4}(E)

Subdivide EE uniformly into nh2n_h^2 equilateral cells KK of side h=1/nhh=1/n_h. Let ChC_h be the Crouzeix–Raviart space [5]: piecewise linear functions that are continuous at the midpoints xex_e of interior edges. Let Πh\Pi_h be the interpolation by edge means, (Πhu)(xe)=∣e∣−1∫eu(\Pi_hu)(x_e)=|e|^{-1}\int_e u. Write ah(u,y)=∑K∫K∇u⋅∇ya_h(u,y)=\sum_K\int_K\nabla u\cdot\nabla y and b(u,y)=∫∂Euyb(u,y)=\int_{\partial E}uy, and let σh,0≤σh,1≤⋯\sigma_{h,0}\leq\sigma_{h,1}\leq\cdots be the finite eigenvalues of the pencil (ah,b)(a_h,b) on ChC_h. The min–max principle holds over subspaces on which bb is positive definite.

Lemma 3.1 (Trace constant). Let KK be an equilateral triangle of side hh, ee an edge of KK, and φ∈H1(K)\varphi\in H^1(K) with ∫eφ=0\int_e\varphi=0. Then

∥φ∥L2(e)2≤43(1π2+1π)h∥∇φ∥L2(K)2.\lVert\varphi\rVert_{L^2(e)}^2\leq\frac{4}{\sqrt{3}}\left(\frac{1}{\pi^2}+\frac{1}{\pi}\right)h\lVert\nabla\varphi\rVert_{L^2(K)}^2.

Proof. Let φ^=φ−φˉK\hat{\varphi}=\varphi-\bar{\varphi}_K. Since φ\varphi has mean zero on ee, ∥φ∥e≤∥φ−ζ∥e\lVert\varphi\rVert_e\leq\lVert\varphi-\zeta\rVert_e for every constant ζ\zeta; in particular ∥φ∥e≤∥φ^∥e\lVert\varphi\rVert_e\leq\lVert\hat{\varphi}\rVert_e. Let OeO_e be the vertex opposite ee. The field (x−Oe)φ^2(x-O_e)\hat{\varphi}^2 has normal component 00 on the two edges through OeO_e, and 2∣K∣/∣e∣2|K|/|e| times φ^2\hat{\varphi}^2 on ee. Its divergence is 2φ^2+2φ^∇φ^⋅(x−Oe)2\hat{\varphi}^2+2\hat{\varphi}\nabla\hat{\varphi}\cdot(x-O_e). The divergence theorem therefore gives

∥φ^∥e2=∣e∣∣K∣(∥φ^∥K2+∫Kφ^∇φ^⋅(x−Oe))≤∣e∣∣K∣(∥φ^∥K2+h∥φ^∥K∥∇φ∥K).\lVert\hat{\varphi}\rVert_e^2=\frac{|e|}{|K|}\left(\lVert\hat{\varphi}\rVert_K^2+\int_K\hat{\varphi}\nabla\hat{\varphi}\cdot(x-O_e)\right)\leq\frac{|e|}{|K|}\left(\lVert\hat{\varphi}\rVert_K^2+h\lVert\hat{\varphi}\rVert_K\lVert\nabla\varphi\rVert_K\right).

By the Payne–Weinberger inequality [15, 3] on the convex set KK of diameter hh, ∥φ^∥K≤hπ∥∇φ∥K\lVert\hat{\varphi}\rVert_K\leq\frac{h}{\pi}\lVert\nabla\varphi\rVert_K. Finally ∣e∣/∣K∣=4/(3h)|e|/|K|=4/(\sqrt{3}h).

Lemma 3.2 (Projection bound; cf. [14, 17]). Let nh≥2n_h\geq2 and Ch2=2⋅43(1π2+1π)hC_h^2=2\cdot\frac{4}{\sqrt{3}}\left(\frac{1}{\pi^2}+\frac{1}{\pi}\right)h. Then for each kk below the number of boundary edges,

σk(E)≥σh,k1+Ch2σh,k.\sigma_k(E)\geq\frac{\sigma_{h,k}}{1+C_h^2\sigma_{h,k}}.

Proof. (i) For every yh∈Chy_h\in C_h, ah(u−Πhu,yh)=0a_h(u-\Pi_hu,y_h)=0. Indeed ∇yh\nabla y_h is constant on each cell KK, and ∫K∇(u−Πhu)=∫∂K(u−Πhu)ν=0\int_K\nabla(u-\Pi_hu)=\int_{\partial K}(u-\Pi_hu)\nu=0 because Πhu∣K\Pi_hu|_K is linear and has the same edge means as uu.

(ii) Write φ=u−Πhu\varphi=u-\Pi_hu. It has mean zero on every edge of every cell. Each boundary edge lies in one cell, and for nh≥2n_h\geq2 a cell has at most two boundary edges (exactly two at the corners). Lemma 3.1 then gives b(φ,φ)≤Ch2ah(φ,φ)b(\varphi,\varphi)\leq C_h^2a_h(\varphi,\varphi).

(iii) Let ZkZ_k be the span of the eigenfunctions of σ0,…,σk\sigma_0,\ldots,\sigma_k, and take u∈Zku\in Z_k with b(u,u)=1b(u,u)=1. Put v=ah(u−Πhu,u−Πhu)v=a_h(u-\Pi_hu,u-\Pi_hu). By (i), ah(Πhu,Πhu)=∫∣∇u∣2−v≤σk−va_h(\Pi_hu,\Pi_hu)=\int|\nabla u|^2-v\leq\sigma_k-v, and by (ii), b(Πhu,Πhu)1/2≥1−Chvb(\Pi_hu,\Pi_hu)^{1/2}\geq1-C_h\sqrt{v}. If Ch2σk≥1C_h^2\sigma_k\geq1 the claim is trivial, since x/(1+Ch2x)<Ch−2x/(1+C_h^2x)<C_h^{-2}. Otherwise Πh\Pi_h is injective on ZkZ_k, and

ah(Πhu,Πhu)b(Πhu,Πhu)≤max⁡0≤v≤σkσk−v(1−Chv)2=σk1−Ch2σk.\frac{a_h(\Pi_hu,\Pi_hu)}{b(\Pi_hu,\Pi_hu)} \leq\max_{0\leq v\leq\sigma_k}\frac{\sigma_k-v}{(1-C_h\sqrt{v})^2} =\frac{\sigma_k}{1-C_h^2\sigma_k}.

