betaFeedback

WorksMathematics

Minimum-Area Enclosing Rectangles and the First Neumann Eigenvalue of Planar Convex Domains

Abstract

Let ΩΩ be a bounded convex domain in the plane and let RR be a minimum-area rectangle containing Ω‾\overlineΩ. We prove that if the ratio of the shorter side of RR to its longer side is at most ρ≈0.443485ρ\approx 0.443485, then the first nonzero Neumann eigenvalue of ΩΩ is bounded from below by that of RR. Moreover, equality holds if and only if ΩΩ coincides with the interior of RR. The proof is based on a flux identity for the means of a test function over vertical sections of ΩΩ, combined with weighted estimates involving the distributional second derivatives of the boundary functions. Finally, we show that the threshold ρ≈0.443485ρ\approx0.443485 is nearly optimal by constructing, for every ratio in (0.457994,1/2)(0.457994,1/2), trapezoids that are arbitrarily close to their minimum-area enclosing rectangle and for which the inequality fails.

Introduction

Let Ω⊂R2\Omega\subset\mathbb{R}^{2} be a bounded open convex domain. We consider the Neumann eigenvalue problem

{−Δu=μuin Ω,∂νu=0on ∂Ω,(1)\begin{cases} -\Delta u = \mu u & \text{in } \Omega,\\ \partial_{\nu}u = 0 & \text{on } \partial\Omega, \tag*{(1)} \end{cases}

understood in the weak sense, where ν\nu denotes the outward unit normal to ∂Ω\partial\Omega. The first nonzero eigenvalue of (1.1) is characterized by

μ1(Ω)=inf⁡u∈H1(Ω)∖{0}{∫Ω∣∇u∣2∫Ωu2:∫Ωu(x) dx=0},(2)\mu_{1}(\Omega)=\inf_{u\in H^{1}(\Omega)\setminus\{0\}}\left\{\frac{\int_{\Omega}|\nabla u|^{2}}{\int_{\Omega}u^{2}}:\int_{\Omega}u(x)\,dx=0\right\}, \tag*{(2)}

so that 1/μ1(Ω)1/\mu_{1}(\Omega) is the optimal constant in the Poincaré inequality for functions of mean zero on Ω\Omega. Throughout the paper, all functions are real-valued.

Estimating μ1(Ω)\mu_{1}(\Omega) in terms of geometric quantities of Ω\Omega, such as its area and diameter, is a classical problem. The Szegő–Weinberger inequality [15 Wei56] asserts that among planar domains of prescribed area, the disk maximizes μ1\mu_{1}. For convex domains, Payne and Weinberger [14] proved the sharp lower bound

μ1(Ω)≥π2diam⁡(Ω)2,\mu_{1}(\Omega)\ge\frac{\pi^{2}}{\operatorname{diam}(\Omega)^{2}},

where the constant is approached by rectangles whose shorter sides tend to zero. Quantitative refinements in terms of the width have recently been obtained by Amato, Bucur and Fragalà [1] and by Wang and Xu [WX24]; see also [2] for nonlinear extensions. For various extensions to curved spaces, see [12], [ZY84], [8], [5], [3], [4], [13], [10, 11, 9], [ZW17], just to mention a few.

For a rectangle RR with side lengths L≥W>0L \ge W > 0, it is well known that μ1(int⁡R)=π2L2\mu_1(\operatorname{int} R) = \frac{\pi^2}{L^2}. This suggests looking for geometric conditions under which the diameter in the Payne–Weinberger inequality can be replaced by the longer side of a rectangle containing Ω\Omega. Inclusion alone is not enough, since Neumann eigenvalues are not monotone with respect to inclusion, even among convex domains. In this direction, Freitas and Kennedy [7] proved that for bounded convex planar domains Ω1⊂Ω2\Omega_1 \subset\Omega_2,

μ1(Ω1)≥π24j0,12μ1(Ω2),\mu_1(\Omega_1) \ge\frac{\pi^2}{4j_{0,1}^2}\mu_1(\Omega_2),

where j0,1j_{0,1} is the first positive zero of the Bessel function J0J_0 and the factor is optimal for general convex inclusions. A related obstruction comes from the work of Cavallina, Funano, Henrot, Lemenant, Lucardesi and Sakaguchi [6], who showed that for a fixed bounded open convex planar domain Ω0\Omega_0, the infimum of μ1\mu_1 over convex subdomains of Ω0\Omega_0 equals π2diam⁡(Ω0)2\frac{\pi^2}{\operatorname{diam}(\Omega_0)^2}. When Ω0\Omega_0 is a rectangle with sides LL and WW, thin rectangles approaching a diagonal of Ω0\Omega_0 therefore have eigenvalues tending to π2L2+W2\frac{\pi^2}{L^2+W^2}, which is smaller than π2L2\frac{\pi^2}{L^2}.

In view of these examples, we impose a condition on the enclosing rectangle itself. By a minimum-area enclosing rectangle of Ω\Omega we mean a rectangle, of arbitrary orientation, which contains Ω‾\overline{\Omega} and has the least area among all such rectangles. Existence follows from the continuity of the two orthogonal widths as functions of the orientation and compactness of the set of orientations. Such a rectangle need not be unique, and we simply fix one of them. Since the function f(s)=scot⁡sf(s)=s\cot s is strictly decreasing from 11 to 00 on (0,π/2)(0,\pi/2), there exists a unique ρ∈(0,1/2)\rho\in(0,1/2) satisfying

πρcot⁡(πρ)=14.(3)\pi\rho\cot(\pi\rho)=\frac{1}{4}. \tag*{(3)}

The main result of this paper states the following.

Theorem 1.1. Let Ω\Omega be a bounded open convex set in the plane, and let RR be a minimum-area enclosing rectangle of Ω\Omega, with side lengths L≥W>0L \ge W > 0. If W/L≤ρW/L \le\rho, then

μ1(Ω)≥π2L2=μ1(int⁡R).\mu_1(\Omega) \ge\frac{\pi^2}{L^2}=\mu_1(\operatorname{int} R).

Equality holds if and only if Ω=int⁡R\Omega=\operatorname{int} R.

Remark 1.2. Numerically, ρ≈0.443485\rho\approx0.443485, so Theorem 1.1 applies whenever the minimum-area enclosing rectangle is sufficiently elongated. The numerical value is not used anywhere in the proof. We only use the defining equation (1.3) and the elementary bounds in Lemma 2.2.

After a rigid motion and scaling, we may assume that

Ω‾⊂Rr:=[0,1]×[0,r],r=WL≤ρ,\overline{\Omega}\subset R_r:=[0,1]\times[0,r],\qquad r=\frac{W}{L}\le\rho,

where RrR_r is a minimum-area enclosing rectangle of Ω\Omega. Thus, proving Theorem 1.1 amounts to proving

μ1(Ω)≥π2(4)\mu_1(\Omega)\ge\pi^2 \tag*{(4)}

and characterizing the case of equality.

The threshold ρ\rho in Theorem 1.1 is close to optimal. In Appendix A, we show that there is a constant r∗≈0.457993r_\ast\approx0.457993 with the following property. For r∈(r∗,1/2)r\in(r_\ast,1/2), consider the trapezoids

Ωr,s={(x,y):0<x<1, 0<y<r(1−sx)},0<s<1.\Omega_{r,s}=\{(x,y):0<x<1,\ 0<y<r(1-sx)\},\qquad0<s<1.

The rectangle RrR_r is a minimum-area enclosing rectangle of each Ωr,s\Omega_{r,s}, and Ω‾r,s\overline{\Omega}_{r,s} converges to RrR_r in the Hausdorff distance as s→0s \to0. Nevertheless,

μ1(Ωr,s)<π2\mu_1(\Omega_{r,s}) < \pi^2

for all sufficiently small s>0s > 0. Hence, the optimal threshold in Theorem 1.1 lies in [ρ,r∗][\rho,r_*], an interval of length less than 0.015. It remains an interesting problem to determine the optimal threshold.

This paper is organized as follows. In § 2, we prove the contact separation estimate, collect some one-dimensional inequalities, and establish the bounds for ρ\rho. § 3 contains the flux identity, the transverse estimate and the shear estimate. We prove Theorem 1.1, including the characterization of the equality case, in § 4. The appendix gives a direct variational proof of the trapezoid counterexamples.

AI disclosure. The core mathematical results and proofs in this paper are the original contributions of the authors. All essential strategies regarding Theorem 1.1 were formulated by the authors. OpenAI’s ChatGPT was used solely as an assistive tool to identify the trapezoids example in Appendix A; all arguments and calculations have been rigorously verified by the authors.

Preliminaries

Contact separation

Let 0<r<10 < r < 1, and let Ω⊂(0,1)×(0,r)\Omega\subset(0,1) \times(0,r) be an open convex set whose projection onto the xx-axis is (0,1)(0,1). Since Ω\Omega is convex, we may write

Ω={(x,y):0<x<1, a(x)<y<b(x)},\Omega= \{(x,y): 0 < x < 1,\ a(x) < y < b(x)\},

and

h(x)=b(x)−a(x),c(x)=a(x)+b(x)2.(5)h(x) = b(x) - a(x), \qquad c(x) = \frac{a(x) + b(x)}{2}. \tag*{(5)}

where a(x)a(x) is convex, b(x)b(x) is concave, and h(x)h(x) is positive and concave on (0,1)(0,1). These functions extend continuously to $[0,1]andsatisfy$0≤a≤b≤rand satisfy \$0 \le a \le b \le r. Note that the heights at the endpoints may vanish and that the boundary slopes may be unbounded near x=0x = 0 and x=1x = 1. We denote by

νa=a′′,νb=−b′′,νa+νb=−h′′\nu_a = a'', \qquad\nu_b = -b'', \qquad\nu_a + \nu_b = -h''

the nonnegative Radon measures on (0,1)(0,1).

