Introduction
Consider the two-dimensional incompressible Euler equations in a bounded Jordan domain D ⊂ R 2 D \subset\mathbb{R}^2 D ⊂ R 2 :
{ ∂ t v + ( v ⋅ ∇ ) v = − ∇ p in D , ∇ ⋅ v = 0 in D , (1) \begin{cases}
\partial_t\mathbf{v}+(\mathbf{v}\cdot\nabla)\mathbf{v}=-\nabla p & \text{in }D,\\
\nabla\cdot\mathbf{v}=0 & \text{in }D, \tag*{(1)}
\end{cases} { ∂ t v + ( v ⋅ ∇ ) v = − ∇ p ∇ ⋅ v = 0 in D , in D , (1)
where v = ( v 1 , v 2 ) T \mathbf{v}=(v_1,v_2)^{\mathsf{T}} v = ( v 1 , v 2 ) T is the fluid velocity and p p p is the pressure. For v ( t ) ∈ L 2 ( D ; R 2 ) \mathbf{v}(t)\in L^2(D;\mathbb{R}^2) v ( t ) ∈ L 2 ( D ; R 2 ) , incompressibility and the slip boundary condition are imposed in the weak form
∫ D v ( t , x ) ⋅ ∇ φ ( x ) d x = 0 ∀ φ ∈ H 1 ( D ) , t ≥ 0. (2) \int_D \mathbf{v}(t,x)\cdot\nabla\varphi(x)\,\mathrm{d}x=0 \qquad\forall\varphi\in H^1(D),\ t\geq0. \tag*{(2)} ∫ D v ( t , x ) ⋅ ∇ φ ( x ) d x = 0 ∀ φ ∈ H 1 ( D ) , t ≥ 0. (2)
It is not hard to verify that this equation recovers the usual slip boundary condition v ⋅ n = 0 \mathbf{v}\cdot\mathbf{n}=0 v ⋅ n = 0 for a smooth domain, where n \mathbf{n} n is the outward unit normal.
Under mild regularity assumptions, it is convenient to work with the vorticity formulation of the Euler equations:
∂ t ω + v ⋅ ∇ ω = 0 , v = ∇ ⊥ G ω , (3) \partial_t\omega+\mathbf{v}\cdot\nabla\omega=0,\qquad\mathbf{v}=\nabla^\perp\mathcal{G}\omega, \tag*{(3)} ∂ t ω + v ⋅ ∇ ω = 0 , v = ∇ ⊥ G ω , (3)
where ω : = ∂ 1 v 2 − ∂ 2 v 1 \omega:=\partial_1v_2-\partial_2v_1 ω := ∂ 1 v 2 − ∂ 2 v 1 is the vorticity, ∇ ⊥ = ( ∂ 2 , − ∂ 1 ) \nabla^\perp=(\partial_2,-\partial_1) ∇ ⊥ = ( ∂ 2 , − ∂ 1 ) and G = ( − Δ D ) − 1 \mathcal{G}=(-\Delta_D)^{-1} G = ( − Δ D ) − 1 is the Green operator for − Δ -\Delta − Δ with the homogeneous Dirichlet boundary condition. We note that
ψ = G ω ∈ H 0 1 ( D ) , ∀ ω ∈ H − 1 ( D ) \psi=\mathcal{G}\omega\in H^1_0(D),\qquad\forall\omega\in H^{-1}(D) ψ = G ω ∈ H 0 1 ( D ) , ∀ ω ∈ H − 1 ( D )
and v = ∇ ⊥ ψ \mathbf{v}=\nabla^\perp\psi v = ∇ ⊥ ψ satisfies (2) .
Yudovich’s theorem [29 ] asserts that L ∞ L^\infty L ∞ vorticity defines an Euler flow in smooth bounded domain globally in time. Thus the vorticity is transported by the Euler flow. Specifically, if the vorticity distribution takes a patch form initially, the solution retains the patch form for all time. Consequently, the vortex patch dynamics is reduced to the boundary dynamics of vortical domains. The global regularity of vortex patches in the plane was first proved by Chemin in [16 ] provided the initial patch is of class C 1 , α C^{1,\alpha} C 1 , α , 0 < α < 1 0<\alpha<1 0 < α < 1 , see also [4 , 24 ] for another elementary proof. Global regularity and small-scale formation in smooth bounded domains are studied in [20 ] , whereas the ill-posedness of C 2 C^2 C 2 -vortex patch is established in [21 ] .
A vortex patch is called steady when its vorticity is independent of time, or equivalently
∫ D ω ∇ ⊥ G ω ⋅ ∇ φ d x = 0 , φ ∈ C c ∞ ( D ) . \int_D \omega\nabla^\perp G\omega\cdot\nabla\varphi\,dx=0,\qquad\varphi\in C^\infty_c(D). ∫ D ω ∇ ⊥ G ω ⋅ ∇ φ d x = 0 , φ ∈ C c ∞ ( D ) .
Following the Kelvin-Benjamin variational principle that steady flows arise as critical points of kinetic energy over rearrangements of vorticity, Turkington [27 ] constructed steady vortex patches by maximizing the kinetic energy
E k i n ( ω ) = 1 2 ∫ D ∣ v ∣ 2 d x = 1 2 ∫ D ω G ω d x (4) \mathcal{E}_{\mathrm{kin}}(\omega)=\frac{1}{2}\int_D |\mathbf{v}|^2\,dx=\frac{1}{2}\int_D\omega G\omega\,dx \tag*{(4)} E kin ( ω ) = 2 1 ∫ D ∣ v ∣ 2 d x = 2 1 ∫ D ω G ω d x (4)
on the admissible class
A λ = { ω ∈ L ∞ ( D ) : 0 ≤ ω ≤ λ , ∫ D ω d x = 1 } . \mathcal{A}_\lambda=\left\{\omega\in L^\infty(D):0\leq\omega\leq\lambda,\ \int_D\omega\,dx=1\right\}. A λ = { ω ∈ L ∞ ( D ) : 0 ≤ ω ≤ λ , ∫ D ω d x = 1 } .
Here λ > ∣ D ∣ − 1 \lambda>|D|^{-1} λ > ∣ D ∣ − 1 is a prescribed vorticity bound, while the circulation is normalized to one. We note that A λ \mathcal{A}_\lambda A λ consists of 1 ∣ D ∣ 1 D \frac{1}{|D|}1_D ∣ D ∣ 1 1 D only when λ = ∣ D ∣ − 1 \lambda=|D|^{-1} λ = ∣ D ∣ − 1 .
In the rearrangement point of view [5 , 6 ] , the class of patches
R λ : = { λ 1 B : B ⊂ D measurable , ∣ B ∣ = λ − 1 } \mathcal{R}_\lambda:=\{\lambda1_B:B\subset D\ \text{measurable},\ |B|=\lambda^{-1}\} R λ := { λ 1 B : B ⊂ D measurable , ∣ B ∣ = λ − 1 }
is a rearrangement class, and A λ \mathcal{A}_\lambda A λ is its relaxed admissible class; more precisely, A λ \mathcal{A}_\lambda A λ is the weak L 2 ( D ) L^2(D) L 2 ( D ) closure of R λ \mathcal{R}_\lambda R λ . In the smooth-domain setting, Turkington [27 ] proved that the maximum is attained and that every maximizer is a steady vortex patch of the form
E k i n ( ω λ ) = max ω ∈ A λ E k i n ( ω ) , ω λ = λ 1 A λ , \mathcal{E}_{\mathrm{kin}}(\omega_\lambda)=\max_{\omega\in\mathcal{A}_\lambda}\mathcal{E}_{\mathrm{kin}}(\omega),\qquad\omega_\lambda=\lambda1_{A_\lambda}, E kin ( ω λ ) = ω ∈ A λ max E kin ( ω ) , ω λ = λ 1 A λ ,
where
A λ = { G ω λ > μ λ } ⊂ D , μ λ > 0 , ∣ A λ ∣ = λ − 1 . A_\lambda=\{G\omega_\lambda>\mu_\lambda\}\subset D,\qquad\mu_\lambda>0,\qquad|A_\lambda|=\lambda^{-1}. A λ = { G ω λ > μ λ } ⊂ D , μ λ > 0 , ∣ A λ ∣ = λ − 1 .
The threshold μ λ \mu_\lambda μ λ is the Lagrange multiplier associated with the circulation constraint, and we refer to these maximizers as energy-maximizing vortex patches . For sufficiently large λ > 0 \lambda>0 λ > 0 , Turkington proved that ∂ A λ \partial A_\lambda ∂ A λ is a single simple closed C 1 C^1 C 1 curve and that the rescaled patches converge to the unit disk [27 ] . For arbitrary λ > ∣ D ∣ − 1 \lambda>|D|^{-1} λ > ∣ D ∣ − 1 , he conjectured that the set of irregular free boundary points Σ λ : = { x ∈ ∂ A λ : ∇ G ω λ ( x ) = 0 } \Sigma_\lambda:=\{x\in\partial A_\lambda:\nabla G\omega_\lambda(x)=0\} Σ λ := { x ∈ ∂ A λ : ∇ G ω λ ( x ) = 0 } would satisfy
# Σ λ < ∞ or H 1 ( Σ λ ) = 0. \#\Sigma_\lambda<\infty\quad\text{or}\quad\mathcal{H}^1(\Sigma_\lambda)=0. # Σ λ < ∞ or H 1 ( Σ λ ) = 0.
Our aim is to establish boundary regularity for every energy-maximizing vortex patch in A λ \mathcal{A}_\lambda A λ . The existence and threshold argument are given in Proposition 2.1. We denote the bounded component of R 2 ∖ Γ \mathbb{R}^2\setminus\Gamma R 2 ∖ Γ by Int ( Γ ) \operatorname{Int}(\Gamma) Int ( Γ ) when Γ \Gamma Γ is a Jordan curve, and use ⨆ \bigsqcup ⨆ to address a pairwise disjoint union. The first result of this paper is given as follow:
Theorem 1.1. Let D ⊂ R 2 D \subset\mathbb{R}^{2} D ⊂ R 2 be a bounded Jordan domain and λ > ∣ D ∣ − 1 \lambda> |D|^{-1} λ > ∣ D ∣ − 1 . The energy functional attains its maximum on A λ \mathcal{A}_{\lambda} A λ , and every maximizer has the representation
ω λ = λ 1 A λ a.e. , u λ = G 1 A λ − q λ , q λ > 0 , \omega_{\lambda}=\lambda1_{A_{\lambda}}\quad\text{a.e.},\qquad u_{\lambda}=G1_{A_{\lambda}}-q_{\lambda},\qquad q_{\lambda}>0, ω λ = λ 1 A λ a.e. , u λ = G 1 A λ − q λ , q λ > 0 , where A λ = { u λ > 0 } A_{\lambda}=\{u_{\lambda}>0\} A λ = { u λ > 0 } , A λ ‾ ⊂ D \overline{A_{\lambda}}\subset D A λ ⊂ D , and
− Δ u λ = 1 A λ in D , u λ = − q λ on ∂ D . -\Delta u_{\lambda}=1_{A_{\lambda}}\quad\text{in }D,\qquad u_{\lambda}=-q_{\lambda}\quad\text{on }\partial D. − Δ u λ = 1 A λ in D , u λ = − q λ on ∂ D . The zero level-set of u λ u_{\lambda} u λ coincides with ∂ A λ \partial A_{\lambda} ∂ A λ . There are finitely many pairwise disjoint real-analytic Jordan curves Γ 1 , … , Γ N \Gamma_{1},\ldots,\Gamma_{N} Γ 1 , … , Γ N , N ≥ 1 N\ge1 N ≥ 1 , such that
A λ = ⋃ j = 1 N Int ( Γ j ) , ∂ A λ = ⋃ j = 1 N Γ j , min ∂ A λ ∣ ∇ u λ ∣ > 0. A_{\lambda}=\bigcup_{j=1}^{N}\operatorname{Int}(\Gamma_{j}),\qquad\partial A_{\lambda}=\bigcup_{j=1}^{N}\Gamma_{j},\qquad\min_{\partial A_{\lambda}}|\nabla u_{\lambda}|>0. A λ = j = 1 ⋃ N Int ( Γ j ) , ∂ A λ = j = 1 ⋃ N Γ j , ∂ A λ min ∣∇ u λ ∣ > 0. For each 0 < α < 1 0<\alpha<1 0 < α < 1 , there are C , r ∗ > 0 C,r_{*}>0 C , r ∗ > 0 , depending on the fixed patch and α \alpha α , such that
sup p ∈ ∂ A λ ∣ ∣ A λ ∩ B r ( p ) ∣ π r 2 − 1 2 ∣ ≤ C r α ( 0 < r < r ∗ ) . (5) \sup_{p\in\partial A_{\lambda}}\left|\frac{|A_{\lambda}\cap B_{r}(p)|}{\pi r^{2}}-\frac{1}{2}\right|\le Cr^{\alpha}\qquad(0<r<r_{*}). \tag*{(5)} p ∈ ∂ A λ sup π r 2 ∣ A λ ∩ B r ( p ) ∣ − 2 1 ≤ C r α ( 0 < r < r ∗ ) . (5)
Remark 1.2. A C 0 C^{0} C 0 boundary is understood in the local graph sense: after a change of coordinates, for a neighborhood U U U and a continuous function h h h ,
D ∩ U = { ( x 1 , x 2 ) ∈ U : x 2 > h ( x 1 ) } , ∂ D ∩ U = { ( x 1 , x 2 ) ∈ U : x 2 = h ( x 1 ) } . D\cap U=\{(x_{1},x_{2})\in U:x_{2}>h(x_{1})\},\qquad\partial D\cap U=\{(x_{1},x_{2})\in U:x_{2}=h(x_{1})\}. D ∩ U = {( x 1 , x 2 ) ∈ U : x 2 > h ( x 1 )} , ∂ D ∩ U = {( x 1 , x 2 ) ∈ U : x 2 = h ( x 1 )} .
The super level-set representation of energy-maximizing vortex patches reduces their boundary regularity to the unstable obstacle equation
− Δ u = 1 { u > 0 } , (6) -\Delta u=1_{\{u>0\}}, \tag*{(6)} − Δ u = 1 { u > 0 } , (6)
whose singular solutions and variational regularity are studied in [3 , 2 , 1 , 23 ] . With the normalized energy E ( ρ ) = 1 2 ⟨ ρ , G ρ ⟩ \mathcal{E}(\rho)=\frac{1}{2}\langle\rho,G\rho\rangle E ( ρ ) = 2 1 ⟨ ρ , Gρ ⟩ and admissible class
A m = { ρ ∈ L ∞ ( D ) : 0 ≤ ρ ≤ 1 , ∫ D ρ = m } . (7) \mathcal{A}_{m}=\left\{\rho\in L^{\infty}(D):0\le\rho\le1,\ \int_{D}\rho=m\right\}. \tag*{(7)} A m = { ρ ∈ L ∞ ( D ) : 0 ≤ ρ ≤ 1 , ∫ D ρ = m } . (7)
By the change of normalization
m = λ − 1 , ω = λ ρ , E k i n ( ω ) = λ 2 E ( ρ ) , m=\lambda^{-1},\qquad\omega=\lambda\rho,\qquad\mathcal{E}_{\mathrm{kin}}(\omega)=\lambda^{2}\mathcal{E}(\rho), m = λ − 1 , ω = λ ρ , E kin ( ω ) = λ 2 E ( ρ ) ,
the normalized stream function u = λ − 1 ( G ω λ − μ λ ) u=\lambda^{-1}(G\omega_{\lambda}-\mu_{\lambda}) u = λ − 1 ( G ω λ − μ λ ) satisfies equation (6) . For the boundary regularity of uniformly rotating vortex patches, one has a sign-changing elliptic free-boundary problem, see for example [19 , 12 , 28 ] .
Remark 1.3. At the endpoint λ = ∣ D ∣ − 1 \lambda=|D|^{-1} λ = ∣ D ∣ − 1 , the admissible class consists only of λ 1 D \lambda1_{D} λ 1 D , whose boundary inherits the regularity of D D D . For every nontrivial bound λ > ∣ D ∣ − 1 \lambda>|D|^{-1} λ > ∣ D ∣ − 1 , Theorem 1.1 gives Σ λ = ∅ \Sigma_{\lambda}=\varnothing Σ λ = ∅ and hence completely resolves Turkington’s free-boundary regularity question. The theorem concerns each connected component and does not assert that the patch is connected; see Remark 4.3.
Our approach follows the weighted trace and annular arguments developed by Chanillo, Kenig, and To [14 ] for the composite membrane problem. For the nonlocal kinetic energy, we derive the area-constrained second-variation inequality (30) from the exact quadratic energy expansion. The variations act on arbitrary finite collections of regular boundary arcs under a single total-area constraint, retaining the Green interactions between distinct pieces. Strong convergence of the rescaled-vorticity layers in H − 1 ( D ) H^{-1}(D) H − 1 ( D ) justifies passage to the quadratic term. The resulting weighted trace inequality yields annular summability, while a torsion identity gives the scale-independent lower bound 4 π 4\pi 4 π for the weighted boundary integral of each closed regular component. Together with disk comparison, these estimates exclude singular points on the free boundary for every m ∈ ( 0 , ∣ D ∣ ) m \in(0, |D|) m ∈ ( 0 , ∣ D ∣ ) . Recently, Ferreri, Mazari-Fouquer, and Prunier [18 ] establish regularity for volume-constrained optimal control problems with bilinear state equations of the form − Δ Θ m = m Θ m + B ( x , Θ m ) -\Delta\Theta_m = m\Theta_m + B(x, \Theta_m) − Δ Θ m = m Θ m + B ( x , Θ m ) , while the present problem has the fixed Poisson state equation − Δ ψ = ρ -\Delta\psi= \rho − Δ ψ = ρ and the quadratic functional E ( ρ ) = 1 2 ⟨ ρ , G ρ ⟩ \mathcal{E}(\rho) = \frac{1}{2}\langle\rho, G\rho\rangle E ( ρ ) = 2 1 ⟨ ρ , Gρ ⟩ . Throughout this paper, the free boundary is the topological boundary of the open superlevel representative, and Lemma 2.2 identifies it with the boundary defined by positivity of both phase measures in every ball enclosing the point.
The regularity theorem also makes it possible to count distinct energy maximizers and to study their stability. To formulate these additional conclusions, we define
V ( m ) : = max ρ ∈ A m E ( ρ ) , M m : = { ρ ∈ A m : E ( ρ ) = V ( m ) } , argmax ω ∈ A λ E k i n ( ω ) : = { λ ρ : ρ ∈ M 1 / λ } . (8) \begin{aligned}
V(m) &:= \max_{\rho\in\mathcal{A}_m} \mathcal{E}(\rho), \qquad\mathcal{M}_m := \{\rho\in\mathcal{A}_m : \mathcal{E}(\rho) = V(m)\},\\
\operatorname*{argmax}_{\omega\in\mathcal{A}_\lambda} \mathcal{E}_{\mathrm{kin}}(\omega) &:= \{\lambda\rho: \rho\in\mathcal{M}_{1/\lambda}\}. \tag*{(8)}
\end{aligned} V ( m ) ω ∈ A λ argmax E kin ( ω ) := ρ ∈ A m max E ( ρ ) , M m := { ρ ∈ A m : E ( ρ ) = V ( m )} , := { λ ρ : ρ ∈ M 1/ λ } . (8)
Here we identify patches with their characteristic functions modulo null-measure sets, and write A ∈ M m A \in\mathcal{M}_m A ∈ M m when 1 A ∈ M m \mathbf{1}_A \in\mathcal{M}_m 1 A ∈ M m . Thus # M m \#\mathcal{M}_m # M m counts distinct maximizers, whereas the integer N N N in Theorem 1.1 counts number of connected-components of energy-maximizing vortex patches.
The following theorem gives a uniform count over the full mass interval in a smooth convex domain; the boundary may contain line segments. Smoothness is retained for the endpoint estimates and for the stability assertion.
Theorem 1.4. Let D ⊂ R 2 D \subset\mathbb{R}^2 D ⊂ R 2 be a smooth convex bounded domain . There exists an integer N D ≥ 1 N_D \ge1 N D ≥ 1 , depending only on D D D , such that
1 ≤ # M m ≤ N D ( 0 < m < ∣ D ∣ ) . 1 \le\#\mathcal{M}_m \le N_D \qquad(0 < m < |D|). 1 ≤ # M m ≤ N D ( 0 < m < ∣ D ∣ ) . Moreover, # M m = 1 \#\mathcal{M}_m = 1 # M m = 1 for all sufficiently small m > 0 m > 0 m > 0 and all m < ∣ D ∣ m < |D| m < ∣ D ∣ sufficiently close to ∣ D ∣ |D| ∣ D ∣ .
The proof combines boundary regularity with a nodal argument and a bilinear Pohozaev identity, followed by analytic reduction and compactness. It does not require nondegeneracy of the full mass-constrained second variation. Small-mass uniqueness follows from [7 ] , while uniqueness near full mass is derived from a boundary collar estimate.
No additional regularity assumption on the boundary is required. The proof in § 7 also gives
sup m ∈ I # M m < ∞ I ⋐ ( 0 , ∣ D ∣ ) . \sup_{m\in I} \#\mathcal{M}_m < \infty\qquad I \Subset(0, |D|). m ∈ I sup # M m < ∞ I ⋐ ( 0 , ∣ D ∣ ) .
The endpoint estimates used for the mass-independent bound in Theorem 1.4 require a smooth fluid boundary.
The finiteness of the set of energy maximizers implies that each maximizer is isolated and hence is a strict local maximizer, with respect to the L p L^p L p norm for each 1 < p < ∞ 1 < p < \infty 1 < p < ∞ , of the kinetic energy on its rearrangement class. Burton’s stability theorem [6 Theorem 5] then yields the following stability result.
Corollary 1.6. Under the assumptions of theorem 1.4, let λ > ∣ D ∣ − 1 \lambda> |D|^{-1} λ > ∣ D ∣ − 1 and let ω ∗ \omega_* ω ∗ maximize E k i n \mathcal{E}_{\mathrm{kin}} E kin on A λ \mathcal{A}_\lambda A λ . For every 1 < p < ∞ 1 < p < \infty 1 < p < ∞ and ε > 0 \varepsilon> 0 ε > 0 , there is δ > 0 \delta> 0 δ > 0 such that any Yudovich solution ω ( t ) \omega(t) ω ( t ) of (1.3) satisfies
∥ ω 0 − ω ∗ ∥ L p ( D ) < δ ⟹ sup t ≥ 0 ∥ ω ( t ) − ω ∗ ∥ L p ( D ) < ε . \|\omega_0-\omega_*\|_{L^p(D)}<\delta
\quad\Longrightarrow\quad
\sup_{t\geq0}\|\omega(t)-\omega_*\|_{L^p(D)}<\varepsilon. ∥ ω 0 − ω ∗ ∥ L p ( D ) < δ ⟹ t ≥ 0 sup ∥ ω ( t ) − ω ∗ ∥ L p ( D ) < ε .
The remainder of the paper is organized as follows. § 2 fixes the superlevel representation and the elliptic estimates. The variational and geometric arguments in § 3 lead to the proof of the main regularity theorem in § 4 . The Dirichlet zero-mode analysis in § 5 and the analytic reduction in § 6 yield fixed-mass finiteness and uniform bounds on compact mass intervals for smooth convex domains. § 7 extends these counts to arbitrary bounded convex domains, then establishes the endpoint estimates under the smooth boundary assumption. The appendix records conditional extensions to general smooth simply connected domains and identities for smooth families of maximizing patches.
Preliminaries
Throughout § 2 to § 4 , D D D is a bounded Jordan domain. We first specify the Dirichlet realization and its boundary continuity, which will keep the maximizing patch away from ∂ D \partial D ∂ D without requiring boundary differentiability. Set H − 1 ( D ) = ( H 0 1 ( D ) ) ∗ H^{-1}(D)=(H_0^1(D))^* H − 1 ( D ) = ( H 0 1 ( D ) ) ∗ , where H 0 1 ( D ) H_0^1(D) H 0 1 ( D ) is the closure of C c ∞ ( D ) C_c^\infty(D) C c ∞ ( D ) in H 1 ( D ) H^1(D) H 1 ( D ) . Poincaré’s inequality and the weak Dirichlet problem define G f ∈ H 0 1 ( D ) \mathcal{G}f\in H_0^1(D) G f ∈ H 0 1 ( D ) by
∫ D ∇ G f ⋅ ∇ z d x = ⟨ f , z ⟩ ( z ∈ H 0 1 ( D ) ) . (9) \int_D \nabla\mathcal{G}f\cdot\nabla z\,\mathrm{d}x=\langle f,z\rangle
\qquad(z\in H_0^1(D)). \tag*{(9)} ∫ D ∇ G f ⋅ ∇ z d x = ⟨ f , z ⟩ ( z ∈ H 0 1 ( D )) . (9)
The Dirichlet Green function is represented as
G D ( x , y ) = Φ ( x − y ) − H D ( x , y ) , Φ ( z ) = − 1 2 π log ∣ z ∣ , G_D(x,y)=\Phi(x-y)-H_D(x,y),\qquad\Phi(z)=-\frac{1}{2\pi}\log|z|, G D ( x , y ) = Φ ( x − y ) − H D ( x , y ) , Φ ( z ) = − 2 π 1 log ∣ z ∣ ,
where H D H_D H D is the regular part of the Green function and R D ( x ) : = H D ( x , x ) R_D(x):=H_D(x,x) R D ( x ) := H D ( x , x ) is the Robin function. For f ∈ H − 1 ( D ) f\in H^{-1}(D) f ∈ H − 1 ( D ) ,
∥ f ∥ H ˙ D − 1 2 : = ⟨ f , G f ⟩ = ∫ D ∣ ∇ G f ∣ 2 d x = sup 0 ≠ v ∈ H 0 1 ( D ) ⟨ f , v ⟩ 2 ∫ D ∣ ∇ v ∣ 2 d x . (10) \|f\|_{\dot H^{-1}_D}^{2}:=\langle f,\mathcal{G}f\rangle
=\int_D|\nabla\mathcal{G}f|^2\,\mathrm{d}x
=\sup_{0\ne v\in H_0^1(D)}
\frac{\langle f,v\rangle^2}{\int_D|\nabla v|^2\,\mathrm{d}x}. \tag*{(10)} ∥ f ∥ H ˙ D − 1 2 := ⟨ f , G f ⟩ = ∫ D ∣∇ G f ∣ 2 d x = 0 = v ∈ H 0 1 ( D ) sup ∫ D ∣∇ v ∣ 2 d x ⟨ f , v ⟩ 2 . (10)
To verify continuity up to the possibly rough fluid boundary, let D = { z ∈ C : ∣ z ∣ < 1 } \mathbb{D}=\{z\in\mathbb{C}:|z|<1\} D = { z ∈ C : ∣ z ∣ < 1 } . The Riemann mapping theorem and Carathéodory’s theorem [32 ] give a conformal map
F : D ⟶ D , F : D ‾ ⟶ D ‾ a homeomorphism . F:\mathbb{D}\longrightarrow D,\qquad F:\overline{\mathbb{D}}\longrightarrow\overline{D}\text{ a homeomorphism}. F : D ⟶ D , F : D ⟶ D a homeomorphism .
For f ∈ L ∞ ( D ) f\in L^\infty(D) f ∈ L ∞ ( D ) , extend f f f by zero and define its Newtonian potential and pullback by
N f ( x ) : = ∫ D Φ ( x − y ) f ( y ) d y , w f = N f ∘ F . N_f(x):=\int_D\Phi(x-y)f(y)\,\mathrm{d}y,\qquad w_f=N_f\circ F. N f ( x ) := ∫ D Φ ( x − y ) f ( y ) d y , w f = N f ∘ F .
The potential is C 1 , α C^{1,\alpha} C 1 , α on bounded subsets of R 2 \mathbb{R}^2 R 2 for every 0 < α < 1 0<\alpha<1 0 < α < 1 . Conformal invariance of the Dirichlet integral therefore gives
w f ∈ H 1 ( D ) ∩ C ( D ‾ ) , ∫ D ∣ ∇ w f ∣ 2 = ∫ D ∣ ∇ N f ∣ 2 < ∞ . w_f\in H^1(\mathbb{D})\cap C(\overline{\mathbb{D}}),\qquad\int_{\mathbb{D}}|\nabla w_f|^2=\int_D|\nabla N_f|^2<\infty. w f ∈ H 1 ( D ) ∩ C ( D ) , ∫ D ∣∇ w f ∣ 2 = ∫ D ∣∇ N f ∣ 2 < ∞.
Let h f h_f h f be the harmonic extension of its boundary values. The Poisson formula and the Dirichlet principle on the disk yield
h f ( z ) = 1 2 π ∫ 0 2 π 1 − ∣ z ∣ 2 ∣ e i t − z ∣ 2 w f ( e i t ) d t , h f ∈ H 1 ( D ) ∩ C ( D ‾ ) , h_f(z)=\frac{1}{2\pi}\int_0^{2\pi}\frac{1-|z|^2}{|e^{it}-z|^2}w_f(e^{it})\,\mathrm{d}t,\qquad h_f\in H^1(\mathbb{D})\cap C(\overline{\mathbb{D}}), h f ( z ) = 2 π 1 ∫ 0 2 π ∣ e i t − z ∣ 2 1 − ∣ z ∣ 2 w f ( e i t ) d t , h f ∈ H 1 ( D ) ∩ C ( D ) ,
w f − h f ∈ H 0 1 ( D ) , ∫ D ∣ ∇ h f ∣ 2 ≤ ∫ D ∣ ∇ w f ∣ 2 . w_f-h_f\in H_0^1(\mathbb{D}),\qquad\int_{\mathbb{D}}|\nabla h_f|^2\leq\int_{\mathbb{D}}|\nabla w_f|^2. w f − h f ∈ H 0 1 ( D ) , ∫ D ∣∇ h f ∣ 2 ≤ ∫ D ∣∇ w f ∣ 2 .
Approximating w f − h f w_f-h_f w f − h f by functions in C c ∞ ( D ) C_c^\infty(\mathbb{D}) C c ∞ ( D ) , then composing with F − 1 F^{-1} F − 1 , preserves convergence of the Dirichlet integrals; Poincaré’s inequality on D D D also gives convergence in L 2 ( D ) L^2(D) L 2 ( D ) . Hence
( w f − h f ) ∘ F − 1 ∈ H 0 1 ( D ) , − Δ ( N f − h f ∘ F − 1 ) = f . (w_f-h_f)\circ F^{-1}\in H_0^1(D),\qquad-\Delta(N_f-h_f\circ F^{-1})=f. ( w f − h f ) ∘ F − 1 ∈ H 0 1 ( D ) , − Δ ( N f − h f ∘ F − 1 ) = f .
Uniqueness in (9) identifies the weak solution with this continuous potential:
G f = N f − h f ∘ F − 1 ∈ H 0 1 ( D ) ∩ C ( D ‾ ) ∩ W l o c 2 , p ( D ) ( 1 < p < ∞ ) , \mathcal{G}f=N_f-h_f\circ F^{-1}\in H_0^1(D)\cap C(\overline{D})\cap W_{\mathrm{loc}}^{2,p}(D)\qquad(1<p<\infty), G f = N f − h f ∘ F − 1 ∈ H 0 1 ( D ) ∩ C ( D ) ∩ W loc 2 , p ( D ) ( 1 < p < ∞ ) ,
G f ∣ ∂ D = 0 , ∣ G f ∣ ≤ ∥ f ∥ L ∞ ( D ) τ , τ : = G 1 D ∈ C ( D ‾ ) . (11) \mathcal{G}f|_{\partial D}=0,\qquad|\mathcal{G}f|\leq\|f\|_{L^\infty(D)}\tau,\qquad\tau:=\mathcal{G}1_D\in C(\overline{D}). \tag*{(11)} G f ∣ ∂ D = 0 , ∣ G f ∣ ≤ ∥ f ∥ L ∞ ( D ) τ , τ := G 1 D ∈ C ( D ) . (11)
The last bound follows from the weak comparison principle. Interior elliptic estimates consequently give
∥ G f ∥ W 2 , p ( D ′ ) ≤ C D ′ , D , p ∥ f ∥ L ∞ ( D ) ( D ′ ‾ ⊂ D ) . (12) \|\mathcal{G}f\|_{W^{2,p}(D')}\leq C_{D',D,p}\|f\|_{L^\infty(D)}\qquad(\overline{D'}\subset D). \tag*{(12)} ∥ G f ∥ W 2 , p ( D ′ ) ≤ C D ′ , D , p ∥ f ∥ L ∞ ( D ) ( D ′ ⊂ D ) . (12)
We next verify that the Dirichlet stream function enforces (2) on the rough domain. Given ψ ∈ H 0 1 ( D ) \psi\in H_0^1(D) ψ ∈ H 0 1 ( D ) , choose
ψ k ∈ C c ∞ ( D ) , ψ k ⟶ ψ in H 1 ( D ) . \psi_k\in C_c^\infty(D),\qquad\psi_k\longrightarrow\psi\text{ in }H^1(D). ψ k ∈ C c ∞ ( D ) , ψ k ⟶ ψ in H 1 ( D ) .
For every φ ∈ H 1 ( D ) \varphi\in H^1(D) φ ∈ H 1 ( D ) , integration by parts is confined to the compact support of ψ k \psi_k ψ k , and therefore
∫ D ∇ ⊥ ψ k ⋅ ∇ φ d x = − ∫ D φ ∇ ⋅ ( ∇ ⊥ ψ k ) d x = 0. \int_D\nabla^\perp\psi_k\cdot\nabla\varphi\,\mathrm{d}x=-\int_D\varphi\,\nabla\cdot(\nabla^\perp\psi_k)\,\mathrm{d}x=0. ∫ D ∇ ⊥ ψ k ⋅ ∇ φ d x = − ∫ D φ ∇ ⋅ ( ∇ ⊥ ψ k ) d x = 0.
Passing to the limit gives the required identity without a boundary trace of the velocity:
∣ ∫ D ∇ ⊥ ψ ⋅ ∇ φ d x ∣ ≤ ∥ ∇ ( ψ − ψ k ) ∥ L 2 ( D ) ∥ ∇ φ ∥ L 2 ( D ) ⟶ 0. (13) \left|\int_D\nabla^\perp\psi\cdot\nabla\varphi\,\mathrm{d}x\right|\leq\|\nabla(\psi-\psi_k)\|_{L^2(D)}\|\nabla\varphi\|_{L^2(D)}\longrightarrow0. \tag*{(13)} ∫ D ∇ ⊥ ψ ⋅ ∇ φ d x ≤ ∥∇ ( ψ − ψ k ) ∥ L 2 ( D ) ∥∇ φ ∥ L 2 ( D ) ⟶ 0. (13)
In particular, v = ∇ ⊥ G ω \mathbf{v}=\nabla^\perp G\omega v = ∇ ⊥ G ω satisfies (2) for every bounded vorticity ω \omega ω . Testing with φ ∈ C c ∞ ( D ) \varphi\in C^\infty_c(D) φ ∈ C c ∞ ( D ) also yields
∇ ⋅ v = 0 in D ′ ( D ) . \nabla\cdot\mathbf{v}=0\quad\text{in }\mathcal{D}'(D). ∇ ⋅ v = 0 in D ′ ( D ) .
Let H 1 \mathcal{H}^1 H 1 denote the one-dimensional Hausdorff measure and let \mu\mathbin{\vrule height 1.4ex depth -0.3ex width 0.07ex\vrule height 0.07ex depth -0.02ex width 0.7ex}E denote restriction to E E E , see for example [25 ] . We note that d H 1 = d s \mathrm{d}\mathcal{H}^1=\mathrm{d}s d H 1 = d s for regular curves. On smooth boundaries, n n n denotes the outward unit normal of D D D and ν \nu ν that of a patch. For a smooth compact curve Γ ⊂ D \Gamma\subset D Γ ⊂ D and φ ∈ L 2 ( Γ ) \varphi\in L^2(\Gamma) φ ∈ L 2 ( Γ ) , the notation φ d s ∣ Γ \varphi\mathrm{d}s|_\Gamma φ d s ∣ Γ denotes the distribution z ↦ ∫ Γ φ z d s z\mapsto\int_\Gamma\varphi z\,\mathrm{d}s z ↦ ∫ Γ φ z d s . For a measurable E ⊂ D E\subset D E ⊂ D , we define
θ E ( x , r ) : = ∣ E ∩ B r ( x ) ∣ π r 2 , B r ( x ) ⊂ D , (14) \theta_E(\mathbf{x},r):=\frac{|E\cap B_r(\mathbf{x})|}{\pi r^2},\qquad B_r(\mathbf{x})\subset D, \tag*{(14)} θ E ( x , r ) := π r 2 ∣ E ∩ B r ( x ) ∣ , B r ( x ) ⊂ D , (14)
and set
E ( t ) : = { x ∈ D : lim r ↓ 0 θ E ( x , r ) = t } , t ∈ { 0 , 1 } . (15) E^{(t)}:=\left\{\mathbf{x}\in D:\lim_{r\downarrow0}\theta_E(\mathbf{x},r)=t\right\},\qquad t\in\{0,1\}. \tag*{(15)} E ( t ) := { x ∈ D : r ↓ 0 lim θ E ( x , r ) = t } , t ∈ { 0 , 1 } . (15)
The measure-theoretic boundary is defined as:
∂ m t E : = D ∖ ( E ( 0 ) ∪ E ( 1 ) ) . (16) \partial_{\mathrm{mt}}E:=D\setminus\left(E^{(0)}\cup E^{(1)}\right). \tag*{(16)} ∂ mt E := D ∖ ( E ( 0 ) ∪ E ( 1 ) ) . (16)
Since θ D ∖ E = 1 − θ E \theta_{D\setminus E}=1-\theta_E θ D ∖ E = 1 − θ E , one has
∂ m t E = { lim sup r ↓ 0 θ E > 0 , lim sup r ↓ 0 θ D ∖ E > 0 } ⊂ ∂ E . (17) \partial_{\mathrm{mt}}E=\left\{\limsup_{r\downarrow0}\theta_E>0,\ \limsup_{r\downarrow0}\theta_{D\setminus E}>0\right\}\subset\partial E. \tag*{(17)} ∂ mt E = { r ↓ 0 lim sup θ E > 0 , r ↓ 0 lim sup θ D ∖ E > 0 } ⊂ ∂ E . (17)
Here ∂ E \partial E ∂ E is the topological boundary of the chosen representative. The density classes are unchanged by null-set modifications of E E E .
The following proposition selects an open representative of each maximizing patch whose boundary lies strictly inside D D D , allowing us to study the free boundary through interior estimates for the stream function. It extends the existence and threshold argument of [27 Theorem 2.1 and Corollary 2.3] to bounded Jordan domains; the additional boundary-continuity input is (11) .
Proposition 2.1. Let D ⊂ R 2 D\subset\mathbb{R}^2 D ⊂ R 2 be a bounded Jordan domain, and let λ > ∣ D ∣ − 1 \lambda>|D|^{-1} λ > ∣ D ∣ − 1 . The energy E k i n \mathcal{E}_{\mathrm{kin}} E kin attains its maximum on A λ \mathcal{A}_\lambda A λ . Every maximizer satisfies
ω λ = λ 1 A λ a.e. , A λ = { G ω λ > μ λ } ⊂ D , ∣ A λ ∣ = λ − 1 , (18) \omega_\lambda=\lambda\mathbf{1}_{A_\lambda}\quad\text{a.e.},\qquad A_\lambda=\{G\omega_\lambda>\mu_\lambda\}\subset D,\qquad|A_\lambda|=\lambda^{-1}, \tag*{(18)} ω λ = λ 1 A λ a.e. , A λ = { G ω λ > μ λ } ⊂ D , ∣ A λ ∣ = λ − 1 , (18) where μ λ > 0 \mu_\lambda>0 μ λ > 0 is uniquely determined by ω λ \omega_\lambda ω λ . Moreover, A λ ‾ ⊂ D \overline{A_\lambda}\subset D A λ ⊂ D . Each such maximizer is a weak solution of the steady Euler equation.