The maximum is attained at v=Chσk\sqrt{v}=C_h\sigma_k. The min–max principle on the (k+1)(k+1)-dimensional space ΠhZk\Pi_hZ_k gives σh,k≤σk/(1−Ch2σk)\sigma_{h,k}\leq\sigma_k/(1-C_h^2\sigma_k), which rearranges to the claim.

Proposition 3.3. σ4(E)>3.5617858371486>23\sigma_4(E)>3.5617858371486>2\sqrt{3}.

Proof. Take nh=24n_h=24, which gives 900 degrees of freedom and Ch2≈0.080758C_h^2\approx0.080758. Let ShS_h and BhB_h be the matrices of aha_h and bb in the edge basis of ChC_h. On an equilateral cell the element stiffness matrix is

23(2−1−1−12−1−1−12).\frac{2}{\sqrt{3}} \begin{pmatrix} 2 & -1 & -1\\ -1 & 2 & -1\\ -1 & -1 & 2 \end{pmatrix}.

@ 񟿿 On a boundary edge, the basis function of that edge is ≡1\equiv1 and the other two are ±\pm the linear function rising from −1-1 to 11 along the edge, so the element boundary-mass entries are hh, h/3h/3 and −h/3-h/3. Let O\mathbf{O} be a fixed matrix with binary64 entries (computed once as approximate eigenvectors of Sh−5BhS_h-5B_h). In ball arithmetic we check two things:

  • the Gershgorin discs of OTO\mathbf{O}^{T}\mathbf{O} exclude 00, so O\mathbf{O} is nonsingular;

  • the Gershgorin discs of OT(Sh−5Bh)O\mathbf{O}^{T}(S_h-5B_h)\mathbf{O} consist of 44 discs in (−∞,0)(-\infty,0) and 896896 discs in (0,∞)(0,\infty).

Gershgorin’s counting theorem and Sylvester’s law of inertia give exactly four negative eigenvalues of Sh−5BhS_h-5B_h. If σh,4<5\sigma_{h,4}<5, the pencil would be negative definite at this shift on the span of five discrete eigenfunctions, including the constant. Thus σh,4≥5\sigma_{h,4}\ge5, and Lemma 3.2 gives σ4(E)≥5/(1+5Ch2)>3.5617858371486\sigma_4(E)\ge5/(1+5C_h^2)>3.5617858371486.

Proposition 3.4. For the side-one equilateral triangle,

σ1(E)=σ2(E)=:σ∗∈J,σ3(E)=23<σ4(E).\sigma_1(E)=\sigma_2(E)=:\sigma^*\in J,\qquad\sigma_3(E)=2\sqrt{3}<\sigma_4(E).

In particular the third positive eigenvalue is simple and

Σ(E)=3σ∗∈[3.8724615957722,3.8724615957734].\Sigma(E)=3\sigma^*\in[3.8724615957722,3.8724615957734].

Proof. Use Cartesian coordinates centred at the centroid and put r=1/(23)r=1/(2\sqrt{3}). The harmonic polynomial u=Im⁡((x+iy)3)+2r3u=\operatorname{Im}((x+iy)^3)+2r^3 satisfies ∂νu=−3x2+3r2=u/r\partial_\nu u=-3x^2+3r^2=u/r on y=−ry=-r. Rotation by 2π/32\pi/3 preserves uu, so the same boundary condition holds on every edge. Thus 1/r1/r is an eigenvalue. The inclusion interval JJ lies strictly between 00 and 1/r1/r and contains at least two eigenvalues counted with multiplicity. Together with 00 and 1/r1/r these account, with multiplicity, for all eigenvalues below the lower bound in Proposition 3.3. Hence JJ contains precisely σ1,σ2\sigma_1,\sigma_2, and σ3=1/r\sigma_3=1/r is simple. Take σlo=23\sigma^{\mathrm{lo}}=2\sqrt{3} in Lemma 2.2. Let PP be the spectral projection onto JJ. It commutes with the D3D_3-action, so (Pu1,Pu2)(Pu_1,Pu_2) is again an equivariant pair. Since σ~>0\widetilde{\sigma}>0, Lemma 2.2 gives ∥(I−P)ua∥≤η∥ua∥/(σlo−σ~)<10−12∥ua∥\lVert(I-P)u_a\rVert\le\eta\lVert u_a\rVert/(\sigma^{\mathrm{lo}}-\widetilde{\sigma})<10^{-12}\lVert u_a\rVert; since u1⊥u2u_1\perp u_2 with equal norms, Pu1,Pu2Pu_1,Pu_2 are linearly independent. They therefore span the two-dimensional range of PP, which carries the irreducible representation of D3D_3. By Schur’s lemma Λ\Lambda acts on this range as a scalar.

Local maximality

Let U\mathcal{U} be the eigenspace of σ∗\sigma^*, with basis ua∗=Puau_a^*=Pu_a (a=1,2a=1,2). For T=T(p,q)T=T(p,q), p=12+ρcos⁡ψp=\frac{1}{2}+\rho\cos\psi, q=32+ρsin⁡ψq=\frac{\sqrt{3}}{2}+\rho\sin\psi, we have T=A(E)T=A(E) with

A=I+ρ(0α0β),α=2cos⁡ψ3,β=2sin⁡ψ3,d:=det⁡A=1+ρβ.A=I+\rho\begin{pmatrix}0&\alpha\\0&\beta\end{pmatrix},\qquad\alpha=\frac{2\cos\psi}{\sqrt{3}},\qquad\beta=\frac{2\sin\psi}{\sqrt{3}},\qquad d:=\det A=1+\rho\beta.