If RrR_r is a minimum-area enclosing rectangle of Ω\Omega, then each side of RrR_r intersects Ω‾\overline{\Omega}. Consequently, the bottom and top contact sets

I−(Ω)={x∈[0,1]:(x,0)∈Ω‾},I+(Ω)={x∈[0,1]:(x,r)∈Ω‾}I_-(\Omega) = \{x \in[0,1] : (x,0) \in\overline{\Omega}\}, \qquad I_+(\Omega) = \{x \in[0,1] : (x,r) \in\overline{\Omega}\}

are nonempty closed intervals, possibly reduced to points.

In the proof of Lemma 2.1 we need an elementary fact for differentiating extrema. Suppose that fθf_\theta and gg are continuous functions on Ω‾\overline{\Omega} such that fθ=f0+θg+o(θ)f_\theta= f_0 + \theta g + o(\theta) uniformly on Ω‾\overline{\Omega} as θ→0\theta\to0. Then

ddθ∣0+max⁡Ω‾fθ=max⁡{p∈Ω‾:f0(p)=max⁡Ω‾f0}g(p).\left.\frac{d}{d\theta}\right|_{0+} \max_{\overline{\Omega}} f_\theta = \max_{\{p \in\overline{\Omega}: f_0(p) = \max_{\overline{\Omega}} f_0\}} g(p).

Lemma 2.1. Let RrR_r be a minimum-area enclosing rectangle of Ω\Omega. If I−(Ω)I_-(\Omega) and I+(Ω)I_+(\Omega) are disjoint, replace xx by 1−x1-x if necessary so that

α:=max⁡I−(Ω)<β:=min⁡I+(Ω).\alpha:= \max I_-(\Omega) < \beta:= \min I_+(\Omega).

:::`

Then

β−α≤r2.\beta-\alpha\le r^2.

Moreover, a′b′≤0a'b' \le0 almost everywhere outside [α,β][\alpha,\beta]. If I−(Ω)∩I+(Ω)≠∅I_-(\Omega) \cap I_+(\Omega) \ne\varnothing, then a′b′≤0a'b' \le0 almost everywhere on (0,1)(0,1).

Proof. Suppose I−(Ω)∩I+(Ω)=∅I_-(\Omega) \cap I_+(\Omega) = \varnothing and set d=β−αd=\beta-\alpha. For θ∈R\theta\in\mathbb{R}, introduce the rotated coordinates

Xθ(x,y)=xcos⁡θ+ysin⁡θ,Yθ(x,y)=−xsin⁡θ+ycos⁡θ.X_\theta(x,y)=x\cos\theta+y\sin\theta,\qquad Y_\theta(x,y)=-x\sin\theta+y\cos\theta.

and define the widths of Ω\Omega in these two orthogonal directions by

WX(θ):=max⁡(x,y)∈Ω‾Xθ(x,y)−min⁡(x,y)∈Ω‾Xθ(x,y),W_X(\theta):=\max_{(x,y)\in\overline{\Omega}}X_\theta(x,y)-\min_{(x,y)\in\overline{\Omega}}X_\theta(x,y),
WY(θ):=max⁡(x,y)∈Ω‾Yθ(x,y)−min⁡(x,y)∈Ω‾Yθ(x,y).W_Y(\theta):=\max_{(x,y)\in\overline{\Omega}}Y_\theta(x,y)-\min_{(x,y)\in\overline{\Omega}}Y_\theta(x,y).

Thus, WX(0)=1W_X(0)=1 and WY(0)=rW_Y(0)=r. Their product W(θ):=WX(θ)WY(θ)W(\theta):=W_X(\theta)W_Y(\theta) is the area of the smallest enclosing rectangle whose sides are parallel to the rotated coordinate axes. The minimum-area condition gives that for every θ∈R\theta\in\mathbb{R},

W(θ)≥W(0)=r.W(\theta)\ge W(0)=r.

We now compute the right derivatives at θ=0\theta=0. Note that uniformly on Ω‾\overline{\Omega},

Xθ=x+θy+O(θ2),Yθ=y−θx+O(θ2).X_\theta=x+\theta y+O(\theta^2),\qquad Y_\theta=y-\theta x+O(\theta^2).

The maximizers of Y0Y_0 are the top contacts and its minimizers are the bottom contacts. Hence,

ddθ∣0+max⁡Ω‾Yθ=−min⁡I+(Ω)=−β,\left.\frac{d}{d\theta}\right|_{0+}\max_{\overline{\Omega}}Y_\theta=-\min I_+(\Omega)=-\beta,
ddθ∣0+min⁡Ω‾Yθ=−max⁡I−(Ω)=−α.\left.\frac{d}{d\theta}\right|_{0+}\min_{\overline{\Omega}}Y_\theta=-\max I_-(\Omega)=-\alpha.

Therefore, WY′(0+)=α−β=−dW_Y'(0+)=\alpha-\beta=-d. Similarly, we have

WX′(0+)=max⁡{y:(1,y)∈Ω‾}−min⁡{y:(0,y)∈Ω‾}≤r.W_X'(0+)=\max\{y:(1,y)\in\overline{\Omega}\}-\min\{y:(0,y)\in\overline{\Omega}\}\le r.

Differentiating the area product yields

0≤W′(0+)=WX′(0+)WY(0)+WX(0)WY′(0+)=rWX′(0+)−d≤r2−d.\begin{aligned} 0\le W'(0+)=W_X'(0+)W_Y(0)+W_X(0)W_Y'(0+) \\ &=rW_X'(0+)-d\le r^2-d. \end{aligned}

Thus, we have β−α=d≤r2\beta-\alpha=d\le r^2.

It remains to prove the sign conditions. Since aa attains its minimum at α\alpha and bb attains its maximum at β\beta, convexity forces a′≤0,b′≥0a'\le0,b'\ge0 on [0,α][0,\alpha], and a′≥0,b′≤0a'\ge0,b'\le0 on [β,1][\beta,1]. Thus a′b′≤0a'b'\le0 almost everywhere outside [α,β][\alpha,\beta]. If I−(Ω)∩I+(Ω)≠∅I_-(\Omega)\cap I_+(\Omega)\ne\varnothing, we choose x0∈I−(Ω)∩I+(Ω)x_0\in I_-(\Omega)\cap I_+(\Omega). Then the same sign conditions hold to the left and to the right of x0x_0 respectively, and give a′b′≤0a'b'\le0 almost everywhere on (0,1)(0,1). This argument also covers the cases x0=0x_0=0 and x0=1x_0=1. □

One-dimensional inequalities

Let ϕ(x)=sin⁡(πx)\phi(x)=\sin(\pi x). For any z∈H01(0,1)z\in H_0^1(0,1), a direct calculation gives

∣z′∣2−π2z2=ϕ2∣(zϕ)′∣2+(ϕ′ϕz2)′.|z'|^2-\pi^2z^2=\phi^2\left|\left(\frac{z}{\phi}\right)'\right|^2+\left(\frac{\phi'}{\phi}z^2\right)'.

and integrating above identity over (0,1)(0,1) yields

∫01∣z′∣2 dx−π2∫01z2 dx=∫01ϕ2∣(zϕ)′∣2 dx(6)\int_0^1 |z'|^2\,dx-\pi^2\int_0^1 z^2\,dx=\int_0^1\phi^2\left|\left(\frac{z}{\phi}\right)'\right|^2\,dx \tag*{(6)}

We will use the above identity frequently in the following, and also use the following Poincaré inequality for functions of mean zero. If II is an interval of length tt and v∈H1(I)v\in H^1(I) has mean zero on II, then

∫Iv2≤t2π2∫I∣v′∣2.(7)\int_I v^2\leq\frac{t^2}{\pi^2}\int_I |v'|^2. \tag*{(7)}

For z∈Cc∞(0,1)z\in C_c^\infty(0,1), integration by parts and the Cauchy–Schwarz inequality give

∫01z2x2 dx=2∫01zz′x dx≤2(∫01z2x2 dx)1/2(∫01∣z′∣2 dx)1/2.\int_0^1\frac{z^2}{x^2}\,dx=2\int_0^1\frac{zz'}{x}\,dx\leq2\left(\int_0^1\frac{z^2}{x^2}\,dx\right)^{1/2}\left(\int_0^1|z'|^2\,dx\right)^{1/2}.

By density, the resulting bound extends to H01(0,1)H_0^1(0,1). Adding its reflection about x=1/2x=1/2 gives the following Hardy’s inequality

∫01z2min⁡{x,1−x}2 dx≤8∫01∣z′∣2 dx,z∈H01(0,1).(8)\int_0^1\frac{z^2}{\min\{x,1-x\}^2}\,dx\leq8\int_0^1|z'|^2\,dx,\qquad z\in H_0^1(0,1). \tag*{(8)}

Bounds for the constant ρ\rho

Lemma 2.2. The constant ρ\rho defined by (1.3) satisfies

ρ2<15,\rho^2<\frac{1}{5},

and

κρ:=5πρ7tan⁡πρ2<1.\kappa_\rho:=\frac{5\pi\rho}{7}\tan\frac{\pi\rho}{2}<1.