Proof. Existence follows from weak-* compactness once continuity of the energy has been established: the assumption λ > ∣ D ∣ − 1 \lambda>|D|^{-1} λ > ∣ D ∣ − 1 makes the constant density admissible, while the mass and box constraints are preserved under weak-* limits:
∣ D ∣ − 1 1 D ∈ A λ , sup ω ∈ A λ ∥ ω ∥ L ∞ ≤ λ , |D|^{-1}\mathbf{1}_D\in\mathcal{A}_\lambda,\qquad\sup_{\omega\in\mathcal{A}_\lambda}\|\omega\|_{L^\infty}\leq\lambda, ∣ D ∣ − 1 1 D ∈ A λ , ω ∈ A λ sup ∥ ω ∥ L ∞ ≤ λ ,
ω n ∈ A λ , ω n ⇀ ∗ ω ⟹ ω ∈ A λ . \omega_n\in\mathcal{A}_\lambda,\qquad\omega_n\mathop{\rightharpoonup}^{*}\omega\Longrightarrow\omega\in\mathcal{A}_\lambda. ω n ∈ A λ , ω n ⇀ ∗ ω ⟹ ω ∈ A λ .
To pass the quadratic energy to the limit, we use compactness of H 0 1 ( D ) ↪ L 2 ( D ) H^1_0(D)\hookrightarrow L^2(D) H 0 1 ( D ) ↪ L 2 ( D ) , which holds for every bounded domain. For Ψ n : = G ω n \Psi_n:=\mathcal{G}\omega_n Ψ n := G ω n and Ψ : = G ω \Psi:=\mathcal{G}\omega Ψ := G ω , the weak Dirichlet equation, Poincaré’s inequality, and uniqueness give
sup n ∥ ∇ Ψ n ∥ L 2 ( D ) ≤ C D λ ∣ D ∣ 1 / 2 , Ψ n ⇀ Ψ in H 0 1 ( D ) , \sup_n\|\nabla\Psi_n\|_{L^2(D)}\le C_D\lambda|D|^{1/2},\qquad\Psi_n\rightharpoonup\Psi\quad\text{in }H^1_0(D), n sup ∥∇ Ψ n ∥ L 2 ( D ) ≤ C D λ ∣ D ∣ 1/2 , Ψ n ⇀ Ψ in H 0 1 ( D ) ,
Ψ n ⟶ Ψ in L 2 ( D ) , \Psi_n\longrightarrow\Psi\quad\text{in }L^2(D), Ψ n ⟶ Ψ in L 2 ( D ) ,
∥ ∇ ( Ψ n − Ψ ) ∥ L 2 ( D ) 2 = ∫ D ( ω n − ω ) ( Ψ n − Ψ ) d x ≤ 2 λ ∣ D ∣ 1 / 2 ∥ Ψ n − Ψ ∥ L 2 ( D ) ⟶ 0. \|\nabla(\Psi_n-\Psi)\|_{L^2(D)}^2=\int_D(\omega_n-\omega)(\Psi_n-\Psi)\,\mathrm{d}x\le2\lambda|D|^{1/2}\|\Psi_n-\Psi\|_{L^2(D)}\longrightarrow0. ∥∇ ( Ψ n − Ψ ) ∥ L 2 ( D ) 2 = ∫ D ( ω n − ω ) ( Ψ n − Ψ ) d x ≤ 2 λ ∣ D ∣ 1/2 ∥ Ψ n − Ψ ∥ L 2 ( D ) ⟶ 0.
Pairing this strong convergence with the bounded densities gives
2 ( E k i n ( ω n ) − E k i n ( ω ) ) = ∫ D ω n ( Ψ n − Ψ ) d x + ∫ D ( ω n − ω ) Ψ d x ⟶ 0. 2\bigl(\mathcal{E}_{\mathrm{kin}}(\omega_n)-\mathcal{E}_{\mathrm{kin}}(\omega)\bigr)=\int_D\omega_n(\Psi_n-\Psi)\,\mathrm{d}x+\int_D(\omega_n-\omega)\Psi\,\mathrm{d}x\longrightarrow0. 2 ( E kin ( ω n ) − E kin ( ω ) ) = ∫ D ω n ( Ψ n − Ψ ) d x + ∫ D ( ω n − ω ) Ψ d x ⟶ 0.
For a maximizing sequence, weak-* sequential compactness yields
ω n k ⇀ ∗ ω ∈ A λ , E k i n ( ω ) = lim k → ∞ E k i n ( ω n k ) = sup A λ E k i n . \omega_{n_k}\overset{*}{\rightharpoonup}\omega\in\mathcal{A}_\lambda,\qquad\mathcal{E}_{\mathrm{kin}}(\omega)=\lim_{k\to\infty}\mathcal{E}_{\mathrm{kin}}(\omega_{n_k})=\sup_{\mathcal{A}_\lambda}\mathcal{E}_{\mathrm{kin}}. ω n k ⇀ ∗ ω ∈ A λ , E kin ( ω ) = k → ∞ lim E kin ( ω n k ) = A λ sup E kin .
For a fixed maximizer ω \omega ω , write Ψ = G ω \Psi=\mathcal{G}\omega Ψ = G ω and use convexity of A λ \mathcal{A}_\lambda A λ to take one-sided variations towards any η ∈ A λ \eta\in\mathcal{A}_\lambda η ∈ A λ :
0 ≥ d d t ∣ t = 0 + E k i n ( ω + t ( η − ω ) ) = ∫ D Ψ ( η − ω ) d x . 0\geq\left.\frac{\mathrm{d}}{\mathrm{d}t}\right|_{t=0+}\mathcal{E}_{\mathrm{kin}}\bigl(\omega+t(\eta-\omega)\bigr)=\int_D\Psi(\eta-\omega)\,\mathrm{d}x. 0 ≥ d t d t = 0 + E kin ( ω + t ( η − ω ) ) = ∫ D Ψ ( η − ω ) d x .
Consequently, ω \omega ω also maximizes the linear functional η ↦ ∫ D Ψ η d x \eta\mapsto\int_D\Psi\eta\,\mathrm{d}x η ↦ ∫ D Ψ η d x , so the bathtub principle [22 Theorem 1.14] supplies a competitor supported on the upper level sets of Ψ \Psi Ψ : there are μ ∈ R \mu\in\mathbb{R} μ ∈ R and β ∈ [ 0 , λ ] \beta\in[0,\lambda] β ∈ [ 0 , λ ] such that
S : = { Ψ = μ } , η ∗ : = λ 1 { Ψ > μ } + β 1 S ∈ A λ . S:=\{\Psi=\mu\},\qquad\eta_*:=\lambda\mathbf{1}_{\{\Psi>\mu\}}+\beta\mathbf{1}_S\in\mathcal{A}_\lambda. S := { Ψ = μ } , η ∗ := λ 1 { Ψ > μ } + β 1 S ∈ A λ .
The mass constraint removes the constant μ \mu μ , while the pointwise sign of the remaining integrand forces saturation away from the threshold:
0 ≥ ∫ D Ψ ( η ∗ − ω ) d x = ∫ D ( Ψ − μ ) ( η ∗ − ω ) d x ≥ 0 , 0\geq\int_D\Psi(\eta_*-\omega)\,\mathrm{d}x=\int_D(\Psi-\mu)(\eta_*-\omega)\,\mathrm{d}x\geq0, 0 ≥ ∫ D Ψ ( η ∗ − ω ) d x = ∫ D ( Ψ − μ ) ( η ∗ − ω ) d x ≥ 0 ,
and hence
( λ − ω ) 1 { Ψ > μ } = 0 , ω 1 { Ψ < μ } = 0 a.e. (\lambda-\omega)\mathbf{1}_{\{\Psi>\mu\}}=0,\qquad\omega\mathbf{1}_{\{\Psi<\mu\}}=0\quad\text{a.e.} ( λ − ω ) 1 { Ψ > μ } = 0 , ω 1 { Ψ < μ } = 0 a.e.
To exclude vorticity on the possible plateau S S S , we use the Sobolev level-set identity [17 Theorem 4.4(iv)] :
f ∈ W 1 , p ( D ) , c ∈ R ⟹ 1 { f = c } ∇ f = 0 a.e. f\in W^{1,p}(D),\ c\in\mathbb{R}\quad\Longrightarrow\quad\mathbf{1}_{\{f=c\}}\nabla f=0\quad\text{a.e.} f ∈ W 1 , p ( D ) , c ∈ R ⟹ 1 { f = c } ∇ f = 0 a.e.
Indeed, for χ ∈ C c ∞ ( ( − 1 , 1 ) ) \chi\in C^\infty_c((-1,1)) χ ∈ C c ∞ (( − 1 , 1 )) , 0 ≤ χ ≤ 1 0\leq\chi\leq1 0 ≤ χ ≤ 1 , χ ( 0 ) = 1 \chi(0)=1 χ ( 0 ) = 1 ,
H ε ( t ) : = ∫ 0 t − c χ ( s / ε ) d s , H ε ( f ) ⟶ 0 in L p ( D ) , H_\varepsilon(t):=\int_0^{t-c}\chi(s/\varepsilon)\,\mathrm{d}s,\qquad H_\varepsilon(f)\longrightarrow0\quad\text{in }L^p(D), H ε ( t ) := ∫ 0 t − c χ ( s / ε ) d s , H ε ( f ) ⟶ 0 in L p ( D ) ,
∇ H ε ( f ) = χ ( ( f − c ) / ε ) ∇ f ⟶ 1 { f = c } ∇ f in L p ( D ) , \nabla H_\varepsilon(f)=\chi((f-c)/\varepsilon)\nabla f\longrightarrow\mathbf{1}_{\{f=c\}}\nabla f\quad\text{in }L^p(D), ∇ H ε ( f ) = χ (( f − c ) / ε ) ∇ f ⟶ 1 { f = c } ∇ f in L p ( D ) ,
∇ H ε ( f ) ⟶ 0 in D ′ ( D ) ⟹ 1 { f = c } ∇ f = 0 a.e. \nabla H_\varepsilon(f)\longrightarrow0\quad\text{in }\mathcal{D}'(D)\quad\Longrightarrow\quad\mathbf{1}_{\{f=c\}}\nabla f=0\quad\text{a.e.} ∇ H ε ( f ) ⟶ 0 in D ′ ( D ) ⟹ 1 { f = c } ∇ f = 0 a.e.
Since Ψ ∈ W l o c 2 , p ( D ) \Psi\in W_{\mathrm{loc}}^{2,p}(D) Ψ ∈ W loc 2 , p ( D ) , we apply this identity on relatively compact subdomains, first to Ψ \Psi Ψ and then to each ∂ i Ψ \partial_i\Psi ∂ i Ψ . Exhausting D D D eliminates its second derivatives on the plateau. With S i : = { ∂ i Ψ = 0 } S_i := \{\partial_i\Psi= 0\} S i := { ∂ i Ψ = 0 } , i = 1 , 2 i = 1,2 i = 1 , 2 ,
1 S ∇ Ψ = 0 , ∣ S ∖ S i ∣ = 0 , 1 S i ∇ ∂ i Ψ = 0 a.e. , \mathbf{1}_S\nabla\Psi= 0,\qquad|S\setminus S_i| = 0,\qquad\mathbf{1}_{S_i}\nabla\partial_i\Psi= 0\quad\text{a.e.}, 1 S ∇Ψ = 0 , ∣ S ∖ S i ∣ = 0 , 1 S i ∇ ∂ i Ψ = 0 a.e. ,
1 S D 2 Ψ = 0 ⟹ 1 S ω = − 1 S Δ Ψ = 0 a.e. \mathbf{1}_S D^2\Psi= 0 \quad\Longrightarrow\quad\mathbf{1}_S\omega= -\mathbf{1}_S\Delta\Psi= 0\quad\text{a.e.} 1 S D 2 Ψ = 0 ⟹ 1 S ω = − 1 S ΔΨ = 0 a.e.
Therefore
ω = λ 1 { Ψ > μ } a.e. , ∣ { Ψ > μ } ∣ = λ − 1 . \omega= \lambda\mathbf{1}_{\{\Psi>\mu\}}\quad\text{a.e.},\qquad|\{\Psi>\mu\}| = \lambda^{-1}. ω = λ 1 { Ψ > μ } a.e. , ∣ { Ψ > μ } ∣ = λ − 1 .
The threshold must be positive, which in turn separates the patch from ∂ D \partial D ∂ D . Indeed, the strong maximum principle gives Ψ > 0 \Psi>0 Ψ > 0 in D D D because ω ≥ 0 \omega\geq0 ω ≥ 0 and ∫ D ω = 1 \int_D\omega=1 ∫ D ω = 1 , whereas, with M : = max D ‾ Ψ M:=\max_{\overline{D}}\Psi M := max D Ψ , either alternative outside ( 0 , M ) (0,M) ( 0 , M ) contradicts the mass constraint. Therefore
0 < μ < M , { Ψ > μ } ‾ ⊂ { Ψ ≥ μ } ⊂ D . 0<\mu<M,\qquad\overline{\{\Psi>\mu\}}\subset\{\Psi\geq\mu\}\subset D. 0 < μ < M , { Ψ > μ } ⊂ { Ψ ≥ μ } ⊂ D .
Uniqueness follows from strict decrease of the upper-level-set area on ( 0 , M ) (0,M) ( 0 , M ) , since connectedness of D D D , continuity of Ψ \Psi Ψ , and Ψ ∣ ∂ D = 0 \Psi|_{\partial D}=0 Ψ ∣ ∂ D = 0 imply
0 < a < b < M ⟹ { a < Ψ < b } is nonempty and open , 0<a<b<M\quad\Longrightarrow\quad\{a<\Psi<b\}\text{ is nonempty and open}, 0 < a < b < M ⟹ { a < Ψ < b } is nonempty and open ,
∣ { Ψ > a } ∣ − ∣ { Ψ > b } ∣ ≥ ∣ { a < Ψ < b } ∣ > 0. |\{\Psi>a\}|-|\{\Psi>b\}|\geq|\{a<\Psi<b\}|>0. ∣ { Ψ > a } ∣ − ∣ { Ψ > b } ∣ ≥ ∣ { a < Ψ < b } ∣ > 0.
Thus ∣ { Ψ > μ 1 } ∣ = ∣ { Ψ > μ 2 } ∣ = λ − 1 |\{\Psi>\mu_1\}|=|\{\Psi>\mu_2\}|=\lambda^{-1} ∣ { Ψ > μ 1 } ∣ = ∣ { Ψ > μ 2 } ∣ = λ − 1 implies μ 1 = μ 2 \mu_1=\mu_2 μ 1 = μ 2 .
The superlevel-set representation also gives the steady Euler equation without requiring regularity of the patch boundary. Indeed, the Sobolev chain rule yields
∇ ⊥ ( Ψ − μ ) + = 1 { Ψ > μ } ∇ ⊥ Ψ a.e. , \nabla^\perp(\Psi-\mu)_+ = \mathbf{1}_{\{\Psi>\mu\}}\nabla^\perp\Psi\quad\text{a.e.}, ∇ ⊥ ( Ψ − μ ) + = 1 { Ψ > μ } ∇ ⊥ Ψ a.e. ,
and therefore, for every φ ∈ C c ∞ ( D ) \varphi\in C_c^\infty(D) φ ∈ C c ∞ ( D ) ,
∫ D ω ∇ ⊥ Ψ ⋅ ∇ φ d x = λ ∫ D ∇ ⊥ ( Ψ − μ ) + ⋅ ∇ φ d x = 0 , \int_D\omega\nabla^\perp\Psi\cdot\nabla\varphi\,\mathrm{d}x
=\lambda\int_D\nabla^\perp(\Psi-\mu)_+\cdot\nabla\varphi\,\mathrm{d}x=0, ∫ D ω ∇ ⊥ Ψ ⋅ ∇ φ d x = λ ∫ D ∇ ⊥ ( Ψ − μ ) + ⋅ ∇ φ d x = 0 ,
since the distributional divergence of a perpendicular gradient is zero. □
Set m = λ − 1 m=\lambda^{-1} m = λ − 1 and ρ = ω / λ \rho=\omega/\lambda ρ = ω / λ . The normalized problem is
max ρ ∈ A m E ( ρ ) , E ( ρ ) = 1 2 ∫ D ρ G ρ d x . (19) \max_{\rho\in A_m} E(\rho),\qquad E(\rho)=\frac{1}{2}\int_D\rho G\rho\,\mathrm{d}x. \tag*{(19)} ρ ∈ A m max E ( ρ ) , E ( ρ ) = 2 1 ∫ D ρGρ d x . (19)
By proposition 2.1, every energy-maximizing vortex patch has the superlevel-set representation
ρ = 1 A , ψ = G 1 A , u = ψ − q , A = { u > 0 } ⊂ D , q = μ λ / λ > 0. (20) \rho=\mathbf{1}_A,\qquad\psi=G\mathbf{1}_A,\qquad u=\psi-q,\qquad A=\{u>0\}\subset D,\qquad q=\mu\lambda/\lambda>0. \tag*{(20)} ρ = 1 A , ψ = G 1 A , u = ψ − q , A = { u > 0 } ⊂ D , q = μ λ / λ > 0. (20)
All subsequent boundaries refer to this open representative. The positive threshold and the zero Dirichlet boundary values of ψ \psi ψ give the compactness that will be used in the boundary arguments:
A ‾ ⊂ { ψ ≥ q } ⊂ D , dist ( A ‾ , ∂ D ) > 0. \overline{A}\subset\{\psi\geq q\}\subset D,\qquad\operatorname{dist}(\overline{A},\partial D)>0. A ⊂ { ψ ≥ q } ⊂ D , dist ( A , ∂ D ) > 0.
The normalized stream function satisfies
− Δ u = 1 A = 1 { u > 0 } in D , u = − q on ∂ D . (21) -\Delta u=\mathbf{1}_A=\mathbf{1}_{\{u>0\}}\quad\text{in }D,\qquad u=-q\quad\text{on }\partial D. \tag*{(21)} − Δ u = 1 A = 1 { u > 0 } in D , u = − q on ∂ D . (21)
where the boundary value is attained continuously and u + q ∈ H 0 1 ( D ) u+q\in H^1_0(D) u + q ∈ H 0 1 ( D ) . By (11) and (12) ,
u ∈ C ( D ‾ ) ∩ W l o c 2 , p ( D ) ∩ C l o c 1 , α ( D ) ( 1 < p < ∞ , 0 < α < 1 ) . (22) u\in C(\overline{D})\cap W_{\mathrm{loc}}^{2,p}(D)\cap C_{\mathrm{loc}}^{1,\alpha}(D)\qquad(1<p<\infty,\ 0<\alpha<1). \tag*{(22)} u ∈ C ( D ) ∩ W loc 2 , p ( D ) ∩ C loc 1 , α ( D ) ( 1 < p < ∞ , 0 < α < 1 ) . (22)
We shall use the following modulus to control the gradient near a possible singular point on the free boundary, without assuming regularity of ∂ A \partial A ∂ A . At x 0 ∈ ∂ A x_0\in\partial A x 0 ∈ ∂ A with ∇ u ( x 0 ) = 0 \nabla u(x_0)=0 ∇ u ( x 0 ) = 0 ,
∣ ∇ u ( x ) ∣ ≤ C ∣ x − x 0 ∣ ( 1 + log r 0 ∣ x − x 0 ∣ ) , 0 < ∣ x − x 0 ∣ < r 0 / 2 , (23) |\nabla u(x)|\le C|x-x_0|\left(1+\log\frac{r_0}{|x-x_0|}\right),\qquad0<|x-x_0|<r_0/2, \tag*{(23)} ∣∇ u ( x ) ∣ ≤ C ∣ x − x 0 ∣ ( 1 + log ∣ x − x 0 ∣ r 0 ) , 0 < ∣ x − x 0 ∣ < r 0 /2 , (23)
where B r 0 ( x 0 ) ⊂ D B_{r_0}(x_0)\subset D B r 0 ( x 0 ) ⊂ D . To isolate the singular part of the potential, choose χ ∈ C c ∞ ( B r 0 ( x 0 ) ) \chi\in C^\infty_c(B_{r_0}(x_0)) χ ∈ C c ∞ ( B r 0 ( x 0 )) , 0 ≤ χ ≤ 1 0\le\chi\le1 0 ≤ χ ≤ 1 , χ = 1 \chi=1 χ = 1 on B 3 r 0 / 4 ( x 0 ) B_{3r_0/4}(x_0) B 3 r 0 /4 ( x 0 ) , and write
v = Φ ∗ ( χ 1 A ) , u = v + h , Δ h = 0 in B 3 r 0 / 4 ( x 0 ) . v=\Phi*(\chi1_A),\qquad u=v+h,\qquad\Delta h=0\ \text{ in }B_{3r_0/4}(x_0). v = Φ ∗ ( χ 1 A ) , u = v + h , Δ h = 0 in B 3 r 0 /4 ( x 0 ) .
The harmonic remainder has a Lipschitz gradient on the smaller ball. For the potential, split at twice the separation δ = ∣ x − x 0 ∣ < r 0 / 2 \delta=|x-x_0|<r_0/2 δ = ∣ x − x 0 ∣ < r 0 /2 : use integrability of the gradient kernel in the near region and the mean-value theorem in the far region. The bounds ∣ ∇ Φ ( z ) ∣ ≤ C ∣ z ∣ − 1 |\nabla\Phi(z)|\le C|z|^{-1} ∣∇Φ ( z ) ∣ ≤ C ∣ z ∣ − 1 and ∣ D 2 Φ ( z ) ∣ ≤ C ∣ z ∣ − 2 |D^2\Phi(z)|\le C|z|^{-2} ∣ D 2 Φ ( z ) ∣ ≤ C ∣ z ∣ − 2 give
∣ ∇ v ( x ) − ∇ v ( x 0 ) ∣ ≤ C ∫ B 2 δ ( x 0 ) ( 1 ∣ x − y ∣ + 1 ∣ x 0 − y ∣ ) d y + C δ ∫ B r 0 ( x 0 ) ∖ B 2 δ ( x 0 ) d y ∣ x 0 − y ∣ 2 ≤ C δ + C δ ∫ 2 δ r 0 d t t ≤ C δ ( 1 + log r 0 δ ) , \begin{aligned}
|\nabla v(x)-\nabla v(x_0)|
&\le C\int_{B_{2\delta}(x_0)}\left(\frac{1}{|x-y|}+\frac{1}{|x_0-y|}\right)\,dy\\
&\quad+C\delta\int_{B_{r_0}(x_0)\setminus B_{2\delta}(x_0)}\frac{dy}{|x_0-y|^2}\\
&\le C\delta+C\delta\int_{2\delta}^{r_0}\frac{dt}{t}
\le C\delta\left(1+\log\frac{r_0}{\delta}\right),
\end{aligned} ∣∇ v ( x ) − ∇ v ( x 0 ) ∣ ≤ C ∫ B 2 δ ( x 0 ) ( ∣ x − y ∣ 1 + ∣ x 0 − y ∣ 1 ) d y + C δ ∫ B r 0 ( x 0 ) ∖ B 2 δ ( x 0 ) ∣ x 0 − y ∣ 2 d y ≤ C δ + C δ ∫ 2 δ r 0 t d t ≤ C δ ( 1 + log δ r 0 ) ,
∣ ∇ h ( x ) − ∇ h ( x 0 ) ∣ ≤ ∥ D 2 h ∥ L ∞ ( B r 0 / 2 ( x 0 ) ) δ ≤ C δ . |\nabla h(x)-\nabla h(x_0)|\le\|D^2h\|_{L^\infty(B_{r_0/2}(x_0))}\delta\le C\delta. ∣∇ h ( x ) − ∇ h ( x 0 ) ∣ ≤ ∥ D 2 h ∥ L ∞ ( B r 0 /2 ( x 0 )) δ ≤ C δ .
Note that ∇ u ( x 0 ) = 0 \nabla u(x_0)=0 ∇ u ( x 0 ) = 0 , equation (23) is then proved.
The following lemma relates the topological boundary to the two phases without assuming any regularity condition on the boundary, and then establishes local analyticity wherever the stream-function gradient is nonzero. Together, these conclusions reduce the remaining regularity problem to critical points of u u u on ∂ A \partial A ∂ A .
Lemma 2.2. For an energy-maximizing vortex patch 1 A 1_A 1 A represented by (20) , with supports taken relative to D D D ,
∂ A = supp D ( 1 A d x ) ∩ supp D ( 1 D ∖ A d x ) . (24) \partial A=\operatorname{supp}_D(1_A\,dx)\cap\operatorname{supp}_D(1_{D\setminus A}\,dx). \tag*{(24)} ∂ A = supp D ( 1 A d x ) ∩ supp D ( 1 D ∖ A d x ) . (24) Equivalently, for x ∈ D x\in D x ∈ D ,
x ∈ ∂ A ⟺ { ∣ A ∩ B r ( x ) ∣ > 0 , ∣ B r ( x ) ∖ A ∣ > 0 for every 0 < r < dist ( x , ∂ D ) . x\in\partial A\quad\Longleftrightarrow\quad
\begin{cases}
|A\cap B_r(x)|>0,\\
|B_r(x)\setminus A|>0
\end{cases}
\qquad\text{for every }0<r<\operatorname{dist}(x,\partial D). x ∈ ∂ A ⟺ { ∣ A ∩ B r ( x ) ∣ > 0 , ∣ B r ( x ) ∖ A ∣ > 0 for every 0 < r < dist ( x , ∂ D ) . If x 0 ∈ ∂ A x_0\in\partial A x 0 ∈ ∂ A and ∇ u ( x 0 ) ≠ 0 \nabla u(x_0)\ne0 ∇ u ( x 0 ) = 0 , then ∂ A \partial A ∂ A is a single real-analytic arc near x 0 x_0 x 0 .
Proof. The boundary identity uses openness of A = { u > 0 } A=\{u>0\} A = { u > 0 } together with the equation for u u u . If x ∈ D ∖ ∂ A x\in D\setminus\partial A x ∈ D ∖ ∂ A , a sufficiently small ball lies in one phase, excluding x x x from the intersection of the two supports. Conversely, given x ∈ ∂ A x \in\partial A x ∈ ∂ A and B r ( x ) ⊂ D B_r(x) \subset D B r ( x ) ⊂ D , choose y ∈ A ∩ B r / 2 ( x ) y \in A \cap B_{r/2}(x) y ∈ A ∩ B r /2 ( x ) ; openness then supplies ε ∈ ( 0 , r / 2 ) \varepsilon\in(0,r/2) ε ∈ ( 0 , r /2 ) such that
B ε ( y ) ⊂ A ∩ B r ( x ) ⟹ ∣ A ∩ B r ( x ) ∣ ≥ π ε 2 > 0. B_\varepsilon(y) \subset A \cap B_r(x) \quad\Longrightarrow\quad|A \cap B_r(x)| \ge\pi\varepsilon^2 > 0. B ε ( y ) ⊂ A ∩ B r ( x ) ⟹ ∣ A ∩ B r ( x ) ∣ ≥ π ε 2 > 0.
No regularity of ∂ A \partial A ∂ A is needed for this inclusion.
To prove positivity of the other phase, suppose ∣ B r ( x ) ∖ A ∣ = 0 |B_r(x) \setminus A| = 0 ∣ B r ( x ) ∖ A ∣ = 0 . Continuity of u u u and u > 0 u > 0 u > 0 a.e. in the ball imply
u ≥ 0 on B r ( x ) ‾ , − Δ u = 1 in B r ( x ) , u ( x ) = 0. u \ge0 \quad\text{on } \overline{B_r(x)}, \qquad-\Delta u = 1 \quad\text{in } B_r(x), \qquad u(x) = 0. u ≥ 0 on B r ( x ) , − Δ u = 1 in B r ( x ) , u ( x ) = 0.
The disk torsion function
v ( z ) : = r 2 − ∣ z − x ∣ 2 4 , − Δ v = 1 in B r ( x ) , v = 0 on ∂ B r ( x ) v(z) := \frac{r^2-|z-x|^2}{4}, \qquad-\Delta v = 1 \quad\text{in } B_r(x), \qquad v = 0 \quad\text{on } \partial B_r(x) v ( z ) := 4 r 2 − ∣ z − x ∣ 2 , − Δ v = 1 in B r ( x ) , v = 0 on ∂ B r ( x )
then gives a contradiction by the maximum principle:
− Δ ( u − v ) = 0 in B r ( x ) , u − v ≥ 0 on ∂ B r ( x ) , -\Delta(u-v)=0 \quad\text{in } B_r(x), \qquad u-v \ge0 \quad\text{on } \partial B_r(x), − Δ ( u − v ) = 0 in B r ( x ) , u − v ≥ 0 on ∂ B r ( x ) ,
u ≥ v in B r ( x ) ⟹ 0 = u ( x ) ≥ v ( x ) = r 2 4 > 0. u \ge v \quad\text{in } B_r(x) \quad\Longrightarrow\quad0=u(x)\ge v(x)=\frac{r^2}{4}>0. u ≥ v in B r ( x ) ⟹ 0 = u ( x ) ≥ v ( x ) = 4 r 2 > 0.
This establishes (24) , ensuring that both phases have positive measure on every fixed ball centered at the boundary; the uniform half-density estimate will follow later from corollary 4.2.
At a noncritical point, the implicit function theorem identifies the boundary with a single C 1 , α C^{1,\alpha} C 1 , α zero-level arc, whose regularity can be improved to real analyticity by applying [13 Lemma 3] to
Δ u = h ( u , ∇ u ) , h ( s , p ) = − 1 { s > 0 } , ∂ p α h = 0 ( ∣ α ∣ ≥ 1 ) . \Delta u=h(u,\nabla u), \qquad h(s,\mathbf{p})=-\mathbf{1}_{\{s>0\}}, \qquad\partial_{\mathbf{p}}^\alpha h=0 \quad(|\alpha|\ge1). Δ u = h ( u , ∇ u ) , h ( s , p ) = − 1 { s > 0 } , ∂ p α h = 0 ( ∣ α ∣ ≥ 1 ) .
The hypotheses of that lemma hold since ∥ h ∥ L ∞ = 1 \lVert h\rVert_{L^\infty}=1 ∥ h ∥ L ∞ = 1 and h ( s , ⋅ ) h(s,\cdot) h ( s , ⋅ ) is entire; analyticity in the gradient variable is sufficient, despite the jump in s s s . □
Define
R : = ∂ A ∩ { ∣ ∇ u ∣ > 0 } . (25) R:=\partial A\cap\{|\nabla u|>0\}. \tag*{(25)} R := ∂ A ∩ { ∣∇ u ∣ > 0 } . (25)
By lemma 2.2, every x ∈ R x\in R x ∈ R has a neighborhood U ⊂ D U\subset D U ⊂ D and a real-analytic embedding γ : ( − 1 , 1 ) → U \gamma:(-1,1)\to U γ : ( − 1 , 1 ) → U such that
inf U ∣ ∇ u ∣ > 0 , U ∩ ∂ A = U ∩ R = γ ( ( − 1 , 1 ) ) . \inf_U|\nabla u|>0, \qquad U\cap\partial A=U\cap R=\gamma((-1,1)). U inf ∣∇ u ∣ > 0 , U ∩ ∂ A = U ∩ R = γ (( − 1 , 1 )) .
Thus R R R is an embedded real-analytic one-dimensional manifold without boundary, and U U U meets exactly one component of R R R .
Variational estimates and boundary geometry
Throughout, 1 A \mathbf{1}_A 1 A is an energy-maximizing vortex patch, and A = { u > 0 } A=\{u>0\} A = { u > 0 } is its open vortex core, as chosen in § 2 .
We begin by quantifying how closely the patch is approximated by a half-plane near a noncritical boundary point, keeping explicit the dependence on ∣ ∇ u ( p ) ∣ |\nabla u(p)| ∣∇ u ( p ) ∣ so that the density estimate can later be made uniform once critical boundary points have been excluded.
Lemma 3.1. Let 0 < α < 1 0 < \alpha< 1 0 < α < 1 , p ∈ ∂ A \mathbf{p} \in\partial A p ∈ ∂ A , and
g p : = ∣ ∇ u ( p ) ∣ > 0 , e p : = ∇ u ( p ) / g p , B r 0 ( p ) ⊂ D , [ ∇ u ] C 0 , α ( B r 0 ( p ) ) ≤ L . g_{\mathbf{p}} := |\nabla u(\mathbf{p})| > 0,\qquad\mathbf{e}_{\mathbf{p}} := \nabla u(\mathbf{p})/g_{\mathbf{p}},\qquad B_{r_0}(\mathbf{p}) \subset D,\qquad[\nabla u]_{C^{0,\alpha}(B_{r_0}(\mathbf{p}))} \leq L. g p := ∣∇ u ( p ) ∣ > 0 , e p := ∇ u ( p ) / g p , B r 0 ( p ) ⊂ D , [ ∇ u ] C 0 , α ( B r 0 ( p )) ≤ L . For 0 < r < r 0 0 < r < r_0 0 < r < r 0 , set
A p , r : = { y ∈ B 1 : p + r y ∈ A } , H p + : = { y ∈ B 1 : e p ⋅ y > 0 } . A_{\mathbf{p},r} := \{\mathbf{y} \in B_1 : \mathbf{p} + r\mathbf{y} \in A\},\qquad H_{\mathbf{p}}^+ := \{\mathbf{y} \in B_1 : \mathbf{e}_{\mathbf{p}} \cdot\mathbf{y} > 0\}. A p , r := { y ∈ B 1 : p + r y ∈ A } , H p + := { y ∈ B 1 : e p ⋅ y > 0 } . Then
∣ A p , r △ H p + ∣ ≤ 4 L g p r α , ∣ θ A ( p , r ) − 1 2 ∣ ≤ 4 L π g p r α . (26) |A_{\mathbf{p},r} \mathbin{\triangle} H_{\mathbf{p}}^+| \leq\frac{4L}{g_{\mathbf{p}}}r^\alpha,\qquad\left|\theta_A(\mathbf{p},r) - \frac{1}{2}\right| \leq\frac{4L}{\pi g_{\mathbf{p}}}r^\alpha. \tag*{(26)} ∣ A p , r △ H p + ∣ ≤ g p 4 L r α , θ A ( p , r ) − 2 1 ≤ π g p 4 L r α . (26) Proof. Rescaling to B 1 B_1 B 1 reduces the density estimate to comparison with the half-disk determined by ∇ u ( p ) \nabla u(\mathbf{p}) ∇ u ( p ) . Since multiplication by r − 1 > 0 r^{-1} > 0 r − 1 > 0 preserves the positivity set, Taylor’s formula and u ( p ) = 0 u(\mathbf{p}) = 0 u ( p ) = 0 give
∣ u ( p + r y ) r − ∇ u ( p ) ⋅ y ∣ = ∣ ∫ 0 1 [ ∇ u ( p + t r y ) − ∇ u ( p ) ] ⋅ y d t ∣ \left|\frac{u(\mathbf{p}+r\mathbf{y})}{r} - \nabla u(\mathbf{p}) \cdot\mathbf{y}\right| = \left|\int_0^1 [\nabla u(\mathbf{p}+tr\mathbf{y})-\nabla u(\mathbf{p})]\cdot\mathbf{y}\,\mathrm{d}t\right| r u ( p + r y ) − ∇ u ( p ) ⋅ y = ∫ 0 1 [ ∇ u ( p + t r y ) − ∇ u ( p )] ⋅ y d t ≤ L 1 + α r α ∣ y ∣ 1 + α ≤ L r α ( y ∈ B 1 ) . \leq\frac{L}{1+\alpha}r^\alpha|\mathbf{y}|^{1+\alpha} \leq Lr^\alpha\qquad(\mathbf{y} \in B_1). ≤ 1 + α L r α ∣ y ∣ 1 + α ≤ L r α ( y ∈ B 1 ) . The signs of u ( p + r y ) u(\mathbf{p}+r\mathbf{y}) u ( p + r y ) and e p ⋅ y \mathbf{e}_{\mathbf{p}}\cdot\mathbf{y} e p ⋅ y agree wherever the magnitude of the linear term exceeds the error. Thus, with a = L r α / g p a = Lr^\alpha/g_{\mathbf{p}} a = L r α / g p ,
A p , r △ H p + ⊂ S a : = { y ∈ B 1 : ∣ e p ⋅ y ∣ ≤ a } , A_{\mathbf{p},r} \mathbin{\triangle} H_{\mathbf{p}}^+ \subset S_a := \{\mathbf{y} \in B_1 : |\mathbf{e}_{\mathbf{p}}\cdot\mathbf{y}| \leq a\}, A p , r △ H p + ⊂ S a := { y ∈ B 1 : ∣ e p ⋅ y ∣ ≤ a } , ∣ S a ∣ = ∫ − min { a , 1 } min { a , 1 } 2 1 − s 2 d s ≤ 4 a , |S_a| = \int_{-\min\{a,1\}}^{\min\{a,1\}} 2\sqrt{1-s^2}\,\mathrm{d}s \leq4a, ∣ S a ∣ = ∫ − m i n { a , 1 } m i n { a , 1 } 2 1 − s 2 d s ≤ 4 a , ∣ θ A ( p , r ) − 1 2 ∣ = 1 π ∣ A p , r ∣ − ∣ H p + ∣ ≤ 1 π ∣ A p , r △ H p + ∣ ≤ 4 a π . \left|\theta_A(\mathbf{p},r)-\frac{1}{2}\right| = \frac{1}{\pi}|A_{\mathbf{p},r}|-|H_{\mathbf{p}}^+| \leq\frac{1}{\pi}|A_{\mathbf{p},r}\mathbin{\triangle}H_{\mathbf{p}}^+| \leq\frac{4a}{\pi}. θ A ( p , r ) − 2 1 = π 1 ∣ A p , r ∣ − ∣ H p + ∣ ≤ π 1 ∣ A p , r △ H p + ∣ ≤ π 4 a .
To turn energy maximization into estimates on the boundary geometry, we record an exact inequality for competitors of the same mass, which will then be applied to area-preserving boundary variations.
Proposition 3.2. For every η ∈ L ∞ ( D ) \eta\in L^\infty(D) η ∈ L ∞ ( D ) satisfying 0 ≤ η ≤ 1 0 \leq\eta\leq1 0 ≤ η ≤ 1 and ∫ D η d x = ∣ A ∣ \int_D \eta\,\mathrm{d}x = |A| ∫ D η d x = ∣ A ∣ ,
∫ D u ( η − 1 A ) d x + 1 2 ∥ η − 1 A ∥ H ˙ D 1 2 ≤ 0. (27) \int_D u(\eta-\mathbf{1}_A)\,\mathrm{d}x + \frac{1}{2}\|\eta-\mathbf{1}_A\|_{\dot{H}^{1}_D}^2 \leq0. \tag*{(27)} ∫ D u ( η − 1 A ) d x + 2 1 ∥ η − 1 A ∥ H ˙ D 1 2 ≤ 0. (27) Proof. Writing h = η − 1 A h=\eta-\mathbf{1}_A h = η − 1 A , the quadratic structure of the energy and self-adjointness of G \mathcal{G} G give an expansion with no remainder, while the mass constraint ∫ D h = 0 \int_D h=0 ∫ D h = 0 eliminates the constant q q q in G 1 A = u + q \mathcal{G}\mathbf{1}_A=u+q G 1 A = u + q :
0 ≥ E ( η ) − E ( 1 A ) = ∫ D ( G 1 A ) h d x + 1 2 ⟨ h , G h ⟩ = ∫ D u h d x + 1 2 ∥ h ∥ H ˙ D 1 2 . (28) 0 \geq\mathcal{E}(\eta)-\mathcal{E}(\mathbf{1}_A)=\int_D(\mathcal{G}\mathbf{1}_A)h\,\mathrm{d}x+\frac{1}{2}\langle h,\mathcal{G}h\rangle=\int_D uh\,\mathrm{d}x+\frac{1}{2}\|h\|_{\dot{H}^{1}_D}^2. \tag*{(28)} 0 ≥ E ( η ) − E ( 1 A ) = ∫ D ( G 1 A ) h d x + 2 1 ⟨ h , G h ⟩ = ∫ D u h d x + 2 1 ∥ h ∥ H ˙ D 1 2 . (28)
Applying this inequality to normal variations supported on finitely many regular boundary pieces yields an estimate that couples all the pieces through a single total-area constraint. A regular arc is a compact embedded subarc of R \mathcal{R} R of positive length; a closed regular curve is a simple closed subcurve of R \mathcal{R} R . Let
Γ = ⋃ ℓ = 1 M Γ ℓ , g = ∣ ∇ u ∣ Γ , \Gamma= \bigcup_{\ell=1}^{M} \Gamma_\ell,\qquad g = |\nabla u|_{\Gamma}, Γ = ℓ = 1 ⋃ M Γ ℓ , g = ∣∇ u ∣ Γ ,
be a nonempty finite disjoint union of such pieces. The normal ν \nu ν points out of A A A ; thus
min Γ g > 0 , ν = − ∇ u ∣ ∇ u ∣ Γ , ∂ ν u = − g . \min_{\Gamma} g > 0,\qquad\nu= -\frac{\nabla u}{|\nabla u|}_{\Gamma},\qquad\partial_\nu u = -g. Γ min g > 0 , ν = − ∣∇ u ∣ ∇ u Γ , ∂ ν u = − g .