Number the edges e0=[(0,0),(1,0)]e_0=[(0,0),(1,0)], e1=[(1,0),apex]e_1=[(1,0),\text{apex}] and e2=[apex,(0,0)]e_2=[\text{apex},(0,0)]. Then s0=1s_0=1, while

s1=∣(ρcos⁡ψ−12,32+ρsin⁡ψ)∣,s2=∣(ρcos⁡ψ+12,32+ρsin⁡ψ)∣s_1=\left|\left(\rho\cos\psi-\frac{1}{2},\frac{\sqrt{3}}{2}+\rho\sin\psi\right)\right|,\qquad s_2=\left|\left(\rho\cos\psi+\frac{1}{2},\frac{\sqrt{3}}{2}+\rho\sin\psi\right)\right|

are the side lengths of TT, and LT=1+s1+s2L_T=1+s_1+s_2.

The moments of U\mathcal{U} are the following 2×22\times2 matrices and vectors, indexed by a,ba,b: the energy matrices Qij=(∫E∂iua∗∂jub∗)Q_{ij}=\left(\int_E\partial_i u_a^*\partial_j u_b^*\right), the edge Gram matrices ge=(∫eua∗ub∗)g^e=\left(\int_e u_a^*u_b^*\right), and the edge-mean vectors μe=(∫eua∗)\mu^e=\left(\int_e u_a^*\right). They satisfy two exact identities:

∑eμe=0,Q11+Q22=σ∗∑ege.(3)\sum_e\mu^e=0,\qquad Q_{11}+Q_{22}=\sigma^*\sum_e g^e. \tag*{(3)}

The first holds because eigenfunctions of Λ\Lambda with σ∗≠0\sigma^{*} \ne0 are orthogonal to constants. The second is the weak eigen-equation on U\mathcal{U}.

Lemma 4.1 (Variational comparison). Let a trial space on a polygon TT have energy matrix NN and positive definite boundary-variance matrix DD. For a reference value SS put X=(S/L(T))D−NX = (S/L(T))D - N. Then

S−Σ(T)≥L(T)max⁡ξ≠0ξTXξξTDξ.S - \Sigma(T) \ge L(T) \max_{\xi\ne0} \frac{\xi^{T}X\xi}{\xi^{T}D\xi}.

In particular, if X−tDX - tD has a positive direction, the deficit exceeds L(T)tL(T)t.

Proof. Apply (1) to each trial function and subtract the resulting bound from SS.

Apply the lemma with S=3σ∗S = 3\sigma^{*}, N=NAN = N_A and D=DAD = D_A on U\mathcal{U}. At ρ=0\rho= 0 the eigen-equation gives X=0X = 0. The following factorisation isolates the leading term.

Lemma 4.2 (X=ρYX = \rho Y). Let λe=(se−1)/ρ\lambda_e = (s_e - 1)/\rho. Then λe=(κe+ρ)/(1+se)\lambda_e = (\kappa_e + \rho)/(1 + s_e), where κ1=−cos⁡ψ+3sin⁡ψ\kappa_1 = -\cos\psi+ \sqrt{3}\sin\psi and κ2=cos⁡ψ+3sin⁡ψ\kappa_2 = \cos\psi+ \sqrt{3}\sin\psi. Assume (3). Then X=ρYX = \rho Y, where

Y=σ∗(∑eωege−3ρLT2vvT)−∑i,jHijQij,H=(β+ρα2/d−α/d−α/d−β/d),Y = \sigma^{*}\left(\sum_e \omega_e g^e - \frac{3\rho}{L_T^2}vv^T\right) - \sum_{i,j} H_{ij}Q_{ij}, \qquad H = \begin{pmatrix} \beta+ \rho\alpha^2/d & -\alpha/d \\ -\alpha/d & -\beta/d \end{pmatrix},

with ω0=−(λ1+λ2)/LT\omega_0 = -(\lambda_1 + \lambda_2)/L_T, ω1=(2λ1−λ2)/LT\omega_1 = (2\lambda_1 - \lambda_2)/L_T, ω2=(2λ2−λ1)/LT\omega_2 = (2\lambda_2 - \lambda_1)/L_T, and v=λ1μ1+λ2μ2v = \lambda_1\mu^1 + \lambda_2\mu^2.

Proof. First, se2=1+ρκe+ρ2s_e^2 = 1 + \rho\kappa_e + \rho^2, so se−1=ρ(κe+ρ)/(1+se)s_e - 1 = \rho(\kappa_e + \rho)/(1 + s_e).

Next, A−1=(1−ρα/d01/d1/d)A^{-1} = \begin{pmatrix} 1-\rho\alpha/d & 0 \\ 1/d & 1/d \end{pmatrix}, which gives dG−I=ρHdG - I = \rho H.

For the boundary terms, ∑eseμe=∑e(se−1)μe=ρv\sum_e s_e\mu^e = \sum_e(s_e - 1)\mu^e = \rho v by (3). Also (3/LT)se−1=(3se−LT)/LT=ρωe(3/L_T)s_e - 1 = (3s_e - L_T)/L_T = \rho\omega_e, using s0=1s_0 = 1.

Hence σ∗(3/LT)DA−σ∗∑ege=ρσ∗(∑eωege−3ρvvT/LT2)\sigma^{*}(3/L_T)D_A-\sigma^{*}\sum_e g^e = \rho\sigma^{*}\left(\sum_e\omega_e g^e - 3\rho vv^T/L_T^2\right), and NA−(Q11+Q22)=ρ∑HijQijN_A-(Q_{11}+Q_{22}) = \rho\sum H_{ij}Q_{ij}. Subtracting these two equations and using (3) gives X=ρYX = \rho Y.

Lemma 4.3 (Moment enclosures). Let δa=ua−ua∗\delta_a = u_a - u_a^{*} and γ=23−σ~\gamma= 2\sqrt{3} - \tilde{\sigma}. Then

∥δa∥L2(∂E)≤ε∂:=η∥ua∥∂E/γ,∥∇δa∥L2(E)≤ε∇:=2 3η∥ua∥∂E/γ.\lVert\delta_a\rVert_{L^2(\partial E)} \le\varepsilon_{\partial} := \eta\lVert u_a\rVert_{\partial E}/\gamma,\qquad\lVert\nabla\delta_a\rVert_{L^2(E)} \le\varepsilon_{\nabla} := \sqrt{2}\,3\eta\lVert u_a\rVert_{\partial E}/\gamma.