Proof. Set γ=πρ\gamma=\pi\rho. Then 0<γ<π/20<\gamma<\pi/2 and

γcot⁡γ=14.\gamma\cot\gamma=\frac{1}{4}.

We first prove ρ2<1/5\rho^2<1/5. Since f(s)=scot⁡sf(s)=s\cot s is strictly decreasing on (0,π/2)(0,\pi/2) and f(π/3)=π/(33)>1/4f(\pi/3)=\pi/(3\sqrt{3})>1/4, we obtain γ>π/3\gamma>\pi/3. For 0<x<10<x<1, we have the elementary inequality

tan⁡x≤x1−x2.\tan x\leq\frac{x}{1-x^2}.

Let x=π/2−γ<1x=\pi/2-\gamma<1. Then

14γ≤π/2−γ1−(π/2−γ)2.\frac{1}{4\gamma}\leq\frac{\pi/2-\gamma}{1-(\pi/2-\gamma)^2}.

Therefore, we have

3γ2−πγ+1−π24≤0.3\gamma^2-\pi\gamma+1-\frac{\pi^2}{4}\leq0.

Since ρ=γ/π\rho=\gamma/\pi and 1/π2>1/101/\pi^2>1/10, we obtain

3ρ2−ρ−320<0,3\rho^2-\rho-\frac{3}{20}<0,

Thus,

ρ2<(5+7030)2<15.(9)\rho^2 < \left(\frac{5+\sqrt{70}}{30}\right)^2 < \frac{1}{5}. \tag*{(9)}

By the half-angle formula and tan⁡γ=4γ\tan\gamma= 4\gamma, we have

tan⁡γ2=1+16γ2−14γ.\tan\frac{\gamma}{2} = \frac{\sqrt{1+16\gamma^2}-1}{4\gamma}.

Thus,

5πρ7tan⁡πρ2=5γ7tan⁡γ2=528(1+16γ2−1).\frac{5\pi\rho}{7}\tan\frac{\pi\rho}{2} = \frac{5\gamma}{7}\tan\frac{\gamma}{2} = \frac{5}{28}\left(\sqrt{1+16\gamma^2}-1\right).

Consequently,

κρ=5πρ7tan⁡πρ2<528(33−1)<1,\kappa_\rho= \frac{5\pi\rho}{7}\tan\frac{\pi\rho}{2} < \frac{5}{28}\left(\sqrt{33}-1\right)<1,

where we used γ2=π2ρ2<2\gamma^2=\pi^2\rho^2<2.

Flux inequalities

Throughout this section we use the notation of Section 2, with 0<r≤ρ0<r\leq\rho. No smoothness of the boundary is assumed. The distributional second derivatives of the boundary functions enter only through the measures νa\nu_a and νb\nu_b.

The flux identity

3.1. The flux identity. Fix u∈H1(Ω)u\in H^1(\Omega) with ∫Ωu dx=0\int_\Omega u\,dx=0. We decompose uu into its sectional mean mm and a fluctuation ww which has mean zero on each vertical section, that is,

m(x)=1h(x)∫a(x)b(x)u(x,y) dy,w(x,y)=u(x,y)−m(x).m(x)=\frac{1}{h(x)}\int_{a(x)}^{b(x)}u(x,y)\,dy,\qquad w(x,y)=u(x,y)-m(x).

where h(x)h(x) is defined in (2.1). For almost every xx, w∈H1(a(x),b(x))w\in H^1(a(x),b(x)) and satisfies wy=uyw_y=u_y on that section. Let MM and QQ denote the contributions of these two components to the squared L2(Ω)L^2(\Omega) norm, namely

M=∫01hm2,Q=∫Ωw2.M=\int_0^1 hm^2,\qquad Q=\int_\Omega w^2.

Since ww is orthogonal to constants on each section, we have

∫Ωu2=M+Q.\int_\Omega u^2=M+Q.

We define the sectional flux FF, its normalization zz, and the squared L2(0,1)L^2(0,1) norm ZZ of zz by

F(x)=−∫0xh(t)m(t) dt,z(x)=F(x)h(x),Z=∫01z2.(10)F(x)=-\int_0^x h(t)m(t)\,dt,\qquad z(x)=\frac{F(x)}{\sqrt{h(x)}},\qquad Z=\int_0^1 z^2. \tag*{(10)}

Since uu has mean zero, F(0)=F(1)=0F(0)=F(1)=0. The normalization by h\sqrt{h} is chosen so that the weight hh disappears from the leading derivative term in the identity below. To measure how far zz is from being a multiple of ϕ(x)=sin⁡(πx)\phi(x)=\sin(\pi x), we put

E=∫01ϕ2∣(zϕ)′∣2 dx.(11)E=\int_0^1\phi^2\left|\left(\frac{z}{\phi}\right)'\right|^2\,dx. \tag*{(11)}

We also put

Va=12∫(0,1)z2h dνa,Vb=12∫(0,1)z2h dνb,V=Va+Vb.(12)V_a=\frac{1}{2}\int_{(0,1)}\frac{z^2}{h}\,d\nu_a,\qquad V_b=\frac{1}{2}\int_{(0,1)}\frac{z^2}{h}\,d\nu_b,\qquad V=V_a+V_b. \tag*{(12)}

where VaV_a and VbV_b are the weighted curvature contributions of the lower and upper boundaries, respectively. The curvature integrals are taken in the sense of measures on (0,1)(0,1), so that integration by parts on compact subintervals remains valid when the boundary functions are not smooth.

Lemma 3.1. The normalized flux zz defined in (10) belongs to H01(0,1)H^1_0(0,1). All the quantities in (11) and (12) are finite, and

M=π2Z+E+V+34∫01(h′h)2z2.(13)M = \pi^2 Z + E + V + \frac{3}{4} \int_0^1 \left(\frac{h'}{h}\right)^2 z^2. \tag*{(13)}

Proof. On compact subintervals of (0,1)(0,1), the identity F′=−hmF'=-hm can be rewritten as

h m=−z′−h′2hz.\sqrt{h}\,m=-z'-\frac{h'}{2h}z.

We choose truncation points ε\varepsilon and 1−ε1-\varepsilon at which h′h' has no jump. Using the above identity and integrating by parts, we obtain

∫ε1−εhm2=∫ε1−ε∣z′∣2+34∫ε1−ε(h′h)2z2+12∫(ε,1−ε)z2h d(−h′′)+[h′2hz2]ε1−ε.(14)\int_{\varepsilon}^{1-\varepsilon} hm^2 = \int_{\varepsilon}^{1-\varepsilon} |z'|^2 +\frac{3}{4}\int_{\varepsilon}^{1-\varepsilon}\left(\frac{h'}{h}\right)^2z^2 +\frac{1}{2}\int_{(\varepsilon,1-\varepsilon)}\frac{z^2}{h}\,d(-h'') +\left[\frac{h'}{2h}z^2\right]_{\varepsilon}^{1-\varepsilon}. \tag*{(14)}

We now show that the boundary term tends to zero. For 0≤t≤x≤1/20\leq t\leq x\leq1/2, the concavity of hh gives h(t)≤(1−t)h(x)/(1−x)≤2h(x)h(t)\leq(1-t)h(x)/(1-x)\leq2h(x). Hence, by Cauchy–Schwarz inequality,

z(x)2≤1h(x)∫0xh∫0xhm2≤2x∫0xhm2=o(x).z(x)^2\leq\frac{1}{h(x)}\int_0^x h\int_0^x hm^2\leq2x\int_0^x hm^2=o(x).

Similarly, z(x)2=o(1−x)z(x)^2=o(1-x) as x→1x\to1. On the other hand, the concavity of hh also gives

−11−x≤h′h≤1x,-\frac{1}{1-x}\leq\frac{h'}{h}\leq\frac{1}{x},

so the boundary term in (14) tends to zero. All the other terms on the right-hand side of (14) are nonnegative. Letting the truncated intervals exhaust (0,1)(0,1), we see that these terms are finite and that z∈H01(0,1)z\in H^1_0(0,1). The identity (13) then follows from (6). □\square

The transverse estimate

We now estimate the transverse part of the energy. For 0<t<10<t<1, let

A(t)=2tan⁡(πt/2)π,B(t)=1π2t2−cot⁡(πt/2)2πt.(15)A(t)=\frac{2\tan(\pi t/2)}{\pi},\qquad B(t)=\frac{1}{\pi^2t^2}-\frac{\cot(\pi t/2)}{2\pi t}. \tag*{(15)}

Both functions extend continuously to t=0t=0, with values 11 and 1/121/12, respectively. For the functions in (5), we set

T=∫01z2(A(h)c′2+B(h)h′2).T=\int_0^1 z^2\left(A(h)c'^2+B(h)h'^2\right).