Proposition 3.3. For every ϕ ∈ L 2 ( Γ ) \phi\in L^2(\Gamma) ϕ ∈ L 2 ( Γ ) with
∫ Γ ϕ d H 1 = 0 , (29) \int_{\Gamma} \phi\,\mathrm{d}\mathcal{H}^1 = 0, \tag*{(29)} ∫ Γ ϕ d H 1 = 0 , (29)
one has
\left\|\phi\mathcal{H}^1\mathbin{\vrule height 1.4ex depth -0.3ex width 0.07ex}_{\Gamma}\right\|_{\dot{H}^{-1}_{D}}^2 \le\int_{\Gamma} g\phi^2\,\mathrm{d}\mathcal{H}^1. \tag*{(30)}
Proof. Fix 0 < α < 1 0<\alpha<1 0 < α < 1 and initially take ϕ \phi ϕ smooth, supported away from arc endpoints, and periodic on closed curves. After slightly extending each arc within R \mathcal{R} R , lemma 2.2 provides a neighborhood of each point of Γ \Gamma Γ in which the boundary is a single analytic arc. Compactness and pairwise disjointness then give disjoint tubes, without requiring regularity of the remaining boundary, and in arclength coordinates we have
F ℓ ( s , r ) = γ ℓ ( s ) + r ν ℓ ( s ) , A = { r < 0 } , J ℓ ( s , r ) = 1 + κ ℓ ( s ) r ∈ [ 1 / 2 , 3 / 2 ] ( ∣ r ∣ < ε ) , F_\ell(s,r)=\gamma_\ell(s)+r\nu_\ell(s),\qquad A=\{r<0\},\qquad J_\ell(s,r)=1+\kappa_\ell(s)r\in[1/2,3/2]\quad(|r|<\varepsilon), F ℓ ( s , r ) = γ ℓ ( s ) + r ν ℓ ( s ) , A = { r < 0 } , J ℓ ( s , r ) = 1 + κ ℓ ( s ) r ∈ [ 1/2 , 3/2 ] ( ∣ r ∣ < ε ) ,
where κ ℓ \kappa_\ell κ ℓ is the signed curvature determined by the displayed Jacobian. Although (29) cancels the first-order area change, an additional correction is needed to preserve the area exactly; for this purpose, choose smooth ζ ℓ \zeta_\ell ζ ℓ with the same support conventions and
B 0 : = ∑ ℓ ∫ Γ ℓ ζ ℓ d H 1 ≠ 0 , b ℓ ( t , c ) : = t ϕ ℓ + c ζ ℓ . B_0:=\sum_\ell\int_{\Gamma_\ell}\zeta_\ell\,\mathrm{d}\mathcal{H}^1\ne0,\qquad b_\ell(t,c):=t\phi_\ell+c\zeta_\ell. B 0 := ℓ ∑ ∫ Γ ℓ ζ ℓ d H 1 = 0 , b ℓ ( t , c ) := t ϕ ℓ + c ζ ℓ .
The signed area change is
V ( t , c ) = ∑ ℓ ∫ Γ ℓ ( b ℓ ( t , c ) + κ ℓ 2 b ℓ ( t , c ) 2 ) d H 1 . (31) \mathcal{V}(t,c)=\sum_\ell\int_{\Gamma_\ell}\left(b_\ell(t,c)+\frac{\kappa_\ell}{2}b_\ell(t,c)^2\right)\,\mathrm{d}\mathcal{H}^1. \tag*{(31)} V ( t , c ) = ℓ ∑ ∫ Γ ℓ ( b ℓ ( t , c ) + 2 κ ℓ b ℓ ( t , c ) 2 ) d H 1 . (31)
The implicit function theorem gives
V ( 0 , 0 ) = V t ( 0 , 0 ) = 0 , V c ( 0 , 0 ) = B 0 ≠ 0 , \mathcal{V}(0,0)=\mathcal{V}_t(0,0)=0,\qquad\mathcal{V}_c(0,0)=B_0\ne0, V ( 0 , 0 ) = V t ( 0 , 0 ) = 0 , V c ( 0 , 0 ) = B 0 = 0 ,
V ( t , c ( t ) ) = 0 , c ( 0 ) = c ′ ( 0 ) = 0 , c ′ ′ ( 0 ) = − 1 B 0 ∑ ℓ ∫ Γ ℓ κ ℓ ϕ ℓ 2 d H 1 . \mathcal{V}(t,c(t))=0,\qquad c(0)=c'(0)=0,\qquad c''(0)=-\frac{1}{B_0}\sum_\ell\int_{\Gamma_\ell}\kappa_\ell\phi_\ell^2\,\mathrm{d}\mathcal{H}^1. V ( t , c ( t )) = 0 , c ( 0 ) = c ′ ( 0 ) = 0 , c ′′ ( 0 ) = − B 0 1 ℓ ∑ ∫ Γ ℓ κ ℓ ϕ ℓ 2 d H 1 .
Hence
c ( t ) = − t 2 2 B 0 ∑ ℓ ∫ Γ ℓ κ ℓ ϕ ℓ 2 d H 1 + O ( t 3 ) . (32) c(t)=-\frac{t^2}{2B_0}\sum_\ell\int_{\Gamma_\ell}\kappa_\ell\phi_\ell^2\,\mathrm{d}\mathcal{H}^1+O(t^3). \tag*{(32)} c ( t ) = − 2 B 0 t 2 ℓ ∑ ∫ Γ ℓ κ ℓ ϕ ℓ 2 d H 1 + O ( t 3 ) . (32)
With b ℓ , t = b ℓ ( t , c ( t ) ) b_{\ell,t}=b_\ell(t,c(t)) b ℓ , t = b ℓ ( t , c ( t )) , define A t A_t A t by replacing r < 0 r<0 r < 0 with r < b ℓ , t ( s ) r<b_{\ell,t}(s) r < b ℓ , t ( s ) in the tubes and leaving A A A unchanged elsewhere. For ∣ t ∣ < t 0 |t|<t_0 ∣ t ∣ < t 0 , the displacements vanish near the tube ends and satisfy
∥ b ℓ , t ∥ ∞ ≤ C 0 ∣ t ∣ < ε / 2 , ∣ A t ∣ = ∣ A ∣ . \|b_{\ell,t}\|_\infty\le C_0|t|<\varepsilon/2,\qquad|A_t|=|A|. ∥ b ℓ , t ∥ ∞ ≤ C 0 ∣ t ∣ < ε /2 , ∣ A t ∣ = ∣ A ∣.
Thus the correction preserves the first-order normal displacement ϕ \phi ϕ while enforcing the area constraint for every small t t t .
Put h t = 1 A t − 1 A h_t=\mathbf{1}_{A_t}-\mathbf{1}_A h t = 1 A t − 1 A and \mu_\phi=\phi\mathcal{H}^1\mathbin{\vrule height 1.4ex depth -0.3ex width 0.07ex\vrule height 0.07ex depth -0.02ex width 0.8ex}\Gamma . We identify the quadratic energy term by proving strong convergence of t − 1 h t t^{-1}h_t t − 1 h t to μ ϕ \mu_\phi μ ϕ in H − 1 ( D ) H^{-1}(D) H − 1 ( D ) ; weak convergence alone would give only lower semicontinuity of the squared norm. For z ∈ H 0 1 ( D ) z\in H_0^1(D) z ∈ H 0 1 ( D ) , write z ℓ = z ∘ F ℓ z_\ell=z\circ F_\ell z ℓ = z ∘ F ℓ . Signed integration gives
⟨ h t t , z ⟩ = ∑ ℓ ∫ Γ ℓ 1 t ∫ 0 b ℓ , t ( s ) z ℓ ( s , r ) ( 1 + κ ℓ r ) d r d s . (33) \left\langle\frac{h_t}{t},z\right\rangle
=\sum_\ell\int_{\Gamma_\ell}\frac{1}{t}\int_0^{b_{\ell,t}(s)}z_\ell(s,r)(1+\kappa_\ell r)\,dr\,ds. \tag*{(33)} ⟨ t h t , z ⟩ = ℓ ∑ ∫ Γ ℓ t 1 ∫ 0 b ℓ , t ( s ) z ℓ ( s , r ) ( 1 + κ ℓ r ) d r d s . (33)
The Sobolev slice estimate is
∣ z ℓ ( s , r ) − z ℓ ( s , 0 ) ∣ ≤ ∣ r ∣ 1 / 2 ( ∫ − ∣ r ∣ ∣ r ∣ ∣ ∂ r z ℓ ( s , ρ ) ∣ 2 d ρ ) 1 / 2 . |z_\ell(s,r)-z_\ell(s,0)|\le|r|^{1/2}\left(\int_{-|r|}^{|r|}|\partial_r z_\ell(s,\rho)|^2\,d\rho\right)^{1/2}. ∣ z ℓ ( s , r ) − z ℓ ( s , 0 ) ∣ ≤ ∣ r ∣ 1/2 ( ∫ − ∣ r ∣ ∣ r ∣ ∣ ∂ r z ℓ ( s , ρ ) ∣ 2 d ρ ) 1/2 .
If I ℓ I_\ell I ℓ is the arclength interval, with endpoints identified for a closed curve, define
U δ : = ⋃ ℓ F ℓ ( I ℓ × ( − δ , δ ) ) , Q δ ( z ) 2 : = ∑ ℓ ∫ I ℓ ∫ − δ δ ∣ ∂ r z ℓ ∣ 2 d r d s . U_\delta:=\bigcup_\ell F_\ell(I_\ell\times(-\delta,\delta)),\qquad
Q_\delta(z)^2:=\sum_\ell\int_{I_\ell}\int_{-\delta}^{\delta}|\partial_rz_\ell|^2\,dr\,ds. U δ := ℓ ⋃ F ℓ ( I ℓ × ( − δ , δ )) , Q δ ( z ) 2 := ℓ ∑ ∫ I ℓ ∫ − δ δ ∣ ∂ r z ℓ ∣ 2 d r d s .
Then
∣ U δ ∣ ≤ 3 δ H 1 ( Γ ) , Q δ ( z ) 2 ≤ 2 ∥ ∇ z ∥ L 2 ( D ) 2 . |U_\delta|\le3\delta\mathcal{H}^1(\Gamma),\qquad
Q_\delta(z)^2\le2\|\nabla z\|_{L^2(D)}^2. ∣ U δ ∣ ≤ 3 δ H 1 ( Γ ) , Q δ ( z ) 2 ≤ 2∥∇ z ∥ L 2 ( D ) 2 .
Subtracting ⟨ μ ϕ , z ⟩ \langle\mu_\phi,z\rangle ⟨ μ ϕ , z ⟩ in (33) yields I 1 + I 2 + I 3 I_1+I_2+I_3 I 1 + I 2 + I 3 , where
I 1 = ∑ ℓ ∫ Γ ℓ c ( t ) t ζ ℓ z ℓ ( s , 0 ) d s , I_1=\sum_\ell\int_{\Gamma_\ell}\frac{c(t)}{t}\zeta_\ell z_\ell(s,0)\,ds, I 1 = ℓ ∑ ∫ Γ ℓ t c ( t ) ζ ℓ z ℓ ( s , 0 ) d s ,
I 2 = ∑ ℓ ∫ Γ ℓ 1 t ∫ 0 b ℓ , t ( z ℓ ( s , r ) − z ℓ ( s , 0 ) ) d r d s , I_2=\sum_\ell\int_{\Gamma_\ell}\frac{1}{t}\int_0^{b_{\ell,t}}\bigl(z_\ell(s,r)-z_\ell(s,0)\bigr)\,dr\,ds, I 2 = ℓ ∑ ∫ Γ ℓ t 1 ∫ 0 b ℓ , t ( z ℓ ( s , r ) − z ℓ ( s , 0 ) ) d r d s ,
I 3 = ∑ ℓ ∫ Γ ℓ 1 t ∫ 0 b ℓ , t κ ℓ r z ℓ ( s , r ) d r d s . I_3=\sum_\ell\int_{\Gamma_\ell}\frac{1}{t}\int_0^{b_{\ell,t}}\kappa_\ell r z_\ell(s,r)\,dr\,ds. I 3 = ℓ ∑ ∫ Γ ℓ t 1 ∫ 0 b ℓ , t κ ℓ r z ℓ ( s , r ) d r d s .
The trace, slice, and Poincaré inequalities imply
∣ I 1 ∣ ≤ C ∣ c ( t ) ∣ ∣ t ∣ ∥ ∇ z ∥ L 2 ( D ) , ∣ I 2 ∣ ≤ C ∣ t ∣ 1 / 2 Q C 0 ∣ t ∣ ( z ) ≤ C ∣ t ∣ 1 / 2 ∥ ∇ z ∥ L 2 ( D ) , ∣ I 3 ∣ ≤ C ∥ z ∥ L 1 ( U C 0 ∣ t ∣ ) ≤ C ∣ t ∣ 1 / 2 ∥ ∇ z ∥ L 2 ( D ) . (34) \begin{aligned}
|I_1|&\le C\frac{|c(t)|}{|t|}\|\nabla z\|_{L^2(D)},\\
|I_2|&\le C|t|^{1/2}Q_{C_0|t|}(z)\le C|t|^{1/2}\|\nabla z\|_{L^2(D)},\\
|I_3|&\le C\|z\|_{L^1(U_{C_0|t|})}\le C|t|^{1/2}\|\nabla z\|_{L^2(D)}. \tag*{(34)}
\end{aligned} ∣ I 1 ∣ ∣ I 2 ∣ ∣ I 3 ∣ ≤ C ∣ t ∣ ∣ c ( t ) ∣ ∥∇ z ∥ L 2 ( D ) , ≤ C ∣ t ∣ 1/2 Q C 0 ∣ t ∣ ( z ) ≤ C ∣ t ∣ 1/2 ∥∇ z ∥ L 2 ( D ) , ≤ C ∥ z ∥ L 1 ( U C 0 ∣ t ∣ ) ≤ C ∣ t ∣ 1/2 ∥∇ z ∥ L 2 ( D ) . (34)
Here C C C is independent of t , z t,z t , z . Consequently,
∥ t − 1 h t − μ ϕ ∥ H ˙ D − 1 ≤ C ( ∣ t ∣ 1 / 2 + ∣ c ( t ) ∣ ∣ t ∣ ) ≤ C ∣ t ∣ 1 / 2 ⟶ 0. (35) \|t^{-1}h_t-\mu_\phi\|_{\dot H^{-1}_D}
\le C\left(|t|^{1/2}+\frac{|c(t)|}{|t|}\right)
\le C|t|^{1/2}\longrightarrow0. \tag*{(35)} ∥ t − 1 h t − μ ϕ ∥ H ˙ D − 1 ≤ C ( ∣ t ∣ 1/2 + ∣ t ∣ ∣ c ( t ) ∣ ) ≤ C ∣ t ∣ 1/2 ⟶ 0. (35)
and therefore
∥ h t ∥ H ˙ D − 1 2 = t 2 ∥ μ ϕ ∥ H ˙ D − 1 2 + o ( t 2 ) . (36) \|h_t\|_{\dot{H}_D^{-1}}^2=t^2\|\mu_\phi\|_{\dot{H}_D^{-1}}^2+o(t^2). \tag*{(36)} ∥ h t ∥ H ˙ D − 1 2 = t 2 ∥ μ ϕ ∥ H ˙ D − 1 2 + o ( t 2 ) . (36)
The linear energy term depends on the first normal derivative of u u u . Since u = 0 u=0 u = 0 on Γ \Gamma Γ and ∂ ν u = − g \partial_\nu u=-g ∂ ν u = − g , u ∘ F ℓ ( s , r ) = − g ( s ) r + O ( ∣ r ∣ 1 + α ) u\circ F_\ell(s,r)=-g(s)r+O(|r|^{1+\alpha}) u ∘ F ℓ ( s , r ) = − g ( s ) r + O ( ∣ r ∣ 1 + α ) uniformly on the fixed family of tubes. Signed integration, valid for both positive and negative displacements, gives
∫ D u h t d x = ∑ ℓ ∫ Γ ℓ ∫ 0 b ℓ , t ( u ∘ F ℓ ) ( s , r ) ( 1 + κ ℓ r ) d r d s = − 1 2 ∑ ℓ ∫ Γ ℓ g b ℓ , t 2 d H 1 + O ( ∣ t ∣ 2 + α + ∣ t ∣ 3 ) = − t 2 2 ∫ Γ g ϕ 2 d H 1 + o ( t 2 ) . (37) \begin{aligned}
\int_D uh_t\,\mathrm{d}x
&=\sum_{\ell}\int_{\Gamma_\ell}\int_0^{b_{\ell,t}}(u\circ F_\ell)(s,r)(1+\kappa_\ell r)\,\mathrm{d}r\,\mathrm{d}s\\
&=-\frac{1}{2}\sum_{\ell}\int_{\Gamma_\ell}g b_{\ell,t}^2\,\mathrm{d}\mathcal{H}^1+O(|t|^{2+\alpha}+|t|^3)\\
&=-\frac{t^2}{2}\int_\Gamma g\phi^2\,\mathrm{d}\mathcal{H}^1+o(t^2). \tag*{(37)}
\end{aligned} ∫ D u h t d x = ℓ ∑ ∫ Γ ℓ ∫ 0 b ℓ , t ( u ∘ F ℓ ) ( s , r ) ( 1 + κ ℓ r ) d r d s = − 2 1 ℓ ∑ ∫ Γ ℓ g b ℓ , t 2 d H 1 + O ( ∣ t ∣ 2 + α + ∣ t ∣ 3 ) = − 2 t 2 ∫ Γ g ϕ 2 d H 1 + o ( t 2 ) . (37)
Since the construction preserves the area exactly, we may apply (27) with η = 1 A t \eta=\mathbf{1}_{A_t} η = 1 A t and combine the two energy expansions to obtain
0 ≥ 2 t 2 ( E ( 1 A t ) − E ( 1 A ) ) = 2 t 2 ∫ D u h t d x + ∥ t − 1 h t ∥ H ˙ D − 1 2 ⟶ − ∫ Γ g ϕ 2 d H 1 + ∥ μ ϕ ∥ H ˙ D − 1 2 . 0\geq\frac{2}{t^2}\bigl(\mathcal{E}(\mathbf{1}_{A_t})-\mathcal{E}(\mathbf{1}_A)\bigr)=\frac{2}{t^2}\int_D uh_t\,\mathrm{d}x+\|t^{-1}h_t\|_{\dot{H}_D^{-1}}^2\longrightarrow-\int_\Gamma g\phi^2\,\mathrm{d}\mathcal{H}^1+\|\mu_\phi\|_{\dot{H}_D^{-1}}^2. 0 ≥ t 2 2 ( E ( 1 A t ) − E ( 1 A ) ) = t 2 2 ∫ D u h t d x + ∥ t − 1 h t ∥ H ˙ D − 1 2 ⟶ − ∫ Γ g ϕ 2 d H 1 + ∥ μ ϕ ∥ H ˙ D − 1 2 .
To extend the inequality to L 2 L^2 L 2 displacements, approximate while retaining the total-flux constraint. Choose smooth ψ n → ϕ \psi_n\to\phi ψ n → ϕ in L 2 ( Γ ) L^2(\Gamma) L 2 ( Γ ) with the preceding support conventions, and fix such a smooth ζ \zeta ζ with ∫ Γ ζ = 1 \int_\Gamma\zeta=1 ∫ Γ ζ = 1 . Then
ϕ n : = ψ n − ( ∫ Γ ψ n d H 1 ) ζ , ∫ Γ ϕ n = 0 , ϕ n → ϕ in L 2 ( Γ ) . \phi_n:=\psi_n-\left(\int_\Gamma\psi_n\,\mathrm{d}\mathcal{H}^1\right)\zeta,\qquad\int_\Gamma\phi_n=0,\qquad\phi_n\to\phi\text{ in }L^2(\Gamma). ϕ n := ψ n − ( ∫ Γ ψ n d H 1 ) ζ , ∫ Γ ϕ n = 0 , ϕ n → ϕ in L 2 ( Γ ) .
The trace theorem gives, for f ∈ L 2 ( Γ ) f\in L^2(\Gamma) f ∈ L 2 ( Γ ) ,
∥ f H 1 ⌞ Γ ∥ H ˙ D − 1 = sup z ∈ H 0 1 ( D ) ∥ ∇ z ∥ L 2 ( D ) = 1 ∣ ∫ Γ f z d H 1 ∣ ≤ ∥ f ∥ L 2 ( Γ ) sup z ∈ H 0 1 ( D ) ∥ ∇ z ∥ L 2 ( D ) = 1 ∥ z ∥ L 2 ( Γ ) ≤ C Γ ∥ f ∥ L 2 ( Γ ) . \|f\mathcal{H}^1\mathbin{\llcorner}\Gamma\|_{\dot{H}_D^{-1}}
=\sup_{\substack{z\in H^1_0(D)\\ \|\nabla z\|_{L^2(D)}=1}}\left|\int_\Gamma fz\,\mathrm{d}\mathcal{H}^1\right|
\leq\|f\|_{L^2(\Gamma)}\sup_{\substack{z\in H^1_0(D)\\ \|\nabla z\|_{L^2(D)}=1}}\|z\|_{L^2(\Gamma)}
\leq C_\Gamma\|f\|_{L^2(\Gamma)}. ∥ f H 1 └ Γ ∥ H ˙ D − 1 = z ∈ H 0 1 ( D ) ∥∇ z ∥ L 2 ( D ) = 1 sup ∫ Γ f z d H 1 ≤ ∥ f ∥ L 2 ( Γ ) z ∈ H 0 1 ( D ) ∥∇ z ∥ L 2 ( D ) = 1 sup ∥ z ∥ L 2 ( Γ ) ≤ C Γ ∥ f ∥ L 2 ( Γ ) .
Thus
∥ ( ϕ n − ϕ ) H 1 ⌞ Γ ∥ H ˙ D − 1 ≤ C Γ ∥ ϕ n − ϕ ∥ L 2 ( Γ ) → 0 , ∫ Γ g ϕ n 2 d H 1 → ∫ Γ g ϕ 2 d H 1 . \|(\phi_n-\phi)\mathcal{H}^1\mathbin{\llcorner}\Gamma\|_{\dot{H}_D^{-1}}\leq C_\Gamma\|\phi_n-\phi\|_{L^2(\Gamma)}\to0,\qquad\int_\Gamma g\phi_n^2\,\mathrm{d}\mathcal{H}^1\to\int_\Gamma g\phi^2\,\mathrm{d}\mathcal{H}^1. ∥ ( ϕ n − ϕ ) H 1 └ Γ ∥ H ˙ D − 1 ≤ C Γ ∥ ϕ n − ϕ ∥ L 2 ( Γ ) → 0 , ∫ Γ g ϕ n 2 d H 1 → ∫ Γ g ϕ 2 d H 1 .
Taking t → 0 t\to0 t → 0 for each smooth ϕ n \phi_n ϕ n and then n → ∞ n\to\infty n → ∞ using these two convergences proves (30) . Because the constraint acts on the total flux ∫ Γ ϕ \int_\Gamma\phi ∫ Γ ϕ , it allows flux to be transferred between distinct pieces, while the quadratic term retains their Green interactions:
∥ μ ϕ ∥ H ˙ D − 1 2 = ∑ ℓ , k = 1 M ∫ Γ ℓ ∫ Γ k G D ( x , y ) ϕ ℓ ( x ) ϕ k ( y ) d H 1 ( y ) d H 1 ( x ) . \|\mu_\phi\|_{\dot{H}_D^{-1}}^2=\sum_{\ell,k=1}^{M}\int_{\Gamma_\ell}\int_{\Gamma_k}G_D(x,y)\phi_\ell(x)\phi_k(y)\,\mathrm{d}\mathcal{H}^1(y)\,\mathrm{d}\mathcal{H}^1(x). ∥ μ ϕ ∥ H ˙ D − 1 2 = ℓ , k = 1 ∑ M ∫ Γ ℓ ∫ Γ k G D ( x , y ) ϕ ℓ ( x ) ϕ k ( y ) d H 1 ( y ) d H 1 ( x ) .
The variation estimate can now be expressed as a weighted trace inequality for test functions in H 0 1 ( D ) H^1_0(D) H 0 1 ( D ) , in a form suited to controlling the weighted boundary measure g − 1 d H 1 g^{-1}d\mathcal{H}^1 g − 1 d H 1 on families of regular arcs that approach a possible critical point.
Corollary 3.4. For v ∈ H 0 1 ( D ) v\in H^1_0(D) v ∈ H 0 1 ( D ) , set
c Γ , v : = ∫ Γ v / g d H 1 ∫ Γ 1 / g d H 1 . c_{\Gamma,v}:=\frac{\int_{\Gamma}v/g\,d\mathcal{H}^1}{\int_{\Gamma}1/g\,d\mathcal{H}^1}. c Γ , v := ∫ Γ 1/ g d H 1 ∫ Γ v / g d H 1 . Then
∫ Γ ( v − c Γ , v ) 2 g d H 1 ≤ ∫ D ∣ ∇ v ∣ 2 d x . (38) \int_{\Gamma}\frac{(v-c_{\Gamma,v})^2}{g}\,d\mathcal{H}^1\leq\int_D|\nabla v|^2\,dx. \tag*{(38)} ∫ Γ g ( v − c Γ , v ) 2 d H 1 ≤ ∫ D ∣∇ v ∣ 2 d x . (38)
Proof. Subtracting the common weighted mean makes the displacement below satisfy the single area constraint. Since min Γ g > 0 \min_{\Gamma}g>0 min Γ g > 0 , the trace of v v v defines
c : = c Γ , v , ϕ : = v − c g ∈ L 2 ( Γ ) , ∫ Γ ϕ = 0 , X : = ∫ Γ g ϕ 2 d H 1 . c:=c_{\Gamma,v},\qquad\phi:=\frac{v-c}{g}\in L^2(\Gamma),\qquad\int_{\Gamma}\phi=0,\qquad X:=\int_{\Gamma}g\phi^2\,d\mathcal{H}^1. c := c Γ , v , ϕ := g v − c ∈ L 2 ( Γ ) , ∫ Γ ϕ = 0 , X := ∫ Γ g ϕ 2 d H 1 .
Pairing this displacement with v v v recovers the weighted boundary variance X X X , to which Dirichlet duality and proposition 3.3 apply:
X=\int_{\Gamma}(v-c)\phi\,d\mathcal{H}^1=\int_{\Gamma}v\phi\,d\mathcal{H}^1\leq\|\phi\mathcal{H}^1\mathbin{\vrule height 1.4ex depth -0.4ex width 0.07ex\vrule height 0.07ex depth -0.0ex width 0.7ex\Gamma}\|_{\dot H^{-1}_D}\|\nabla v\|_{L^2(D)}\leq X^{1/2}\|\nabla v\|_{L^2(D)}.
For X = 0 X=0 X = 0 the claim is immediate; for X > 0 X>0 X > 0 , division by X 1 / 2 X^{1/2} X 1/2 gives X ≤ ∥ ∇ v ∥ L 2 ( D ) 2 X\leq\|\nabla v\|_{L^2(D)}^2 X ≤ ∥∇ v ∥ L 2 ( D ) 2 . □
The weighted trace inequality controls regular boundary pieces approaching a common point through the following summability criterion, which will be used to exclude accumulation of regular curves at a critical point.
Lemma 3.5. Let x 0 ∈ D x_0\in D x 0 ∈ D and
R 1 > R 2 > ⋯ ↓ 0 , B R 1 ( x 0 ) ⊂ D . R_1>R_2>\cdots\downarrow0,\qquad B_{R_1}(x_0)\subset D. R 1 > R 2 > ⋯ ↓ 0 , B R 1 ( x 0 ) ⊂ D . Suppose that γ j \gamma_j γ j are pairwise disjoint compact regular arcs or closed regular curves with
γ j ⊂ B R j ( x 0 ) ‾ ∖ B R j + 1 ( x 0 ) . \gamma_j\subset\overline{B_{R_j}(x_0)}\setminus B_{R_{j+1}}(x_0). γ j ⊂ B R j ( x 0 ) ∖ B R j + 1 ( x 0 ) . Then
∑ j = 1 ∞ ∫ γ j 1 ∣ ∇ u ∣ d H 1 < ∞ . (39) \sum_{j=1}^{\infty}\int_{\gamma_j}\frac{1}{|\nabla u|}\,d\mathcal{H}^1<\infty. \tag*{(39)} j = 1 ∑ ∞ ∫ γ j ∣∇ u ∣ 1 d H 1 < ∞. (39) If ∇ u ( x 0 ) = 0 \nabla u(x_0)=0 ∇ u ( x 0 ) = 0 , there are no R , c > 0 R,c>0 R , c > 0 and regular arcs satisfying, for every sufficiently large k k k ,
γ k ⊂ B 2 − k R ( x 0 ) ‾ ∖ B 2 − k − 1 R ( x 0 ) , H 1 ( γ k ) ≥ c 2 − k R . (40) \gamma_k\subset\overline{B_{2^{-k}R}(x_0)}\setminus B_{2^{-k-1}R}(x_0),\qquad\mathcal{H}^1(\gamma_k)\geq c2^{-k}R. \tag*{(40)} γ k ⊂ B 2 − k R ( x 0 ) ∖ B 2 − k − 1 R ( x 0 ) , H 1 ( γ k ) ≥ c 2 − k R . (40)
Proof. Set
g = ∣ ∇ u ∣ , a j = ∫ γ j g − 1 d H 1 ∈ ( 0 , ∞ ) , A K = ∑ j = 1 K a j . g=|\nabla u|,\qquad a_j=\int_{\gamma_j}g^{-1}\,d\mathcal{H}^1\in(0,\infty),\qquad A_K=\sum_{j=1}^{K}a_j. g = ∣∇ u ∣ , a j = ∫ γ j g − 1 d H 1 ∈ ( 0 , ∞ ) , A K = j = 1 ∑ K a j .
We apply the weighted trace inequality to two separated families with equal g − 1 d H 1 g^{-1}\,d\mathcal{H}^{1} g − 1 d H 1 mass, choosing a cutoff that equals 0 on one family and 2 on the other. Its weighted mean is then 1, whereas its Dirichlet energy is independent of the spatial scale in dimension two, giving the contradiction below if the weighted masses diverge. Suppose A K → ∞ A_K \to\infty A K → ∞ . Choose
V ∈ C c ∞ ( B 1 ) , 0 ≤ V ≤ 2 , V ∣ B 1 / 2 = 2 , V \in C^\infty_c(B_1), \qquad0 \le V \le2, \qquad V\big|_{B_{1/2}}=2, V ∈ C c ∞ ( B 1 ) , 0 ≤ V ≤ 2 , V B 1/2 = 2 ,
v K ( x ) = V ( x − x 0 R K + 1 ) . v_K(x)=V\left(\frac{x-x_0}{R_{K+1}}\right). v K ( x ) = V ( R K + 1 x − x 0 ) .
For M > K + 1 M>K+1 M > K + 1 such that R M < R K + 1 / 2 R_M<R_{K+1}/2 R M < R K + 1 /2 ,
v K ∣ γ j = 0 ( j ≤ K ) , v K ∣ γ j = 2 ( j ≥ M ) , ∑ j = M ∞ a j = ∞ . v_K\big|_{\gamma_j}=0\quad(j\le K), \qquad v_K\big|_{\gamma_j}=2\quad(j\ge M), \qquad\sum_{j=M}^{\infty}a_j=\infty. v K γ j = 0 ( j ≤ K ) , v K γ j = 2 ( j ≥ M ) , j = M ∑ ∞ a j = ∞.
Since the tail has infinite weighted mass, we can truncate it to match the mass A K A_K A K of the first K K K pieces. Let
L = min { ℓ ≥ M : ∑ j = M ℓ a j ≥ A K } , β K = A K − ∑ j = M L − 1 a j ∈ ( 0 , a L ] . L=\min\left\{\ell\ge M:\sum_{j=M}^{\ell}a_j\ge A_K\right\}, \qquad\beta_K=A_K-\sum_{j=M}^{L-1}a_j\in(0,a_L]. L = min ⎩ ⎨ ⎧ ℓ ≥ M : j = M ∑ ℓ a j ≥ A K ⎭ ⎬ ⎫ , β K = A K − j = M ∑ L − 1 a j ∈ ( 0 , a L ] .
For an arclength parametrization γ L : [ 0 , H 1 ( γ L ) ] → γ L \gamma_L:[0,\mathcal{H}^{1}(\gamma_L)]\to\gamma_L γ L : [ 0 , H 1 ( γ L )] → γ L , choose s K s_K s K by
∫ 0 s K d s g ( γ L ( s ) ) = β K . \int_0^{s_K}\frac{ds}{g(\gamma_L(s))}=\beta_K. ∫ 0 s K g ( γ L ( s )) d s = β K .
The integral is continuous and strictly increasing from 0 to a L a_L a L , so this choice realizes the remaining mass β K \beta_K β K exactly. Set
T K = ⋃ j = M L − 1 γ j ∪ γ L ( [ 0 , s K ] ) , Γ K = ⋃ j = 1 K γ j ∪ T K , T_K=\bigcup_{j=M}^{L-1}\gamma_j\cup\gamma_L([0,s_K]), \qquad\Gamma_K=\bigcup_{j=1}^{K}\gamma_j\cup T_K, T K = j = M ⋃ L − 1 γ j ∪ γ L ([ 0 , s K ]) , Γ K = j = 1 ⋃ K γ j ∪ T K ,
∫ T K g − 1 d H 1 = A K . \int_{T_K}g^{-1}\,d\mathcal{H}^{1}=A_K. ∫ T K g − 1 d H 1 = A K .
Then
∫ Γ K g − 1 d H 1 = 2 A K , ∫ Γ K v K g − 1 d H 1 = 2 A K , c Γ K , v K = 1. \int_{\Gamma_K}g^{-1}\,d\mathcal{H}^{1}=2A_K, \qquad\int_{\Gamma_K}v_Kg^{-1}\,d\mathcal{H}^{1}=2A_K, \qquad c_{\Gamma_K,v_K}=1. ∫ Γ K g − 1 d H 1 = 2 A K , ∫ Γ K v K g − 1 d H 1 = 2 A K , c Γ K , v K = 1.
The last piece is a compact subarc or the whole closed curve. By corollary 3.4 and two-dimensional Dirichlet scaling,
2 A K = ∫ Γ K ( v K − 1 ) 2 g d H 1 ≤ ∫ D ∣ ∇ v K ∣ 2 d x = ∫ R 2 ∣ ∇ V ∣ 2 d x . (41) 2A_K=\int_{\Gamma_K}\frac{(v_K-1)^2}{g}\,d\mathcal{H}^{1}\le\int_D|\nabla v_K|^2\,dx=\int_{\mathbb{R}^2}|\nabla V|^2\,dx. \tag*{(41)} 2 A K = ∫ Γ K g ( v K − 1 ) 2 d H 1 ≤ ∫ D ∣∇ v K ∣ 2 d x = ∫ R 2 ∣∇ V ∣ 2 d x . (41)
contrary to A K → ∞ A_K\to\infty A K → ∞ . The last equality follows because the change of variables x = x 0 + R K + 1 y x=x_0+R_{K+1}y x = x 0 + R K + 1 y cancels the factor R K + 1 − 2 R_{K+1}^{-2} R K + 1 − 2 from ∣ ∇ v K ∣ 2 |\nabla v_K|^2 ∣∇ v K ∣ 2 against the Jacobian R K + 1 2 R_{K+1}^2 R K + 1 2 , so no comparison between consecutive radii is required.
For the second assertion, suppose (3.15) holds and set ρ j = 2 − 2 j R \rho_j=2^{-2j}R ρ j = 2 − 2 j R . Retaining the even indices separates the arcs by a positive radial gap, including when consecutive dyadic arcs meet on a common circle. Then
x 0 ∈ ⋃ j γ 2 j ‾ ⊂ ∂ A , \mathbf{x}_0 \in\overline{\bigcup_j \gamma_{2j}} \subset\partial A, x 0 ∈ j ⋃ γ 2 j ⊂ ∂ A ,
γ 2 j ⊂ B ρ j ( x 0 ) ‾ ∖ B ρ j + 1 ( x 0 ) , H 1 ( γ 2 j ) ≥ c ρ j . \gamma_{2j} \subset\overline{B_{\rho_j}(\mathbf{x}_0)} \setminus B_{\rho_{j+1}}(\mathbf{x}_0), \qquad\mathcal{H}^1(\gamma_{2j}) \ge c\rho_j. γ 2 j ⊂ B ρ j ( x 0 ) ∖ B ρ j + 1 ( x 0 ) , H 1 ( γ 2 j ) ≥ c ρ j .
The gradient estimate at a critical point converts the lower bound on arc length into a divergent lower bound on weighted mass. By (23) , for j ≥ j 0 j \ge j_0 j ≥ j 0 with j 0 j_0 j 0 sufficiently large,
sup γ 2 j g ≤ C ρ j ( 1 + log ( 2 r 0 / ρ j ) ) ≤ C ′ ( j + 1 ) ρ j , ∫ γ 2 j g − 1 d H 1 ≥ c ′ j + 1 . \sup_{\gamma_{2j}} g \le C\rho_j(1+\log(2r_0/\rho_j)) \le C'(j+1)\rho_j, \qquad\int_{\gamma_{2j}} g^{-1}\,d\mathcal{H}^1 \ge\frac{c'}{j+1}. γ 2 j sup g ≤ C ρ j ( 1 + log ( 2 r 0 / ρ j )) ≤ C ′ ( j + 1 ) ρ j , ∫ γ 2 j g − 1 d H 1 ≥ j + 1 c ′ .
Thus
∞ = ∑ j ≥ j 0 c ′ j + 1 ≤ ∑ j ≥ j 0 ∫ γ 2 j g − 1 d H 1 < ∞ , \infty= \sum_{j\ge j_0}\frac{c'}{j+1} \le\sum_{j\ge j_0}\int_{\gamma_{2j}}g^{-1}\,d\mathcal{H}^1 < \infty, ∞ = j ≥ j 0 ∑ j + 1 c ′ ≤ j ≥ j 0 ∑ ∫ γ 2 j g − 1 d H 1 < ∞ ,
contradicting (39) . This argument follows the device of [14 Lemmas 3.1–3.2] . □
To recover the entire boundary from its regular part, we first show that every boundary point can be approximated by regular points.
Lemma 3.6. The regular set R \mathcal{R} R is dense in ∂ A \partial A ∂ A .
Proof. An interior disk tangent to the patch boundary provides a point with nonzero gradient, because comparison with its explicit torsion function gives a derivative bound without requiring regularity at the contact point. To construct such a point near any given p ∈ ∂ A \mathbf{p}\in\partial A p ∈ ∂ A , fix B ε ( p ) ⊂ D B_\varepsilon(\mathbf{p})\subset D B ε ( p ) ⊂ D and choose
a ∈ A ∩ B ε / 4 ( p ) , z ∈ ∂ A , \mathbf{a}\in A\cap B_{\varepsilon/4}(\mathbf{p}), \qquad\mathbf{z}\in\partial A, a ∈ A ∩ B ε /4 ( p ) , z ∈ ∂ A ,
0 < d : = dist ( a , ∂ A ) = ∣ z − a ∣ ≤ ∣ a − p ∣ < ε / 4. 0<d:=\operatorname{dist}(\mathbf{a},\partial A)=|\mathbf{z}-\mathbf{a}|\le|\mathbf{a}-\mathbf{p}|<\varepsilon/4. 0 < d := dist ( a , ∂ A ) = ∣ z − a ∣ ≤ ∣ a − p ∣ < ε /4.
