Introduction
The dynamics of water waves constitute a central subject in the mathematical theory of fluid mechanics. In the irrotational setting, substantial progress has been made in understanding the existence, regularity, stability, and long-time dynamics of gravity and capillary-gravity waves, see for example [21 ] . By contrast, the corresponding questions for rotational flows remain considerably more challenging. The presence of vorticity introduces an additional transport mechanism in the fluid bulk and leads to a nontrivial coupling between the interior fluid motion and the evolution of the free surface, we refer to interesting readers to the monograph [20 ] .
Consider the two-dimensional incompressible Euler equations with gravity in a fluid domain of finite depth. The unknown fluid domain is given by
Ω ( t ) = { ( x 1 , x 2 ) ∈ R 2 : − 1 < x 2 < 1 + ζ ( t , x 1 ) } , \begin{equation}
\Omega(t) = \left\{
(x_1,x_2)\in\mathbb R^2:
-1<x_2<1+\zeta(t,x_1)
\right\}, \tag*{(1.1)}
\end{equation} Ω ( t ) = { ( x 1 , x 2 ) ∈ R 2 : − 1 < x 2 < 1 + ζ ( t , x 1 ) } , (1.1)
where ζ ( t , x 1 ) \zeta(t,x_1) ζ ( t , x 1 ) denotes the displacement of the free surface from the rest position x 2 = 1 x_2=1 x 2 = 1 . Suppose that ζ ( t , ⋅ ) ∈ H s ( R ) \zeta(t,\cdot)\in H^s(\mathbb R) ζ ( t , ⋅ ) ∈ H s ( R ) , s > 3 / 2 s>3/2 s > 3/2 , and that
∥ ζ ( t , ⋅ ) ∥ L ∞ ( R ) ≤ 1 2 . \|\zeta(t,\cdot)\|_{L^\infty(\mathbb R)}
\leq \frac12. ∥ ζ ( t , ⋅ ) ∥ L ∞ ( R ) ≤ 2 1 .
In particular, the free surface remains well-separated from the flat bottom
B = { ( x 1 , x 2 ) : x 2 = − 1 } . \mathbf B=\{(x_1,x_2):x_2=-1\}. B = {( x 1 , x 2 ) : x 2 = − 1 } .
The free boundary will be denoted by
S ( t ) = { ( x 1 , 1 + ζ ( t , x 1 ) ) : x 1 ∈ R } . S(t)=\{(x_1,1+\zeta(t,x_1)):x_1\in\mathbb R\}. S ( t ) = {( x 1 , 1 + ζ ( t , x 1 )) : x 1 ∈ R } .
Let v = ( v 1 , v 2 ) v=(v_1,v_2) v = ( v 1 , v 2 ) be the fluid velocity and p p p the pressure. The governing equations are
\begin{equation}
\begin{cases}
\partial_t v+(v\cdot\nabla)v+\nabla p
=-g\vec e_2,
& x\in\Omega(t),\\
\nabla\cdot v=0,
& x\in\Omega(t),
\end{cases}
\end{equation}
together with the kinematic boundary condition
\begin{equation}
\partial_t\zeta
=
v_2-v_1\partial_{x_1}\zeta
\qquad\text{on }S(t),
\end{equation}
and the impermeability condition
\begin{equation}
v_2=0
\qquad\text{on }\mathbf B.
\end{equation}
On the free surface, the Young--Laplace law gives
\begin{equation}
p=\alpha^2\kappa,
\qquad
\kappa
=
-\frac{\partial_{x_1}^2\zeta}
{\bigl(1+(\partial_{x_1}\zeta)^2\bigr)^{3/2}},
\end{equation}
where $g>0$ is the gravitational acceleration and
$\alpha^2>0$ is the surface-tension coefficient.
The pressure in the region above the free surface has been normalized to zero.
The local well-posedness theory for free-boundary incompressible Euler equations, including flows with nontrivial vorticity, has been developed in a number of works; see, for example, [6 , 5 , 7 , 23 , 8 , 9 ] and the references therein. Nevertheless, the construction and qualitative analysis of coherent structures in rotational water-wave dynamics remain relatively less understood. One of the principal difficulties is that the free surface is affected not only by the velocity field but also, nonlocally, by the nonlinear interactions of vorticity within the fluid. We also want to point out that the rotational water-wave equations admit a noncanonical Hamiltonian formulation [24 ] with a generalized Dirichlet–Neumann operator. The corresponding energy is given as:
E ( v , ζ ) = 1 2 ∫ Ω ( t ) ∣ v ∣ 2 d x + g 2 ∫ R ζ 2 d x 1 + α 2 ∫ R ( 1 + ( ∂ x 1 ζ ) 2 − 1 ) d x 1 . \begin{equation}
\begin{aligned}
E(v,\zeta)
=\frac12\int_{\Omega(t)}|v|^2\,dx
+\frac g2\int_{\mathbb R}\zeta^2\,dx_1
+\alpha^2\int_{\mathbb R}
\left(
\sqrt{1+(\partial_{x_1}\zeta)^2}-1
\right)\,dx_1.
\end{aligned} \tag*{(1.2)}
\end{equation} E ( v , ζ ) = 2 1 ∫ Ω ( t ) ∣ v ∣ 2 d x + 2 g ∫ R ζ 2 d x 1 + α 2 ∫ R ( 1 + ( ∂ x 1 ζ ) 2 − 1 ) d x 1 . (1.2)
This structure provides a natural framework for studying finite-energy solutions, for which the kinetic energy is finite and the free-surface perturbation decays at spatial infinity.
A particularly interesting problem is to understand how localized vorticity structures, which have been extensively studied in the incompressible Euler equations in fixed domains [26 , 25 , 27 ] , can coexist with a free surface, while retaining a finite total energy. In a finite-depth fluid, concentrated vorticity interacts with both the rigid bottom and the water-air interface, thereby producing a delicate balance between nonlinear interior dynamics and boundary deformation. It is therefore natural to investigate whether several strongly concentrated vorticity structures can be arranged into a stationary configuration. Such solutions need not be small in velocity or vorticity while the resulting deformation of the free surface remains small.
A fundamental class of coherent structures consists of traveling waves:
v ( t , x 1 , x 2 ) = v ( x 1 − c t , x 2 ) , ζ ( t , x 1 ) = ζ ( x 1 − c t ) , \begin{equation}
v(t,x_1,x_2)
=v(x_1-\mathbf ct,x_2),
\qquad
\zeta(t,x_1)
=
\zeta(x_1-\mathbf ct), \tag*{(1.3)}
\end{equation} v ( t , x 1 , x 2 ) = v ( x 1 − c t , x 2 ) , ζ ( t , x 1 ) = ζ ( x 1 − c t ) , (1.3)
where c ∈ R \mathbf c\in\mathbb R c ∈ R is the propagation speed of the wave and the free-surface displacement tends to zero as ∣ x 1 ∣ → ∞ |x_1|\to\infty ∣ x 1 ∣ → ∞ .
In this paper, we restrict attention to stationary solutions, corresponding to c = 0 \mathbf c=0 c = 0 . We seek localized configurations satisfying
v ∈ L 2 ( Ω ) , ζ ∈ H 2 ( R ) , v\in L^2(\Omega),
\qquad
\zeta\in H^2(\mathbb R), v ∈ L 2 ( Ω ) , ζ ∈ H 2 ( R ) ,
where
Ω = { ( x 1 , x 2 ) : − 1 < x 2 < 1 + ζ ( x 1 ) } . \Omega
=
\{(x_1,x_2):-1<x_2<1+\zeta(x_1)\}. Ω = {( x 1 , x 2 ) : − 1 < x 2 < 1 + ζ ( x 1 )} .
For two-dimensional incompressible flows, the scalar vorticity is defined by
ω = ∂ x 1 v 2 − ∂ x 2 v 1 . \omega
=
\partial_{x_1}v_2-\partial_{x_2}v_1. ω = ∂ x 1 v 2 − ∂ x 2 v 1 .
Taking the curl of the Euler equations gives the vorticity transport equation
∂ t ω + v ⋅ ∇ ω = 0. \begin{equation}
\partial_t\omega+v\cdot\nabla\omega=0. \tag*{(1.4)}
\end{equation} ∂ t ω + v ⋅ ∇ ω = 0. (1.4)
Consequently, a stationary solution satisfies
{ v ⋅ ∇ ω = 0 , x ∈ Ω , v 2 − v 1 ∂ x 1 ζ = 0 , x ∈ S . \begin{equation}
\begin{cases}
v\cdot\nabla\omega=0,
& x\in\Omega,\\
v_2-v_1\partial_{x_1}\zeta=0,
& x\in S.
\end{cases} \tag*{(1.5)}
\end{equation} { v ⋅ ∇ ω = 0 , v 2 − v 1 ∂ x 1 ζ = 0 , x ∈ Ω , x ∈ S . (1.5)
Since the fluid is incompressible, we introduce a stream function Ψ \Psi Ψ through
v = ∇ ⊥ Ψ = ( − ∂ x 2 Ψ , ∂ x 1 Ψ ) . v=\nabla^\perp\Psi
=
(-\partial_{x_2}\Psi,\partial_{x_1}\Psi). v = ∇ ⊥ Ψ = ( − ∂ x 2 Ψ , ∂ x 1 Ψ ) .
Then
ω = Δ Ψ , \omega=\Delta\Psi, ω = ΔΨ ,
and the stationary vorticity equation becomes
∇ ⊥ Ψ ⋅ ∇ Δ Ψ = 0 in Ω . \begin{equation}
\nabla^\perp\Psi\cdot\nabla\Delta\Psi=0
\qquad\text{in }\Omega. \tag*{(1.6)}
\end{equation} ∇ ⊥ Ψ ⋅ ∇ΔΨ = 0 in Ω. (1.6)
The kinematic boundary condition implies that Ψ \Psi Ψ is constant along the free surface. Likewise, the impermeability condition implies that Ψ \Psi Ψ is constant along the rigid bottom. For localized stationary flows with finite kinetic energy, the horizontal volume flux vanishes, and hence these two boundary constants agree. After adding a constant to the stream function, we may therefore impose
Ψ = 0 on ∂ Ω . \begin{equation}
\Psi=0
\qquad\text{on }\partial\Omega. \tag*{(1.7)}
\end{equation} Ψ = 0 on ∂ Ω. (1.7)
To obtain a formulation of elliptic equations, it is sufficient to prescribe the vorticity as a nonlinear function of the stream function. More precisely, we consider
ω = 1 δ 2 γ ( Ψ ) , \begin{equation}
\omega
=
\frac1{\delta^2}\gamma(\Psi), \tag*{(1.8)}
\end{equation} ω = δ 2 1 γ ( Ψ ) , (1.8)
where 0 < δ ≪ 1 0<\delta\ll1 0 < δ ≪ 1 is a concentration parameter and γ \gamma γ is a given smooth nonlinearity.
Since
∇ ⊥ Ψ ⋅ ∇ γ ( Ψ ) = 0 , \nabla^\perp\Psi\cdot\nabla\gamma(\Psi)=0, ∇ ⊥ Ψ ⋅ ∇ γ ( Ψ ) = 0 ,
the stationary vorticity equation is automatically satisfied whenever Ψ \Psi Ψ solves
{ Δ Ψ = 1 δ 2 γ ( Ψ ) , x ∈ Ω , Ψ = 0 , x ∈ ∂ Ω . \begin{equation}
\begin{cases}
\displaystyle
\Delta\Psi=\frac1{\delta^2}\gamma(\Psi),
& x\in\Omega,\\[2mm]
\Psi=0,
& x\in\partial\Omega.
\end{cases} \tag*{(1.9)}
\end{equation} ⎩ ⎨ ⎧ ΔΨ = δ 2 1 γ ( Ψ ) , Ψ = 0 , x ∈ Ω , x ∈ ∂ Ω. (1.9)
The remaining condition is furnished by the Bernoulli equation on the free surface. Let
Γ ( t ) = ∫ 0 t γ ( s ) d s . \Gamma(t)=\int_0^t\gamma(s)\,ds. Γ ( t ) = ∫ 0 t γ ( s ) d s .
For the stationary Euler flow satisfying (1.8) , the momentum equation implies
∇ ( 1 2 ∣ ∇ Ψ ∣ 2 + p + g x 2 − 1 δ 2 Γ ( Ψ ) ) = 0. \nabla\left(
\frac12|\nabla\Psi|^2
+p+gx_2
-\frac1{\delta^2}\Gamma(\Psi)
\right)=0. ∇ ( 2 1 ∣∇Ψ ∣ 2 + p + g x 2 − δ 2 1 Γ ( Ψ ) ) = 0.
Assuming that the velocity and the surface displacement vanish at spatial infinity, the constant is equal to g g g . Since Ψ = 0 \Psi=0 Ψ = 0 on S S S and Γ ( 0 ) = 0 \Gamma(0)=0 Γ ( 0 ) = 0 , the Young–Laplace condition yields
1 2 ∣ ∇ Ψ ∣ 2 + g ζ + α 2 κ = 0 on S . \begin{equation}
\frac12|\nabla\Psi|^2
+g\zeta+\alpha^2\kappa=0
\qquad\text{on }S. \tag*{(1.10)}
\end{equation} 2 1 ∣∇Ψ ∣ 2 + g ζ + α 2 κ = 0 on S . (1.10)
Thus, the construction of stationary rotational capillary-gravity waves is reduced to the coupled elliptic free-boundary system
{ Δ Ψ = 1 δ 2 γ ( Ψ ) , x ∈ Ω , Ψ = 0 , x ∈ ∂ Ω , 1 2 ∣ ∇ Ψ ∣ 2 + g ζ + α 2 κ = 0 , x ∈ S , \begin{equation}
\begin{cases}
\displaystyle
\Delta\Psi=\frac1{\delta^2}\gamma(\Psi),
& x\in\Omega,\\[2mm]
\Psi=0,
& x\in\partial\Omega,\\[2mm]
\displaystyle
\frac12|\nabla\Psi|^2
+g\zeta+\alpha^2\kappa=0,
& x\in S,
\end{cases} \tag*{(1.11)}
\end{equation} ⎩ ⎨ ⎧ ΔΨ = δ 2 1 γ ( Ψ ) , Ψ = 0 , 2 1 ∣∇Ψ ∣ 2 + g ζ + α 2 κ = 0 , x ∈ Ω , x ∈ ∂ Ω , x ∈ S , (1.11)
where the fluid domain Ω \Omega Ω is itself unknown.
The main difficulty is that the geometry of the domain depends on the solution through the nonlinear Bernoulli condition. In particular, the vorticity concentrations produce a velocity field whose trace on the free surface determines the resulting boundary deformation.
The existence and qualitative properties of steady water waves carrying localized vorticity have attracted considerable attention in recent years. An important contribution was made by Shatah, Walsh, and Zeng [22 ] , who established the existence of traveling capillary-gravity water waves with compactly supported vorticity, including configurations involving submerged point vortices and vortex patches. In finite depth, Varholm [13 ] constructed solitary gravity-capillary waves carrying finitely many point vortices arranged along a vertical line.The existence, nonexistence, and far-field asymptotics of deep-water solitary waves with loca lized vorticity were subsequently investigated by Chen, Walsh, and Wheeler [14 ] .
Further developments include the construction and instability analysis of solitary waves carrying a finite vortex dipole [15 ] , the conditional orbital stability of solitary waves with a submerged point vortex [16 ] , and the long-time dynamics of water waves interacting with point vortices [18 ] . More recently, Chen, Varholm, Walsh, and Wheeler [17 ] established the existence of large-amplitude solitary gravity waves carrying a submerged point vortex, without requiring surface tension.
A particularly relevant result for the present work was obtained by Ehrnstrom, Walsh, and Zeng [1 ] . They constructed smooth, finite-energy stationary capillary-gravity waves with exponentially localized vorticity. Their approach revealed a connection between stationary rotational water waves and singularly perturbed semilinear elliptic equations. In particular, the construction is based on a localized ground-state profile, allowing the kinetic energy to remain nonvanishing even when the free-surface displacement is exponentially small. Motivated by this development, we investigate stationary water waves possessing multiple localized stream-function spikes with alternating signs.
More precisely, we are interested in solutions for which the vorticity is concentrated near finitely many points in the fluid interior. The natural building block is a positive, radially symmetric, spatially localized solution of the entire-space semilinear elliptic equation
Δ U = γ ( U ) in R 2 . \begin{equation}
\Delta U=\gamma(U)
\qquad\text{in }\mathbb R^2. \tag*{(1.12)}
\end{equation} Δ U = γ ( U ) in R 2 . (1.12)
The concentration parameter δ \delta δ determines the characteristic size of each localized core. Indeed, if x 0 ∈ Ω x_0\in\Omega x 0 ∈ Ω , then
U ( x − x 0 δ ) U\left(\frac{x-x_0}{\delta}\right) U ( δ x − x 0 )
solves the rescaled equation
Δ U δ = 1 δ 2 γ ( U δ ) \Delta U_\delta
=
\frac1{\delta^2}\gamma(U_\delta) Δ U δ = δ 2 1 γ ( U δ )
in the whole plane.
We impose the following assumptions on the nonlinearity and its ground state.
There exists an integer k 0 ≥ 3 k_0\geq3 k 0 ≥ 3 such that γ ∈ C k 0 ( R ) \gamma\in C^{k_0}(\mathbb R) γ ∈ C k 0 ( R ) and
γ ( 0 ) = 0 , γ ′ ( 0 ) = 1 , γ ′ ′ ( t ) < 0 for t > 0. \gamma(0)=0,\qquad
\gamma'(0)=1,\qquad
\gamma''(t)<0\quad\text{for }t>0. γ ( 0 ) = 0 , γ ′ ( 0 ) = 1 , γ ′′ ( t ) < 0 for t > 0. Moreover, γ \gamma γ is odd:
γ ( − t ) = − γ ( t ) . \gamma(-t)=-\gamma(t). γ ( − t ) = − γ ( t ) . Equation (1.12) admits a nontrivial positive radial solution U ∈ C k 0 + 2 ( R 2 ) U\in C^{k_0+2}(\mathbb R^2) U ∈ C k 0 + 2 ( R 2 ) satisfying
U ( x ) , ∇ U ( x ) ⟶ 0 as ∣ x ∣ → ∞ . U(x),\nabla U(x)\longrightarrow0
\qquad\text{as }|x|\to\infty. U ( x ) , ∇ U ( x ) ⟶ 0 as ∣ x ∣ → ∞. The ground state U U U is nondegenerate modulo translations, in the sense that
ker ( − Δ + γ ′ ( U ) ) = span { ∂ x 1 U , ∂ x 2 U } \ker\left(
-\Delta+\gamma'(U)
\right)
=
\operatorname{span}
\left\{
\partial_{x_1}U,\partial_{x_2}U
\right\} ker ( − Δ + γ ′ ( U ) ) = span { ∂ x 1 U , ∂ x 2 U } as an operator from H 2 ( R 2 ) H^2(\mathbb R^2) H 2 ( R 2 ) into L 2 ( R 2 ) L^2(\mathbb R^2) L 2 ( R 2 ) .
A representative example is
γ ( t ) = t − t 3 , \gamma(t)=t-t^3, γ ( t ) = t − t 3 ,
for which (1.12) becomes
− Δ U + U = U 3 . -\Delta U+U=U^3. − Δ U + U = U 3 .
More generally, nonlinearities of the form
γ ( t ) = t − ∣ t ∣ p t \gamma(t)=t-|t|^pt γ ( t ) = t − ∣ t ∣ p t
may be considered whenever the exponent and the corresponding ground state satisfy the regularity and nondegeneracy assumptions above.
Under these assumptions, the ground state and its derivatives of all orders decay exponentially at infinity. The nondegeneracy property is essential for the perturbative construction, since the two translational modes are the only elements of the kernel of the linearized operator.
The oddness of γ \gamma γ is particularly important in the present problem. Indeed, if U U U solves (1.12) , then − U -U − U is also a solution, and
γ ′ ( − U ) = γ ′ ( U ) . \gamma'(-U)=\gamma'(U). γ ′ ( − U ) = γ ′ ( U ) .
Thus positive and negative stream-function spikes have identical linearized operators. This permits the construction of alternating-sign configurations using a common ground-state profile.
The purpose of this paper is to construct a family of finite-energy stationary capillary-gravity water waves containing an arbitrary prescribed number of localized vorticity concentrations. More precisely, we establish the existence of solutions whose stream functions exhibit multiple vertically aligned spikes with alternating signs. The vorticity becomes increasingly concentrated as a small parameter tends to zero, whereas the free-surface deformation is exponentially small. The construction reveals an interaction mechanism that involves neighboring vorticity concentrations and the two boundaries of the fluid domain.
To this end, we shall construct solutions whose leading-order stream functions take the form
Ψ ( x ) = ∑ j = 1 n ( − 1 ) j + 1 U ( x − x j δ ) + lower-order terms , \begin{equation}
\Psi(x)
=
\sum_{j=1}^n
(-1)^{j+1} U\left(\frac{x-x_j}{\delta}\right)
+\text{lower-order terms}, \tag*{(1.13)}
\end{equation} Ψ ( x ) = j = 1 ∑ n ( − 1 ) j + 1 U ( δ x − x j ) + lower-order terms , (1.13)
where x j x_j x j are the concentration points to be determined.
Correspondingly, the vorticity has the leading structure
ω ( x ) ≈ 1 δ 2 ∑ j = 1 n ( − 1 ) j + 1 Δ U ( x − x j δ ) . \omega(x)
\approx
\frac1{\delta^2}
\sum_{j=1}^n
(-1)^{j+1}
\Delta U\left(\frac{x-x_j}{\delta}\right). ω ( x ) ≈ δ 2 1 j = 1 ∑ n ( − 1 ) j + 1 Δ U ( δ x − x j ) .
The localization of the ground state implies that the different cores interact only exponentially weakly when their mutual distances remain bounded away from zero. These exponentially small interactions play a decisive role in determining the admissible positions of the spikes.
Our main theorem establishes the existence of stationary water waves with an arbitrary finite number of vertically aligned, alternating-sign stream-function spikes.
Theorem 1.1. Assume that γ \gamma γ satisfies (A) and (B). Let g > 0 g>0 g > 0 , α > 0 \alpha>0 α > 0 , and let n ≥ 1 n\geq1 n ≥ 1 be an integer.
There exists δ 0 > 0 \delta_0>0 δ 0 > 0 such that, for every δ ∈ ( 0 , δ 0 ) \delta\in(0,\delta_0) δ ∈ ( 0 , δ 0 ) , the stationary capillary-gravity water-wave problem (1.11) admits a solution ( Ψ δ , ζ δ ) ∈ ( H 2 ( Ω ) ∩ H 0 1 ( Ω ) ) × H 2 ( R ) (\Psi_\delta,\zeta_\delta)
\in
\bigl(
H^2(\Omega)
\cap H_0^1(\Omega)
\bigr)
\times H^2(\mathbb R) ( Ψ δ , ζ δ ) ∈ ( H 2 ( Ω ) ∩ H 0 1 ( Ω ) ) × H 2 ( R ) possessing exactly n n n localized stream-function spikes.
More precisely, there exist concentration heights τ = ( τ 1 , … , τ n ) ∈ R n \boldsymbol{\tau}
=
(\tau_1,\ldots,\tau_n)\in\mathbb R^n τ = ( τ 1 , … , τ n ) ∈ R n satisfying ∣ τ j − n − 2 j + 1 n ∣ ≤ C δ ∣ log δ ∣ , j = 1 , … , n ,
\left|
\tau_j-\frac{n-2j+1}{n}
\right|
\leq
C\delta|\log\delta|,
\qquad j=1,\ldots,n,
τ j − n n − 2 j + 1 ≤ C δ ∣ log δ ∣ , j = 1 , … , n , such that Ψ δ ( x ) = ∑ j = 1 n σ j [ U j ( x δ ) − ( U j ) b c ( x δ ) ] + v ~ ( x δ ) ,
\Psi_\delta(x)
=
\sum_{j=1}^n
\sigma_j
\left[
U_j\left(\frac{x}{\delta}\right)
-
(U_j)_{\mathrm{bc}}
\left(\frac{x}{\delta}\right)
\right]
+
\widetilde v\left(\frac{x}{\delta}\right),
Ψ δ ( x ) = ∑ j = 1 n σ j [ U j ( δ x ) − ( U j ) bc ( δ x ) ] + v ( δ x ) , where U j ( y ) = U ( y − τ j δ e ⃗ 2 ) , σ j = ( − 1 ) j + 1 . U_j(y)
=
U\left(
y-\frac{\tau_j}{\delta}\vec e_2
\right),
\qquad
\sigma_j=(-1)^{j+1}. U j ( y ) = U ( y − δ τ j e 2 ) , σ j = ( − 1 ) j + 1 . Here ( U j ) b c (U_j)_{\mathrm{bc}} ( U j ) bc denotes the corresponding boundary correction, as defined in § 2.2 .
Furthermore, the correction term and the free-surface displacement satisfy ∥ v ~ ( ⋅ / δ ) ∥ H 2 ( Ω ) ≤ C δ β exp ( − 2 n δ )
\left\|
\widetilde v(\cdot/\delta)
\right\|_{H^2(\Omega)}
\leq
C\delta^{\beta}
\exp\left(-\frac{2}{n\delta}\right)
∥ v ( ⋅ / δ ) ∥ H 2 ( Ω ) ≤ C δ β exp ( − n δ 2 ) and ∥ ζ δ ∥ H 2 ( R ) ≤ C δ β ∣ log δ ∣ exp ( − 2 n δ ) .
\|\zeta_\delta\|_{H^2(\mathbb R)}
\leq
C\delta^{\beta}|\log\delta|
\exp\left(-\frac{2}{n\delta}\right).
∥ ζ δ ∥ H 2 ( R ) ≤ C δ β ∣ log δ ∣ exp ( − n δ 2 ) . The constants C C C and β \beta β are independent of sufficiently small δ \delta δ .
The theorem extends the single-spike construction of [1 ] to configurations containing an arbitrary finite number of alternating-sign spikes. In contrast to a single concentration, the multi-spike construction requires a detailed analysis of the interactions among neighboring cores and between the cores and the fluid boundaries.
An important feature of the result is the asymptotic distribution of the concentration points. By (1.14) ,
τ j ⟶ n − 2 j + 1 n as δ → 0. \tau_j
\longrightarrow
\frac{n-2j+1}{n}
\qquad\text{as }\delta\to0. τ j ⟶ n n − 2 j + 1 as δ → 0.
Hence the limiting concentration points are equally spaced along the vertical segment joining the bottom and the undisturbed free surface. In particular,
τ j − τ j + 1 = 2 n + O ( δ ∣ log δ ∣ ) , j = 1 , … , n − 1. \tau_j-\tau_{j+1}
=
\frac2n+O(\delta|\log\delta|),
\qquad j=1,\ldots,n-1. τ j − τ j + 1 = n 2 + O ( δ ∣ log δ ∣ ) , j = 1 , … , n − 1.
The distance between each outermost concentration point and the corresponding boundary tends to 1 / n 1/n 1/ n . Thus the nearest-neighbor separation is twice the limiting distance from an outermost spike to the adjacent boundary.
This geometric arrangement reflects the balance between the exponentially weak interactions of neighboring spikes and the boundary effects. Although these interactions disappear in the leading-order limit, they determine the finite-dimensional conditions governing the positions of the concentrations.
We emphasize several consequences of the main theorem. First, the constructed solutions have finite, nonvanishing kinetic energy in the singular limit δ → 0 \delta\to0 δ → 0 . Indeed, the spikes are well separated, and their mutual interactions are exponentially small. The scaling invariance of the Dirichlet integral in two dimensions yields
∫ Ω δ ∣ ∇ Ψ δ ∣ 2 d x = ∑ j = 1 n ∫ R 2 ∣ ∇ U ∣ 2 d y + o ( 1 ) = n ∥ ∇ U ∥ L 2 ( R 2 ) 2 + o ( 1 ) . \begin{equation}
\begin{aligned}
\int_{\Omega_\delta}
|\nabla\Psi_\delta|^2\,dx
&=
\sum_{j=1}^n
\int_{\mathbb R^2}|\nabla U|^2\,dy
+o(1)\\
&=
n\|\nabla U\|_{L^2(\mathbb R^2)}^2
+o(1).
\end{aligned} \tag*{(1.18)}
\end{equation} ∫ Ω δ ∣∇ Ψ δ ∣ 2 d x = j = 1 ∑ n ∫ R 2 ∣∇ U ∣ 2 d y + o ( 1 ) = n ∥∇ U ∥ L 2 ( R 2 ) 2 + o ( 1 ) . (1.18)
Consequently, the kinetic energy follows
1 2 ∫ Ω δ ∣ v δ ∣ 2 d x ⟶ n 2 ∫ R 2 ∣ ∇ U ∣ 2 d y . \frac12
\int_{\Omega_\delta}|v_\delta|^2\,dx
\longrightarrow
\frac n2
\int_{\mathbb R^2}|\nabla U|^2\,dy. 2 1 ∫ Ω δ ∣ v δ ∣ 2 d x ⟶ 2 n ∫ R 2 ∣∇ U ∣ 2 d y .
Thus, although the fluid velocity becomes strongly localized, its total kinetic energy remains of order one.
Second, the vorticity becomes increasingly concentrated. Since
ω δ = 1 δ 2 γ ( Ψ δ ) , \omega_\delta = \frac1{\delta^2}\gamma(\Psi_\delta), ω δ = δ 2 1 γ ( Ψ δ ) ,
its magnitude satisfies
∥ ω δ ∥ L ∞ ( Ω δ ) ≃ δ − 2 . \|\omega_\delta\|_{L^\infty(\Omega_\delta)}
\simeq \delta^{-2}. ∥ ω δ ∥ L ∞ ( Ω δ ) ≃ δ − 2 .
Moreover, the localization and separation of the spikes imply
∥ ω δ ∥ L 1 ( Ω δ ) = n ∥ Δ U ∥ L 1 ( R 2 ) + o ( 1 ) . \begin{equation}
\|\omega_\delta\|_{L^1(\Omega_\delta)}
=
n\|\Delta U\|_{L^1(\mathbb R^2)}
+o(1). \tag*{(1.19)}
\end{equation} ∥ ω δ ∥ L 1 ( Ω δ ) = n ∥Δ U ∥ L 1 ( R 2 ) + o ( 1 ) . (1.19)
On the other hand, integrating (1.12) over the whole plane gives
∫ R 2 Δ U d x = 0. \int_{\mathbb R^2}\Delta U\,dx=0. ∫ R 2 Δ U d x = 0.
Therefore each isolated ground-state core has zero net circulation, although its absolute vorticity is nonzero.
The total circulation of the constructed configuration is consequently asymptotically small:
∫ Ω δ ω δ d x = ∫ ∂ Ω δ ∂ N Ψ δ d S = o ( 1 ) . \begin{equation}
\int_{\Omega_\delta}\omega_\delta\,dx
=
\int_{\partial\Omega_\delta}
\partial_N\Psi_\delta\,dS
=o(1). \tag*{(1.20)}
\end{equation} ∫ Ω δ ω δ d x = ∫ ∂ Ω δ ∂ N Ψ δ d S = o ( 1 ) . (1.20)
In fact, the exponential localization of the spikes and their separation from the boundary yield an exponentially small boundary flux.
These properties distinguish the solutions from classical point-vortex configurations with prescribed nonzero circulations. The present solutions instead describe localized, sign-changing vorticity structures generated by nonlinear elliptic ground states. Third, despite the strong concentration of the interior vorticity, the resulting displacement of the free surface is exponentially small. Indeed, (1.17) implies
∥ ζ δ ∥ H 2 ( R ) ⟶ 0 \|\zeta_\delta\|_{H^2(\mathbb R)}
\longrightarrow0 ∥ ζ δ ∥ H 2 ( R ) ⟶ 0
faster than any algebraic power of δ \delta δ . The underlying mechanism can be seen directly from the Bernoulli condition. Since the spikes remain at positive distances from the free surface, their induced velocity on the surface is exponentially small. Accordingly, the boundary forcing
− 1 2 ∣ ∇ Ψ δ ∣ 2 -\frac12|\nabla\Psi_\delta|^2 − 2 1 ∣∇ Ψ δ ∣ 2
is exponentially small.
To leading order, the free-surface equation takes the form
( g − α 2 ∂ x 1 2 ) ζ δ ∼ − 1 2 ∣ ∇ Ψ δ ∣ 2 ∣ S . \begin{equation}
(g-\alpha^2\partial_{x_1}^2)
\zeta_\delta
\sim -\left.\frac12
|\nabla\Psi_\delta|^2\right|_{S}. \tag*{(1.21)}
\end{equation} ( g − α 2 ∂ x 1 2 ) ζ δ ∼ − 2 1 ∣∇ Ψ δ ∣ 2 S . (1.21)
The positivity of the inverse of g − α 2 ∂ x 1 2 g-\alpha^2\partial_{x_1}^2 g − α 2 ∂ x 1 2 explains why the leading-order deformation is a depression rather than an elevation. Thus the constructed solutions exhibit a separation of scales: the vorticity amplitude is of order δ − 2 \delta^{-2} δ − 2 , the kinetic energy is of order one, and the deformation of the free surface is exponentially small. In contrast to a single-spike configuration, the presence of several concentrations introduces additional interaction effects. The positions of the individual spikes must be determined simultaneously through a balance involving neighboring spikes and the boundaries of the fluid domain.
The main contribution in this paper is to construct, for every prescribed integer n ≥ 1 n\geq 1 n ≥ 1 , a family of finite-energy stationary capillary-gravity waves whose stream functions exhibit exactly n n n well-separated, alternating-sign spikes. The concentration centers approach an equally spaced vertical configuration, while the free-surface deformation remains exponentially small. Our construction is based on a nonlinear gluing procedure combining the nondegeneracy of the entire-space ground state with a finite-dimensional reduction for the concentration parameters.
Although the resulting vorticity is strongly concentrated, each individual core may contain both positive and negative vorticity. Thus the alternating-sign structure refers specifically to the stream-function spikes. The resulting family provides a new class of stationary rotational water waves in which the internal vorticity distribution exhibits an increasingly complicated multi-core structure, while the free surface remains close to its undisturbed position.