Lemma 3.2. The quantity TT is finite, and for every u∈H1(Ω)u\in H^1(\Omega) with ∫Ωu=0\int_\Omega u=0,

M2≤(Z+T)(∫Ω∣∇u∣2−π2Q).(16)M^2\leq(Z+T)\left(\int_\Omega|\nabla u|^2-\pi^2 Q\right). \tag*{(16)}

In particular, if M≥π2(Z+T)M\geq\pi^2(Z+T) for every mean-zero u∈H1(Ω)u\in H^1(\Omega), then μ1(Ω)≥π2\mu_1(\Omega)\geq\pi^2.

Proof. For 0<t<10<t<1, let

It=(−t/2,t/2),Ht={v∈H1(It):∫Itv=0}.I_t=(-t/2,t/2),\qquad\mathcal{H}_t=\left\{v\in H^1(I_t):\int_{I_t}v=0\right\}.

and define the symmetric bilinear form

qt(v1,v2)=∫It(v1′v2′−π2v1v2),qt(v)=qt(v,v).q_t(v_1,v_2)=\int_{I_t}(v_1'v_2'-\pi^2v_1v_2),\qquad q_t(v)=q_t(v,v).

By (7),

qt(v)≥(1−t2)∫It∣v′∣2,v∈Ht.q_t(v)\ge(1-t^2)\int_{I_t}|v'|^2,\qquad v\in\mathcal{H}_t.

Thus, qtq_t is positive definite on Ht\mathcal{H}_t. Consider the odd and even functions

ψo(s)=sin⁡(πs)πcos⁡(πt/2),\psi_o(s)=\frac{\sin(\pi s)}{\pi\cos(\pi t/2)},

and

ψe(s)=−cos⁡(πs)2πsin⁡(πt/2)+1π2t.\psi_e(s)=-\frac{\cos(\pi s)}{2\pi\sin(\pi t/2)}+\frac{1}{\pi^2t}.

Both functions belong to Ht\mathcal{H}_t, and direct differentiation gives

−ψo′′−π2ψo=0,-\psi_o''-\pi^2\psi_o=0,

and

−ψe′′−π2ψe=−1t,-\psi_e''-\pi^2\psi_e=-\frac{1}{t},

together with the endpoint conditions

ψo′(±t/2)=1,ψe′(±t/2)=±12.\psi_o'(\pm t/2)=1,\qquad\psi_e'(\pm t/2)=\pm\frac{1}{2}.

Integrating by parts, we obtain that for every v∈Htv\in\mathcal{H}_t,

qt(ψo,v)=∫Itv′,qt(ψe,v)=∫Itstv′.q_t(\psi_o,v)=\int_{I_t}v', \qquad q_t(\psi_e,v)=\int_{I_t}\frac{s}{t}v'.

By parity, qt(ψo,ψe)=0q_t(\psi_o,\psi_e)=0, while integration by parts yields

qt(ψo)=ψo(t/2)−ψo(−t/2)=2πtan⁡πt2=tA(t),q_t(\psi_o)=\psi_o(t/2)-\psi_o(-t/2)=\frac{2}{\pi}\tan\frac{\pi t}{2}=tA(t),

and

qt(ψe)=ψe(t/2)=1π2t−cot⁡(πt/2)2π=tB(t).q_t(\psi_e)=\psi_e(t/2)=\frac{1}{\pi^2t}-\frac{\cot(\pi t/2)}{2\pi}=tB(t).

Applying the Cauchy–Schwarz inequality for qtq_t to qt(ξ1ψo+ξ2ψe,v)q_t(\xi_1\psi_o+\xi_2\psi_e,v), we obtain

∣∫It(ξ1+ξ2ts)v′(s) ds∣2≤t{A(t)ξ12+B(t)ξ22}qt(v),ξ1,ξ2∈R.(17)\left|\int_{I_t}\left(\xi_1+\frac{\xi_2}{t}s\right)v'(s)\,ds\right|^2\le t\{A(t)\xi_1^2+B(t)\xi_2^2\}q_t(v),\qquad\xi_1,\xi_2\in\mathbb{R}. \tag*{(17)}

Since aa is convex, bb is concave and both take values in [0,r][0,r], we have

∣a′(x)∣+∣b′(x)∣≤2rmin⁡{x,1−x}(18)|a'(x)|+|b'(x)|\le\frac{2r}{\min\{x,1-x\}} \tag*{(18)}

for almost every x∈(0,1)x\in(0,1). The coefficients AA and BB extend continuously to t=0t=0 and are therefore bounded on [0,r][0,r]. Hence, Hardy’s inequality (8) and Lemma 3.1 give

T≤Cr∫01z2min⁡{x,1−x}2 dx≤Cr∫01∣z′∣2 dx≤CrM,(19)T\le C_r\int_0^1\frac{z^2}{\min\{x,1-x\}^2}\,dx\le C_r\int_0^1|z'|^2\,dx\le C_rM, \tag*{(19)}

where Cr>0C_r>0 denotes a constant depending only on rr, whose value may change from one occurrence to the next. In particular, TT is finite.

Assume first that uu is smooth in a neighborhood of Ω‾\overline{\Omega}. Parametrizing each vertical section by y=c(x)+h(x)sy=c(x)+h(x)s, we have

m(x)=∫−1/21/2u(x,c(x)+h(x)s) ds,m(x)=\int_{-1/2}^{1/2}u(x,c(x)+h(x)s)\,ds,

and differentiation gives that for almost every x∈(0,1)x\in(0,1),

m′(x)=1h∫ab[ux+(c′+h′h(y−c))uy] dy.m'(x)=\frac{1}{h}\int_a^b\left[u_x+\left(c'+\frac{h'}{h}(y-c)\right)u_y\right]\,dy.

Since F′=−hmF'=-hm, integration by parts yields

∫ε1−εhm2=∫ε1−εFm′−[Fm]ε1−ε.(20)\int_{\varepsilon}^{1-\varepsilon}hm^2=\int_{\varepsilon}^{1-\varepsilon}Fm'-[Fm]_{\varepsilon}^{1-\varepsilon}. \tag*{(20)}

Since mm is bounded and F(0)=F(1)=0F(0)=F(1)=0, the boundary term vanishes as ε→0\varepsilon\to0. Moreover, the bounds used in (3.10) also imply

∫Ω(Fh)2[1+(c′+h′h(y−c))2]=∫01z2(1+c′2+h′212) dx<∞.\int_{\Omega}\left(\frac{F}{h}\right)^2\left[1+\left(c'+\frac{h'}{h}(y-c)\right)^2\right]=\int_0^1z^2\left(1+c'^2+\frac{h'^2}{12}\right)\,dx<\infty.

Therefore, we can choose ε→0\varepsilon\to0 in (3.11). Since wy=uyw_y=u_y, we obtain

M=∫01Fm′=∫ΩFh[ux+(c′+h′h(y−c))wy].(21)M=\int_0^1Fm'=\int_{\Omega}\frac{F}{h}\left[u_x+\left(c'+\frac{h'}{h}(y-c)\right)w_y\right]. \tag*{(21)}

Since ∫Ω(F/h)2=Z\int_{\Omega}(F/h)^2=Z, the first term is bounded in absolute value by Z∥ux∥2\sqrt{Z}\left\lVert u_x\right\rVert_2. For the second term, we apply (3.8) for almost every xx with

t=h(x),v(s)=w(x,c(x)+s),ξ1=c′(x),ξ2=h′(x).t=h(x),\qquad v(s)=w(x,c(x)+s),\qquad\xi_1=c'(x),\qquad\xi_2=h'(x).

By Cauchy–Schwarz inequality, we have

∣∫ΩFh[(c′+h′h(y−c))wy]∣≤T(∫Ω(wy2−π2w2))1/2.\left|\int_{\Omega}\frac{F}{h}\left[\left(c'+\frac{h'}{h}(y-c)\right)w_y\right]\right|\le\sqrt{T}\left(\int_{\Omega}(w_y^2-\pi^2w^2)\right)^{1/2}.

Consequently, (3.12) gives

M≤Z(∫Ωux2)1/2+T(∫Ω(wy2−π2w2))1/2.M\le\sqrt{Z}\left(\int_{\Omega}u_x^2\right)^{1/2}+\sqrt{T}\left(\int_{\Omega}(w_y^2-\pi^2w^2)\right)^{1/2}.

Applying the Cauchy–Schwarz inequality once more to these two terms and using wy=uyw_y=u_y, we obtain (3.7) for every mean-zero function uu which is smooth in a neighborhood of Ω‾\overline{\Omega}.

For a general u∈H1(Ω)u\in H^1(\Omega) with mean zero, we choose smooth mean-zero functions unu_n converging to uu in H1(Ω)H^1(\Omega), and we denote the corresponding quantities by mnm_n, znz_n, MnM_n, QnQ_n, ZnZ_n and TnT_n. Then

∫01h∣mn−m∣2≤∫Ω∣un−u∣2.\int_0^1h|m_n-m|^2\le\int_{\Omega}|u_n-u|^2.

Applying Lemma 3.1 and (3.10) to un−uu_n-u, we get

∫01∣zn′−z′∣2 dx+∫01∣zn−z∣2{A(h)c′2+B(h)h′2} dx≤Cr∫Ω∣un−u∣2.\int_0^1|z_n'-z'|^2\,dx+\int_0^1|z_n-z|^2\{A(h)c'^2+B(h)h'^2\}\,dx\le C_r\int_{\Omega}|u_n-u|^2.