Then
B d ( a ) ⊂ A , B d ( a ) ‾ ⊂ B ε / 2 ( p ) , v ( x ) : = d 2 − ∣ x − a ∣ 2 4 . B_d(\mathbf{a})\subset A,\qquad\overline{B_d(\mathbf{a})}\subset B_{\varepsilon/2}(\mathbf{p}),\qquad v(\mathbf{x}):=\frac{d^2-|\mathbf{x}-\mathbf{a}|^2}{4}. B d ( a ) ⊂ A , B d ( a ) ⊂ B ε /2 ( p ) , v ( x ) := 4 d 2 − ∣ x − a ∣ 2 .
The maximum principle gives
{ Δ ( u − v ) = 0 in B d ( a ) , u − v ≥ 0 on ∂ B d ( a ) , ⟹ u ≥ v in B d ( a ) . \begin{cases}
\Delta(u-v)=0 & \text{in } B_d(\mathbf{a}),\\
u-v\ge0 & \text{on } \partial B_d(\mathbf{a}),
\end{cases}
\qquad\Longrightarrow\qquad u\ge v\quad\text{in }B_d(\mathbf{a}). { Δ ( u − v ) = 0 u − v ≥ 0 in B d ( a ) , on ∂ B d ( a ) , ⟹ u ≥ v in B d ( a ) .
For n = ( z − a ) / d \mathbf{n}=(\mathbf{z}-\mathbf{a})/d n = ( z − a ) / d ,
u ( z − t n ) ≥ d t 2 − t 2 4 ( 0 < t < d ) , − ∇ u ( z ) ⋅ n = lim t → 0 u ( z − t n ) t ≥ d 2 > 0. u(\mathbf{z}-t\mathbf{n})\ge\frac{dt}{2}-\frac{t^2}{4}\quad(0<t<d),\qquad-\nabla u(\mathbf{z})\cdot\mathbf{n}=\lim_{t\to0}\frac{u(\mathbf{z}-t\mathbf{n})}{t}\ge\frac d2>0. u ( z − t n ) ≥ 2 d t − 4 t 2 ( 0 < t < d ) , − ∇ u ( z ) ⋅ n = t → 0 lim t u ( z − t n ) ≥ 2 d > 0.
Therefore
R ∩ B ε / 2 ( p ) ∋ z , p ∈ R ‾ . \mathcal{R}\cap B_{\varepsilon/2}(\mathbf{p})\ni\mathbf{z},\qquad\mathbf{p}\in\overline{\mathcal{R}}. R ∩ B ε /2 ( p ) ∋ z , p ∈ R .
Here ε > 0 \varepsilon>0 ε > 0 can be arbitrarily small. □
Having established density of the regular set, we use the annular criterion to show that each of its components closes up inside D D D , rather than accumulating at a critical boundary point.
Lemma 3.7. Every connected component of R \mathcal{R} R is a simple closed real-analytic curve compactly contained in D D D .
Proof. For a connected component C C C of R \mathcal{R} R , accumulation at a critical point would force C C C to cross every sufficiently small dyadic annulus, contrary to lemma 3.5. We make this argument precise using
C ‾ ⊂ ∂ A ⊂ D . \overline{C} \subset\partial A \subset D. C ⊂ ∂ A ⊂ D .
Suppose p ∈ C ‾ ∖ R \mathbf{p} \in\overline{C} \setminus\mathcal{R} p ∈ C ∖ R , so ∇ u ( p ) = 0 \nabla u(\mathbf{p}) = 0 ∇ u ( p ) = 0 . Choose
z ∈ C , B ρ 0 ( p ) ⊂ D , ρ k = 2 − k ρ 0 . \mathbf{z} \in C,\qquad B_{\rho_0}(\mathbf{p}) \subset D,\qquad\rho_k = 2^{-k}\rho_0. z ∈ C , B ρ 0 ( p ) ⊂ D , ρ k = 2 − k ρ 0 .
For all sufficiently large k k k ,
y k ∈ C ∩ B ρ k / 2 ( p ) , ∣ z − p ∣ > ρ k . \mathbf{y}_k \in C \cap B_{\rho_k/2}(\mathbf{p}),\qquad|\mathbf{z}-\mathbf{p}| > \rho_k. y k ∈ C ∩ B ρ k /2 ( p ) , ∣ z − p ∣ > ρ k .
Arcwise connectedness of C C C gives a smooth embedded arc
σ k : [ 0 , 1 ] → C , σ k ( 0 ) = y k , σ k ( 1 ) = z . \sigma_k : [0,1] \to C,\qquad\sigma_k(0)=\mathbf{y}_k,\qquad\sigma_k(1)=\mathbf{z}. σ k : [ 0 , 1 ] → C , σ k ( 0 ) = y k , σ k ( 1 ) = z .
Set
b k = min { t : ∣ σ k ( t ) − p ∣ = ρ k } , b_k = \min\{t : |\sigma_k(t)-\mathbf{p}|=\rho_k\}, b k = min { t : ∣ σ k ( t ) − p ∣ = ρ k } ,
a k = max { t ≤ b k : ∣ σ k ( t ) − p ∣ = ρ k / 2 } , a_k = \max\{t\leq b_k : |\sigma_k(t)-\mathbf{p}|=\rho_k/2\}, a k = max { t ≤ b k : ∣ σ k ( t ) − p ∣ = ρ k /2 } ,
γ k = σ k ( [ a k , b k ] ) . \gamma_k = \sigma_k([a_k,b_k]). γ k = σ k ([ a k , b k ]) .
Then a k < b k a_k < b_k a k < b k and
γ k ⊂ B ρ k ( p ) ∖ B ρ k / 2 ( p ) , H 1 ( γ k ) ≥ ∣ σ k ( b k ) − σ k ( a k ) ∣ ≥ ρ k / 2. \gamma_k \subset B_{\rho_k}(\mathbf{p}) \setminus B_{\rho_k/2}(\mathbf{p}),\qquad\mathcal{H}^{1}(\gamma_k) \geq|\sigma_k(b_k)-\sigma_k(a_k)| \geq\rho_k/2. γ k ⊂ B ρ k ( p ) ∖ B ρ k /2 ( p ) , H 1 ( γ k ) ≥ ∣ σ k ( b k ) − σ k ( a k ) ∣ ≥ ρ k /2.
Thus lemma 3.5 gives
C ‾ ∖ R = ∅ . \overline{C} \setminus\mathcal{R} = \varnothing. C ∖ R = ∅ .
It remains to exclude accumulation on a different regular component. For p ∈ C ‾ \mathbf{p} \in\overline{C} p ∈ C , choose a coordinate neighborhood U U U such that U ∩ R U \cap\mathcal{R} U ∩ R is one connected arc. Then
( U ∩ R ) ∩ C ≠ ∅ ⟹ U ∩ R ⊂ C ⟹ p ∈ C . (U\cap\mathcal{R})\cap C\ne\varnothing\quad\Longrightarrow\quad U\cap\mathcal{R}\subset C \quad\Longrightarrow\quad\mathbf{p}\in C. ( U ∩ R ) ∩ C = ∅ ⟹ U ∩ R ⊂ C ⟹ p ∈ C .
Consequently,
C = C ‾ ⊂ D , 0 < H 1 ( C ) < ∞ , min C ∣ ∇ u ∣ > 0. C=\overline{C}\subset D,\qquad0<\mathcal{H}^{1}(C)<\infty,\qquad\min_C|\nabla u|>0. C = C ⊂ D , 0 < H 1 ( C ) < ∞ , C min ∣∇ u ∣ > 0.
The classification of compact connected one-dimensional manifolds without boundary shows that C C C is diffeomorphic to S 1 S^1 S 1 ; together with its embedding and lemma 2.2, this identifies C C C as a real-analytic Jordan curve. □
To control the number of these closed curves, we identify the phase enclosed by each curve and obtain a lower bound on its weighted boundary integral that is independent of its size.
Lemma 3.8. For every connected component Γ \Gamma Γ of R \mathcal{R} R , its Jordan interior Ω Γ = Int ( Γ ) \Omega_\Gamma=\operatorname{Int}(\Gamma) Ω Γ = Int ( Γ ) satisfies
Ω Γ ‾ ⊂ D , Ω Γ ⊂ A , (42) \overline{\Omega_\Gamma}\subset D,\qquad\Omega_\Gamma\subset A, \tag*{(42)} Ω Γ ⊂ D , Ω Γ ⊂ A , (42)
and
∫ Γ ∣ ∇ u ∣ d H 1 = ∣ Ω Γ ∣ , ∫ Γ 1 ∣ ∇ u ∣ d H 1 ≥ 4 π . (43) \int_{\Gamma} |\nabla u|\,d\mathcal{H}^{1}=|\Omega_{\Gamma}|,\qquad
\int_{\Gamma}\frac{1}{|\nabla u|}\,d\mathcal{H}^{1}\ge4\pi. \tag*{(43)} ∫ Γ ∣∇ u ∣ d H 1 = ∣ Ω Γ ∣ , ∫ Γ ∣∇ u ∣ 1 d H 1 ≥ 4 π . (43)
Proof. Before testing the equation on the Jordan interior, we use simple connectedness of D D D to ensure that this interior lies within the domain of the equation. Denoting the winding number by Ind ( Γ , y ) \operatorname{Ind}(\Gamma,y) Ind ( Γ , y ) , a point y ∈ Ω Γ ∖ D y\in\Omega_{\Gamma}\setminus D y ∈ Ω Γ ∖ D would give, by Jordan separation and contraction of Γ \Gamma Γ in D D D , respectively,
∣ Ind ( Γ , y ) ∣ = 1 , Ind ( Γ , y ) = 0 , |\operatorname{Ind}(\Gamma,y)|=1,\qquad\operatorname{Ind}(\Gamma,y)=0, ∣ Ind ( Γ , y ) ∣ = 1 , Ind ( Γ , y ) = 0 ,
a contradiction. Therefore
Ω Γ ⊂ D , Ω Γ ‾ = Ω Γ ∪ Γ ⊂ D . \Omega_{\Gamma}\subset D,\qquad\overline{\Omega_{\Gamma}}=\Omega_{\Gamma}\cup\Gamma\subset D. Ω Γ ⊂ D , Ω Γ = Ω Γ ∪ Γ ⊂ D .
Since u = 0 u=0 u = 0 on Γ \Gamma Γ , its negative part u − : = max { − u , 0 } u^{-}:=\max\{-u,0\} u − := max { − u , 0 } belongs to H 0 1 ( Ω Γ ) H^1_0(\Omega_{\Gamma}) H 0 1 ( Ω Γ ) and is therefore an admissible test function; using 1 A u − = 0 \mathbf{1}_A u^{-}=0 1 A u − = 0 , which follows from A = { u > 0 } A=\{u>0\} A = { u > 0 } , we obtain
− ∫ Ω Γ ∣ ∇ u − ∣ 2 d x = ∫ Ω Γ 1 A u − d x = 0 , u − ≡ 0. -\int_{\Omega_{\Gamma}}|\nabla u^{-}|^2\,dx=\int_{\Omega_{\Gamma}}\mathbf{1}_A u^{-}\,dx=0,\qquad u^{-}\equiv0. − ∫ Ω Γ ∣∇ u − ∣ 2 d x = ∫ Ω Γ 1 A u − d x = 0 , u − ≡ 0.
Since
u ≡ 0 in Ω Γ ⟹ ∇ u ∣ Γ = 0 ⟹ Γ ∩ R = ∅ , u\equiv0\text{ in }\Omega_{\Gamma}\quad\Longrightarrow\quad\nabla u|_{\Gamma}=0\quad\Longrightarrow\quad\Gamma\cap\mathcal{R}=\varnothing, u ≡ 0 in Ω Γ ⟹ ∇ u ∣ Γ = 0 ⟹ Γ ∩ R = ∅ ,
u ≢ 0 u\not\equiv0 u ≡ 0 , so the strong minimum principle for the nonnegative superharmonic function u u u excludes every interior zero and gives
u > 0 , − Δ u = 1 in Ω Γ , ∂ ν u = − ∣ ∇ u ∣ on Γ , u>0,\qquad-\Delta u=1\text{ in }\Omega_{\Gamma},\qquad\partial_{\nu}u=-|\nabla u|\text{ on }\Gamma, u > 0 , − Δ u = 1 in Ω Γ , ∂ ν u = − ∣∇ u ∣ on Γ ,
where ν \nu ν is the outward unit normal. Thus
∣ Ω Γ ∣ = − ∫ Ω Γ Δ u d x = − ∫ Γ ∂ ν u d H 1 = ∫ Γ ∣ ∇ u ∣ d H 1 . |\Omega_{\Gamma}|=-\int_{\Omega_{\Gamma}}\Delta u\,dx=-\int_{\Gamma}\partial_{\nu}u\,d\mathcal{H}^{1}=\int_{\Gamma}|\nabla u|\,d\mathcal{H}^{1}. ∣ Ω Γ ∣ = − ∫ Ω Γ Δ u d x = − ∫ Γ ∂ ν u d H 1 = ∫ Γ ∣∇ u ∣ d H 1 .
Combining this relation between boundary flux and enclosed area with the isoperimetric and Cauchy–Schwarz inequalities gives
4 π ∣ Ω Γ ∣ ≤ H 1 ( Γ ) 2 ≤ ( ∫ Γ ∣ ∇ u ∣ d H 1 ) ( ∫ Γ 1 ∣ ∇ u ∣ d H 1 ) = ∣ Ω Γ ∣ ∫ Γ 1 ∣ ∇ u ∣ d H 1 . 4\pi|\Omega_{\Gamma}|\le\mathcal{H}^{1}(\Gamma)^2\le\left(\int_{\Gamma}|\nabla u|\,d\mathcal{H}^{1}\right)\left(\int_{\Gamma}\frac{1}{|\nabla u|}\,d\mathcal{H}^{1}\right)=|\Omega_{\Gamma}|\int_{\Gamma}\frac{1}{|\nabla u|}\,d\mathcal{H}^{1}. 4 π ∣ Ω Γ ∣ ≤ H 1 ( Γ ) 2 ≤ ( ∫ Γ ∣∇ u ∣ d H 1 ) ( ∫ Γ ∣∇ u ∣ 1 d H 1 ) = ∣ Ω Γ ∣ ∫ Γ ∣∇ u ∣ 1 d H 1 .
Division by ∣ Ω Γ ∣ > 0 |\Omega_{\Gamma}|>0 ∣ Ω Γ ∣ > 0 proves (43) , whose lower bound 4 π 4\pi 4 π is independent of both the size and shape of the component. □
The uniform lower bound in lemma 3.8, together with annular summability, now rules out infinitely many regular components.
Proposition 3.9. The regular set R \mathcal{R} R has only finitely many connected components.
Proof. If infinitely many components accumulate, either all but finitely many of these curves leave some fixed neighborhood or every neighborhood contains infinitely many entire curves. The first possibility produces long annular arcs; the second contradicts summability through the uniform torsion bound. Suppose { Γ i } i ≥ 1 \{\Gamma_i\}_{i\ge1} { Γ i } i ≥ 1 are distinct components. Compactness of ∂ A \partial A ∂ A gives, after passage to a subsequence,
x i ∈ Γ i , x i ⟶ x ∞ ∈ ∂ A . x_i\in\Gamma_i,\qquad x_i\longrightarrow x_{\infty}\in\partial A. x i ∈ Γ i , x i ⟶ x ∞ ∈ ∂ A .
A coordinate neighborhood of a regular point meets only one component; therefore
∇ u ( x ∞ ) = 0 , dist ( x ∞ , Γ i ) > 0 for every i . \nabla u(x_\infty)=0,\qquad\operatorname{dist}(x_\infty,\Gamma_i)>0\quad\text{for every }i. ∇ u ( x ∞ ) = 0 , dist ( x ∞ , Γ i ) > 0 for every i .
We first claim
# { i : Γ i ⊂ B R ( x ∞ ) } = ∞ whenever B R ( x ∞ ) ⊂ D . (44) \#\{i:\Gamma_i\subset B_R(x_\infty)\}=\infty\qquad\text{whenever }B_R(x_\infty)\subset D. \tag*{(44)} # { i : Γ i ⊂ B R ( x ∞ )} = ∞ whenever B R ( x ∞ ) ⊂ D . (44)
If (44) fails, discarding finitely many curves gives
Γ i ∖ B R ( x ∞ ) ≠ ∅ ( i ≥ 1 ) , x i → x ∞ . \Gamma_i\setminus B_R(x_\infty)\ne\varnothing\quad(i\ge1),\qquad x_i\to x_\infty. Γ i ∖ B R ( x ∞ ) = ∅ ( i ≥ 1 ) , x i → x ∞ .
Choose distinct i ( k ) i(k) i ( k ) such that
Γ i ( k ) ∩ B 2 − k − 1 R ( x ∞ ) ≠ ∅ , Γ i ( k ) ∖ B R ( x ∞ ) ≠ ∅ . \Gamma_{i(k)}\cap B_{2^{-k-1}R}(x_\infty)\ne\varnothing,\qquad\Gamma_{i(k)}\setminus B_R(x_\infty)\ne\varnothing. Γ i ( k ) ∩ B 2 − k − 1 R ( x ∞ ) = ∅ , Γ i ( k ) ∖ B R ( x ∞ ) = ∅ .
The hitting-time construction in lemma 3.7 supplies compact regular arcs
γ k ⊂ Γ i ( k ) ∩ ( B 2 − k R ( x ∞ ) ∖ B 2 − k − 1 R ( x ∞ ) ) , H 1 ( γ k ) ≥ 2 − k − 1 R . \gamma_k\subset\Gamma_{i(k)}\cap\left(B_{2^{-k}R}(x_\infty)\setminus B_{2^{-k-1}R}(x_\infty)\right),\qquad\mathcal{H}^1(\gamma_k)\ge2^{-k-1}R. γ k ⊂ Γ i ( k ) ∩ ( B 2 − k R ( x ∞ ) ∖ B 2 − k − 1 R ( x ∞ ) ) , H 1 ( γ k ) ≥ 2 − k − 1 R .
Lemma 3.5 gives a contradiction, proving (44) .
The preceding claim allows us to place entire curves in separated annuli, with each new curve lying closer to x ∞ x_\infty x ∞ than the previously selected one. Choose B R 1 ( x ∞ ) ⊂ D B_{R_1}(x_\infty)\subset D B R 1 ( x ∞ ) ⊂ D and inductively select
Γ ^ j ⊂ B R j ( x ∞ ) , d j = dist ( x ∞ , Γ ^ j ) , R j + 1 = d j / 2. \widehat{\Gamma}_j\subset B_{R_j}(x_\infty),\qquad d_j=\operatorname{dist}(x_\infty,\widehat{\Gamma}_j),\qquad R_{j+1}=d_j/2. Γ j ⊂ B R j ( x ∞ ) , d j = dist ( x ∞ , Γ j ) , R j + 1 = d j /2.
Then
0 < R j + 1 < R j / 2 , Γ ^ j ⊂ B R j ( x ∞ ) ∖ B R j + 1 ( x ∞ ) ‾ . 0<R_{j+1}<R_j/2,\qquad\widehat{\Gamma}_j\subset B_{R_j}(x_\infty)\setminus\overline{B_{R_{j+1}}(x_\infty)}. 0 < R j + 1 < R j /2 , Γ j ⊂ B R j ( x ∞ ) ∖ B R j + 1 ( x ∞ ) .
The annuli are disjoint, whereas every selected curve contributes at least 4 π 4\pi 4 π to the weighted integral. By Lemmas 3.5 and 3.8 , we have
∑ j = 1 ∞ 4 π ≤ ∑ j = 1 ∞ ∫ Γ ^ j 1 ∣ ∇ u ∣ d H 1 < ∞ , \sum_{j=1}^{\infty}4\pi\le\sum_{j=1}^{\infty}\int_{\widehat{\Gamma}_j}\frac{1}{|\nabla u|}\,\mathrm{d}\mathcal{H}^1<\infty, j = 1 ∑ ∞ 4 π ≤ j = 1 ∑ ∞ ∫ Γ j ∣∇ u ∣ 1 d H 1 < ∞ ,
which is a contradiction. □
Proof of the boundary regularity
Combining density of regular points with finiteness of their closed components identifies the entire boundary, from which we recover the positive phase without imposing regularity on the remaining boundary points.
Proposition 4.1. There are finitely many pairwise disjoint real-analytic Jordan curves Γ 1 , … , Γ N \Gamma_1,\ldots,\Gamma_N Γ 1 , … , Γ N such that
∂ A = ⋃ j = 1 N Γ j , A = ⋃ j = 1 N Int ( Γ j ) . (45) \partial A=\bigcup_{j=1}^{N}\Gamma_j,\qquad A=\bigcup_{j=1}^{N}\operatorname{Int}(\Gamma_j). \tag*{(45)} ∂ A = j = 1 ⋃ N Γ j , A = j = 1 ⋃ N Int ( Γ j ) . (45) Every component of A A A is simply connected, and
∂ A = R , ∂ A ∩ { ∇ u = 0 } = ∅ . (46) \partial A=\mathcal{R},\qquad\partial A\cap\{\nabla u=0\}=\varnothing. \tag*{(46)} ∂ A = R , ∂ A ∩ { ∇ u = 0 } = ∅ . (46) Moreover,
{ u = 0 } = ∂ A , u < 0 in D ∖ A ‾ . (47) \{u=0\}=\partial A,\qquad u<0\quad\text{in }D\setminus\overline{A}. \tag*{(47)} { u = 0 } = ∂ A , u < 0 in D ∖ A . (47)
Proof. Since the regular set consists of finitely many compact components, their union is closed; density therefore leaves no additional boundary points. More precisely, lemmas 3.6 and 3.7 and proposition 3.9 give
1 ≤ N < ∞ , R = ⋃ j = 1 N Γ j ⊂ D , R ‾ = R , ∂ A = R ‾ = R . 1 \le N < \infty,\qquad\mathcal{R} = \bigcup_{j=1}^{N} \Gamma_j \subset D,\qquad\overline{\mathcal{R}} = \mathcal{R},\qquad\partial A = \overline{\mathcal{R}} = \mathcal{R}. 1 ≤ N < ∞ , R = j = 1 ⋃ N Γ j ⊂ D , R = R , ∂ A = R = R .
Continuity of the gradient on this compact set then gives
∂ A ∩ { ∇ u = 0 } = ∅ , min ∂ A ∣ ∇ u ∣ > 0. \partial A \cap\{\nabla u = 0\} = \varnothing,\qquad\min_{\partial A} |\nabla u| > 0. ∂ A ∩ { ∇ u = 0 } = ∅ , ∂ A min ∣∇ u ∣ > 0.
To recover the patch from these curves, set Ω j = Int ( Γ j ) \Omega_j = \operatorname{Int}(\Gamma_j) Ω j = Int ( Γ j ) and recall from lemma 3.8 that each interior lies in the positive phase. Since two disjoint Jordan curves can have overlapping interiors only when one lies inside the other, Jordan separation gives
Ω j ⊂ A , u ∣ Γ j = 0 , u ∣ Ω j > 0 , \Omega_j \subset A,\qquad u|_{\Gamma_j} = 0,\qquad u|_{\Omega_j} > 0, Ω j ⊂ A , u ∣ Γ j = 0 , u ∣ Ω j > 0 ,
i ≠ j , Ω i ∩ Ω j ≠ ∅ ⟹ Γ i ⊂ Ω j or Γ j ⊂ Ω i ⟹ ( Γ i ∪ Γ j ) ∩ A ≠ ∅ . i \ne j,\quad\Omega_i \cap\Omega_j \ne\varnothing\Longrightarrow\Gamma_i \subset\Omega_j\ \text{or}\ \Gamma_j \subset\Omega_i \Longrightarrow(\Gamma_i \cup\Gamma_j) \cap A \ne\varnothing. i = j , Ω i ∩ Ω j = ∅ ⟹ Γ i ⊂ Ω j or Γ j ⊂ Ω i ⟹ ( Γ i ∪ Γ j ) ∩ A = ∅ .
The identity ( Γ i ∪ Γ j ) ∩ A = ∅ (\Gamma_i \cup\Gamma_j) \cap A = \varnothing ( Γ i ∪ Γ j ) ∩ A = ∅ excludes such nesting, so the interiors are pairwise disjoint; moreover, each Ω j \Omega_j Ω j is connected and both open and closed relative to A A A , whence
Ω j ‾ ∩ A = Ω j , Ω j is open and connected ⟹ Ω j is a connected component of A . \overline{\Omega_j} \cap A = \Omega_j,\qquad\Omega_j\ \text{is open and connected}\quad\Longrightarrow\quad\Omega_j\ \text{is a connected component of }A. Ω j ∩ A = Ω j , Ω j is open and connected ⟹ Ω j is a connected component of A .
It remains to exclude a component of A A A not represented by these interiors. For any component A 0 A_0 A 0 of A A A , openness and boundedness give
∅ ≠ ∂ A 0 ⊂ ∂ A = ⋃ j = 1 N Γ j . \varnothing\ne\partial A_0 \subset\partial A = \bigcup_{j=1}^{N} \Gamma_j. ∅ = ∂ A 0 ⊂ ∂ A = j = 1 ⋃ N Γ j .
Choose x ∈ ∂ A 0 ∩ Γ j x \in\partial A_0 \cap\Gamma_j x ∈ ∂ A 0 ∩ Γ j and a regular coordinate neighborhood U U U of x x x , where the positive side agrees with Ω j \Omega_j Ω j . Consequently,
∅ ≠ U ∩ A 0 ⊂ U ∩ A = U ∩ Ω j ⟹ A 0 = Ω j , A = ⋃ j = 1 N Ω j . \varnothing\ne U \cap A_0 \subset U \cap A = U \cap\Omega_j \quad\Longrightarrow\quad A_0 = \Omega_j,\qquad A = \bigcup_{j=1}^{N} \Omega_j. ∅ = U ∩ A 0 ⊂ U ∩ A = U ∩ Ω j ⟹ A 0 = Ω j , A = j = 1 ⋃ N Ω j .
Each Ω j \Omega_j Ω j is simply connected by the Jordan–Schoenflies theorem. To identify the zero set outside the patch, observe that the pairwise disjoint compact sets Ω j ‾ \overline{\Omega_j} Ω j lie in D D D and have smooth boundaries admitting disjoint exterior collars. A path in D D D can therefore be perturbed to meet these boundaries transversely and detoured through the collars whenever it crosses an interior, proving that E : = D ∖ A ‾ E := D \setminus\overline{A} E := D ∖ A is connected. On this exterior phase,
∂ E = ∂ D ∪ ⋃ j = 1 N Γ j , Δ u = 0 in E , u ∣ ∂ D = − q < 0 , u ∣ Γ j = 0 , \partial E = \partial D \cup\bigcup_{j=1}^{N} \Gamma_j,\qquad\Delta u = 0\ \text{in }E,\qquad u|_{\partial D} = -q < 0,\qquad u|_{\Gamma_j} = 0, ∂ E = ∂ D ∪ j = 1 ⋃ N Γ j , Δ u = 0 in E , u ∣ ∂ D = − q < 0 , u ∣ Γ j = 0 ,
strong maximum principle ⟹ u < 0 in E ⟹ { u = 0 } = ∂ A . \text{strong maximum principle}\quad\Longrightarrow\quad u < 0\ \text{in }E\quad\Longrightarrow\quad\{u = 0\} = \partial A. strong maximum principle ⟹ u < 0 in E ⟹ { u = 0 } = ∂ A .
Since the gradient has a positive minimum on the compact boundary of the fixed patch, the local half-plane approximation in lemma 3.1 now yields a uniform density estimate.
Corollary 4.2. For each 0 < α < 1 0 < \alpha< 1 0 < α < 1 there are C , r ∗ > 0 C, r_\ast> 0 C , r ∗ > 0 , depending on the fixed patch, such that
sup p ∈ ∂ A ∣ θ A ( p , r ) − 1 2 ∣ ≤ C r α ( 0 < r < r ∗ ) . (48) \sup_{p \in\partial A} \left|\theta_A(p,r) - \frac{1}{2}\right| \le Cr^\alpha\qquad(0 < r < r_\ast). \tag*{(48)} p ∈ ∂ A sup θ A ( p , r ) − 2 1 ≤ C r α ( 0 < r < r ∗ ) . (48) In particular,
lim r ↓ 0 θ A ( p , r ) = 1 2 ( p ∈ ∂ A ) , (49) \lim_{r \downarrow0} \theta_A(p,r) = \frac{1}{2} \qquad(p \in\partial A), \tag*{(49)} r ↓ 0 lim θ A ( p , r ) = 2 1 ( p ∈ ∂ A ) , (49) and
∂ A ∩ ( A ( 0 ) ∪ A ( 1 ) ) = ∅ , ∂ A = ∂ m t A . (50) \partial A \cap\left(A^{(0)} \cup A^{(1)}\right) = \varnothing,\qquad\partial A = \partial_{\mathrm{mt}} A. \tag*{(50)} ∂ A ∩ ( A ( 0 ) ∪ A ( 1 ) ) = ∅ , ∂ A = ∂ mt A . (50)
Proof. Choose r 0 > 0 r_0 > 0 r 0 > 0 so small that the compact neighborhood
K : = { x ∈ R 2 : dist ( x , ∂ A ) ≤ 2 r 0 } ⊂ D . K := \{x \in\mathbb{R}^2 : \operatorname{dist}(x,\partial A) \le2r_0\} \subset D. K := { x ∈ R 2 : dist ( x , ∂ A ) ≤ 2 r 0 } ⊂ D .
By proposition 4.1 and the interior regularity in (22) ,
g 0 : = min ∂ A ∣ ∇ u ∣ > 0 , L : = [ ∇ u ] C 0 , α ( K ) < ∞ . g_0 := \min_{\partial A} |\nabla u| > 0,\qquad L := [\nabla u]_{C^{0,\alpha}(K)} < \infty. g 0 := ∂ A min ∣∇ u ∣ > 0 , L := [ ∇ u ] C 0 , α ( K ) < ∞.
Choose 0 < r ∗ ≤ r 0 0 < r_\ast\le r_0 0 < r ∗ ≤ r 0 . Lemma 3.1 gives
sup p ∈ ∂ A ∣ θ A ( p , r ) − 1 2 ∣ ≤ 4 L π g 0 r α ( 0 < r < r ∗ ) . \sup_{p \in\partial A} \left|\theta_A(p,r) - \frac{1}{2}\right| \le\frac{4L}{\pi g_0}r^\alpha\qquad(0 < r < r_\ast). p ∈ ∂ A sup θ A ( p , r ) − 2 1 ≤ π g 0 4 L r α ( 0 < r < r ∗ ) .
The limiting density 1 / 2 1/2 1/2 places every topological boundary point in the measure-theoretic boundary, while (17) supplies the reverse inclusion; hence
∂ A ⊂ ∂ m t A ⊂ ∂ A , ∂ A ∩ ( A ( 0 ) ∪ A ( 1 ) ) = ∅ . \partial A \subset\partial_{\mathrm{mt}} A \subset\partial A,\qquad\partial A \cap\left(A^{(0)} \cup A^{(1)}\right) = \varnothing. ∂ A ⊂ ∂ mt A ⊂ ∂ A , ∂ A ∩ ( A ( 0 ) ∪ A ( 1 ) ) = ∅ .
□
Proof of theorem 1.1. To transfer the preceding conclusions from the normalized problem to circulation one, we use proposition 2.1 together with the energy normalization:
m = λ − 1 , ω λ = λ 1 A λ , E ( 1 A λ ) = λ − 2 E k i n ( ω λ ) = max ρ ∈ A m E ( ρ ) , m = \lambda^{-1},\qquad\omega_\lambda= \lambda1_{A_\lambda},\qquad\mathcal{E}(1_{A_\lambda}) = \lambda^{-2}\mathcal{E}_{\mathrm{kin}}(\omega_\lambda) = \max_{\rho\in\mathcal{A}_m} \mathcal{E}(\rho), m = λ − 1 , ω λ = λ 1 A λ , E ( 1 A λ ) = λ − 2 E kin ( ω λ ) = ρ ∈ A m max E ( ρ ) ,
q λ = μ λ / λ > 0 , u λ = G 1 A λ − q λ , A λ = { u λ > 0 } ⊂ D . q_\lambda= \mu_\lambda/\lambda> 0,\qquad u_\lambda= \mathcal{G}1_{A_\lambda} - q_\lambda,\qquad A_\lambda= \{u_\lambda> 0\} \subset D. q λ = μ λ / λ > 0 , u λ = G 1 A λ − q λ , A λ = { u λ > 0 } ⊂ D .
Applying proposition 4.1 and corollary 4.2 to this normalized maximizer gives all the asserted conclusions.
□
Remark 4.3. Simple connectedness also follows directly from the equation, without using boundary regularity or finiteness: by the planar Jordan-curve criterion, it suffices to show that every smooth Jordan curve in a component encloses only points of that component. Given a component A j A_j A j and such a curve γ ⊂ A j \gamma\subset A_j γ ⊂ A j , simple connectedness of D D D places its closed interior in D D D , allowing us to set
Ω : = Int ( γ ) , Ω ‾ ⊂ D , c γ : = min γ u > 0 , η = ( c γ − u ) + ∈ H 0 1 ( Ω ) . \Omega:= \operatorname{Int}(\gamma),\qquad\overline{\Omega} \subset D,\qquad c_\gamma:= \min_{\gamma} u > 0,\qquad\eta= (c_\gamma- u)_+ \in H^1_0(\Omega). Ω := Int ( γ ) , Ω ⊂ D , c γ := γ min u > 0 , η = ( c γ − u ) + ∈ H 0 1 ( Ω ) .
The positive minimum on γ \gamma γ makes η \eta η vanish on ∂ Ω \partial\Omega ∂ Ω , so it is an admissible weak test function. Testing − Δ u = 1 A -\Delta u = \mathbf{1}_A − Δ u = 1 A gives
− ∫ Ω ∣ ∇ η ∣ 2 d x = ∫ Ω 1 A η d x ≥ 0 ⟹ u ≥ c γ > 0 on Ω ‾ . -\int_{\Omega}|\nabla\eta|^2\,dx=\int_{\Omega}\mathbf{1}_A\eta\,dx\ge0\quad\Longrightarrow\quad u\ge c_\gamma>0\quad\text{on }\overline{\Omega}. − ∫ Ω ∣∇ η ∣ 2 d x = ∫ Ω 1 A η d x ≥ 0 ⟹ u ≥ c γ > 0 on Ω .
Continuity extends the inequality throughout Ω ‾ \overline{\Omega} Ω ; since this set is connected and meets A j A_j A j along γ \gamma γ , we obtain
Ω ‾ ⊂ A ⟹ Ω ‾ ⊂ A j . \overline{\Omega}\subset A\quad\Longrightarrow\quad\overline{\Omega}\subset A_j. Ω ⊂ A ⟹ Ω ⊂ A j .
The planar Jordan-curve criterion gives simple connectedness of A j A_j A j .
Dirichlet zero modes and convexity
In this section and § 6 , D D D is a bounded simply connected domain with C ∞ C^\infty C ∞ boundary; convexity is imposed where stated. The extension of fixed-mass finiteness and compact-interval bounds to arbitrary bounded convex domains is proved in § 7 . The regularity theorem above applies to every bounded Jordan domain. We combine the fixed-mass trace inequality with a nodal argument as in [30 ] . Fix a maximizing patch A A A , and write Γ = ∂ A \Gamma=\partial A Γ = ∂ A , u = G 1 A − q u=G\mathbf{1}_A-q u = G 1 A − q , and g = ∣ ∇ u ∣ g=|\nabla u| g = ∣∇ u ∣ on Γ \Gamma Γ . Theorem 1.1 gives min Γ g > 0 \min_\Gamma g>0 min Γ g > 0 . Define the single-layer trace and the constrained quadratic form by
S Γ ϕ = G ( ϕ d s ∣ Γ ) ∣ Γ , X Γ = { ϕ ∈ L 2 ( Γ ) : ∫ Γ ϕ d s = 0 } , (51) S_\Gamma\phi=\mathcal{G}(\phi\,ds|_\Gamma)|_\Gamma,\qquad X_\Gamma=\left\{\phi\in L^2(\Gamma):\int_\Gamma\phi\,ds=0\right\}, \tag*{(51)} S Γ ϕ = G ( ϕ d s ∣ Γ ) ∣ Γ , X Γ = { ϕ ∈ L 2 ( Γ ) : ∫ Γ ϕ d s = 0 } , (51)
Q A ( ϕ ) = ∫ Γ g ϕ 2 d s − ∫ Γ ϕ S Γ ϕ d s ≥ 0 ( ϕ ∈ X Γ ) . (52) Q_A(\phi)=\int_\Gamma g\phi^2\,ds-\int_\Gamma\phi S_\Gamma\phi\,ds\ge0\qquad(\phi\in X_\Gamma). \tag*{(52)} Q A ( ϕ ) = ∫ Γ g ϕ 2 d s − ∫ Γ ϕ S Γ ϕ d s ≥ 0 ( ϕ ∈ X Γ ) . (52)
The inequality is proposition 3.3 on the full boundary. Set
K A = { ϕ ∈ X Γ : Q A ( ϕ ) = 0 } , K A 0 = { ϕ ∈ X Γ : S Γ ϕ = g ϕ } . K_A=\{\phi\in X_\Gamma:Q_A(\phi)=0\},\qquad K_A^0=\{\phi\in X_\Gamma:S_\Gamma\phi=g\phi\}. K A = { ϕ ∈ X Γ : Q A ( ϕ ) = 0 } , K A 0 = { ϕ ∈ X Γ : S Γ ϕ = g ϕ } .
Polarization on X Γ X_\Gamma X Γ gives
ϕ ∈ K A ⟹ S Γ ϕ − g ϕ = c for a constant c . (53) \phi\in K_A\quad\Longrightarrow\quad S_\Gamma\phi-g\phi=c\quad\text{for a constant }c. \tag*{(53)} ϕ ∈ K A ⟹ S Γ ϕ − g ϕ = c for a constant c . (53)
Write c A ( ϕ ) = c c_A(\phi)=c c A ( ϕ ) = c ; then K A 0 = ker c A K_A^0=\ker c_A K A 0 = ker c A . We call these the threshold-preserving zero modes, since their first-order variation leaves the level q q q fixed; see (60) .
Lemma 5.1. Every ϕ ∈ K A \phi\in K_A ϕ ∈ K A is smooth. For 0 < α < 1 0<\alpha<1 0 < α < 1 ,
∥ ϕ ∥ C 1 , α ( Γ ) + ∣ c A ( ϕ ) ∣ ≤ C A , α ∥ ϕ ∥ L 2 ( Γ ) . \|\phi\|_{C^{1,\alpha}(\Gamma)}+|c_A(\phi)|\le C_{A,\alpha}\|\phi\|_{L^2(\Gamma)}. ∥ ϕ ∥ C 1 , α ( Γ ) + ∣ c A ( ϕ ) ∣ ≤ C A , α ∥ ϕ ∥ L 2 ( Γ ) .
Proof. The logarithmic Green kernel on a fixed smooth curve satisfies
sup x ∈ Γ ∥ G D ( x , ⋅ ) ∥ L 2 ( Γ ) ≤ C , \sup_{x\in\Gamma}\|G_D(x,\cdot)\|_{L^2(\Gamma)}\le C, x ∈ Γ sup ∥ G D ( x , ⋅ ) ∥ L 2 ( Γ ) ≤ C ,
∥ G D ( x , ⋅ ) − G D ( y , ⋅ ) ∥ L 2 ( Γ ) 2 ≤ C δ ( 1 + ∣ log δ ∣ ) 2 , δ = ∣ x − y ∣ ≪ 1. \|G_D(x,\cdot)-G_D(y,\cdot)\|_{L^2(\Gamma)}^2\le C\delta(1+|\log\delta|)^2,\qquad\delta=|x-y|\ll1. ∥ G D ( x , ⋅ ) − G D ( y , ⋅ ) ∥ L 2 ( Γ ) 2 ≤ C δ ( 1 + ∣ log δ ∣ ) 2 , δ = ∣ x − y ∣ ≪ 1.