Consequently each moment of UU differs from the corresponding moment of (u1,u2)(u_1,u_2) by at most the following amounts:

  • ε∇(∥∇ua∥+∥∇ub∥+ε∇)\varepsilon_{\nabla}(\lVert\nabla u_a\rVert+ \lVert\nabla u_b\rVert+ \varepsilon_{\nabla}) for the entry (a,b)(a,b) of QijQ_{ij};

  • ε∂(∥ua∥∂E+∥ub∥∂E+ε∂)\varepsilon_{\partial}(\lVert u_a\rVert_{\partial E} + \lVert u_b\rVert_{\partial E} + \varepsilon_{\partial}) for the entry (a,b)(a,b) of geg^e;

  • ∣e∣1/2ε∂\lvert e\rvert^{1/2}\varepsilon_{\partial} for the entry aa of μe\mu^e.

Proof. Apply Lemma 2.2 to f=uaf = u_a, with ∥r∥≤η∥ua∥\lVert r\rVert\le\eta\lVert u_a\rVert; its hypotheses were verified in the proof of Proposition 3.4. The moment bounds follow from Cauchy–Schwarz.

Here ∥ua∥∂E2≈1.577442\lVert u_a\rVert_{\partial E}^2 \approx1.577442, so ε∂≤1.14×10−13\varepsilon_{\partial} \le1.14 \times10^{-13} and ε∇≤2.11×10−13\varepsilon_{\nabla} \le2.11 \times10^{-13}. We compute the moments of uau_a exactly by Green’s formula ∫Eg dA=12i∮∂Ezˉg dz\int_E g\,\mathrm{d}A = \frac{1}{2\mathrm{i}}\oint_{\partial E}\bar{z}g\,\mathrm{d}z and ∫Efbfˉa dA=12i∮∂Efˉbfa dz\int_E f_b\bar{f}_a\,\mathrm{d}A = \frac{1}{2\mathrm{i}}\oint_{\partial E}\bar{f}_b f_a\,\mathrm{d}z for holomorphic gg. This turns every area integral into exact polynomial line integrals.

Proposition 4.4. Let ρ1\rho_1 be the binary64 number nearest 0.60.6. For 0<ρ≤ρ10 < \rho\le\rho_1 and every direction, Σ(E)−Σ(T(p,q))>ρ/2\Sigma(E) - \Sigma(T(p,q)) > \rho/2.

Proof. We cover [0,ρ1]×[0,ψmax⁡][0,\rho_1]\times[0,\psi_{\max}] by 907 closed boxes, where ψmax⁡\psi_{\max} is the binary64 number nearest 6.3 and exceeds 2π2\pi. The boxes lie in the rectangle, have disjoint interiors, and their exact rational areas sum to its area. They cover the rectangle since a nonempty relatively open complement would have positive area. We evaluate YY and DAD_A on every box using Lemma 4.3. For Z=2Y−DA/2Z=2Y-D_A/2 the checks give

(DA)11>0,det⁡DA>0,tr⁡Z>0 or det⁡Z<0.(D_A)_{11}>0,\qquad\det D_A>0,\qquad\operatorname{tr}Z>0\ \text{or}\ \det Z<0.

Thus Y−DA/4Y-D_A/4 has a positive direction. Since X=ρYX=\rho Y, Lemma 4.1 gives a deficit greater than LTρ/4L_T\rho/4. The other two sides have total length greater than the unit base, so LT>2L_T>2.

Theorem 4.5 (Splitting of the first pair). For a fixed unit base let the apex move from (1/2,3/2)(1/2,\sqrt{3}/2) by h=ρ(cos⁡ψ,sin⁡ψ)\mathbf{h}=\rho(\cos\psi,\sin\psi). As ρ↓0\rho\downarrow0, uniformly in ψ\psi,

LTσ1(T)=Σ(E)−cρ+O(ρ2),LTσ2(T)=Σ(E)+cρ+O(ρ2),L_T\sigma_1(T)=\Sigma(E)-c\rho+O(\rho^2),\qquad L_T\sigma_2(T)=\Sigma(E)+c\rho+O(\rho^2),

where the exact constant satisfies 3.5332113353<c<3.53321133553.5332113353<c<3.5332113355.

Proof. Work on H=H1(E)/RH=H^1(E)/\mathbb{R} with the Dirichlet inner product. On this fixed space let ah=LTNAa_h=L_TN_A and bh=DAb_h=D_A be the pulled-back forms of (2), now acting on all of HH. Their representing operators Ph,BhP_h,B_h are real analytic in the Cartesian components of hh: the energy coefficients are rational in AA, and the edge lengths and perimeter are positive near h=0h=0. The operator PhP_h is uniformly positive there. The boundary variance is unchanged by adding constants and the trace is compact into L2(∂E)L^2(\partial E), so BhB_h is compact. Positive eigenvalues of Kh=Ph−1/2BhPh−1/2K_h=P_h^{-1/2}B_hP_h^{-1/2} are the reciprocals of the normalised positive Steklov eigenvalues, by the weak equation on the quotient (choose the representative of zero weighted boundary mean). The inverse square root is analytic near P0=3IP_0=3I by its power series.

At h=0h=0 the top eigenvalue of K0K_0 is isolated and double by Proposition 3.4. The resolvent on a small enclosing circle has a convergent Neumann series for small ∣h∣|h|. The Riesz projection defined by this contour is real analytic and has rank two by continuity. Project a basis at 0 and apply Ph−1/2P_h^{-1/2} to obtain an analytic frame of the corresponding pencil subspace. Multiplication by the inverse square root of its positive bhb_h-Gram matrix makes the frame bhb_h-orthonormal. In this frame the matrix M(h)M(h) of aha_h is symmetric and real analytic, with eigenvalues LTσ1(T),LTσ2(T)L_T\sigma_1(T),L_T\sigma_2(T). Write

M(h)=Σ(E)I+F(h)+O(∣h∣2),M(h)=\Sigma(E)I+F(h)+O(|h|^2),

where FF is linear and the remainder is in matrix norm, uniformly in direction.