It follows that Mn→MM_n\to M, Qn→QQ_n\to Q, Zn→ZZ_n\to Z and Tn→TT_n\to T. Therefore, (3.7) passes to the limit.

Finally, suppose that M≥π2(Z+T)M \ge\pi^2(Z+T) for every mean-zero test function, and fix a nonzero function u∈H1(Ω)u \in H^1(\Omega) with mean zero. If M>0M>0, then (3.7) implies

M2≤Mπ2(∫Ω∣∇u∣2−π2Q),M^2 \le\frac{M}{\pi^2}\left(\int_{\Omega}|\nabla u|^2-\pi^2 Q\right),

and hence

∫Ω∣∇u∣2≥π2(M+Q)=π2∫Ωu2.\int_{\Omega}|\nabla u|^2 \ge\pi^2(M+Q)=\pi^2\int_{\Omega}u^2.

If M=0M=0, then uu has mean zero on almost every vertical section. Since h≤r<1h\le r<1, applying (2.3) on each section gives

∫Ω∣∇u∣2≥π2r2∫Ωu2>π2∫Ωu2.(22)\int_{\Omega}|\nabla u|^2 \ge\frac{\pi^2}{r^2}\int_{\Omega}u^2>\pi^2\int_{\Omega}u^2. \tag*{(22)}

In both cases, the variational characterization (1.2) proves the last assertion. □

The shear estimate

For the rest of this section, RrR_r is a minimum-area enclosing rectangle of Ω\Omega. The quantity to be controlled is

S=∫01(a′b′)2 dx.S=\int_0^1(a'b')^2\,dx.

It is finite by the slope bound in the proof of Lemma 3.2 and Hardy’s inequality. Recall that EE, VaV_a and VbV_b are finite by Lemma 3.1.

The following argument uses the product of the two gaps between the boundary functions and the horizontal sides of the rectangle. This product vanishes at the contacts, and its second derivative relates the shear directly to the boundary measures.

Lemma 3.3. Let RrR_r be a minimum-area enclosing rectangle of Ω\Omega. Then

S≤514r2(E+V).(23)S\le\frac{5}{14}r^2(E+V). \tag*{(23)}

Proof. If the contact intervals intersect, then S=0S=0 by Lemma 2.1 and (3.14) holds. Otherwise, after replacing xx by 1−x1-x if necessary, we may assume as in Lemma 2.1 that

α=max⁡I−(Ω),β=min⁡I+(Ω),0≤α<β≤1,d=β−α.\alpha=\max I_-(\Omega),\qquad\beta=\min I_+(\Omega),\qquad0\le\alpha<\beta\le1,\qquad d=\beta-\alpha.

By Lemma 2.1 and 2.2, we have d<1/5d<1/5. On the interval J=(α,β)J=(\alpha,\beta), put

p=a′b′,g=a(r−b),ℓ=(r−b)a′+ab′,p=a'b',\qquad g=a(r-b),\qquad\ell=(r-b)a'+ab',

where gg is the product of the distances from the lower and upper boundaries to the bottom and top sides of RrR_r, and it vanishes at both contacts. Since aa is convex with minimum a(α)=0a(\alpha)=0 and bb is concave with maximum b(β)=rb(\beta)=r, we have a′≥0a'\ge0 and b′≥0b'\ge0 almost everywhere on JJ. This gives

S=∫Jpz2 dx.S=\int_J pz^2\,dx.

Convexity and the supporting-line inequalities yield the following inequalities almost everywhere on JJ.

0≤a≤(x−α)a′,a+(1−x)a′≤r,0≤r−b≤(β−x)b′,r−b+xb′≤r.(24)\begin{aligned} 0\le a&\le(x-\alpha)a',&\qquad a+(1-x)a'&\le r,\\ 0\le r-b&\le(\beta-x)b',&\qquad r-b+xb'&\le r. \tag*{(24)} \end{aligned}

Consequently,

0≤g≤d24p,0≤ℓ≤dp.0\le g\le\frac{d^2}{4}p,\qquad0\le\ell\le dp.

A direct computation gives

(a+(1−x)a′)(r−b+xb′)−ℓ=(xa′−a)((1−x)b′−(r−b))≥0.\left(a+(1-x)a'\right)\left(r-b+xb'\right)-\ell=\left(xa'-a\right)\left((1-x)b'-(r-b)\right)\ge0.

Together with (3.15), this yields

ℓ≤r2,ℓ2≤r2dp.(25)\ell\le r^2,\qquad\ell^2\le r^2dp. \tag*{(25)}

In particular, ∣g′∣≤ℓ≤r2|g'|\le\ell\le r^2. Since gg vanishes at α\alpha and β\beta, its extension by zero outside JJ is Lipschitz.

On JJ, the product rule for distributions gives

g′′=(r−b)νa+aνb−2p dx.g''=(r-b)\nu_a+a\nu_b-2p\,dx.

Note that

a′(α+)≤νa{α}(α>0),b′(β−)≤νb{β}(β<1).a'(\alpha+)\le\nu_a\{\alpha\}\quad(\alpha>0),\qquad b'(\beta-)\le\nu_b\{\beta\}\quad(\beta<1).

Therefore, multiplying by z2z^2 and integrating by parts, we obtain

S≤12∫[α,β]∩(0,1)(r−b)z2 dνa+12∫[α,β]∩(0,1)az2 dνb+∫Jg′z′ dx.S\le\frac{1}{2}\int_{[\alpha,\beta]\cap(0,1)}(r-b)z^2\,d\nu_a+\frac{1}{2}\int_{[\alpha,\beta]\cap(0,1)}az^2\,d\nu_b+\int_J g'z'\,dx.

These computations remain valid when α\alpha or β\beta is an endpoint of [0,1][0,1]. Indeed, g′g' is bounded and zz has zero trace, so the boundary term g′z2g'z^2 of the truncated integrals tends to zero.

Since

(r−b)h≤(r−b)b≤r24,ah≤a(r−a)≤r24,(r-b)h\le(r-b)b\le\frac{r^2}{4},\qquad ah\le a(r-a)\le\frac{r^2}{4},

we obtain

S≤r24V+∫Jg′z′ dx.(26)S\le\frac{r^2}{4}V+\int_J g'z'\,dx. \tag*{(26)}

It remains to estimate the last integral in (3.17). Since ϕ=sin⁡(πx)\phi=\sin(\pi x) is positive and concave on (0,1)(0,1), we have

−11−x≤ϕ′ϕ≤1x.-\frac{1}{1-x}\le\frac{\phi'}{\phi}\le\frac{1}{x}.

Combining these inequalities with (3.15), we obtain

ϕ′ϕa≤a′,−ϕ′ϕ(r−b)≤b′.\frac{\phi'}{\phi}a\le a',\qquad-\frac{\phi'}{\phi}(r-b)\le b'.

Hence,

∣g′−2ϕ′ϕg∣=∣(r−b)(a′−ϕ′ϕa)−a(b′+ϕ′ϕ(r−b))∣≤ℓ≤rdp.(27)\left|g'-2\frac{\phi'}{\phi}g\right| =\left|(r-b)\left(a'-\frac{\phi'}{\phi}a\right)-a\left(b'+\frac{\phi'}{\phi}(r-b)\right)\right| \le\ell\le r\sqrt{dp}. \tag*{(27)}

Write v=z/ϕv=z/\phi. Since gg vanishes at α,β\alpha,\beta, integration by parts gives

∫Jg′z′ dx=−∫Jg(ϕϕ′)′v2 dx+∫J(g′−2ϕ′ϕg)zϕv′ dx.(28)\int_J g'z'\,dx=-\int_J g(\phi\phi')'v^2\,dx+\int_J\left(g'-2\frac{\phi'}{\phi}g\right)z\phi v'\,dx. \tag*{(28)}

If α=0\alpha=0 or β=1\beta=1, the additional boundary term g(ϕ′/ϕ)z2g(\phi'/\phi)z^2 also vanishes there. Indeed, the Lipschitz bound on gg cancels the endpoint singularity of ϕ′/ϕ\phi'/\phi, and zz has zero trace. The remaining integrals near the endpoints are finite by Hardy’s inequality.

Since 0≤g≤d24p0\le g\le\frac{d^2}{4}p and (ϕϕ′)′=ϕ′2−π2ϕ2≥−π2ϕ2(\phi\phi')'=\phi'^2-\pi^2\phi^2\ge-\pi^2\phi^2, we have

−∫Jg(ϕϕ′)′v2 dx≤π2∫Jgz2 dx≤π2d24S.-\int_J g(\phi\phi')'v^2\,dx\le\pi^2\int_J gz^2\,dx\le\frac{\pi^2d^2}{4}S.

By (3.18) and Cauchy–Schwarz inequality, we obtain

∫J(g′−2ϕ′ϕg)zϕv′ dx≤rd(∫01pz2 dx)1/2(∫01ϕ2∣v′∣2 dx)1/2=rdSE.\int_J \left(g' - 2\frac{\phi'}{\phi}g\right)z\phi v'\,dx \le r\sqrt{d}\left(\int_0^1 pz^2\,dx\right)^{1/2}\left(\int_0^1 \phi^2|v'|^2\,dx\right)^{1/2}=r\sqrt{dSE}.