Outline of the construction
The proof is based on a nonlinear gluing scheme for the coupled elliptic free-boundary problem, building on methods developed for localized solutions of semilinear elliptic equations; see, for example, [10 , 11 ] .
We begin with a superposition of n n n alternating-sign copies of the entire-space ground state, positioned along a vertical line in the reference strip
Ω 0 = R × ( − 1 , 1 ) . \Omega_0=\mathbb R\times(-1,1). Ω 0 = R × ( − 1 , 1 ) .
Each spike is modified by a suitable boundary correction so that the resulting approximation satisfies the Dirichlet condition on the fluid boundary.
The next step is to analyze the linearized elliptic operator around this approximate multi-spike configuration. The nondegeneracy assumption (B) identifies the translational modes as the only leading-order obstructions to invertibility. Suitable orthogonality conditions are then imposed to control these modes and to solve for the infinite-dimensional correction.
A further difficulty arises from the unknown free boundary. The Bernoulli condition couples the surface profile to the velocity generated by the concentrated vorticity. This coupling is handled perturbatively, taking advantage of the exponential separation between the spike cores and the free surface.
Finally, the solvability conditions associated with the translational modes lead to a finite-dimensional system for the spike locations. Its dominant terms describe the interactions between neighboring spikes and their reflections across the fluid boundaries. Solving this system yields the asymptotic location formula (1.14) and closes the gluing construction.
The resulting approach demonstrates how the nonlinear gluing techniques for elliptic spike solutions can be adapted to a nonlinear free-boundary fluid problem. It also suggests possible extensions to other concentrated-vorticity configurations, including more general multi-core arrangements and rotational traveling waves. We note that this approach are also adopted to construct overhanging solitary water waves with weak gravity [19 ] .
Notation. Throughout the paper, C C C and c c c denote positive constants which may change from line to line but are independent of sufficiently small δ \delta δ . We write a ≲ b a\lesssim b a ≲ b if a ≤ C b a\leq Cb a ≤ C b , and a ≃ b a\simeq b a ≃ b if both a ≲ b a\lesssim b a ≲ b and b ≲ a b\lesssim a b ≲ a . For a function f f f in an L 2 L^2 L 2 space, f ⊥ f^\perp f ⊥ denotes an appropriate L 2 L^2 L 2 -orthogonal complement when the underlying subspace is clear from the context.
Reformulation of the elliptic free boundary problem
The steady water-wave problem is a classic elliptic free-boundary problem, where the fluid domain itself constitutes part of the unknowns. To address this difficulty, we employ a conformal mapping to transform the unknown fluid domain into a fixed reference domain.
Conformal Mapping
A conformal parametrization is introduced to reformulate the free-boundary problem on a fixed reference domain. In this paper, we work with the water waves with a flat bottom.
The fluid domain is defined by
Ω = { ( x 1 , x 2 ) ∈ R 2 ∣ − 1 < x 2 < 1 + ζ ( x 1 ) } . \Omega = \{(x_1,x_2)\in {\mathbb{R}}^2\;|\; -1< x_2 < 1 + \zeta(x_1) \}. Ω = {( x 1 , x 2 ) ∈ R 2 ∣ − 1 < x 2 < 1 + ζ ( x 1 )} .
The complex strip
S : = { z 1 + i z 2 ∈ C ∣ ∣ z 2 ∣ < 1 } \mathcal{S}:=\{z_1+\mathrm iz_2\in {\mathbb{C}}\;|\;|z_2|<1\} S := { z 1 + i z 2 ∈ C ∣ ∣ z 2 ∣ < 1 }
serves as the reference domain. We now seek a holomorphic mapping Γ \Gamma Γ such that Ω \Omega Ω is the image of reference domain S \mathcal{S} S under this mapping.
Next, we follow the construction from [1 Section 2] . Let Γ = Γ 1 + i Γ 2 : S → C \Gamma=\Gamma_1+\mathrm i\Gamma_2:\mathcal{S}\to {\mathbb{C}} Γ = Γ 1 + i Γ 2 : S → C be a holomorphic mapping satisfying Γ 2 ∣ { z 2 = − 1 } = 0 {\Gamma_2|_{\{z_2=-1\}}=0} Γ 2 ∣ { z 2 = − 1 } = 0 . Since Γ 2 \Gamma_2 Γ 2 is harmonic, it is uniquely determined in the space H 2 ( S ) H^2(\mathcal{S}) H 2 ( S ) by its trace Γ 2 ( ⋅ , ± 1 ) \Gamma_2(\cdot,\pm1) Γ 2 ( ⋅ , ± 1 ) . By Γ 2 ( ⋅ , − 1 ) = 0 \Gamma_2(\cdot,-1)=0 Γ 2 ( ⋅ , − 1 ) = 0 , we have
Γ 2 = sinh ( ∣ ∂ x 1 ∣ ( x 2 + 1 ) ) sinh ( 2 ∣ ∂ x 1 ∣ ) Γ 2 ( ⋅ , 1 ) in S , \begin{equation}
\Gamma_2=\frac{\sinh{\left(|\partial_{x_1}|(x_2+1)\right)}}{\sinh{\left(2|\partial_{x_1}|\right)} } \Gamma_2(\cdot,1)\qquad\textrm{in } \mathcal{S}, \tag*{(2.1)}
\end{equation} Γ 2 = sinh ( 2∣ ∂ x 1 ∣ ) sinh ( ∣ ∂ x 1 ∣ ( x 2 + 1 ) ) Γ 2 ( ⋅ , 1 ) in S , (2.1)
and
∂ x 2 Γ 2 = ∣ ∂ x 1 ∣ coth ( 2 ∣ ∂ x 1 ∣ ) Γ 2 on { x 2 = 1 } . \begin{equation}
\partial_{x_2}\Gamma_2=|\partial_{x_1}|\coth{\left(2|\partial_{x_1}|\right)} \Gamma_2 \qquad\text{on }\{x_2 = 1\}. \tag*{(2.2)}
\end{equation} ∂ x 2 Γ 2 = ∣ ∂ x 1 ∣ coth ( 2∣ ∂ x 1 ∣ ) Γ 2 on { x 2 = 1 } . (2.2)
Assume that Γ 2 ( ⋅ , 1 ) \Gamma_2(\cdot,1) Γ 2 ( ⋅ , 1 ) is even in x 1 x_1 x 1 . Then Γ 2 \Gamma_2 Γ 2 is also even in x 1 x_1 x 1 . Since Γ 1 \Gamma_1 Γ 1 is harmonic conjugate to Γ 2 \Gamma_2 Γ 2 , by Cauchy–Riemann equation, Γ 1 \Gamma_1 Γ 1 is odd in x 1 x_1 x 1 . Then Γ 1 \Gamma_1 Γ 1 is uniquely determined by Γ 2 \Gamma_2 Γ 2 . Therefore, we can obtain Γ \Gamma Γ from trace Γ 2 ( ⋅ , 1 ) \Gamma_2(\cdot,1) Γ 2 ( ⋅ , 1 ) .
We can now write the fluid domain as
Ω = ( i d + Γ ) ( { ( x 1 , x 2 ) ∈ R 2 ∣ ∣ x 2 ∣ < 1 } ) = { ( x 1 + Γ 1 ( x 1 , x 2 ) , x 2 + Γ 2 ( x 1 , x 2 ) ) ∈ R 2 ∣ ∣ x 2 ∣ < 1 } . \begin{align*}
\Omega&=(id+\Gamma)\left(\{(x_1,x_2)\in {\mathbb{R}}^2\;|\;|x_2|< 1\}\right)\\
&=\left\{\left(x_1+\Gamma_1(x_1,x_2),x_2+\Gamma_2(x_1,x_2)\right)\in {\mathbb{R}}^2\;|\;|x_2|< 1\right\}.
\end{align*} Ω = ( i d + Γ ) ( {( x 1 , x 2 ) ∈ R 2 ∣ ∣ x 2 ∣ < 1 } ) = { ( x 1 + Γ 1 ( x 1 , x 2 ) , x 2 + Γ 2 ( x 1 , x 2 ) ) ∈ R 2 ∣ ∣ x 2 ∣ < 1 } .
Pulling back the stream function gives
Φ = Ψ ∘ ( i d + Γ ) : { ( x 1 , x 2 ) ∈ R 2 ∣ ∣ x 2 ∣ < 1 } → R , \begin{equation}
\Phi=\Psi\circ(id+\Gamma):\{(x_1,x_2)\in {\mathbb{R}}^2\;|\;|x_2|< 1\}\to {\mathbb{R}}, \tag*{(2.3)}
\end{equation} Φ = Ψ ∘ ( i d + Γ ) : {( x 1 , x 2 ) ∈ R 2 ∣ ∣ x 2 ∣ < 1 } → R , (2.3)
hence problem (1.6) , (1.10) and (1.9) become
{ δ 2 Δ Φ = ∣ 1 + Γ ′ ∣ 2 γ ( Φ ) in { ( x 1 , x 2 ) ∈ R 2 ∣ ∣ x 2 ∣ < 1 } , Φ = 0 on { ( x 1 , x 2 ) ∈ R 2 ∣ ∣ x 2 ∣ = 1 } , \begin{equation}
\left\{\begin{aligned}
\delta^2\Delta\Phi&=|1+\Gamma^\prime|^2\gamma(\Phi) &\qquad\text{in }\{(x_1,x_2)\in {\mathbb{R}}^2\;|\;|x_2|< 1\},\\
\Phi&=0 &\qquad\text{on }\{(x_1,x_2)\in {\mathbb{R}}^2\;|\;|x_2|= 1\},
\end{aligned}\right. \tag*{(2.4)}
\end{equation} { δ 2 ΔΦ Φ = ∣1 + Γ ′ ∣ 2 γ ( Φ ) = 0 in {( x 1 , x 2 ) ∈ R 2 ∣ ∣ x 2 ∣ < 1 } , on {( x 1 , x 2 ) ∈ R 2 ∣ ∣ x 2 ∣ = 1 } , (2.4)
and the Bernoulli condition (1.10) becomes
1 2 ( ∂ x 2 Φ ) 2 ∣ 1 + Γ ′ ∣ 2 − α 2 I m ( Γ ′ ′ ( 1 + Γ ′ ‾ ) ) ∣ 1 + Γ ′ ∣ 3 + g Γ 2 = 0 on { ( x 1 , x 2 ) ∈ R 2 ∣ ∣ x 2 ∣ = 1 } , \begin{equation}
\frac{1}{2}\frac{(\partial_{x_2}\Phi)^2}{|1+\Gamma^\prime|^{2}}-\alpha^2\frac{Im(\Gamma^{\prime\prime}(1+\overline{\Gamma^\prime}))}{|1+\Gamma^\prime|^{3}}+g\Gamma_2=0\qquad\text{on }\{(x_1,x_2)\in {\mathbb{R}}^2\;|\;|x_2|= 1\}, \tag*{(2.5)}
\end{equation} 2 1 ∣1 + Γ ′ ∣ 2 ( ∂ x 2 Φ ) 2 − α 2 ∣1 + Γ ′ ∣ 3 I m ( Γ ′′ ( 1 + Γ ′ )) + g Γ 2 = 0 on {( x 1 , x 2 ) ∈ R 2 ∣ ∣ x 2 ∣ = 1 } , (2.5)
where Γ ′ = ∂ x 1 Γ \Gamma^\prime=\partial_{x_1}\Gamma Γ ′ = ∂ x 1 Γ .
After scaling, define
φ = Φ ( δ ⋅ ) , \begin{equation}
\varphi=\Phi(\delta\,\cdot), \tag*{(2.6)}
\end{equation} φ = Φ ( δ ⋅ ) , (2.6)
then equation (2.4) translates to
{ Δ φ = ∣ 1 + Γ ′ ( δ ⋅ ) ∣ 2 γ ( φ ) in S δ , φ = 0 on ∂ S δ , \begin{equation}
\left\{\begin{aligned}
&\Delta\varphi=|1+\Gamma^\prime(\delta\,\cdot)|^2\gamma(\varphi)\qquad&&\text{in }\mathcal{S}_\delta,\\
&\varphi=0\qquad&&\textrm{on }\partial\mathcal{S}_\delta,
\end{aligned}\right. \tag*{(2.7)}
\end{equation} { Δ φ = ∣1 + Γ ′ ( δ ⋅ ) ∣ 2 γ ( φ ) φ = 0 in S δ , on ∂ S δ , (2.7)
where the slab S δ \mathcal{S}_\delta S δ is defined as
S δ = { x ∈ R 2 ∣ ∣ x 2 ∣ < 1 δ } . \begin{equation}
\mathcal{S}_\delta=\left\{x\in\mathbb{R}^2\;|\;|x_2|<\tfrac{1}{\delta}\right\}. \tag*{(2.8)}
\end{equation} S δ = { x ∈ R 2 ∣ ∣ x 2 ∣ < δ 1 } . (2.8)
The Bernoulli condition is transformed into
1 2 δ 2 ( ∂ x 2 φ ( ⋅ , 1 δ ) ) 2 ∣ 1 + Γ ′ ( δ ⋅ , 1 ) ∣ 2 − α 2 I m ( Γ ′ ′ ( δ ⋅ , 1 ) ( 1 + Γ ′ ( δ ⋅ , 1 ) ‾ ) ) ∣ 1 + Γ ′ ( δ ⋅ , 1 ) ∣ 3 + g Γ 2 ( δ ⋅ , 1 ) = 0. \begin{equation}
\frac{1}{2\delta^{2}}\frac{\left(\partial_{x_2}\varphi(\cdot,\tfrac{1}{\delta})\right)^2}{|1+\Gamma^\prime(\delta\,\cdot,1)|^{2}}-\alpha^2\frac{Im (\Gamma^{\prime\prime}(\delta\,\cdot,1)(1+ \overline{\Gamma^\prime(\delta\,\cdot,1)}))}{|1+\Gamma^\prime(\delta\,\cdot,1)|^3}+g\Gamma_2(\delta\,\cdot,1)=0. \tag*{(2.9)}
\end{equation} 2 δ 2 1 ∣1 + Γ ′ ( δ ⋅ , 1 ) ∣ 2 ( ∂ x 2 φ ( ⋅ , δ 1 ) ) 2 − α 2 ∣1 + Γ ′ ( δ ⋅ , 1 ) ∣ 3 I m ( Γ ′′ ( δ ⋅ , 1 ) ( 1 + Γ ′ ( δ ⋅ , 1 ) )) + g Γ 2 ( δ ⋅ , 1 ) = 0. (2.9)
Approximate Solution
We want to solve for φ \varphi φ modeled on U U U . Although U U U has exponential localization, it does not satisfy the homogeneous boundary conditions in (2.7) . We therefore perform a boundary correction, that is, we subtract from U U U a function with the same trace but with very small interior effect.
Since U U U is exponentially localized and γ ′ ( 0 ) = 1 \gamma'(0)=1 γ ′ ( 0 ) = 1 , away from the origin, the linearized operator γ ′ ( U ) − Δ \gamma^\prime(U)-\Delta γ ′ ( U ) − Δ behaves like 1 − Δ 1-\Delta 1 − Δ . Hence for U ∈ H s ( ∂ S δ ) U\in H^s(\partial\mathcal{S}_\delta) U ∈ H s ( ∂ S δ ) , define the boundary correction U b c U_{\mathrm{bc}} U bc of U U U to be the unique solution of the following equation in the space H s ( S δ ) H^s(\mathcal{S}_\delta) H s ( S δ ) :
{ ( 1 − Δ ) U b c = 0 in S δ , U b c = U on ∂ S δ . \begin{equation}
\left\{\begin{aligned}
&(1-\Delta)U_{\mathrm{bc}}=0\qquad&&\textrm{in }\mathcal{S}_{\delta},\\
&U_{\mathrm{bc}}=U\qquad&&\textrm{on }\partial\mathcal{S}_\delta.
\end{aligned}\right. \tag*{(2.10)}
\end{equation} { ( 1 − Δ ) U bc = 0 U bc = U in S δ , on ∂ S δ . (2.10)
By elliptic regularity estimates (see [12 Theorem 8.13] ), for any s > 0 s>0 s > 0 ,
∥ U b c ∥ H s ( S δ ) ≲ ∥ U ∥ H s ( ∂ S δ ) . \begin{equation}
\|U_{\mathrm{bc}}\|_{H^{s}(\mathcal{S}_\delta)}\lesssim\|U\|_{H^s (\partial\mathcal{S}_\delta)}. \tag*{(2.11)}
\end{equation} ∥ U bc ∥ H s ( S δ ) ≲ ∥ U ∥ H s ( ∂ S δ ) . (2.11)
A more detailed analysis of U b c U_{\mathrm{bc}} U bc can be found in § 2.3 .
For τ ∈ ( − 1 , 1 ) \tau\in(-1,1) τ ∈ ( − 1 , 1 ) , let U ( ⋅ ; τ ) : = U ( ⋅ − τ δ e 2 ) U(\,\cdot\,;\tau):=U\left(\,\cdot-\tfrac{\tau}{\delta}e_2\right) U ( ⋅ ; τ ) := U ( ⋅ − δ τ e 2 ) with U U U as in Assumption assumption:A . For a fixed small positive constant ε \varepsilon ε , define
P ε : = { τ = ( τ 1 , τ 2 , … , τ n ) ∣ τ i ∈ ( − 1 , 1 ) , ∣ τ i ± 1 ∣ > ε , ∣ τ i − τ j ∣ > ε , i , j = 1 , … , n , i ≠ j } . P_{\varepsilon}:=\{\boldsymbol{\tau}=(\tau_1,\tau_2,\dots,\tau_n)\;|\;\tau_i\in (-1,1), |\tau_i\pm1|>\varepsilon,|\tau_i-\tau_j|>\varepsilon,i,j=1,\dots,n,i\neq j\}. P ε := { τ = ( τ 1 , τ 2 , … , τ n ) ∣ τ i ∈ ( − 1 , 1 ) , ∣ τ i ± 1∣ > ε , ∣ τ i − τ j ∣ > ε , i , j = 1 , … , n , i = j } .
Given τ ∈ P δ \boldsymbol{\tau}\in P_{\delta} τ ∈ P δ , let
σ i : = ( − 1 ) i + 1 , i = 1 , … , n , \sigma_i:=(-1)^{i+1},\qquad i=1,\dots,n, σ i := ( − 1 ) i + 1 , i = 1 , … , n ,
we construct the approximate solution as
ω τ = ∑ i = 1 n σ i ( U ( ⋅ ; τ i ) − U ( ⋅ ; τ i ) b c ) , \omega_{\boldsymbol{\tau}}=\sum_{i=1}^n\sigma_i\left(U(\,\cdot\,;\tau_i)-U(\,\cdot\,;\tau_i)_{\mathrm{bc}}\right), ω τ = i = 1 ∑ n σ i ( U ( ⋅ ; τ i ) − U ( ⋅ ; τ i ) bc ) ,
where U ( ⋅ ; τ i ) b c U(\,\cdot\,;\tau_i)_{\mathrm{bc}} U ( ⋅ ; τ i ) bc is the solution of the equation (2.10) . For simplicity of notation, we denote U ( ⋅ ; τ i ) U(\,\cdot\,;\tau_i) U ( ⋅ ; τ i ) as U i U_i U i . Then we make the ansatz
φ = ω τ + u . \begin{equation}
\varphi=\omega_{\boldsymbol{\tau}}+u. \tag*{(2.12)}
\end{equation} φ = ω τ + u . (2.12)
Substituting it into (2.7) , by Δ σ i U i = σ i γ ( U i ) \Delta \sigma_iU_i =\sigma_i \gamma(U_i) Δ σ i U i = σ i γ ( U i ) and Δ σ i ( U i ) b c = σ i ( U i ) b c \Delta \sigma_i(U_i)_\mathrm{bc} = \sigma_i(U_i)_\mathrm{bc} Δ σ i ( U i ) bc = σ i ( U i ) bc , u u u satisfies the following equation:
Δ u = ∣ 1 + Γ ′ ( δ ⋅ ) ∣ 2 γ ( ω τ + u ) − ∑ i = 1 n σ i γ ( U i ) + ∑ i = 1 n σ i ( U i ) b c = γ ′ ( ∑ i = 1 n σ i U i ) u + ∣ 1 + Γ ′ ( δ ⋅ ) ∣ 2 γ ( ω τ + u ) − ∑ i = 1 n σ i γ ( U i ) + ∑ i = 1 n σ i ( U i ) b c − γ ′ ( ∑ i = 1 n σ i U i ) u in S δ . \begin{equation}
\\
\begin{aligned}
\Delta u=&\left|1+\Gamma^\prime(\delta\,\cdot\,)\right|^2 \gamma(\omega_{\boldsymbol{\tau}}+u)-\sum_{i=1}^n\sigma_i\gamma(U_i)+\sum_{i=1}^n\sigma_i(U_i)_\mathrm{bc}\\
=&\gamma^\prime\left(\sum_{i=1}^n\sigma_iU_i\right)u+\left|1+\Gamma^\prime(\delta \,\cdot\,)\right|^2\gamma(\omega_{\boldsymbol{\tau}}+u)\\
&-\sum_{i=1}^n\sigma_i\gamma(U_i)+\sum_{i=1}^n\sigma_i(U_i)_\mathrm{bc}-\gamma^\prime\left(\sum_{i=1}^n\sigma_iU_i\right)u\qquad\text{in }\mathcal{S}_\delta.
\end{aligned} \tag*{(2.13)}
\end{equation} Δ u = = ∣ 1 + Γ ′ ( δ ⋅ ) ∣ 2 γ ( ω τ + u ) − i = 1 ∑ n σ i γ ( U i ) + i = 1 ∑ n σ i ( U i ) bc γ ′ ( i = 1 ∑ n σ i U i ) u + ∣ 1 + Γ ′ ( δ ⋅ ) ∣ 2 γ ( ω τ + u ) − i = 1 ∑ n σ i γ ( U i ) + i = 1 ∑ n σ i ( U i ) bc − γ ′ ( i = 1 ∑ n σ i U i ) u in S δ . (2.13)
Since ω τ ∣ ∂ S δ = 0 \omega_{\boldsymbol{\tau}}|_{\partial\mathcal{S}_\delta}=0 ω τ ∣ ∂ S δ = 0 , the boundary value condition in (2.7) reduces to
u = 0 on ∂ S δ . \begin{equation}
u=0 \qquad\textrm{on }\partial\mathcal{S}_\delta. \tag*{(2.14)}
\end{equation} u = 0 on ∂ S δ . (2.14)
We next turn to the Bernoulli condition (2.9) . Let Γ s = Γ 2 ( ⋅ , 1 ) \Gamma_{\mathrm{s}}=\Gamma_2(\,\cdot\,,1) Γ s = Γ 2 ( ⋅ , 1 ) , using the Cauchy-Riemann equations,
Γ ′ = ∂ x 1 Γ 1 + i ∂ x 1 Γ 2 = ∂ x 2 Γ 2 + i ∂ x 1 Γ 2 . \Gamma^\prime=\partial_{x_1}\Gamma_1+\mathrm i\partial_{x_1} \Gamma_2=\partial_{x_2}\Gamma_2+\mathrm i\partial_{x_1} \Gamma_2. Γ ′ = ∂ x 1 Γ 1 + i ∂ x 1 Γ 2 = ∂ x 2 Γ 2 + i ∂ x 1 Γ 2 .
Together with (2.12) and (2.2) , we have
1 2 δ 2 ( ∂ x 2 ( ω τ + u ) ( ⋅ δ , 1 δ ) ) 2 ( 1 + m ( ∂ x 1 ) Γ s ) 2 + Γ s ′ 2 − α 2 ( 1 + m ( ∂ x 1 ) Γ s ) Γ s ′ ′ − Γ s ′ m ( ∂ x 1 ) Γ s ′ ( ( 1 + m ( ∂ x 1 ) Γ s ) 2 + Γ s ′ 2 ) 3 / 2 + g Γ s = 0 , \begin{equation}
\frac{1}{2\delta^{2}}\frac{\left(\partial_{x_2}(\omega_{\boldsymbol{\tau}}+u)\left(\frac{\,\cdot\,}{\delta},\frac{1}{\delta} \right)\right)^2}{\left(1+m({\partial_{x_1}})\Gamma_{\mathrm{s}}\right)^2+{\Gamma_{\mathrm{s}}^\prime}^2}-\alpha^2\frac{(1+m({\partial_{x_1}})\Gamma_{\mathrm{s}})\Gamma_{\mathrm{s}}^{\prime\prime}-\Gamma_{\mathrm{s}}^\prime m({\partial_{x_1}})\Gamma_{\mathrm{s}}^\prime}{\left((1+m({\partial_{x_1}})\Gamma_{\mathrm{s}})^2+{\Gamma_{\mathrm{s}}^\prime}^2\right)^{3/2}}+g\Gamma_{\mathrm{s}}=0, \tag*{(2.15)}
\end{equation} 2 δ 2 1 ( 1 + m ( ∂ x 1 ) Γ s ) 2 + Γ s ′ 2 ( ∂ x 2 ( ω τ + u ) ( δ ⋅ , δ 1 ) ) 2 − α 2 ( ( 1 + m ( ∂ x 1 ) Γ s ) 2 + Γ s ′ 2 ) 3/2 ( 1 + m ( ∂ x 1 ) Γ s ) Γ s ′′ − Γ s ′ m ( ∂ x 1 ) Γ s ′ + g Γ s = 0 , (2.15)
where m ( ∂ x 1 ) = ∣ ∂ x 1 ∣ coth ( 2 ∣ ∂ x 1 ∣ ) m({\partial_{x_1}})=|{\partial_{x_1}}|\coth{(2|{\partial_{x_1}}|)} m ( ∂ x 1 ) = ∣ ∂ x 1 ∣ coth ( 2∣ ∂ x 1 ∣ ) . If ∥ Γ s ∥ H 2 \|\Gamma_{\mathrm{s}}\|_{H^2} ∥ Γ s ∥ H 2 small, the leading term of the operator in (2.15) acting on Γ s \Gamma_{\mathrm{s}} Γ s is an invertible operator g − α 2 ∂ x 1 2 : H k ( R ) → H k − 2 ( R ) , k ≥ 2 g-\alpha^2 \partial_{x_1}^2:H^k( {\mathbb{R}})\to H^{k-2}( {\mathbb{R}}),\,k\geq2 g − α 2 ∂ x 1 2 : H k ( R ) → H k − 2 ( R ) , k ≥ 2 . Define operator A ( Γ s ) A(\Gamma_{\mathrm{s}}) A ( Γ s ) as
A ( Γ s ) = ( g − α 2 ( 1 + m ( ∂ x 1 ) Γ s ) ( ∂ x 1 ) 2 − Γ s ′ m ( ∂ x 1 ) ∂ x 1 ( ( 1 + m ( ∂ x 1 ) Γ s ) 2 + Γ s ′ 2 ) − 3 2 ) ( g − α 2 ∂ x 1 2 ) − 1 . A(\Gamma_{\mathrm{s}})=\left(g-
\alpha^2\frac{\left(1+m({\partial_{x_1}})\Gamma_{\mathrm{s}}\right)({\partial_{x_1}})^2-\Gamma_{\mathrm{s}}^\prime m({\partial_{x_1}}){\partial_{x_1}}}{\left((1+m({\partial_{x_1}})\Gamma_{\mathrm{s}})^2+{\Gamma_{\mathrm{s}}^\prime}^2 \right)^{-\frac{3}{2}}}
\right)(g-\alpha^2\partial_{x_1}^2)^{-1}. A ( Γ s ) = g − α 2 ( ( 1 + m ( ∂ x 1 ) Γ s ) 2 + Γ s ′ 2 ) − 2 3 ( 1 + m ( ∂ x 1 ) Γ s ) ( ∂ x 1 ) 2 − Γ s ′ m ( ∂ x 1 ) ∂ x 1 ( g − α 2 ∂ x 1 2 ) − 1 .
For ∥ Γ s ∥ H 2 \|\Gamma_{\mathrm{s}}\|_{H^{2}} ∥ Γ s ∥ H 2 small, since A ( 0 ) = i d A(0)=id A ( 0 ) = i d and
∥ A ( Γ s ) − i d ∥ L ( H 2 ) ≲ ∥ Γ s ∥ H k , \|A(\Gamma_{\mathrm{s}})-id\|_{\mathcal L(H^{2})}\lesssim \|\Gamma_{\mathrm{s}}\|_{H^{k}}, ∥ A ( Γ s ) − i d ∥ L ( H 2 ) ≲ ∥ Γ s ∥ H k ,
then A ( Γ s ) A(\Gamma_{\mathrm{s}}) A ( Γ s ) is invertible. Hence we can write (2.15) as
( g − α 2 ( ∂ x 1 ) 2 ) Γ s + 1 2 δ 2 A ( Γ s ) − 1 [ ( ∂ x 2 ( ω τ + u ) ( ⋅ δ , 1 δ ) ) 2 ( 1 + m ( ∂ x 1 ) Γ s ) 2 + Γ s ′ 2 ] = 0. \begin{equation}
\begin{aligned}
(g-\alpha^2({\partial_{x_1}})^2)\Gamma_{\mathrm{s}}+\frac{1}{2\delta^{2}}A(\Gamma_{\mathrm{s}})^{-1}\left[\frac{\left(\partial_{x_2}(\omega_{\boldsymbol{\tau}}+u)\left(\frac{\,\cdot\,}{\delta},\frac{1}{\delta}\right)\right)^2}{(1+m({\partial_{x_1}})\Gamma_{\mathrm{s}})^2+{\Gamma_{\mathrm{s}}^\prime}^2}\right]=0.
\end{aligned} \tag*{(2.16)}
\end{equation} ( g − α 2 ( ∂ x 1 ) 2 ) Γ s + 2 δ 2 1 A ( Γ s ) − 1 [ ( 1 + m ( ∂ x 1 ) Γ s ) 2 + Γ s ′ 2 ( ∂ x 2 ( ω τ + u ) ( δ ⋅ , δ 1 ) ) 2 ] = 0. (2.16)
From the above analysis, if u u u , Γ s \Gamma_{\mathrm{s}} Γ s and τ \boldsymbol{\tau} τ satisfy
( − Δ + γ ′ ( ∑ i = 1 n σ i U i ) ) u + F ( τ , u , Γ s ) = 0 in S δ ( g − α 2 ( ∂ x 1 ) 2 ) Γ s + G ( τ , u , Γ s ) = 0 on R , \begin{align}
\left(-\Delta+\gamma^\prime\left(\textstyle\sum_{i=1}^n\sigma_iU_i\right)\right)u + F(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}})&=0 \qquad\textrm{in }\mathcal{S}_\delta \\
\left(g-\alpha^2({\partial_{x_1}})^2\right)\Gamma_{\mathrm{s}}+G(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}})&=0 \qquad\textrm{on }\mathbb{R},
\end{align} ( − Δ + γ ′ ( ∑ i = 1 n σ i U i ) ) u + F ( τ , u , Γ s ) ( g − α 2 ( ∂ x 1 ) 2 ) Γ s + G ( τ , u , Γ s ) = 0 in S δ = 0 on R ,
where
F ( τ , u , Γ s ) : = ∣ 1 + Γ ′ ( δ ⋅ ) ∣ 2 γ ( ω τ + u ) − ∑ i = 1 n σ i γ ( U i ) + ∑ i = 1 n σ i ( U i ) b c − γ ′ ( ∑ i = 1 n σ i U i ) u \begin{equation}
F(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}}):=\left|1+\Gamma^\prime(\delta \,\cdot\,)\right|^2\gamma(\omega_{\boldsymbol{\tau}}+u)
-\sum_{i=1}^n\sigma_i\gamma(U_i)+\sum_{i=1}^n\sigma_i(U_i)_\mathrm{bc}-\gamma^\prime\left(\sum_{i=1}^n\sigma_iU_i\right)u \tag*{(2.17)}
\end{equation} F ( τ , u , Γ s ) := ∣ 1 + Γ ′ ( δ ⋅ ) ∣ 2 γ ( ω τ + u ) − i = 1 ∑ n σ i γ ( U i ) + i = 1 ∑ n σ i ( U i ) bc − γ ′ ( i = 1 ∑ n σ i U i ) u (2.17)
and
G ( τ , u , Γ s ) : = 1 2 δ 2 A ( Γ s ) − 1 [ ( ∂ x 2 ( ω τ + u ) ( ⋅ δ , 1 δ ) ) 2 ( 1 + ∣ ∂ x 1 ∣ coth ( 2 ∣ ∂ x 1 ∣ ) Γ s ) 2 + Γ s ′ 2 ] , \begin{equation}
G(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}}):=
\frac{1}{2\delta^{2}}A(\Gamma_{\mathrm{s}})^{-1}\left[\frac{\left(\partial_{x_2}(\omega_{\boldsymbol{\tau}}+u)\left(\frac{\,\cdot\,}{\delta},\frac{1}{\delta}\right)\right)^2}{(1+|{\partial_{x_1}}|\coth{(2|{\partial_{x_1}}|)}\Gamma_{\mathrm{s}})^2+{\Gamma_{\mathrm{s}}^\prime}^2}\right], \tag*{(2.18)}
\end{equation} G ( τ , u , Γ s ) := 2 δ 2 1 A ( Γ s ) − 1 [ ( 1 + ∣ ∂ x 1 ∣ coth ( 2∣ ∂ x 1 ∣ ) Γ s ) 2 + Γ s ′ 2 ( ∂ x 2 ( ω τ + u ) ( δ ⋅ , δ 1 ) ) 2 ] , (2.18)
then ( Ψ , ζ ) (\Psi,\zeta) ( Ψ , ζ ) obtained from (2.1) , (2.3) , (2.6) , (2.12) will solve the stationary water wave problem(1.6) , (1.10) and (1.9) .
Basic properties of U and boundary correction
In this subsection, we collect the estimates for U U U and its boundary correction that will be used in the sequel. These estimates are consequences of the exponential decay for U U U . We state them here.
Lemma 2.1. Under Assumptions assumption:A and assumption:B , there exists λ > 0 \lambda> 0 λ > 0 such that lim r → ∞ ( − 1 ) k r 1 2 e r ∂ r k U ( r ) = λ , for all 0 ≤ k ≤ k 0 + 2.