The second estimate follows by splitting the local arcs at scale δ \delta δ : the near part is O ( δ ( 1 + ∣ log δ ∣ ) 2 ) O(\delta(1+|\log\delta|)^2) O ( δ ( 1 + ∣ log δ ∣ ) 2 ) and the far part is bounded by C δ 2 ∫ δ 1 r − 2 d r C\delta^2\int_\delta^1r^{-2}\,dr C δ 2 ∫ δ 1 r − 2 d r . Thus, for 0 < β < 1 / 2 0<\beta<1/2 0 < β < 1/2 ,
∥ S Γ ϕ ∥ C 0 , β ( Γ ) ≤ C β ∥ ϕ ∥ 2 , \|S_\Gamma\phi\|_{C^{0,\beta}(\Gamma)}\le C_\beta\|\phi\|_2, ∥ S Γ ϕ ∥ C 0 , β ( Γ ) ≤ C β ∥ ϕ ∥ 2 ,
c A ( ϕ ) = 1 H 1 ( Γ ) ∫ Γ ( S Γ ϕ − g ϕ ) d s , ∣ c A ( ϕ ) ∣ ≤ C ∥ ϕ ∥ 2 . c_A(\phi)=\frac{1}{\mathcal{H}^1(\Gamma)}\int_{\Gamma}(S_\Gamma\phi-g\phi)\,\mathrm{d}s,\qquad|c_A(\phi)|\leq C\|\phi\|_2. c A ( ϕ ) = H 1 ( Γ ) 1 ∫ Γ ( S Γ ϕ − g ϕ ) d s , ∣ c A ( ϕ ) ∣ ≤ C ∥ ϕ ∥ 2 .
One-sided Dirichlet regularity gives g ∈ C ∞ ( Γ ) g\in C^\infty(\Gamma) g ∈ C ∞ ( Γ ) and min Γ g > 0 \min_\Gamma g>0 min Γ g > 0 . Consequently,
ϕ = S Γ ϕ − c A ( ϕ ) g , S Γ : C k , β ( Γ ) ⟶ C k + 1 , β ( Γ ) ( k ≥ 0 ) \phi=\frac{S_\Gamma\phi-c_A(\phi)}{g},\qquad S_\Gamma:C^{k,\beta}(\Gamma)\longrightarrow C^{k+1,\beta}(\Gamma)\quad(k\geq0) ϕ = g S Γ ϕ − c A ( ϕ ) , S Γ : C k , β ( Γ ) ⟶ C k + 1 , β ( Γ ) ( k ≥ 0 )
imply ϕ ∈ C ∞ ( Γ ) \phi\in C^\infty(\Gamma) ϕ ∈ C ∞ ( Γ ) . Quantitatively, fix β ∈ ( 0 , 1 / 2 ) \beta\in(0,1/2) β ∈ ( 0 , 1/2 ) and use
∥ ϕ ∥ C 0 , α ≤ C α ∥ ϕ ∥ C 1 , β ≤ C A , α ∥ ϕ ∥ 2 , ∥ ϕ ∥ C 1 , α ≤ C A , α ( ∥ ϕ ∥ C 0 , α + ∣ c A ( ϕ ) ∣ ) . \|\phi\|_{C^{0,\alpha}}\leq C_\alpha\|\phi\|_{C^{1,\beta}}\leq C_{A,\alpha}\|\phi\|_2,\qquad\|\phi\|_{C^{1,\alpha}}\leq C_{A,\alpha}(\|\phi\|_{C^{0,\alpha}}+|c_A(\phi)|). ∥ ϕ ∥ C 0 , α ≤ C α ∥ ϕ ∥ C 1 , β ≤ C A , α ∥ ϕ ∥ 2 , ∥ ϕ ∥ C 1 , α ≤ C A , α ( ∥ ϕ ∥ C 0 , α + ∣ c A ( ϕ ) ∣ ) .
□ \square □
For ϕ ∈ K A 0 \phi\in K_A^0 ϕ ∈ K A 0 , its potential v = G ( ϕ d s ∣ Γ ) v=\mathcal{G}(\phi\,\mathrm{d}s|_\Gamma) v = G ( ϕ d s ∣ Γ ) satisfies
v ∈ H 0 1 ( D ) , ∫ D ∇ v ⋅ ∇ z d x = ∫ Γ v z g d s ( z ∈ H 0 1 ( D ) ) , ∫ Γ v g d s = 0. (54) v\in H_0^1(D),\qquad\int_D\nabla v\cdot\nabla z\,\mathrm{d}x=\int_\Gamma\frac{vz}{g}\,\mathrm{d}s\quad(z\in H_0^1(D)),\qquad\int_\Gamma\frac{v}{g}\,\mathrm{d}s=0. \tag*{(54)} v ∈ H 0 1 ( D ) , ∫ D ∇ v ⋅ ∇ z d x = ∫ Γ g v z d s ( z ∈ H 0 1 ( D )) , ∫ Γ g v d s = 0. (54)
Conversely, (54) gives ϕ = v / g ∈ K A 0 \phi=v/g\in K_A^0 ϕ = v / g ∈ K A 0 . By lemma 5.1, v v v is continuous on D ‾ \overline{D} D , smooth up to each side of Γ \Gamma Γ , and harmonic in D ∖ Γ ‾ D\setminus\overline{\Gamma} D ∖ Γ .
We now use the trace equation to control the nodal domains of its potential and the zeros of its outer normal derivative.
Lemma 5.2. Every nonzero solution of (54) has exactly two nodal domains.
Proof. The trace of v v v takes both signs by (54) . For a nodal domain O i O_i O i of sign s i ∈ { − 1 , 1 } s_i\in\{-1,1\} s i ∈ { − 1 , 1 } and ε > 0 \varepsilon>0 ε > 0 , continuity of v v v on D ‾ \overline{D} D and its zero boundary values imply that
K i , ε : = { x ∈ O i ‾ : s i v ( x ) ≥ ε } K_{i,\varepsilon}:=\{x\in\overline{O_i}:s_i v(x)\geq\varepsilon\} K i , ε := { x ∈ O i : s i v ( x ) ≥ ε } is a compact subset of O i O_i O i .
Choose χ i , ε ∈ C ∞ ( O i ) \chi_{i,\varepsilon}\in C^\infty(O_i) χ i , ε ∈ C ∞ ( O i ) with 0 ≤ χ i , ε ≤ 1 0\leq\chi_{i,\varepsilon}\leq1 0 ≤ χ i , ε ≤ 1 and χ i , ε = 1 \chi_{i,\varepsilon}=1 χ i , ε = 1 in a neighborhood of K i , ε K_{i,\varepsilon} K i , ε . This cutoff realizes the nodal truncation without assuming any regularity of ∂ O i \partial O_i ∂ O i :
v i , ε : = s i ( s i v − ε ) + χ i , ε = s i ( s i v − ε ) + 1 O i ∈ H 0 1 ( D ) , v_{i,\varepsilon}:=s_i(s_i v-\varepsilon)_+\chi_{i,\varepsilon}=s_i(s_i v-\varepsilon)_+1_{O_i}\in H_0^1(D), v i , ε := s i ( s i v − ε ) + χ i , ε = s i ( s i v − ε ) + 1 O i ∈ H 0 1 ( D ) , ∇ v i , ε = 1 O i ∩ { s i v > ε } ∇ v . \nabla v_{i,\varepsilon}=1_{O_i\cap\{s_i v>\varepsilon\}}\nabla v. ∇ v i , ε = 1 O i ∩ { s i v > ε } ∇ v . Indeed, the term containing ∇ χ i , ε \nabla\chi_{i,\varepsilon} ∇ χ i , ε vanishes where the positive part is nonzero. For v i : = v 1 O i v_i:=v1_{O_i} v i := v 1 O i ,
∥ v i , ε − v i ∥ L 2 ( D ) ≤ ε ∣ D ∣ 1 / 2 , \|v_{i,\varepsilon}-v_i\|_{L^2(D)}\leq\varepsilon|D|^{1/2}, ∥ v i , ε − v i ∥ L 2 ( D ) ≤ ε ∣ D ∣ 1/2 , ∥ ∇ v i , ε − 1 O i ∇ v ∥ L 2 ( D ) 2 = ∫ O i ∩ { 0 < s i v ≤ ε } ∣ ∇ v ∣ 2 d x ⟶ 0. \|\nabla v_{i,\varepsilon}-1_{O_i}\nabla v\|_{L^2(D)}^2=\int_{O_i\cap\{0<s_i v\leq\varepsilon\}}|\nabla v|^2\,\mathrm{d}x\longrightarrow0. ∥∇ v i , ε − 1 O i ∇ v ∥ L 2 ( D ) 2 = ∫ O i ∩ { 0 < s i v ≤ ε } ∣∇ v ∣ 2 d x ⟶ 0. Thus the truncations converge strongly in H 1 ( D ) H^1(D) H 1 ( D ) , and
v i ∈ H 0 1 ( D ) , ∇ v i = 1 O i ∇ v , v_i\in H_0^1(D),\qquad\nabla v_i=1_{O_i}\nabla v, v i ∈ H 0 1 ( D ) , ∇ v i = 1 O i ∇ v , so testing (54) with v i v_i v i gives
∫ D ∣ ∇ v i ∣ 2 d x = ∫ Γ v i 2 g d s > 0 , I i : = ∫ Γ v i g d s ≠ 0. \int_D|\nabla v_i|^2\,\mathrm{d}x=\int_\Gamma\frac{v_i^2}{g}\,\mathrm{d}s>0,\qquad I_i:=\int_\Gamma\frac{v_i}{g}\,\mathrm{d}s\neq0. ∫ D ∣∇ v i ∣ 2 d x = ∫ Γ g v i 2 d s > 0 , I i := ∫ Γ g v i d s = 0. If three nodal domains existed, put w = I 2 v 1 − I 1 v 2 w=I_2v_1-I_1v_2 w = I 2 v 1 − I 1 v 2 . Then w ≠ 0 w\neq0 w = 0 , w = 0 w=0 w = 0 on O 3 O_3 O 3 , and
∫ Γ w g d s = 0 , ∫ D ∣ ∇ w ∣ 2 d x = ∫ Γ w 2 g d s . \int_\Gamma\frac{w}{g}\,\mathrm{d}s=0,\qquad\int_D|\nabla w|^2\,\mathrm{d}x=\int_\Gamma\frac{w^2}{g}\,\mathrm{d}s. ∫ Γ g w d s = 0 , ∫ D ∣∇ w ∣ 2 d x = ∫ Γ g w 2 d s .
Polarizing (38) at w w w gives
∫ D ∇ w ⋅ ∇ z d x = ∫ Γ w ( z − z ˉ ) / g d s = ∫ Γ w z / g d s ( z ∈ H 0 1 ( D ) ) . \int_D \nabla w \cdot\nabla z\,\mathrm{d}x = \int_\Gamma w(z-\bar{z})/g\,\mathrm{d}s = \int_\Gamma wz/g\,\mathrm{d}s \qquad(z \in H^1_0(D)). ∫ D ∇ w ⋅ ∇ z d x = ∫ Γ w ( z − z ˉ ) / g d s = ∫ Γ w z / g d s ( z ∈ H 0 1 ( D )) .
Here z ˉ = ( ∫ Γ z / g ) / ( ∫ Γ 1 / g ) \bar{z} = \left(\int_\Gamma z/g\right)/\left(\int_\Gamma1/g\right) z ˉ = ( ∫ Γ z / g ) / ( ∫ Γ 1/ g ) .
If w w w vanishes on an open subset of a phase component, harmonic unique continuation makes it vanish throughout that component. On each adjacent interface arc its trace, hence the source w / g w/g w / g , is zero. The weak equation then makes w w w harmonic across the arc, propagating the zero to the next component. The finitely many Jordan interiors and their connected exterior have a connected adjacency graph. Since O 3 ∖ Γ O_3 \setminus\Gamma O 3 ∖ Γ contains an open set, w ≡ 0 w \equiv0 w ≡ 0 in D D D , a contradiction. □
Proposition 5.3. If 0 ≠ ϕ ∈ K A 0 0 \ne\phi\in K^0_A 0 = ϕ ∈ K A 0 and v = G ( ϕ d s ∣ Γ ) v = \mathcal{G}(\phi\,\mathrm{d}s|_\Gamma) v = G ( ϕ d s ∣ Γ ) , then ∂ n v \partial_n v ∂ n v has exactly two zeros on ∂ D \partial D ∂ D . Both are simple in the boundary arclength parameter. In particular, dim K A 0 ≤ 2 \dim K^0_A \le2 dim K A 0 ≤ 2 .
Proof. Let F : D → D F:\mathbb{D} \to D F : D → D be conformal. Its smooth boundary extension has F ′ ≠ 0 F' \ne0 F ′ = 0 on ∂ D \partial D ∂ D [32 ] . The function w = v ∘ F w = v \circ F w = v ∘ F is harmonic in an annular collar and vanishes on the circle. Odd harmonic reflection therefore makes
b ( θ ) : = ∂ r w ( e i θ ) = ∣ F ′ ( e i θ ) ∣ ∂ n v ( F ( e i θ ) ) b(\theta) := \partial_r w(e^{i\theta}) = \lvert F'(e^{i\theta})\rvert\,\partial_n v(F(e^{i\theta})) b ( θ ) := ∂ r w ( e i θ ) = ∣ F ′ ( e i θ )∣ ∂ n v ( F ( e i θ ))
real analytic. Moreover,
∫ 0 2 π b ( θ ) d θ = ∫ ∂ D ∂ n v d s = − ∫ Γ ϕ d s = 0. \int_0^{2\pi} b(\theta)\,\mathrm{d}\theta= \int_{\partial D} \partial_n v\,\mathrm{d}s = -\int_\Gamma\phi\,\mathrm{d}s = 0. ∫ 0 2 π b ( θ ) d θ = ∫ ∂ D ∂ n v d s = − ∫ Γ ϕ d s = 0.
If b ≡ 0 b \equiv0 b ≡ 0 , zero Cauchy data and the continuation argument in lemma 5.2 give v ≡ 0 v \equiv0 v ≡ 0 . Thus b b b takes both signs and has finitely many zeros. The conformal homeomorphism preserves the two nodal domains of v v v .
Suppose b b b has a zero of order k ≥ 2 k \ge2 k ≥ 2 at p p p on the circle. In a local conformal coordinate ζ = T ( x ) \zeta= T(x) ζ = T ( x ) mapping the disk to the upper half-plane and p p p to zero, reflection gives
w ∘ T − 1 ( ζ ) = c Im ( ζ k + 1 ) + O ( ∣ ζ ∣ k + 2 ) , c ∈ R ∖ { 0 } . w \circ T^{-1}(\zeta) = c\,\operatorname{Im}(\zeta^{k+1}) + O(\lvert\zeta\rvert^{k+2}), \qquad c \in\mathbb{R} \setminus\{0\}. w ∘ T − 1 ( ζ ) = c Im ( ζ k + 1 ) + O (∣ ζ ∣ k + 2 ) , c ∈ R ∖ { 0 } .
There are at least three alternating sign sectors. Choose rays 0 < θ 1 < θ 2 < π 0 < \theta_1 < \theta_2 < \pi 0 < θ 1 < θ 2 < π in two sectors of the same sign s s s , enclosing an opposite-sign sector. For small r > 0 r > 0 r > 0 , the arcs
α i ( t ) = T − 1 ( t e i θ i ) , 0 ≤ t ≤ r , i = 1 , 2 , \alpha_i(t) = T^{-1}(te^{i\theta_i}), \qquad0 \le t \le r,\qquad i=1,2, α i ( t ) = T − 1 ( t e i θ i ) , 0 ≤ t ≤ r , i = 1 , 2 ,
meet only at p p p and otherwise lie in the unique sign-s s s nodal domain N N N . Choose a simple path η : [ 0 , 1 ] → N \eta:[0,1] \to N η : [ 0 , 1 ] → N joining α 1 ( r ) \alpha_1(r) α 1 ( r ) to α 2 ( r ) \alpha_2(r) α 2 ( r ) , and trim it at
t 2 = min { t : η ( t ) ∈ α 2 ( [ 0 , r ] ) } , t_2 = \min\{t:\eta(t) \in\alpha_2([0,r])\}, t 2 = min { t : η ( t ) ∈ α 2 ([ 0 , r ])} ,
t 1 = max { t ≤ t 2 : η ( t ) ∈ α 1 ( [ 0 , r ] ) } , η ( t i ) = α i ( r i ) . t_1 = \max\{t \le t_2:\eta(t) \in\alpha_1([0,r])\}, \qquad\eta(t_i) = \alpha_i(r_i). t 1 = max { t ≤ t 2 : η ( t ) ∈ α 1 ([ 0 , r ])} , η ( t i ) = α i ( r i ) .
Because η \eta η avoids p p p , one has t 1 < t 2 t_1 < t_2 t 1 < t 2 and r i > 0 r_i > 0 r i > 0 . Hence
J = α 1 ( [ 0 , r 1 ] ) ∪ η ( [ t 1 , t 2 ] ) ∪ α 2 ( [ 0 , r 2 ] ) J = \alpha_1([0,r_1]) \cup\eta([t_1,t_2]) \cup\alpha_2([0,r_2]) J = α 1 ([ 0 , r 1 ]) ∪ η ([ t 1 , t 2 ]) ∪ α 2 ([ 0 , r 2 ])
is a Jordan curve with J ∖ { p } ⊂ N J \setminus\{p\} \subset N J ∖ { p } ⊂ N and J ∩ ∂ D = { p } J \cap\partial D = \{p\} J ∩ ∂ D = { p } . Its interior Ω J \Omega_J Ω J lies in the disk. Near p p p , the complementary sector outside ( θ 1 , θ 2 ) (\theta_1,\theta_2) ( θ 1 , θ 2 ) contains points outside the disk and is therefore on the unbounded side of J J J . The intervening sign-− s -s − s sector lies in Ω J \Omega_J Ω J . Since b b b takes both signs, there is a compact boundary subarc I I I away from p p p on which s b > 0 sb>0 s b > 0 . Uniformly for ξ ∈ I \xi\in I ξ ∈ I ,
w ( ( 1 − δ ) ξ ) = − δ b ( ξ ) + O ( δ 2 ) , δ ↓ 0. w((1-\delta)\xi)=-\delta b(\xi)+O(\delta^2),\qquad\delta\downarrow0. w (( 1 − δ ) ξ ) = − δ b ( ξ ) + O ( δ 2 ) , δ ↓ 0.
These collar points have sign − s -s − s and lie outside J J J , since dist ( I , J ) > 0 \operatorname{dist}(I,J)>0 dist ( I , J ) > 0 . A path joining them to the sign-− s -s − s sector inside J J J would cross J ∖ { p } ⊂ N J\setminus\{p\}\subset N J ∖ { p } ⊂ N . This contradicts the uniqueness of the sign-− s -s − s nodal domain. Every zero of b b b is simple.
If there were at least four zeros, choose cyclically ordered boundary points ξ 1 , … , ξ 4 \xi_1,\ldots,\xi_4 ξ 1 , … , ξ 4 whose inner collars have signs s , − s , s , − s s,-s,s,-s s , − s , s , − s . Join short inward radial segments at ξ 1 , ξ 3 \xi_1,\xi_3 ξ 1 , ξ 3 by a simple path in N N N , and apply the same first-hit/last-hit trimming. This gives a simple crosscut C C C joining ξ 1 \xi_1 ξ 1 to ξ 3 \xi_3 ξ 3 , with its interior in N N N . Adjoining either circle arc between the endpoints gives a Jordan curve; therefore C C C separates the two boundary arcs, and the sign-− s -s − s collar points at ξ 2 , ξ 4 \xi_2,\xi_4 ξ 2 , ξ 4 lie in different components of D ∖ C D\setminus C D ∖ C . They cannot lie in one sign-− s -s − s nodal domain, a contradiction. Thus b b b has exactly two simple zeros. The smooth boundary parametrization and the positive factor ∣ F ′ ∣ |F'| ∣ F ′ ∣ preserve their simplicity for ∂ n v \partial_n v ∂ n v .
For fixed θ 0 \theta_0 θ 0 , the linear map
K A 0 ⟶ R 2 , ϕ ⟼ ( b ( θ 0 ) , b ′ ( θ 0 ) ) K_A^0 \longrightarrow\mathbb{R}^2,\qquad\phi\longmapsto(b(\theta_0),b'(\theta_0)) K A 0 ⟶ R 2 , ϕ ⟼ ( b ( θ 0 ) , b ′ ( θ 0 ))
is injective, since a nonzero element of its kernel would give a multiple boundary zero. Hence dim K A 0 ≤ 2 \dim K_A^0\le2 dim K A 0 ≤ 2 .
The nodal information becomes an obstruction to zero modes once it is combined with the following bilinear Pohozaev identity.
Lemma 5.4. Let U = G 1 A U=\mathcal{G}1_A U = G 1 A , P = − ∂ n U > 0 P=-\partial_nU>0 P = − ∂ n U > 0 on ∂ D \partial D ∂ D , and let ϕ ∈ K A 0 \phi\in K_A^0 ϕ ∈ K A 0 , v = G ( ϕ d s ∣ Γ ) v=\mathcal{G}(\phi\,ds|_\Gamma) v = G ( ϕ d s ∣ Γ ) , b = ∂ n v b=\partial_n v b = ∂ n v . For every smooth vector field Z Z Z on D ‾ \overline{D} D satisfying D Z + ( D Z ) T = ( div Z ) I DZ+(DZ)^T=(\operatorname{div}Z)I D Z + ( D Z ) T = ( div Z ) I , one has
∫ ∂ D P b ( Z ⋅ n ) d s = − ∫ A v div Z d x . (55) \int_{\partial D}Pb(Z\cdot n)\,ds=-\int_Av\operatorname{div}Z\,dx. \tag*{(55)} ∫ ∂ D P b ( Z ⋅ n ) d s = − ∫ A v div Z d x . (55) In particular,
∫ ∂ D P b ( a + σ x ) ⋅ n d s = 0 ( a ∈ R 2 , σ ∈ R ) . (56) \int_{\partial D}Pb(a+\sigma x)\cdot n\,ds=0\qquad(a\in\mathbb{R}^2,\ \sigma\in\mathbb{R}). \tag*{(56)} ∫ ∂ D P b ( a + σ x ) ⋅ n d s = 0 ( a ∈ R 2 , σ ∈ R ) . (56)
Proof. For the bilinear stress vector
B = ( Z ⋅ ∇ U ) ∇ v + ( Z ⋅ ∇ v ) ∇ U − ( ∇ U ⋅ ∇ v ) Z , B=(Z\cdot\nabla U)\nabla v+(Z\cdot\nabla v)\nabla U-(\nabla U\cdot\nabla v)Z, B = ( Z ⋅ ∇ U ) ∇ v + ( Z ⋅ ∇ v ) ∇ U − ( ∇ U ⋅ ∇ v ) Z ,
the terms involving first derivatives of Z Z Z cancel. Since − Δ U = 1 A -\Delta U=1_A − Δ U = 1 A , − Δ v = ϕ d s ∣ Γ -\Delta v=\phi\,ds|_\Gamma − Δ v = ϕ d s ∣ Γ , ∇ U = − g ν \nabla U=-g\nu ∇ U = − g ν on Γ \Gamma Γ , and v = g ϕ v=g\phi v = g ϕ there,
div B = − 1 A Z ⋅ ∇ v + v ( Z ⋅ ν ) d s ∣ Γ \operatorname{div}B=-1_AZ\cdot\nabla v+v(Z\cdot\nu)\,ds|_\Gamma div B = − 1 A Z ⋅ ∇ v + v ( Z ⋅ ν ) d s ∣ Γ
in distributions. This also follows by integration in the two phases and the jump condition for the normal derivative of v v v . Integration over D D D and then by parts in A A A gives
∫ ∂ D B ⋅ n d s = ∫ A v div Z d x . \int_{\partial D}B\cdot n\,ds=\int_Av\operatorname{div}Z\,dx. ∫ ∂ D B ⋅ n d s = ∫ A v div Z d x .
On ∂ D \partial D ∂ D , both potentials vanish and B ⋅ n = − P b ( Z ⋅ n ) B \cdot n = -P b(Z \cdot n) B ⋅ n = − P b ( Z ⋅ n ) , which proves (55) . Finally, Green symmetry gives
∫ A v d x = ∫ Γ U ϕ d s = q ∫ Γ ϕ d s = 0. \int_A v\,dx = \int_\Gamma U\phi\,ds = q\int_\Gamma\phi\,ds = 0. ∫ A v d x = ∫ Γ U ϕ d s = q ∫ Γ ϕ d s = 0.
The field Z = a + σ x Z = a + \sigma x Z = a + σ x has constant divergence 2 σ 2\sigma 2 σ , proving (56) .
Lemma 5.5. Let D D D be smooth and convex, and let b ∈ C ( ∂ D ) b \in C(\partial D) b ∈ C ( ∂ D ) have exactly two zeros and opposite signs on the complementary boundary arcs. There is a field Z ( x ) = a + σ x Z(x) = a + \sigma x Z ( x ) = a + σ x such that b ( Z ⋅ n ) ≥ 0 b(Z \cdot n) \ge0 b ( Z ⋅ n ) ≥ 0 on ∂ D \partial D ∂ D , with strict inequality on a nonempty open arc.
Proof. Let p 1 , p 2 p_1,p_2 p 1 , p 2 be the zeros. Write ℓ i \ell_i ℓ i for the supporting tangent line at p i p_i p i and F i = ℓ i ∩ D ‾ F_i = \ell_i \cap\overline{D} F i = ℓ i ∩ D for its contact face, which can be a point or a closed segment.
If ℓ 1 = ℓ 2 \ell_1 = \ell_2 ℓ 1 = ℓ 2 , set z = ( p 1 + p 2 ) / 2 z = (p_1 + p_2)/2 z = ( p 1 + p 2 ) /2 . For Z = x − z Z = x - z Z = x − z , convexity gives Z ⋅ n ≥ 0 Z \cdot n \ge0 Z ⋅ n ≥ 0 , with equality exactly on the common face. Indeed, any supporting line through z z z is its unique tangent line by smoothness. On the complementary boundary arc b b b has a constant nonzero sign. Changing the sign of Z Z Z proves the assertion.
If the supporting lines are distinct and not parallel, let z z z be their intersection. Then z ∉ D ‾ z \notin\overline{D} z ∈ / D , since two distinct supporting lines cannot pass through a smooth boundary point. From this exterior point there are exactly two supporting lines. Thus ( x − z ) ⋅ n (x-z) \cdot n ( x − z ) ⋅ n vanishes precisely on F 1 ∪ F 2 F_1 \cup F_2 F 1 ∪ F 2 . It is negative at a nearest boundary point to z z z and positive at a farthest one, so it has opposite signs on the two open arcs connecting the faces. The signs of b b b are also opposite and constant on these arcs. One overall choice of sign for Z = x − z Z = x-z Z = x − z gives the required comparison.
If the supporting lines are distinct and parallel, their outer normals are opposite. Choose a nonzero constant vector a a a parallel to them. The zero set of a ⋅ n a \cdot n a ⋅ n is exactly F 1 ∪ F 2 F_1 \cup F_2 F 1 ∪ F 2 , and its signs on the two connecting arcs are opposite. The same comparison applies.
Proposition 5.6. In a smooth bounded convex domain, every maximizing patch satisfies K A 0 = { 0 } K_A^0 = \{0\} K A 0 = { 0 } . Consequently, the map ϕ ↦ c \phi\mapsto c ϕ ↦ c in (53) is injective on K A K_A K A , and dim K A ≤ 1 \dim K_A \le1 dim K A ≤ 1 .
Proof. If 0 ≠ ϕ ∈ K A 0 0 \ne\phi\in K_A^0 0 = ϕ ∈ K A 0 , Proposition 5.3 supplies the two sign changes required by Lemma 5.5. The resulting field Z = a + σ x Z = a + \sigma x Z = a + σ x satisfies
∫ ∂ D P b ( Z ⋅ n ) d s > 0 , \int_{\partial D} P b(Z \cdot n)\,ds > 0, ∫ ∂ D P b ( Z ⋅ n ) d s > 0 , because P > 0 P > 0 P > 0 . This contradicts (56) . The assertion about the full constrained kernel follows immediately from (53) .
Analytic reduction and finiteness
Throughout this section, D D D is smooth, bounded, and convex. For A ∈ M m A \in\mathcal{M}_m A ∈ M m , write
U A = G 1 A , q A > 0 , { U A = q A } = ∂ A , g A = − ∂ ν U A > 0 on ∂ A . U_A = G1_A,\qquad q_A > 0,\qquad\{U_A = q_A\} = \partial A,\qquad g_A = -\partial_\nu U_A > 0\ \text{on }\partial A. U A = G 1 A , q A > 0 , { U A = q A } = ∂ A , g A = − ∂ ν U A > 0 on ∂ A .
Lemma 6.1. For every compact interval I ⊂ ( 0 , ∣ D ∣ ) I \subset(0, |D|) I ⊂ ( 0 , ∣ D ∣ ) , the set
M I = { ( m , 1 A ) : m ∈ I , A ∈ M m } \mathcal{M}_I = \{(m,1_A): m \in I,\ A \in\mathcal{M}_m\} M I = {( m , 1 A ) : m ∈ I , A ∈ M m }
is compact in I × L 1 ( D ) I \times L^1(D) I × L 1 ( D ) , and ( m , A ) ↦ q A (m,A) \mapsto q_A ( m , A ) ↦ q A is continuous. If ( m j , 1 A j ) → ( m , 1 A ) (m_j,\mathbf{1}_{A_j}) \to(m,\mathbf{1}_A) ( m j , 1 A j ) → ( m , 1 A ) , then ∂ A j \partial A_j ∂ A j is eventually a normal graph over ∂ A \partial A ∂ A , with graph functions converging to zero in C 1 , α ( ∂ A ) C^{1,\alpha}(\partial A) C 1 , α ( ∂ A ) for every 0 < α < 1 0 < \alpha< 1 0 < α < 1 .
Proof. Let τ = G 1 \tau= \mathcal{G}1 τ = G 1 . Green symmetry and 0 ≤ G ( ρ + σ ) ≤ 2 τ 0 \le\mathcal{G}(\rho+\sigma) \le2\tau 0 ≤ G ( ρ + σ ) ≤ 2 τ give
∣ E ( ρ ) − E ( σ ) ∣ = 1 2 ∣ ∫ D ( ρ − σ ) G ( ρ + σ ) d x ∣ ≤ ∥ τ ∥ ∞ ∥ ρ − σ ∥ 1 ( 0 ≤ ρ , σ ≤ 1 ) . |\mathcal{E}(\rho)-\mathcal{E}(\sigma)| = \frac{1}{2}\left|\int_D (\rho-\sigma)\mathcal{G}(\rho+\sigma)\,\mathrm{d}x\right|
\le\|\tau\|_\infty\|\rho-\sigma\|_1 \qquad(0 \le\rho,\sigma\le1). ∣ E ( ρ ) − E ( σ ) ∣ = 2 1 ∫ D ( ρ − σ ) G ( ρ + σ ) d x ≤ ∥ τ ∥ ∞ ∥ ρ − σ ∥ 1 ( 0 ≤ ρ , σ ≤ 1 ) .
For ρ ∈ A m \rho\in\mathcal{A}_m ρ ∈ A m , choose
σ = { ( m ′ / m ) ρ , m ′ < m , ρ + m ′ − m ∣ D ∣ − m ( 1 − ρ ) , m ′ > m . \sigma=
\begin{cases}
(m'/m)\rho, & m' < m,\\
\rho+ \dfrac{m'-m}{|D|-m}(1-\rho), & m' > m.
\end{cases} σ = ⎩ ⎨ ⎧ ( m ′ / m ) ρ , ρ + ∣ D ∣ − m m ′ − m ( 1 − ρ ) , m ′ < m , m ′ > m .
σ ∈ A m ′ \sigma\in\mathcal{A}_{m'} σ ∈ A m ′ , ∥ σ − ρ ∥ 1 = ∣ m ′ − m ∣ \|\sigma-\rho\|_1 = |m'-m| ∥ σ − ρ ∥ 1 = ∣ m ′ − m ∣ .
Reversing m , m ′ m,m' m , m ′ yields
∣ V ( m ′ ) − V ( m ) ∣ ≤ ∥ τ ∥ ∞ ∣ m ′ − m ∣ . (57) |V(m')-V(m)| \le\|\tau\|_\infty|m'-m|. \tag*{(57)} ∣ V ( m ′ ) − V ( m ) ∣ ≤ ∥ τ ∥ ∞ ∣ m ′ − m ∣. (57)
If m j → m m_j \to m m j → m and 1 A j ⇀ ∗ ρ \mathbf{1}_{A_j} \mathrel{\mathop{\rightharpoonup}\limits^*} \rho 1 A j ⇀ ∗ ρ , weak-* energy continuity gives ρ ∈ M m \rho\in\mathcal{M}_m ρ ∈ M m . Thus ρ = 1 A \rho= \mathbf{1}_A ρ = 1 A and
∥ 1 A j − 1 A ∥ 1 = m j + m − 2 ∫ A 1 A j d x ⟶ 0. \|\mathbf{1}_{A_j}-\mathbf{1}_A\|_1 = m_j+m-2\int_A \mathbf{1}_{A_j}\,\mathrm{d}x \longrightarrow0. ∥ 1 A j − 1 A ∥ 1 = m j + m − 2 ∫ A 1 A j d x ⟶ 0.
This proves compactness. For every finite p p p ,
∥ 1 A j − 1 A ∥ p = ∥ 1 A j − 1 A ∥ 1 1 / p ⟶ 0 , ∥ U A j − U A ∥ W 2 , p ≤ C p ∥ 1 A j − 1 A ∥ p . \|\mathbf{1}_{A_j}-\mathbf{1}_A\|_p = \|\mathbf{1}_{A_j}-\mathbf{1}_A\|_1^{1/p} \longrightarrow0,\qquad
\|U_{A_j}-U_A\|_{W^{2,p}} \le C_p\|\mathbf{1}_{A_j}-\mathbf{1}_A\|_p. ∥ 1 A j − 1 A ∥ p = ∥ 1 A j − 1 A ∥ 1 1/ p ⟶ 0 , ∥ U A j − U A ∥ W 2 , p ≤ C p ∥ 1 A j − 1 A ∥ p .
Consequently
U A j ⟶ U A in C 1 , β ( D ‾ ) ( 0 < β < 1 ) . (58) U_{A_j} \longrightarrow U_A \quad\text{in } C^{1,\beta}(\overline{D}) \qquad(0 < \beta< 1). \tag*{(58)} U A j ⟶ U A in C 1 , β ( D ) ( 0 < β < 1 ) . (58)
Since 0 < q A j ≤ ∥ U A j ∥ ∞ ≤ ∥ τ ∥ ∞ 0 < q_{A_j} \le\|U_{A_j}\|_\infty\le\|\tau\|_\infty 0 < q A j ≤ ∥ U A j ∥ ∞ ≤ ∥ τ ∥ ∞ , every subsequence has a threshold cluster value q ∗ q_* q ∗ . Uniform convergence gives
1 A = 1 a.e. on { U A > q ∗ } , 1 A = 0 a.e. on { U A < q ∗ } . \mathbf{1}_A = 1 \text{ a.e. on }\{U_A>q_*\},\qquad
\mathbf{1}_A = 0 \text{ a.e. on }\{U_A<q_*\}. 1 A = 1 a.e. on { U A > q ∗ } , 1 A = 0 a.e. on { U A < q ∗ } .
If q ∗ ≠ q A q_* \ne q_A q ∗ = q A , a sufficiently thin strip along the noncritical level ∂ A \partial A ∂ A contradicts one of these identities. Hence q A j → q A q_{A_j} \to q_A q A j → q A .
In disjoint normal tubes around ∂ A \partial A ∂ A , choose the normal coordinate r r r so that
∂ r U A ≤ − 1 2 min ∂ A g A < 0. \partial_r U_A \le-\frac{1}{2}\min_{\partial A} g_A < 0. ∂ r U A ≤ − 2 1 ∂ A min g A < 0.
Equation (47) bounds ∣ U A − q A ∣ |U_A-q_A| ∣ U A − q A ∣ away from zero on the complement of the tubes. Convergence therefore gives exactly one root r = h j ( x ) r=h_j(x) r = h j ( x ) on every normal fiber and no boundary points elsewhere. The implicit-function formula and (58) yield
∥ h j ∥ C 1 ⟶ 0 , sup j ∥ h j ∥ C 1 , β < ∞ ⟹ ∥ h j ∥ C 1 , α ⟶ 0 ( α < β < 1 ) . \|h_j\|_{C^1} \longrightarrow0,\qquad
\sup_j\|h_j\|_{C^{1,\beta}} < \infty
\quad\Longrightarrow\quad
\|h_j\|_{C^{1,\alpha}} \longrightarrow0
\qquad(\alpha< \beta< 1). ∥ h j ∥ C 1 ⟶ 0 , j sup ∥ h j ∥ C 1 , β < ∞ ⟹ ∥ h j ∥ C 1 , α ⟶ 0 ( α < β < 1 ) .
Fix 0 < α < 1 0 < \alpha< 1 0 < α < 1 and A 0 ∈ M m 0 A_0 \in\mathcal{M}_{m_0} A 0 ∈ M m 0 , and set
Γ 0 = ∂ A 0 , q 0 = q A 0 , g 0 = − ∂ ν U A 0 , B = C 1 , α ( Γ 0 ) . \Gamma_0 = \partial A_0,\qquad q_0 = q_{A_0},\qquad g_0 = -\partial_\nu U_{A_0},\qquad B = C^{1,\alpha}(\Gamma_0). Γ 0 = ∂ A 0 , q 0 = q A 0 , g 0 = − ∂ ν U A 0 , B = C 1 , α ( Γ 0 ) .
In disjoint normal tubes, use x = γ ( ξ ) + r ν ( ξ ) x = \gamma(\xi) + r\nu(\xi) x = γ ( ξ ) + r ν ( ξ ) with Jacobian 1 + κ ( ξ ) r 1 + \kappa(\xi)r 1 + κ ( ξ ) r . For ∥ h ∥ B \lVert h\rVert_B ∥ h ∥ B small, put
X h ( x ) = x + h ( x ) ν ( x ) , ∂ A h = X h ( Γ 0 ) . X_h(x) = x + h(x)\nu(x),\qquad\partial A_h = X_h(\Gamma_0). X h ( x ) = x + h ( x ) ν ( x ) , ∂ A h = X h ( Γ 0 ) .
Here A h A_h A h is the image of A 0 A_0 A 0 under the C 1 , α C^{1,\alpha} C 1 , α diffeomorphism
( ξ , r ) ⟼ ( ξ , r + χ ( r ) h ( ξ ) ) , (\xi,r) \longmapsto(\xi,r+\chi(r)h(\xi)), ( ξ , r ) ⟼ ( ξ , r + χ ( r ) h ( ξ )) ,
where χ = 1 \chi=1 χ = 1 near 0 0 0 and is supported in the tubes; the map is the identity elsewhere. Thus A h ⊂ D A_h\subset D A h ⊂ D has the topology of A 0 A_0 A 0 . For two small graphs,
∣ A h △ A k ∣ = ∫ Γ 0 ∣ ∫ k ( x ) h ( x ) ( 1 + κ ( x ) r ) d r ∣ d s ≍ ∥ h − k ∥ L 1 ( Γ 0 ) . |A_h\mathbin{\triangle} A_k|=\int_{\Gamma_0}\left|\int_{k(x)}^{h(x)}(1+\kappa(x)r)\,dr\right|\,ds\asymp\lVert h-k\rVert_{L^1(\Gamma_0)}. ∣ A h △ A k ∣ = ∫ Γ 0 ∫ k ( x ) h ( x ) ( 1 + κ ( x ) r ) d r d s ≍ ∥ h − k ∥ L 1 ( Γ 0 ) .