For a unit vector e=(cos⁡ψ,sin⁡ψ)e=(\cos\psi,\sin\psi) the matrix F(e)F(e) represents ∂e(ah−Σ(E)bh)∣0\left.\partial_e(a_h-\Sigma(E)b_h)\right|_0 on the unperturbed eigenspace in a b0b_0-orthonormal basis. Indeed, differentiating the moving frame introduces two further terms, each zero by the weak eigen-equation against every vector of HH. In the original basis ua∗u^*_a the same derivative is −3Y(0,ψ)-3Y(0,\psi), since ah−Σ(E)bh=−LTρYa_h-\Sigma(E)b_h=-L_T\rho Y on that fixed space. Consequently

det⁡F(e)=9det⁡Y(0,ψ)det⁡DI.\det F(e)=9\frac{\det Y(0,\psi)}{\det D_I}.

To see why the slope is independent of ee, identify the apex with z=p+iqz=p+iq. Cyclic relabelling of the vertices followed by a similarity replaces zz by f(z)=1/(1−z)f(z)=1/(1-z). It fixes z∗=eiπ/3z_* = e^{i\pi/3} and f′(z∗)=e2πi/3f'(z_*)=e^{2\pi i/3}. Thus the two normalised eigenvalues agree as an unordered pair at zz and f(z)f(z). Both the trace and the determinant of M(h)−Σ(E)IM(h)-\Sigma(E)I are therefore invariant under h↦f(z∗+h)−z∗h\mapsto f(z_*+h)-z_*. Comparing the terms of degree one in the trace and degree two in the determinant shows that tr⁡F(h)\operatorname{tr}F(h) and det⁡F(h)\det F(h) are invariant under rotation by 2π/32\pi/3. The only invariant linear form is zero, and every invariant real quadratic form is a multiple of ∣h∣2|h|^2 (its symmetric matrix commutes with that rotation). Hence F(h)F(h) has eigenvalues ±c∣h∣\pm c|h|, where

c=3−det⁡Y(0,0)/det⁡DI.c = 3\sqrt{-\det Y(0,0)/\det D_I}.

Ball evaluation using the enclosed moments of the projected eigenfunctions gives the stated positive enclosure for cc. The min–max principle bounds the change in each ordered eigenvalue by the norm of the matrix remainder, proving both expansions.

The far region and the proof of the theorem

For 0≤p≤120 \le p \le\frac{1}{2} put

qmax⁡(p)=min⁡(2p−p2,32−ρ12−(p−12)2).q_{\max}(p)=\min\left(\sqrt{2p-p^2},\frac{\sqrt{3}}{2}-\sqrt{\rho_1^2-\left(p-\frac{1}{2}\right)^2}\right).

Lemma 5.1. (i) F∩{ρ≥ρ1}={(p,q):0<p≤12, 0<q≤qmax⁡(p)}F\cap\{\rho\ge\rho_1\}=\{(p,q):0<p\le\frac{1}{2},\ 0<q\le q_{\max}(p)\}. (ii) For (p,q)∈F(p,q)\in F, with ℓ1=∣(p,q)∣\ell_1=|(p,q)| and ℓ2=∣(1−p,q)∣\ell_2=|(1-p,q)|,

Σ(T(p,q))≤π2qΞ(p)(1+ℓ1+ℓ2),Ξ(p)=I(p)+I(1−p),\Sigma(T(p,q))\le\pi^2q\Xi(p)(1+\ell_1+\ell_2),\qquad\Xi(p)=\mathcal{I}(p)+\mathcal{I}(1-p),

where

I(t)=1t∫0txsin⁡2(πx) dx=t4−sin⁡2πt4π+sin⁡(πt)sinc⁡(πt)4π\mathcal{I}(t)=\frac{1}{t}\int_0^t x\sin^2(\pi x)\,\mathrm{d}x=\frac{t}{4}-\frac{\sin2\pi t}{4\pi}+\frac{\sin(\pi t)\operatorname{sinc}(\pi t)}{4\pi}

and sinc⁡x=sin⁡x/x\operatorname{sinc}x=\sin x/x.

Proof. (i) For (p,q)∈F(p,q)\in F we have q≤2p−p2≤32q\le\sqrt{2p-p^2}\le\frac{\sqrt{3}}{2}, and ρ≥ρ1\rho\ge\rho_1 means (q−32)2≥ρ12−(p−12)2\left(q-\frac{\sqrt{3}}{2}\right)^2\ge\rho_1^2-\left(p-\frac{1}{2}\right)^2, where the right side is positive because ∣p−12∣≤12<ρ1\left|p-\frac{1}{2}\right|\le\frac{1}{2}<\rho_1. Since q<32+ρ12−(p−12)2q<\frac{\sqrt{3}}{2}+\sqrt{\rho_1^2-\left(p-\frac{1}{2}\right)^2}, this is equivalent to q≤32−ρ12−(p−12)2q\le\frac{\sqrt{3}}{2}-\sqrt{\rho_1^2-\left(p-\frac{1}{2}\right)^2}.

(ii) Take u=cos⁡(πx)u=\cos(\pi x) in (1). The vertical section of TT at abscissa xx has length qx/pqx/p for x≤px\le p and q(1−x)/(1−p)q(1-x)/(1-p) for x≥px\ge p, so the substitution x↦1−xx\mapsto1-x on [p,1][p,1] gives

∫T∣∇u∣2=π2q(1p∫0pxsin⁡2πx dx+11−p∫01−pxsin⁡2πx dx)=π2qΞ(p).\int_T|\nabla u|^2=\pi^2q\left(\frac{1}{p}\int_0^p x\sin^2\pi x\,\mathrm{d}x+\frac{1}{1-p}\int_0^{1-p}x\sin^2\pi x\,\mathrm{d}x\right)=\pi^2q\Xi(p).