\lim_{r\to\infty}(-1)^{k}r^{\frac{1}{2}}{\mathrm e}^{r}\partial_r^k U(r)=\lambda, \qquad\text{for all } 0\le k\le k_0+2.
lim r → ∞ ( − 1 ) k r 2 1 e r ∂ r k U ( r ) = λ , for all 0 ≤ k ≤ k 0 + 2. In particular, for radial functions, we have ∂ x 2 U ≂ − x 2 r − 3 2 e − r , r ≫ 1.
\partial_{x_2}U\eqsim- x_2 r^{-\frac{3}{2}}{\mathrm e}^{-r},\qquad r\gg 1.
∂ x 2 U ≂ − x 2 r − 2 3 e − r , r ≫ 1.
The proof can be found in [1 Proposition 3.1] . Since U i U_i U i is obtained by translating U U U , the estimates for U i U_i U i and its derivatives also follow from the above lemma. Next, we consider the boundary correction ( U i ) b c (U_i)_{\mathrm{bc}} ( U i ) bc . From (2.10) , it satisfies
{ ( 1 − Δ ) ( U i ) b c = 0 in S δ , ( U i ) b c = U i on ∂ S δ . \begin{equation}
\left\{\begin{aligned}
&(1-\Delta)(U_i)_\mathrm{bc}=0\qquad&&\textrm{in }\mathcal{S}_{\delta},\\
&(U_i)_\mathrm{bc}=U_i\qquad&&\textrm{on }\partial\mathcal{S}_\delta.
\end{aligned}\right. \tag*{(2.21)}
\end{equation} { ( 1 − Δ ) ( U i ) bc = 0 ( U i ) bc = U i in S δ , on ∂ S δ . (2.21)
To obtain ( U i ) b c (U_i)_{\mathbf{bc}} ( U i ) bc , we need the Green’s function corresponding to the Dirichlet problem for the operator 1 − Δ 1-\Delta 1 − Δ in S δ \mathcal{S}_{\delta} S δ .
We first consider the Green’s function in R 2 {\mathbb{R}}^2 R 2 . Let G 0 G_0 G 0 be the fundamental solution satisfying
− Δ G 0 + G 0 = δ 0 in R 2 . -\Delta G_0+G_0=\delta_0\qquad\text{in } {\mathbb{R}}^2. − Δ G 0 + G 0 = δ 0 in R 2 .
Taking the Fourier transform, we obtain
G 0 ( x ) = 1 ( 2 π ) 2 ∫ R 2 e i ξ ⋅ x 1 + ∣ ξ ∣ 2 d ξ . G_0(x)=\frac{1}{(2\pi)^2}\int_{ {\mathbb{R}}^2}\frac{e^{i\xi\cdot x}}{1+|\xi|^2}\,d\xi. G 0 ( x ) = ( 2 π ) 2 1 ∫ R 2 1 + ∣ ξ ∣ 2 e i ξ ⋅ x d ξ .
Since G 0 G_0 G 0 is radially symmetric, we write G 0 ( x ) = G 0 ( ∣ x ∣ ) = G 0 ( r ) G_0(x)=G_0(|x|)=G_0(r) G 0 ( x ) = G 0 ( ∣ x ∣ ) = G 0 ( r ) . By an argument similar to Lemma 2.1 and [4 ] ,
G 0 ( r ) = O ( ∣ log r ∣ ) as r → 0 , ∂ r k G 0 ( r ) = O ( r − k ) as r → 0 , k ≥ 1 , ∂ r k G 0 ( r ) = O ( r − 1 / 2 e − r ) as r → ∞ , k ≥ 0. \begin{equation}
\begin{aligned}
G_0(r)&=O(|\log r|)&&\quad\text{as }r\to0,\\
\partial_r^kG_0(r)&=O(r^{-k})&&\quad\text{as }r\to0,k\geq 1,\\
\partial_r^kG_0(r)&=O(r^{-1/2}e^{-r})&&\quad\text{as }r\to\infty,k\geq 0.
\end{aligned} \tag*{(2.22)}
\end{equation} G 0 ( r ) ∂ r k G 0 ( r ) ∂ r k G 0 ( r ) = O ( ∣ log r ∣ ) = O ( r − k ) = O ( r − 1/2 e − r ) as r → 0 , as r → 0 , k ≥ 1 , as r → ∞ , k ≥ 0. (2.22)
Next, we construct the Green’s function in the strip S δ \mathcal{S}_\delta S δ by the method of images. Recall that the boundary of S δ \mathcal{S}_\delta S δ consists of two horizontal lines. In order to enforce the Dirichlet boundary condition, we reflect the singularity with respect to each boundary and alternate the sign of the reflected terms.
More precisely, define
G ~ ( x , y ) : = ∑ m ∈ Z [ G 0 ( x 1 − y 1 , x 2 − y 2 + 4 m δ ) − G 0 ( x 1 − y 1 , x 2 + y 2 − 2 δ + 4 m δ ) ] . \begin{equation}
\begin{aligned}
\widetilde G(x,y):=\sum_{m\in\mathbb Z}\Bigg[&G_0\left(x_1-y_1,
x_2-y_2+\frac{4m}{\delta}\right)\\
&-G_0\left(x_1-y_1,x_2+y_2-\frac{2}{\delta}+\frac{4m}{\delta}
\right)\Bigg].
\end{aligned} \tag*{(2.23)}
\end{equation} G ( x , y ) := m ∈ Z ∑ [ G 0 ( x 1 − y 1 , x 2 − y 2 + δ 4 m ) − G 0 ( x 1 − y 1 , x 2 + y 2 − δ 2 + δ 4 m ) ] . (2.23)
The alternating signs guarantee the cancellation of the Green function on the two boundary components of ∂ S δ \partial\mathcal{S}_\delta ∂ S δ . Indeed, on upper boundary x 2 = 1 / δ x_2=1/\delta x 2 = 1/ δ , using the symmetry of G 0 ( r ) G_0(r) G 0 ( r ) , the two sums cancel. Similarly, on lower boundary x 2 = − 1 / δ x_2=-1/\delta x 2 = − 1/ δ , the two sums cancel after using the symmetry of G 0 G_0 G 0 and changing the index m ↦ m + 1 m\mapsto m+1 m ↦ m + 1 of the latter term. Moreover, only the m = 0 m=0 m = 0 term in (2.23) has a singularity at x = y x=y x = y , while all other image terms are smooth in S δ \mathcal{S}_\delta S δ . Hence, G ~ ( x , y ) \widetilde G(x,y) G ( x , y ) satisfies
{ − Δ G ~ ( x , y ) + G ~ ( x , y ) = δ y ( x ) , x ∈ S δ , G ~ ( x , y ) = 0 , x ∈ ∂ S δ . \begin{cases}
-\Delta \widetilde G(x,y)+\widetilde G(x,y)=\delta_y(x),&x\in\mathcal{S}_\delta,\\
\widetilde G(x,y)=0,&x\in\partial\mathcal{S}_\delta.
\end{cases} { − Δ G ( x , y ) + G ( x , y ) = δ y ( x ) , G ( x , y ) = 0 , x ∈ S δ , x ∈ ∂ S δ .
We next define the Poisson kernel associated with the Green’s function G ~ ( x , y ) \widetilde G(x,y) G ( x , y ) by
P ( x , y ) : = − ∂ ν y G ~ ( x , y ) , y ∈ ∂ S δ , P(x,y):=-\partial_{\nu_y}\widetilde G(x,y),\qquad y\in\partial\mathcal{S}_\delta, P ( x , y ) := − ∂ ν y G ( x , y ) , y ∈ ∂ S δ ,
where ν y \nu_y ν y denotes the unit outward normal to ∂ S δ \partial\mathcal{S}_\delta ∂ S δ . By the Poisson integral formula, we obtain
( U i ) b c ( x ) = ∫ ∂ S δ P ( x , y ) U i ( y ) d S = − ∫ R ∂ y 2 G ~ ( x , ( y 1 , 1 δ ) ) U i ( y 1 , 1 δ ) d y 1 + ∫ R ∂ y 2 G ~ ( x , ( y 1 , − 1 δ ) ) U i ( y 1 , − 1 δ ) d y 1 . \begin{align}
&(U_i)_{\mathrm{bc}}(x)=\int_{\partial\mathcal{S}_\delta}P(x,y)
U_i(y)\,dS \tag*{(2.24)}\\
&=-\int_{ {\mathbb{R}}}\partial_{y_2}\widetilde G\left(x,\left(y_1,\frac{1}{\delta}\right)\right)U_i\left(y_1,\frac{1}{\delta}\right)\,dy_1+\int_{ {\mathbb{R}}}\partial_{y_2}\widetilde G\left(x,\left(y_1,-\frac{1}{\delta}\right)\right)U_i\left(y_1,-\frac{1}{\delta}\right)\,dy_1.\notag
\end{align} ( U i ) bc ( x ) = ∫ ∂ S δ P ( x , y ) U i ( y ) d S = − ∫ R ∂ y 2 G ( x , ( y 1 , δ 1 ) ) U i ( y 1 , δ 1 ) d y 1 + ∫ R ∂ y 2 G ( x , ( y 1 , − δ 1 ) ) U i ( y 1 , − δ 1 ) d y 1 . (2.24)
We proceed to derive some estimates for ( U i ) b c (U_i)_{\mathrm{bc}} ( U i ) bc that will be used later.
Corollary. For any ∣ τ i ∣ , ∣ τ j ∣ < 1 − ε |\tau_i|,\,|\tau_j|< 1-\varepsilon ∣ τ i ∣ , ∣ τ j ∣ < 1 − ε , U ( ⋅ ; τ i ) b c U(\,\cdot\,;\tau_i)_\mathrm{bc} U ( ⋅ ; τ i ) bc satisfy ∥ U ( ⋅ ; τ i ) b c ∥ H k 0 + 2 ( S δ ) ≲ e − 1 − ∣ τ i ∣ δ , ∣ ( U ( ⋅ ; τ i ) b c , U ( ⋅ ; τ j ) ) L 2 ( S δ ) ∣ ≲ δ − 5 2 e − 2 − ∣ τ i ∣ − ∣ τ j ∣ δ , ∣ ( ∂ x 2 U ( ⋅ ; τ i ) b c , U ( ⋅ ; τ j ) ) L 2 ( S δ ) ∣ ≲ δ − 5 2 e − 2 − ∣ τ i ∣ − ∣ τ j ∣ δ . \begin{aligned}
\left\|U(\,\cdot\,;\tau_i)_\mathrm{bc}\right\|_{H^{k_0+2} (\mathcal{S}_\delta)}&\lesssim{\mathrm e}^{-\frac{1-|\tau_i|}\delta},\\
\left|\left(U(\,\cdot\,;\tau_i)_\mathrm{bc},U(\,\cdot\,;\tau_j)\right)_{L^2(\mathcal{S}_\delta)}\right|&\lesssim\delta^{-\frac{5}{2}}{\mathrm e}^{-\frac{2-|\tau_i|-|\tau_j|}{\delta}},\\
\left|\left(\partial_{x_2}U(\,\cdot\,;\tau_i)_\mathrm{bc},U(\,\cdot\,;\tau_j)\right)_{L^2(\mathcal{S}_\delta)}\right|&\lesssim\delta^{-\frac{5}{2}}{\mathrm e}^{-\frac{2-|\tau_i|-|\tau_j|}{\delta}}.
\end{aligned} ∥ U ( ⋅ ; τ i ) bc ∥ H k 0 + 2 ( S δ ) ( U ( ⋅ ; τ i ) bc , U ( ⋅ ; τ j ) ) L 2 ( S δ ) ( ∂ x 2 U ( ⋅ ; τ i ) bc , U ( ⋅ ; τ j ) ) L 2 ( S δ ) ≲ e − δ 1 − ∣ τ i ∣ , ≲ δ − 2 5 e − δ 2 − ∣ τ i ∣ − ∣ τ j ∣ , ≲ δ − 2 5 e − δ 2 − ∣ τ i ∣ − ∣ τ j ∣ . Here and in what follows, ( f , g ) L 2 \left(f,g\right)_{L^2} ( f , g ) L 2 denotes the L 2 L^2 L 2 inner product of f f f and g g g .
Proof. By (2.11) and Lemma 2.1 , a direct estimate yields
∣ U ( ⋅ ; τ i ) b c ∣ H k 0 + 2 ( S δ ) 2 ≲ ∣ U ( ⋅ ; τ i ) ∣ H k 0 + 2 ( ∂ S δ ) 2 ≲ ∫ R ∣ U ( x , 1 δ − τ i δ ) ∣ 2 d x + ∫ R ∣ U ( x , − 1 δ − τ i δ ) ∣ 2 d x ≲ δ ∫ { ∣ x ∣ ≤ 1 − τ i δ } e − 2 x 2 + ( 1 δ − τ i δ ) 2 d x + δ ∫ { ∣ x ∣ ≤ 1 + τ i δ } e − 2 x 2 + ( 1 δ + τ i δ ) 2 d x + δ ∫ { ∣ x ∣ > 1 − τ i δ } e − 2 x 2 + ( 1 δ − τ i δ ) 2 d x + δ ∫ { ∣ x ∣ > 1 + τ i δ } e − 2 x 2 + ( 1 δ + τ i δ ) 2 d x ≲ δ ∫ { ∣ x ∣ ≤ 1 − τ i δ } e − 2 − 2 τ i δ d x + δ ∫ { ∣ x ∣ ≤ 1 + τ i δ } e − 2 + 2 τ i δ d x + δ ∫ 1 − τ i δ ∞ e − 2 x d x + δ ∫ 1 + τ i δ ∞ e − 2 x d x ≲ e − 2 − 2 τ i δ + e − 2 + 2 τ i δ ≲ e − 2 − 2 ∣ τ i ∣ δ . \begin{align*}
&\left| U(\,\cdot\,;\tau_i)_\mathrm{bc}\right|^2_{H^{k_0+2} (\mathcal{S}_\delta)}\lesssim\left|U(\,\cdot\,;\tau_i)\right|^2_{H^{k_0+2} (\partial\mathcal{S}_\delta)}\\
&\lesssim\displaystyle\int_{\mathbb R}\left|U\left(x,\frac{1}{\delta}-\frac{\tau_i}{\delta}\right)\right|^2\,dx+\displaystyle\int_{\mathbb R}\left|U\left(x,-\frac{1}{\delta}-\frac{\tau_i}{\delta}\right)\right|^2\,dx\\
&\lesssim\delta\displaystyle\int_{\left\{|x|\leq\frac{1-\tau_i}{\delta}\right\}}{\mathrm e}^{-2\sqrt{x^2+(\frac{1}{\delta}-\frac{\tau_i}{\delta})^2}}\,dx+\delta\displaystyle\int_{\left\{|x|\leq\frac{1+\tau_i}{\delta}\right\}}{\mathrm e}^{-2\sqrt{x^2+(\frac{1}{\delta}+\frac{\tau_i}{\delta})^2}}\,dx\\
&\quad\,+\delta\displaystyle\int_{\left\{|x|>\frac{1-\tau_i}{\delta}\right\}}{\mathrm e}^{-2\sqrt{x^2+(\frac{1}{\delta}-\frac{\tau_i}{\delta})^2}}\,dx+\delta\displaystyle\int_{\left\{|x|>\frac{1+\tau_i}{\delta}\right\}}{\mathrm e}^{-2\sqrt{x^2+(\frac{1}{\delta}+\frac{\tau_i}{\delta})^2}}\,dx\\
&\lesssim\delta\displaystyle\int_{\left\{|x|\leq\frac{1-\tau_i}{\delta}\right\}}{\mathrm e}^{-\frac{2-2\tau_i}{\delta}}\,dx+\delta\displaystyle\int_{\left\{|x|\leq\frac{1+\tau_i}{\delta}\right\}}{\mathrm e}^{-\frac{2+2\tau_i}{\delta}}\,dx+\delta\displaystyle\int_{\frac{1-\tau_i}{\delta}}^{\infty}{\mathrm e}^{-2x}\,dx+\delta\displaystyle\int_{\frac{1+\tau_i}{\delta}}^{\infty}{\mathrm e}^{-2x}\,dx\\
&\lesssim{\mathrm e}^{-\frac{2-2\tau_i}{\delta}}+{\mathrm e}^{-\frac{2+2\tau_i}{\delta}}\lesssim{\mathrm e}^{-\frac{2-2|\tau_i|}{\delta}}.
\end{align*} ∣ U ( ⋅ ; τ i ) bc ∣ H k 0 + 2 ( S δ ) 2 ≲ ∣ U ( ⋅ ; τ i ) ∣ H k 0 + 2 ( ∂ S δ ) 2 ≲ ∫ R U ( x , δ 1 − δ τ i ) 2 d x + ∫ R U ( x , − δ 1 − δ τ i ) 2 d x ≲ δ ∫ { ∣ x ∣ ≤ δ 1 − τ i } e − 2 x 2 + ( δ 1 − δ τ i ) 2 d x + δ ∫ { ∣ x ∣ ≤ δ 1 + τ i } e − 2 x 2 + ( δ 1 + δ τ i ) 2 d x + δ ∫ { ∣ x ∣ > δ 1 − τ i } e − 2 x 2 + ( δ 1 − δ τ i ) 2 d x + δ ∫ { ∣ x ∣ > δ 1 + τ i } e − 2 x 2 + ( δ 1 + δ τ i ) 2 d x ≲ δ ∫ { ∣ x ∣ ≤ δ 1 − τ i } e − δ 2 − 2 τ i d x + δ ∫ { ∣ x ∣ ≤ δ 1 + τ i } e − δ 2 + 2 τ i d x + δ ∫ δ 1 − τ i ∞ e − 2 x d x + δ ∫ δ 1 + τ i ∞ e − 2 x d x ≲ e − δ 2 − 2 τ i + e − δ 2 + 2 τ i ≲ e − δ 2 − 2∣ τ i ∣ . This implies (2.25) . We next consider 2024 . Using (2.24) ,
( U ( ⋅ ; τ i ) b c , U ( ⋅ ; τ j ) ) L 2 ( S δ ) = ∫ S δ ∫ ∂ S δ P ( x , y ) U ( y ; τ i ) U ( x ; τ j ) d y d x . \left(U(\,\cdot\,;\tau_i)_\mathrm{bc},U(\,\cdot\,;\tau_j)\right)_{L^2(\mathcal{S}_\delta)}=\displaystyle\int_{\mathcal{S}_\delta}\displaystyle\int_{\partial\mathcal{S}_\delta}P(x,y)U(y\,;\tau_i)U(x\,;\tau_j)\,dy\,dx. ( U ( ⋅ ; τ i ) bc , U ( ⋅ ; τ j ) ) L 2 ( S δ ) = ∫ S δ ∫ ∂ S δ P ( x , y ) U ( y ; τ i ) U ( x ; τ j ) d y d x . By Fubini’s theorem, (2.22) and Lemma 2.1 ,
∣ ∫ S δ ∫ ∂ S δ P ( x , y ) U ( y ; τ i ) U ( x ; τ j ) d y d x ∣ ≲ ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ S δ ∣ P ( x , y ) ∣ ∣ U ( x ; τ j ) ∣ d x ≲ ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ( ∫ { ∣ x − y ∣ ≤ μ } ∩ S δ + ∫ { ∣ x − y ∣ > μ } ∩ S δ ) ∣ P ( x , y ) ∣ ∣ U ( x ; τ j ) ∣ d x ≲ ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ { ∣ x − y ∣ ≤ μ } ∩ S δ ∣ x − y ∣ − 1 ∣ x − τ j δ e 2 ∣ − 1 2 e − ∣ x − τ j δ e 2 ∣ d x + ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ { ∣ x − y ∣ > μ } ∩ S δ ( 1 + ∣ x − y ∣ ) − 1 2 ( 1 + ∣ x − τ j δ e 2 ∣ ) − 1 2 e − ∣ x − y ∣ − ∣ x − τ j δ e 2 ∣ d x = : I 1 + I 2 , \begin{align*}
&\left|\displaystyle\int_{\mathcal{S}_\delta}\displaystyle\int_{\partial\mathcal{S}_\delta}P(x,y)U(y\,;\tau_i)U(x\,;\tau_j)\,dy\,dx\right|\lesssim\displaystyle\int_{\partial\mathcal{S}_\delta}|U(y\,;\tau_i)|\,dy\displaystyle\int_{\mathcal{S}_\delta}|P(x,y)||U(x\,;\tau_j)|\,dx\\
&\lesssim\displaystyle\int_{\partial\mathcal{S}_\delta}|U(y\,;\tau_i)|\,dy\left(\displaystyle\int_{\{|x-y|\leq\mu\}\cap\mathcal{S}_\delta}+\displaystyle\int_{\{|x-y|>\mu\}\cap\mathcal{S}_\delta}\right)|P(x,y)||U(x\,;\tau_j)|\,dx\\
&\lesssim\displaystyle\int_{\partial\mathcal{S}_\delta}|U(y\,;\tau_i)|\,dy\displaystyle\int_{\{|x-y|\leq\mu\}\cap\mathcal{S}_\delta}\left|x-y\right|^{-1}\left|x-\tfrac{\tau_j}{\delta}e_2\right|^{-\frac{1}{2}}{\mathrm e}^{-\left|x-\frac{\tau_j}{\delta}e_2\right|}\,dx\\
&\quad\,+\displaystyle\int_{\partial\mathcal{S}_\delta}|U(y\,;\tau_i)|\,dy\displaystyle\int_{\{|x-y|>\mu\}\cap\mathcal{S}_\delta}\left(1+\left|x-y\right|\right)^{-\frac{1}{2}}\left(1+\left|x-\tfrac{\tau_j}{\delta}e_2\right|\right)^{-\frac{1}{2}}{\mathrm e}^{-|x-y|-\left|x-\frac{\tau_j}{\delta}e_2\right|}\,dx\\
&=:I_1+I_2,
\end{align*} ∫ S δ ∫ ∂ S δ P ( x , y ) U ( y ; τ i ) U ( x ; τ j ) d y d x ≲ ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ S δ ∣ P ( x , y ) ∣∣ U ( x ; τ j ) ∣ d x ≲ ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ( ∫ { ∣ x − y ∣ ≤ μ } ∩ S δ + ∫ { ∣ x − y ∣ > μ } ∩ S δ ) ∣ P ( x , y ) ∣∣ U ( x ; τ j ) ∣ d x ≲ ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ { ∣ x − y ∣ ≤ μ } ∩ S δ ∣ x − y ∣ − 1 x − δ τ j e 2 − 2 1 e − ∣ x − δ τ j e 2 ∣ d x + ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ { ∣ x − y ∣ > μ } ∩ S δ ( 1 + ∣ x − y ∣ ) − 2 1 ( 1 + x − δ τ j e 2 ) − 2 1 e − ∣ x − y ∣ − ∣ x − δ τ j e 2 ∣ d x =: I 1 + I 2 , where μ \mu μ is a fixed small constant.
For I 1 I_1 I 1 , by Lemma 2.1 , it follows that
∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ { ∣ x − y ∣ ≤ μ } ∩ S δ ∣ x − y ∣ − 1 ∣ x − τ j δ e 2 ∣ − 1 2 e − ∣ x − τ j δ e 2 ∣ d x ≲ ∫ ∂ S δ ∣ y − τ i δ e 2 ∣ − 1 2 e − ∣ y − τ i δ e 2 ∣ ( ∣ y − τ j δ e 2 ∣ − μ ) − 1 2 e − ∣ y − τ j δ e 2 ∣ + μ d y ≲ δ ∫ ∂ S δ e − ∣ y − τ i δ e 2 ∣ e − ∣ y − τ j δ e 2 ∣ d y = δ ∫ R e − y 1 2 + ( 1 δ − τ i δ ) 2 e − y 1 2 + ( 1 δ − τ j δ ) 2 d y 1 + δ ∫ R e − y 1 2 + ( 1 δ + τ i δ ) 2 e − y 1 2 + ( 1 δ + τ j δ ) 2 d y 1 ≲ δ ∫ − 2 δ 2 δ e − y 1 2 + ( 1 δ − τ i δ ) 2 e − y 1 2 + ( 1 δ − τ j δ ) 2 d y 1 + δ ∫ − 2 δ 2 δ e − y 1 2 + ( 1 δ + τ i δ ) 2 e − y 1 2 + ( 1 δ + τ j δ ) 2 d y 1 + δ ∫ 2 δ ∞ e − 2 y 1 d y 1 ≲ e − 2 − τ i − τ j δ + e − 2 + τ i + τ j δ + δ e − 4 δ ≲ e − 2 − ∣ τ i ∣ − ∣ τ j ∣ δ . \begin{align*}
&\displaystyle\int_{\partial\mathcal{S}_\delta}|U(y\,;\tau_i)|\,dy\displaystyle\int_{\{|x-y|\leq\mu\}\cap\mathcal{S}_\delta}\left|x-y\right|^{-1}\left|x-\tfrac{\tau_j}{\delta}e_2\right|^{-\frac{1}{2}}{\mathrm e}^{-\left|x-\frac{\tau_j}{\delta}e_2\right|}\,dx\\
&\lesssim\displaystyle\int_{\partial\mathcal{S}_\delta}\left|y-\tfrac{\tau_i}{\delta}e_2\right|^{-\frac{1}{2}}{\mathrm e}^{-\left|y-\frac{\tau_i}{\delta}e_2\right|}\left(\left|y-\tfrac{\tau_j}{\delta}e_2\right|-\mu\right)^{-\frac{1}{2}}{\mathrm e}^{-\left|y-\frac{\tau_j}{\delta}e_2\right|+\mu}\,dy\\
&\lesssim\delta\displaystyle\int_{\partial\mathcal{S}_\delta}{\mathrm e}^{-\left|y-\frac{\tau_i}{\delta}e_2\right|}{\mathrm e}^{-\left|y-\frac{\tau_j}{\delta}e_2\right|}\,dy\\
&=\delta\displaystyle\int_{ {\mathbb{R}}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}-\frac{\tau_i}{\delta})^2}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}-\frac{\tau_j}{\delta})^2}}\,dy_1+\delta\displaystyle\int_{ {\mathbb{R}}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}+\frac{\tau_i}{\delta})^2}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}+\frac{\tau_j}{\delta})^2}}\,dy_1\\
&\lesssim\delta\displaystyle\int_{-\frac{2}{\delta}}^{\frac{2}{\delta}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}-\frac{\tau_i}{\delta})^2}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}-\frac{\tau_j}{\delta})^2}}\,dy_1+\delta\displaystyle\int_{-\frac{2}{\delta}}^{\frac{2}{\delta}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}+\frac{\tau_i}{\delta})^2}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}+\frac{\tau_j}{\delta})^2}}\,dy_1\\
&\quad\,+\delta\displaystyle\int_{\frac{2}{\delta}}^{\infty}{\mathrm e}^{-2y_1}\,dy_1\\
&\lesssim{\mathrm e}^{-\frac{2-\tau_i-\tau_j}{\delta}}+{\mathrm e}^{-\frac{2+\tau_i+\tau_j}{\delta}}+\delta{\mathrm e}^{-\frac{4}{\delta}}\lesssim{\mathrm e}^{-\frac{2-|\tau_i|-|\tau_j|}{\delta}}.
\end{align*} ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ { ∣ x − y ∣ ≤ μ } ∩ S δ ∣ x − y ∣ − 1 x − δ τ j e 2 − 2 1 e − ∣ x − δ τ j e 2 ∣ d x ≲ ∫ ∂ S δ y − δ τ i e 2 − 2 1 e − ∣ y − δ τ i e 2 ∣ ( y − δ τ j e 2 − μ ) − 2 1 e − ∣ y − δ τ j e 2 ∣ + μ d y ≲ δ ∫ ∂ S δ e − ∣ y − δ τ i e 2 ∣ e − ∣ y − δ τ j e 2 ∣ d y = δ ∫ R e − y 1 2 + ( δ 1 − δ τ i ) 2 e − y 1 2 + ( δ 1 − δ τ j ) 2 d y 1 + δ ∫ R e − y 1 2 + ( δ 1 + δ τ i ) 2 e − y 1 2 + ( δ 1 + δ τ j ) 2 d y 1 ≲ δ ∫ − δ 2 δ 2 e − y 1 2 + ( δ 1 − δ τ i ) 2 e − y 1 2 + ( δ 1 − δ τ j ) 2 d y 1 + δ ∫ − δ 2 δ 2 e − y 1 2 + ( δ 1 + δ τ i ) 2 e − y 1 2 + ( δ 1 + δ τ j ) 2 d y 1 + δ ∫ δ 2 ∞ e − 2 y 1 d y 1 ≲ e − δ 2 − τ i − τ j + e − δ 2 + τ i + τ j + δ e − δ 4 ≲ e − δ 2 − ∣ τ i ∣ − ∣ τ j ∣ . For I 2 I_2 I 2 , using ∣ x − y ∣ + ∣ x − τ j δ e 2 ∣ ≥ ∣ y − τ j δ e 2 ∣ |x-y|+\left|x-\frac{\tau_j}{\delta}e_2\right|\geq|y-\frac{\tau_j}{\delta}e_2| ∣ x − y ∣ + x − δ τ j e 2 ≥ ∣ y − δ τ j e 2 ∣ , we have
∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ { ∣ x − y ∣ > μ } ∩ S δ ( 1 + ∣ x − y ∣ ) − 1 2 ( 1 + ∣ x − τ j δ e 2 ∣ ) − 1 2 e − ∣ x − y ∣ − ∣ x − τ j δ e 2 ∣ d x ≲ ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ − 4 δ 4 δ ∫ − 1 δ 1 δ e − ∣ y − τ j δ e 2 ∣ d x 2 d x 1 + ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ { ∣ x 1 ∣ > 4 δ } ∩ S δ e − 4 δ − ∣ x − y ∣ d x ≲ δ − 3 2 ∫ R e − y 1 2 + ( 1 δ − τ i δ ) 2 e − y 1 2 + ( 1 δ − τ j δ ) 2 d y 1 + δ − 3 2 ∫ R e − y 1 2 + ( 1 δ + τ i δ ) 2 e − y 1 2 + ( 1 δ + τ j δ ) 2 d y 1 + δ 1 2 e − 4 δ ∫ R e − y 1 2 + ( 1 δ − τ i δ ) 2 d y 1 + δ 1 2 e − 4 δ ∫ R e − y 1 2 + ( 1 δ + τ i δ ) 2 d y 1 ≲ δ − 5 2 e − 2 − τ i − τ j δ + δ − 5 2 e − 2 + τ i + τ j δ + δ − 1 2 e − 5 − τ i δ + δ − 1 2 e − 5 + τ i δ ≲ δ − 5 2 e − 2 − ∣ τ i ∣ − ∣ τ j ∣ δ . \begin{align*}
&\displaystyle\int_{\partial\mathcal{S}_\delta}|U(y\,;\tau_i)|\,dy\displaystyle\int_{\{|x-y|>\mu\}\cap\mathcal{S}_\delta}\left(1+\left|x-y\right|\right)^{-\frac{1}{2}}\left(1+\left|x-\tfrac{\tau_j}{\delta}e_2\right|\right)^{-\frac{1}{2}}{\mathrm e}^{-|x-y|-\left|x-\frac{\tau_j}{\delta}e_2\right|}\,dx\\
&\lesssim\displaystyle\int_{\partial\mathcal{S}_\delta}|U(y\,;\tau_i)|\,dy\displaystyle\int_{-\frac{4}{\delta}}^{\frac{4}{\delta}}\displaystyle\int_{-\frac{1}{\delta}}^{\frac{1}{\delta}}{\mathrm e}^{-\left|y-\frac{\tau_j}{\delta}e_2\right|}\,dx_2\,dx_1\\
&\quad\,+\displaystyle\int_{\partial\mathcal{S}_\delta}|U(y\,;\tau_i)|\,dy\displaystyle\int_{\left\{|x_1|>\frac{4}{\delta}\right\}\cap\mathcal{S}_\delta}{\mathrm e}^{-\frac{4}{\delta}-\left|x-y\right|}\,dx\\
&\lesssim\delta^{-\frac{3}{2}}\displaystyle\int_{ {\mathbb{R}}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}-\frac{\tau_i}{\delta})^2}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}-\frac{\tau_j}{\delta})^2}}\,dy_1+\delta^{-\frac{3}{2}}\displaystyle\int_{ {\mathbb{R}}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}+\frac{\tau_i}{\delta})^2}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}+\frac{\tau_j}{\delta})^2}}\,dy_1\\
&\quad\,+\delta^{\frac{1}{2}}{\mathrm e}^{-\frac{4}{\delta}}\displaystyle\int_{ {\mathbb{R}}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}-\frac{\tau_i}{\delta})^2}}\,dy_1+\delta^{\frac{1}{2}}{\mathrm e}^{-\frac{4}{\delta}}\displaystyle\int_{ {\mathbb{R}}}{\mathrm e}^{-\sqrt{y_1^2+(\frac{1}{\delta}+\frac{\tau_i}{\delta})^2}}\,dy_1\\
&\lesssim\delta^{-\frac{5}{2}}{\mathrm e}^{-\frac{2-\tau_i-\tau_j}{\delta}}+\delta^{-\frac{5}{2}}{\mathrm e}^{-\frac{2+\tau_i+\tau_j}{\delta}}+\delta^{-\frac{1}{2}}{\mathrm e}^{-\frac{5-\tau_i}{\delta}}+\delta^{-\frac{1}{2}}{\mathrm e}^{-\frac{5+\tau_i}{\delta}}\lesssim\delta^{-\frac{5}{2}}{\mathrm e}^{-\frac{2-|\tau_i|-|\tau_j|}{\delta}}.