In particular, the chart is injective modulo null sets. Define
F ( h , t , s ) = ( ( G 1 A h ) ∘ X h − t , ∣ A h ∣ − s ) ∈ Y : = B × R (59) \mathcal{F}(h,t,s)=\left((\mathcal{G}1_{A_h})\circ X_h-t,\ |A_h|-s\right)\in Y:=B\times\mathbb{R} \tag*{(59)} F ( h , t , s ) = ( ( G 1 A h ) ∘ X h − t , ∣ A h ∣ − s ) ∈ Y := B × R (59)
near ( 0 , q 0 , m 0 ) (0,q_0,m_0) ( 0 , q 0 , m 0 ) .
Lemma 6.2. The map F \mathcal{F} F is real analytic. Its shape derivative
L ϕ : = D h F ( 0 , q 0 , m 0 ) ϕ = ( S Γ 0 ϕ − g 0 ϕ , ∫ Γ 0 ϕ d s ) (60) L\phi:=D_h\mathcal{F}(0,q_0,m_0)\phi=\left(S_{\Gamma_0}\phi-g_0\phi,\ \int_{\Gamma_0}\phi\,ds\right) \tag*{(60)} L ϕ := D h F ( 0 , q 0 , m 0 ) ϕ = ( S Γ 0 ϕ − g 0 ϕ , ∫ Γ 0 ϕ d s ) (60) is injective and Fredholm of index − 1 -1 − 1 .
Proof. Set p ( y ) = − ∣ y ∣ 2 / 4 p(y)=-|y|^2/4 p ( y ) = − ∣ y ∣ 2 /4 . Since − Δ p = 1 -\Delta p=1 − Δ p = 1 , Green’s formula gives
G 1 A h ( x ) = p ( x ) + ∫ ∂ A h [ p ( y ) ∂ ν y G D ( x , y ) − G D ( x , y ) ∂ ν h p ( y ) ] d s y ( x ∈ A h ) . \mathcal{G}1_{A_h}(x)=p(x)+\int_{\partial A_h}\left[p(y)\partial_{\nu_y}G_D(x,y)-G_D(x,y)\partial_{\nu_h}p(y)\right]\,ds_y\qquad(x\in A_h). G 1 A h ( x ) = p ( x ) + ∫ ∂ A h [ p ( y ) ∂ ν y G D ( x , y ) − G D ( x , y ) ∂ ν h p ( y ) ] d s y ( x ∈ A h ) .
Let S D S^D S D have kernel G D G_D G D and K D K^D K D the principal-value kernel ∂ ν y G D \partial_{\nu_y}G_D ∂ ν y G D . The interior double-layer trace is − 1 2 I + K D -\frac12 I+K^D − 2 1 I + K D , so
( G 1 A h ) ∣ ∂ A h = 1 2 p + K ∂ A h D p − S ∂ A h D ∂ ν h p (61) (\mathcal{G}1_{A_h})|_{\partial A_h}=\frac12p+K^D_{\partial A_h}p-S^D_{\partial A_h}\partial_{\nu_h}p \tag*{(61)} ( G 1 A h ) ∣ ∂ A h = 2 1 p + K ∂ A h D p − S ∂ A h D ∂ ν h p (61)
For the fundamental kernel Φ ( x ) = − ( 2 π ) − 1 log ∣ x ∣ \Phi(x)=-(2\pi)^{-1}\log|x| Φ ( x ) = − ( 2 π ) − 1 log ∣ x ∣ , [31 Theorem 3.2] gives real-analytic dependence on C 1 , α C^{1,\alpha} C 1 , α embeddings in the spaces
S : C 0 , α ⟶ C 1 , α , K : C 1 , α ⟶ C 1 , α . S:C^{0,\alpha}\longrightarrow C^{1,\alpha},\qquad K:C^{1,\alpha}\longrightarrow C^{1,\alpha}. S : C 0 , α ⟶ C 1 , α , K : C 1 , α ⟶ C 1 , α .
Apply this theorem to each Jordan interior of Γ 0 \Gamma_0 Γ 0 , whose exterior is connected. Write J h d s = X h ∗ ( d s ∣ ∂ A h ) J_h\,ds=X_h^*(ds|_{\partial A_h}) J h d s = X h ∗ ( d s ∣ ∂ A h ) and ν ^ h = ν h ∘ X h \widehat{\nu}_h=\nu_h\circ X_h ν h = ν h ∘ X h . For a unit tangent T 0 T_0 T 0 to Γ 0 \Gamma_0 Γ 0 , the normal-coordinate convention above gives
J h = ( 1 + κ h ) 2 + ∣ ∂ s h ∣ 2 , J h ν ^ h = ( 1 + κ h ) ν − ( ∂ s h ) T 0 . J_h=\sqrt{(1+\kappa h)^2+|\partial_s h|^2},\qquad J_h\widehat{\nu}_h=(1+\kappa h)\nu-(\partial_s h)T_0. J h = ( 1 + κh ) 2 + ∣ ∂ s h ∣ 2 , J h ν h = ( 1 + κh ) ν − ( ∂ s h ) T 0 .
For ∥ h ∥ B \lVert h\rVert_B ∥ h ∥ B small, J h J_h J h is bounded away from zero. Products and the positive square root in the Banach algebra C 0 , α ( Γ 0 ) C^{0,\alpha}(\Gamma_0) C 0 , α ( Γ 0 ) therefore show that the pulled-back data satisfy
h ⟼ p ∘ X h ∈ C 1 , α , h ⟼ ν ^ h , J h , ( ∂ ν h p ) ∘ X h ∈ C 0 , α h\longmapsto p\circ X_h\in C^{1,\alpha},\qquad h\longmapsto\widehat{\nu}_h,\ J_h,\ (\partial_{\nu_h}p)\circ X_h\in C^{0,\alpha} h ⟼ p ∘ X h ∈ C 1 , α , h ⟼ ν h , J h , ( ∂ ν h p ) ∘ X h ∈ C 0 , α
real analytically. The regular Green part H D = Φ − G D H_D=\Phi-G_D H D = Φ − G D is smooth across the diagonal and harmonic in each variable. Hence
( Δ x + Δ y ) H D = 0 , H D ∈ C ω ( D × D ) . (\Delta_x+\Delta_y)H_D=0,\qquad H_D\in C^\omega(D\times D). ( Δ x + Δ y ) H D = 0 , H D ∈ C ω ( D × D ) .
Choose a compact set K ⊂ D K\subset D K ⊂ D whose interior contains all small deformed closures. Interior analytic estimates give constants M , R > 0 M,R>0 M , R > 0 such that, for all multiindices β , γ ∈ N 0 2 \beta,\gamma\in\mathbb{N}_0^2 β , γ ∈ N 0 2 ,
sup K × K ∣ ∂ x β ∂ y γ H D ∣ ≤ M R − ∣ β ∣ − ∣ γ ∣ ( ∣ β ∣ + ∣ γ ∣ ) ! . (62) \sup_{K\times K}|\partial_x^\beta\partial_y^\gamma H_D|\le MR^{-|\beta|-|\gamma|}(|\beta|+|\gamma|)!. \tag*{(62)} K × K sup ∣ ∂ x β ∂ y γ H D ∣ ≤ M R − ∣ β ∣ − ∣ γ ∣ ( ∣ β ∣ + ∣ γ ∣ )! . (62)
For distinct components, the same estimate holds for Φ ( x − y ) \Phi(x-y) Φ ( x − y ) on the product of their disjoint fixed normal tubes. Thus these bounds require no regularity of the outer boundary. To obtain analyticity in operator norm, write Γ 0 = ⋃ i = 1 N Γ i \Gamma_0=\bigcup_{i=1}^N\Gamma_i Γ 0 = ⋃ i = 1 N Γ i and let P ( x , y ) P(x,y) P ( x , y ) denote either H D ( x , y ) H_D(x,y) H D ( x , y ) , or Φ ( x − y ) \Phi(x-y) Φ ( x − y ) when i ≠ j i\ne j i = j . Define
T P i j ( h ) η ( x ) = ∫ Γ j P ( X h ( x ) , X h ( y ) ) η ( y ) J h ( y ) d s y , T_P^{ij}(h)\eta(x)=\int_{\Gamma_j}P(X_h(x),X_h(y))\eta(y)J_h(y)\,{\rm d}s_y, T P ij ( h ) η ( x ) = ∫ Γ j P ( X h ( x ) , X h ( y )) η ( y ) J h ( y ) d s y ,
K P i j ( h ) ξ ( x ) = ∫ Γ j ∇ y P ( X h ( x ) , X h ( y ) ) ⋅ ν ^ h ( y ) ξ ( y ) J h ( y ) d s y , x ∈ Γ i . K_P^{ij}(h)\xi(x)=\int_{\Gamma_j}\nabla_yP(X_h(x),X_h(y))\cdot\hat{\nu}_h(y)\xi(y)J_h(y)\,{\rm d}s_y,\qquad x\in\Gamma_i. K P ij ( h ) ξ ( x ) = ∫ Γ j ∇ y P ( X h ( x ) , X h ( y )) ⋅ ν ^ h ( y ) ξ ( y ) J h ( y ) d s y , x ∈ Γ i .
Set V f ( x , y ) = ( f ( x ) ν ( x ) , f ( y ) ν ( y ) ) V_f(x,y)=(f(x)\nu(x),f(y)\nu(y)) V f ( x , y ) = ( f ( x ) ν ( x ) , f ( y ) ν ( y )) . Since X h X_h X h is affine in h h h ,
D h k [ P ( X h ( x ) , X h ( y ) ) ] [ f 1 , … , f k ] = D k P ( X h ( x ) , X h ( y ) ) [ V f 1 ( x , y ) , … , V f k ( x , y ) ] . D_h^k[P(X_h(x),X_h(y))][f_1,\ldots,f_k]
=
D^kP(X_h(x),X_h(y))[V_{f_1}(x,y),\ldots,V_{f_k}(x,y)]. D h k [ P ( X h ( x ) , X h ( y ))] [ f 1 , … , f k ] = D k P ( X h ( x ) , X h ( y )) [ V f 1 ( x , y ) , … , V f k ( x , y )] .
The bounds (62) , the Hölder product estimates, and the formulas for J h J_h J h and J h ν ^ h J_h\hat{\nu}_h J h ν ^ h consequently imply
∥ D h k T P i j ( h ) [ f 1 , … , f k ] ∥ L ( C 0 , α ( Γ j ) , C 1 , α ( Γ i ) ) + ∥ D h k K P i j ( h ) [ f 1 , … , f k ] ∥ L ( C 1 , α ( Γ j ) , C 1 , α ( Γ i ) ) ≤ C 0 C 1 k k ! ∏ ℓ = 1 k ∥ f ℓ ∥ B . (63) \begin{aligned}
&\left\|D_h^kT_P^{ij}(h)[f_1,\ldots,f_k]\right\|_{\mathcal{L}(C^{0,\alpha}(\Gamma_j),C^{1,\alpha}(\Gamma_i))}\\
&\quad+\left\|D_h^kK_P^{ij}(h)[f_1,\ldots,f_k]\right\|_{\mathcal{L}(C^{1,\alpha}(\Gamma_j),C^{1,\alpha}(\Gamma_i))}
\le C_0C_1^k k!\prod_{\ell=1}^k\|f_\ell\|_B. \tag*{(63)}
\end{aligned} D h k T P ij ( h ) [ f 1 , … , f k ] L ( C 0 , α ( Γ j ) , C 1 , α ( Γ i )) + D h k K P ij ( h ) [ f 1 , … , f k ] L ( C 1 , α ( Γ j ) , C 1 , α ( Γ i )) ≤ C 0 C 1 k k ! ℓ = 1 ∏ k ∥ f ℓ ∥ B . (63)
The constants are uniform on a smaller ball about 0 in B B B . Indeed, differentiating once in the target variable and estimating its Hölder seminorm require at most three additional derivatives of P P P ; the resulting factor ( k + 3 ) ! (k+3)! ( k + 3 )! is absorbed into C 0 C 1 k k ! C_0C_1^k k! C 0 C 1 k k ! . For either operator T = T P i j T=T_P^{ij} T = T P ij or K P i j K_P^{ij} K P ij , Taylor’s formula in its indicated operator norm now gives
∥ T ( h ) − ∑ k = 0 n 1 k ! D h k T ( 0 ) [ h , … , h ] ∥ ≤ C 0 ( C 1 ∥ h ∥ B ) n + 1 ⟶ 0 ( C 1 ∥ h ∥ B < 1 ) . \left\|T(h)-\sum_{k=0}^{n}\frac{1}{k!}D_h^kT(0)[h,\ldots,h]\right\|
\le C_0(C_1\|h\|_B)^{n+1}\longrightarrow0\qquad(C_1\|h\|_B<1). T ( h ) − k = 0 ∑ n k ! 1 D h k T ( 0 ) [ h , … , h ] ≤ C 0 ( C 1 ∥ h ∥ B ) n + 1 ⟶ 0 ( C 1 ∥ h ∥ B < 1 ) .
Together with the singular diagonal blocks supplied by [31 ] , this proves analyticity of the full pulled-back layer operators. Equation (61) , together with
∣ A h ∣ = m 0 + ∫ Γ 0 ( h + k h 2 2 ) d s , |A_h|=m_0+\int_{\Gamma_0}\left(h+\frac{kh^2}{2}\right)\,{\rm d}s, ∣ A h ∣ = m 0 + ∫ Γ 0 ( h + 2 k h 2 ) d s ,
proves analyticity of F \mathcal{F} F .
For h = ε ϕ h=\varepsilon\phi h = εϕ , write U ε = G 1 A ε ϕ U_\varepsilon=\mathcal{G}\mathbf{1}_{A_{\varepsilon\phi}} U ε = G 1 A εϕ . The signed-strip formula yields, at each x ∈ Γ 0 x\in\Gamma_0 x ∈ Γ 0 ,
U ε ( x ) − U 0 ( x ) ε ⟶ ∫ Γ 0 G D ( x , y ) ϕ ( y ) d s y . \frac{U_\varepsilon(x)-U_0(x)}{\varepsilon}\longrightarrow\int_{\Gamma_0}G_D(x,y)\phi(y)\,ds_y. ε U ε ( x ) − U 0 ( x ) ⟶ ∫ Γ 0 G D ( x , y ) ϕ ( y ) d s y .
Indeed, away from an arc of length 2 δ 2\delta 2 δ about x x x , ordinary transport applies; the arc contributes at most C δ ( 1 + ∣ log δ ∣ ) C\delta(1+\lvert\log\delta\rvert) C δ ( 1 + ∣ log δ ∣) to the difference quotient for small ∣ ε ∣ \lvert\varepsilon\rvert ∣ ε ∣ . First let ε → 0 \varepsilon\to0 ε → 0 , then δ → 0 \delta\to0 δ → 0 . Also U ε → U 0 U_\varepsilon\to U_0 U ε → U 0 in C 1 ( D ‾ ) C^1(\overline{D}) C 1 ( D ) by Dirichlet estimates. Therefore
U ε ( X ε ϕ ( x ) ) − U 0 ( x ) ε = U ε ( x ) − U 0 ( x ) ε + ∫ 0 1 ∇ U ε ( x + θ ε ϕ ( x ) ν ( x ) ) ⋅ ϕ ( x ) ν ( x ) d θ ⟶ S Γ 0 ϕ ( x ) − g 0 ( x ) ϕ ( x ) . \frac{U_\varepsilon(X_{\varepsilon\phi}(x))-U_0(x)}{\varepsilon}
=
\frac{U_\varepsilon(x)-U_0(x)}{\varepsilon}
+\int_0^1\nabla U_\varepsilon(x+\theta\varepsilon\phi(x)\nu(x))\cdot\phi(x)\nu(x)\,d\theta
\longrightarrow S_{\Gamma_0}\phi(x)-g_0(x)\phi(x). ε U ε ( X εϕ ( x )) − U 0 ( x ) = ε U ε ( x ) − U 0 ( x ) + ∫ 0 1 ∇ U ε ( x + θ εϕ ( x ) ν ( x )) ⋅ ϕ ( x ) ν ( x ) d θ ⟶ S Γ 0 ϕ ( x ) − g 0 ( x ) ϕ ( x ) .
Analyticity into B B B identifies this pointwise formula with the Banach-space derivative, proving (60) .
One-sided elliptic regularity gives g 0 ∈ C ∞ ( Γ 0 ) g_0\in C^\infty(\Gamma_0) g 0 ∈ C ∞ ( Γ 0 ) , and min Γ 0 g 0 > 0 \min_{\Gamma_0}g_0>0 min Γ 0 g 0 > 0 . The operator
L 0 ϕ = ( − g 0 ϕ , ∫ Γ 0 ϕ d s ) L_0\phi=\left(-g_0\phi,\int_{\Gamma_0}\phi\,ds\right) L 0 ϕ = ( − g 0 ϕ , ∫ Γ 0 ϕ d s )
is an isomorphism from B B B onto
{ ( f , b ) ∈ Y : b + ∫ Γ 0 f g 0 d s = 0 } . \left\{(f,b)\in Y:b+\int_{\Gamma_0}\frac{f}{g_0}\,ds=0\right\}. { ( f , b ) ∈ Y : b + ∫ Γ 0 g 0 f d s = 0 } .
Thus ind L 0 = − 1 \operatorname{ind}L_0=-1 ind L 0 = − 1 . Since
B ↪ C 0 , α ( Γ 0 ) compactly , S Γ 0 : C 0 , α ⟶ B boundedly , B\hookrightarrow C^{0,\alpha}(\Gamma_0)\quad\text{compactly},\qquad S_{\Gamma_0}:C^{0,\alpha}\longrightarrow B\quad\text{boundedly}, B ↪ C 0 , α ( Γ 0 ) compactly , S Γ 0 : C 0 , α ⟶ B boundedly ,
L − L 0 L-L_0 L − L 0 is compact and ind L = − 1 \operatorname{ind}L=-1 ind L = − 1 . Finally,
L ϕ = 0 ⟹ S Γ 0 ϕ = g 0 ϕ , ∫ Γ 0 ϕ d s = 0 ⟹ ϕ ∈ K A 0 0 = { 0 } L\phi=0\quad\Longrightarrow\quad S_{\Gamma_0}\phi=g_0\phi,\ \int_{\Gamma_0}\phi\,ds=0\quad\Longrightarrow\quad\phi\in K^0_{A_0}=\{0\} L ϕ = 0 ⟹ S Γ 0 ϕ = g 0 ϕ , ∫ Γ 0 ϕ d s = 0 ⟹ ϕ ∈ K A 0 0 = { 0 }
by proposition 5.6.
Proposition 6.3 (Scalar analytic reduction). There are open intervals J ∋ q 0 J\ni q_0 J ∋ q 0 , I 0 ∋ m 0 I_0\ni m_0 I 0 ∋ m 0 , a neighborhood U ∋ 0 U\ni0 U ∋ 0 in B B B , and real-analytic maps H : J × I 0 → B H:J\times I_0\to B H : J × I 0 → B , r : J × I 0 → R r:J\times I_0\to\mathbb{R} r : J × I 0 → R , with H ( q 0 , m 0 ) = 0 H(q_0,m_0)=0 H ( q 0 , m 0 ) = 0 , such that
F ( h , t , s ) = 0 ⟺ h = H ( t , s ) , r ( t , s ) = 0 ( ( h , t , s ) ∈ U × J × I 0 ) . (64) \mathcal{F}(h,t,s)=0\quad\Longleftrightarrow\quad h=H(t,s),\quad r(t,s)=0\qquad((h,t,s)\in U\times J\times I_0). \tag*{(64)} F ( h , t , s ) = 0 ⟺ h = H ( t , s ) , r ( t , s ) = 0 (( h , t , s ) ∈ U × J × I 0 ) . (64) At fixed mass and threshold, there is at most one maximizing patch in this chart.
Proof. The range Y 0 = L ( B ) Y_0=L(B) Y 0 = L ( B ) is closed of codimension one. Choose
ℓ ∈ Y ∗ , ker ℓ = Y 0 , e ∈ Y , ℓ ( e ) = 1 , P y = y − ℓ ( y ) e . \ell\in Y^*,\qquad\ker\ell=Y_0,\qquad e\in Y,\qquad\ell(e)=1,\qquad Py=y-\ell(y)e. ℓ ∈ Y ∗ , ker ℓ = Y 0 , e ∈ Y , ℓ ( e ) = 1 , P y = y − ℓ ( y ) e .
Then D h ( P F ) = L : B → Y 0 D_h(P\mathcal{F})=L:B\to Y_0 D h ( P F ) = L : B → Y 0 is an isomorphism. The analytic implicit-function theorem gives h = H ( t , s ) h=H(t,s) h = H ( t , s ) as the unique solution of P F = 0 P\mathcal{F}=0 P F = 0 in product neighborhoods. Shrink them so that H ( J × I 0 ) ⊂ U H(J\times I_0)\subset U H ( J × I 0 ) ⊂ U , J ⊂ ( 0 , ∞ ) J\subset(0,\infty) J ⊂ ( 0 , ∞ ) , and I 0 ⊂ ( 0 , ∣ D ∣ ) I_0\subset(0,\lvert D\rvert) I 0 ⊂ ( 0 , ∣ D ∣) . Setting
r ( t , s ) = ℓ F ( H ( t , s ) , t , s ) r(t,s)=\ell\mathcal{F}(H(t,s),t,s) r ( t , s ) = ℓ F ( H ( t , s ) , t , s )
gives
F ( H ( t , s ) , t , s ) = r ( t , s ) e , \mathcal{F}(H(t,s),t,s)=r(t,s)e, F ( H ( t , s ) , t , s ) = r ( t , s ) e ,
which proves (64) . By lemma 6.1, all maximizing pairs sufficiently close to ( m 0 , 1 A 0 ) (m_0,\mathbf{1}_{A_0}) ( m 0 , 1 A 0 ) in mass and L 1 L^1 L 1 lie in this chart, with thresholds in J J J . Injectivity of the graph chart gives the last assertion.
Lemma 6.4. If r ( t , m 0 ) r(t,m_0) r ( t , m 0 ) vanishes identically near q 0 q_0 q 0 , then, on a smaller open interval J 1 ∋ q 0 J_1 \ni q_0 J 1 ∋ q 0 , the patches A ( t ) : = A H ( t , m 0 ) A(t):=A_{H(t,m_0)} A ( t ) := A H ( t , m 0 ) are maximizers with thresholds t t t . They are distinct for distinct t t t and continuous in L 1 ( D ) L^1(D) L 1 ( D ) .
Proof. The full equation gives
∣ A ( t ) ∣ = m 0 , U A ( t ) = t on ∂ A ( t ) . |A(t)|=m_0,\qquad U_{A(t)}=t\quad\text{on }\partial A(t). ∣ A ( t ) ∣ = m 0 , U A ( t ) = t on ∂ A ( t ) . For their normal velocity V t V_t V t ,
d d t E ( 1 A ( t ) ) = ∫ ∂ A ( t ) U A ( t ) V t d s = t ∫ ∂ A ( t ) V t d s = t d d t ∣ A ( t ) ∣ = 0. \frac{d}{dt}\mathcal{E}(\mathbf{1}_{A(t)})=\int_{\partial A(t)}U_{A(t)}V_t\,\mathrm{d}s=t\int_{\partial A(t)}V_t\,\mathrm{d}s=t\frac{d}{dt}|A(t)|=0. d t d E ( 1 A ( t ) ) = ∫ ∂ A ( t ) U A ( t ) V t d s = t ∫ ∂ A ( t ) V t d s = t d t d ∣ A ( t ) ∣ = 0. To justify the first equality, put f = 1 A ( t + δ ) − 1 A ( t ) f=\mathbf{1}_{A(t+\delta)}-\mathbf{1}_{A(t)} f = 1 A ( t + δ ) − 1 A ( t ) . Since H H H is C 1 C^1 C 1 ,
∣ A ( t + δ ) △ A ( t ) ∣ = O ( ∣ δ ∣ ) , ∥ f ∥ H ˙ D − 1 2 ≤ C p ∥ f ∥ p 2 = O ( ∣ δ ∣ 2 / p ) = o ( ∣ δ ∣ ) ( 1 < p < 2 ) . |A(t+\delta)\mathbin{\triangle} A(t)|=O(|\delta|),\qquad\lVert f\rVert_{\dot H^{-1}_D}^{2}\le C_p\lVert f\rVert_p^2=O(|\delta|^{2/p})=o(|\delta|)\quad(1<p<2). ∣ A ( t + δ ) △ A ( t ) ∣ = O ( ∣ δ ∣ ) , ∥ f ∥ H ˙ D − 1 2 ≤ C p ∥ f ∥ p 2 = O ( ∣ δ ∣ 2/ p ) = o ( ∣ δ ∣ ) ( 1 < p < 2 ) . The exact energy expansion reduces the derivative to transport of the continuous potential U A ( t ) U_{A(t)} U A ( t ) . As A ( q 0 ) = A 0 A(q_0)=A_0 A ( q 0 ) = A 0 ,
E ( 1 A ( t ) ) = V ( m 0 ) , A ( t ) ∈ M m 0 . \mathcal{E}(\mathbf{1}_{A(t)})=V(m_0),\qquad A(t)\in\mathcal{M}_{m_0}. E ( 1 A ( t ) ) = V ( m 0 ) , A ( t ) ∈ M m 0 . Its unique threshold is t t t : if q A ( t ) ≠ t q_{A(t)}\ne t q A ( t ) = t , continuity of U A ( t ) U_{A(t)} U A ( t ) and positive-area neighborhoods on both sides of its C 1 , α C^{1,\alpha} C 1 , α boundary contradict 1 A ( t ) = 1 { U A ( t ) > q A ( t ) } \mathbf{1}_{A(t)}=\mathbf{1}_{\{U_{A(t)}>q_{A(t)}\}} 1 A ( t ) = 1 { U A ( t ) > q A ( t ) } a.e. Threshold uniqueness and the signed-strip estimate now give distinctness and L 1 L^1 L 1 continuity.
Proposition 6.5. For every m ∈ ( 0 , ∣ D ∣ ) m\in(0,|D|) m ∈ ( 0 , ∣ D ∣ ) , the set M m \mathcal{M}_m M m is finite.
Proof. Suppose the compact metric space ( M m , d ) (\mathcal{M}_m,d) ( M m , d ) , d ( A , B ) = ∥ 1 A − 1 B ∥ 1 d(A,B)=\lVert\mathbf{1}_A-\mathbf{1}_B\rVert_1 d ( A , B ) = ∥ 1 A − 1 B ∥ 1 , is infinite. Its derived set M m ′ \mathcal{M}'_m M m ′ is nonempty and compact. Choose
A 0 ∈ M m ′ , q 0 = q A 0 = max A ∈ M m ′ q A , A j → A 0 , A_0\in\mathcal{M}'_m,\qquad q_0=q_{A_0}=\max_{A\in\mathcal{M}'_m}q_A,\qquad A_j\to A_0, A 0 ∈ M m ′ , q 0 = q A 0 = A ∈ M m ′ max q A , A j → A 0 , with A j ∈ M m ∖ { A 0 } A_j\in\mathcal{M}_m\setminus\{A_0\} A j ∈ M m ∖ { A 0 } pairwise distinct. By the local reduction, for large j j j ,
A j = A H ( q A j , m ) , r ( q A j , m ) = 0 , q A j → q 0 . A_j=A_{H(q_{A_j},m)},\qquad r(q_{A_j},m)=0,\qquad q_{A_j}\to q_0. A j = A H ( q A j , m ) , r ( q A j , m ) = 0 , q A j → q 0 . The thresholds are pairwise distinct, since fixed mass and threshold determine at most one shape in this chart. The one-variable identity theorem gives r ( t , m ) ≡ 0 r(t,m)\equiv0 r ( t , m ) ≡ 0 near q 0 q_0 q 0 . By lemma 6.4, there is an open interval J 1 ∋ q 0 J_1\ni q_0 J 1 ∋ q 0 of distinct maximizers A H ( t , m ) A_{H(t,m)} A H ( t , m ) with thresholds t t t . Their L 1 L^1 L 1 continuity implies
A H ( t , m ) ∈ M m ′ ( t ∈ J 1 ) . A_{H(t,m)}\in\mathcal{M}'_m\qquad(t\in J_1). A H ( t , m ) ∈ M m ′ ( t ∈ J 1 ) . Taking t > q 0 t>q_0 t > q 0 contradicts the choice of A 0 A_0 A 0 .
Corollary 6.6. For every compact interval I ⊂ ( 0 , ∣ D ∣ ) I\subset(0,|D|) I ⊂ ( 0 , ∣ D ∣ ) there is N I < ∞ N_I<\infty N I < ∞ such that
# M m ≤ N I ( m ∈ I ) . \#\mathcal{M}_m\le N_I\qquad(m\in I). # M m ≤ N I ( m ∈ I ) .
Proof. At ( m 0 , A 0 ) ∈ M I (m_0,A_0) \in\mathcal{M}_I ( m 0 , A 0 ) ∈ M I , fixed-mass finiteness and lemma 6.4 imply r ( ⋅ , m 0 ) ≢ 0 r(\cdot,m_0) \not\equiv0 r ( ⋅ , m 0 ) ≡ 0 . Its zero at q 0 q_0 q 0 has finite order k ≥ 1 k \ge1 k ≥ 1 . Weierstrass preparation gives
r ( t , s ) = a ( t , s ) [ ( t − q 0 ) k + ∑ j = 0 k − 1 a j ( s ) ( t − q 0 ) j ] , a ≠ 0 , a j ( m 0 ) = 0. r(t,s)=a(t,s)\left[(t-q_0)^k+\sum_{j=0}^{k-1}a_j(s)(t-q_0)^j\right],\qquad a\ne0,\quad a_j(m_0)=0. r ( t , s ) = a ( t , s ) [ ( t − q 0 ) k + j = 0 ∑ k − 1 a j ( s ) ( t − q 0 ) j ] , a = 0 , a j ( m 0 ) = 0.
Each mass fiber has at most k k k roots, hence at most k k k maximizing patches in this chart. By lemma 6.1, the smaller chart on which this preparation holds contains a relatively open neighborhood of ( m 0 , A 0 ) (m_0,A_0) ( m 0 , A 0 ) in M I \mathcal{M}_I M I . A finite subcover, with orders k 1 , … , k L k_1,\ldots,k_L k 1 , … , k L , yields
# M m ≤ ∑ ℓ = 1 L k ℓ = : N I ( m ∈ I ) . \#\mathcal{M}_m\le\sum_{\ell=1}^{L}k_\ell=:N_I\qquad(m\in I). # M m ≤ ℓ = 1 ∑ L k ℓ =: N I ( m ∈ I ) .
Uniform counting
We first prove remark 1.5, allowing any bounded convex fluid domain. The boundary argument is transferred to smooth convex subdomains, while the shape analysis remains confined to the interior. We then impose a smooth boundary to control the two mass endpoints and prove theorem 1.4.
Lemma 7.1. Let D ⊂ R 2 D\subset\mathbb{R}^2 D ⊂ R 2 be a bounded convex domain and A ∈ M m A\in\mathcal{M}_m A ∈ M m , where 0 < m < ∣ D ∣ 0<m<|D| 0 < m < ∣ D ∣ . With the notation of (5.1)–(5.3),
K A 0 = { 0 } , dim K A ≤ 1. K_A^0=\{0\},\qquad\dim K_A\le1. K A 0 = { 0 } , dim K A ≤ 1.
Proof. Suppose 0 ≠ ϕ ∈ K A 0 0\ne\phi\in K_A^0 0 = ϕ ∈ K A 0 and set
Γ = ∂ A , U = G 1 A , v = G ( ϕ d s ∣ Γ ) , g = ∣ ∇ U ∣ Γ . \Gamma=\partial A,\qquad U=\mathcal{G}1_A,\qquad v=\mathcal{G}(\phi\,\mathrm{d}s|_\Gamma),\qquad g=|\nabla U|_\Gamma. Γ = ∂ A , U = G 1 A , v = G ( ϕ d s ∣ Γ ) , g = ∣∇ U ∣ Γ .
By theorem 1.1, Γ \Gamma Γ is a finite union of analytic curves contained in D D D , and g > 0 g>0 g > 0 . The proof of lemma 5.1 uses only layer estimates on these interior curves, so ϕ ∈ C ∞ ( Γ ) \phi\in C^\infty(\Gamma) ϕ ∈ C ∞ ( Γ ) . The interior trace theorem and Poincaré’s inequality give
∣ ∫ Γ ϕ z d s ∣ ≤ C Γ ∥ ϕ ∥ L 2 ( Γ ) ∥ ∇ z ∥ L 2 ( D ) ( z ∈ H 0 1 ( D ) ) , \left|\int_\Gamma\phi z\,\mathrm{d}s\right|\le C_\Gamma\|\phi\|_{L^2(\Gamma)}\|\nabla z\|_{L^2(D)}\qquad(z\in H_0^1(D)), ∫ Γ ϕ z d s ≤ C Γ ∥ ϕ ∥ L 2 ( Γ ) ∥∇ z ∥ L 2 ( D ) ( z ∈ H 0 1 ( D )) ,
so the curve source belongs to H − 1 ( D ) H^{-1}(D) H − 1 ( D ) and defines a unique weak Dirichlet potential. To verify its continuity at the outer boundary, consider the Newtonian single-layer potential
N ϕ ( x ) = ∫ Γ Φ ( x − y ) ϕ ( y ) d s y . N_\phi(x)=\int_\Gamma\Phi(x-y)\phi(y)\,\mathrm{d}s_y. N ϕ ( x ) = ∫ Γ Φ ( x − y ) ϕ ( y ) d s y .
Single-layer regularity on the smooth interior curves gives a continuous potential with C 1 , α C^{1,\alpha} C 1 , α extensions from either side of each curve [31 Section 2] . Since N ϕ N_\phi N ϕ is smooth away from Γ \Gamma Γ , its matching traces and bounded one-sided gradients imply
N ϕ ∈ H 1 ( D ) ∩ C ( D ‾ ) , − Δ N ϕ = ϕ d s ∣ Γ in D ′ ( D ) . N_\phi\in H^1(D)\cap C(\overline{D}),\qquad-\Delta N_\phi=\phi\,\mathrm{d}s|_\Gamma\quad\text{in }\mathcal{D}'(D). N ϕ ∈ H 1 ( D ) ∩ C ( D ) , − Δ N ϕ = ϕ d s ∣ Γ in D ′ ( D ) .
Identify R 2 \mathbb{R}^{2} R 2 with C \mathbb{C} C , write T = ∂ D \mathbb{T}=\partial D T = ∂ D , and let F : D → D F:D\to D F : D → D be conformal. Its continuous extension is a homeomorphism of the closures. Let P \mathcal{P} P denote the Poisson extension on D D D , and set
W ϕ = N ϕ ∘ F , h ϕ = P [ W ϕ ∣ T ] . W_{\phi}=N_{\phi}\circ F,\qquad h_{\phi}=\mathcal{P}[W_{\phi}|_{\mathbb{T}}]. W ϕ = N ϕ ∘ F , h ϕ = P [ W ϕ ∣ T ] .
Conformal invariance of Dirichlet energy and the Dirichlet principle on the disk yield
W ϕ , h ϕ ∈ H 1 ( D ) ∩ C ( D ‾ ) , W ϕ − h ϕ ∈ H 0 1 ( D ) , W_{\phi},h_{\phi}\in H^{1}(D)\cap C(\overline{D}),\qquad W_{\phi}-h_{\phi}\in H_{0}^{1}(D), W ϕ , h ϕ ∈ H 1 ( D ) ∩ C ( D ) , W ϕ − h ϕ ∈ H 0 1 ( D ) ,
∫ D ∣ ∇ h ϕ ∣ 2 d x ≤ ∫ D ∣ ∇ W ϕ ∣ 2 d x = ∫ D ∣ ∇ N ϕ ∣ 2 d x . \int_{D}|\nabla h_{\phi}|^{2}\,dx\leq\int_{D}|\nabla W_{\phi}|^{2}\,dx=\int_{D}|\nabla N_{\phi}|^{2}\,dx. ∫ D ∣∇ h ϕ ∣ 2 d x ≤ ∫ D ∣∇ W ϕ ∣ 2 d x = ∫ D ∣∇ N ϕ ∣ 2 d x .
Choose f k ∈ C c ∞ ( D ) f_k\in C_{c}^{\infty}(D) f k ∈ C c ∞ ( D ) converging to W ϕ − h ϕ W_{\phi}-h_{\phi} W ϕ − h ϕ in H 1 ( D ) H^{1}(D) H 1 ( D ) . Then
f k ∘ F − 1 ∈ C c ∞ ( D ) , f_k\circ F^{-1}\in C_{c}^{\infty}(D), f k ∘ F − 1 ∈ C c ∞ ( D ) ,
∥ ∇ ( f k ∘ F − 1 − f ℓ ∘ F − 1 ) ∥ L 2 ( D ) = ∥ ∇ ( f k − f ℓ ) ∥ L 2 ( D ) ⟶ 0. \|\nabla(f_k\circ F^{-1}-f_{\ell}\circ F^{-1})\|_{L^{2}(D)}
=\|\nabla(f_k-f_{\ell})\|_{L^{2}(D)}\longrightarrow0. ∥∇ ( f k ∘ F − 1 − f ℓ ∘ F − 1 ) ∥ L 2 ( D ) = ∥∇ ( f k − f ℓ ) ∥ L 2 ( D ) ⟶ 0.
Poincaré’s inequality makes these pullbacks Cauchy in H 0 1 ( D ) H_{0}^{1}(D) H 0 1 ( D ) ; local convergence identifies their limit as
v ~ = N ϕ − h ϕ ∘ F − 1 ∈ H 0 1 ( D ) ∩ C ( D ‾ ) , v ~ ∣ ∂ D = 0. \widetilde{v}=N_{\phi}-h_{\phi}\circ F^{-1}\in H_{0}^{1}(D)\cap C(\overline{D}),\qquad\widetilde{v}|_{\partial D}=0. v = N ϕ − h ϕ ∘ F − 1 ∈ H 0 1 ( D ) ∩ C ( D ) , v ∣ ∂ D = 0.
The harmonicity of h ϕ ∘ F − 1 h_{\phi}\circ F^{-1} h ϕ ∘ F − 1 gives − Δ v ~ = ϕ d s ∣ Γ -\Delta\widetilde{v}=\phi\,ds|_{\Gamma} − Δ v = ϕ d s ∣ Γ . Weak uniqueness therefore identifies v ~ = v \widetilde{v}=v v = v , and the defining equation for K A 0 K_{A}^{0} K A 0 gives
v ∈ H 0 1 ( D ) ∩ C ( D ‾ ) , v ∣ ∂ D = 0 , v = g ϕ on Γ . v\in H_{0}^{1}(D)\cap C(\overline{D}),\qquad v|_{\partial D}=0,\qquad v=g\phi\text{ on }\Gamma. v ∈ H 0 1 ( D ) ∩ C ( D ) , v ∣ ∂ D = 0 , v = g ϕ on Γ.
The truncation and continuation proof of lemma 5.2 therefore applies without a smooth outer boundary: v v v has exactly two nodal domains. Now set
W = U ∘ F , w = v ∘ F . W=U\circ F,\qquad w=v\circ F. W = U ∘ F , w = v ∘ F .
Both functions are harmonic in an annulus r 0 < ∣ z ∣ < 1 r_{0}<|z|<1 r 0 < ∣ z ∣ < 1 and vanish on T \mathbb{T} T . Reflection across the circle makes their radial derivatives real analytic there. The Hopf lemma on the disk collar and the zero total source give
P ∗ ( θ ) : = − ∂ r W ( e i θ ) > 0 , β ( θ ) : = ∂ r w ( e i θ ) , ∫ 0 2 π β ( θ ) d θ = − ∫ Γ ϕ d s = 0. (65) P_{*}(\theta):=-\partial_{r}W(e^{i\theta})>0,\qquad
\beta(\theta):=\partial_{r}w(e^{i\theta}),\qquad
\int_{0}^{2\pi}\beta(\theta)\,d\theta=-\int_{\Gamma}\phi\,ds=0. \tag*{(65)} P ∗ ( θ ) := − ∂ r W ( e i θ ) > 0 , β ( θ ) := ∂ r w ( e i θ ) , ∫ 0 2 π β ( θ ) d θ = − ∫ Γ ϕ d s = 0. (65)
The disk argument in the proof of proposition 5.3 now applies to w w w : β \beta β has exactly two simple zeros. This conclusion concerns the radial derivative on the circle and does not require a pointwise normal derivative on ∂ D \partial D ∂ D .