Combining above inequalities with Young’s inequality, (3.19) gives

∫Jg′zz′ dx≤π2d24S+dS+r24E.\int_J g' z z'\,dx \le\frac{\pi^2d^2}{4}S+dS+\frac{r^2}{4}E.

Substitution into (3.17) yields

(1−d−π2d24)S≤r24(E+V).(29)\left(1-d-\frac{\pi^2d^2}{4}\right)S\le\frac{r^2}{4}(E+V). \tag*{(29)}

Since d<1/5d<1/5, we have

1−d−π2d24>710,1-d-\frac{\pi^2d^2}{4}>\frac{7}{10},

and (3.14) follows. □

Proof of the main theorem

In this section, we combine the results of Sections 2 and 3 to prove Theorem 1.1. We establish the remainder inequality (4.2) directly on Ω\Omega and deduce the lower bound, and then we use the vanishing of the nonnegative remainder terms to characterize equality.

Proof of Theorem 1.1. We normalize the minimum-area enclosing rectangle to Rr=[0,1]×[0,r]R_r=[0,1]\times[0,r], where 0<r≤ρ0<r\le\rho. We fix a mean-zero function u∈H1(Ω)u\in H^1(\Omega) and use the notation of Section 3. The identities

c2=h24+a′b′,A(t)4+B(t)=1−πtcot⁡(πt)π2t2c^2=\frac{h^2}{4}+a'b',\qquad\frac{\mathcal{A}(t)}{4}+\mathcal{B}(t)=\frac{1-\pi t\cot(\pi t)}{\pi^2t^2}

give

π2T=∫01(1−πhcot⁡(πh))(h′h)2z2+π2∫01A(h)a′b′z2.\pi^2T=\int_0^1\left(1-\pi h\cot(\pi h)\right)\left(\frac{h'}{h}\right)^2z^2+\pi^2\int_0^1\mathcal{A}(h)a'b'z^2.

Both integrals are finite. Indeed, the bound (3.9) and Hardy’s inequality (2.4) give

∫01A(h)∣a′b′∣z2 dx≤Cr∫01z(x)2min⁡{x,1−x}2 dx≤Cr∫01∣z′∣2 dx<∞,\int_0^1\mathcal{A}(h)|a'b'|z^2\,dx\le C_r\int_0^1\frac{z(x)^2}{\min\{x,1-x\}^2}\,dx\le C_r\int_0^1|z'|^2\,dx<\infty,

and then the first integral is finite since T<+∞T<+\infty.

Write G:=M−π2(Z+T)\mathcal{G}:=M-\pi^2(Z+T). From (3.4), we obtain

G=E+V+∫01(πhcot⁡(πh)−14)(h′h)2z2−π2∫01A(h)a′b′z2.(30)\mathcal{G}=E+V+\int_0^1\left(\pi h\cot(\pi h)-\frac14\right)\left(\frac{h'}{h}\right)^2z^2-\pi^2\int_0^1\mathcal{A}(h)a'b'z^2. \tag*{(30)}

Since A\mathcal{A} is positive and increasing on (0,1)(0,1) and 0≤h≤r0\le h\le r, we have

∫01A(h)a′b′z2≤A(r)∫01(a′b′)+z2=A(r)S.\int_0^1\mathcal{A}(h)a'b'z^2\le\mathcal{A}(r)\int_0^1(a'b')_+z^2=\mathcal{A}(r)S.

By Lemma 3.3, we obtain

G≥(1−κr)(E+V)+∫01(πhcot⁡(πh)−14)(h′h)2z2.(31)\mathcal{G}\ge(1-\kappa_r)(E+V)+\int_0^1\left(\pi h\cot(\pi h)-\frac14\right)\left(\frac{h'}{h}\right)^2z^2. \tag*{(31)}

where

κr=514π2A(r)r2=5πr7tan⁡πr2.\kappa_r=\frac{5}{14}\pi^2\mathcal{A}(r)r^2=\frac{5\pi r}{7}\tan\frac{\pi r}{2}.

Since f(s)=scot⁡sf(s)=s\cot s is strictly decreasing on (0,π/2)(0,\pi/2) and h≤r≤ρh\le r\le\rho, we obtain

πhcot⁡(πh)−14≥0.\pi h\cot(\pi h)-\frac{1}{4}\ge0.

Moreover, κr\kappa_r is increasing in rr, and Lemma 2.2 gives κr≤κρ<1\kappa_r\le\kappa_\rho<1. It follows that

G≥(1−κρ)(E+V)≥0,\mathcal{G}\ge(1-\kappa_\rho)(E+V)\ge0,

that is,

M≥π2(Z+T).(32)M\ge\pi^2(Z+T). \tag*{(32)}

Since the mean-zero function uu was arbitrary, Lemma 3.2 yields μ1(Ω)≥π2\mu_1(\Omega)\ge\pi^2.

It remains to show the equality case. Suppose that μ1(Ω)=π2\mu_1(\Omega)=\pi^2. A bounded convex domain with nonempty interior is a Lipschitz domain, so the embedding H1(Ω)↪L2(Ω)H^1(\Omega)\hookrightarrow L^2(\Omega) is compact, and the infimum in (1.2) is attained. Let uu be a nonzero first eigenfunction, which has mean zero. Then

∫Ω∣∇u∣2=π2(M+Q).\int_\Omega|\nabla u|^2=\pi^2(M+Q).

By (22), we have M>0M>0. Substituting the preceding identity into (16), we obtain

M2≤π2M(Z+T),M^2\le\pi^2M(Z+T),

that is

M≤π2(Z+T).M\le\pi^2(Z+T).

Combining with (32), we have G=0\mathcal{G}=0. Thus, every nonnegative term on the right-hand side of (31) vanishes. Since 1−κr>01-\kappa_r>0, this implies

E=Va=Vb=0,E=V_a=V_b=0,

and

∫01(πhcot⁡(πh)−14)(h′h)2z2=0.(33)\int_0^1\left(\pi h\cot(\pi h)-\frac{1}{4}\right)\left(\frac{h'}{h}\right)^2z^2=0. \tag*{(33)}

The identity E=0E=0 shows that z/ϕz/\phi is constant on (0,1)(0,1), so that

z(x)=λsin⁡(πx).z(x)=\lambda\sin(\pi x).

The constant λ\lambda is nonzero, since otherwise F=h z=0F=\sqrt{h}\,z=0 and F′=−hmF'=-hm would give M=0M=0. Therefore z2/hz^2/h is continuous and strictly positive in (0,1)(0,1). Since νa\nu_a and νb\nu_b are nonnegative and Va=Vb=0V_a=V_b=0, it follows that

νa=νb=0on (0,1).\nu_a=\nu_b=0\quad\text{on }(0,1).

Thus aa, bb and hh are affine. Since h(x)≤r≤ρh(x)\le r\le\rho, (4.4) gives

(πhcot⁡(πh)−14)(h′h)2z2=0on (0,1),\left(\pi h\cot(\pi h)-\frac{1}{4}\right)\left(\frac{h'}{h}\right)^2z^2=0\quad\text{on }(0,1),

which implies hh is constant and a′=b′a'=b' in (0,1)(0,1). The convexity of aa and −b-b gives a′′=b′′=0a''=b''=0 and a′=b′=ka'=b'=k for some constant kk. By Lemma 2.1, k=0k=0, so the boundary functions aa and bb are constant. Since every side of RrR_r intersects Ω‾\overline{\Omega}, it follows that a≡0a\equiv0 and b≡rb\equiv r, and hence Ω=int⁡Rr\Omega=\operatorname{int}R_r. Conversely, μ1(int⁡Rr)=π2\mu_1(\operatorname{int}R_r)=\pi^2 because r<1r<1. Scaling back proves the theorem. □

Appendix A. Near-optimality of the aspect-ratio bound

In this appendix, we show that the threshold ρ\rho in Theorem 1.1 cannot be replaced by any number larger than r∗≈0.457993r_\ast\approx0.457993. This implies that the optimal threshold lies in [ρ,r∗][\rho,r_\ast]. The counterexamples are trapezoids obtained from the rectangle RrR_r by slightly tilting its top side. Their construction is governed by the following function. For 0<r<1/20<r<1/2, define

C(r)=8rπ{cot⁡(πr)−2∑k=1∞coth⁡(πr4k2−1)(4k2−1)5/2}.(34)C(r)=\frac{8r}{\pi}\left\{\cot(\pi r)-2\sum_{k=1}^{\infty}\frac{\coth(\pi r\sqrt{4k^2-1})}{(4k^2-1)^{5/2}}\right\}. \tag*{(34)}

Lemma A.1. The function CC is continuous and strictly decreasing on (0,1/2)(0,1/2). It has a unique zero r∗∈(0,1/2)r_\ast\in(0,1/2), and C(r)<0C(r)<0 for r∗<r<1/2r_\ast<r<1/2.

Proof. Since coth⁡\coth is decreasing on (0,∞)(0,\infty), the series in (A.1) converges uniformly on compact subintervals of (0,1/2](0,1/2]. Thus, CC is a continuous function on (0,1/2](0,1/2]. Set tk=πr4k2−1t_k=\pi r\sqrt{4k^2-1} and rewrite (A.1) as

C(r)=8π2{πrcot⁡(πr)−2∑k=1∞tkcoth⁡tk(4k2−1)3}.C(r)=\frac{8}{\pi^2}\left\{\pi r\cot(\pi r)-2\sum_{k=1}^{\infty}\frac{t_k\coth t_k}{(4k^2-1)^3}\right\}.