The antiderivative x24−xsin⁡2πx4π−cos⁡2πx8π2\frac{x^2}{4}-\frac{x\sin2\pi x}{4\pi}-\frac{\cos2\pi x}{8\pi^2} and the identity (1−cos⁡2πt)/t=2πsin⁡(πt)sinc⁡(πt)(1-\cos2\pi t)/t=2\pi\sin(\pi t)\operatorname{sinc}(\pi t) give the closed form of I\mathcal{I}. The denominator of (1) is min⁡ζ∫∂T(u−ζ)2\min_{\zeta}\int_{\partial T}(u-\zeta)^2. On the side from (0,0)(0,0) to (p,q)(p,q) arc length is (ℓ1/p) dx(\ell_1/p)\,\mathrm{d}x with ℓ1/p≥1\ell_1/p\ge1, and similarly on the third side. So for every constant ζ\zeta,

∫∂T(u−ζ)2≥2∫01(cos⁡πx−ζ)2 dx=1+2ζ2≥1.\int_{\partial T}(u-\zeta)^2\ge2\int_0^1(\cos\pi x-\zeta)^2\,\mathrm{d}x=1+2\zeta^2\ge1.

Hence Σ(T)≤L∫T∣∇u∣2\Sigma(T)\le L\int_T|\nabla u|^2 with L=1+ℓ1+ℓ2L=1+\ell_1+\ell_2.

Proposition 5.2. For every (p,q)∈F(p,q)\in F with ρ≥ρ1\rho\ge\rho_1, Σ(T(p,q))≤3.114382798607<Σ(E)\Sigma(T(p,q))\le3.114382798607<\Sigma(E).

Proof. By Lemma 5.1(i) these are the points with 0<p≤120<p\le\frac{1}{2}, 0<q≤qmax⁡(p)0<q\le q_{\max}(p). The bound of Lemma 5.1(ii) is nondecreasing in qq, and Ξ≥0\Xi\ge0. On an interval p−≤p≤p+p_-\le p\le p_+, the first entry of qmax⁡q_{\max} is increasing in pp and the second is decreasing, so

qmax⁡(p)≤q+:=min⁡(2p+−p+2,32−ρ12−(p−−12)2).q_{\max}(p)\le q_+:=\min\left(\sqrt{2p_+-p_+^2},\frac{\sqrt{3}}{2}-\sqrt{\rho_1^2-\left(p_--\frac{1}{2}\right)^2}\right).

Moreover ℓ1≤(p+2+q+2)1/2\ell_1\le(p_+^2+q_+^2)^{1/2} and ℓ2≤((1−p−)2+q+2)1/2\ell_2\le((1-p_-)^2+q_+^2)^{1/2}. Hence on that interval

Σ(T)≤π2q+sup⁡[p−,p+]Ξ(1+p+2+q+2+(1−p−)2+q+2).\Sigma(T)\le\pi^2q_+\sup_{[p_-,p_+]}\Xi\left(1+\sqrt{p_+^2+q_+^2}+\sqrt{(1-p_-)^2+q_+^2}\right).

We evaluate the right side in ball arithmetic on the 16 intervals [k32,k+132]\left[\frac{k}{32},\frac{k+1}{32}\right], 0≤k≤150\le k\le15, which cover [0,12]\left[0,\frac{1}{2}\right]. Each upper endpoint is below 3.1143827986073.114382798607.

Proof of Theorem 1.1 Use the normalisation of Lemma 2.3. The deficit vanishes at ρ=0\rho=0 and exceeds ρ/2\rho/2 for 0<ρ≤ρ10<\rho\leq\rho_1 by Proposition 4.4. For ρ≥ρ1\rho\geq\rho_1, Proposition 5.2 gives a deficit at least 3.8724615957722−3.114382798607>1/23.8724615957722-3.114382798607>1/2. Throughout FF we have 0<p≤1/20<p\leq1/2 and 0<q≤3/20<q\leq\sqrt{3}/2, so ρ<1\rho<1. Finally ∣(p,q)∣=ℓmin⁡/ℓmax⁡\lvert(p,q)\rvert=\ell_{\min}/\ell_{\max} in these coordinates, and the reverse triangle inequality gives 1−∣(p,q)∣≤ρ1-\lvert(p,q)\rvert\leq\rho. This proves both deficit bounds and the equality case. The extremal value is enclosed in Proposition 3.4.

Corollary 5.3 (Fixed-profile thin limit). Fix 0<p<10<p<1, and set Φp(x)=min⁡{x/p,(1−x)/(1−p)}\Phi_p(x)=\min\{x/p,(1-x)/(1-p)\} on [0,1][0,1]. Define

θ(p)=inf⁡0≠v∈H1(0,1)∫01v=0∫01Φp(x)∣v′(x)∣2 dx2∫01∣v(x)∣2 dx.\theta(p)=\inf_{\substack{0\ne v\in H^1(0,1)\\ \int_0^1 v=0}}\frac{\int_0^1\Phi_p(x)\lvert v'(x)\rvert^2\,\mathrm{d}x}{2\int_0^1\lvert v(x)\rvert^2\,\mathrm{d}x}.

Then σ1(T(p,q))=qθ(p)+o(q)\sigma_1(T(p,q))=q\theta(p)+o(q) and Σ(T(p,q))=2qθ(p)+o(q)\Sigma(T(p,q))=2q\theta(p)+o(q) as q↓0q\downarrow0.

Proof. In Henrot and Michetti’s thin-domain lemma [12 Lemma 3.2] take the upper profile Φp\Phi_p, the lower profile zero, and thickness qq. Their lower-growth hypothesis holds since Φp(x)≥x(1−x)\Phi_p(x)\geq x(1-x). Their first one-dimensional eigenvalue is 2θ(p)2\theta(p), and their asymptotic formula multiplies it by q/2q/2. Now use L(T(p,q))→2L(T(p,q))\to2.

Implementation

All certified steps use Arb [2] (python-flint 0.9.0), at 64–200 bits.

Certificate data. Floating-point arithmetic is used only to produce candidate data, which are then fixed and treated as exact binary numbers: the coefficients ckc_k and the congruence matrix O\mathbf{O}. These are committed to the repository, and the certified steps read only them. The shift σ~\widetilde{\sigma} is computed from the ckc_k in ball arithmetic and rounded to binary64. Hence the certified results do not depend on the linear-algebra library or the processor; regenerating the data is a separate, optional step. Every number of this paper that a computation checks is kept in a single claims file. The checking script compares each computed enclosure with these claims in exact rational arithmetic, and the paper’s macros are generated from the same file.