Convexity supplies the uniform estimate needed to pass the stress identity to the limiting boundary. Indeed, the conformal convexity criterion and its Herglotz representation give a probability measure μ \mu μ on T \mathbb{T} T such that
Re ( 1 + z F ′ ′ ( z ) F ′ ( z ) ) > 0 , log F ′ ( z ) F ′ ( 0 ) = − 2 ∫ T log ( 1 − ξ ‾ z ) d μ ( ξ ) . \operatorname{Re}\left(1+\frac{zF''(z)}{F'(z)}\right)>0,\qquad
\log\frac{F'(z)}{F'(0)}=-2\int_{\mathbb{T}}\log(1-\overline{\xi}z)\,d\mu(\xi). Re ( 1 + F ′ ( z ) z F ′′ ( z ) ) > 0 , log F ′ ( 0 ) F ′ ( z ) = − 2 ∫ T log ( 1 − ξ z ) d μ ( ξ ) .
Taking real parts yields
∣ F ′ ( z ) ∣ ≥ ∣ F ′ ( 0 ) ∣ ( 1 + ∣ z ∣ ) 2 , ∥ 1 F ′ ∥ L ∞ ( D ) ≤ 4 ∣ F ′ ( 0 ) ∣ . (66) |F'(z)|\geq\frac{|F'(0)|}{(1+|z|)^{2}},\qquad
\left\|\frac{1}{F'}\right\|_{L^{\infty}(D)}\leq\frac{4}{|F'(0)|}. \tag*{(66)} ∣ F ′ ( z ) ∣ ≥ ( 1 + ∣ z ∣ ) 2 ∣ F ′ ( 0 ) ∣ , F ′ 1 L ∞ ( D ) ≤ ∣ F ′ ( 0 ) ∣ 4 . (66)
The same convexity criterion shows that
D r : = F ( r D ) ( 0 < r < 1 ) D_r := F(r\mathbb{D}) \qquad(0 < r < 1) D r := F ( r D ) ( 0 < r < 1 )
is a smooth convex domain, and D r ↑ D D_r \uparrow D D r ↑ D . Since ∂ D \partial D ∂ D is rectifiable, F ′ F' F ′ belongs to the Hardy class with exponent one [32 ] . In particular, its radial boundary values satisfy
sup 0 < r < 1 ∫ 0 2 π ∣ F ′ ( r e i θ ) ∣ d θ < ∞ , \sup_{0<r<1}\int_{0}^{2\pi} \left|F'(re^{i\theta})\right|\,d\theta< \infty, 0 < r < 1 sup ∫ 0 2 π F ′ ( r e i θ ) d θ < ∞ ,
F ′ ( r e i ⋅ ) ⟶ F ′ ( e i ⋅ ) in L 1 ( 0 , 2 π ) and a.e. as r ↑ 1. F'(re^{i\cdot}) \longrightarrow F'(e^{i\cdot}) \quad\text{in } L^1(0,2\pi) \text{ and a.e. as } r \uparrow1. F ′ ( r e i ⋅ ) ⟶ F ′ ( e i ⋅ ) in L 1 ( 0 , 2 π ) and a.e. as r ↑ 1.
The boundary parametrization γ ( θ ) = F ( e i θ ) \gamma(\theta) = F(e^{i\theta}) γ ( θ ) = F ( e i θ ) is absolutely continuous, with γ ′ ( θ ) = i e i θ F ′ ( e i θ ) \gamma'(\theta) = ie^{i\theta}F'(e^{i\theta}) γ ′ ( θ ) = i e i θ F ′ ( e i θ ) a.e. Together with (66) , this gives, for almost every θ \theta θ and ζ = e i θ \zeta= e^{i\theta} ζ = e i θ ,
0 < ∣ F ′ ( ζ ) ∣ < ∞ , d s = ∣ F ′ ( ζ ) ∣ d θ , n ( F ( ζ ) ) = ζ F ′ ( ζ ) ∣ F ′ ( ζ ) ∣ . (67) 0 < |F'(\zeta)| < \infty,\qquad ds = |F'(\zeta)|\,d\theta,\qquad n(F(\zeta)) = \frac{\zeta F'(\zeta)}{|F'(\zeta)|}. \tag*{(67)} 0 < ∣ F ′ ( ζ ) ∣ < ∞ , d s = ∣ F ′ ( ζ ) ∣ d θ , n ( F ( ζ )) = ∣ F ′ ( ζ ) ∣ ζ F ′ ( ζ ) . (67)
Here F ′ ( ζ ) F'(\zeta) F ′ ( ζ ) denotes the radial limit and n n n the almost-everywhere outward normal of the convex domain. More explicitly, the arclength change of variables reads
∫ ∂ D f ( x ) d s x = ∫ 0 2 π f ( F ( e i θ ) ) ∣ F ′ ( e i θ ) ∣ d θ ( f ≥ 0 measurable ) . (68) \int_{\partial D} f(x)\,ds_x = \int_{0}^{2\pi} f(F(e^{i\theta}))|F'(e^{i\theta})|\,d\theta\qquad(f \ge0\ \text{measurable}). \tag*{(68)} ∫ ∂ D f ( x ) d s x = ∫ 0 2 π f ( F ( e i θ )) ∣ F ′ ( e i θ ) ∣ d θ ( f ≥ 0 measurable ) . (68)
Since the weight is positive and finite a.e., a property holds for arclength-a.e. point of ∂ D \partial D ∂ D if and only if its pullback holds for a.e. θ \theta θ .
For Z ( x ) = a + σ x Z(x) = a + \sigma x Z ( x ) = a + σ x , a ∈ R 2 a \in\mathbb{R}^2 a ∈ R 2 , σ ∈ R \sigma\in\mathbb{R} σ ∈ R , introduce the bilinear stress vector
B Z = ( Z ⋅ ∇ U ) ∇ v + ( Z ⋅ ∇ v ) ∇ U − ( ∇ U ⋅ ∇ v ) Z . B_Z = (Z \cdot\nabla U)\nabla v + (Z \cdot\nabla v)\nabla U - (\nabla U \cdot\nabla v)Z. B Z = ( Z ⋅ ∇ U ) ∇ v + ( Z ⋅ ∇ v ) ∇ U − ( ∇ U ⋅ ∇ v ) Z .
The distributional calculation in lemma 5.4 is confined to the interior and gives
div B Z = − 1 A Z ⋅ ∇ v + v ( Z ⋅ ν ) d s ∣ Γ . \operatorname{div} B_Z = -1_A Z \cdot\nabla v + v(Z \cdot\nu)\,ds|_\Gamma. div B Z = − 1 A Z ⋅ ∇ v + v ( Z ⋅ ν ) d s ∣ Γ .
For r r r sufficiently close to one, A ‾ ⊂ D r \overline{A} \subset D_r A ⊂ D r ; hence
∫ ∂ D r B Z ⋅ n r d s = 2 σ ∫ A v d x = 2 σ q A ∫ Γ ϕ d s = 0. (69) \int_{\partial D_r} B_Z \cdot n_r\,ds = 2\sigma\int_A v\,dx = 2\sigma q_A\int_\Gamma\phi\,ds = 0. \tag*{(69)} ∫ ∂ D r B Z ⋅ n r d s = 2 σ ∫ A v d x = 2 σ q A ∫ Γ ϕ d s = 0. (69)
Since U U U and v v v need not vanish on ∂ D r \partial D_r ∂ D r , we retain all terms in the stress vector. With
C Z ( r , θ ) : = a + σ F ( r e i θ ) e i θ F ′ ( r e i θ ) , C_Z(r,\theta) := \frac{a+\sigma F(re^{i\theta})}{e^{i\theta}F'(re^{i\theta})}, C Z ( r , θ ) := e i θ F ′ ( r e i θ ) a + σ F ( r e i θ ) ,
the normal and tangential components in (69) become
0 = ∫ 0 2 π Re C Z ( r , θ ) ( r W r w r − r − 1 W θ w θ ) d θ + ∫ 0 2 π Im C Z ( r , θ ) ( W θ w r + W r w θ ) d θ . 0 = \int_{0}^{2\pi} \operatorname{Re} C_Z(r,\theta)\left(rW_r w_r-r^{-1}W_\theta w_\theta\right)\,d\theta
+\int_{0}^{2\pi} \operatorname{Im} C_Z(r,\theta)\left(W_\theta w_r+W_r w_\theta\right)\,d\theta. 0 = ∫ 0 2 π Re C Z ( r , θ ) ( r W r w r − r − 1 W θ w θ ) d θ + ∫ 0 2 π Im C Z ( r , θ ) ( W θ w r + W r w θ ) d θ .
All derivatives here are evaluated at r e i θ re^{i\theta} r e i θ . Reflection gives, uniformly in θ \theta θ ,
W θ , w θ = O ( 1 − r ) , W r = − P ∗ + O ( 1 − r ) , w r = β + O ( 1 − r ) . W_\theta,w_\theta= O(1-r),\qquad W_r = -P_* + O(1-r),\qquad w_r = \beta+ O(1-r). W θ , w θ = O ( 1 − r ) , W r = − P ∗ + O ( 1 − r ) , w r = β + O ( 1 − r ) .
The distortion estimate (66) gives
sup 0 < r < 1 ∥ C Z ( r , ⋅ ) ∥ L ∞ ( 0 , 2 π ) ≤ 4 ∣ F ′ ( 0 ) ∣ ( ∣ a ∣ + ∣ σ ∣ sup x ∈ D ∣ x ∣ ) . \sup_{0<r<1}\|C_Z(r,\cdot)\|_{L^\infty(0,2\pi)}\leq\frac{4}{|F'(0)|}\left(|a|+|\sigma|\sup_{x\in D}|x|\right). 0 < r < 1 sup ∥ C Z ( r , ⋅ ) ∥ L ∞ ( 0 , 2 π ) ≤ ∣ F ′ ( 0 ) ∣ 4 ( ∣ a ∣ + ∣ σ ∣ x ∈ D sup ∣ x ∣ ) .
Dominated convergence therefore yields the boundary identity
∫ 0 2 π P ∗ ( θ ) β ( θ ) c Z ( θ ) d θ = 0 , c Z ( θ ) : = Re a + σ F ( e i θ ) e i θ F ′ ( e i θ ) . (70) \int_0^{2\pi}P_*(\theta)\beta(\theta)c_Z(\theta)\,d\theta=0,\qquad c_Z(\theta):=\operatorname{Re}\frac{a+\sigma F(e^{i\theta})}{e^{i\theta}F'(e^{i\theta})}. \tag*{(70)} ∫ 0 2 π P ∗ ( θ ) β ( θ ) c Z ( θ ) d θ = 0 , c Z ( θ ) := Re e i θ F ′ ( e i θ ) a + σ F ( e i θ ) . (70)
To obtain a field with the required sign, apply lemma 5.5 on D r D_r D r to the boundary function b r ( F ( r e i θ ) ) = β ( θ ) b_r(F(re^{i\theta}))=\beta(\theta) b r ( F ( r e i θ )) = β ( θ ) . After a positive rescaling, the resulting field Z r ( x ) = a r + σ r x Z_r(x)=a_r+\sigma_r x Z r ( x ) = a r + σ r x satisfies
∣ a r ∣ 2 + σ r 2 = 1 , β ( θ ) Re C Z r ( r , θ ) ≥ 0. |a_r|^2+\sigma_r^2=1,\qquad\beta(\theta)\operatorname{Re}C_{Z_r}(r,\theta)\geq0. ∣ a r ∣ 2 + σ r 2 = 1 , β ( θ ) Re C Z r ( r , θ ) ≥ 0.
Choose r j ↑ 1 r_j\uparrow1 r j ↑ 1 so that ( a r j , σ r j ) → ( a , σ ) (a_{r_j},\sigma_{r_j})\to(a,\sigma) ( a r j , σ r j ) → ( a , σ ) . Then
∣ a ∣ 2 + σ 2 = 1 , β c Z ≥ 0 a.e. on T . |a|^2+\sigma^2=1,\qquad\beta c_Z\geq0\quad\text{a.e. on }\mathbb{T}. ∣ a ∣ 2 + σ 2 = 1 , β c Z ≥ 0 a.e. on T .
Since P ∗ > 0 P_*>0 P ∗ > 0 , (70) implies β c Z = 0 \beta c_Z=0 β c Z = 0 a.e.; the two-zero property of β \beta β gives c Z = 0 c_Z=0 c Z = 0 a.e. By (67) and (68) ,
( Z ⋅ n ) ∘ F = ∣ F ′ ∣ c Z = 0 ⟹ Z ⋅ n = 0 for arclength-a.e. x ∈ ∂ D . (Z\cdot n)\circ F=|F'|c_Z=0\quad\Longrightarrow\quad Z\cdot n=0\quad\text{for arclength-a.e. }x\in\partial D. ( Z ⋅ n ) ∘ F = ∣ F ′ ∣ c Z = 0 ⟹ Z ⋅ n = 0 for arclength-a.e. x ∈ ∂ D .
A bounded convex domain is Lipschitz. The divergence theorem thus gives
0 = ∫ ∂ D Z ⋅ n d s = 2 σ ∣ D ∣ , 0 = ∫ ∂ D ( a ⋅ x ) ( a ⋅ n ) d s = ∣ a ∣ 2 ∣ D ∣ , 0=\int_{\partial D}Z\cdot n\,ds=2\sigma|D|,\qquad0=\int_{\partial D}(a\cdot x)(a\cdot n)\,ds=|a|^2|D|, 0 = ∫ ∂ D Z ⋅ n d s = 2 σ ∣ D ∣ , 0 = ∫ ∂ D ( a ⋅ x ) ( a ⋅ n ) d s = ∣ a ∣ 2 ∣ D ∣ ,
where the second equality uses σ = 0 \sigma=0 σ = 0 . This contradicts ∣ a ∣ 2 + σ 2 = 1 |a|^2+\sigma^2=1 ∣ a ∣ 2 + σ 2 = 1 , proving K A 0 = { 0 } K_A^0=\{0\} K A 0 = { 0 } . The bound on dim K A \dim K_A dim K A follows from (53) . □
The remaining extension concerns compactness and the analytic shape map. Uniform convergence of the potentials up to the wall, together with interior convergence of their gradients, is sufficient for both.
Proof of Remark 1.5 . Let D D D be bounded and convex. Comparison with the Green function of a containing disk gives
0 ≤ G D ( x , y ) ≤ C D ( 1 + ∣ log ∣ x − y ∣ ∣ ) . 0\leq G_D(x,y)\leq C_D(1+|\log|x-y||). 0 ≤ G D ( x , y ) ≤ C D ( 1 + ∣ log ∣ x − y ∣∣ ) .
Hölder’s inequality and interior elliptic estimates imply, for 1 < p < ∞ 1<p<\infty 1 < p < ∞ and D ′ ‾ ⊂ D \overline{D'}\subset D D ′ ⊂ D ,
∥ G f ∥ C ( D ‾ ) ≤ C D , p ∥ f ∥ L p ( D ) , ∥ G f ∥ W 2 , p ( D ′ ) ≤ C D ′ , D , p ∥ f ∥ L p ( D ) . (71) \|Gf\|_{C(\overline{D})}\leq C_{D,p}\|f\|_{L^p(D)},\qquad\|Gf\|_{W^{2,p}(D')}\leq C_{D',D,p}\|f\|_{L^p(D)}. \tag*{(71)} ∥ G f ∥ C ( D ) ≤ C D , p ∥ f ∥ L p ( D ) , ∥ G f ∥ W 2 , p ( D ′ ) ≤ C D ′ , D , p ∥ f ∥ L p ( D ) . (71)
For general f ∈ L p ( D ) f\in L^p(D) f ∈ L p ( D ) , the continuous representative follows by approximating f f f by bounded functions and using the first estimate.
Fix a compact interval I ⊂ ( 0 , ∣ D ∣ ) I\subset(0,|D|) I ⊂ ( 0 , ∣ D ∣ ) . The value estimate (57) and weak-star energy continuity are valid on D D D . As in lemma 6.1, they give compactness of M I \mathcal{M}_I M I in I × L 1 ( D ) I\times L^1(D) I × L 1 ( D ) . For a convergent sequence ( m j , 1 A j ) → ( m , 1 A ) (m_j,1_{A_j})\to(m,1_A) ( m j , 1 A j ) → ( m , 1 A ) , (71) gives
U A j ⟶ U A in C ( D ‾ ) and in C l o c 1 , α ( D ) ( 0 < α < 1 ) . (72) U_{A_j}\longrightarrow U_A\quad\text{in }C(\overline{D})\text{ and in }C_{\mathrm{loc}}^{1,\alpha}(D)\qquad(0<\alpha<1). \tag*{(72)} U A j ⟶ U A in C ( D ) and in C loc 1 , α ( D ) ( 0 < α < 1 ) . (72)
Every cluster value q ∗ q_* q ∗ of q A j q_{A_j} q A j consequently satisfies
1 A = 1 a.e. on { U A > q ∗ } , 1 A = 0 a.e. on { U A < q ∗ } . \mathbf{1}_A=1\ \text{a.e. on }\{U_A>q_*\},\qquad\mathbf{1}_A=0\ \text{a.e. on }\{U_A<q_*\}. 1 A = 1 a.e. on { U A > q ∗ } , 1 A = 0 a.e. on { U A < q ∗ } .
If q ∗ ≠ q A q_*\ne q_A q ∗ = q A , a thin strip on one side of the noncritical level ∂ A \partial A ∂ A contradicts these identities. Hence q A j → q A q_{A_j}\to q_A q A j → q A , and compactness yields
q I : = min ( m , A ) ∈ M I q A > 0 , A ‾ ⊂ K I : = { x ∈ D ‾ : τ ( x ) ≥ q I } ⊂ D , τ = G 1. (73) q_I:=\min_{(m,A)\in\mathcal{M}_I}q_A>0,\qquad\overline{A}\subset K_I:=\{x\in\overline{D}:\tau(x)\ge q_I\}\subset D,\qquad\tau=\mathcal{G}1. \tag*{(73)} q I := ( m , A ) ∈ M I min q A > 0 , A ⊂ K I := { x ∈ D : τ ( x ) ≥ q I } ⊂ D , τ = G 1. (73)
The set K I K_I K I is compact. Since { U A = q A } = ∂ A \{U_A=q_A\}=\partial A { U A = q A } = ∂ A , the potential has a nonzero gap from q A q_A q A outside fixed interior normal tubes. Equations (72) and (73) therefore give the same normal-graph convergence as lemma 6.1.
All graphs in a shape chart have closures in a compact subset of D D D . The regular Green part satisfies H D ∈ C ω ( D × D ) H_D\in C^\omega(D\times D) H D ∈ C ω ( D × D ) , so the layer-potential proof of analyticity in lemma 6.2 applies unchanged. In the derivative calculation, local C 1 C^1 C 1 convergence from (72) replaces convergence on D ‾ \overline{D} D . Thus
D h F ( 0 , q A , m ) ϕ = ( S Γ ϕ − g ϕ , ∫ Γ ϕ d s ) , ind D h F = − 1 , ker D h F = K A 0 = { 0 } . D_h\mathcal{F}(0,q_A,m)\phi=\left(S_\Gamma\phi-g\phi,\int_\Gamma\phi\,\mathrm{d}s\right),\qquad\operatorname{ind}D_h\mathcal{F}=-1,\qquad\ker D_h\mathcal{F}=K_A^0=\{0\}. D h F ( 0 , q A , m ) ϕ = ( S Γ ϕ − g ϕ , ∫ Γ ϕ d s ) , ind D h F = − 1 , ker D h F = K A 0 = { 0 } .
where the last equality is lemma 7.1. The scalar reduction (64) and the constant-energy argument of lemma 6.4 are consequently valid on D D D .
These are precisely the compactness and analytic-reduction properties used in the proof of proposition 6.5. Its maximum-threshold argument on the derived set therefore gives # M m < ∞ \#\mathcal{M}_m<\infty # M m < ∞ for each m ∈ ( 0 , ∣ D ∣ ) m\in(0,|D|) m ∈ ( 0 , ∣ D ∣ ) .
For the compact-interval bound, the scalar residual at ( m 0 , A 0 ) ∈ M I (m_0,A_0)\in\mathcal{M}_I ( m 0 , A 0 ) ∈ M I has finite vanishing order
k m 0 , A 0 : = ord t = q A 0 r ( t , m 0 ) < ∞ ; k_{m_0,A_0}:=\operatorname{ord}_{t=q_{A_0}}r(t,m_0)<\infty; k m 0 , A 0 := ord t = q A 0 r ( t , m 0 ) < ∞ ;
otherwise lemma 6.4 would give infinitely many maximizers at m 0 m_0 m 0 . Weierstrass preparation, as in corollary 6.6, bounds the number of maximizing patches at each nearby mass in this chart by k m 0 , A 0 k_{m_0,A_0} k m 0 , A 0 . Normal-graph compactness supplies finitely many such charts covering M I \mathcal{M}_I M I , with orders k 1 , … , k L k_1,\ldots,k_L k 1 , … , k L ; hence
1 ≤ # M m ≤ ∑ ℓ = 1 L k ℓ = : N D , I ( m ∈ I ) . 1\le\#\mathcal{M}_m\le\sum_{\ell=1}^{L}k_\ell=:N_{D,I}\qquad(m\in I). 1 ≤ # M m ≤ ℓ = 1 ∑ L k ℓ =: N D , I ( m ∈ I ) .
For the rest of this section, D D D has C ∞ C^\infty C ∞ boundary, and convexity is imposed where stated. We combine the compact-interval count with uniqueness near m = 0 m=0 m = 0 and m = ∣ D ∣ m=|D| m = ∣ D ∣ . The full-mass argument uses a fixed normal collar and a positive lower bound for the normal derivative of the torsion function on the entire fluid boundary.
Proposition 7.2. Let D D D be smooth, bounded, and simply connected. There are δ D , c D > 0 \delta_D,c_D>0 δ D , c D > 0 such that, for ∣ D ∣ − δ D < m < ∣ D ∣ |D|-\delta_D<m<|D| ∣ D ∣ − δ D < m < ∣ D ∣ , M m = { 1 A } \mathcal{M}_m=\{\mathbf{1}_A\} M m = { 1 A } and
Q A ( ϕ ) ≥ c D ∥ ϕ ∥ L 2 ( ∂ A ) 2 ( ϕ ∈ L 2 ( ∂ A ) ) . Q_A(\phi)\ge c_D\|\phi\|_{L^2(\partial A)}^2\qquad(\phi\in L^2(\partial A)). Q A ( ϕ ) ≥ c D ∥ ϕ ∥ L 2 ( ∂ A ) 2 ( ϕ ∈ L 2 ( ∂ A )) .
Proof. Set τ = G 1 \tau= \mathcal{G}1 τ = G 1 , ε = ∣ D ∣ − m \varepsilon= |D| - m ε = ∣ D ∣ − m , and fix p > 2 p > 2 p > 2 . Uniformly over A ∈ M m A \in\mathcal{M}_m A ∈ M m , elliptic estimates give
0 ≤ U A ≤ τ , ∥ U A − τ ∥ C 1 ( D ‾ ) ≤ C D , p ∥ 1 A − 1 ∥ p = C D , p ε 1 / p , ∥ ∇ U A ∥ ∞ ≤ M . 0 \le U_A \le\tau,\qquad\|U_A-\tau\|_{C^1(\overline{D})} \le C_{D,p}\|1_A-1\|_p = C_{D,p}\varepsilon^{1/p},\qquad\|\nabla U_A\|_\infty\le M. 0 ≤ U A ≤ τ , ∥ U A − τ ∥ C 1 ( D ) ≤ C D , p ∥ 1 A − 1 ∥ p = C D , p ε 1/ p , ∥∇ U A ∥ ∞ ≤ M .
Moreover, for each fixed θ > 0 \theta> 0 θ > 0 ,
q A ≥ θ ⟹ A ∩ { τ ≤ θ } = ∅ ⟹ ε ≥ ∣ { τ ≤ θ } ∣ > 0. q_A \ge\theta\ \Longrightarrow\ A \cap\{\tau\le\theta\} = \varnothing\ \Longrightarrow\ \varepsilon\ge|\{\tau\le\theta\}| > 0. q A ≥ θ ⟹ A ∩ { τ ≤ θ } = ∅ ⟹ ε ≥ ∣ { τ ≤ θ } ∣ > 0.
Hence q A → 0 q_A \to0 q A → 0 uniformly as ε ↓ 0 \varepsilon\downarrow0 ε ↓ 0 .
Use inward normal coordinates Ψ ( s , r ) \Psi(s,r) Ψ ( s , r ) in a fixed collar 0 ≤ r ≤ r 0 0 \le r \le r_0 0 ≤ r ≤ r 0 , with s s s arclength on ∂ D \partial D ∂ D and J ( s , r ) = ∣ ∂ s Ψ ( s , r ) ∣ J(s,r) = |\partial_s\Psi(s,r)| J ( s , r ) = ∣ ∂ s Ψ ( s , r ) ∣ . The Hopf lemma allows fixed constants a 0 a_0 a 0 , J − J_- J − , J + > 0 J_+ > 0 J + > 0 such that
∂ r ( τ ∘ Ψ ) ≥ 2 a 0 , J − ≤ J ≤ J + . \partial_r(\tau\circ\Psi) \ge2a_0,\qquad J_- \le J \le J_+. ∂ r ( τ ∘ Ψ ) ≥ 2 a 0 , J − ≤ J ≤ J + .
Choose r 0 r_0 r 0 so that the complementary compact core is nonempty. Since min c o r e τ > 0 \min_{\mathrm{core}}\tau> 0 min core τ > 0 , for sufficiently small ε \varepsilon ε ,
∂ r ( U A ∘ Ψ ) ≥ a 0 in the collar , U A > q A on the core . \partial_r(U_A \circ\Psi) \ge a_0\quad\text{in the collar},\qquad U_A > q_A\quad\text{on the core}. ∂ r ( U A ∘ Ψ ) ≥ a 0 in the collar , U A > q A on the core .
Each normal fiber therefore has a unique crossing r = a ( s ) ∈ ( 0 , r 0 ) r=a(s)\in(0,r_0) r = a ( s ) ∈ ( 0 , r 0 ) , and
D ∖ A = { Ψ ( s , r ) : 0 < r ≤ a ( s ) } . D\setminus A = \{\Psi(s,r):0<r\le a(s)\}. D ∖ A = { Ψ ( s , r ) : 0 < r ≤ a ( s )} .
Write L = ∣ ∂ D ∣ L=|\partial D| L = ∣ ∂ D ∣ . The level equation and the mass constraint give
a 0 a ( s ) ≤ q A ≤ M a ( s ) , ε = ∫ ∂ D ∫ 0 a ( s ) J ( s , r ) d r d s ≥ J − L M q A , a_0a(s)\le q_A\le Ma(s),\qquad\varepsilon=\int_{\partial D}\int_0^{a(s)}J(s,r)\,dr\,ds\ge\frac{J_-L}{M}q_A, a 0 a ( s ) ≤ q A ≤ M a ( s ) , ε = ∫ ∂ D ∫ 0 a ( s ) J ( s , r ) d r d s ≥ M J − L q A ,
and consequently
∥ a ∥ ∞ ≤ C w ε , C w = M a 0 J − L . \|a\|_\infty\le C_w\varepsilon,\qquad C_w=\frac{M}{a_0J_-L}. ∥ a ∥ ∞ ≤ C w ε , C w = a 0 J − L M .
Differentiating ( U A ∘ Ψ ) ( s , a ( s ) ) = q A (U_A\circ\Psi)(s,a(s))=q_A ( U A ∘ Ψ ) ( s , a ( s )) = q A yields
∣ a ′ ∣ ≤ J + M a 0 , d s ∂ A = j a ( s ) d s , |a'|\le\frac{J_+M}{a_0},\qquad ds_{\partial A}=j_a(s)\,ds, ∣ a ′ ∣ ≤ a 0 J + M , d s ∂ A = j a ( s ) d s ,
j a : = J ( s , a ) 2 + ∣ a ′ ∣ 2 ≤ L 0 : = J + 1 + ( M / a 0 ) 2 . j_a:=\sqrt{J(s,a)^2+|a'|^2}\le L_0:=J_+\sqrt{1+(M/a_0)^2}. j a := J ( s , a ) 2 + ∣ a ′ ∣ 2 ≤ L 0 := J + 1 + ( M / a 0 ) 2 .
For A , B ∈ M m A,B\in\mathcal{M}_m A , B ∈ M m with graphs a , b a,b a , b , set h = b − a h=b-a h = b − a , f = 1 B − 1 A f=1_B-1_A f = 1 B − 1 A , and δ = C w ε < r 0 \delta=C_w\varepsilon<r_0 δ = C w ε < r 0 . On the strip between the graphs,
u A f ≤ 0 , ∣ u A ( Ψ ( s , r ) ) ∣ ≥ a 0 ∣ r − a ( s ) ∣ . u_Af\le0,\qquad|u_A(\Psi(s,r))|\ge a_0|r-a(s)|. u A f ≤ 0 , ∣ u A ( Ψ ( s , r )) ∣ ≥ a 0 ∣ r − a ( s ) ∣.
Fiberwise integration therefore gives
− 2 ∫ D u A f d x ≥ a 0 J − ∥ h ∥ L 2 ( ∂ D ) 2 . (74) -2\int_Du_Af\,dx\ge a_0J_-\|h\|_{L^2(\partial D)}^2. \tag*{(74)} − 2 ∫ D u A f d x ≥ a 0 J − ∥ h ∥ L 2 ( ∂ D ) 2 . (74)
For z ∈ H 0 1 ( D ) z\in H^1_0(D) z ∈ H 0 1 ( D ) , its pullback z ~ = z ∘ Ψ \widetilde z=z\circ\Psi z = z ∘ Ψ satisfies
∣ z ~ ( s , r ) ∣ 2 ≤ δ ∫ 0 δ ∣ ∂ r z ~ ( s , t ) ∣ 2 d t ( 0 ≤ r ≤ δ ) . |\widetilde z(s,r)|^2\le\delta\int_0^\delta|\partial_r\widetilde z(s,t)|^2\,dt\qquad(0\le r\le\delta). ∣ z ( s , r ) ∣ 2 ≤ δ ∫ 0 δ ∣ ∂ r z ( s , t ) ∣ 2 d t ( 0 ≤ r ≤ δ ) .
This holds first for smooth zero-trace functions and then by Sobolev slicing. Cauchy–Schwarz gives
∣ ∫ D f z d x ∣ ≤ J + δ ∥ h ∥ L 2 ( ∂ D ) ( ∫ ∂ D ∫ 0 δ ∣ ∂ r z ∣ 2 d r d s ) 1 / 2 ≤ J + δ J − ∥ h ∥ L 2 ( ∂ D ) ∥ ∇ z ∥ 2 . \left|\int_D fz\,\mathrm{d}x\right| \le J_+ \sqrt{\delta}\|h\|_{L^2(\partial D)}\left(\int_{\partial D}\int_0^\delta|\partial_r z|^2\,\mathrm{d}r\,\mathrm{d}s\right)^{1/2}
\le\frac{J_+ \sqrt{\delta}}{\sqrt{J_-}}\|h\|_{L^2(\partial D)}\|\nabla z\|_2. ∫ D f z d x ≤ J + δ ∥ h ∥ L 2 ( ∂ D ) ( ∫ ∂ D ∫ 0 δ ∣ ∂ r z ∣ 2 d r d s ) 1/2 ≤ J − J + δ ∥ h ∥ L 2 ( ∂ D ) ∥∇ z ∥ 2 .
Thus
∥ f ∥ H ˙ D − 1 2 ≤ J + 2 J − δ ∥ h ∥ L 2 ( ∂ D ) 2 . (75) \|f\|_{\dot H_D^{-1}}^2 \le\frac{J_+^2}{J_-}\delta\|h\|_{L^2(\partial D)}^2. \tag*{(75)} ∥ f ∥ H ˙ D − 1 2 ≤ J − J + 2 δ ∥ h ∥ L 2 ( ∂ D ) 2 . (75)
Since ∣ A ∣ = ∣ B ∣ |A|=|B| ∣ A ∣ = ∣ B ∣ and E ( 1 A ) = E ( 1 B ) \mathcal{E}(1_A)=\mathcal{E}(1_B) E ( 1 A ) = E ( 1 B ) , (28) yields
− 2 ∫ D u A f d x = ∥ f ∥ H ˙ D − 1 2 , ( a 0 J − − J + 2 J − δ ) ∥ h ∥ L 2 ( ∂ D ) 2 ≤ 0. -2\int_D u_A f\,\mathrm{d}x=\|f\|_{\dot H_D^{-1}}^2,\qquad\left(a_0J_- - \frac{J_+^2}{J_-}\delta\right)\|h\|_{L^2(\partial D)}^2\le0. − 2 ∫ D u A f d x = ∥ f ∥ H ˙ D − 1 2 , ( a 0 J − − J − J + 2 δ ) ∥ h ∥ L 2 ( ∂ D ) 2 ≤ 0.
Hence A = B A=B A = B for sufficiently small δ \delta δ .
The same normal-ray estimate gives, with Γ = ∂ A \Gamma=\partial A Γ = ∂ A ,
∫ Γ z 2 d s ≤ L 0 J − δ ∫ D ∣ ∇ z ∣ 2 d x , ∥ φ d s ∣ Γ ∥ H ˙ D − 1 2 ≤ L 0 J − δ ∥ φ ∥ L 2 ( Γ ) 2 . \int_\Gamma z^2\,\mathrm{d}s\le\frac{L_0}{J_-}\delta\int_D|\nabla z|^2\,\mathrm{d}x,\qquad\|\varphi\,\mathrm{d}s|_\Gamma\|_{\dot H_D^{-1}}^2\le\frac{L_0}{J_-}\delta\|\varphi\|_{L^2(\Gamma)}^2. ∫ Γ z 2 d s ≤ J − L 0 δ ∫ D ∣∇ z ∣ 2 d x , ∥ φ d s ∣ Γ ∥ H ˙ D − 1 2 ≤ J − L 0 δ ∥ φ ∥ L 2 ( Γ ) 2 .
Since g ≥ a 0 g\ge a_0 g ≥ a 0 on Γ \Gamma Γ ,
Q A ( φ ) ≥ ( a 0 − L 0 J − δ ) ∥ φ ∥ L 2 ( Γ ) 2 . Q_A(\varphi)\ge\left(a_0-\frac{L_0}{J_-}\delta\right)\|\varphi\|_{L^2(\Gamma)}^2. Q A ( φ ) ≥ ( a 0 − J − L 0 δ ) ∥ φ ∥ L 2 ( Γ ) 2 .
Choose δ D > 0 \delta_D>0 δ D > 0 below the preliminary mass-deficit bound and so that
C w δ D ≤ min { r 0 2 , a 0 J − 2 2 J + 2 , a 0 J − 2 L 0 } . C_w\delta_D\le\min\left\{\frac{r_0}{2},\frac{a_0J_-^2}{2J_+^2},\frac{a_0J_-}{2L_0}\right\}. C w δ D ≤ min { 2 r 0 , 2 J + 2 a 0 J − 2 , 2 L 0 a 0 J − } .
Both conclusions follow with c D = a 0 / 2 c_D=a_0/2 c D = a 0 /2 .
Proposition 7.3. If D D D is smooth, bounded, and convex, there is m D > 0 m_D>0 m D > 0 such that # M m = 1 \#\mathcal{M}_m=1 # M m = 1 for 0 < m < m D 0<m<m_D 0 < m < m D .
Proof. We verify the concentration, single-core, and strength hypotheses of [7 (1.12)–(1.13), Theorems 1.2–1.3] . For m j ↓ 0 m_j\downarrow0 m j ↓ 0 and A j ∈ M m j A_j\in\mathcal{M}_{m_j} A j ∈ M m j , put
λ j = m j − 1 , ω j = λ j 1 A j , ψ j = λ j U A j , μ j = λ j q A j . \lambda_j=m_j^{-1},\qquad\omega_j=\lambda_j1_{A_j},\qquad\psi_j=\lambda_jU_{A_j},\qquad\mu_j=\lambda_jq_{A_j}. λ j = m j − 1 , ω j = λ j 1 A j , ψ j = λ j U A j , μ j = λ j q A j .
The scaling
ρ ⟼ λ j ρ : A m j ⟶ { 0 ≤ ω ≤ λ j : ∫ D ω = 1 } , E ( λ j ρ ) = λ j 2 E ( ρ ) \rho\longmapsto\lambda_j\rho:\mathcal{A}_{m_j}\longrightarrow\left\{0\le\omega\le\lambda_j:\int_D\omega=1\right\},\qquad\mathcal{E}(\lambda_j\rho)=\lambda_j^2\mathcal{E}(\rho) ρ ⟼ λ j ρ : A m j ⟶ { 0 ≤ ω ≤ λ j : ∫ D ω = 1 } , E ( λ j ρ ) = λ j 2 E ( ρ )
is bijective, so ω j \omega_j ω j is an absolute maximizer in the circulation-one class and
− Δ ψ j = λ j 1 A j , ψ j ∣ ∂ D = 0 , A j = { ψ j > μ j } , λ j ∣ A j ∣ = 1. -\Delta\psi_j=\lambda_j1_{A_j},\qquad\psi_j|_{\partial D}=0,\qquad A_j=\{\psi_j>\mu_j\},\qquad\lambda_j|A_j|=1. − Δ ψ j = λ j 1 A j , ψ j ∣ ∂ D = 0 , A j = { ψ j > μ j } , λ j ∣ A j ∣ = 1.
The Robin function R D R_D R D has a unique critical point x D x_D x D , with D 2 R D ( x D ) > 0 D^2R_D(x_D)>0 D 2 R D ( x D ) > 0 ; this is the Robin-function input to [7 Theorem 1.3] and requires no strict boundary curvature assumption. Turkington’s concentration theorem [27 ] , stated in [9 Theorem 2.1] , gives, after subsequence extraction,
x j : = ∫ D x ω j d x ⟶ x D , diam A j ≤ C λ j − 1 / 2 ⟶ 0. x_j := \int_D x\omega_j\,\mathrm{d}x \longrightarrow x_D,\qquad\operatorname{diam} A_j \le C\lambda_j^{-1/2} \longrightarrow0. x j := ∫ D x ω j d x ⟶ x D , diam A j ≤ C λ j − 1/2 ⟶ 0.
Thus
A j ⊂ B ∣ x j − x D ∣ + C λ j − 1 / 2 ( x D ) ⊂ B o ( 1 ) ( x D ) . A_j \subset B_{\lvert x_j-x_D\rvert+C\lambda_j^{-1/2}}(x_D) \subset B_{o(1)}(x_D). A j ⊂ B ∣ x j − x D ∣ + C λ j − 1/2 ( x D ) ⊂ B o ( 1 ) ( x D ) .
Since H D H_D H D is uniformly bounded near ( x D , x D ) (x_D,x_D) ( x D , x D ) and ∫ D ω j = 1 \int_D\omega_j=1 ∫ D ω j = 1 , for x ∈ ∂ A j x\in\partial A_j x ∈ ∂ A j ,
μ j = ψ j ( x ) = ∫ D G D ( x , y ) ω j ( y ) d y ≥ log λ j 4 π − C ⟶ + ∞ . \mu_j=\psi_j(x)=\int_D G_D(x,y)\omega_j(y)\,\mathrm{d}y\ge\frac{\log\lambda_j}{4\pi}-C\longrightarrow+\infty. μ j = ψ j ( x ) = ∫ D G D ( x , y ) ω j ( y ) d y ≥ 4 π log λ j − C ⟶ + ∞.
Turkington’s single-core result [27 Corollary 4.6, p. 1026] gives, for large j j j ,
A j is simply connected , ∂ A j ∈ C 1 . A_j\text{ is simply connected},\qquad\partial A_j\in C^1. A j is simply connected , ∂ A j ∈ C 1 .
The nonvanishing normal derivative also follows from theorem 1.1. For a fixed small ball B d ( x D ) ⊂ D B_d(x_D)\subset D B d ( x D ) ⊂ D ,
A j = B d ( x D ) ∩ { ψ j > μ j } ⊂ B d ( x D ) . A_j=B_d(x_D)\cap\{\psi_j>\mu_j\}\subset B_d(x_D). A j = B d ( x D ) ∩ { ψ j > μ j } ⊂ B d ( x D ) .
Thus [7 Theorem 1.2] applies with k = 1 k=1 k = 1 and κ 1 = 1 \kappa_1=1 κ 1 = 1 , including the simply connected core required in its published formulation. Its threshold is an unknown.
If the conclusion failed, choose distinct A j , B j ∈ M m j A_j,B_j\in\mathcal{M}_{m_j} A j , B j ∈ M m j at m j ↓ 0 m_j\downarrow0 m j ↓ 0 . Apply the preceding results to both sequences. They concentrate at the same nondegenerate point x D x_D x D , with the same λ j \lambda_j λ j and circulation. Local uniqueness gives ψ A j = ψ B j \psi_{A_j}=\psi_{B_j} ψ A j = ψ B j and A j = B j A_j=B_j A j = B j for large j j j , a contradiction. □
Proof of Theorem 1.4. Existence follows from proposition 2.1. Choose
d = min { m D / 2 , δ D / 2 , ∣ D ∣ / 4 } > 0 , I = [ d , ∣ D ∣ − d ] . d=\min\{m_D/2,\delta_D/2,\lvert D\rvert/4\}>0,\qquad I=[d,\lvert D\rvert-d]. d = min { m D /2 , δ D /2 , ∣ D ∣ /4 } > 0 , I = [ d , ∣ D ∣ − d ] .
By corollary 6.6 and propositions 7.2 and 7.3,
1 ≤ # M m ≤ { N I , m ∈ I , 1 , m ∈ ( 0 , ∣ D ∣ ) ∖ I . 1\le\#\mathcal{M}_m\le
\begin{cases}
N_I,&m\in I,\\
1,&m\in(0,\lvert D\rvert)\setminus I.
\end{cases} 1 ≤ # M m ≤ { N I , 1 , m ∈ I , m ∈ ( 0 , ∣ D ∣) ∖ I .
Hence N D = max { 1 , N I } N_D=\max\{1,N_I\} N D = max { 1 , N I } depends only on D D D . □
Appendix A. General simply connected domains
Throughout this appendix, D D D is a bounded simply connected domain with C ∞ C^\infty C ∞ boundary. The regularity, variational, and nodal arguments above apply in any smooth bounded simply connected domain. Convexity enters through the separator in lemma 5.5, which proves K A 0 = { 0 } K_A^0=\{0\} K A 0 = { 0 } . Without it, the bound dim K A 0 ≤ 2 \dim K_A^0\le2 dim K A 0 ≤ 2 alone does not imply isolation.
A conformal change of variables makes the additional term in the boundary comparison explicit.
Identify R 2 \mathbb{R}^2 R 2 with C \mathbb{C} C . Fix A ∈ M m A\in\mathcal{M}_m A ∈ M m and 0 ≠ ϕ ∈ K A 0 0\ne\phi\in K_A^0 0 = ϕ ∈ K A 0 , and set
U = G 1 A , v = G ( ϕ d s ∣ Γ ) , P = − ∂ n U > 0 , b = ∂ n v . U=\mathcal{G}1_A,\qquad v=\mathcal{G}(\phi\,\mathrm{d}s|_\Gamma),\qquad P=-\partial_nU>0,\qquad b=\partial_nv. U = G 1 A , v = G ( ϕ d s ∣ Γ ) , P = − ∂ n U > 0 , b = ∂ n v .
Let F : D → D F:D\to D F : D → D be conformal, smoothly extended to the closed disk, with F ′ ≠ 0 F'\ne0 F ′ = 0 there. Put
α = ∣ F ′ ∣ 2 , Ω = F − 1 ( A ) , U ~ = U ∘ F , v ~ = v ∘ F . \alpha=\lvert F'\rvert^2,\qquad\Omega=F^{-1}(A),\qquad\widetilde U=U\circ F,\qquad\widetilde v=v\circ F. α = ∣ F ′ ∣ 2 , Ω = F − 1 ( A ) , U = U ∘ F , v = v ∘ F .
Then
− Δ w U ~ = α 1 Ω , m = ∫ Ω α d w . -\Delta_w\widetilde U=\alpha1_{\Omega},\qquad m=\int_{\Omega}\alpha\,dw. − Δ w U = α 1 Ω , m = ∫ Ω α d w .
On w = e i θ w=e^{i\theta} w = e i θ , write P ~ = − ∂ r U ~ \widetilde P=-\partial_r\widetilde U P = − ∂ r U and b ~ = ∂ r v ~ \widetilde b=\partial_r\widetilde v b = ∂ r v . For Y ( w ) = a + σ w Y(w)=a+\sigma w Y ( w ) = a + σ w , a ∈ C a\in\mathbb C a ∈ C , σ ∈ R \sigma\in\mathbb R σ ∈ R , and Z ( F ( w ) ) = F ′ ( w ) Y ( w ) Z(F(w))=F'(w)Y(w) Z ( F ( w )) = F ′ ( w ) Y ( w ) ,
P ∘ F = P ~ / ∣ F ′ ∣ , b ∘ F = b ~ / ∣ F ′ ∣ , d s ∂ D = ∣ F ′ ∣ d θ , P\circ F=\widetilde P/\lvert F'\rvert,\qquad b\circ F=\widetilde b/\lvert F'\rvert,\qquad ds_{\partial D}=\lvert F'\rvert\,d\theta, P ∘ F = P / ∣ F ′ ∣ , b ∘ F = b / ∣ F ′ ∣ , d s ∂ D = ∣ F ′ ∣ d θ ,
( Z ⋅ n ) ∘ F = ∣ F ′ ∣ y , y ( θ ) = Re ( a e − i θ ) + σ . (Z\cdot n)\circ F=\lvert F'\rvert y,\qquad y(\theta)=\operatorname{Re}(ae^{-i\theta})+\sigma. ( Z ⋅ n ) ∘ F = ∣ F ′ ∣ y , y ( θ ) = Re ( a e − i θ ) + σ .
By proposition 5.3, b ~ \widetilde b b has two simple zeros. Choose lifts θ 1 < θ 2 < θ 1 + 2 π \theta_1<\theta_2<\theta_1+2\pi θ 1 < θ 2 < θ 1 + 2 π with b ~ > 0 \widetilde b>0 b > 0 on ( θ 1 , θ 2 ) (\theta_1,\theta_2) ( θ 1 , θ 2 ) , and take
c = θ 1 + θ 2 2 , β = θ 2 − θ 1 2 ∈ ( 0 , π ) , Y ( w ) = e i c − cos β w . c=\frac{\theta_1+\theta_2}{2},\qquad\beta=\frac{\theta_2-\theta_1}{2}\in(0,\pi),\qquad Y(w)=e^{ic}-\cos\beta\,w. c = 2 θ 1 + θ 2 , β = 2 θ 2 − θ 1 ∈ ( 0 , π ) , Y ( w ) = e i c − cos β w .
Thus y = cos ( θ − c ) − cos β y=\cos(\theta-c)-\cos\beta y = cos ( θ − c ) − cos β has the sign of b ~ \widetilde b b , and
∫ ∂ D P b ( Z ⋅ n ) d s = ∫ 0 2 π P ~ b ~ y d θ > 0. \int_{\partial D}Pb(Z\cdot n)\,ds=\int_0^{2\pi}\widetilde P\widetilde b\,y\,d\theta>0. ∫ ∂ D P b ( Z ⋅ n ) d s = ∫ 0 2 π P b y d θ > 0.
The transformed divergence and zero moment are
( div x Z ) ∘ F = 2 Re ( σ + F ′ ′ F ′ Y ) , α ( div x Z ) ∘ F = div w ( α Y ) , (\operatorname{div}_x Z)\circ F=2\operatorname{Re}\left(\sigma+\frac{F''}{F'}Y\right),\qquad\alpha(\operatorname{div}_x Z)\circ F=\operatorname{div}_w(\alpha Y), ( div x Z ) ∘ F = 2 Re ( σ + F ′ F ′′ Y ) , α ( div x Z ) ∘ F = div w ( α Y ) ,
∫ Ω α v ~ d w = ∫ A v d x = ∫ Γ U ϕ d s = q ∫ Γ ϕ d s = 0. \int_{\Omega}\alpha\widetilde v\,dw=\int_Av\,dx=\int_{\Gamma}U\phi\,ds=q\int_{\Gamma}\phi\,ds=0. ∫ Ω α v d w = ∫ A v d x = ∫ Γ U ϕ d s = q ∫ Γ ϕ d s = 0.
Consequently (55) becomes
0 < ∫ ∂ D P b ( Z ⋅ n ) d s = − ∫ Ω v ~ div w ( α Y ) d w = − ∫ Ω v ~ Y ⋅ ∇ α d w . 0<\int_{\partial D}Pb(Z\cdot n)\,ds=-\int_{\Omega}\widetilde v\,\operatorname{div}_w(\alpha Y)\,dw=-\int_{\Omega}\widetilde vY\cdot\nabla\alpha\,dw. 0 < ∫ ∂ D P b ( Z ⋅ n ) d s = − ∫ Ω v div w ( α Y ) d w = − ∫ Ω v Y ⋅ ∇ α d w .
The mass constraint does not cancel this weighted moment.
For conformal fields on the physical domain,
div Z ≡ 2 σ ⟺ Z ( z ) = a + ( σ + i ω ) z , a ∈ C , σ , ω ∈ R , \operatorname{div}Z\equiv2\sigma\quad\Longleftrightarrow\quad Z(z)=a+(\sigma+i\omega)z,\qquad a\in\mathbb C,\ \sigma,\omega\in\mathbb R, div Z ≡ 2 σ ⟺ Z ( z ) = a + ( σ + iω ) z , a ∈ C , σ , ω ∈ R ,
since a holomorphic derivative with constant real part is constant. Such fields satisfy ∫ ∂ D P b ( Z ⋅ n ) d s = 0 \int_{\partial D}Pb(Z\cdot n)\,ds=0 ∫ ∂ D P b ( Z ⋅ n ) d s = 0 , but the required sign comparison is established only for convex D D D . A conformal field tangent to ∂ D \partial D ∂ D has zero boundary and bulk terms in (55) , so adding it does not alter the obstruction.
For a general smooth field, the bilinear stress identity instead gives
∫ ∂ D P b ( Z ⋅ n ) d s = − ∫ A v div Z d x − ∫ D ( E Z ∇ U ) ⋅ ∇ v d x , \int_{\partial D}Pb(Z\cdot n)\,ds=-\int_Av\operatorname{div}Z\,dx-\int_D(\mathcal E_Z\nabla U)\cdot\nabla v\,dx, ∫ ∂ D P b ( Z ⋅ n ) d s = − ∫ A v div Z d x − ∫ D ( E Z ∇ U ) ⋅ ∇ v d x ,
E Z = D Z + ( D Z ) T − ( div Z ) I . \mathcal E_Z=DZ+(DZ)^T-(\operatorname{div}Z)I. E Z = D Z + ( D Z ) T − ( div Z ) I .
Prescribing divergence therefore introduces a strain term whose sign is not controlled by the present trace inequality.
For smooth periodic families, the same stress identity includes a term arising from the variation of the comparison field.
Proposition A.1. Let A t ∈ M m A_t \in\mathcal{M}_m A t ∈ M m be a C 2 C^2 C 2 family of period T T T in local C 1 , α C^{1,\alpha} C 1 , α normal graph coordinates. Set
U t = G 1 A t , u t = U t − q ( t ) , Γ t = ∂ A t , g t = ∣ ∇ U t ∣ Γ t , U_t = \mathcal{G}\mathbf{1}_{A_t}, \qquad u_t = U_t-q(t), \qquad\Gamma_t = \partial A_t, \qquad g_t = \lvert\nabla U_t\rvert_{\Gamma_t}, U t = G 1 A t , u t = U t − q ( t ) , Γ t = ∂ A t , g t = ∣ ∇ U t ∣ Γ t , and let ϕ t \phi_t ϕ t be the outward normal velocity. Then q ∈ C 2 q \in C^2 q ∈ C 2 and
∫ Γ t ϕ t d s = 0 , w t : = U ˙ t = G ( ϕ t d s ∣ Γ t ) , v t : = u ˙ t = w t − q ˙ ( t ) , \int_{\Gamma_t}\phi_t\,\mathrm{d}s=0, \qquad w_t:=\dot{U}_t=\mathcal{G}(\phi_t\,\mathrm{d}s\vert_{\Gamma_t}), \qquad v_t:=\dot{u}_t=w_t-\dot{q}(t), ∫ Γ t ϕ t d s = 0 , w t := U ˙ t = G ( ϕ t d s ∣ Γ t ) , v t := u ˙ t = w t − q ˙ ( t ) , S Γ t ϕ t − g t ϕ t = q ˙ ( t ) , ϕ t ∈ K A t , S_{\Gamma_t}\phi_t-g_t\phi_t=\dot{q}(t), \qquad\phi_t\in K_{A_t}, S Γ t ϕ t − g t ϕ t = q ˙ ( t ) , ϕ t ∈ K A t , ∫ A t w t d x = 0 , ∫ A t v t d x = − m q ˙ ( t ) . (76) \int_{A_t}w_t\,\mathrm{d}x=0, \qquad\int_{A_t}v_t\,\mathrm{d}x=-m\dot{q}(t). \tag*{(76)} ∫ A t w t d x = 0 , ∫ A t v t d x = − m q ˙ ( t ) . (76) For P t = − ∂ n U t > 0 P_t=-\partial_nU_t>0 P t = − ∂ n U t > 0 and b t = ∂ n w t b_t=\partial_nw_t b t = ∂ n w t , P ˙ t = − b t \dot{P}_t=-b_t P ˙ t = − b t . Every C 1 C^1 C 1 periodic family Z t Z_t Z t of smooth conformal fields on D ‾ \overline{D} D satisfies
∫ 0 T ∫ ∂ D P t b t ( Z t ⋅ n ) d s d t = ∫ 0 T ∫ A t u t div Z ˙ t d x d t . (77) \int_0^T\int_{\partial D}P_tb_t(Z_t\cdot n)\,\mathrm{d}s\,\mathrm{d}t=\int_0^T\int_{A_t}u_t\operatorname{div}\dot{Z}_t\,\mathrm{d}x\,\mathrm{d}t. \tag*{(77)} ∫ 0 T ∫ ∂ D P t b t ( Z t ⋅ n ) d s d t = ∫ 0 T ∫ A t u t div Z ˙ t d x d t . (77) Both sides equal
1 2 ∫ 0 T ∫ ∂ D P t 2 ( Z ˙ t ⋅ n ) d s d t . (78) \frac{1}{2}\int_0^T\int_{\partial D}P_t^2(\dot{Z}_t\cdot n)\,\mathrm{d}s\,\mathrm{d}t. \tag*{(78)} 2 1 ∫ 0 T ∫ ∂ D P t 2 ( Z ˙ t ⋅ n ) d s d t . (78)
Proof. In a fixed normal chart, write
Ξ ( ξ , r ) = x ( ξ ) + r ν ∗ ( ξ ) , J ( ξ , r ) = 1 + κ ( ξ ) r , A h = { r < h ( ξ ) } , \Xi(\xi,r)=x(\xi)+r\nu_*(\xi), \qquad J(\xi,r)=1+\kappa(\xi)r, \qquad A_h=\{r<h(\xi)\}, Ξ ( ξ , r ) = x ( ξ ) + r ν ∗ ( ξ ) , J ( ξ , r ) = 1 + κ ( ξ ) r , A h = { r < h ( ξ )} ,
and define
⟨ T h k , z ⟩ = ∫ Γ ∗ k ( ξ ) J ( ξ , h ( ξ ) ) z ( Ξ ( ξ , h ( ξ ) ) ) d s ∗ . \langle T_hk,z\rangle=\int_{\Gamma_*}k(\xi)J(\xi,h(\xi))z(\Xi(\xi,h(\xi)))\,\mathrm{d}s_*. ⟨ T h k , z ⟩ = ∫ Γ ∗ k ( ξ ) J ( ξ , h ( ξ )) z ( Ξ ( ξ , h ( ξ ))) d s ∗ .
The trace and normal-slice estimates of proposition 3.3 give, uniformly on a small graph ball,
∥ 1 A h + k − 1 A h − T h k ∥ H ˙ D − 1 ≤ C ( ∥ k ∥ ∞ 3 / 2 + ∥ k ∥ ∞ 2 ) , \left\|\mathbf{1}_{A_{h+k}}-\mathbf{1}_{A_h}-T_hk\right\|_{\dot{H}^{-1}_D}\leq C\left(\|k\|_{\infty}^{3/2}+\|k\|_{\infty}^2\right), ∥ 1 A h + k − 1 A h − T h k ∥ H ˙ D − 1 ≤ C ( ∥ k ∥ ∞ 3/2 + ∥ k ∥ ∞ 2 ) ,
∥ T h − T ℓ ∥ C 1 , α , H ˙ D − 1 ≤ C ( ∥ h − ℓ ∥ ∞ 1 / 2 + ∥ h − ℓ ∥ ∞ ) . \|T_h-T_\ell\|_{C^{1,\alpha},\dot{H}^{-1}_D}\leq C\left(\|h-\ell\|_{\infty}^{1/2}+\|h-\ell\|_{\infty}\right). ∥ T h − T ℓ ∥ C 1 , α , H ˙ D − 1 ≤ C ( ∥ h − ℓ ∥ ∞ 1/2 + ∥ h − ℓ ∥ ∞ ) .
Thus h ↦ 1 A h h\mapsto\mathbf{1}_{A_h} h ↦ 1 A h is C 1 C^1 C 1 into H ˙ D − 1 \dot{H}^{-1}_D H ˙ D − 1 . Since
j h = ( J ( ξ , h ( ξ ) ) 2 + ∣ ∂ ξ h ∣ 2 ) 1 / 2 , ( ϕ t ∘ X h ) j h = h ˙ J h , j_h=\left(J(\xi,h(\xi))^2+\lvert\partial_\xi h\rvert^2\right)^{1/2}, \qquad(\phi_t\circ X_h)j_h=\dot{h}J_h, j h = ( J ( ξ , h ( ξ ) ) 2 + ∣ ∂ ξ h ∣ 2 ) 1/2 , ( ϕ t ∘ X h ) j h = h ˙ J h ,
we obtain T h h ˙ = ϕ t d s ∣ Γ t T_h\dot{h}=\phi_t\,\mathrm{d}s\vert_{\Gamma_t} T h h ˙ = ϕ t d s ∣ Γ t and the formula for w t w_t w t in H 0 1 ( D ) H^1_0(D) H 0 1 ( D ) .
The material trace is analytic by (61) . Its constant value is q ( t ) q(t) q ( t ) , so q ∈ C 2 q\in C^2 q ∈ C 2 ; differentiating it gives
S Γ t ϕ t − g t ϕ t = q ˙ ( t ) . S_{\Gamma_t}\phi_t-g_t\phi_t=\dot{q}(t). S Γ t ϕ t − g t ϕ t = q ˙ ( t ) .
Equivalently, the fixed-boundary trace derivative is S Γ t ϕ t S_{\Gamma_t}\phi_t S Γ t ϕ t , and the evaluation derivative is ∇ U t ⋅ ( ϕ t ν t ) = − g t ϕ t \nabla U_t\cdot(\phi_t\nu_t)=-g_t\phi_t ∇ U t ⋅ ( ϕ t ν t ) = − g t ϕ t . Therefore
Q A t ( ϕ t ) = − q ˙ ( t ) , ∫ Γ t ϕ t d s = 0 , ∫ A t w t d x = ∫ Γ t U t ϕ t d s = q ( t ) ∫ Γ t ϕ t d s = 0. Q_{A_t}(\phi_t)=-\dot{q}(t), \qquad\int_{\Gamma_t}\phi_t\,\mathrm{d}s=0, \qquad\int_{A_t}w_t\,\mathrm{d}x=\int_{\Gamma_t}U_t\phi_t\,\mathrm{d}s=q(t)\int_{\Gamma_t}\phi_t\,\mathrm{d}s=0. Q A t ( ϕ t ) = − q ˙ ( t ) , ∫ Γ t ϕ t d s = 0 , ∫ A t w t d x = ∫ Γ t U t ϕ t d s = q ( t ) ∫ Γ t ϕ t d s = 0.
which proves (A.1).
A finite chart cover of the parameter circle gives ⋃ t A t ‾ ⊂ K ⊂ D \bigcup_t \overline{A_t} \subset K \subset D ⋃ t A t ⊂ K ⊂ D . In a fixed outer collar O \mathcal{O} O , the Green kernel and its derivatives are smooth for y ∈ K y \in K y ∈ K . Differentiating the chart volume integrals twice yields
t ↦ U t ∈ C 2 ( R / T Z ; C j + 1 ( O ‾ ) ) ( j ≥ 0 ) , P ˙ t = − b t . t \mapsto U_t \in C^2(\mathbb{R}/T\mathbb{Z}; C^{j+1}(\overline{\mathcal{O}})) \quad(j \ge0), \qquad\dot{P}_t=-b_t. t ↦ U t ∈ C 2 ( R / T Z ; C j + 1 ( O )) ( j ≥ 0 ) , P ˙ t = − b t .
No second source derivative in H ˙ D − 1 \dot{H}^{-1}_D H ˙ D − 1 is used.
For a fixed conformal field Z Z Z , the state stress vector ( Z ⋅ ∇ U t ) ∇ U t − 1 2 ∣ ∇ U t ∣ 2 Z (Z \cdot\nabla U_t)\nabla U_t-\frac{1}{2}|\nabla U_t|^2Z ( Z ⋅ ∇ U t ) ∇ U t − 2 1 ∣∇ U t ∣ 2 Z gives
1 2 ∫ ∂ D P t 2 ( Z ⋅ n ) d s = − ∫ A t Z ⋅ ∇ u t d x = ∫ A t u t div Z d x . (79) \frac{1}{2}\int_{\partial D}P_t^2(Z\cdot n)\,ds=-\int_{A_t}Z\cdot\nabla u_t\,dx=\int_{A_t}u_t\operatorname{div}Z\,dx. \tag*{(79)} 2 1 ∫ ∂ D P t 2 ( Z ⋅ n ) d s = − ∫ A t Z ⋅ ∇ u t d x = ∫ A t u t div Z d x . (79)
Since u t ∣ Γ t = 0 u_t|_{\Gamma_t}=0 u t ∣ Γ t = 0 , differentiation gives
∫ ∂ D P t b t ( Z ⋅ n ) d s = − ∫ A t v t div Z d x . \int_{\partial D}P_tb_t(Z\cdot n)\,ds=-\int_{A_t}v_t\operatorname{div}Z\,dx. ∫ ∂ D P t b t ( Z ⋅ n ) d s = − ∫ A t v t div Z d x .
For J ( t ) = ∫ A t u t div Z t d x J(t)=\int_{A_t}u_t\operatorname{div}Z_t\,dx J ( t ) = ∫ A t u t div Z t d x ,
J ′ ( t ) = − ∫ ∂ D P t b t ( Z t ⋅ n ) d s + ∫ A t u t div Z ˙ t d x , J ( T ) = J ( 0 ) . J'(t)=-\int_{\partial D}P_tb_t(Z_t\cdot n)\,ds+\int_{A_t}u_t\operatorname{div}\dot{Z}_t\,dx,\qquad J(T)=J(0). J ′ ( t ) = − ∫ ∂ D P t b t ( Z t ⋅ n ) d s + ∫ A t u t div Z ˙ t d x , J ( T ) = J ( 0 ) .
Integration proves (A.2); (A.4) with Z = Z ˙ t Z=\dot{Z}_t Z = Z ˙ t proves (A.3). □ \square □
For translations, rotations, and dilations,
Z ( z ) = a + ( σ + i ω ) z ⟹ ∫ ∂ D P t b t ( Z ⋅ n ) d s = 2 σ m q ˙ ( t ) . Z(z)=a+(\sigma+i\omega)z \quad\Longrightarrow\quad\int_{\partial D}P_tb_t(Z\cdot n)\,ds=2\sigma m\dot{q}(t). Z ( z ) = a + ( σ + iω ) z ⟹ ∫ ∂ D P t b t ( Z ⋅ n ) d s = 2 σ m q ˙ ( t ) .
In particular, dilation orthogonality at a fixed parameter requires q ˙ ( t ) = 0 \dot{q}(t)=0 q ˙ ( t ) = 0 .
Proposition A.2. In proposition A.1, assume q q q is constant and ϕ t ≠ 0 \phi_t\ne0 ϕ t = 0 for every t t t . Fix F : D → D F:D\to D F : D → D conformal and put
B ( t , θ ) = ∣ F ′ ( e i θ ) ∣ b t ( F ( e i θ ) ) . B(t,\theta)=|F'(e^{i\theta})|b_t(F(e^{i\theta})). B ( t , θ ) = ∣ F ′ ( e i θ ) ∣ b t ( F ( e i θ )) . Its zeros admit C 1 C^1 C 1 periodic circle-valued labels z ± = e i θ ± z_\pm=e^{i\theta_\pm} z ± = e i θ ± , distinguished by ± ∂ θ B ( t , θ ± ) > 0 \pm\partial_\theta B(t,\theta_\pm)>0 ± ∂ θ B ( t , θ ± ) > 0 . They are uniformly separated. Set
a = min B = 0 ∣ ∂ θ B ∣ > 0 , M = max ∣ ∂ t B ∣ . a=\min_{B=0}|\partial_\theta B|>0,\qquad M=\max|\partial_tB|. a = B = 0 min ∣ ∂ θ B ∣ > 0 , M = max ∣ ∂ t B ∣. Local angular lifts satisfy
θ ˙ ± = r ± : = − ∂ t B ( t , θ ± ) ∂ θ B ( t , θ ± ) , ∣ r ± ∣ ≤ M / a . \dot{\theta}_\pm=r_\pm:=-\frac{\partial_tB(t,\theta_\pm)}{\partial_\theta B(t,\theta_\pm)},\qquad|r_\pm|\le M/a. θ ˙ ± = r ± := − ∂ θ B ( t , θ ± ) ∂ t B ( t , θ ± ) , ∣ r ± ∣ ≤ M / a . With
ℓ = Arg ( 0 , 2 π ) ( z − / z + ) , β = ℓ / 2 , E = z + e i β , \ell=\operatorname{Arg}_{(0,2\pi)}(z_-/z_+),\qquad\beta=\ell/2,\qquad E=z_+e^{i\beta}, ℓ = Arg ( 0 , 2 π ) ( z − / z + ) , β = ℓ /2 , E = z + e i β , one has
ℓ ˙ = r − − r + , E ˙ = i 2 ( r + + r − ) E . \dot{\ell}=r_--r_+,\qquad\dot{E}=\frac{i}{2}(r_++r_-)E. ℓ ˙ = r − − r + , E ˙ = 2 i ( r + + r − ) E . The labels have the same integer winding over a period. The fields
Y t ( w ) = E ( t ) − cos β ( t ) w , Z t ( F ( w ) ) = F ′ ( w ) Y t ( w ) Y_t(w)=E(t)-\cos\beta(t)w,\qquad Z_t(F(w))=F'(w)Y_t(w) Y t ( w ) = E ( t ) − cos β ( t ) w , Z t ( F ( w )) = F ′ ( w ) Y t ( w )
are C 1 C^1 C 1 and periodic in every C j ( D ) ‾ \overline{C^j(D)} C j ( D ) , with
sup t ( ∥ Z t ∥ C j + ∥ Z ˙ t ∥ C j ) ≤ C F , j ( 1 + M / a ) . \sup_t(\lVert Z_t\rVert_{C^j}+\lVert\dot{Z}_t\rVert_{C^j})\le C_{F,j}(1+M/a). t sup (∥ Z t ∥ C j + ∥ Z ˙ t ∥ C j ) ≤ C F , j ( 1 + M / a ) .
Their boundary integrand in (A.2) is positive away from the two zeros.
Proof. By (A.1), 0 ≠ ϕ t ∈ K A t 0 0\ne\phi_t\in K^0_{A_t} 0 = ϕ t ∈ K A t 0 ; hence proposition 5.3 gives two simple zeros. Joint C 1 C^1 C 1 regularity and compactness of the parameter torus imply a > 0 a>0 a > 0 . Uniform continuity of ∂ θ B \partial_\theta B ∂ θ B gives a common separation ρ > 0 \rho>0 ρ > 0 . The implicit-function branches glue uniquely by the derivative signs, including at t = 0 , T t=0,T t = 0 , T , and
ρ ≤ ℓ ≤ 2 π − ρ , θ − ( T ) − θ − ( 0 ) = θ + ( T ) − θ + ( 0 ) = 2 π k , k ∈ Z . \rho\le\ell\le2\pi-\rho,\qquad\theta_-(T)-\theta_-(0)=\theta_+(T)-\theta_+(0)=2\pi k,\qquad k\in\mathbb{Z}. ρ ≤ ℓ ≤ 2 π − ρ , θ − ( T ) − θ − ( 0 ) = θ + ( T ) − θ + ( 0 ) = 2 π k , k ∈ Z .
Differentiate θ − = θ + + ℓ \theta_-=\theta_++\ell θ − = θ + + ℓ and E = z + e i ℓ / 2 E=z_+e^{i\ell/2} E = z + e i ℓ /2 to obtain the formulas. Smoothness of F , F − 1 F,F^{-1} F , F − 1 gives the field bounds. Finally, with E = e i c E=e^{ic} E = e i c locally,
Y t ⋅ n = cos ( θ − c ) − cos β Y_t\cdot n=\cos(\theta-c)-\cos\beta Y t ⋅ n = cos ( θ − c ) − cos β
has the sign of B ( t , θ ) B(t,\theta) B ( t , θ ) .
The resulting positive period integral equals the moving-field term (A.3); periodicity does not force either to vanish. No existence of such a maximizing family is asserted.
The analytic reduction also applies without convexity. Its finite-dimensional kernel describes the remaining obstruction to finiteness.
Compactness, graph convergence, and shape analyticity in lemmas 6.1 and 6.2 require no convexity. For A ∈ M m A\in\mathcal{M}_m A ∈ M m , set
B = C 1 , α ( Γ ) , X = Y = B × R , F ^ ( h , η ) = F ( h , q A + η , m ) . B=C^{1,\alpha}(\Gamma),\qquad\mathcal{X}=\mathcal{Y}=B\times\mathbb{R},\qquad\widehat{\mathcal{F}}(h,\eta)=\mathcal{F}(h,q_A+\eta,m). B = C 1 , α ( Γ ) , X = Y = B × R , F ( h , η ) = F ( h , q A + η , m ) .
Then
L ^ ( ϕ , η ) = ( S Γ ϕ − g ϕ − η , ∫ Γ ϕ d s ) . \widehat{L}(\phi,\eta)=\left(S_\Gamma\phi-g\phi-\eta,\int_\Gamma\phi\,\mathrm{d}s\right). L ( ϕ , η ) = ( S Γ ϕ − g ϕ − η , ∫ Γ ϕ d s ) .
For L ^ 0 ( ϕ , η ) = ( − g ϕ − η , ∫ Γ ϕ d s ) \widehat{L}_0(\phi,\eta)=(-g\phi-\eta,\int_\Gamma\phi\,\mathrm{d}s) L 0 ( ϕ , η ) = ( − g ϕ − η , ∫ Γ ϕ d s ) ,
L ^ 0 − 1 ( f , b ) = ( ϕ , η ) , η = − b + ∫ Γ f / g d s ∫ Γ 1 / g d s , ϕ = − f + η g . \widehat{L}_0^{-1}(f,b)=(\phi,\eta),\qquad\eta=-\frac{b+\int_\Gamma f/g\,\mathrm{d}s}{\int_\Gamma1/g\,\mathrm{d}s},\qquad\phi=-\frac{f+\eta}{g}. L 0 − 1 ( f , b ) = ( ϕ , η ) , η = − ∫ Γ 1/ g d s b + ∫ Γ f / g d s , ϕ = − g f + η .
Since S Γ : B → B S_\Gamma:B\to B S Γ : B → B is compact, L ^ \widehat{L} L is Fredholm of index zero. Polarization and lemma 5.1 give
ker L ^ = { ( ϕ , c A ( ϕ ) ) : ϕ ∈ K A } , d : = dim ker L ^ = dim K A ≤ dim K A 0 + 1 ≤ 3. \ker\widehat{L}=\{(\phi,c_A(\phi)):\phi\in K_A\},\qquad d:=\dim\ker\widehat{L}=\dim K_A\le\dim K_A^0+1\le3. ker L = {( ϕ , c A ( ϕ )) : ϕ ∈ K A } , d := dim ker L = dim K A ≤ dim K A 0 + 1 ≤ 3.
Choose closed splittings
X = N ⊕ X 1 , Y = Ran L ^ ⊕ C , N = ker L ^ , dim C = d , \mathcal{X}=N\oplus X_1,\qquad\mathcal{Y}=\operatorname{Ran}\widehat{L}\oplus C,\qquad N=\ker\widehat{L},\qquad\dim C=d, X = N ⊕ X 1 , Y = Ran L ⊕ C , N = ker L , dim C = d ,
and let Π \Pi Π project onto Ran L ^ \operatorname{Ran}\widehat{L} Ran L . The analytic implicit-function theorem gives a local analytic map Ψ : N → X 1 \Psi:N\to X_1 Ψ : N → X 1 , Ψ ( 0 ) = D Ψ ( 0 ) = 0 \Psi(0)=D\Psi(0)=0 Ψ ( 0 ) = D Ψ ( 0 ) = 0 , such that
R r e d ( k ) = ( I − Π ) F ^ ( k + Ψ ( k ) ) , R_{\mathrm{red}}(k)=(I-\Pi)\widehat{\mathcal{F}}(k+\Psi(k)), R red ( k ) = ( I − Π ) F ( k + Ψ ( k )) ,
F ^ ( k + x ) = 0 ⟺ x = Ψ ( k ) , R r e d ( k ) = 0. \widehat{\mathcal{F}}(k+x)=0\iff x=\Psi(k),\qquad R_{\mathrm{red}}(k)=0. F ( k + x ) = 0 ⟺ x = Ψ ( k ) , R red ( k ) = 0.
Thus the stationary set is locally the analytic zero set of R r e d : R d → R d R_{\mathrm{red}}:\mathbb{R}^{d}\to\mathbb{R}^{d} R red : R d → R d , d ≤ 3 d\leq3 d ≤ 3 ; it need not be a manifold.
Proposition A.3. Let D D D be smooth, bounded, and simply connected. If K A 0 = { 0 } K^{0}_{A}=\{0\} K A 0 = { 0 } for every A ∈ M m A\in\mathcal{M}_{m} A ∈ M m at a fixed mass, then # M m < ∞ \#\mathcal{M}_{m}<\infty # M m < ∞ . If this exclusion holds at every maximizing pair with mass in a compact interval I ⊂ ( 0 , ∣ D ∣ ) I\subset(0,|D|) I ⊂ ( 0 , ∣ D ∣ ) , then sup m ∈ I # M m < ∞ \sup_{m\in I}\#\mathcal{M}_{m}<\infty sup m ∈ I # M m < ∞ . If it holds at every admissible mass and the Robin function has finitely many nondegenerate global minima, then sup 0 < m < ∣ D ∣ # M m < ∞ \sup_{0<m<|D|}\#\mathcal{M}_{m}<\infty sup 0 < m < ∣ D ∣ # M m < ∞ .
Proof. The condition K A 0 = 0 K^{0}_{A}=0 K A 0 = 0 makes the shape-only derivative injective, with index − 1 -1 − 1 , so proposition 6.3 applies. Were M m \mathcal{M}_{m} M m infinite, its compact derived set M m ′ \mathcal{M}'_{m} M m ′ would contain a threshold maximum A 0 A_{0} A 0 . Distinct maximizers approaching A 0 A_{0} A 0 give distinct zeros q j → q 0 q_{j}\to q_{0} q j → q 0 of the scalar residual. Therefore
r ( ⋅ , m ) ≡ 0 near q 0 ⟹ A H ( t , m ) ∈ M m ′ , q A H ( t , m ) = t r(\mathord{\cdot},m)\equiv0\text{ near }q_{0}\quad\Longrightarrow\quad A_{H(t,m)}\in\mathcal{M}'_{m},\quad q_{A_{H(t,m)}}=t r ( ⋅ , m ) ≡ 0 near q 0 ⟹ A H ( t , m ) ∈ M m ′ , q A H ( t , m ) = t
by lemma 6.4. Taking t > q 0 t>q_{0} t > q 0 is a contradiction. This proves fixed-mass finiteness as in proposition 6.5.
The residual consequently has finite vanishing order at every maximizing pair. Weierstrass preparation and joint compactness give the bound on I I I by the proof of corollary 6.6. Near full mass, proposition 7.2 gives uniqueness without convexity. If the global Robin minima are p 1 , … , p J p_{1},\ldots,p_{J} p 1 , … , p J , Turkington’s concentration theorem and single-core conclusion [27 Corollary 4.6] , applied as in proposition 7.3, place every small-mass maximizer in one of their single-patch regimes. Nondegeneracy and local uniqueness [7 ] give # M m ≤ J \#\mathcal{M}_{m}\leq J # M m ≤ J for all sufficiently small m m m . These endpoint bounds and an interior interval prove the last claim. □
The same local argument shows
# M m = ∞ , A 0 ∈ arg max A ∈ M m ′ q A ⟹ K A 0 0 ≠ 0. \#\mathcal{M}_{m}=\infty,\quad A_{0}\in\underset{A\in\mathcal{M}'_{m}}{\operatorname{arg\,max}}q_{A}\quad\Longrightarrow\quad K^{0}_{A_{0}}\neq0. # M m = ∞ , A 0 ∈ A ∈ M m ′ arg max q A ⟹ K A 0 0 = 0.
For a C 1 C^{1} C 1 fixed-mass stationary arc through a maximizer,
d d t E ( 1 A t ) = q ( t ) d d t ∣ A t ∣ = 0 , \frac{d}{dt}\mathcal{E}(1_{A_{t}})=q(t)\frac{d}{dt}|A_{t}|=0, d t d E ( 1 A t ) = q ( t ) d t d ∣ A t ∣ = 0 ,
so the entire connected arc consists of maximizers. Its velocity satisfies
φ t ∈ K A t , q ˙ ( t ) = 0 , φ t ≠ 0 ⟹ 0 ≠ φ t ∈ K A t 0 . \varphi_{t}\in K_{A_{t}},\quad\dot{q}(t)=0,\quad\varphi_{t}\neq0\quad\Longrightarrow\quad0\neq\varphi_{t}\in K^{0}_{A_{t}}. φ t ∈ K A t , q ˙ ( t ) = 0 , φ t = 0 ⟹ 0 = φ t ∈ K A t 0 .
For an immersed maximizing manifold N \mathcal{N} N and a critical point A A A of q ∣ N q|_{\mathcal{N}} q ∣ N , normal velocities give an injection
T A N ↪ K A 0 , dim N ≤ 2. T_{A}\mathcal{N}\hookrightarrow K^{0}_{A},\quad\dim\mathcal{N}\leq2. T A N ↪ K A 0 , dim N ≤ 2.
These are conditional and first-order statements; unconditional finiteness for general simply connected domains is not established.
Statements and Declarations
Funding. D. Cao was supported by National Key R&D Program (Grant 2023YFA1010001) and NNSF of China (Grant 12371212). G. Wang was supported by NNSF of China (Grant 12471101) and Fundamental Research Funds for the Central Universities (Grant DUT26YC106). Y. Wang was supported by NNSF of China (Grant 12471110).
Data availability. No datasets were generated or analysed during this study.
Use of AI-assisted tools. During preparation of this manuscript, ChatGPT 6.0 (OpenAI) was used for assistance with structural and language revision, LaTeX editing, and checking parts of the mathematical exposition. The authors are responsible for verifying all arguments and references and for the final content of the manuscript.
References
State Key Laboratory of Mathematical Sciences, Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190, P.R. China, and University of Chinese Academy of Sciences, Beijing 100049, P.R. China
Email address: dmcao@amt.ac.cn
School of Mathematical Sciences, Dalian University of Technology, Dalian 116024, P.R. China
Email address: gdw@dlut.edu.cn
School of Mathematics, Tianjin Normal University, Tianjin, China
Email address: wangyuchen@mail.nankai.edu.cn