\end{align*} ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ { ∣ x − y ∣ > μ } ∩ S δ ( 1 + ∣ x − y ∣ ) − 2 1 ( 1 + x − δ τ j e 2 ) − 2 1 e − ∣ x − y ∣ − ∣ x − δ τ j e 2 ∣ d x ≲ ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ − δ 4 δ 4 ∫ − δ 1 δ 1 e − ∣ y − δ τ j e 2 ∣ d x 2 d x 1 + ∫ ∂ S δ ∣ U ( y ; τ i ) ∣ d y ∫ { ∣ x 1 ∣ > δ 4 } ∩ S δ e − δ 4 − ∣ x − y ∣ d x ≲ δ − 2 3 ∫ R e − y 1 2 + ( δ 1 − δ τ i ) 2 e − y 1 2 + ( δ 1 − δ τ j ) 2 d y 1 + δ − 2 3 ∫ R e − y 1 2 + ( δ 1 + δ τ i ) 2 e − y 1 2 + ( δ 1 + δ τ j ) 2 d y 1 + δ 2 1 e − δ 4 ∫ R e − y 1 2 + ( δ 1 − δ τ i ) 2 d y 1 + δ 2 1 e − δ 4 ∫ R e − y 1 2 + ( δ 1 + δ τ i ) 2 d y 1 ≲ δ − 2 5 e − δ 2 − τ i − τ j + δ − 2 5 e − δ 2 + τ i + τ j + δ − 2 1 e − δ 5 − τ i + δ − 2 1 e − δ 5 + τ i ≲ δ − 2 5 e − δ 2 − ∣ τ i ∣ − ∣ τ j ∣ . Summarizing the estimates for I 1 I_1 I 1 and I 2 I_2 I 2 gives 2024 .
Finally, consider 2025 . Integration by parts gives
∣ ( ∂ x 2 U ( ⋅ ; τ i ) b c , U ( ⋅ ; τ j ) ) L 2 ( S δ ) ∣ = ∣ ∫ S δ ∂ x 2 U ( x ; τ i ) b c U ( x ; τ j ) d x ∣ ≤ ∣ ∫ R ( U i ) b c ( x 1 , 1 δ ) U j ( x 1 , 1 δ ) d x 1 ∣ + ∣ ∫ R ( U i ) b c ( x 1 , − 1 δ ) U j ( x 1 , − 1 δ ) d x 1 ∣ + ∣ ∫ S δ ( U i ) b c ( x ) ∂ x 2 U j ( x ) d x ∣ = ∣ ∫ R U i ( x 1 , 1 δ ) U j ( x 1 , 1 δ ) d x 1 ∣ + ∣ ∫ R U i ( x 1 , − 1 δ ) U j ( x 1 , − 1 δ ) d x 1 ∣ + ∣ ∫ S δ ( U i ) b c ( x ) ∂ x 2 U j ( x ) d x ∣ = : J 1 + J 2 + J 3 , \begin{align*}
&\left|\left(\partial_{x_2}U(\,\cdot\,;\tau_i)_\mathrm{bc},U(\,\cdot\,;\tau_j)\right)_{L^2(\mathcal{S}_\delta)}\right|=\left|\displaystyle\int_{\mathcal{S}_\delta}\partial_{x_2}U(x\,;\tau_i)_{\mathrm{bc}}U(x\,;\tau_j)\,dx\right|\\
&\leq\left|\int_{ {\mathbb{R}}}(U_i)_{\mathrm{bc}}\left(x_1,\frac{1}{\delta}\right)U_j\left(x_1,\frac{1}\delta\right)\,dx_1\right|+\left|\int_{ {\mathbb{R}}}(U_i)_{\mathrm{bc}}\left(x_1,-\frac{1}{\delta}\right)U_j\left(x_1,-\frac{1}\delta\right)\,dx_1\right|\\
&\quad\,+\left|\displaystyle\int_{\mathcal{S}_\delta}(U_i)_{\mathrm{bc}}(x)\partial_{x_2}U_j(x)\,dx\right|\\
&=\left|\int_{ {\mathbb{R}}}U_i\left(x_1,\frac{1}{\delta}\right)U_j\left(x_1,\frac{1}\delta\right)\,dx_1\right|+\left|\int_{ {\mathbb{R}}}U_i\left(x_1,-\frac{1}{\delta}\right)U_j\left(x_1,-\frac{1}\delta\right)\,dx_1\right|\\
&\quad\,+\left|\displaystyle\int_{\mathcal{S}_\delta}(U_i)_{\mathrm{bc}}(x)\partial_{x_2}U_j(x)\,dx\right|\\
&=:J_1+J_2+J_3,
\end{align*} ( ∂ x 2 U ( ⋅ ; τ i ) bc , U ( ⋅ ; τ j ) ) L 2 ( S δ ) = ∫ S δ ∂ x 2 U ( x ; τ i ) bc U ( x ; τ j ) d x ≤ ∫ R ( U i ) bc ( x 1 , δ 1 ) U j ( x 1 , δ 1 ) d x 1 + ∫ R ( U i ) bc ( x 1 , − δ 1 ) U j ( x 1 , − δ 1 ) d x 1 + ∫ S δ ( U i ) bc ( x ) ∂ x 2 U j ( x ) d x = ∫ R U i ( x 1 , δ 1 ) U j ( x 1 , δ 1 ) d x 1 + ∫ R U i ( x 1 , − δ 1 ) U j ( x 1 , − δ 1 ) d x 1 + ∫ S δ ( U i ) bc ( x ) ∂ x 2 U j ( x ) d x =: J 1 + J 2 + J 3 , where the boundary condition in (2.21) is used.
For J 1 J_1 J 1 , when ∣ x 1 ∣ ≤ 4 δ |x_1|\leq\frac 4\delta ∣ x 1 ∣ ≤ δ 4 , the following inequality holds
x 1 2 + ( 1 − τ i δ ) 2 − ∣ 1 − τ i δ ∣ = x 1 2 x 1 2 + ( 1 − τ i δ ) 2 + ∣ 1 − τ i δ ∣ ≥ δ x 1 2 2 ( 5 + 1 ) . \begin{align*}
\sqrt{x_1^2+\left(\frac {1-\tau_i}\delta\right)^2}-\left|\frac {1-\tau_i}\delta\right|=\frac{x_1^2}{\sqrt{x_1^2+\left(\frac {1-\tau_i}\delta\right)^2}+\left|\frac {1-\tau_i}\delta\right|}\geq\frac{\delta x_1^2}{2(\sqrt{5}+1)}.
\end{align*} x 1 2 + ( δ 1 − τ i ) 2 − δ 1 − τ i = x 1 2 + ( δ 1 − τ i ) 2 + δ 1 − τ i x 1 2 ≥ 2 ( 5 + 1 ) δ x 1 2 . Therefore, by Lemma 2.1 ,
J 1 ≲ ∫ − 4 δ 4 δ ( x 1 2 + ( 1 − τ i δ ) 2 ) − 1 4 ( x 1 2 + ( 1 − τ j δ ) 2 ) − 1 4 e − ( x 1 2 + ( 1 − τ i δ ) 2 + x 1 2 + ( 1 − τ j δ ) 2 ) d x 1 + ∫ 4 δ ∞ ( x 1 2 + ( 1 − τ i δ ) 2 ) − 1 4 ( x 1 2 + ( 1 − τ j δ ) 2 ) − 1 4 e − 2 x 1 d x 1 ≲ δ ∫ − 4 δ 4 δ e − ( ∣ 1 − τ i δ ∣ + ∣ 1 − τ j δ ∣ + δ x 1 2 5 + 1 ) d x + δ e − 8 δ ≲ δ 1 2 e − 2 − ∣ τ i ∣ − ∣ τ j ∣ δ . \begin{align*}
J_1&\lesssim\displaystyle\int_{-\frac 4\delta}^{\frac{4}{\delta}}\left(x_1^2+\left(\tfrac{1-\tau_i}\delta\right)^2\right)^{-\frac{1}{4}}\left(x_1^2+\left(\tfrac{1-\tau_j}\delta\right)^2\right)^{-\frac{1}{4}}{\mathrm e}^{-\left(\sqrt{x_1^2+\left(\frac {1-\tau_i}\delta\right)^2}+\sqrt{x_1^2+\left(\frac {1-\tau_j}\delta\right)^2}\right)}\,dx_1\\
&\quad\,+\displaystyle\int_{\frac{4}{\delta}}^{\infty}\left(x_1^2+\left(\tfrac {1-\tau_i}\delta\right)^2\right)^{-\frac{1}{4}} \left(x_1^2+\left(\tfrac {1-\tau_j}\delta\right)^2\right)^{-\frac{1}{4}} {\mathrm e}^{-2x_1}\,dx_1\\
&\lesssim\delta\displaystyle\int_{-\frac 4\delta}^{\frac{4}{\delta}}{\mathrm e}^{-\left(\left|\frac {1-\tau_i}\delta\right|+\left|\frac {1-\tau_j}\delta\right|+\frac{\delta x_1^2}{\sqrt{5}+1}\right)}\,dx+\delta{\mathrm e}^{-\frac{8}{\delta}}\lesssim\delta^{\frac{1}{2}}{\mathrm e}^{-\frac{2-|\tau_i|-|\tau_j|}{\delta}}.
\end{align*} J 1 ≲ ∫ − δ 4 δ 4 ( x 1 2 + ( δ 1 − τ i ) 2 ) − 4 1 ( x 1 2 + ( δ 1 − τ j ) 2 ) − 4 1 e − ( x 1 2 + ( δ 1 − τ i ) 2 + x 1 2 + ( δ 1 − τ j ) 2 ) d x 1 + ∫ δ 4 ∞ ( x 1 2 + ( δ 1 − τ i ) 2 ) − 4 1 ( x 1 2 + ( δ 1 − τ j ) 2 ) − 4 1 e − 2 x 1 d x 1 ≲ δ ∫ − δ 4 δ 4 e − ( ∣ δ 1 − τ i ∣ + δ 1 − τ j + 5 + 1 δ x 1 2 ) d x + δ e − δ 8 ≲ δ 2 1 e − δ 2 − ∣ τ i ∣ − ∣ τ j ∣ . Similarly,
J 2 ≲ δ 1 2 e − 2 − ∣ τ i ∣ − ∣ τ j ∣ δ . J_2\lesssim\delta^{\frac{1}{2}}{\mathrm e}^{-\frac{2-|\tau_i|-|\tau_j|}{\delta}}. J 2 ≲ δ 2 1 e − δ 2 − ∣ τ i ∣ − ∣ τ j ∣ . For J 3 J_3 J 3 , by Lemma 2.1 , the decay estimates for ∂ x 2 U j \partial_{x_2}U_j ∂ x 2 U j and U j U_j U j are the same, then we have
J 3 ≲ ∣ ( U ( ⋅ ; τ i ) b c , U ( ⋅ ; τ j ) ) L 2 ( S δ ) ∣ ≲ δ − 5 2 e − 2 − ∣ τ i ∣ − ∣ τ j ∣ δ . J_3\lesssim\left|\left(U(\,\cdot\,;\tau_i)_\mathrm{bc},U(\,\cdot\,;\tau_j)\right)_{L^2(\mathcal{S}_\delta)}\right|\lesssim\delta^{-\frac{5}{2}}{\mathrm e}^{-\frac{2-|\tau_i|-|\tau_j|}{\delta}}. J 3 ≲ ( U ( ⋅ ; τ i ) bc , U ( ⋅ ; τ j ) ) L 2 ( S δ ) ≲ δ − 2 5 e − δ 2 − ∣ τ i ∣ − ∣ τ j ∣ . It follows that 2025 holds. ◻
At the end of this section, we consider the construction of ( U i ) b c (U_i)_{\mathrm{bc}} ( U i ) bc . Since U i U_i U i satisfies Δ U i = γ ( U i ) \Delta U_i=\gamma(U_i) Δ U i = γ ( U i ) and γ ( 0 ) = 0 \gamma(0)=0 γ ( 0 ) = 0 , γ ′ ( 0 ) = 1 \gamma'(0)=1 γ ′ ( 0 ) = 1 , then for U i U_i U i small,
( 1 − Δ ) U i = U i − γ ( U i ) = O ( U i 2 ) . (1-\Delta)U_i=U_i-\gamma(U_i)=O(U_i^2). ( 1 − Δ ) U i = U i − γ ( U i ) = O ( U i 2 ) .
Thus, away from the center of the i i i -th spike where U i U_i U i is exponentially small, satisfies the following homogeneous linear equation to leading order:
( 1 − Δ ) U i = 0. (1-\Delta)U_i=0. ( 1 − Δ ) U i = 0.
This observation suggests that the leading-order behavior of ( U i ) b c (U_i)_{\mathrm{bc}} ( U i ) bc can be described by the corresponding linear boundary-value problem. In particular, the boundary data of U i U_i U i are propagated into the strip through the Poisson kernel, and the leading-order contribution is given by the reflections of U i U_i U i across the upper and lower boundaries.
Denote
( U i ) b c ( x ) = U i ( x 1 , 2 δ − x 2 ) + U i ( x 1 , − 2 δ − x 2 ) + R i ( x ) , \begin{equation}
(U_i)_{\mathrm{bc}}(x)=U_i\left(x_1,\tfrac2\delta-x_2\right)+U_i\left(x_1,-\tfrac2\delta-x_2\right)+\mathcal{R}_i(x), \tag*{(2.26)}
\end{equation} ( U i ) bc ( x ) = U i ( x 1 , δ 2 − x 2 ) + U i ( x 1 , − δ 2 − x 2 ) + R i ( x ) , (2.26)
where R i ( x ) \mathcal{R}_i(x) R i ( x ) satisfies
{ ( 1 − Δ ) R i ( x ) = − ( 1 − Δ ) U i ( x 1 , 2 δ − x 2 ) − ( 1 − Δ ) U i ( x 1 , − 2 δ − x 2 ) , x ∈ S δ , R i ( x 1 , ± 1 δ ) = U i ( x 1 , ∓ 3 δ ) . \begin{cases} (1-\Delta)\mathcal{R}_i(x)=-(1-\Delta)U_i\left(x_1,\frac2\delta-x_2\right)-(1-\Delta)U_i\left(x_1,-\frac2\delta-x_2\right),\quad x\in\mathcal{S}_\delta,\\ \mathcal{R}_i\left(x_1,\pm\frac1\delta\right)=U_i\left(x_1,\mp\frac3\delta\right).
\end{cases} { ( 1 − Δ ) R i ( x ) = − ( 1 − Δ ) U i ( x 1 , δ 2 − x 2 ) − ( 1 − Δ ) U i ( x 1 , − δ 2 − x 2 ) , x ∈ S δ , R i ( x 1 , ± δ 1 ) = U i ( x 1 , ∓ δ 3 ) .
We now consider the estimates for R i \mathcal{R}_i R i . By the Green’s function representation above
R i ( x ) = − ∫ S δ G ~ ( x , y ) ( ( 1 − Δ ) U i ( y 1 , 2 δ − y 2 ) + ( 1 − Δ ) U i ( y 1 , − 2 δ − y 2 ) ) d y + ∫ R ∂ y 2 G ~ ( x , ( y 1 , 1 δ ) ) U i ( y 1 , − 3 δ ) d y 1 − ∫ R ∂ y 2 G ~ ( x , ( y 1 , − 1 δ ) ) U i ( y 1 , 3 δ ) d y 1 . \begin{align*}
\mathcal{R}_i(x)=&-\int_{\mathcal{S}_\delta}\widetilde G(x,y)
\left((1-\Delta)U_i\left(y_1,\tfrac2\delta-y_2\right)+(1-\Delta)U_i\left(y_1,-\tfrac2\delta-y_2\right)\right)\,dy\\
&+\int_{ {\mathbb{R}}}\partial_{y_2}\widetilde G\left(x,\left(y_1,\tfrac{1}{\delta}\right)\right)U_i\left(y_1,-\tfrac{3}{\delta}\right)\,dy_1-\int_{ {\mathbb{R}}}\partial_{y_2}\widetilde G\left(x,\left(y_1,-\tfrac{1}{\delta}\right)\right)U_i\left(y_1,\tfrac{3}{\delta}\right)\,dy_1.
\end{align*} R i ( x ) = − ∫ S δ G ( x , y ) ( ( 1 − Δ ) U i ( y 1 , δ 2 − y 2 ) + ( 1 − Δ ) U i ( y 1 , − δ 2 − y 2 ) ) d y + ∫ R ∂ y 2 G ( x , ( y 1 , δ 1 ) ) U i ( y 1 , − δ 3 ) d y 1 − ∫ R ∂ y 2 G ( x , ( y 1 , − δ 1 ) ) U i ( y 1 , δ 3 ) d y 1 .
Applying the elliptic regularity estimate used in (2.11) and the argument in Corollary (?) for ∥ U ( ⋅ ; τ i ) ∥ H k 0 + 2 ( ∂ S δ ) \left\|U(\,\cdot\,;\tau_i)\right\|_{H^{k_0+2} (\partial\mathcal{S}_\delta)} ∥ U ( ⋅ ; τ i ) ∥ H k 0 + 2 ( ∂ S δ ) , we have
∥ R i ∥ H k 0 + 2 ( S δ ) ≲ ∥ ( 1 − Δ ) U i ( ⋅ , 2 δ − ⋅ ) ∥ H k 0 ( S δ ) + ∥ ( 1 − Δ ) U i ( ⋅ , − 2 δ − ⋅ ) ∥ H k 0 ( S δ ) + ∥ U ( ⋅ , − 3 + τ i δ ) ∥ H k 0 + 2 ( R ) + ∥ U ( ⋅ , 3 − τ i δ ) ∥ H k 0 + 2 ( R ) ≲ ∥ U i 2 ( ⋅ , 2 δ − ⋅ ) ∥ H k 0 ( S δ ) + ∥ U i 2 ( ⋅ , − 2 δ − ⋅ ) ∥ H k 0 ( S δ ) + e − 3 − ∣ τ i ∣ δ . \begin{align*}
\left\|{R}_i\right\|_{H^{k_0+2}(\mathcal{S}_\delta)}&\lesssim\left\|(1-\Delta)U_i\left(\,\cdot\,,\tfrac2\delta-\,\cdot\,\right)\right\|_{H^{k_0} (\mathcal{S}_\delta)}+\left\|(1-\Delta)U_i\left(\,\cdot\,,-\tfrac2\delta-\,\cdot\,\right)\right\|_{H^{k_0} (\mathcal{S}_\delta)}\\
&\quad\,+\left\|U(\,\cdot\,,-\tfrac{3+\tau_i}\delta)\right\|_{H^{k_0+2} ( {\mathbb{R}})}+\left\|U(\,\cdot\,,\tfrac{3-\tau_i}\delta)\right\|_{H^{k_0+2}( {\mathbb{R}})}\\
&\lesssim\left\|U_i^2\left(\,\cdot\,,\tfrac2\delta-\,\cdot\,\right)\right\|_{H^{k_0} (\mathcal{S}_\delta)}+\left\|U_i^2\left(\,\cdot\,,-\tfrac2\delta-\,\cdot\,\right)\right\|_{H^{k_0} (\mathcal{S}_\delta)}+{\mathrm e}^{-\frac{3-|\tau_i|}{\delta}}.
\end{align*} ∥ R i ∥ H k 0 + 2 ( S δ ) ≲ ( 1 − Δ ) U i ( ⋅ , δ 2 − ⋅ ) H k 0 ( S δ ) + ( 1 − Δ ) U i ( ⋅ , − δ 2 − ⋅ ) H k 0 ( S δ ) + U ( ⋅ , − δ 3 + τ i ) H k 0 + 2 ( R ) + U ( ⋅ , δ 3 − τ i ) H k 0 + 2 ( R ) ≲ U i 2 ( ⋅ , δ 2 − ⋅ ) H k 0 ( S δ ) + U i 2 ( ⋅ , − δ 2 − ⋅ ) H k 0 ( S δ ) + e − δ 3 − ∣ τ i ∣ .
For ∣ x 1 ∣ ≤ 2 δ |x_1|\leq\frac2\delta ∣ x 1 ∣ ≤ δ 2 , ∣ x 2 ∣ ≤ 1 δ |x_2|\leq\frac1\delta ∣ x 2 ∣ ≤ δ 1 and ∣ τ i ∣ δ < 1 − ε δ \frac{|\tau_i|}\delta<\frac{1-\varepsilon}{\delta} δ ∣ τ i ∣ < δ 1 − ε ,
∣ x − 2 − τ i δ e 2 ∣ = x 1 2 + ( x 2 − 2 − τ i δ ) 2 ≥ x 1 2 + ( 1 − τ i δ ) 2 . \begin{align*}
\left|x-\frac{2-\tau_i}\delta e_2\right|&=\sqrt{x_1^2+\left(x_2-\frac {2-\tau_i}\delta\right)^2}\geq\sqrt{x_1^2+\left(\frac{1-\tau_i}{\delta}\right)^2}.
\end{align*} x − δ 2 − τ i e 2 = x 1 2 + ( x 2 − δ 2 − τ i ) 2 ≥ x 1 2 + ( δ 1 − τ i ) 2 .
Since
x 1 2 + ( 1 − τ i δ ) 2 − 1 − τ i δ = x 1 2 x 1 2 + ( 1 − τ i δ ) 2 + 1 − τ i δ ≥ δ x 1 2 2 ( 2 + 1 ) , \begin{align*}
\sqrt{x_1^2+\left(\frac{1-\tau_i}{\delta}\right)^2}-\frac{1-\tau_i}{\delta}=\frac{x_1^2}{\sqrt{x_1^2+\left(\frac{1-\tau_i}{\delta}\right)^2}+\frac{1-\tau_i}{\delta}}\geq\frac{\delta x_1^2}{2(\sqrt{2}+1)},
\end{align*} x 1 2 + ( δ 1 − τ i ) 2 − δ 1 − τ i = x 1 2 + ( δ 1 − τ i ) 2 + δ 1 − τ i x 1 2 ≥ 2 ( 2 + 1 ) δ x 1 2 ,
we have
∣ x − 2 − τ i δ e 2 ∣ ≥ 1 − τ i δ + δ x 1 2 2 ( 2 + 1 ) . \left|x-\frac{2-\tau_i}\delta e_2\right|\geq\frac{1-\tau_i}{\delta}+\frac{\delta x_1^2}{2(\sqrt{2}+1)}. x − δ 2 − τ i e 2 ≥ δ 1 − τ i + 2 ( 2 + 1 ) δ x 1 2 .
Then from Lemma 2.1 ,
∣ U i 2 ( ⋅ , 2 δ − ⋅ ) ∣ H k 0 ( S δ ) 2 ≲ ∫ S δ ∣ x − 2 − τ i δ e 2 ∣ − 2 e − 4 ∣ x − 2 − τ i δ e 2 ∣ d x ≲ ∫ { ∣ x 1 ∣ ≤ 2 δ } ∫ − 1 δ 1 δ ( 2 − τ i δ − x 2 ) − 2 e − 4 ( 1 − τ i δ + δ x 1 2 2 ( 2 + 1 ) ) d x 2 d x 1 + 2 ∫ 2 δ ∞ ∫ − 1 δ 1 δ ( 2 − τ i δ − x 2 ) − 2 e − 4 x 1 d x 2 d x 1 ≲ δ 1 2 e − 4 − 4 τ i δ + δ e − 8 δ ≲ δ 1 2 e − 4 − 4 ∣ τ i ∣ δ . \begin{align*}
\left|U_i^2\left(\,\cdot\,,\tfrac2\delta-\,\cdot\,\right)\right|^2_{H^{k_0} (\mathcal{S}_\delta)}&\lesssim\int_{\mathcal{S}_\delta}\left|x-\frac{2-\tau_i}\delta e_2\right|^{-2}{\mathrm e}^{-4\left|x-\frac{2-\tau_i}\delta e_2\right|}\,dx\\
&\lesssim\int_{\left\{|x_1|\leq\frac2\delta\right\}}\int_{-\frac1\delta}^{\frac1\delta}\left(\frac{2-\tau_i}\delta-x_2\right)^{-2}{\mathrm e}^{-4\left(\frac{1-\tau_i}{\delta}+\frac{\delta x_1^2}{2(\sqrt{2}+1)}\right)}\,dx_2\,dx_1\\
&\quad\,+2\int_{\frac2\delta}^{\infty}\int_{-\frac1\delta}^{\frac1\delta}\left(\frac{2-\tau_i}\delta-x_2\right)^{-2}{\mathrm e}^{-4x_1}\,dx_2\,dx_1\\
&\lesssim\delta^{\frac{1}{2}}{\mathrm e}^{-\frac{4-4\tau_i}{\delta}}+\delta{\mathrm e}^{-\frac{8}{\delta}}\lesssim\delta^{\frac{1}{2}}{\mathrm e}^{-\frac{4-4|\tau_i|}{\delta}}.
\end{align*} U i 2 ( ⋅ , δ 2 − ⋅ ) H k 0 ( S δ ) 2 ≲ ∫ S δ x − δ 2 − τ i e 2 − 2 e − 4 ∣ x − δ 2 − τ i e 2 ∣ d x ≲ ∫ { ∣ x 1 ∣ ≤ δ 2 } ∫ − δ 1 δ 1 ( δ 2 − τ i − x 2 ) − 2 e − 4 ( δ 1 − τ i + 2 ( 2 + 1 ) δ x 1 2 ) d x 2 d x 1 + 2 ∫ δ 2 ∞ ∫ − δ 1 δ 1 ( δ 2 − τ i − x 2 ) − 2 e − 4 x 1 d x 2 d x 1 ≲ δ 2 1 e − δ 4 − 4 τ i + δ e − δ 8 ≲ δ 2 1 e − δ 4 − 4∣ τ i ∣ .
Analogously,
∣ U i 2 ( ⋅ , − 2 δ − ⋅ ) ∣ H k 0 ( S δ ) 2 ≲ δ 1 2 e − 4 − 4 ∣ τ i ∣ δ . \left|U_i^2\left(\,\cdot\,,-\tfrac2\delta-\,\cdot\,\right)\right|^2_{H^{k_0} (\mathcal{S}_\delta)}\lesssim\delta^{\frac{1}{2}}{\mathrm e}^{-\frac{4-4|\tau_i|}{\delta}}. U i 2 ( ⋅ , − δ 2 − ⋅ ) H k 0 ( S δ ) 2 ≲ δ 2 1 e − δ 4 − 4∣ τ i ∣ .
Therefore,
∥ R i ∥ H k 0 + 2 ( S δ ) ≲ δ 1 4 e − 2 − 2 ∣ τ i ∣ δ . \begin{equation}
\left\|{R}_i\right\|_{H^{k_0+2}(\mathcal{S}_\delta)}\lesssim\delta^{\frac{1}{4}}{\mathrm e}^{-\frac{2-2|\tau_i|}{\delta}}. \tag*{(2.27)}
\end{equation} ∥ R i ∥ H k 0 + 2 ( S δ ) ≲ δ 4 1 e − δ 2 − 2∣ τ i ∣ . (2.27)
Linear theory
The inner-outer gluing system
Motivated by [10 , 11 ] , we use an inner–outer gluing method to construct the perturbation term u u u in (2.12) . For each concentration point, we consider an inner problem corresponding to the spike U i U_i U i , while the outer problem is mainly concerned with the correction outside the all spikes and the interactions between the spikes.
From (2.12) , we split u u u into an outer part and inner parts:
u ( x ) = ∑ i = 1 n η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ϕ i ( x ) + ψ ( x ) , \begin{align*}
u(x)=\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\phi_i\left(x\right)+\psi(x),
\end{align*} u ( x ) = i = 1 ∑ n η ( R δ x − ( 0 , δ τ i ) ) ϕ i ( x ) + ψ ( x ) ,
where R is a small enough constant and η ( x ) \eta(x) η ( x ) is a smooth cutoff function satisfying
η ( x ) = { 1 , ∣ x ∣ ≤ 1 , 0 , ∣ x ∣ ≥ 2. \eta(x) =
\begin{cases}
1, ~&|x|\leq1,\\
0, ~&|x|\geq2.
\end{cases} η ( x ) = { 1 , 0 , ∣ x ∣ ≤ 1 , ∣ x ∣ ≥ 2.
By perturbed u eq , for i = 1 , … , n i=1,\dots,n i = 1 , … , n , x ∈ B i , 2 : = { x ∈ R 2 ∣ ∣ x − ( 0 , τ i δ ) ∣ ≤ 2 R δ } x\in B_{i,2}:=\left\{x\in {\mathbb{R}}^2\;\left|\;\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\leq \frac{2R}{\delta}\right.\right\} x ∈ B i , 2 := { x ∈ R 2 x − ( 0 , δ τ i ) ≤ δ 2 R } , the inner problem is
( − Δ + γ ′ ( U i ) ) ϕ i + F ( τ , u , Γ s ) + ( γ ′ ( ∑ i = 1 n σ i U i ) − 1 ) ψ = 0 , for i = 1 , … , n . \begin{align}
\left(-\Delta+\gamma^\prime\left(U_i\right)\right)\phi_i + F(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}})+\left(\gamma^\prime\left(\sum_{i=1}^n\sigma_iU_i\right)-1\right)\psi=0,\quad \text{for }i=1,\dots,n. \tag*{(3.1)}
\end{align} ( − Δ + γ ′ ( U i ) ) ϕ i + F ( τ , u , Γ s ) + ( γ ′ ( i = 1 ∑ n σ i U i ) − 1 ) ψ = 0 , for i = 1 , … , n . (3.1)
The outer problem is given by
{ Δ ψ ( x ) − ψ ( x ) = ( 1 − ∑ i = 1 n η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ) ( γ ′ ( ∑ i = 1 n σ i U i ) − 1 ) ψ ( x ) + ( 1 − ∑ i = 1 n η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ) F − 2 δ R ∑ i = 1 n ∇ η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ⋅ ∇ ϕ i ( x ) − δ 2 R 2 ∑ i = 1 n Δ η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ϕ i ( x ) + ∑ i = 1 n η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ( γ ′ ( ∑ i = 1 n σ i U i ) − γ ′ ( U i ) ) ϕ i ( x ) , x ∈ S δ , ψ ( x ) = 0 , x ∈ ∂ S δ . \begin{align}
\begin{cases}
&\Delta\psi(x)-\psi(x)\\
&=\left(1-\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\right)\left(\gamma^\prime\left(\sum_{i=1}^n\sigma_iU_i\right)-1\right)\psi(x)\\[3mm]
&\quad\,+\left(1-\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\right)F\\[3mm]
&\quad\,-\cfrac{2\delta}{R}\sum_{i=1}^n\nabla\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\cdot\nabla\phi_i(x)-\cfrac{\delta^2}{R^2}\sum_{i=1}^n\Delta\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\phi_i(x)\\[3mm]
&\quad\,+\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\left(\gamma^\prime\left(\sum_{i=1}^n\sigma_iU_i\right)-\gamma^\prime\left(U_i\right)\right)\phi_i(x),\quad x\in \mathcal{S}_\delta,\\[4mm]
&\psi(x)=0,\quad x\in \partial\mathcal{S}_\delta.
\end{cases} \tag*{(3.2)}
\end{align} ⎩ ⎨ ⎧ Δ ψ ( x ) − ψ ( x ) = ( 1 − ∑ i = 1 n η ( R δ x − ( 0 , δ τ i ) ) ) ( γ ′ ( ∑ i = 1 n σ i U i ) − 1 ) ψ ( x ) + ( 1 − ∑ i = 1 n η ( R δ x − ( 0 , δ τ i ) ) ) F − R 2 δ ∑ i = 1 n ∇ η ( R δ x − ( 0 , δ τ i ) ) ⋅ ∇ ϕ i ( x ) − R 2 δ 2 ∑ i = 1 n Δ η ( R δ x − ( 0 , δ τ i ) ) ϕ i ( x ) + ∑ i = 1 n η ( R δ x − ( 0 , δ τ i ) ) ( γ ′ ( ∑ i = 1 n σ i U i ) − γ ′ ( U i ) ) ϕ i ( x ) , x ∈ S δ , ψ ( x ) = 0 , x ∈ ∂ S δ . (3.2)
The linear theory of inner problem
In this subsection, we consider the linear operator corresponding to problem (3.1) . Note that no boundary conditions are imposed, so we only need to find a solution ϕ i \phi_i ϕ i defined on S δ \mathcal{S}_\delta S δ for each i i i .
Define
L τ i : = − Δ + γ ′ ( U i ) . L_{\boldsymbol{\tau_i}} := -\Delta + \gamma^\prime\left(U_i\right). L τ i := − Δ + γ ′ ( U i ) .
For given h ∈ L e 2 ( S δ ) h\in L_{\mathrm{e}}^2(\mathcal{S}_\delta) h ∈ L e 2 ( S δ ) , we consider
L τ i [ ϕ i ] = h in S δ . L_{\boldsymbol{\tau_i}}[\phi_i] = h\quad \text{in }\mathcal{S}_\delta. L τ i [ ϕ i ] = h in S δ .
By Assumption assumption:B , the kernel of L τ i L_{\boldsymbol{\tau}_i} L τ i in H e 2 ( R 2 ) H_{\mathrm e}^2(\mathbb R^2) H e 2 ( R 2 ) is spanned by ∂ x 2 U i \partial_{x_2} U_i ∂ x 2 U i , since ∂ x 1 U i {\partial} _{x_1} U_i ∂ x 1 U i is odd in x 1 x_1 x 1 . Let P i P_i P i denote the orthogonal projection of L e 2 ( R 2 ) L_{\textrm{e}}^2(\mathbb{R}^2) L e 2 ( R 2 ) onto s p a n { ∂ x 2 U i } span{\{ {\partial} _{x_2} U_i\}} s p an { ∂ x 2 U i } and we use the same symbol for the projection H e 2 ( R 2 ) H_{\textrm{e}}^2(\mathbb{R}^2) H e 2 ( R 2 ) onto s p a n { ∂ x 2 U i } span{\{ {\partial} _{x_2} U_i\}} s p an { ∂ x 2 U i } . The non-degeneracy result of L τ i L_{\boldsymbol{\tau_i}} L τ i in R 2 \mathbb{R}^2 R 2 implies that
L τ i : ( I − P i ) H e 2 ( R 2 ) ⟶ ( I − P i ) L e 2 ( R 2 ) L_{\boldsymbol{\tau}_i}:(I-P_i)H_{\mathrm e}^2(\mathbb R^2)\longrightarrow(I-P_i)L_{\mathrm e}^2(\mathbb R^2) L τ i : ( I − P i ) H e 2 ( R 2 ) ⟶ ( I − P i ) L e 2 ( R 2 )
is an isomorphism, with uniformly bounded inverse for ∣ τ i ∣ < 1 − ε |\tau_i| < 1-\varepsilon ∣ τ i ∣ < 1 − ε .
Now given a function f ∈ L e 2 ( S δ ) f\in L_{\mathrm{e}}^2(\mathcal{S}_\delta) f ∈ L e 2 ( S δ ) , and define its zero extension E f Ef E f by
E f = { f ( x ) , x ∈ S δ , 0 , x ∈ R 2 \ S δ . Ef=\begin{cases}
f(x),&\quad x\in \mathcal{S}_\delta,\\
0,&\quad x\in {\mathbb{R}}^2\backslash\mathcal{S}_\delta.