Recall that tcot⁡tt\cot t is strictly decreasing on (0,π/2)(0,\pi/2). Moreover, since for t>0t>0,

ddt(tcoth⁡t)=sinh⁡tcosh⁡t−tsinh⁡2t>0,\frac{d}{dt}(t\coth t)=\frac{\sinh t\cosh t-t}{\sinh^2t}>0,

it follows that CC is strictly decreasing on (0,1/2](0,1/2]. Since tcot⁡t→1t\cot t\to1 and tcoth⁡t→1t\coth t\to1 as t→0t\to0, we have

lim⁡r→0C(r)=8π2{1−2∑k=1∞1(4k2−1)3}>0.\lim_{r\to0}C(r)=\frac{8}{\pi^2}\left\{1-2\sum_{k=1}^{\infty}\frac{1}{(4k^2-1)^3}\right\}>0.

On the other hand, cot⁡(π/2)=0\cot(\pi/2)=0, so C(1/2)<0C(1/2)<0. Since CC is continuous and strictly decreasing on (0,1/2](0,1/2], it has a unique zero r∗∈(0,1/2)r_\ast\in(0,1/2), and C(r)<0C(r)<0 for r∗<r<1/2r_\ast<r<1/2. □

Proposition A.2. Let r∗<r<1/2r_\ast<r<1/2. For 0<s<10<s<1, consider the convex trapezoid

Ωr,s={(x,y):0<x<1, 0<y<r(1−sx)}.\Omega_{r,s}=\{(x,y):0<x<1,\ 0<y<r(1-sx)\}.

Then for every 0<s<10<s<1, the rectangle Rr=[0,1]×[0,r]R_r=[0,1]\times[0,r] is a minimum-area enclosing rectangle of Ωr,s\Omega_{r,s}, and

dH(Ω‾r,s,Rr)≤rs,d_H(\overline{\Omega}_{r,s},R_r)\le rs,

where dHd_H denotes the Hausdorff distance between compact sets. Moreover,

μ1(Ωr,s)<π2\mu_1(\Omega_{r,s})<\pi^2

for all sufficiently small $s>0.

The test functions in the proof of Proposition A.2 are perturbations of the first nonconstant Neumann eigenfunction of the rectangle. Put

Dr=(0,1)×(0,r),Ψ(x)=2cos⁡(πx).D_r=(0,1)\times(0,r),\qquad\Psi(x)=\sqrt{2}\cos(\pi x).

Then −ΔΨ=π2Ψ-\Delta\Psi=\pi^2\Psi and ∫DrΨ2=r\int_{D_r}\Psi^2=r. On Ωr,s\Omega_{r,s}, however, Ψ\Psi does not have mean zero, and its normal derivative does not vanish on the tilted side. We eliminate both obstructions by subtracting sχrs\chi_r, where χr\chi_r is the function on DrD_r given by the following lemma.

Lemma A.3. For each 0<r<1/20<r<1/2, there is a function χr∈C1(Dr‾)\chi_r\in C^1(\overline{D_r}) such that

∫Dr(∇χr⋅∇η−π2χrη)=r∫01Ψ′(x)η(x,r) dx.(35)\int_{D_r}\left(\nabla\chi_r\cdot\nabla\eta-\pi^2\chi_r\eta\right)=r\int_0^1\Psi'(x)\eta(x,r)\,dx. \tag*{(35)}

for every η∈H1(Dr‾)\eta\in H^1(\overline{D_r}). Moreover,

−π2∫Drχr=r(Ψ(1)−Ψ(0))=−22r.-\pi^2\int_{D_r}\chi_r=r\left(\Psi(1)-\Psi(0)\right)=-2\sqrt{2}r.
∫Dr(∣∇χr∣2−π2χr2)=−rC(r).(36)\int_{D_r}\left(\lvert\nabla\chi_r\rvert^2-\pi^2\chi_r^2\right)=-rC(r). \tag*{(36)}

Proof. On (0,1)(0,1), the function Ψ′(x)=−2πsin⁡(πx)\Psi'(x)=-\sqrt{2}\pi\sin(\pi x) has the cosine expansion

Ψ′(x)=−22+42∑k=1∞cos⁡(2kπx)4k2−1,\Psi'(x)=-2\sqrt{2}+4\sqrt{2}\sum_{k=1}^{\infty}\frac{\cos(2k\pi x)}{4k^2-1},

which converges absolutely and uniformly on [0,1][0,1]. Set

ωk=π4k2−1,k≥1,\omega_k=\pi\sqrt{4k^2-1},\qquad k\geq1,

and define

χr(x,y)=22rcos⁡(πy)πsin⁡(πr)+42r∑k=1∞cos⁡(2kπx)cosh⁡(ωky)(4k2−1)ωksinh⁡(ωkr).(37)\chi_r(x,y)=\frac{2\sqrt{2}r\cos(\pi y)}{\pi\sin(\pi r)}+4\sqrt{2}r\sum_{k=1}^{\infty}\frac{\cos(2k\pi x)\cosh(\omega_k y)}{(4k^2-1)\omega_k\sinh(\omega_k r)}. \tag*{(37)}

Since the kk-th term of the series in (A.4) and its first derivatives are bounded on Dr‾\overline{D_r} by Crk−2C_rk^{-2}, we have χr∈C1(Dr‾)\chi_r\in C^1(\overline{D_r}). Note that every term in (A.4) satisfies −Δv=π2v-\Delta v=\pi^2v. Thus,

−Δχr=π2χrin Dr.-\Delta\chi_r=\pi^2\chi_r\quad\text{in }D_r.

Differentiating termwise, we see that ∂νχr=0\partial_\nu\chi_r=0 on the bottom and vertical sides of DrD_r. On the top side, the cosine expansion of Ψ′\Psi' gives

∂νχr(x,r)=∂yχr(x,r)=rΨ′(x).\partial_\nu\chi_r(x,r)=\partial_y\chi_r(x,r)=r\Psi'(x).

Integrating by parts, we obtain (A.2).

Taking η=1\eta=1 in (A.2), we get

−π2∫Drχr=r(Ψ(1)−Ψ(0))=−22r.-\pi^2\int_{D_r}\chi_r=r\left(\Psi(1)-\Psi(0)\right)=-2\sqrt{2}r.

On the other hand, a direct computation gives

∫DrxΨ=2r∫01xcos⁡(πx) dx=−22rπ2.\int_{D_r}x\Psi=\sqrt{2}r\int_0^1x\cos(\pi x)\,dx=-\frac{2\sqrt{2}r}{\pi^2}.

This proves the first identity in (A.3). Taking η=χr\eta=\chi_r in (A.2), we get

∫Dr(∣∇χr∣2−π2χr2)=r∫01Ψ′(x)χr(x,r) dx.\int_{D_r}\left(\lvert\nabla\chi_r\rvert^2-\pi^2\chi_r^2\right)=r\int_0^1\Psi'(x)\chi_r(x,r)\,dx.

By the orthogonality of the functions 11 and cos⁡(2kπx)\cos(2k\pi x) in L2(0,1)L^2(0,1), we have

∫01Ψ′(x)χr(x,r) dx=−8rπcot⁡(πr)+16rπ∑k=1∞coth⁡(ωkr)(4k2−1)5/2\int_0^1\Psi'(x)\chi_r(x,r)\,dx=-\frac{8r}{\pi}\cot(\pi r)+\frac{16r}{\pi}\sum_{k=1}^{\infty}\frac{\coth(\omega_k r)}{(4k^2-1)^{5/2}}
=−C(r),=-C(r),

which yields the second identity. □

Proof of Proposition A.2. We first show that RrR_r is a minimum-area enclosing rectangle of Ωr,s\Omega_{r,s}. Since r(1−sx)≥r(1−x)r(1-sx) \ge r(1-x) for 0≤x≤10 \le x \le1, the closure of Ωr,s\Omega_{r,s} contains the triangle with vertices

(0,0),(1,0),(0,r)(0,0), \qquad(1,0), \qquad(0,r)

whose area is r/2r/2. Note that a triangle contained in a rectangle has at most half the area of the rectangle. Hence, every rectangle containing Ωr,s\Omega_{r,s} has area at least rr, which implies RrR_r is a minimum-area enclosing rectangle. Moreover, each point of RrR_r lies within vertical distance rsx≤rsrsx \le rs of Ω‾r,s\overline{\Omega}_{r,s}, which gives the Hausdorff estimate.

We now turn to the eigenvalue. Since r>r∗r > r_*, Lemma A.1 gives C(r)<0C(r) < 0. Let χr\chi_r be the function from Lemma A.3, and consider

Us(x,y)=Ψ(x)−sχr(x,y).U_s(x,y) = \Psi(x) - s\chi_r(x,y).

Here and below, Or(sk)O_r(s^k) denotes a quantity whose absolute value is bounded by CrskC_rs^k for 0<s<10 < s < 1.