Boxes. Box endpoints are exact binary numbers. A parameter box [x0,x1][x_0,x_1] enters as the Arb union of its two exact endpoints, which contains the whole interval. Square roots of nonnegative quantities whose balls straddle 0 are clamped to [0,⋅][0,\sqrt{\cdot}]. In the code, ρ1\rho_1 is the binary double nearest 0.6, which is 0.59999999999999997779…0.59999999999999997779\ldots, slightly below 0.6. The local and the far computations use this same value, so the two regions meet exactly. The tiling of the local boxes is checked with exact float comparisons and an exact rational area sum.

Consistency checks (not part of the proof).

  • The Green-formula moments agree with brute-force quadrature on random harmonic polynomials to 5×10−145\times10^{-14}.

  • The transplant bound of § 5, and the full Rayleigh quotient of cos⁡(πx)\cos(\pi x), agree with direct quadrature on TT to about 10−1410^{-14}.

  • At 20 sample triangles both bounds lie above the floating-point Ritz value of Σ\Sigma, as upper bounds must.

  • The Crouzeix–Raviart discrete values are σh,1=1.28042\sigma_{h,1}=1.28042 for nh=8n_h=8 and 1.287901.28790 for nh=16n_h=16, increasing towards σ∗=1.29082\sigma^{*}=1.29082.

Code names. Some identifiers differ from the notation here. In the Python code, σ~\widetilde{\sigma} is sig, the edge means μe\mu^{e} are mm, λ1,λ2\lambda_1,\lambda_2 are d1, d2, the ωe\omega_e are ce, and DA\mathcal{D}_A is D. In Lean, the μe\mu^{e} are m0x, ..., m2y, the ωe\omega_e are c0, c1, c2, and NA,DA\mathcal{N}_A,\mathcal{D}_A are NA, DA.

Lean. A Lean 4/Mathlib development formalises the elementary steps:

  • the triangle isoperimetric inequality in Heron form, with its equality case, and the deduction of Corollary 1.2;

  • the algebra of Lemma 4.2;

  • the integral defining I\mathcal{I} in Lemma 5.1, and the two edge integrals of cos⁡(πx)\cos(\pi x) used in the consistency checks.

It uses no sorry, and its only axioms are propext, Classical.choice and Quot.sound. It does not formalise the spectral theory, the trace and Payne–Weinberger inequalities, or the ball-arithmetic computations.

Availability. The code, the logs, the box certificates and the Lean files are in the accompanying repository2. The single command scripts/verify.py reruns all of them and writes a PASS/FAIL summary.

Numerical observations (not rigorous)

Except for the certified upper bounds in Table 1, the statements in this section are numerical. They use the harmonic-polynomial Ritz method, with trial space {Re⁡wk,Im⁡wk}k≤n\{\operatorname{Re} w^k,\operatorname{Im} w^k\}_{k\le n} and all integrals computed on the boundary.

polygoncertified upper bound for Σ\Sigmaestimate of Σ\Sigma
60–60–603.87246159577293.8724615957728 ±1×10−13\pm1 \times10^{-13}
50–50–802.87773098573772.87773098574 ±3×10−11\pm3 \times10^{-11}
70–70–402.57307657289182.573076572892 ±2×10−12\pm2 \times10^{-12}
40–60–802.53503581509352.53503581509 ±2×10−11\pm2 \times10^{-11}
45–45–902.49607237004762.4960723700475 ±1×10−13\pm1 \times10^{-13}
30–60–901.90138175772561.9013817577255 ±1×10−13\pm1 \times10^{-13}
30–30–1201.56034408001201.5603441 ±3×10−7\pm3 \times10^{-7}
square5.50602193869025.5060219386901 ±1×10−13\pm1 \times10^{-13}
regular pentagon5.97747255504395.9774726 ±4×10−7\pm4 \times10^{-7}
regular hexagon6.13557014559366.135570 ±5×10−6\pm5 \times10^{-6}

Table 1. Values of Σ=σ1L\Sigma= \sigma_1 L. The ±\pm values are heuristic error estimates. Rounding the central estimate to the displayed precision can place it slightly above the certified upper bound.

Observation 7.1 (Global picture). On a 25×3025 \times30 grid over FF (double precision, n=30n=30) the largest value of Σ\Sigma is attained at the equilateral apex. Table 1 lists selected values, sorted by Σ\Sigma, together with the regular NN-gons for N=4,5,6N=4,5,6. They are computed as follows:

  • the Ritz matrices are exact polynomial line integrals, evaluated in ball arithmetic;

  • the eigenproblem is solved in 50- and 80-digit arithmetic with the full basis of 2n2n directions, for n=30,40,50n=30,40,50;

  • the estimate is the value for n=50n=50;

  • its error is estimated (not rigorously) as twice the largest of the last change in nn, a geometric and a power-law extrapolation of the remaining changes, and the change between the two precisions; the estimate is printed only to the digits this error justifies;

  • the certified upper bound is the Rayleigh quotient of the fixed n=50n=50 trial function, evaluated in ball arithmetic on the exact polygon.

Convergence is super-geometric for EE, the square, and the right-angled triangles. It is slowest for the polygons with an angle of 108∘108^\circ or 120∘120^\circ (the 30–30–120 triangle, the pentagon and the hexagon), whose corners produce the strongest singularities.

The regular NN-gon values increase with NN, consistent with [4].

Observation 7.2 (Thin triangles). For the limiting endpoint profile Φ0(x)=1−x\Phi_0(x) = 1 - x in the quotient of Corollary 5.3, the Euler–Lagrange equation ((1−x)u′)′+2θu=0((1 - x)u')' + 2\theta u = 0 gives θ(0)=j1,12/8≈1.83525\theta(0) = j_{1,1}^2/8 \approx1.83525. Numerically θ(0.05)=1.93168\theta(0.05) = 1.93168, θ(0.25)=2.40957\theta(0.25) = 2.40957 and θ(12)=2.89159\theta(\frac{1}{2}) = 2.89159. The double-precision Ritz values (n=40n = 40) of Σ/(2q)\Sigma/(2q) at q=0.01q = 0.01 agree with these to 4×10−44 \times10^{-4} relative. In particular Σ(T)≈2.89 ϑ(T)\Sigma(T) \approx2.89\,\vartheta(T) for thin isosceles triangles, compared with the rigorous constant 45.6 of [8].