\end{cases} E f = { f ( x ) , 0 , x ∈ S δ , x ∈ R 2 \ S δ .
Then E f ∈ L e 2 ( R 2 ) Ef\in L_{\mathrm e}^2(\mathbb R^2) E f ∈ L e 2 ( R 2 ) and
∥ E f ∥ L 2 ( R 2 ) = ∥ f ∥ L 2 ( S δ ) . \left\|Ef\right\|_{L^2( {\mathbb{R}}^2)}=\|f\|_{L^2(\mathcal{S}_\delta)}. ∥ E f ∥ L 2 ( R 2 ) = ∥ f ∥ L 2 ( S δ ) .
If
( ∂ x 2 U i , f ) L 2 ( S δ ) = ( ∂ x 2 U i , E f ) L 2 ( R 2 ) = 0 , \left( {\partial} _{x_2} U_i,f\right)_{L^2(\mathcal{S}_\delta)}=\left( {\partial} _{x_2} U_i,Ef\right)_{L^2( {\mathbb{R}}^2)}=0, ( ∂ x 2 U i , f ) L 2 ( S δ ) = ( ∂ x 2 U i , E f ) L 2 ( R 2 ) = 0 ,
then E f ∈ ( I − P i ) L e 2 ( R 2 ) Ef\in(I-P_i)L_{\mathrm e}^2(\mathbb R^2) E f ∈ ( I − P i ) L e 2 ( R 2 ) . By the nondegeneracy of L τ i L_{\boldsymbol{\tau}_i} L τ i in R 2 {\mathbb{R}}^2 R 2 , there exists a unique ϕ ∈ ( I − P i ) H e 2 ( R 2 ) \phi\in(I-P_i)H_{\mathrm e}^2(\mathbb R^2) ϕ ∈ ( I − P i ) H e 2 ( R 2 ) satisfying
L τ i ϕ = E f in R 2 . L_{\boldsymbol{\tau}_i}\phi=Ef\quad\text{in }\mathbb R^2. L τ i ϕ = E f in R 2 .
Consequently, the restriction of ϕ \phi ϕ to S δ \mathcal{S}_\delta S δ satisfies
L τ i ϕ = f in S δ , L_{\boldsymbol{\tau_i}}\phi=f\quad\text{in }\mathcal{S}_\delta, L τ i ϕ = f in S δ ,
and the nondegeneracy estimate extends as follows.
Lemma. Suppose that ∥ h ∥ L 2 ( S δ ) < + ∞ \|h\|_{L^2(\mathcal{S}_\delta)}<+\infty ∥ h ∥ L 2 ( S δ ) < + ∞ and satisfies orthogonality condition ( ∂ x 2 U i , E h ) L 2 ( R 2 ) = ( ∂ x 2 U i , h ) L 2 ( S δ ) = 0. \begin{aligned}
\left( {\partial} _{x_2} U_i\,,\,Eh\right)_{L^2( {\mathbb{R}}^2)}=\left( {\partial} _{x_2} U_i\,,\,h\right)_{L^2(\mathcal{S}_\delta)}=0.
\end{aligned} ( ∂ x 2 U i , E h ) L 2 ( R 2 ) = ( ∂ x 2 U i , h ) L 2 ( S δ ) = 0. Then there exists δ 0 > 0 \delta_0 >0 δ 0 > 0 and c 0 > 0 c_0 > 0 c 0 > 0 such that, for all δ ∈ ( 0 , δ 0 ) \delta \in (0,\delta_0) δ ∈ ( 0 , δ 0 ) and ∣ τ i ∣ < 1 − ε |\tau_i|<1-\varepsilon ∣ τ i ∣ < 1 − ε , problem L τ i [ ϕ i ] = E h L_{\boldsymbol{\tau_i}}[\phi_i] = Eh L τ i [ ϕ i ] = E h has a unique solution ϕ i ∈ ( I − P i ) H e 2 ( R 2 ) \phi_i\in (I-P_i)H_{\textrm{e}}^2(\mathbb{R}^2) ϕ i ∈ ( I − P i ) H e 2 ( R 2 ) that satisfies the estimate ∥ ϕ ∥ H 2 ( S δ ) ≤ c 0 ∥ h ∥ L 2 ( S δ )
\|\phi\|_{H^2(\mathcal{S}_\delta)}\leq c_0\|h\|_{L^2(\mathcal{S}_\delta)}
∥ ϕ ∥ H 2 ( S δ ) ≤ c 0 ∥ h ∥ L 2 ( S δ )
Proof. We will follow the method used in [2 ] . The existence of a solution has already been discussed above. We now prove the estimate (3.4) . Since ∥ h ∥ L 2 ( S δ ) = ∥ E h ∥ L 2 ( R 2 ) \|h\|_{L^2(\mathcal{S}_\delta)}=\|Eh\|_{L^2( {\mathbb{R}}^2)} ∥ h ∥ L 2 ( S δ ) = ∥ E h ∥ L 2 ( R 2 ) , it suffices to prove
∥ ϕ ∥ H 2 ( S δ ) ≤ c 0 ∥ E h ∥ L 2 ( R 2 ) . \begin{equation}
\|\phi\|_{H^2(\mathcal{S}_\delta)}\leq c_0\|Eh\|_{L^2( {\mathbb{R}}^2)}. \tag*{(3.5)}
\end{equation} ∥ ϕ ∥ H 2 ( S δ ) ≤ c 0 ∥ E h ∥ L 2 ( R 2 ) . (3.5) Suppose that (3.5) is false. Then there exist sequences { δ k } \{\delta_k\} { δ k } , { τ i , k } \{\tau_{i,k}\} { τ i , k } , { ϕ k } \{\phi_k\} { ϕ k } and h k h_k h k (k = 1 , 2 , … k=1,2,\dots k = 1 , 2 , … ) with ∣ τ i , k ∣ < 1 − ε |\tau_{i,k}|<1-\varepsilon ∣ τ i , k ∣ < 1 − ε , ϕ k ∈ ( I − P i , k ) H e 2 ( R 2 ) \phi_k\in(I-P_{i,k})H_{\textrm{e}}^2(\mathbb{R}^2) ϕ k ∈ ( I − P i , k ) H e 2 ( R 2 ) such that
∥ h k ∥ L 2 ( S δ k ) → 0 , \begin{equation}
\|h_k\|_{L^2(\mathcal{S}_{\delta_k})}\to 0, \tag*{(3.6)}
\end{equation} ∥ h k ∥ L 2 ( S δ k ) → 0 , (3.6) ∥ ϕ k ∥ H 2 ( S δ k ) = 1 , \begin{equation}
\|\phi_k\|_{H^2(\mathcal{S}_{\delta_k})}=1, \tag*{(3.7)}
\end{equation} ∥ ϕ k ∥ H 2 ( S δ k ) = 1 , (3.7) where P i , k P_{i,k} P i , k denote the orthogonal projection of L e 2 ( R 2 ) L_{\textrm{e}}^2(\mathbb{R}^2) L e 2 ( R 2 ) onto s p a n { ∂ x 2 U ( ⋅ − τ i , k δ k e 2 ) } span\left\{ {\partial} _{x_2} U(\,\cdot-\tfrac{\tau_{i,k}}{\delta_k}e_2)\right\} s p an { ∂ x 2 U ( ⋅ − δ k τ i , k e 2 ) } and we use the same symbol for the projection H e 2 ( R 2 ) H_{\textrm{e}}^2(\mathbb{R}^2) H e 2 ( R 2 ) onto s p a n { ∂ x 2 U ( ⋅ − τ i , k δ k e 2 ) } span\left\{ {\partial} _{x_2} U(\,\cdot-\tfrac{\tau_{i,k}}{\delta_k}e_2)\right\} s p an { ∂ x 2 U ( ⋅ − δ k τ i , k e 2 ) } .
We introduce new sequences { Φ i , k } \{\Phi_{i,k}\} { Φ i , k } by
Φ i , k ( x ) = χ ( δ k x ) ϕ k ( x + ( τ i , k / δ k ) e 2 ) , \Phi_{i,k}(x)=\chi(\delta_kx)\phi_k(x+(\tau_{i,k}/\delta_k)e_2), Φ i , k ( x ) = χ ( δ k x ) ϕ k ( x + ( τ i , k / δ k ) e 2 ) , where χ ( z ) \chi(z) χ ( z ) is a cut-off function such that χ ( z ) = 1 \chi(z)=1 χ ( z ) = 1 for ∣ z ∣ ≤ R |z|\leq R ∣ z ∣ ≤ R and χ ( z ) = 0 \chi(z)=0 χ ( z ) = 0 for ∣ z ∣ > 2 R |z|>2R ∣ z ∣ > 2 R .
It follows from (3.7) and the smoothness of χ \chi χ that
∥ Φ i , k ∥ H 2 ( R 2 ) ≤ C \|\Phi_{i,k}\|_{H^2( {\mathbb{R}}^2)}\leq C ∥ Φ i , k ∥ H 2 ( R 2 ) ≤ C for all k k k sufficiently large. Therefore there exists a subsequence, again denoted by { Φ i , k } \{\Phi_{i,k}\} { Φ i , k } which converges weakly to a limit Φ i , ∞ \Phi_{i,\infty} Φ i , ∞ as k → ∞ k\to\infty k → ∞ in H 2 ( R 2 ) H^2( {\mathbb{R}}^2) H 2 ( R 2 ) . We are now going to show that Φ i , ∞ ≡ 0 \Phi_{i,\infty}\equiv 0 Φ i , ∞ ≡ 0 . As a first step we deduce
∫ R 2 Φ i , ∞ ∂ x 2 U = 0 , i = 1 , … , n . \begin{equation}
\displaystyle\int_{ {\mathbb{R}}^2}\Phi_{i,\infty}\partial_{x_2}U=0,\quad i=1,\dots,n. \tag*{(3.8)}
\end{equation} ∫ R 2 Φ i , ∞ ∂ x 2 U = 0 , i = 1 , … , n . (3.8) This statement is shown as
∫ R 2 Φ i , k ( x ) ∂ x 2 U ( x ) d x = ∫ R 2 χ ( δ k x ) ϕ k ( x + ( τ i , k / δ k ) e 2 ) ∂ x 2 U ( x ) d x = ∫ R 2 χ ( δ k x − τ i , k e 2 ) ϕ k ( x ) ∂ x 2 U ( x − ( τ i , k / δ k ) e 2 ) d x = ∫ R 2 ϕ k ( x ) ∂ x 2 U ( x − ( τ i , k / δ k ) e 2 ) d x + o ( 1 ) = o ( 1 ) . \begin{align*}
\displaystyle\int_{ {\mathbb{R}}^2}\Phi_{i,k}(x)\partial_{x_2}U(x)\,dx&=\displaystyle\int_{ {\mathbb{R}}^2}\chi(\delta_kx)\phi_k(x+(\tau_{i,k}/\delta_k)e_2)\partial_{x_2}U(x)\,dx\\
&=\displaystyle\int_{ {\mathbb{R}}^2}\chi(\delta_kx-\tau_{i,k}e_2)\phi_k(x)\partial_{x_2}U(x-(\tau_{i,k}/\delta_k)e_2)\,dx\\
&=\displaystyle\int_{ {\mathbb{R}}^2}\phi_k(x)\partial_{x_2}U(x-(\tau_{i,k}/\delta_k)e_2)\,dx+o(1)=o(1).
\end{align*} ∫ R 2 Φ i , k ( x ) ∂ x 2 U ( x ) d x = ∫ R 2 χ ( δ k x ) ϕ k ( x + ( τ i , k / δ k ) e 2 ) ∂ x 2 U ( x ) d x = ∫ R 2 χ ( δ k x − τ i , k e 2 ) ϕ k ( x ) ∂ x 2 U ( x − ( τ i , k / δ k ) e 2 ) d x = ∫ R 2 ϕ k ( x ) ∂ x 2 U ( x − ( τ i , k / δ k ) e 2 ) d x + o ( 1 ) = o ( 1 ) . Here we have used the facts that ϕ k ∈ ( I − P i , k ) H e 2 ( R 2 ) \phi_k\in(I-P_{i,k})H_{\textrm{e}}^2(\mathbb{R}^2) ϕ k ∈ ( I − P i , k ) H e 2 ( R 2 ) and the exponential decay of ∂ x 2 U \partial_{x_2}U ∂ x 2 U in Lemma (2.19) . This implies (3.8) .
Using K 0 = s p a n { ∂ x 2 U } \mathcal{K}_0=span{\{ {\partial} _{x_2} U\}} K 0 = s p an { ∂ x 2 U } to denote the kernel and cokernel of the linear operator − Δ + γ ′ ( U ) -\Delta + \gamma^\prime(U) − Δ + γ ′ ( U ) . Equation (3.8) implies that Φ i , ∞ ∈ K 0 ⊥ \Phi_{i,\infty}\in \mathcal{K}_0^\perp Φ i , ∞ ∈ K 0 ⊥ . By the exponential decay of U U U and (3.6) , we have after possibly taking a further subsequence that
− Δ Φ i , ∞ + γ ′ ( U ) Φ i , ∞ = 0 , -\Delta\Phi_{i,\infty} + \gamma^\prime(U)\Phi_{i,\infty}=0, − Δ Φ i , ∞ + γ ′ ( U ) Φ i , ∞ = 0 , i.e., Φ i , ∞ ∈ K 0 \Phi_{i,\infty}\in\mathcal{K}_0 Φ i , ∞ ∈ K 0 . Therefore Φ i , ∞ = 0 \Phi_{i,\infty} = 0 Φ i , ∞ = 0 .
Hence
Φ i , k ⇀ 0 in H 2 ( R 2 ) and Φ i , k → 0 in L 2 ( R 2 ) \Phi_{i,k}\rightharpoonup 0\quad\text{in }H^2( {\mathbb{R}}^2)\quad \text{and}\quad \Phi_{i,k}\to 0\quad\text{in }L^2( {\mathbb{R}}^2) Φ i , k ⇀ 0 in H 2 ( R 2 ) and Φ i , k → 0 in L 2 ( R 2 ) as k → ∞ k\to\infty k → ∞ . By the definition of Φ i , k \Phi_{i,k} Φ i , k and the estimates of γ ′ ( U ) = 1 + O ( U ) \gamma^\prime(U) = 1 + O(U) γ ′ ( U ) = 1 + O ( U ) for small U U U we get
∥ ( γ ′ ( U ( ⋅ − ( τ i , k / δ k ) e 2 ) ) − 1 ) ϕ k ∥ L 2 ( R 2 ) → 0 as k → ∞ . \left\|\left(\gamma^\prime\left(U(\,\cdot\,-(\tau_{i,k}/\delta_k)e_2)\right)-1\right)\phi_k\right\|_{L^2( {\mathbb{R}}^2)}\to 0\quad\text{as }k\to\infty. ∥ ( γ ′ ( U ( ⋅ − ( τ i , k / δ k ) e 2 ) ) − 1 ) ϕ k ∥ L 2 ( R 2 ) → 0 as k → ∞. Furthermore,
∥ ( − Δ + 1 ) ϕ k ∥ L 2 ( R 2 ) → 0 as k → ∞ . \left\|\left(-\Delta+1\right)\phi_k\right\|_{L^2( {\mathbb{R}}^2)}\to 0\quad\text{as }k\to\infty. ∥ ( − Δ + 1 ) ϕ k ∥ L 2 ( R 2 ) → 0 as k → ∞. In summary,
∥ Δ ϕ k ∥ L 2 ( R 2 ) → 0 and ∥ ϕ k ∥ L 2 ( R 2 ) → 0. \begin{equation}
\left\|\Delta \phi_k\right\|_{L^2( {\mathbb{R}}^2)}\to 0\qquad\text{and}\qquad \left\|\phi_k\right\|_{L^2( {\mathbb{R}}^2)}\to 0. \tag*{(3.9)}
\end{equation} ∥ Δ ϕ k ∥ L 2 ( R 2 ) → 0 and ∥ ϕ k ∥ L 2 ( R 2 ) → 0. (3.9) From (3.9) and the following elliptic regularity estimate (see [3 Theorem 11.2.1] )
∥ ϕ k ∥ H 2 ( S δ k ) ≤ C ( ∥ Δ ϕ k ∥ L 2 ( R 2 ) + ∥ ϕ k ∥ L 2 ( R 2 ) ) . \left\|\phi_k\right\|_{H^2(\mathcal{S}_{\delta_k})}\leq C \left(\left\|\Delta \phi_k\right\|_{L^2( {\mathbb{R}}^2)}+\left\|\phi_k\right\|_{L^2( {\mathbb{R}}^2)}\right). ∥ ϕ k ∥ H 2 ( S δ k ) ≤ C ( ∥ Δ ϕ k ∥ L 2 ( R 2 ) + ∥ ϕ k ∥ L 2 ( R 2 ) ) . Then we imply that
∥ ϕ k ∥ H 2 ( S δ k ) → 0 as k → ∞ . \left\|\phi_k\right\|_{H^2(\mathcal{S}_{\delta_k})}\to 0\quad\text{as }k\to\infty. ∥ ϕ k ∥ H 2 ( S δ k ) → 0 as k → ∞. This contradicts the assumption
∥ ϕ k ∥ H 2 ( S δ k ) = 1 \left\|\phi_k\right\|_{H^2(\mathcal{S}_{\delta_k})}= 1 ∥ ϕ k ∥ H 2 ( S δ k ) = 1 and the proof of Lemma (?) is completed. ◻
By elliptic theory, the nondegeneracy also holds in higher Sobolev spaces.
Corollary. Suppose that ∥ h ∥ H k ( S δ ) < + ∞ \|h\|_{H^k(\mathcal{S}_\delta)}<+\infty ∥ h ∥ H k ( S δ ) < + ∞ and satisfies orthogonality condition ( ∂ x 2 U i , E h ) L 2 ( R 2 ) = ( ∂ x 2 U i , h ) L 2 ( S δ ) = 0. \left( {\partial} _{x_2} U_i\,,\,Eh\right)_{L^2( {\mathbb{R}}^2)}=\left( {\partial} _{x_2} U_i\,,\,h\right)_{L^2(\mathcal{S}_\delta)}=0. ( ∂ x 2 U i , E h ) L 2 ( R 2 ) = ( ∂ x 2 U i , h ) L 2 ( S δ ) = 0. Then there exists δ 0 > 0 \delta_0 >0 δ 0 > 0 and c 0 > 0 c_0 > 0 c 0 > 0 such that, for all δ ∈ ( 0 , δ 0 ) \delta \in (0,\delta_0) δ ∈ ( 0 , δ 0 ) and ∣ τ i ∣ < 1 − ε |\tau_i|<1-\varepsilon ∣ τ i ∣ < 1 − ε , problem L τ i [ ϕ i ] = E h L_{\boldsymbol{\tau_i}}[\phi_i] = Eh L τ i [ ϕ i ] = E h has a unique solution ϕ i ∈ ( I − P i ) H e k + 2 ( R 2 ) \phi_i\in (I-P_i)H_{\textrm{e}}^{k+2}(\mathbb{R}^2) ϕ i ∈ ( I − P i ) H e k + 2 ( R 2 ) that satisfies the estimate ∥ ϕ ∥ H k + 2 ( S δ ) ≤ c 0 ∥ h ∥ H k ( S δ )
\|\phi\|_{H^{k+2}(\mathcal{S}_\delta)}\leq c_0\|h\|_{H^k(\mathcal{S}_\delta)}
∥ ϕ ∥ H k + 2 ( S δ ) ≤ c 0 ∥ h ∥ H k ( S δ )
Furthermore, by elliptic regularity estimate (see [3 Theorem 11.3.2] ), if h ∈ C 0 ( S δ ) h\in C^0(\mathcal{S}_\delta) h ∈ C 0 ( S δ ) , then ϕ i ∈ C 1 , α ( S δ ) \phi_i\in C^{1,\alpha}(\mathcal{S}_\delta) ϕ i ∈ C 1 , α ( S δ ) and
∥ ϕ i ∥ C 1 , α ( B i , 2 ) ≲ ∥ h ∥ C 0 ( S δ ) + ∥ ϕ i ∥ L 2 ( S δ ) ≲ ∥ h ∥ C 0 ( S δ ) + ∥ h ∥ L 2 ( S δ ) . \left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})}\lesssim\left\|h\right\|_{C^0(\mathcal{S}_\delta)}+\left\|\phi_i\right\|_{L^2(\mathcal{S}_\delta)}\lesssim\left\|h\right\|_{C^0(\mathcal{S}_\delta)}+\left\|h\right\|_{L^2(\mathcal{S}_\delta)}. ∥ ϕ i ∥ C 1 , α ( B i , 2 ) ≲ ∥ h ∥ C 0 ( S δ ) + ∥ ϕ i ∥ L 2 ( S δ ) ≲ ∥ h ∥ C 0 ( S δ ) + ∥ h ∥ L 2 ( S δ ) .
If h ∈ H m ( S δ ) h\in H^m(\mathcal{S}_\delta) h ∈ H m ( S δ ) with m > 1 m>1 m > 1 , by Sobolev embedding,
∥ ϕ i ∥ C 1 , α ( B i , 2 ) ≲ ∥ h ∥ H m ( S δ ) , 0 < α < min { 1 , m − 1 } . \begin{equation}
\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})}\lesssim\left\|h\right\|_{H^m(\mathcal{S}_\delta)},\qquad 0<\alpha<\min\{1,m-1\}. \tag*{(3.10)}
\end{equation} ∥ ϕ i ∥ C 1 , α ( B i , 2 ) ≲ ∥ h ∥ H m ( S δ ) , 0 < α < min { 1 , m − 1 } . (3.10)
The linear theory of the outer problem
To solve the outer problem (3.2) , we consider the associated linear problem
{ Δ ψ − ψ = h , x ∈ S δ , ψ = 0 , x ∈ ∂ S δ . \begin{equation}
\begin{cases}
\Delta\psi-\psi=h,
&\quad x\in\mathcal{S}_\delta,\\[2mm]
\psi=0,
&\quad x\in\partial\mathcal{S}_\delta .
\end{cases} \tag*{(3.11)}
\end{equation} ⎩ ⎨ ⎧ Δ ψ − ψ = h , ψ = 0 , x ∈ S δ , x ∈ ∂ S δ . (3.11)
Recall § 2.3 , the solution of (3.11) can be represented as
ψ ( x ) = ∫ S δ G ~ ( x , ξ ) h ( ξ ) d ξ . \begin{equation}
\psi(x)=\int_{\mathcal{S}_\delta}\widetilde G(x,\xi)h(\xi)\,d\xi. \tag*{(3.12)}
\end{equation} ψ ( x ) = ∫ S δ G ( x , ξ ) h ( ξ ) d ξ . (3.12)
Moreover, from the exponential decay of G 0 G_0 G 0 in (2.22) , the series defining G ~ \widetilde G G converges absolutely and inherits the same decay properties away from the singularity. Consequently, standard elliptic estimates yield
∥ ψ ∥ H k ( S δ ) ≲ ∥ h ∥ H k − 2 ( S δ ) , k ≥ 2. \|\psi\|_{H^k(\mathcal{S}_\delta)}\lesssim\|h\|_{H^{k-2}(\mathcal{S}_\delta)},\qquad k\ge2. ∥ ψ ∥ H k ( S δ ) ≲ ∥ h ∥ H k − 2 ( S δ ) , k ≥ 2.
Solve the inner-outer gluing system
In this section, we use the fixed‑point argument to solve the inner-outer gluing system (3.1) , (3.2) and perturbed Gamma eq . Thus obtaining the functions ( ϕ 1 , … , ϕ n , ψ , Γ s ) (\phi_1,\dots,\phi_n,\psi,\Gamma_{\mathrm{s}}) ( ϕ 1 , … , ϕ n , ψ , Γ s ) and reducing the system to a finite-dimensional problem.
Solving the outer problem
Before considering the outer problem (3.2) , we first employ the fixed-point method to determine Γ s \Gamma_{\mathrm{s}} Γ s from equation perturbed Gamma eq . This requires us to consider the bound on G ( τ , 0 , 0 ) G(\boldsymbol{\tau}, 0, 0) G ( τ , 0 , 0 ) , D u G D_u G D u G and D Γ s G D_{\Gamma_{\mathrm{s}}}G D Γ s G . Here the operators D u G D_u G D u G and D Γ s G D_{\Gamma_{\mathrm{s}}}G D Γ s G are defined respectively as
D u G ( τ , u , Γ s ) [ v ] = d d t G ( τ , u + t v , Γ s ) ∣ t = 0 , v ∈ H k ( S δ ) D_uG(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}})[v]=\frac{\mathrm{d}}{\mathrm{d}t}G(\boldsymbol{\tau},u+tv,\Gamma_{\mathrm{s}})\bigg|_{t=0},\qquad v\in H^k(\mathcal{S}_\delta) D u G ( τ , u , Γ s ) [ v ] = d t d G ( τ , u + t v , Γ s ) t = 0 , v ∈ H k ( S δ )
and
D Γ s G ( τ , u , Γ s ) [ Λ s ] = d d t G ( τ , u , Γ s + t Λ s ) ∣ t = 0 , Λ s ∈ H k ( R ) . D_{\Gamma_{\mathrm{s}}}G(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}})[\Lambda_{\mathrm{s}}]=\frac{\mathrm{d}}{\mathrm{d}t}G(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}}+t\Lambda_{\mathrm{s}})\bigg|_{t=0},\qquad \Lambda_{\mathrm{s}}\in H^k( {\mathbb{R}}). D Γ s G ( τ , u , Γ s ) [ Λ s ] = d t d G ( τ , u , Γ s + t Λ s ) t = 0 , Λ s ∈ H k ( R ) .
For G ( τ , 0 , 0 ) G(\tau, 0, 0) G ( τ , 0 , 0 ) , one can directly compute
G ( τ , 0 , 0 ) = 1 2 δ 2 ( ∂ x 2 ω τ ( ⋅ δ , 1 δ ) ) 2 . \begin{equation*}
G(\boldsymbol{\tau}, 0, 0)
= \frac 1{2 \delta^2}\left( {\partial} _{x_2}\omega_{\boldsymbol{\tau}}\left(\tfrac{\,\cdot\,}{\delta}, \tfrac 1\delta \right)\right)^2.
\end{equation*} G ( τ , 0 , 0 ) = 2 δ 2 1 ( ∂ x 2 ω τ ( δ ⋅ , δ 1 ) ) 2 .
By Corollary (?) and the Sobolev trace theorem, for any k 0 + 1 2 ≥ k ≥ 0 k_0+\frac 12\ge k\ge 0 k 0 + 2 1 ≥ k ≥ 0 , i = 1 , … , n i=1,\dots,n i = 1 , … , n , we have
∥ ∂ x 2 ( U i − ( U i ) b c ) ( ⋅ , 1 δ ) − 2 ( ∂ x 2 U i ) ( ⋅ , 1 δ ) ∥ H k ( R ) ≲ ∥ ( ∂ x 2 U i ) ( ⋅ , − 3 δ ) ∥ H k ( R ) + δ 3 4 e − 2 ( 1 − ∣ τ i ∣ ) δ . \begin{equation*}
\left \| {\partial} _{x_2} (U_i - (U_i)_{\mathrm{bc}}) (\,\cdot\,, \tfrac 1\delta)
-2( {\partial} _{x_2} U_i) (\,\cdot\,, \tfrac 1\delta) \right\|_{H^k ( {\mathbb{R}})} \lesssim \|( {\partial} _{x_2} U_i) ( \,\cdot\,, -\tfrac 3\delta)\|_{H^k ( {\mathbb{R}})} + \delta^{\frac 34} e^{-\frac {2(1-|\tau_i|)}\delta}.
\end{equation*} ∂ x 2 ( U i − ( U i ) bc ) ( ⋅ , δ 1 ) − 2 ( ∂ x 2 U i ) ( ⋅ , δ 1 ) H k ( R ) ≲ ∥ ( ∂ x 2 U i ) ( ⋅ , − δ 3 ) ∥ H k ( R ) + δ 4 3 e − δ 2 ( 1 − ∣ τ i ∣ ) .
From Proposition 2.1 , we can estimate ( ∂ x 2 U i ) ( ⋅ , 1 δ ) ( {\partial} _{x_2} U_i) (\,\cdot\,, \frac 1\delta) ( ∂ x 2 U i ) ( ⋅ , δ 1 )
∥ ( ∂ x 2 U i ) ( ⋅ , 1 δ ) ∥ H k ( R ) 2 ≲ ( ∫ 0 δ − 1 2 + ∫ δ − 1 2 ∞ ) ( x 1 2 + ( 1 − τ i δ ) 2 ) − 1 2 e − 2 ( x 1 2 + ( 1 − τ i δ ) 2 ) 1 2 d x 1 ≲ δ 1 2 e − 2 ( 1 − τ i ) δ + ∫ ( ( 1 − τ i ) 2 + δ ) 1 2 δ ∞ d ρ x 1 ( ρ ) e 2 ρ ≲ δ 1 2 e − 2 ( 1 − τ i ) δ , \begin{equation*}
\begin{split}
\left \|( {\partial} _{x_2} U_i) ( \,\cdot\,, \tfrac 1\delta) \right\|_{H^k ( {\mathbb{R}})}^2 \lesssim & \left(\int_0^{\delta^{-\frac 12}} + \int_{\delta^{-\frac 12}}^\infty\right) \left(x_1^2 + (\tfrac {1-\tau_i}\delta)^2 \right)^{-\frac 12} e^{-2(x_1^2 + (\frac {1-\tau_i} \delta)^2)^{\frac 12}} \, dx_1\\
\lesssim & \delta^{\frac 12} e^{-\frac {2(1-\tau_i)}\delta} + \int_{\frac {((1-\tau_i)^2 +\delta)^{\frac 12}}\delta }^\infty \frac {d\rho} {x_1(\rho) e^{2\rho}}\lesssim \delta^{\frac 12} e^{-\frac {2(1-\tau_i)}\delta},
\end{split}
\end{equation*} ( ∂ x 2 U i ) ( ⋅ , δ 1 ) H k ( R ) 2 ≲ ≲ ∫ 0 δ − 2 1 + ∫ δ − 2 1 ∞ ( x 1 2 + ( δ 1 − τ i ) 2 ) − 2 1 e − 2 ( x 1 2 + ( δ 1 − τ i ) 2 ) 2 1 d x 1 δ 2 1 e − δ 2 ( 1 − τ i ) + ∫ δ (( 1 − τ i ) 2 + δ ) 2 1 ∞ x 1 ( ρ ) e 2 ρ d ρ ≲ δ 2 1 e − δ 2 ( 1 − τ i ) ,
where x 1 ( ρ ) = ( ρ 2 − ( 1 − τ i δ ) 2 ) 1 2 x_1 (\rho) = (\rho^2 - (\frac {1-\tau_i}\delta)^2)^{\frac 12} x 1 ( ρ ) = ( ρ 2 − ( δ 1 − τ i ) 2 ) 2 1 . This implies that
∥ ( ∂ x 2 ( U i − ( U i ) b c ) ) ( ⋅ , 1 δ ) ∥ H k ( R ) ≲ δ 1 4 e − 1 − τ i δ + δ 3 4 e − 2 ( 1 − ∣ τ i ∣ ) δ for all i = 1 , … , n . \begin{equation}
\left\|( {\partial} _{x_2} (U_i - (U_i)_{\mathrm{bc}})) (\,\cdot\,, \tfrac 1\delta) \right\|_{H^k ( {\mathbb{R}})} \lesssim \delta^{\frac 14} e^{-\frac {1-\tau_i}\delta}+ \delta^{\frac 34} e^{-\frac {2(1-|\tau_i|)}\delta}\quad\text{for all }i=1,\dots,n. \tag*{(4.1)}
\end{equation} ( ∂ x 2 ( U i − ( U i ) bc )) ( ⋅ , δ 1 ) H k ( R ) ≲ δ 4 1 e − δ 1 − τ i + δ 4 3 e − δ 2 ( 1 − ∣ τ i ∣ ) for all i = 1 , … , n . (4.1)
From the one-dimensional scaling property,
∥ f ( ⋅ δ ) ∥ H ˙ s ( R ) = δ − s + 1 2 ∥ f ∥ H ˙ s ( R ) for all s ∈ R . \|f(\tfrac{\cdot}{\delta})\|_{\dot H^s ( {\mathbb{R}})} = \delta^{-s+\frac 12} \|f\|_{\dot H^s ( {\mathbb{R}})} \quad\text{for all }s\in \mathbb{R}. ∥ f ( δ ⋅ ) ∥ H ˙ s ( R ) = δ − s + 2 1 ∥ f ∥ H ˙ s ( R ) for all s ∈ R .