Since −ΔΨ=π2Ψ-\Delta\Psi= \pi^2\Psi and Ψ′(0)=Ψ′(1)=0\Psi'(0) = \Psi'(1) = 0, it follows that for every η∈H1(Ωr,s)\eta\in H^1(\Omega_{r,s}),

∫Ωr,s(∇Ψ⋅∇η−π2Ψη)=rs∫01Ψ′(x)η(x,r(1−sx)) dx.(38)\int_{\Omega_{r,s}}(\nabla\Psi\cdot\nabla\eta-\pi^2\Psi\eta) =rs\int_0^1\Psi'(x)\eta(x,r(1-sx))\,dx. \tag*{(38)}

For η=Ψ\eta=\Psi, (A.5) gives

∫Ωr,s(∣∇Ψ∣2−π2Ψ2)=0.\int_{\Omega_{r,s}}(|\nabla\Psi|^2-\pi^2\Psi^2)=0.

Expanding the quadratic form and applying (A.5) with η=χr\eta=\chi_r, we obtain

∫Ωr,s(∣∇Us∣2−π2Us2)=s2∫Ωr,s(∣∇χr∣2−π2χr2)−2rs2∫01Ψ′(x)χr(x,r(1−sx)) dx.\int_{\Omega_{r,s}}(|\nabla U_s|^2-\pi^2U_s^2) =s^2\int_{\Omega_{r,s}}(|\nabla\chi_r|^2-\pi^2\chi_r^2) -2rs^2\int_0^1\Psi'(x)\chi_r(x,r(1-sx))\,dx.

Since χr\chi_r and its first derivatives are bounded on Dr‾\overline{D_r} and ∣Dr∖Ωr,s∣=rs/2|D_r\setminus\Omega_{r,s}|=rs/2, replacing Ωr,s\Omega_{r,s} by DrD_r in the first integral and r(1−sx)r(1-sx) by rr in the second changes the right-hand side only by Or(s3)O_r(s^3). Moreover, (A.2) with η=χr\eta=\chi_r gives

r∫01Ψ′(x)χr(x,r) dx=∫Dr(∣∇χr∣2−π2χr2).r\int_0^1\Psi'(x)\chi_r(x,r)\,dx =\int_{D_r}(|\nabla\chi_r|^2-\pi^2\chi_r^2).

Therefore, by (A.3),

∫Ωr,s(∣∇Us∣2−π2Us2)=−s2∫Dr(∣∇χr∣2−π2χr2)+Or(s3)=rC(r)s2+Or(s3).(39)\begin{aligned} \int_{\Omega_{r,s}}(|\nabla U_s|^2-\pi^2U_s^2) &=-s^2\int_{D_r}(|\nabla\chi_r|^2-\pi^2\chi_r^2)+O_r(s^3)\\ &=rC(r)s^2+O_r(s^3). \tag*{(39)} \end{aligned}

Since ∫01Ψ=0\int_0^1\Psi=0, the first identity in (A.3) gives

∫Ωr,sΨ=−s∫Drχr=s∫Drχr.\int_{\Omega_{r,s}}\Psi=-s\int_{D_r}\chi_r=s\int_{D_r}\chi_r.

Consequently,

∫Ωr,sUs=s∫Dr∖Ωr,sχr=Or(s2).\int_{\Omega_{r,s}}U_s=s\int_{D_r\setminus\Omega_{r,s}}\chi_r=O_r(s^2).

Let

U^s=Us−1∣Ωr,s∣∫Ωr,sUs\widehat{U}_s=U_s-\frac{1}{|\Omega_{r,s}|}\int_{\Omega_{r,s}}U_s

be the mean-zero part of UsU_s. Then ∇U^s=∇Us\nabla\widehat{U}_s=\nabla U_s and

∫Ωr,sU^s2=∫Ωr,sUs2−1∣Ωr,s∣(∫Ωr,sUs)2=∫Ωr,sUs2+Or(s4).\int_{\Omega_{r,s}} \widehat{U}_s^2=\int_{\Omega_{r,s}} U_s^2-\frac{1}{|\Omega_{r,s}|}\left(\int_{\Omega_{r,s}} U_s\right)^2=\int_{\Omega_{r,s}} U_s^2+O_r(s^4).

Hence, (A.6) gives

∫Ωr,s∣∇U^s∣2−π2∫Ωr,sU^s2=rC(r)s2+Or(s3),\int_{\Omega_{r,s}}|\nabla\widehat{U}_s|^2-\pi^2\int_{\Omega_{r,s}}\widehat{U}_s^2=rC(r)s^2+O_r(s^3),

which is negative for all sufficiently small s>0s>0, since C(r)<0C(r)<0. On the other hand, as s→0s\to0,

∫Ωr,sU^s2⟶∫DrΨ2=r>0.\int_{\Omega_{r,s}}\widehat{U}_s^2\longrightarrow\int_{D_r}\Psi^2=r>0.

Therefore, for all sufficiently small s>0s>0, the function U^s\widehat{U}_s is nonzero, has mean zero on Ωr,s\Omega_{r,s}, and has Rayleigh quotient less than π2\pi^2. The variational characterization (1.2) then gives μ1(Ωr,s)<π2\mu_1(\Omega_{r,s})<\pi^2. □

References

Department of Mathematics and Statistics, Auburn University, Auburn, AL, 36849, USA

Email address: [email protected]

School of Mathematical Sciences, Fudan University, Shanghai, 200433, China

Email address: [email protected]

School of Mathematical Sciences, Soochow University, Suzhou, 215006, China

Email address: [email protected]

References

  1. [1][ABF24] Vincenzo Amato, Dorin Bucur, and Ilaria Fragalà, The geometric size of the fundamental gap, 2024, arXiv preprint arXiv:2407.01341.arxiv.org/abs/2407.01341
  2. [2][ABF26] Vincenzo Amato, Dorin Bucur, and Ilaria Fragalà, A sharp quantitative nonlinear Poincaré inequality on convex domains, Ann. Inst. H. Poincaré C Anal. Non Linéaire, 2026, Published online first 24 June 2026, doi:10.4171/AIHPC/185.DOI
  3. [3][AC13] Ben Andrews and Julie Clutterbuck, Sharp modulus of continuity for parabolic equations on manifolds and lower bounds for the first eigenvalue, Anal. PDE 6 (2013), no. 5, 1013–1024. MR 3125548
  4. [4][AN12] Ben Andrews and Lei Ni, Eigenvalue comparison on Bakry-Emery manifolds, Comm. Partial Differential Equations 37 (2012), no. 11, 2081–2092. MR 3005536arxiv.org/abs/1111.4967
  5. [5][BQ00] Dominique Bakry and Zhongmin Qian, Some new results on eigenvectors via dimension, diameter, and Ricci curvature, Adv. Math. 155 (2000), no. 1, 98–153. MR 1789850DOI
  6. [6][CFH+25] Lorenzo Cavallina, Kei Funano, Antoine Henrot, Antoine Lemenant, Ilaria Lucardesi, and Shigeru Sakaguchi, Two extremum problems for Neumann eigenvalues, J. Anal. Math. 155 (2025), no. 2, 657–697. MR 4905324arxiv.org/abs/2312.13747
  7. [7][FK25] Pedro Freitas and James B. Kennedy, On domain monotonicity of Neumann eigenvalues of convex domains, Proc. Amer. Math. Soc. 153 (2025), no. 12, 5315–5328. MR 4989645arxiv.org/abs/2307.06593
  8. [8][Krö99] Pawel Kröger, On upper bounds for high order Neumann eigenvalues of convex domains in Euclidean space, Proc. Amer. Math. Soc. 127 (1999), no. 6, 1665–1669. MR 1486739DOI
  9. [9][LW21a] Xiaolong Li and Kui Wang, Lower bounds for the first eigenvalue of the Laplacian on Kähler manifolds, Trans. Amer. Math. Soc. 374 (2021), no. 11, 8081–8099. MR 4328692arxiv.org/abs/2010.12792
  10. [10][LW21b] ________, Sharp lower bound for the first eigenvalue of the weighted p-Laplacian I, J. Geom. Anal. 31 (2021), no. 8, 8686–8708. MR 4293545
  11. [11][LW21c] ________, Sharp lower bound for the first eigenvalue of the weighted p-Laplacian II, Math. Res. Lett. 28 (2021), no. 5, 1459–1479. MR 4471716arxiv.org/abs/1911.04596
  12. [12][LY80] Peter Li and Shing Tung Yau, Estimates of eigenvalues of a compact Riemannian manifold, Geometry of the Laplace operator (Proc. Sympos. Pure Math., Univ. Hawaii, Honolulu, Hawaii, 1979), Proc. Sympos. Pure Math., XXXVI, Amer. Math. Soc., Providence, R.I., 1980, pp. 205–239. MR 573435DOI
  13. [13][Ni13] Lei Ni, Estimates on the modulus of expansion for vector fields solving nonlinear equations, J. Math. Pures Appl. (9) 99 (2013), no. 1, 1–16. MR 3003280arxiv.org/abs/1107.2351
  14. [14][PW60] L. E. Payne and H. F. Weinberger, An optimal Poincaré inequality for convex domains, Arch. Rational Mech. Anal. 5 (1960), 286–292. MR 117419DOI
  15. [15][Sze54] G. Szegő, Inequalities for certain eigenvalues of a membrane of given area, J. Rational Mech. Anal. 3 (1954), 343–356. MR 61749DOI

Paper details

vlnc:W_dqcphmv4
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC0-1.0
Usage

views · 0

downloads · 0

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

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents