\pdfoutput=1 \documentclass[11pt]{article} \usepackage[T1]{fontenc} \usepackage{lmodern} \usepackage{amsmath,amssymb,amsthm,mathtools} \usepackage{booktabs} \input{claims} \usepackage[margin=1in]{geometry} \usepackage[colorlinks=true,urlcolor=blue,citecolor=blue]{hyperref} \newtheorem{theorem}{Theorem}[section] \newtheorem{proposition}[theorem]{Proposition} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{corollary}[theorem]{Corollary} \theoremstyle{definition} \newtheorem{definition}[theorem]{Definition} \theoremstyle{remark} \newtheorem{remark}[theorem]{Remark} \newtheorem{observation}[theorem]{Observation} \emergencystretch=3em \newcommand{\R}{\mathbb{R}} \newcommand{\C}{\mathbb{C}} \newcommand{\eps}{\varepsilon} \newcommand{\dd}{\,\mathrm{d}} \newcommand{\Lam}{\Lambda} \DeclareMathOperator{\tr}{tr} \renewcommand{\Re}{\operatorname{Re}} \renewcommand{\Im}{\operatorname{Im}} \DeclareMathOperator{\sinc}{sinc} \title{The equilateral triangle maximises the first Steklov eigenvalue among triangles} \author{Jason Davies\thanks{\href{mailto:jason@jasondavies.com}{\nolinkurl{jason@jasondavies.com}}}} \date{November 19, 2026} \begin{document} \maketitle \begin{abstract} Let $\sigma_1(T)$ be the first nonzero Steklov eigenvalue of a triangle $T$ and $L(T)$ its perimeter. Among triangles the equilateral triangle $E$ uniquely maximises $\sigma_1 L$. If the shortest and longest sides of $T$ have lengths $a\le b$, then $\sigma_1(E)L(E)-\sigma_1(T)L(T)\ge (b-a)/(2b)$. Since $|T|/L(T)^2$ is also maximal at $E$, the equilateral triangle uniquely maximises $\sigma_1\sqrt{|T|}$ as well. This settles the case $n=3$ of the Steklov analogue of the P\'olya--Szeg\H{o} conjecture for $n$-gons, for both normalisations. The extremal value satisfies $3.8724615957722\le \sigma_1(E)L(E)\le 3.8724615957734$. For side length one, $\sigma_1(E)=\sigma_2(E)<\sigma_3(E)=2\sqrt3<\sigma_4(E)$. 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 $E$ we transplant the first eigenspace of $E$ affinely; away from $E$ the trial function $\cos(\pi x)$ suffices. We verify the finite matrix and interval inequalities in ball arithmetic. \end{abstract} \noindent\textbf{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. \section{Introduction} For a bounded Lipschitz domain $\Omega\subset\R^2$, the Steklov problem $\Delta u=0$ in $\Omega$, $\partial_\nu u=\sigma u$ on $\partial\Omega$ has eigenvalues $0=\sigma_0<\sigma_1(\Omega)\le\sigma_2(\Omega)\le\cdots$. They satisfy \begin{equation}\label{eq:var} \sigma_1(\Omega)=\min_{u\in H^1(\Omega)\setminus\R}\mathcal R_\Omega(u),\qquad \mathcal R_\Omega(u)=\frac{\int_\Omega|\nabla u|^2}{\int_{\partial\Omega}u^2-\frac1{|\partial\Omega|}\bigl(\int_{\partial\Omega}u\bigr)^2}. \end{equation} The scale-invariant quantity is \[\Sigma(\Omega):=\sigma_1(\Omega)\,L(\Omega),\qquad L(\Omega)=|\partial\Omega|.\] Weinstock's inequality~\cite{Weinstock} states that $\Sigma(\Omega)\le2\pi$ for simply connected planar domains, with equality only for the disk; see the survey~\cite{GP}. For polygons with a prescribed number of sides, the natural conjecture is the analogue of P\'olya--Szeg\H{o}'s conjecture for the Dirichlet Laplacian: \medskip\noindent\textbf{Problem} (Oudet; AIM problem list \emph{Steklov eigenproblems}, Problem 4.6~\cite{AimPL}). \emph{Do regular $n$-gons maximise $\sigma_1$ among $n$-gons of given area, or of given perimeter?