Remarks

Lemma 4.1 applies to any polygonal trial space for which the energy and boundary-variance forms can be enclosed. At a multiple reference eigenvalue the spectral gap controls the projection error, and the identity X=ρYX = \rho Y makes verification of a linear deficit possible up to ρ=0\rho= 0.

For n≥4n \ge4 the affine group no longer acts transitively on similarity classes of nn-gons, so one needs a richer family of transplantations, for example piecewise affine or Schwarz–Christoffel maps. Parallelograms form a two-parameter affine family of the square. General quadrilaterals have four parameters modulo similarities, so a first-order comparison of a double eigenvalue would also have to address deformations in the kernel of the linear map into symmetric 2×22 \times2 matrices.

Declaration of AI-generated research and authorship

This manuscript is the product of an AI-driven research process. During a broader investigation of open problems from the AIM problem lists, initiated by the human author, an AI system identified this problem as a candidate, found the proof, and wrote the code for the certified computations. The author then directed AI systems to package the result, formalise parts of it in Lean and prepare this manuscript. The AI systems developed the results and their presentation, including the mathematical exposition, the computations, the formalisation, and the organisation and writing of the text; the human author wrote no part of the manuscript. The system was Claude Opus 5.5 (Anthropic), used through Claude Code. OpenAI Codex subsequently revised the proof and computational certificates for stability, spectral isolation and splitting, and incorporated the thin-domain corollary.

The proof is computer-assisted and is reproduced by scripts/verify.py; the Lean development covers only the elementary steps listed in § 7.

The human author has not read this manuscript or checked its arguments. At the time of writing, the mathematics, the code and the formalisation have been checked only by AI systems and by the computer verification described above; no independent human expert has reviewed them. The author takes responsibility for the accuracy of this disclosure.

References

  1. [1]AIM Problem Lists, Steklov eigenproblems, Section 4 (Shape optimisation for Steklov), Problem 4.6 (É. Oudet), http://aimpl.org/stekloveigen/4.
  2. [2]F. Johansson, Arb: efficient arbitrary-precision midpoint-radius interval arithmetic, IEEE Trans. Comput. 66 (2017), 1281–1292. doi:10.1109/TC.2017.2690633.DOI
  3. [3]M. Bebendorf, A note on the Poincaré inequality for convex domains, Z. Anal. Anwend. 22 (2003), 751–756. doi:10.4171/ZAA/1170.DOI
  4. [4]Z. Cheng, C. Gui, Y. Hu, Q. Li, R. Yao, Global strict monotonicity and asymptotics of the first Steklov eigenvalue for regular polygons, arXiv:2603.25116v2 (2026). doi:10.48550/arXiv.2603.25116.DOI
  5. [5]M. Crouzeix, P.-A. Raviart, Conforming and nonconforming finite element methods for solving the stationary Stokes equations I, RAIRO Anal. Numér. 7 (1973), 33–75. doi:10.1051/m2an/197307R300331.DOI
  6. [6]C. Davis, W. M. Kahan, The rotation of eigenvectors by a perturbation. III, SIAM J. Numer. Anal. 7 (1970), 1–46. doi:10.1137/0707001.
  7. [7]S. Dominguez, N. Nigam, B. Shahriari, A combined finite element and Bayesian optimization framework for shape optimization in spectral geometry, Comput. Math. Appl. 74 (2017), no. 11, 2874–2896. doi:10.1016/j.camwa.2017.08.044.DOI
  8. [8]E. B. Dryden, C. Gordon, J. Moreno, J. Rowlett, C. Villegas-Blas, The Steklov spectrum of convex polygonal domains I: spectral finiteness, J. Geom. Anal. 35 (2025), no. 3, Paper No. 91. doi:10.1007/s12220-025-01922-8.DOI
  9. [9]A. Girouard, J. Lagacé, I. Polterovich, A. Savo, The Steklov spectrum of cuboids, Mathematika 65 (2019), no. 2, 272–310. doi:10.1112/S0025579318000414.DOI
  10. [10]A. Girouard, I. Polterovich, Spectral geometry of the Steklov problem, J. Spectr. Theory 7 (2017), 321–359. doi:10.4171/JST/164.DOI
  11. [11]R. S. Laugesen, B. A. Siudeja, Maximizing Neumann fundamental tones of triangles, J. Math. Phys. 50 (2009), 112903. doi:10.1063/1.3246834.DOI
  12. [12]A. Henrot and M. Michetti, A comparison between Neumann and Steklov eigenvalues, J. Spectr. Theory 12 (2022), 1405–1442, doi:10.4171/JST/429.DOI
  13. [13]N. Nigam, Open Problem 8 in Open problems from the miniconference on sharp eigenvalue estimates for partial differential operators (2020), problem list.
  14. [14]X. Liu, A framework of verified eigenvalue bounds for self-adjoint differential operators, Appl. Math. Comput. 267 (2015), 341–355. doi:10.1016/j.amc.2015.03.048.DOI
  15. [15]L. E. Payne, H. F. Weinberger, An optimal Poincaré inequality for convex domains, Arch. Rational Mech. Anal. 5 (1960), 286–292. doi:10.1007/BF00252910.DOI
  16. [16]R. Weinstock, Inequalities for a classical eigenvalue problem, J. Rational Mech. Anal. 3 (1954), 745–753. doi:10.1512/iumj.1954.3.53036.DOI
  17. [17]C. You, H. Xie, X. Liu, Guaranteed eigenvalue bounds for the Steklov eigenvalue problem, SIAM J. Numer. Anal. 57 (2019), 1395–1410. doi:10.1137/18M1189592.DOI

Paper details

vlnc:W_sw5cdart
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Anthropic Claude Code Claude Opus 5.5, OpenAI Codex GPT-6.2-Astra

View statement
Looking for related papers…
Contents