} \medskip Dominguez, Nigam and Shahriari~\cite{DNS} 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~\cite{Nigam}. Among rectangles of given perimeter or area, the square maximises $\sigma_1$: this is the planar case of the cuboid inequality of Girouard, Lagac\'e, Polterovich and Savo~\cite[Thm.~1.6]{GLPS}. Cheng, Gui, Hu, Li and Yao~\cite{CGHLY} state the conjecture for convex polygons with at most $N$ sides at fixed perimeter (their Conjecture~1.1), and leave it open for every $N\ge3$. They prove that $\sigma_1$ of the perimeter-normalised regular $N$-gon increases strictly with $N$, and that its first eigenspace is two-dimensional. Dryden, Gordon, Moreno, Rowlett and Villegas-Blas~\cite{DGMRV} prove $\sigma_k(T)L(T)\le\frac{8\sqrt3}{3}\pi^2k^2\vartheta(T)$ for triangles, where $\vartheta(T)$ is the smallest angle. This shows that degenerate triangles are far from optimal, but it does not compare triangles with $E$. For the Neumann problem, Laugesen and Siudeja~\cite{LS} proved that the equilateral triangle maximises the first nonzero Neumann eigenvalue times $L^2$ 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 $E$ is double, so the maximum is a cone and the comparison degenerates at $E$; and a rigorous lower bound for $\sigma_1(E)$ is needed. We are not aware of a proof of the Steklov case for any $n\ge3$. \begin{theorem}\label{thm:main} Let $E$ be an equilateral triangle. For every triangle $T$, \[\Sigma(T)\le\Sigma(E),\] with equality if and only if $T$ is equilateral. More precisely, normalise a longest side to $[0,1]$ and write the apex as $(p,q)$ with $p\le1/2$ and $q>0$. If $\rho=|(p,q)-(1/2,\sqrt3/2)|$, then \[ \Sigma(E)-\Sigma(T)\ \ge\ \frac{\rho}{2}\ \ge\ \frac{\ell_{\max}-\ell_{\min}}{2\ell_{\max}}, \] where $\ell_{\max}$ and $\ell_{\min}$ are the extreme side lengths of $T$. Moreover \[\clSigmaElo\le\Sigma(E)\le\clSigmaEhi .\] \end{theorem} Girouard and Polterovich computed the first sixty normalised eigenvalues of the equilateral triangle numerically~\cite[Example~3.2.1]{GP}. For side length one we obtain $\sigma_1(E)=\sigma_2(E)<\sigma_3(E)=2\sqrt3<\sigma_4(E)$ (Proposition~\ref{prop:eq}). The double eigenvalue agrees with~\cite[Thm.~1.2]{CGHLY} for $N=3$. \begin{corollary}\label{cor:area} For every triangle $T$, \[\sigma_1(T)\sqrt{|T|}\le\sigma_1(E)\sqrt{|E|}\in[\clareaElo,\,\clareaEhi],\] with equality if and only if $T$ is equilateral. \end{corollary} \begin{proof} $\sigma_1\sqrt{|T|}=\Sigma(T)\,(|T|/L^2)^{1/2}$. By Heron's formula and the AM--GM inequality, $|T|/L^2\le\sqrt3/36$, with equality only for equilateral triangles. Now apply Theorem~\ref{thm:main}. \end{proof} \subsection*{Method} Upper bounds for $\sigma_1(T)$ come directly from~\eqref{eq:var}; a lower bound is needed only for the single domain $E$. We parametrise triangles by the apex $(p,q)$ over the base $[0,1]$ (Lemma~\ref{lem:param}), and let $\rho$ be the distance of $(p,q)$ from the equilateral apex. \begin{enumerate} \item \emph{Near $E$} ($\rho\le\rho_1\approx0.6$). The trial space is the exact first eigenspace of $E$, transplanted by the affine map $E\to T$. The comparison involves only a few moments of the eigenspace and two exact identities. These identities let us divide out the factor $\rho$ (Lemma~\ref{lem:XrhoY}), which turns the degenerate comparison at $\rho=0$ 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 $\sigma_4(E)$. \item \emph{Away from $E$} ($\rho\ge\rho_1$). The single trial function $\cos(\pi x)$ gives $\Sigma(T)\le\clfarBound$, well below $\Sigma(E)>3.8724$. A one-dimensional check on $\clfarIntervals$ intervals suffices. \end{enumerate} \section{Preliminaries} \subsection{The Dirichlet-to-Neumann map} Let $\Omega$ be a bounded Lipschitz domain. The Dirichlet-to-Neumann map $\Lam$ is the self-adjoint operator on $L^2(\partial\Omega)$ associated with the closed form $f\mapsto\int_\Omega|\nabla\mathcal Hf|^2$ on $H^{1/2}(\partial\Omega)$, where $\mathcal Hf$ is the harmonic extension. It has compact resolvent and eigenvalues $\sigma_k$, with an orthonormal basis of eigenvectors $\chi_k$ of $L^2(\partial\Omega)$; $\chi_0$ is constant. If $u$ is a harmonic polynomial with trace $f$, Green's formula gives $f\in\operatorname{dom}\Lam$ and $\Lam f=\partial_\nu u$. For $\tilde\sigma\in\R$ and $\eta>0$ write $J=[\tilde\sigma-\eta,\tilde\sigma+\eta]$. \begin{lemma}[Inclusion]\label{lem:incl} Let $W\subset\operatorname{dom}\Lam$ have dimension $k$, and suppose $\|(\Lam-\tilde\sigma)f\|\le\eta\|f\|$ for all $f\in W$. Then $\Lam$ has at least $k$ eigenvalues, counted with multiplicity, in $J$. \end{lemma} \begin{proof} Suppose not. Then some $f\in W\setminus0$ is orthogonal to the spectral subspace of $J$, and $\|(\Lam-\tilde\sigma)f\|^2=\sum_{|\sigma_j-\tilde\sigma|>\eta}|\sigma_j-\tilde\sigma|^2|\langle f,\chi_j\rangle|^2>\eta^2\|f\|^2$, a contradiction. \end{proof} \begin{lemma}[Eigenvector enclosure; cf.~\cite{DK}]\label{lem:enclosure} Let $P$ be the spectral projection of $\Lam$ onto $J$. Suppose every eigenvalue of $\Lam$ outside $J$ is either $0$ or at least $\sigma^{\rm lo}>\tilde\sigma\ge0$, and put $\gamma=\sigma^{\rm lo}-\tilde\sigma$. Let $f\in\operatorname{dom}\Lam$ be orthogonal to the constants, and $r=(\Lam-\tilde\sigma)f$. Then \[\|(I-P)f\|_{L^2(\partial\Omega)}\le\frac{\|r\|}{\gamma},\qquad \int_\Omega\bigl|\nabla\mathcal H(I-P)f\bigr|^2\le\frac{\sigma^{\rm lo}}{\gamma^2}\|r\|^2.\] \end{lemma} \begin{proof} $(\sigma_j-\tilde\sigma)\langle f,\chi_j\rangle=\langle r,\chi_j\rangle$ for every $j$, and the component along $\chi_0$ vanishes. The remaining $j$ with $\sigma_j\notin J$ have $\sigma_j\ge\sigma^{\rm lo}$, so $\|(I-P)f\|^2\le\|r\|^2/\gamma^2$, and $\int|\nabla\mathcal H(I-P)f|^2=\sum\sigma_j|\langle r,\chi_j\rangle|^2/(\sigma_j-\tilde\sigma)^2\le\sigma^{\rm lo}\|r\|^2/\gamma^2$, because $\sigma/(\sigma-\tilde\sigma)^2$ is decreasing for $\sigma>\tilde\sigma$ when $\tilde\sigma\ge0$ (its derivative is $-(\sigma+\tilde\sigma)/(\sigma-\tilde\sigma)^3$). \end{proof} \subsection{Parametrisation} For $(p,q)$ with $q>0$, write \[T(p,q)=\operatorname{conv}\{(0,0),(1,0),(p,q)\},\] and set \[F=\{(p,q):0
0,\ (1-p)^2+q^2\le1\}.\]
Since $\Sigma$ is invariant under similarities, it suffices to treat the triangles $T(p,q)$, $(p,q)\in F$.
\begin{lemma}\label{lem:param}
Every triangle is similar to $T(p,q)$ for some $(p,q)\in F$, and $T(p,q)$ is equilateral if and only if
$(p,q)=(\frac12,\frac{\sqrt3}{2})$.
\end{lemma}
\begin{proof}
Rescale so that a longest side has length $1$. Move it to $[0,1]\times\{0\}$ with the third vertex
$(p,q)$ in the upper half-plane, and reflect in $x=\frac12$ if necessary so that $p\le\frac12$. The side
from $(1,0)$ to the apex is not longer than the base, so $(1-p)^2+q^2\le1$; this forces $p>0$. The other
side then satisfies $p^2+q^2\le(1-p)^2+q^2\le1$. The equilateral case is clear.
\end{proof}
We write $p=\frac12+\rho\cos\psi$, $q=\frac{\sqrt3}2+\rho\sin\psi$ for polar coordinates about the
equilateral apex; $\rho$ is the Euclidean distance in the $(p,q)$-plane.
\subsection{Affine transplantation}
Let $A\in GL_2^+(\R)$, $u\in H^1(\Omega)$ and $u_A=u\circ A^{-1}\in H^1(A\Omega)$. Write $G=A^{-1}A^{-T}$. For
an edge $e$ of a polygon $\Omega$ with unit tangent $\tau_e$, put $s_e=|A\tau_e|$. Changing variables gives
\begin{equation}\label{eq:transplant}
\begin{gathered}
\int_{A\Omega}|\nabla u_A|^2=\det A\sum_{i,j}G_{ij}\int_\Omega\partial_iu\,\partial_ju,\\
\int_{Ae}u_A^2=s_e\int_eu^2,\qquad \int_{Ae}u_A=s_e\int_eu,
\end{gathered}
\end{equation}
because the arc length on $Ae$ is $s_e$ times that on $e$. For a finite-dimensional space $\mathcal V$ of trial
functions, define the forms
$\mathcal N_A(u,u')=\int\nabla u_A\cdot\nabla u'_A$ and
$\mathcal D_A(u,u')=\sum_es_e\int_euu'-\frac1{L(A\Omega)}(\sum_es_e\int_eu)(\sum_es_e\int_eu')$ on $\mathcal V$.
If $\mathcal D_A$ is positive definite on $\mathcal V$, then~\eqref{eq:var} gives
$\sigma_1(A\Omega)\le\min_{\mathcal V\setminus 0}\mathcal N_A/\mathcal D_A$.
\section{The equilateral triangle}\label{sec:eq}
Let $E=T(\frac12,\frac{\sqrt3}2)$, with side $1$ and $L(E)=3$. Let $z_0$ be its centroid and
$w=-i\sqrt3\,(z-z_0)$, so that the circumradius is $1$ in $w$ and the apex lies on the positive real $w$-axis. The dihedral group $D_3$
acts by $w\mapsto e^{2\pi i/3}w$ and $w\mapsto\bar w$.
\subsection{An exactly equivariant trial pair}
Given real coefficients $c_k$ ($k\le20$, $k\not\equiv0\bmod3$, fourteen in all), put
$\mathfrak f=\sum_{k\equiv1}c_kw^k+\sum_{k\equiv2}c_k\bar w^k$, $u_1=\Re\mathfrak f$ and $u_2=\Im\mathfrak f$. Rotation by
$2\pi/3$ multiplies $\mathfrak f$ by $e^{2\pi i/3}$, and reflection conjugates it. Hence $(u_1,u_2)$ transforms
under $D_3$ exactly as the coordinates $(\Re w,\Im w)$ do, whatever the $c_k$ are. In particular
$\int_{\partial E}u_a=0$; each $u_a$ is the real part of a holomorphic polynomial $\mathfrak f_a$ (since
$\Re\bar w^k=\Re w^k$ and $\Im\bar w^k=-\Im w^k$), and by Schur's lemma all their Gram
matrices are scalar. The $c_k$ come from a floating-point Ritz computation and are then fixed as exact
binary numbers.
Let $\tilde\sigma\approx\clsigmaTilde$ be the Rayleigh quotient of $u_1$, rounded to binary64. All
boundary integrals of products of $u_a$ and $\partial_\nu u_a$ are integrals of polynomials along segments,
and we compute them exactly in ball arithmetic. This gives the relative residual
\[\eta:=\max_{f\in\operatorname{span}(u_1,u_2)\setminus0}\frac{\|\partial_\nu f-\tilde\sigma f\|_{L^2(\partial E)}}{\|f\|_{L^2(\partial E)}}\le\cletaMax.\]
By Lemma~\ref{lem:incl}, $\Lam_E$ has at least two eigenvalues in $J=[\tilde\sigma-\eta,\tilde\sigma+\eta]$.
\subsection{\texorpdfstring{A Crouzeix--Raviart lower bound for $\sigma_4(E)$}{A Crouzeix-Raviart lower bound for sigma4(E)}}
Subdivide $E$ uniformly into $n_h^2$ equilateral cells $K$ of side $h=1/n_h$. Let $\mathcal C_h$ be the
Crouzeix--Raviart space~\cite{CR}: piecewise linear functions that are continuous at the midpoints $x_e$ of
interior edges. Let $\Pi_h$ be the interpolation by edge means, $(\Pi_hu)(x_e)=|e|^{-1}\int_eu$. Write
$\mathfrak a_h(u,y)=\sum_K\int_K\nabla u\cdot\nabla y$ and $\mathfrak b(u,y)=\int_{\partial E}uy$, and let
$\sigma_{h,0}\le\sigma_{h,1}\le\cdots$ be the finite eigenvalues of the pencil $(\mathfrak a_h,\mathfrak b)$
on $\mathcal C_h$. The min--max principle holds over subspaces on which $\mathfrak b$ is positive definite.
\begin{lemma}[Trace constant]\label{lem:trace}
Let $K$ be an equilateral triangle of side $h$, $e$ an edge of $K$, and $\varphi\in H^1(K)$ with
$\int_e\varphi=0$. Then
\[\|\varphi\|^2_{L^2(e)}\le\frac{4}{\sqrt3}\Bigl(\frac1{\pi^2}+\frac1\pi\Bigr)h\,\|\nabla\varphi\|^2_{L^2(K)}.\]
\end{lemma}
\begin{proof}
Let $\hat\varphi=\varphi-\bar\varphi_K$. Since $\varphi$ has mean zero on $e$, $\|\varphi\|_e\le\|\varphi-\zeta\|_e$ for every
constant $\zeta$; in particular $\|\varphi\|_e\le\|\hat\varphi\|_e$. Let $O_e$ be the vertex opposite $e$. The
field $(x-O_e)\hat\varphi^2$ has normal component $0$ on the two edges through $O_e$, and $2|K|/|e|$ times
$\hat\varphi^2$ on $e$. Its divergence is $2\hat\varphi^2+2\hat\varphi\nabla\hat\varphi\cdot(x-O_e)$. The divergence
theorem therefore gives
\[\|\hat\varphi\|_e^2=\frac{|e|}{|K|}\Bigl(\|\hat\varphi\|_K^2+\int_K\hat\varphi\,\nabla\hat\varphi\cdot(x-O_e)\Bigr)
\le\frac{|e|}{|K|}\bigl(\|\hat\varphi\|_K^2+h\|\hat\varphi\|_K\|\nabla\varphi\|_K\bigr).\]
By the Payne--Weinberger inequality~\cite{PW,Beb} on the convex set $K$ of diameter $h$,
$\|\hat\varphi\|_K\le\frac h\pi\|\nabla\varphi\|_K$. Finally $|e|/|K|=4/(\sqrt3h)$.
\end{proof}
\begin{lemma}[Projection bound; cf.~\cite{Liu,YXL}]\label{lem:liu}
Let $n_h\ge2$ and $C_h^2=2\cdot\frac{4}{\sqrt3}(\frac1{\pi^2}+\frac1\pi)h$. Then for each $k$ below the number of
boundary edges,
\[\sigma_k(E)\ge\frac{\sigma_{h,k}}{1+C_h^2\sigma_{h,k}}.\]
\end{lemma}
\begin{proof}
(i) For every $y_h\in\mathcal C_h$, $\mathfrak a_h(u-\Pi_hu,y_h)=0$. Indeed $\nabla y_h$ is constant on each
cell $K$, and $\int_K\nabla(u-\Pi_hu)=\int_{\partial K}(u-\Pi_hu)\nu=0$ because $\Pi_hu|_K$ is linear and has
the same edge means as $u$.
(ii) Write $\varphi=u-\Pi_hu$. It has mean zero on every edge of every cell. Each boundary edge lies in one
cell, and for $n_h\ge2$ a cell has at most two boundary edges (exactly two at the corners). Lemma~\ref{lem:trace} then
gives $\mathfrak b(\varphi,\varphi)\le C_h^2\,\mathfrak a_h(\varphi,\varphi)$.
(iii) Let $\mathcal Z_k$ be the span of the eigenfunctions of $\sigma_0,\dots,\sigma_k$, and take
$u\in\mathcal Z_k$ with $\mathfrak b(u,u)=1$. Put $\upsilon=\mathfrak a_h(u-\Pi_hu,u-\Pi_hu)$. By (i),
$\mathfrak a_h(\Pi_hu,\Pi_hu)=\int|\nabla u|^2-\upsilon\le\sigma_k-\upsilon$, and by (ii),
$\mathfrak b(\Pi_hu,\Pi_hu)^{1/2}\ge1-C_h\sqrt\upsilon$. If $C_h^2\sigma_k\ge1$ the claim is trivial, since
$x/(1+C_h^2x)