By Lemma 2.1 , we obtain the estimate on G ( τ , 0 , 0 ) G(\boldsymbol{\tau}, 0, 0) G ( τ , 0 , 0 ) from the scaling property as
∥ G ( τ , 0 , 0 ) ∥ H k − 2 ( R ) ≲ δ 1 2 − ( k − 2 ) − 2 ∥ ( ( ∂ x 2 ω τ ) ( ⋅ , 1 δ ) ) 2 ∥ H k − 2 ( R ) ≲ δ 1 − k e − 2 ( 1 − ∣ τ ∣ ) δ . \begin{equation}
\|G(\boldsymbol{\tau},0, 0)\|_{H^{k-2} ( {\mathbb{R}})} \lesssim \delta^{\frac 12 -(k-2)-2} \left\| \left(( {\partial} _{x_2}\omega_{\boldsymbol{\tau}}) (\,\cdot\,, \tfrac 1\delta)\right)^2 \right\|_{H^{k-2} ( {\mathbb{R}})} \lesssim \delta^{1-k} e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}. \tag*{(4.2)}
\end{equation} ∥ G ( τ , 0 , 0 ) ∥ H k − 2 ( R ) ≲ δ 2 1 − ( k − 2 ) − 2 ( ( ∂ x 2 ω τ ) ( ⋅ , δ 1 ) ) 2 H k − 2 ( R ) ≲ δ 1 − k e − δ 2 ( 1 − ∣ τ ∣ ) . (4.2)
For D u G D_u G D u G , from the definition of G G G , for any v ∈ H e k ( S δ ) v \in H_{\mathrm{e}}^k(\mathcal{S}_\delta) v ∈ H e k ( S δ )
D u G ( τ , u , Γ s ) v = 1 δ 2 A ( Γ s ) − 1 [ ( ∂ x 2 ( ω τ + u ) ∂ x 2 v ) ( ⋅ δ , 1 δ ) ( 1 + ∣ ∂ x 1 ∣ coth ( 2 ∣ ∂ x 1 ∣ ) Γ s ) 2 + Γ s ′ 2 ] . D_u G(\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}}) v= \frac{1}{\delta^{2}} A(\Gamma_{\mathrm{s}})^{-1} \left[ \frac{( {\partial} _{x_2} (\omega_{\boldsymbol{\tau}}+u) {\partial} _{x_2} v)(\frac{\,\cdot\,} \delta,\frac{1}{\delta}) }{(1 +|{\partial_{x_1}}| \coth{(2|{\partial_{x_1}}|)} \Gamma_{\mathrm{s}})^2 + {\Gamma_{\mathrm{s}}^\prime}^2}\right]. D u G ( τ , u , Γ s ) v = δ 2 1 A ( Γ s ) − 1 [ ( 1 + ∣ ∂ x 1 ∣ coth ( 2∣ ∂ x 1 ∣ ) Γ s ) 2 + Γ s ′ 2 ( ∂ x 2 ( ω τ + u ) ∂ x 2 v ) ( δ ⋅ , δ 1 ) ] .
By the Sobolev trace theorem and (4.1) , we have
∥ ( ∂ x 2 ( ω τ + u ) ∂ x 2 v ) ( ⋅ , 1 δ ) ∥ H k − 2 ( R ) ≲ ∥ ∂ x 2 ( ω τ + u ) ( ⋅ , 1 δ ) ∥ H k ( R ) ∥ ∂ x 2 v ( ⋅ , 1 δ ) ∥ H k ( R ) ≲ ( ∥ u ∥ H k ( S δ ) + δ 1 4 e − 1 − ∣ τ ∣ δ ) ∥ v ∥ H k ( S δ ) . \begin{align*}
& \left\| \left( {\partial} _{x_2} (\omega_{\boldsymbol{\tau}}+u) {\partial} _{x_2} v \right)({\,\cdot\,},\tfrac{1}{\delta})\right\|_{H^{k-2} ( {\mathbb{R}})} \\
&\lesssim\| {\partial} _{x_2} (\omega_{\boldsymbol{\tau}}+u)({\,\cdot\,},\tfrac{1}{\delta})\|_{H^{k} ( {\mathbb{R}})} \| {\partial} _{x_2} v ({\,\cdot\,},\tfrac{1}{\delta})\|_{H^{k} ( {\mathbb{R}})} \lesssim \left(\|u\|_{H^k (\mathcal{S}_\delta)} + \delta^{\frac 14} e^{-\frac {1-|\boldsymbol{\tau}|}\delta} \right) \|v\|_{H^k (\mathcal{S}_\delta)}.
\end{align*} ( ∂ x 2 ( ω τ + u ) ∂ x 2 v ) ( ⋅ , δ 1 ) H k − 2 ( R ) ≲ ∥ ∂ x 2 ( ω τ + u ) ( ⋅ , δ 1 ) ∥ H k ( R ) ∥ ∂ x 2 v ( ⋅ , δ 1 ) ∥ H k ( R ) ≲ ( ∥ u ∥ H k ( S δ ) + δ 4 1 e − δ 1 − ∣ τ ∣ ) ∥ v ∥ H k ( S δ ) .
The scaling property implies the bound on D u G D_u G D u G .
∥ D u G ( τ , u , Γ s ) v ∥ H k − 2 ( R ) ≲ ( δ 1 2 − k ∥ u ∥ H k ( S δ ) + δ 3 4 − k e − 1 − ∣ τ ∣ δ ) ∥ v ∥ H k ( S δ ) . \begin{equation}
\|D_u G(\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}}) v\|_{H^{k-2} ( {\mathbb{R}})}\lesssim \left(\delta^{\frac 12-k}\|u\|_{H^k (\mathcal{S}_\delta)} + \delta^{\frac 34-k} e^{-\frac {1-|\boldsymbol{\tau}|}\delta} \right) \|v\|_{H^k (\mathcal{S}_\delta)}. \tag*{(4.3)}
\end{equation} ∥ D u G ( τ , u , Γ s ) v ∥ H k − 2 ( R ) ≲ ( δ 2 1 − k ∥ u ∥ H k ( S δ ) + δ 4 3 − k e − δ 1 − ∣ τ ∣ ) ∥ v ∥ H k ( S δ ) . (4.3)
Regarding D Γ s G D_{\Gamma_{\mathrm{s}}} G D Γ s G , for any Λ s ∈ H e k ( R ) \Lambda_{\mathrm{s}} \in H_{\mathrm{e}}^k(\mathbb{R}) Λ s ∈ H e k ( R ) ,
2 δ 2 D Γ s G ( τ , u , Γ s ) Λ s = ( D Γ s ( A ( Γ s ) − 1 ) Λ s ) [ ( ∂ x 2 ( ω τ + u ) ) ( ⋅ δ , 1 δ ) 2 ( 1 + ∣ ∂ x 1 ∣ coth ( 2 ∣ ∂ x 1 ∣ ) Γ s ) 2 + Γ s ′ 2 ] − 2 A ( Γ s ) − 1 [ ( ∂ x 2 ( ω τ + u ) ) ( ⋅ δ , 1 δ ) 2 ( ( 1 + ∣ ∂ x 1 ∣ coth ( 2 ∣ ∂ x 1 ∣ ) Γ s ) 2 + Γ s ′ 2 ) 2 × ( ( 1 + ∣ ∂ x 1 ∣ coth ( 2 ∣ ∂ x 1 ∣ ) Γ s ) ∣ ∂ x 1 ∣ coth ( 2 ∣ ∂ x 1 ∣ ) Λ s + Γ s ′ Λ s ′ ) ] . \begin{align*}
2\delta^2 D_{\Gamma_{\mathrm{s}}} G(\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}}) \Lambda_{\mathrm{s}} = & \left( D_{\Gamma_{\mathrm{s}}} (A(\Gamma_{\mathrm{s}})^{-1}) \Lambda_{\mathrm{s}} \right) \left[ \frac{( {\partial} _{x_2} (\omega_{\boldsymbol{\tau}}+u))(\frac{\,\cdot\,} \delta,\frac{1}{\delta})^2 }{(1 +|{\partial_{x_1}}| \coth{(2|{\partial_{x_1}}|)} \Gamma_{\mathrm{s}})^2 + {\Gamma_{\mathrm{s}}^\prime}^2}\right] \\
&-2A(\Gamma_{\mathrm{s}})^{-1}\left[ \frac{( {\partial} _{x_2} (\omega_{\boldsymbol{\tau}}+u))(\frac{\,\cdot\,} \delta,\frac{1}{\delta})^2 }{\left( (1 +|{\partial_{x_1}}| \coth{(2|{\partial_{x_1}}|)} \Gamma_{\mathrm{s}})^2 + {\Gamma_{\mathrm{s}}^\prime}^2\right)^2}\right. \\
&\left. \times \left( (1 +|{\partial_{x_1}}| \coth{(2|{\partial_{x_1}}|)}\Gamma_{\mathrm{s}})|{\partial_{x_1}}| \coth{(2|{\partial_{x_1}}|)} \Lambda_{\mathrm{s}} + \Gamma_{\mathrm{s}}' \Lambda_{\mathrm{s}} '\right)\right].
\end{align*} 2 δ 2 D Γ s G ( τ , u , Γ s ) Λ s = ( D Γ s ( A ( Γ s ) − 1 ) Λ s ) [ ( 1 + ∣ ∂ x 1 ∣ coth ( 2∣ ∂ x 1 ∣ ) Γ s ) 2 + Γ s ′ 2 ( ∂ x 2 ( ω τ + u )) ( δ ⋅ , δ 1 ) 2 ] − 2 A ( Γ s ) − 1 [ ( ( 1 + ∣ ∂ x 1 ∣ coth ( 2∣ ∂ x 1 ∣ ) Γ s ) 2 + Γ s ′ 2 ) 2 ( ∂ x 2 ( ω τ + u )) ( δ ⋅ , δ 1 ) 2 × ( ( 1 + ∣ ∂ x 1 ∣ coth ( 2∣ ∂ x 1 ∣ ) Γ s ) ∣ ∂ x 1 ∣ coth ( 2∣ ∂ x 1 ∣ ) Λ s + Γ s ′ Λ s ′ ) ] .
For small ∥ u ∥ H k ( S δ ) \|u\|_{H^{k}(\mathcal{S}_\delta)} ∥ u ∥ H k ( S δ ) and ∥ Γ s ∥ H k ( R ) \|\Gamma_{\mathrm{s}}\|_{H^k ( {\mathbb{R}})} ∥ Γ s ∥ H k ( R ) , using the scaling property, (4.1) and the Sobolev trace theorem, we obtain
∥ D Γ s G ( τ , u , Γ s ) Λ s ∥ H k − 2 ( R ) ≲ δ − 2 ∥ ( ∂ x 2 ( ω τ + u ) ) ( ⋅ δ , 1 δ ) 2 ∥ H k − 2 ( R ) ∥ Λ s ∥ H k ( R ) ≲ δ 1 2 − k ( ∥ u ∥ H k ( S δ ) 2 + δ 1 2 e − 2 ( 1 − ∣ τ ∣ ) δ ) ∥ Λ s ∥ H k ( R ) . \begin{equation}
\begin{aligned}
\|D_{\Gamma_{\mathrm{s}}} G(\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}}) \Lambda_{\mathrm{s}}\|_{H^{k-2} ( {\mathbb{R}})} \lesssim & \delta^{-2} \|( {\partial} _{x_2} (\omega_{\boldsymbol{\tau}}+u))(\tfrac{\,\cdot\,} \delta,\tfrac{1}{\delta})^2\|_{H^{k-2}( {\mathbb{R}})} \|\Lambda_{\mathrm{s}}\|_{H^k ( {\mathbb{R}})} \\
\lesssim & \delta^{\frac 12-k} \left(\|u\|_{H^{k}(\mathcal{S}_\delta)}^2 + \delta^{\frac 12} e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta} \right) \|\Lambda_{\mathrm{s}}\|_{H^k ( {\mathbb{R}})}.
\end{aligned} \tag*{(4.4)}
\end{equation} ∥ D Γ s G ( τ , u , Γ s ) Λ s ∥ H k − 2 ( R ) ≲ ≲ δ − 2 ∥ ( ∂ x 2 ( ω τ + u )) ( δ ⋅ , δ 1 ) 2 ∥ H k − 2 ( R ) ∥ Λ s ∥ H k ( R ) δ 2 1 − k ( ∥ u ∥ H k ( S δ ) 2 + δ 2 1 e − δ 2 ( 1 − ∣ τ ∣ ) ) ∥ Λ s ∥ H k ( R ) . (4.4)
Using the above estimate, we have
Proposition. Assume ∥ u ∥ H k ( S δ ) ≤ δ k + 1 \|u\|_{H^{k}(\mathcal{S}_\delta)}\leq\delta^{k+1} ∥ u ∥ H k ( S δ ) ≤ δ k + 1 . Then there exist constants C ~ > 0 \widetilde C>0 C > 0 and δ 0 > 0 \delta_0>0 δ 0 > 0 such that for τ ∈ P ε \boldsymbol{\tau} \in P_{\varepsilon} τ ∈ P ε and δ ∈ ( 0 , δ 0 ) \delta\in(0,\delta_0) δ ∈ ( 0 , δ 0 ) , the operator T s [ Γ s ] : = − ( g − α 2 ( ∂ x 1 ) 2 ) − 1 G ( τ , u , Γ s ) \mathrm{T}_s[\Gamma_{\mathrm{s}}]:=-\left(g-\alpha^2({\partial_{x_1}})^2\right)^{-1}G(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}}) T s [ Γ s ] := − ( g − α 2 ( ∂ x 1 ) 2 ) − 1 G ( τ , u , Γ s ) has a fixed point Γ ~ s [ u ] \widetilde{\Gamma}_{\mathrm{s}}[u] Γ s [ u ] in the set B s : = { Γ s ∈ H e k ( R ) ∣ ∥ Γ s ∥ H k ( R ) ≤ δ k } , B_s:=\{\Gamma_{\mathrm{s}}\in H_{\mathrm{e}}^k( {\mathbb{R}})\;|\;\|\Gamma_{\mathrm{s}}\|_{H^k ( {\mathbb{R}})}\leq \delta^{k}\}, B s := { Γ s ∈ H e k ( R ) ∣ ∥ Γ s ∥ H k ( R ) ≤ δ k } , which satisfies ∥ Γ ~ s [ u ] ∥ H k ( R ) ≤ C ~ ( δ 1 − k e − 2 ( 1 − ∣ τ ∣ ) δ + δ 1 2 − k ∥ u ∥ H k ( S δ ) 2 ) . \left\|\widetilde{\Gamma}_{\mathrm{s}}[u]\right\|_{H^k( {\mathbb{R}})}\leq \widetilde C\left(\delta^{1-k} e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{\frac 12-k}\|u\|_{H^k (\mathcal{S}_\delta)}^2\right). Γ s [ u ] H k ( R ) ≤ C ( δ 1 − k e − δ 2 ( 1 − ∣ τ ∣ ) + δ 2 1 − k ∥ u ∥ H k ( S δ ) 2 ) .
Proof. We first prove that T s [ Γ s ] ∈ B s \mathrm{T}_s[\Gamma_{\mathrm{s}}]\in B_s T s [ Γ s ] ∈ B s for any Γ s ∈ B s \Gamma_{\mathrm{s}}\in B_s Γ s ∈ B s . Applying (4.2) , (4.3) and (4.4) , we obtain
∥ ( g − α 2 ( ∂ x 1 ) 2 ) − 1 G ( τ , u , Γ s ) ∥ H k ( R ) ≤ C ( δ 1 − k e − 2 ( 1 − ∣ τ ∣ ) δ + δ k + 5 2 ) . \begin{align*}
\left\|\left(g-\alpha^2({\partial_{x_1}})^2\right)^{-1}G(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}})\right\|_{H^k ( {\mathbb{R}})}\leq C\left(\delta^{1-k} e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{k+\frac{5}{2}}\right).
\end{align*} ( g − α 2 ( ∂ x 1 ) 2 ) − 1 G ( τ , u , Γ s ) H k ( R ) ≤ C ( δ 1 − k e − δ 2 ( 1 − ∣ τ ∣ ) + δ k + 2 5 ) . Choosing δ \delta δ small enough such that C ( δ 1 − k e − 2 ( 1 − ∣ τ ∣ ) δ + δ k + 5 2 ) ≤ δ k C\left(\delta^{1-k} e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{k+\frac{5}{2}}\right)\leq \delta^{k} C ( δ 1 − k e − δ 2 ( 1 − ∣ τ ∣ ) + δ k + 2 5 ) ≤ δ k , hence T s [ Γ s ] ∈ B s \mathrm{T}_s[\Gamma_{\mathrm{s}}]\in B_s T s [ Γ s ] ∈ B s .
Next we prove T s \mathrm{T}_s T s is a contraction on B s B_s B s . Applying (4.4) yields
∥ T s [ Γ s , 1 ] − T s [ Γ s , 2 ] ∥ H k ( R ) ≤ C ∥ D Γ s G ( τ , u , ξ ) [ Γ s , 1 − Γ s , 2 ] ∥ H k − 2 ( R ) ≤ C δ 1 2 − k ( δ 2 k + 2 + δ 1 2 e − 2 ( 1 − ∣ τ ∣ ) δ ) ∥ Γ s , 1 − Γ s , 2 ∥ H k ( R ) . \begin{align*}
\|\mathrm{T}_s[\Gamma_{\mathrm{s},1}]-\mathrm{T}_s[\Gamma_{\mathrm{s},2}]\|_{H^k ( {\mathbb{R}})}&\leq C\|D_{\Gamma_{\mathrm{s}}} G(\boldsymbol{\tau}, u, \xi) [\Gamma_{\mathrm{s},1}-\Gamma_{\mathrm{s},2}]\|_{H^{k-2} ( {\mathbb{R}})}\\
&\leq C\delta^{\frac 12-k} \left(\delta^{2k+2} + \delta^{\frac 12} e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta} \right)\|\Gamma_{\mathrm{s},1}-\Gamma_{\mathrm{s},2}\|_{H^k ( {\mathbb{R}})}.
\end{align*} ∥ T s [ Γ s , 1 ] − T s [ Γ s , 2 ] ∥ H k ( R ) ≤ C ∥ D Γ s G ( τ , u , ξ ) [ Γ s , 1 − Γ s , 2 ] ∥ H k − 2 ( R ) ≤ C δ 2 1 − k ( δ 2 k + 2 + δ 2 1 e − δ 2 ( 1 − ∣ τ ∣ ) ) ∥ Γ s , 1 − Γ s , 2 ∥ H k ( R ) . Finally, choosing δ \delta δ small enough so that
C δ 1 2 − k ( δ 2 k + 2 + δ 1 2 e − 2 ( 1 − ∣ τ ∣ ) δ ) < 1 , C\delta^{\frac 12-k} \left(\delta^{2k+2} + \delta^{\frac 12} e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta} \right)<1, C δ 2 1 − k ( δ 2 k + 2 + δ 2 1 e − δ 2 ( 1 − ∣ τ ∣ ) ) < 1 , then T s \mathrm{T}_s T s is a contraction on B s B_s B s . By the contraction mapping principle, T s \mathrm{T}_s T s possesses a fixed point Γ ~ s [ u ] \widetilde{\Gamma}_{\mathrm{s}}[u] Γ s [ u ] in B s B_s B s . Moreover, we have
∥ Γ ~ s [ u ] ∥ H k ( R ) ≲ ∥ G ( τ , u , 0 ) ∥ H k − 2 ( R ) ≲ ∥ G ( τ , 0 , 0 ) ∥ H k − 2 ( R ) + ∥ D u G ( τ , ξ , 0 ) u ∥ H k − 2 ( R ) ≤ C ~ ( δ 1 − k e − 2 ( 1 − ∣ τ ∣ ) δ + δ 1 2 − k ∥ u ∥ H k ( S δ ) 2 ) . \begin{align*}
\left\|\widetilde{\Gamma}_{\mathrm{s}}[u]\right\|_{H^k( {\mathbb{R}})}\lesssim \|G(\boldsymbol{\tau},u, 0)\|_{H^{k-2} ( {\mathbb{R}})}&\lesssim\|G(\boldsymbol{\tau},0, 0)\|_{H^{k-2} ( {\mathbb{R}})}+\|D_u G(\boldsymbol{\tau},\xi, 0)u\|_{H^{k-2} ( {\mathbb{R}})}\\
&\leq \widetilde C\left(\delta^{1-k} e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{\frac 12-k}\|u\|_{H^k (\mathcal{S}_\delta)}^2\right).
\end{align*} Γ s [ u ] H k ( R ) ≲ ∥ G ( τ , u , 0 ) ∥ H k − 2 ( R ) ≲ ∥ G ( τ , 0 , 0 ) ∥ H k − 2 ( R ) + ∥ D u G ( τ , ξ , 0 ) u ∥ H k − 2 ( R ) ≤ C ( δ 1 − k e − δ 2 ( 1 − ∣ τ ∣ ) + δ 2 1 − k ∥ u ∥ H k ( S δ ) 2 ) . ◻
Now we consider the outer problem (3.2) . Let
L o u t [ ψ ] : = − Δ ψ ( x ) + ψ ( x ) , L_{out}[\psi]:=-\Delta\psi(x)+\psi(x), L o u t [ ψ ] := − Δ ψ ( x ) + ψ ( x ) ,
then the outer problem can be written as
{ L o u t [ ψ ] = f o u t [ ϕ 1 , … , ϕ n , ψ , Γ s ] , in S δ , ψ ( x ) = 0 , on ∂ S δ , \begin{equation}
\begin{aligned}
\begin{cases}
L_{out}[\psi]=f_{out}[\phi_1,\dots,\phi_n,\psi,\Gamma_{\mathrm{s}}],\;&\text{in }\mathcal{S}_\delta,\\
\psi(x)=0,\;&\text{on }\partial\mathcal{S}_\delta,
\end{cases}
\end{aligned} \tag*{(4.6)}
\end{equation} { L o u t [ ψ ] = f o u t [ ϕ 1 , … , ϕ n , ψ , Γ s ] , ψ ( x ) = 0 , in S δ , on ∂ S δ , (4.6)
where
f o u t [ ϕ 1 , … , ϕ n , ψ , Γ s ] = 2 δ R ∑ i = 1 n ∇ η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ⋅ ∇ ϕ i ( x ) + δ 2 R 2 ∑ i = 1 n Δ η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ϕ i ( x ) − ∑ i = 1 n η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ( γ ′ ( ∑ i = 1 n σ i U i ) − γ ′ ( U i ) ) ϕ i ( x ) − ( 1 − ∑ i = 1 n η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ) F ( τ , u , Γ s ) ( x ) − ( 1 − ∑ i = 1 n η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ) ( γ ′ ( ∑ i = 1 n σ i U i ) − 1 ) ψ ( x ) . \begin{align*}
&f_{out}[\phi_1,\dots,\phi_n,\psi,\Gamma_{\mathrm{s}}]\\
&=\cfrac{2\delta}{R}\sum_{i=1}^n\nabla\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\cdot\nabla\phi_i(x)+\cfrac{\delta^2}{R^2}\sum_{i=1}^n\Delta\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\phi_i(x)\\
&\quad\;\,-\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\left(\gamma^\prime\left(\sum_{i=1}^n\sigma_iU_i\right)-\gamma^\prime\left(U_i\right)\right)\phi_i(x)\\
&\quad\;\,-\left(1-\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\right)F(\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}})(x)\\
&\quad\;\,-\left(1-\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\right)\left(\gamma^\prime\left(\sum_{i=1}^n\sigma_iU_i\right)-1\right)\psi(x).
\end{align*} f o u t [ ϕ 1 , … , ϕ n , ψ , Γ s ] = R 2 δ i = 1 ∑ n ∇ η ( R δ x − ( 0 , δ τ i ) ) ⋅ ∇ ϕ i ( x ) + R 2 δ 2 i = 1 ∑ n Δ η ( R δ x − ( 0 , δ τ i ) ) ϕ i ( x ) − i = 1 ∑ n η ( R δ x − ( 0 , δ τ i ) ) ( γ ′ ( i = 1 ∑ n σ i U i ) − γ ′ ( U i ) ) ϕ i ( x ) − ( 1 − i = 1 ∑ n η ( R δ x − ( 0 , δ τ i ) ) ) F ( τ , u , Γ s ) ( x ) − ( 1 − i = 1 ∑ n η ( R δ x − ( 0 , δ τ i ) ) ) ( γ ′ ( i = 1 ∑ n σ i U i ) − 1 ) ψ ( x ) .
then
ψ = ( L o u t ) − 1 ( f o u t [ ϕ 1 , … , ϕ n , ψ , Γ s ] ) . \psi=(L_{out})^{-1}\left(f_{out}[\phi_1,\dots,\phi_n,\psi,\Gamma_{\mathrm{s}}]\right). ψ = ( L o u t ) − 1 ( f o u t [ ϕ 1 , … , ϕ n , ψ , Γ s ] ) .
First we estimate F ( τ , u , Γ s ) F(\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}}) F ( τ , u , Γ s ) . Using γ ( 0 ) = 0 \gamma(0)=0 γ ( 0 ) = 0 and γ ′ ( 0 ) = 1 \gamma'(0)=1 γ ′ ( 0 ) = 1 , we have γ ( s ) = s + O ( s 2 ) \gamma(s)=s+O(s^2) γ ( s ) = s + O ( s 2 ) for ∣ s ∣ |s| ∣ s ∣ sufficiently small. From the definition of F F F , the linear terms cancel out, then
F ( τ , 0 , 0 ) = γ ( ω τ ) − ∑ i = 1 n σ i γ ( U i ) + ∑ i = 1 n σ i ( U i ) b c ≲ ∑ i ≠ j ∣ U i U j ∣ + ∑ i = 1 n ∑ j = 1 n ∣ U i ( U j ) b c ∣ + ∑ i = 1 n ∑ j = 1 n ∣ ( U i ) b c ( U j ) b c ∣ . \begin{equation*}
\begin{aligned}
F(\boldsymbol{\tau}, 0, 0)&=\gamma(\omega_{\boldsymbol{\tau}})-\sum_{i=1}^n\sigma_i\gamma(U_i)+\sum_{i=1}^n\sigma_i(U_i)_\mathrm{bc}\\
&\lesssim\sum_{i\neq j}\left|U_iU_j\right|+\sum_{i=1}^n\sum_{j=1}^n\left|U_i(U_j)_\mathrm{bc}\right|+\sum_{i=1}^n\sum_{j=1}^n\left|(U_i)_\mathrm{bc}(U_j)_\mathrm{bc}\right|.
\end{aligned}
\end{equation*} F ( τ , 0 , 0 ) = γ ( ω τ ) − i = 1 ∑ n σ i γ ( U i ) + i = 1 ∑ n σ i ( U i ) bc ≲ i = j ∑ ∣ U i U j ∣ + i = 1 ∑ n j = 1 ∑ n ∣ U i ( U j ) bc ∣ + i = 1 ∑ n j = 1 ∑ n ∣ ( U i ) bc ( U j ) bc ∣ .
Let ∣ τ ∣ : = max { ∣ τ 1 ∣ , … , ∣ τ n ∣ } |\boldsymbol{\tau}|:=\max\{|\tau_1|,\dots,|\tau_n|\} ∣ τ ∣ := max { ∣ τ 1 ∣ , … , ∣ τ n ∣ } , d m : = min 1 ≤ i < j ≤ n ∣ τ i − τ j ∣ d_m:=\displaystyle\min_{1\leq i< j\leq n}|\tau_i-\tau_j| d m := 1 ≤ i < j ≤ n min ∣ τ i − τ j ∣ and S i : = { x ∈ S δ ∣ ∣ x 2 − τ i δ ∣ ≤ d m 2 δ } \mathcal{S}_i:=\left\{x\in\mathcal{S}_\delta\;\left|\;\left|x_2-\frac {\tau_i}\delta\right|\leq\frac{d_m}{2\delta}\right. \right\} S i := { x ∈ S δ x 2 − δ τ i ≤ 2 δ d m } . According to Assumption assumption:A and Lemma 2.1 , for any j ≠ i j\ne i j = i , 0 ≤ m ≤ k − 2 0\le m\le k-2 0 ≤ m ≤ k − 2 and x ∈ S i x \in \mathcal{S}_i x ∈ S i ,
∣ ∂ m ( U i U j ) ∣ ≲ ( 1 + ∣ x − τ i δ e 2 ∣ ) − 1 2 ∣ x − τ j δ e 2 ∣ − 1 2 e − ( ∣ x − τ i δ e 2 ∣ + ∣ x − τ j δ e 2 ∣ ) . \left|\partial^m\left(U_iU_j\right)\right|\lesssim\left(1+\left|x-\frac{\tau_i}\delta e_2\right|\right)^{-\frac 12}\left|x-\frac{\tau_j}\delta e_2\right|^{-\frac 12}{\mathrm e}^{-\left(\left|x-\frac {\tau_i}\delta e_2\right|+\left|x-\frac{\tau_j}\delta e_2\right|\right)}. ∣ ∂ m ( U i U j ) ∣ ≲ ( 1 + x − δ τ i e 2 ) − 2 1 x − δ τ j e 2 − 2 1 e − ( ∣ x − δ τ i e 2 ∣ + ∣ x − δ τ j e 2 ∣ ) .
For ∣ x 1 ∣ ≤ 4 δ |x_1|\leq\frac 4\delta ∣ x 1 ∣ ≤ δ 4 and ∣ τ i − τ j ∣ δ ≤ 2 δ \frac{|\tau_i-\tau_j|}\delta\leq\frac{2}{\delta} δ ∣ τ i − τ j ∣ ≤ δ 2 , from the convexity of the function f ( x ) = a 2 + x 2 f(x)=\sqrt{a^2+x^2} f ( x ) = a 2 + x 2
∣ x − τ i δ e 2 ∣ + ∣ x − τ j δ e 2 ∣ = x 1 2 + ( x 2 − τ i δ ) 2 + x 1 2 + ( x 2 − τ j δ ) 2 ≥ 4 x 1 2 + ( τ i − τ j δ ) 2 . \begin{align*}
\left|x-\frac{\tau_i}\delta e_2\right|+\left|x-\frac{\tau_j}\delta e_2\right|&=\sqrt{x_1^2+\left(x_2-\frac {\tau_i}\delta\right)^2}+\sqrt{x_1^2+\left(x_2-\frac {\tau_j}\delta\right)^2}\\
&\geq\sqrt{4x_1^2+\left(\frac {\tau_i-\tau_j}{\delta}\right)^2}.
\end{align*} x − δ τ i e 2 + x − δ τ j e 2 = x 1 2 + ( x 2 − δ τ i ) 2 + x 1 2 + ( x 2 − δ τ j ) 2 ≥ 4 x 1 2 + ( δ τ i − τ j ) 2 .
Since
4 x 1 2 + ( τ i − τ j δ ) 2 − ∣ τ i − τ j δ ∣ = 4 x 1 2 4 x 1 2 + ( τ i − τ j δ ) 2 + ∣ τ i − τ j δ ∣ ≥ 2 δ x 1 2 17 + 1 , \begin{align*}
\sqrt{4x_1^2+\left(\frac{\tau_i-\tau_j}{\delta}\right)^2}-\left|\frac{\tau_i-\tau_j}{\delta}\right|=\frac{4x_1^2}{\sqrt{4x_1^2+\left(\frac{\tau_i-\tau_j}{\delta}\right)^2}+\left|\frac{\tau_i-\tau_j}{\delta}\right|}\geq\frac{2\delta x_1^2}{\sqrt{17}+1},
\end{align*} 4 x 1 2 + ( δ τ i − τ j ) 2 − δ τ i − τ j = 4 x 1 2 + ( δ τ i − τ j ) 2 + δ τ i − τ j 4 x 1 2 ≥ 17 + 1 2 δ x 1 2 ,
we have
∣ x − τ i δ e 2 ∣ + ∣ x − τ j δ e 2 ∣ ≥ ∣ τ i − τ j δ ∣ + 2 δ x 1 2 17 + 1 . \left|x-\frac{\tau_i}\delta e_2\right|+\left|x-\frac{\tau_j}\delta e_2\right|\geq\left|\frac{\tau_i-\tau_j}{\delta}\right|+\frac{2\delta x_1^2}{\sqrt{17}+1}. x − δ τ i e 2 + x − δ τ j e 2 ≥ δ τ i − τ j + 17 + 1 2 δ x 1 2 .
Therefore,
∫ S i ( 1 + ∣ x − τ i δ e 2 ∣ ) − 1 ∣ x − τ j δ e 2 ∣ − 1 e − 2 ( ∣ x − τ i δ e 2 ∣ + ∣ x − τ j δ e 2 ∣ ) d x ≲ ∫ { ∣ x 1 ∣ ≤ 4 δ } ∫ 1 δ ( τ i − d m 2 ) 1 δ ( τ i + d m 2 ) ( 1 + ∣ x − τ i δ e 2 ∣ ) − 1 ∣ x − τ j δ e 2 ∣ − 1 e − 2 ( ∣ τ i − τ j ∣ δ + 2 17 + 1 δ x 1 2 ) d x 2 d x 1 + ∫ { ∣ x 1 ∣ ≥ 4 δ } ∫ 1 δ ( τ i − d m 2 ) 1 δ ( τ i + d m 2 ) ( 1 + ∣ x − τ i δ e 2 ∣ ) − 1 ∣ x − τ j δ e 2 ∣ − 1 e − 2 ( 4 δ + ∣ x − τ j δ e 2 ∣ ) d x 2 d x 1 ≲ ∫ { ∣ x 1 ∣ ≤ 4 δ } ∫ 0 d m 2 δ δ ( 1 + x 2 ) − 1 e − 2 ( ∣ τ i − τ j ∣ δ + 2 17 + 1 δ x 1 2 ) d x 2 d x 1 + δ e − 8 δ ≲ δ 1 2 ∣ log δ ∣ e − 2 ∣ τ i − τ j ∣ δ ≲ δ 1 2 ∣ log δ ∣ e − 2 d m δ . \begin{align*}
&\int_{\mathcal{S}_i}\left(1+\left|x-\tfrac{\tau_i}\delta e_2\right|\right)^{-1} \left|x-\tfrac{\tau_j}\delta e_2\right|^{-1} {\mathrm e}^{-2\left(\left|x-\frac {\tau_i}\delta e_2\right|+\left|x-\frac{\tau_j}\delta e_2\right|\right)}\,dx\\
&\lesssim\int_{\left\{|x_1|\leq\frac 4\delta\right\}}\int_{\frac 1\delta(\tau_i-\frac{d_m}{2})}^{\frac 1\delta(\tau_i+\frac{d_m}{2})}\left(1+\left|x-\tfrac{\tau_i}\delta e_2\right|\right)^{-1} \left|x-\tfrac{\tau_j}\delta e_2\right|^{-1}{\mathrm e}^{-2\left(\frac{|\tau_i-\tau_j|}\delta+\frac{2}{\sqrt{17}+1} \delta x_1^2\right)}\,dx_2\,dx_1\\
&\quad\,+\int_{\left\{|x_1|\geq\frac 4\delta\right\}}\int_{\frac 1\delta(\tau_i-\frac{d_m}{2})}^{\frac 1\delta(\tau_i+\frac{d_m}{2})}\left(1+\left|x-\tfrac{\tau_i}\delta e_2\right|\right)^{-1} \left|x-\tfrac{\tau_j}\delta e_2\right|^{-1}{\mathrm e}^{-2\left(\frac{4}\delta+\left|x-\frac{\tau_j}\delta e_2\right|\right)}\,dx_2\,dx_1\\
&\lesssim\int_{\left\{|x_1|\leq\frac 4\delta\right\}}\int_0^{\frac{d_m}{2\delta}} \delta(1+x_2)^{-1}{\mathrm e}^{-2\left(\frac{|\tau_i-\tau_j|}\delta+\frac{2}{\sqrt{17}+1} \delta x_1^2\right)}\,dx_2\,dx_1+\delta{\mathrm e}^{-\frac{8}\delta}\\
&\lesssim\delta^{\frac 12}\left|\log\delta\right|{\mathrm e}^{-\frac{2|\tau_i-\tau_j|}{\delta}}\lesssim\delta^{\frac 12}\left|\log\delta\right|{\mathrm e}^{-\frac{2d_m}{\delta}}.
\end{align*} ∫ S i ( 1 + x − δ τ i e 2 ) − 1 x − δ τ j e 2 − 1 e − 2 ( ∣ x − δ τ i e 2 ∣ + ∣ x − δ τ j e 2 ∣ ) d x ≲ ∫ { ∣ x 1 ∣ ≤ δ 4 } ∫ δ 1 ( τ i − 2 d m ) δ 1 ( τ i + 2 d m ) ( 1 + x − δ τ i e 2 ) − 1 x − δ τ j e 2 − 1 e − 2 ( δ ∣ τ i − τ j ∣ + 17 + 1 2 δ x 1 2 ) d x 2 d x 1 + ∫ { ∣ x 1 ∣ ≥ δ 4 } ∫ δ 1 ( τ i − 2 d m ) δ 1 ( τ i + 2 d m ) ( 1 + x − δ τ i e 2 ) − 1 x − δ τ j e 2 − 1 e − 2 ( δ 4 + ∣ x − δ τ j e 2 ∣ ) d x 2 d x 1 ≲ ∫ { ∣ x 1 ∣ ≤ δ 4 } ∫ 0 2 δ d m δ ( 1 + x 2 ) − 1 e − 2 ( δ ∣ τ i − τ j ∣ + 17 + 1 2 δ x 1 2 ) d x 2 d x 1 + δ e − δ 8 ≲ δ 2 1 ∣ log δ ∣ e − δ 2∣ τ i − τ j ∣ ≲ δ 2 1 ∣ log δ ∣ e − δ 2 d m .
This implies
∥ U i U j ∥ H k − 2 ( S i ) 2 ≲ δ 1 2 ∣ log δ ∣ e − 2 d m δ . \begin{equation}
\|U_iU_j\|_{H^{k-2}(\mathcal{S}_i)}^2\lesssim\delta^{\frac 12}|\log\delta|e^{-\frac{2d_m}\delta}. \tag*{(4.7)}
\end{equation} ∥ U i U j ∥ H k − 2 ( S i ) 2 ≲ δ 2 1 ∣ log δ ∣ e − δ 2 d m . (4.7)
By (2.26) , (2.27) and the derivation of (4.7) , it holds that
∥ U i ( U j ) b c ∥ H k − 2 ( S i ) ≲ ∥ U i U j ( ⋅ , 2 δ − ⋅ ) ∥ H k − 2 ( S i ) + ∥ U i U j ( ⋅ , − 2 δ − ⋅ ) ∥ H k − 2 ( S i ) + ∥ U i R j ∥ H k − 2 ( S i ) ≲ δ 1 4 ∣ log δ ∣ 1 2 e − 2 − τ i − τ j δ + δ 1 4 ∣ log δ ∣ 1 2 e − 2 + τ i + τ j δ + δ 1 4 e − 2 − 2 ∣ τ j ∣ δ ≲ δ 1 4 ∣ log δ ∣ 1 2 e − 2 − 2 ∣ τ ∣ δ . \begin{align*}
&\|U_i(U_j)_\mathrm{bc}\|_{H^{k-2}(\mathcal{S}_i)}\\
&\lesssim\|U_iU_j\left(\,\cdot\,,\tfrac2\delta-\,\cdot\,\right)\|_{H^{k-2}(\mathcal{S}_i)}+\|U_iU_j\left(\,\cdot\,,-\tfrac2\delta-\,\cdot\,\right)\|_{H^{k-2}(\mathcal{S}_i)}+\|U_i\mathcal{R}_j\|_{H^{k-2}(\mathcal{S}_i)}\\
&\lesssim\delta^{\frac 14}\left|\log\delta\right|^{\frac{1}{2}}{\mathrm e}^{-\frac{2-\tau_i-\tau_j}\delta}+\delta^{\frac 14}\left|\log\delta\right|^{\frac{1}{2}}{\mathrm e}^{-\frac{2+\tau_i+\tau_j}\delta}+\delta^{\frac{1}{4}}{\mathrm e}^{-\frac{2-2|\tau_j|}{\delta}}\lesssim\delta^{\frac 14}\left|\log\delta\right|^{\frac{1}{2}}{\mathrm e}^{-\frac{2-2|\boldsymbol{\tau}|}{\delta}}.
\end{align*} ∥ U i ( U j ) bc ∥ H k − 2 ( S i ) ≲ ∥ U i U j ( ⋅ , δ 2 − ⋅ ) ∥ H k − 2 ( S i ) + ∥ U i U j ( ⋅ , − δ 2 − ⋅ ) ∥ H k − 2 ( S i ) + ∥ U i R j ∥ H k − 2 ( S i ) ≲ δ 4 1 ∣ log δ ∣ 2 1 e − δ 2 − τ i − τ j + δ 4 1 ∣ log δ ∣ 2 1 e − δ 2 + τ i + τ j + δ 4 1 e − δ 2 − 2∣ τ j ∣ ≲ δ 4 1 ∣ log δ ∣ 2 1 e − δ 2 − 2∣ τ ∣ .
The remaining terms are estimated similarly, finally we have
∥ F ( τ , 0 , 0 ) ∥ H k − 2 ( S δ ) ≲ δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ 1 4 ∣ log δ ∣ 1 2 e − 2 ( 1 − ∣ τ ∣ ) δ . \begin{equation}
\|F(\boldsymbol{\tau},0,0)\|_{H^{k-2}(\mathcal{S}_\delta)}\lesssim \delta^{\frac 14}\left|\log\delta\right|^{\frac{1}{2}}{\mathrm e}^{-\frac{d_m}\delta}+\delta^{\frac 14}\left|\log\delta\right|^{\frac{1}{2}}{\mathrm e}^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}. \tag*{(4.8)}
\end{equation} ∥ F ( τ , 0 , 0 ) ∥ H k − 2 ( S δ ) ≲ δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ 4 1 ∣ log δ ∣ 2 1 e − δ 2 ( 1 − ∣ τ ∣ ) . (4.8)
Next we turn to D u F D_uF D u F and D Γ s F D_{\Gamma_{\mathrm{s}}}F D Γ s F . Here the operators D u F D_uF D u F and D Γ s F D_{\Gamma_{\mathrm{s}}}F D Γ s F are defined respectively as
D u F ( τ , u , Γ s ) [ v ] = d d t F ( τ , u + t v , Γ s ) ∣ t = 0 , v ∈ H k ( S δ ) D_uF(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}})[v]=\frac{\mathrm{d}}{\mathrm{d}t}F(\boldsymbol{\tau},u+tv,\Gamma_{\mathrm{s}})\bigg|_{t=0},\qquad v\in H^k(\mathcal{S}_\delta) D u F ( τ , u , Γ s ) [ v ] = d t d F ( τ , u + t v , Γ s ) t = 0 , v ∈ H k ( S δ )
and
D Γ s F ( τ , u , Γ s ) [ Λ s ] = d d t F ( τ , u , Γ s + t Λ s ) ∣ t = 0 , Λ s ∈ H k ( R ) . D_{\Gamma_{\mathrm{s}}}F(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}})[\Lambda_{\mathrm{s}}]=\frac{\mathrm{d}}{\mathrm{d}t}F(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}}+t\Lambda_{\mathrm{s}})\bigg|_{t=0},\qquad \Lambda_{\mathrm{s}}\in H^k( {\mathbb{R}}). D Γ s F ( τ , u , Γ s ) [ Λ s ] = d t d F ( τ , u , Γ s + t Λ s ) t = 0 , Λ s ∈ H k ( R ) .
For any v ∈ H e k ( S δ ) v \in H_{\mathrm{e}}^k(\mathcal{S}_\delta) v ∈ H e k ( S δ ) ,
D u F ( τ , u , Γ s ) v = ( ∣ 1 + Γ ′ ( δ ⋅ ) ∣ 2 γ ′ ( ω τ + u ) − γ ′ ( ∑ i = 1 n σ i U i ) ) v = ( 2 Γ 1 ′ ( δ ⋅ ) + ∣ Γ ′ ( δ ⋅ ) ∣ 2 ) γ ′ ( ω τ + u ) v + ( ( u − ∑ i = 1 n σ i ( U i ) b c ) ∫ 0 1 γ ′ ′ ( ∑ i = 1 n σ i U i + s ( u − ∑ i = 1 n σ i ( U i ) b c ) ) d s ) v . \begin{equation*}
\begin{split}
&D_u F(\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}}) v= \left(\left|1+\Gamma^\prime(\delta \,\cdot\, )\right|^2 \gamma'(\omega_{\boldsymbol{\tau}}+u)- \gamma^\prime\left(\sum_{i=1}^n\sigma_iU_i\right)\right) v \\
&=\left(2 \Gamma_1' (\delta \,\cdot\,)+ |\Gamma'(\delta \,\cdot\,)|^2\right)\gamma'(\omega_{\boldsymbol{\tau}}+u)v\\
&\quad\,\,+ \left(\left(u - \sum_{i=1}^n\sigma_i(U_i)_\mathrm{bc}\right)\displaystyle\int_0^1 \gamma''\left(\sum_{i=1}^n\sigma_iU_i + s\left(u - \sum_{i=1}^n\sigma_i(U_i)_\mathrm{bc}\right) \right) \, ds\right) v.
\end{split}
\end{equation*} D u F ( τ , u , Γ s ) v = ( ∣ 1 + Γ ′ ( δ ⋅ ) ∣ 2 γ ′ ( ω τ + u ) − γ ′ ( i = 1 ∑ n σ i U i ) ) v = ( 2 Γ 1 ′ ( δ ⋅ ) + ∣ Γ ′ ( δ ⋅ ) ∣ 2 ) γ ′ ( ω τ + u ) v + ( ( u − i = 1 ∑ n σ i ( U i ) bc ) ∫ 0 1 γ ′′ ( i = 1 ∑ n σ i U i + s ( u − i = 1 ∑ n σ i ( U i ) bc ) ) d s ) v .
By the the scaling identity ∥ f ( δ ⋅ ) ∥ H ˙ s ( S δ ) = δ s − 1 ∥ f ∥ H ˙ s ( S 1 ) \| f(\delta \,\cdot\,) \|_{\dot H^s(\mathcal{S}_\delta)} = \delta^{s-1} \| f \|_{\dot H^s(\mathcal{S}_1)} ∥ f ( δ ⋅ ) ∥ H ˙ s ( S δ ) = δ s − 1 ∥ f ∥ H ˙ s ( S 1 ) and the Sobolev trace theorem, it holds that
∥ 2 Γ 1 ′ ( δ ⋅ ) + ∣ Γ ′ ( δ ⋅ ) ∣ 2 ∥ H k − 1 2 ( S δ ) ≲ δ − 1 ∥ Γ s ∥ H k ( R ) . \left \| 2 \Gamma_1^\prime(\delta \,\cdot\,) + |\Gamma^\prime(\delta \,\cdot\,) |^2 \right\|_{H^{k-\frac{1}{2}}(\mathcal{S}_\delta)} \lesssim \delta^{-1} \|\Gamma_{\mathrm{s}}\|_{H^k(\mathbb{R})}. 2 Γ 1 ′ ( δ ⋅ ) + ∣ Γ ′ ( δ ⋅ ) ∣ 2 H k − 2 1 ( S δ ) ≲ δ − 1 ∥ Γ s ∥ H k ( R ) .
When l > 1 l>1 l > 1 , H l ( S δ ) H^l(\mathcal{S}_\delta) H l ( S δ ) is an algebra, thus for k ≥ 3 2 k\geq\frac{3}{2} k ≥ 2 3 ,
∥ D u F ( τ , u , Γ s ) v ∥ H k − 2 ( S δ ) ≲ ∥ D u F ( τ , u , Γ s ) v ∥ H k − 1 2 ( S δ ) ≲ ( δ − 1 ∥ Γ s ∥ H k ( R ) + ∥ u ∥ H k ( S δ ) + ∑ i = 1 n ∥ ( U i ) b c ∥ H k ( S δ ) ) ∥ v ∥ H k ( S δ ) . \begin{equation}
\begin{aligned}
\|D_u F(\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}}) v\|_{H^{k-2} (\mathcal{S}_\delta)}&\lesssim\|D_u F(\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}}) v\|_{H^{k-\frac{1}{2}} (\mathcal{S}_\delta)}\\
&\lesssim\left( \delta^{-1} \|\Gamma_{\mathrm{s}}\|_{H^{k}( {\mathbb{R}})} + \|u\|_{H^{k} (\mathcal{S}_\delta)} + \sum_{i=1}^n\|(U_i)_\mathrm{bc}\|_{H^{k} (\mathcal{S}_\delta)}\right) \|v\|_{H^k (\mathcal{S}_\delta)}.
\end{aligned} \tag*{(4.9)}
\end{equation} ∥ D u F ( τ , u , Γ s ) v ∥ H k − 2 ( S δ ) ≲ ∥ D u F ( τ , u , Γ s ) v ∥ H k − 2 1 ( S δ ) ≲ ( δ − 1 ∥ Γ s ∥ H k ( R ) + ∥ u ∥ H k ( S δ ) + i = 1 ∑ n ∥ ( U i ) bc ∥ H k ( S δ ) ) ∥ v ∥ H k ( S δ ) . (4.9)
Similarly regarding D Γ s F D_{\Gamma_{\mathrm{s}}} F D Γ s F , we have for any Λ s ∈ H e k ( R ) \Lambda_{\mathrm{s}} \in H_{\mathrm{e}}^k(\mathbb{R}) Λ s ∈ H e k ( R ) that
D Γ s F ( τ , u , Γ s ) Λ s = 2 ( Λ 1 ′ ( δ ⋅ ) + Γ ′ ( δ ⋅ ) ⋅ Λ ′ ( δ ⋅ ) ) γ ( ω τ + u ) , D_{\Gamma_{\mathrm{s}}} F (\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}})\Lambda_{\mathrm{s}} = 2 \left( \Lambda_1' (\delta \,\cdot\,) + \Gamma' (\delta \,\cdot\,) \cdot \Lambda' (\delta \,\cdot\,) \right) \gamma (\omega_{\boldsymbol{\tau}}+u), D Γ s F ( τ , u , Γ s ) Λ s = 2 ( Λ 1 ′ ( δ ⋅ ) + Γ ′ ( δ ⋅ ) ⋅ Λ ′ ( δ ⋅ ) ) γ ( ω τ + u ) ,
where Λ = Λ 1 + i Λ 2 \Lambda = \Lambda_1 + \mathrm i\Lambda_2 Λ = Λ 1 + i Λ 2 is the complex holomorphic function obtained by Λ s \Lambda_{\mathrm{s}} Λ s . Similarly, by the Sobolev trace theorem and the scaling identity above, for k ≥ 3 2 k\geq\frac{3}{2} k ≥ 2 3 ,
∥ D Γ s F ( τ , u , Γ s ) Λ s ∥ H k − 2 ( S δ ) ≲ ∥ D Γ s F ( τ , u , Γ s ) Λ s ∥ H k − 1 2 ( S δ ) ≲ δ − 1 ∥ Λ s ∥ H k ( R ) . \begin{equation}
\|D_{\Gamma_{\mathrm{s}}} F (\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}}) \Lambda_{\mathrm{s}}\|_{H^{k-2} (\mathcal{S}_\delta)}\lesssim\|D_{\Gamma_{\mathrm{s}}} F (\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}}) \Lambda_{\mathrm{s}}\|_{H^{k-\frac{1}{2}} (\mathcal{S}_\delta)}\lesssim\delta^{-1} \|\Lambda_{\mathrm{s}}\|_{H^{k}( {\mathbb{R}})}. \tag*{(4.10)}
\end{equation} ∥ D Γ s F ( τ , u , Γ s ) Λ s ∥ H k − 2 ( S δ ) ≲ ∥ D Γ s F ( τ , u , Γ s ) Λ s ∥ H k − 2 1 ( S δ ) ≲ δ − 1 ∥ Λ s ∥ H k ( R ) . (4.10)
With these estimates, for given solutions ϕ i \phi_i ϕ i , i = 1 , … , n i=1,\dots,n i = 1 , … , n of the inner problems, we can find ψ \psi ψ solving (4.6) .
Proposition. Assume ∥ ϕ i ∥ H k ( S δ ) ≤ δ k + 1 \|\phi_i\|_{H^{k}(\mathcal{S}_\delta)}\leq\delta^{k+1} ∥ ϕ i ∥ H k ( S δ ) ≤ δ k + 1 and ∥ ϕ i ∥ C 1 , α ( B i , 2 ) ≤ δ k + 1 \left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})}\leq\delta^{k+1} ∥ ϕ i ∥ C 1 , α ( B i , 2 ) ≤ δ k + 1 for all i = 1 , … , n i=1,\dots,n i = 1 , … , n . Then there exist constants C ~ > 0 \widetilde C>0 C > 0 and δ 0 > 0 \delta_0>0 δ 0 > 0 such that for τ ∈ P ε \boldsymbol{\tau} \in P_{\varepsilon} τ ∈ P ε and δ ∈ ( 0 , δ 0 ) \delta\in(0,\delta_0) δ ∈ ( 0 , δ 0 ) , the operator T o u t [ ψ ] : = ( L o u t ) − 1 f o u t [ ϕ 1 , … , ϕ n , ψ , Γ ~ s [ u ] ] \mathrm{T}_{out}[\psi]:=(L_{out})^{-1}f_{out}[\phi_1,\dots,\phi_n,\psi,\widetilde{\Gamma}_{\mathrm{s}}[u]] T o u t [ ψ ] := ( L o u t ) − 1 f o u t [ ϕ 1 , … , ϕ n , ψ , Γ s [ u ]] has a fixed point ψ ~ [ ϕ 1 , … , ϕ n ] \widetilde{\psi}[\phi_1,\dots,\phi_n] ψ [ ϕ 1 , … , ϕ n ] in the set B o u t : = { ψ ∈ H e k ( S δ ) ∣ ∥ ψ ∥ H k ( S δ ) ≤ δ k + 1 } , B_{out}:=\{\psi \in H_{\mathrm{e}}^k(\mathcal{S}_\delta) \;|\;\|\psi\|_{H^k(\mathcal{S}_\delta)}\leq \delta^{k+1}\}, B o u t := { ψ ∈ H e k ( S δ ) ∣ ∥ ψ ∥ H k ( S δ ) ≤ δ k + 1 } , which satisfies ∥ ψ ~ [ ϕ 1 , … , ϕ n ] ∥ H k ( S δ ) ≲ δ 1 2 ∑ i = 1 n ∥ ϕ i ∥ H k ( S δ ) + δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ , \begin{aligned}
\left\|\widetilde{\psi}[\phi_1,\dots,\phi_n]\right\|_{H^{k}(\mathcal{S}_\delta)}\lesssim\delta^{\frac{1}{2}}\sum_{i=1}^n\left\|\phi_i\right\|_{H^{k}(\mathcal{S}_\delta)}+\delta^{\frac 14} | \log \delta|^{\frac{1}{2}}e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta},
\end{aligned} ψ [ ϕ 1 , … , ϕ n ] H k ( S δ ) ≲ δ 2 1 i = 1 ∑ n ∥ ϕ i ∥ H k ( S δ ) + δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) , and the following pointwise estimate ∣ ψ ~ [ ϕ 1 , … , ϕ n ] ( x ) ∣ ≤ C ~ { δ − 3 / 2 e − R 2 δ ( e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ + δ − 1 2 − k ∑ i = 1 n ∥ ϕ i ∥ H k ( S δ ) 2 ) + δ − 3 / 2 e − R 2 δ δ ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) + e − d − 2 R δ ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , x ∈ ∪ i = 1 n B i , 1 2 , e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ + δ − 1 2 − k ∑ i = 1 n ∥ ϕ i ∥ H k ( S δ ) 2 + δ ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) + e − d − 2 R δ ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , x ∈ S δ \ ∪ i = 1 n B i , 1 2 , \begin{aligned}
&|\widetilde{\psi}[\phi_1,\dots,\phi_n](x)|\\
&\leq \widetilde C \begin{cases}
\delta^{-3/2}{\mathrm e}^{-\frac{R}{2\delta}}\left(e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{-\frac{1}{2}-k}\sum_{i=1}^n\|\phi_i\|_{H^k (\mathcal{S}_\delta)}^2\right)\\
+\delta^{-3/2}{\mathrm e}^{-\frac{R}{2\delta}}\delta\sum_{i=1}^n\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})}+{\mathrm e}^{-\frac{d-2R}{\delta}}\sum_{i=1}^n\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})},&x\in\cup_{i=1}^nB_{i,\frac{1}{2}},\\[4mm]
e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{-\frac{1}{2}-k}\sum_{i=1}^n\|\phi_i\|_{H^k (\mathcal{S}_\delta)}^2\\
+\delta\sum_{i=1}^n\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})}+{\mathrm e}^{-\frac{d-2R}{\delta}}\sum_{i=1}^n\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})},&x\in \mathcal{S}_\delta\backslash \cup_{i=1}^nB_{i,\frac{1}{2}},
\end{cases}
\end{aligned} ∣ ψ [ ϕ 1 , … , ϕ n ] ( x ) ∣ ≤ C ⎩ ⎨ ⎧ δ − 3/2 e − 2 δ R ( e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) + δ − 2 1 − k ∑ i = 1 n ∥ ϕ i ∥ H k ( S δ ) 2 ) + δ − 3/2 e − 2 δ R δ ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) + e − δ d − 2 R ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) + δ − 2 1 − k ∑ i = 1 n ∥ ϕ i ∥ H k ( S δ ) 2 + δ ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) + e − δ d − 2 R ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , x ∈ ∪ i = 1 n B i , 2 1 , x ∈ S δ \ ∪ i = 1 n B i , 2 1 , where B i , 1 2 : = { x ∈ R 2 ∣ ∣ x − ( 0 , τ i δ ) ∣ ≤ R 2 δ } B_{i,\frac{1}{2}}:=\left\{x\in {\mathbb{R}}^2\;\left|\;\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\leq \frac{R}{2\delta}\right.\right\} B i , 2 1 := { x ∈ R 2 x − ( 0 , δ τ i ) ≤ 2 δ R } .
Proof. We first prove that T o u t [ ψ ] ∈ B o u t \mathrm{T}_{out}[\psi]\in B_{out} T o u t [ ψ ] ∈ B o u t for any ψ ∈ B o u t \psi\in B_{out} ψ ∈ B o u t . From the invertibility of L o u t L_{out} L o u t ,
∥ T o u t [ ψ ] ∥ H k ( S δ ) ≲ ∥ f o u t [ ϕ 1 , … , ϕ n , ψ , Γ ~ s [ u ] ] ∥ H k − 2 ( S δ ) ≲ δ ∑ i = 1 n ∥ ∇ η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ⋅ ∇ ϕ i ( x ) ∥ H k − 2 ( S δ ) + δ 2 ∑ i = 1 n ∥ Δ η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ϕ i ( x ) ∥ H k − 2 ( S δ ) + ∑ i = 1 n ∥ η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ( γ ′ ( ∑ j = 1 n σ j U j ) − γ ′ ( U i ) ) ϕ i ( x ) ∥ H k − 2 ( S δ ) + ∥ ( 1 − ∑ i = 1 n η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ) F ( τ , u , Γ s ) ( x ) ∥ H k − 2 ( S δ ) + ∥ ( 1 − ∑ i = 1 n η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ) ( γ ′ ( ∑ j = 1 n σ j U j ) − 1 ) ψ ( x ) ∥ H k − 2 ( S δ ) . \begin{align*}
&\|\mathrm{T}_{out}[\psi]\|_{H^k (\mathcal{S}_\delta)}\lesssim\left\|f_{out}[\phi_1,\dots,\phi_n,\psi,\widetilde{\Gamma}_{\mathrm{s}}[u]]\right\|_{H^{k-2} (\mathcal{S}_\delta)}\\
&\lesssim\delta\sum_{i=1}^n\left\|\nabla\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\cdot\nabla\phi_i(x)\right\|_{H^{k-2} (\mathcal{S}_\delta)}\\
&\quad\;+\delta^2\sum_{i=1}^n\left\|\Delta\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\phi_i(x)\right\|_{H^{k-2} (\mathcal{S}_\delta)}\\
&\quad\;+\sum_{i=1}^n\left\|\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\left(\gamma^\prime\left(\sum_{j=1}^n\sigma_jU_j\right)-\gamma^\prime\left(U_i\right)\right)\phi_i(x)\right\|_{H^{k-2} (\mathcal{S}_\delta)}\\
&\quad\;+\left\|\left(1-\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\right)F(\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}})(x)\right\|_{H^{k-2} (\mathcal{S}_\delta)}\\
&\quad\;+\left\|\left(1-\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\right)\left(\gamma^\prime\left(\sum_{j=1}^n\sigma_jU_j\right)-1\right)\psi(x)\right\|_{H^{k-2} (\mathcal{S}_\delta)}.
\end{align*} ∥ T o u t [ ψ ] ∥ H k ( S δ ) ≲ f o u t [ ϕ 1 , … , ϕ n , ψ , Γ s [ u ]] H k − 2 ( S δ ) ≲ δ i = 1 ∑ n ∇ η ( R δ x − ( 0 , δ τ i ) ) ⋅ ∇ ϕ i ( x ) H k − 2 ( S δ ) + δ 2 i = 1 ∑ n Δ η ( R δ x − ( 0 , δ τ i ) ) ϕ i ( x ) H k − 2 ( S δ ) + i = 1 ∑ n η ( R δ x − ( 0 , δ τ i ) ) ( γ ′ ( j = 1 ∑ n σ j U j ) − γ ′ ( U i ) ) ϕ i ( x ) H k − 2 ( S δ ) + ( 1 − i = 1 ∑ n η ( R δ x − ( 0 , δ τ i ) ) ) F ( τ , u , Γ s ) ( x ) H k − 2 ( S δ ) + ( 1 − i = 1 ∑ n η ( R δ x − ( 0 , δ τ i ) ) ) ( γ ′ ( j = 1 ∑ n σ j U j ) − 1 ) ψ ( x ) H k − 2 ( S δ ) . Since the support of η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) \eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right) η ( R δ x − ( 0 , δ τ i ) ) is contained in the ball
B i , 2 = { x ∈ R 2 ∣ ∣ x − ( 0 , τ i δ ) ∣ ≤ 2 R δ } , B_{i,2}=\left\{x\in {\mathbb{R}}^2\;\left|\;\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\leq \frac{2R}{\delta}\right.\right\}, B i , 2 = { x ∈ R 2 x − ( 0 , δ τ i ) ≤ δ 2 R } , using the evenness of γ ′ \gamma' γ ′ and the exponential decay of U U U and U b c U_{\mathrm{bc}} U bc , we have
γ ′ ( ∑ j = 1 n σ j U j ) − γ ′ ( U i ) = γ ′ ( ∑ j = 1 n σ j U j ) − γ ′ ( σ i U i ) ≲ ∑ j ≠ i ∣ U j ∣ in B i , 2 . \gamma^\prime\left(\sum_{j=1}^n\sigma_jU_j\right)-\gamma^\prime\left(U_i\right)=\gamma^\prime\left(\sum_{j=1}^n\sigma_jU_j\right)-\gamma^\prime\left(\sigma_iU_i\right)\lesssim
\sum_{j\neq i}|U_j|\quad \text{in }B_{i,2}. γ ′ ( j = 1 ∑ n σ j U j ) − γ ′ ( U i ) = γ ′ ( j = 1 ∑ n σ j U j ) − γ ′ ( σ i U i ) ≲ j = i ∑ ∣ U j ∣ in B i , 2 . Then
∥ η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ( γ ′ ( ∑ j = 1 n σ j U j ) − γ ′ ( U i ) ) ϕ i ∥ H k − 2 ( S δ ) ≲ ( ∑ j ≠ i ∥ U j ∥ H k ( B i , 2 ) ) ∥ ϕ i ∥ H k ( S δ ) . \begin{align*}
&\left\|\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\left(\gamma^\prime\left(\sum_{j=1}^n\sigma_jU_j\right)-\gamma^\prime\left(U_i\right)\right)\phi_i\right\|_{H^{k-2} (\mathcal{S}_\delta)}\\
&\lesssim\left(\sum_{j\neq i}\left\|U_j\right\|_{H^{k} (B_{i,2})}\right)\left\|\phi_i\right\|_{H^{k}(\mathcal{S}_\delta)}.
\end{align*} η ( R δ x − ( 0 , δ τ i ) ) ( γ ′ ( j = 1 ∑ n σ j U j ) − γ ′ ( U i ) ) ϕ i H k − 2 ( S δ ) ≲ j = i ∑ ∥ U j ∥ H k ( B i , 2 ) ∥ ϕ i ∥ H k ( S δ ) . Thus by Lemma 2.1 ,
∥ η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ( γ ′ ( ∑ j = 1 n σ j U j ) − γ ′ ( U i ) ) ϕ i ∥ H k − 2 ( S δ ) ≲ e − d − 2 R δ ∥ ϕ i ∥ H k ( S δ ) , \begin{equation}
\left\|\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\left(\gamma^\prime\left(\sum_{j=1}^n\sigma_jU_j\right)-\gamma^\prime\left(U_i\right)\right)\phi_i\right\|_{H^{k-2} (\mathcal{S}_\delta)}\lesssim{\mathrm e}^{-\frac{d-2R}{\delta}}\left\|\phi_i\right\|_{H^{k}(\mathcal{S}_\delta)}, \tag*{(4.11)}
\end{equation} η ( R δ x − ( 0 , δ τ i ) ) ( γ ′ ( j = 1 ∑ n σ j U j ) − γ ′ ( U i ) ) ϕ i H k − 2 ( S δ ) ≲ e − δ d − 2 R ∥ ϕ i ∥ H k ( S δ ) , (4.11) where d : = min { 2 − 2 ∣ τ ∣ , d m } d:=\min\{2-2|\boldsymbol{\tau}|,d_m\} d := min { 2 − 2∣ τ ∣ , d m } and take R R R small such that d − 2 R > 0 d-2R>0 d − 2 R > 0 .
Similarly, since 1 − ∑ i = 1 n η ( δ R ∣ ξ − ( 0 , τ i δ ) ∣ ) = 0 1-\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|\xi-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)=0 1 − ∑ i = 1 n η ( R δ ξ − ( 0 , δ τ i ) ) = 0 in all ball
B i , 1 : = { x ∈ R 2 ∣ ∣ x − ( 0 , τ i δ ) ∣ ≤ R δ } , i = 1 , … , n , B_{i,1}:=\left\{x\in {\mathbb{R}}^2\;\left|\;\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\leq \frac{R}{\delta}\right.\right\},\quad i=1,\dots,n, B i , 1 := { x ∈ R 2 x − ( 0 , δ τ i ) ≤ δ R } , i = 1 , … , n , using
γ ′ ( ∑ j = 1 n σ j U j ) − 1 ≲ ∑ j = 1 n ∣ U j ∣ in S δ \ ∪ i = 1 n B i , 1 , \gamma^\prime\left(\sum_{j=1}^n\sigma_jU_j\right)-1\lesssim\sum_{j=1}^n|U_j|\quad\text{in }\mathcal{S}_\delta\backslash\cup_{i=1}^nB_{i,1}, γ ′ ( j = 1 ∑ n σ j U j ) − 1 ≲ j = 1 ∑ n ∣ U j ∣ in S δ \ ∪ i = 1 n B i , 1 , we have
∥ ( 1 − ∑ i = 1 n η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) ) ( γ ′ ( ∑ j = 1 n σ j U j ) − 1 ) ψ ( x ) ∥ H k − 2 ( S δ ) ≲ e − R δ ∥ ψ ∥ H k ( S δ ) . \begin{equation}
\left\|\left(1-\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\right)\left(\gamma^\prime\left(\sum_{j=1}^n\sigma_jU_j\right)-1\right)\psi(x)\right\|_{H^{k-2} (\mathcal{S}_\delta)}\lesssim{\mathrm e}^{-\frac {R}\delta}\left\|\psi\right\|_{H^{k}(\mathcal{S}_\delta)}. \tag*{(4.12)}
\end{equation} ( 1 − i = 1 ∑ n η ( R δ x − ( 0 , δ τ i ) ) ) ( γ ′ ( j = 1 ∑ n σ j U j ) − 1 ) ψ ( x ) H k − 2 ( S δ ) ≲ e − δ R ∥ ψ ∥ H k ( S δ ) . (4.12) Thus, by (4.8) , (4.9) , (4.10) and Proposition (?)
∥ T o u t [ ψ ] ∥ H k ( S δ ) ≲ δ ∑ i = 1 n ∥ ϕ i ∥ H k ( S δ ) + δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ + δ − 1 2 − k ∥ u ∥ H k ( S δ ) 2 \|\mathrm{T}_{out}[\psi]\|_{H^k (\mathcal{S}_\delta)}\lesssim\delta\sum_{i=1}^n\left\|\phi_i\right\|_{H^{k}(\mathcal{S}_\delta)}+\delta^{\frac 14} | \log \delta|^{\frac{1}{2}}e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{-\frac{1}{2}-k}\|u\|_{H^k (\mathcal{S}_\delta)}^2 ∥ T o u t [ ψ ] ∥ H k ( S δ ) ≲ δ i = 1 ∑ n ∥ ϕ i ∥ H k ( S δ ) + δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) + δ − 2 1 − k ∥ u ∥ H k ( S δ ) 2 From the above estimates and ∥ ϕ i ∥ H k ( S δ ) ≤ δ k + 1 \|\phi_i\|_{H^{k}(\mathcal{S}_\delta)}\leq\delta^{k+1} ∥ ϕ i ∥ H k ( S δ ) ≤ δ k + 1 for all i = 1 , … , n i=1,\dots,n i = 1 , … , n , choosing δ \delta δ small, then T o u t [ ψ ] ∈ B o u t \mathrm{T}_{out}[\psi]\in B_{out} T o u t [ ψ ] ∈ B o u t .
Next we prove T o u t \mathrm{T}_{out} T o u t is a contraction on B o u t B_{out} B o u t . By (4.9) , (4.10) , (4.5) and (4.12)
∥ T o u t [ ψ 1 ] − T o u t [ ψ 2 ] ∥ H k ( S δ ) ≲ ∥ f o u t [ ϕ 1 , … , ϕ n , ψ 1 , Γ ~ s [ u ] ] − f o u t [ ϕ 1 , … , ϕ n , ψ 2 , Γ ~ s [ u ] ] ∥ H k − 2 ( S δ ) ≲ ∥ D u F ( τ , ζ , Γ ~ s ) ( ψ 1 − ψ 2 ) ∥ H k − 2 ( S δ ) + ∥ D Γ s F ( τ , u , ζ ) D u Γ ~ s [ ζ ] ( ψ 1 − ψ 2 ) ∥ H k − 2 ( S δ ) + ∥ ( 1 − ∑ i = 1 n η ( δ R ∣ ξ − ( 0 , τ i δ ) ∣ ) ) ( γ ′ ( ∑ j = 1 n σ j U j ) − 1 ) ( ψ 1 − ψ 2 ) ∥ H k − 2 ( S δ ) ≲ ( δ − 1 ∥ Γ ~ s ∥ H k ( R ) + ∥ u ∥ H k ( S δ ) + ∑ i = 1 n ∥ ( U i ) b c ∥ H k ( S δ ) ) ∥ ψ 1 − ψ 2 ∥ H k ( S δ ) + e − R δ ∥ ψ 1 − ψ 2 ∥ H k ( S δ ) + δ − 1 ∥ Γ ~ s ∥ H k ( R ) ( δ 1 2 − k ∥ u ∥ H k ( S δ ) + δ 3 4 − k e − 1 − ∣ τ ∣ δ ) ∥ ψ 1 − ψ 2 ∥ H k ( S δ ) . \begin{align*}
&\|\mathrm{T}_{out}[\psi_1]-\mathrm{T}_{out}[\psi_2]\|_{H^{k}(\mathcal{S}_\delta)}\\
&\lesssim\left\|f_{out}[\phi_1,\dots,\phi_n,\psi_1,\widetilde{\Gamma}_{\mathrm{s}}[u]]-f_{out}[\phi_1,\dots,\phi_n,\psi_2,\widetilde{\Gamma}_{\mathrm{s}}[u]]\right\|_{H^{k-2}(\mathcal{S}_\delta)}\\
&\lesssim\left\|D_u F(\boldsymbol{\tau}, \zeta, \widetilde{\Gamma}_{\mathrm{s}})(\psi_1-\psi_2)\right\|_{H^{k-2}(\mathcal{S}_\delta)}+\left\|D_{\Gamma_{\mathrm{s}}} F (\boldsymbol{\tau}, u, \zeta)D_u \widetilde{\Gamma}_{\mathrm{s}}[\zeta](\psi_1-\psi_2)\right\|_{H^{k-2}(\mathcal{S}_\delta)}\\
&\quad\;+\left\|\left(1-\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|\xi-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\right)\left(\gamma^\prime\left(\sum_{j=1}^n\sigma_jU_j\right)-1\right)(\psi_1-\psi_2)\right\|_{H^{k-2}(\mathcal{S}_\delta)}\\
&\lesssim\left(\delta^{-1} \left\|\widetilde{\Gamma}_{\mathrm{s}}\right\|_{H^{k}( {\mathbb{R}})} + \|u\|_{H^{k} (\mathcal{S}_\delta)} + \sum_{i=1}^n\|(U_i)_\mathrm{bc}\|_{H^{k} (\mathcal{S}_\delta)}\right)\|\psi_1-\psi_2\|_{H^{k}(\mathcal{S}_\delta)}+e^{-\frac {R}\delta}\|\psi_1-\psi_2\|_{H^{k} (\mathcal{S}_\delta)}\\
&\quad\;+\delta^{-1} \left\|\widetilde{\Gamma}_{\mathrm{s}}\right\|_{H^{k}( {\mathbb{R}})}\left(\delta^{\frac 12-k}\|u\|_{H^k (\mathcal{S}_\delta)} + \delta^{\frac 34-k} e^{-\frac {1-|\boldsymbol{\tau}|}\delta} \right)\|\psi_1-\psi_2\|_{H^{k} (\mathcal{S}_\delta)}.
\end{align*} ∥ T o u t [ ψ 1 ] − T o u t [ ψ 2 ] ∥ H k ( S δ ) ≲ f o u t [ ϕ 1 , … , ϕ n , ψ 1 , Γ s [ u ]] − f o u t [ ϕ 1 , … , ϕ n , ψ 2 , Γ s [ u ]] H k − 2 ( S δ ) ≲ D u F ( τ , ζ , Γ s ) ( ψ 1 − ψ 2 ) H k − 2 ( S δ ) + D Γ s F ( τ , u , ζ ) D u Γ s [ ζ ] ( ψ 1 − ψ 2 ) H k − 2 ( S δ ) + ( 1 − i = 1 ∑ n η ( R δ ξ − ( 0 , δ τ i ) ) ) ( γ ′ ( j = 1 ∑ n σ j U j ) − 1 ) ( ψ 1 − ψ 2 ) H k − 2 ( S δ ) ≲ ( δ − 1 Γ s H k ( R ) + ∥ u ∥ H k ( S δ ) + i = 1 ∑ n ∥ ( U i ) bc ∥ H k ( S δ ) ) ∥ ψ 1 − ψ 2 ∥ H k ( S δ ) + e − δ R ∥ ψ 1 − ψ 2 ∥ H k ( S δ ) + δ − 1 Γ s H k ( R ) ( δ 2 1 − k ∥ u ∥ H k ( S δ ) + δ 4 3 − k e − δ 1 − ∣ τ ∣ ) ∥ ψ 1 − ψ 2 ∥ H k ( S δ ) . From Corollary (?) and Proposition (?) , choosing δ \delta δ small, then T o u t \mathrm{T}_{out} T o u t is a contraction mapping on B o u t B_{out} B o u t . Hence T o u t \mathrm{T}_{out} T o u t possesses a fixed point ψ ~ [ ϕ 1 , … , ϕ n ] \widetilde{\psi}[\phi_1,\dots,\phi_n] ψ [ ϕ 1 , … , ϕ n ] in B o u t B_{out} B o u t . Moreover, by the above estimates we have
∥ ψ ~ [ ϕ 1 , … , ϕ n ] ∥ H k ( S δ ) ≲ ∥ f o u t [ ϕ 1 , … , ϕ n , 0 , Γ ~ s [ ϕ 1 , … , ϕ n , 0 ] ] ∥ H k − 2 ( S δ ) ≲ δ 1 2 ∑ i = 1 n ∥ ϕ i ∥ H k ( S δ ) + δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ . \begin{align*}
\left\|\widetilde{\psi}[\phi_1,\dots,\phi_n]\right\|_{H^{k}(\mathcal{S}_\delta)}&\lesssim \left\|f_{out}[\phi_1,\dots,\phi_n,0,\widetilde{\Gamma}_{\mathrm{s}}[\phi_1,\dots,\phi_n,0]]\right\|_{H^{k-2}(\mathcal{S}_\delta)}\\
&\lesssim\delta^{\frac{1}{2}}\sum_{i=1}^n\left\|\phi_i\right\|_{H^{k}(\mathcal{S}_\delta)}+\delta^{\frac 14} | \log \delta|^{\frac{1}{2}}e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}.
\end{align*} ψ [ ϕ 1 , … , ϕ n ] H k ( S δ ) ≲ f o u t [ ϕ 1 , … , ϕ n , 0 , Γ s [ ϕ 1 , … , ϕ n , 0 ]] H k − 2 ( S δ ) ≲ δ 2 1 i = 1 ∑ n ∥ ϕ i ∥ H k ( S δ ) + δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) . Finally, we prove a pointwise estimate for the fixed point ψ ~ [ ϕ 1 , … , ϕ n ] \widetilde{\psi}[\phi_1,\dots,\phi_n] ψ [ ϕ 1 , … , ϕ n ] . Applying (3.12) we have
∣ ψ ~ [ ϕ 1 , … , ϕ n ] ( x ) ∣ = ∣ ( L o u t ) − 1 f o u t [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s [ u ] ] ( x ) ∣ ≲ δ ∑ i = 1 n ∣ ∫ S δ G ~ ( x , ξ ) ∇ η ( δ R ∣ ξ − ( 0 , τ i δ ) ∣ ) ⋅ ∇ ϕ i ( ξ ) d ξ ∣ + δ 2 ∑ i = 1 n ∣ ∫ S δ G ~ ( x , ξ ) Δ η ( δ R ∣ ξ − ( 0 , τ i δ ) ∣ ) ϕ i ( ξ ) d ξ ∣ + ∑ i = 1 n ∣ ∫ S δ G ~ ( x , ξ ) η ( δ R ∣ ξ − ( 0 , τ i δ ) ∣ ) ( γ ′ ( ∑ i = 1 n σ i U i ) − γ ′ ( U i ) ) ϕ i ( ξ ) d ξ ∣ + ∣ ∫ S δ G ~ ( x , ξ ) ( 1 − ∑ i = 1 n η ( δ R ∣ ξ − ( 0 , τ i δ ) ∣ ) ) F ( τ , u , Γ s ) ( ξ ) d ξ ∣ . \begin{align}
&\left|\widetilde{\psi}[\phi_1,\dots,\phi_n](x)\right|=\left|(L_{out})^{-1}f_{out}[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde{\Gamma}_{\mathrm{s}}[u]](x)\right|\notag\\
&\lesssim\delta\sum_{i=1}^n\left|\displaystyle\int_{\mathcal{S}_\delta}\widetilde G(x,\xi)\nabla\eta\left(\frac{\delta}{R}\left|\xi-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\cdot\nabla\phi_i(\xi)\;d\xi\right| \tag*{(4.13)}\\
&\quad\;+\delta^2\sum_{i=1}^n\left|\displaystyle\int_{\mathcal{S}_\delta}\widetilde G(x,\xi)\Delta\eta\left(\frac{\delta}{R}\left|\xi-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\phi_i(\xi)\;d\xi\right| \tag*{(4.14)}\\
&\quad\;+\sum_{i=1}^n\left|\displaystyle\int_{\mathcal{S}_\delta}\widetilde G(x,\xi)\eta\left(\frac{\delta}{R}\left|\xi-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\left(\gamma^\prime\left(\sum_{i=1}^n\sigma_iU_i\right)-\gamma^\prime\left(U_i\right)\right)\phi_i(\xi)\;d\xi\right| \tag*{(4.15)}\\
&\quad\;+\left|\displaystyle\int_{\mathcal{S}_\delta}\widetilde G(x,\xi)\left(1-\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|\xi-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\right)F(\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}})(\xi)\;d\xi\right|. \tag*{(4.16)}
\end{align} ψ [ ϕ 1 , … , ϕ n ] ( x ) = ( L o u t ) − 1 f o u t [ ϕ 1 , … , ϕ n , ψ , Γ s [ u ]] ( x ) ≲ δ i = 1 ∑ n ∫ S δ G ( x , ξ ) ∇ η ( R δ ξ − ( 0 , δ τ i ) ) ⋅ ∇ ϕ i ( ξ ) d ξ + δ 2 i = 1 ∑ n ∫ S δ G ( x , ξ ) Δ η ( R δ ξ − ( 0 , δ τ i ) ) ϕ i ( ξ ) d ξ + i = 1 ∑ n ∫ S δ G ( x , ξ ) η ( R δ ξ − ( 0 , δ τ i ) ) ( γ ′ ( i = 1 ∑ n σ i U i ) − γ ′ ( U i ) ) ϕ i ( ξ ) d ξ + ∫ S δ G ( x , ξ ) ( 1 − i = 1 ∑ n η ( R δ ξ − ( 0 , δ τ i ) ) ) F ( τ , u , Γ s ) ( ξ ) d ξ . (4.13) (4.14) (4.15) (4.16) Consider (4.13) , since the support of ∇ η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) \nabla\eta\left(\frac{\delta}{R}\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\right) ∇ η ( R δ x − ( 0 , δ τ i ) ) is contained in the annulus
A i : = { x ∈ R 2 ∣ R δ ≤ ∣ x − ( 0 , τ i δ ) ∣ ≤ 2 R δ } , A_i:=\left\{x\in {\mathbb{R}}^2\;\left|\;\frac{R}{\delta}\leq\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|\leq \frac{2R}{\delta}\right.\right\}, A i := { x ∈ R 2 δ R ≤ x − ( 0 , δ τ i ) ≤ δ 2 R } , by (2.23) and (3.10) ,
∣ ∫ S δ G ~ ( x , ξ ) ∇ η ( δ R ∣ ξ − ( 0 , τ i δ ) ∣ ) ⋅ ∇ ϕ i ( ξ ) d ξ ∣ ≲ ∫ A i ∣ G ~ ( x , ξ ) ∣ ∣ ∇ ϕ i ( ξ ) ∣ d ξ ≲ ∥ ϕ i ∥ C 1 , α ( B i , 2 ) ∫ A i ∑ m ∈ Z ∣ G 0 ( x 1 − y 1 , x 2 − y 2 + 4 m δ ) ∣ d ξ + ∥ ϕ i ∥ C 1 , α ( B i , 2 ) ∫ A i ∑ m ∈ Z ∣ G 0 ( x 1 − y 1 , x 2 + y 2 − 2 δ + 4 m δ ) ∣ d ξ . \begin{align*}
&\left|\displaystyle\int_{\mathcal{S}_\delta}\widetilde G(x,\xi)\nabla\eta\left(\frac{\delta}{R}\left|\xi-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\cdot\nabla\phi_i(\xi)\;d\xi\right|\lesssim\displaystyle\int_{A_i}\left|\widetilde G(x,\xi)\right|\left|\nabla\phi_i(\xi)\right|\;d\xi\\
&\lesssim\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})}\displaystyle\int_{A_i}\sum_{m\in\mathbb Z}\left|G_0\left(x_1-y_1,x_2-y_2+\frac{4m}{\delta}\right)\right|\;d\xi\\
&\quad\,+\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})}\displaystyle\int_{A_i}\sum_{m\in\mathbb Z}\left|G_0\left(x_1-y_1,x_2+y_2-\frac{2}{\delta}+\frac{4m}{\delta}
\right)\right|\;d\xi.
\end{align*} ∫ S δ G ( x , ξ ) ∇ η ( R δ ξ − ( 0 , δ τ i ) ) ⋅ ∇ ϕ i ( ξ ) d ξ ≲ ∫ A i G ( x , ξ ) ∣ ∇ ϕ i ( ξ ) ∣ d ξ ≲ ∥ ϕ i ∥ C 1 , α ( B i , 2 ) ∫ A i m ∈ Z ∑ G 0 ( x 1 − y 1 , x 2 − y 2 + δ 4 m ) d ξ + ∥ ϕ i ∥ C 1 , α ( B i , 2 ) ∫ A i m ∈ Z ∑ G 0 ( x 1 − y 1 , x 2 + y 2 − δ 2 + δ 4 m ) d ξ . For x ∈ B i , 1 2 x\in B_{i,\frac{1}{2}} x ∈ B i , 2 1 , ξ ∈ A i \xi\in A_i ξ ∈ A i , we have R 2 δ ≤ ∣ x − ξ ∣ ≤ 5 R 2 δ \frac{R}{2\delta}\leq|x-\xi|\leq\frac{5R}{2\delta} 2 δ R ≤ ∣ x − ξ ∣ ≤ 2 δ 5 R , then
( x 2 − ξ 2 + 4 m δ ) 2 − ( x 2 − ξ 2 ) 2 ≥ 16 m 2 δ 2 − 20 ∣ m ∣ R δ 2 , m ∈ Z . \left(x_2-\xi_2+\frac{4m}{\delta}\right)^2-(x_2-\xi_2)^2\geq\frac{16m^2}{\delta^2}-\frac{20|m|R}{\delta^2},\qquad m\in\mathbb{Z}. ( x 2 − ξ 2 + δ 4 m ) 2 − ( x 2 − ξ 2 ) 2 ≥ δ 2 16 m 2 − δ 2 20∣ m ∣ R , m ∈ Z . Since R > 0 R>0 R > 0 small, it follows that
( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 + 4 m δ ) 2 − ( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 ) 2 ≥ ( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 ) 2 + 16 m 2 δ 2 − 20 ∣ m ∣ R δ 2 − ( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 ) 2 = 16 m 2 − 20 ∣ m ∣ R δ 2 ( ( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 ) 2 + 16 m 2 δ 2 − 20 ∣ m ∣ R δ 2 + ( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 ) 2 ) ≥ 16 m 2 5 δ R . \begin{align*}
&\sqrt{(x_1-\xi_1)^2+\left(x_2-\xi_2+\frac{4m}{\delta}\right)^2}-\sqrt{(x_1-\xi_1)^2+\left(x_2-\xi_2\right)^2}\\
&\geq\sqrt{(x_1-\xi_1)^2+\left(x_2-\xi_2\right)^2+\frac{16m^2}{\delta^2}-\frac{20|m|R}{\delta^2}}-\sqrt{(x_1-\xi_1)^2+\left(x_2-\xi_2\right)^2}\\
&=\frac{16m^2-20|m|R}{\delta^2\left(\sqrt{(x_1-\xi_1)^2+\left(x_2-\xi_2\right)^2+\frac{16m^2}{\delta^2}-\frac{20|m|R}{\delta^2}}+\sqrt{(x_1-\xi_1)^2+\left(x_2-\xi_2\right)^2}\right)}\geq\frac{16m^2}{5\delta R}.
\end{align*} ( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 + δ 4 m ) 2 − ( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 ) 2 ≥ ( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 ) 2 + δ 2 16 m 2 − δ 2 20∣ m ∣ R − ( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 ) 2 = δ 2 ( ( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 ) 2 + δ 2 16 m 2 − δ 2 20∣ m ∣ R + ( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 ) 2 ) 16 m 2 − 20∣ m ∣ R ≥ 5 δ R 16 m 2 . Using the above inequality and (2.22) , if x ∈ B i , 1 2 x\in B_{i,\frac{1}{2}} x ∈ B i , 2 1 ,
∫ A i ∑ m ∈ Z ∣ G 0 ( x 1 − y 1 , x 2 − y 2 + 4 m δ ) ∣ d ξ ≲ δ 1 2 ∫ A i ∑ m ∈ Z e − ( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 + 4 m δ ) 2 d ξ ≲ δ 1 2 ∫ A i e − ∣ x − ξ ∣ ∑ m ∈ Z e − 16 m 2 5 δ R d ξ ≲ δ − 3 2 e − R 2 δ . \begin{align*}
\displaystyle\int_{A_i}\sum_{m\in\mathbb Z}\left|G_0\left(x_1-y_1,x_2-y_2+\frac{4m}{\delta}\right)\right|\;d\xi&\lesssim\delta^{\frac{1}{2}}\displaystyle\int_{A_i}\sum_{m\in\mathbb Z}{\mathrm e}^{-\sqrt{(x_1-\xi_1)^2+\left(x_2-\xi_2+\frac{4m}{\delta}\right)^2}}\;d\xi\\
&\lesssim\delta^{\frac{1}{2}}\displaystyle\int_{A_i}{\mathrm e}^{-|x-\xi|}\sum_{m\in\mathbb Z}{\mathrm e}^{-\frac{16m^2}{5\delta R}}\;d\xi\\
&\lesssim\delta^{-\frac{3}{2}}{\mathrm e}^{-\frac{R}{2\delta}}.
\end{align*} ∫ A i m ∈ Z ∑ G 0 ( x 1 − y 1 , x 2 − y 2 + δ 4 m ) d ξ ≲ δ 2 1 ∫ A i m ∈ Z ∑ e − ( x 1 − ξ 1 ) 2 + ( x 2 − ξ 2 + δ 4 m ) 2 d ξ ≲ δ 2 1 ∫ A i e − ∣ x − ξ ∣ m ∈ Z ∑ e − 5 δ R 16 m 2 d ξ ≲ δ − 2 3 e − 2 δ R . Similarly,
∫ A i ∑ m ∈ Z ∣ G 0 ( x 1 − y 1 , x 2 + y 2 − 2 δ + 4 m δ ) ∣ d ξ ≲ δ − 3 2 e − R 2 δ . \displaystyle\int_{A_i}\sum_{m\in\mathbb Z}\left|G_0\left(x_1-y_1,x_2+y_2-\frac{2}{\delta}+\frac{4m}{\delta}
\right)\right|\;d\xi\lesssim\delta^{-\frac{3}{2}}{\mathrm e}^{-\frac{R}{2\delta}}. ∫ A i m ∈ Z ∑ G 0 ( x 1 − y 1 , x 2 + y 2 − δ 2 + δ 4 m ) d ξ ≲ δ − 2 3 e − 2 δ R . For x ∈ S δ \ B i , 1 2 x\in \mathcal{S}_\delta\backslash B_{i,\frac{1}{2}} x ∈ S δ \ B i , 2 1 , by absolute integrability of G ~ \widetilde G G , we finally obtain
∣ ∫ S δ G ~ ( x , ξ ) ∇ η ( δ R ∣ ξ − ( 0 , τ i δ ) ∣ ) ⋅ ∇ ϕ i ( ξ ) d ξ ∣ ≲ { δ − 3 / 2 e − R 2 δ ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , x ∈ B i , 1 2 , ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , x ∈ S δ \ B i , 1 2 . \left|\displaystyle\int_{\mathcal{S}_\delta}\widetilde G(x,\xi)\nabla\eta\left(\frac{\delta}{R}\left|\xi-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\cdot\nabla\phi_i(\xi)\;d\xi\right|\lesssim
\begin{cases}
\delta^{-3/2}{\mathrm e}^{-\frac{R}{2\delta}}\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})},&x\in B_{i,\frac{1}{2}},\\
\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})},&x\in \mathcal{S}_\delta\backslash B_{i,\frac{1}{2}}.
\end{cases} ∫ S δ G ( x , ξ ) ∇ η ( R δ ξ − ( 0 , δ τ i ) ) ⋅ ∇ ϕ i ( ξ ) d ξ ≲ { δ − 3/2 e − 2 δ R ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , x ∈ B i , 2 1 , x ∈ S δ \ B i , 2 1 . Using the same method as in (4.13) , for (4.14)
∣ ∫ S δ G ~ ( x , ξ ) Δ η ( δ R ∣ ξ − ( 0 , τ i δ ) ∣ ) ϕ i ( ξ ) d ξ ∣ ≲ { δ − 3 / 2 e − R 2 δ ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , x ∈ B i , 1 2 , ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , x ∈ S δ \ B i , 1 2 . \begin{align*}
\left|\displaystyle\int_{\mathcal{S}_\delta}\widetilde G(x,\xi)\Delta\eta\left(\frac{\delta}{R}\left|\xi-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\phi_i(\xi)\;d\xi\right|\lesssim
\begin{cases}
\delta^{-3/2}{\mathrm e}^{-\frac{R}{2\delta}}\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})},&x\in B_{i,\frac{1}{2}},\\
\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})},&x\in \mathcal{S}_\delta\backslash B_{i,\frac{1}{2}}.
\end{cases}
\end{align*} ∫ S δ G ( x , ξ ) Δ η ( R δ ξ − ( 0 , δ τ i ) ) ϕ i ( ξ ) d ξ ≲ { δ − 3/2 e − 2 δ R ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , x ∈ B i , 2 1 , x ∈ S δ \ B i , 2 1 . For (4.15) , an estimate similar to (4.11) holds
∣ ∫ S δ G ~ ( x , ξ ) η ( δ R ∣ ξ − ( 0 , τ i δ ) ∣ ) ( γ ′ ( ∑ j = 1 n σ j U j ) − γ ′ ( U i ) ) ϕ i ( ξ ) d ξ ∣ ≲ ∫ B i , 2 ∣ G ~ ( x , ξ ) ∣ ∣ ∑ j ≠ i U j ( ξ ) ∣ ∣ ϕ i ( ξ ) ∣ d ξ ≲ e − d − 2 R δ ∥ ϕ i ∥ C 1 , α ( B i , 2 ) . \begin{align*}
&\left|\displaystyle\int_{\mathcal{S}_\delta}\widetilde G(x,\xi)\eta\left(\frac{\delta}{R}\left|\xi-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\left(\gamma^\prime\left(\sum_{j=1}^n\sigma_jU_j\right)-\gamma^\prime\left(U_i\right)\right)\phi_i(\xi)\;d\xi\right|\\
&\lesssim\displaystyle\int_{B_{i,2}}|\widetilde G(x,\xi)|\left|\sum_{j\neq i}U_j(\xi)\right||\phi_i(\xi)|\;d\xi\lesssim{\mathrm e}^{-\frac{d-2R}{\delta}}\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})}.
\end{align*} ∫ S δ G ( x , ξ ) η ( R δ ξ − ( 0 , δ τ i ) ) ( γ ′ ( j = 1 ∑ n σ j U j ) − γ ′ ( U i ) ) ϕ i ( ξ ) d ξ ≲ ∫ B i , 2 ∣ G ( x , ξ ) ∣ j = i ∑ U j ( ξ ) ∣ ϕ i ( ξ ) ∣ d ξ ≲ e − δ d − 2 R ∥ ϕ i ∥ C 1 , α ( B i , 2 ) . For (4.16) , since 1 − ∑ i = 1 n η ( δ R ∣ ξ − ( 0 , τ i δ ) ∣ ) = 0 1-\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|\xi-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)=0 1 − ∑ i = 1 n η ( R δ ξ − ( 0 , δ τ i ) ) = 0 in all B i , 1 , i = 1 , … , n , B_{i,1},\; i=1,\dots,n, B i , 1 , i = 1 , … , n , by (4.8) , (4.9) , (4.10) , Proposition (?) and (5.5) ,
∣ ∫ S δ G ~ ( x , ξ ) ( 1 − ∑ i = 1 n η ( δ R ∣ ξ − ( 0 , τ i δ ) ∣ ) ) F ( τ , u , Γ s ) ( ξ ) d ξ ∣ ≲ ∫ S δ \ ∪ i = 1 n B i , 1 ∣ G ~ ( x , ξ ) ∣ ∣ F ( τ , u , Γ s ) ( ξ ) ∣ d ξ ≲ { δ − 3 / 2 e − R 2 δ ( e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ + δ − 1 2 − k ∥ u ∥ H k ( S δ ) 2 ) , x ∈ B i , 1 2 , e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ + δ − 1 2 − k ∥ u ∥ H k ( S δ ) 2 , x ∈ B i , 1 2 . \begin{align*}
&\left|\displaystyle\int_{\mathcal{S}_\delta}\widetilde G(x,\xi)\left(1-\sum_{i=1}^n\eta\left(\frac{\delta}{R}\left|\xi-\left(0,\frac{\tau_i}{\delta}\right)\right|\right)\right)F(\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}})(\xi)\;d\xi\right|\\
&\lesssim\displaystyle\int_{\mathcal{S}_\delta\backslash\cup_{i=1}^nB_{i,1}}|\widetilde G(x,\xi)||F(\boldsymbol{\tau}, u, \Gamma_{\mathrm{s}})(\xi)|\;d\xi\\
&\lesssim\begin{cases}
\delta^{-3/2}{\mathrm e}^{-\frac{R}{2\delta}}\left(e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{-\frac{1}{2}-k}\|u\|_{H^k (\mathcal{S}_\delta)}^2\right),&x\in B_{i,\frac{1}{2}},\\
e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{-\frac{1}{2}-k}\|u\|_{H^k (\mathcal{S}_\delta)}^2,&x\in B_{i,\frac{1}{2}}.
\end{cases}
\end{align*} ∫ S δ G ( x , ξ ) ( 1 − i = 1 ∑ n η ( R δ ξ − ( 0 , δ τ i ) ) ) F ( τ , u , Γ s ) ( ξ ) d ξ ≲ ∫ S δ \ ∪ i = 1 n B i , 1 ∣ G ( x , ξ ) ∣∣ F ( τ , u , Γ s ) ( ξ ) ∣ d ξ ≲ ⎩ ⎨ ⎧ δ − 3/2 e − 2 δ R ( e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) + δ − 2 1 − k ∥ u ∥ H k ( S δ ) 2 ) , e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) + δ − 2 1 − k ∥ u ∥ H k ( S δ ) 2 , x ∈ B i , 2 1 , x ∈ B i , 2 1 . From the above estimate, we have
∣ ψ ~ [ ϕ 1 , … , ϕ n ] ( x ) ∣ ≤ C ~ { δ − 3 / 2 e − R 2 δ ( e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ + δ − 1 2 − k ∑ i = 1 n ∥ ϕ i ∥ H k ( S δ ) 2 ) + δ − 3 / 2 e − R 2 δ δ ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) + e − d − 2 R δ ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , x ∈ ∪ i = 1 n B i , 1 2 , e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ + δ − 1 2 − k ∑ i = 1 n ∥ ϕ i ∥ H k ( S δ ) 2 + δ ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) + e − d − 2 R δ ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , x ∈ S δ \ ∪ i = 1 n B i , 1 2 . \begin{align*}
&|\widetilde{\psi}[\phi_1,\dots,\phi_n](x)|\\
&\leq \widetilde C \begin{cases}
\delta^{-3/2}{\mathrm e}^{-\frac{R}{2\delta}}\left(e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{-\frac{1}{2}-k}\sum_{i=1}^n\|\phi_i\|_{H^k (\mathcal{S}_\delta)}^2\right)\\
+\delta^{-3/2}{\mathrm e}^{-\frac{R}{2\delta}}\delta\sum_{i=1}^n\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})}+{\mathrm e}^{-\frac{d-2R}{\delta}}\sum_{i=1}^n\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})},&x\in\cup_{i=1}^nB_{i,\frac{1}{2}},\\[4mm]
e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{-\frac{1}{2}-k}\sum_{i=1}^n\|\phi_i\|_{H^k (\mathcal{S}_\delta)}^2\\
+\delta\sum_{i=1}^n\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})}+{\mathrm e}^{-\frac{d-2R}{\delta}}\sum_{i=1}^n\left\|\phi_i\right\|_{C^{1,\alpha}(B_{i,2})},&x\in \mathcal{S}_\delta\backslash \cup_{i=1}^nB_{i,\frac{1}{2}}.
\end{cases}
\end{align*} ∣ ψ [ ϕ 1 , … , ϕ n ] ( x ) ∣ ≤ C ⎩ ⎨ ⎧ δ − 3/2 e − 2 δ R ( e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) + δ − 2 1 − k ∑ i = 1 n ∥ ϕ i ∥ H k ( S δ ) 2 ) + δ − 3/2 e − 2 δ R δ ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) + e − δ d − 2 R ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) + δ − 2 1 − k ∑ i = 1 n ∥ ϕ i ∥ H k ( S δ ) 2 + δ ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) + e − δ d − 2 R ∑ i = 1 n ∥ ϕ i ∥ C 1 , α ( B i , 2 ) , x ∈ ∪ i = 1 n B i , 2 1 , x ∈ S δ \ ∪ i = 1 n B i , 2 1 . ◻
Solving the inner problem
For any given ϕ i \phi_i ϕ i , i = 1 , … , n i=1,\dots,n i = 1 , … , n with ∥ ϕ i ∥ H k ( S δ ) ≤ δ k \|\phi_i\|_{H^{k}(\mathcal{S}_\delta)}\leq\delta^k ∥ ϕ i ∥ H k ( S δ ) ≤ δ k , the outer problem admits a solution ψ ~ [ ϕ 1 , … , ϕ n ] \widetilde{\psi}[\phi_1,\dots,\phi_n] ψ [ ϕ 1 , … , ϕ n ] . After substituting it into the inner problem, we apply the contraction mapping principle to show that, for sufficiently small δ \delta δ and τ ∈ P ε \boldsymbol{\tau}\in P_{\varepsilon} τ ∈ P ε , the inner problem admits a solution.
The inner problem (3.1) is
L i n i [ ϕ i ] = f i n [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] , i = 1 , … , n , \begin{equation}
\begin{aligned}
L_{in}^i[\phi_i]=f_{in}[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}],\quad i=1,\dots,n,
\end{aligned} \tag*{(4.18)}
\end{equation} L in i [ ϕ i ] = f in [ ϕ 1 , … , ϕ n , ψ , Γ s ] , i = 1 , … , n , (4.18)
where
L i n i : = Δ − γ ′ ( U i ) , L_{in}^i:=\Delta-\gamma^\prime\left(U_i\right), L in i := Δ − γ ′ ( U i ) ,
f i n [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] : = F ( τ , ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ) + ( γ ′ ( ∑ i = 1 n σ i U i ) − 1 ) ψ ~ . f_{in}[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]:=F(\boldsymbol{\tau},\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}})+\left(\gamma^\prime\left(\sum_{i=1}^n\sigma_iU_i\right)-1\right)\widetilde\psi. f in [ ϕ 1 , … , ϕ n , ψ , Γ s ] := F ( τ , ϕ 1 , … , ϕ n , ψ , Γ s ) + ( γ ′ ( i = 1 ∑ n σ i U i ) − 1 ) ψ .
Proposition 4.3. For each i = 1 , … , n i=1,\dots,n i = 1 , … , n , assume that ∥ ϕ j ∥ H k ( S δ ) ≤ δ k + 1 \|\phi^j\|_{H^{k}(\mathcal{S}_\delta)}\leq\delta^{k+1} ∥ ϕ j ∥ H k ( S δ ) ≤ δ k + 1 for all j ≠ i j\neq i j = i . Then there exist constants C ~ > 0 \widetilde C >0 C > 0 and δ 0 > 0 \delta_0 > 0 δ 0 > 0 such that for τ ∈ P ε \boldsymbol{\tau} \in P_{\varepsilon} τ ∈ P ε and δ ∈ ( 0 , δ 0 ) \delta\in(0,\delta_0) δ ∈ ( 0 , δ 0 ) , the operator T i n i [ ϕ i ] : = ( L i n i ) − 1 [ f i n [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] + c i [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] ∂ x 2 U i ] \mathrm{T}^i_{in}[\phi_i]:=(L_{in}^i)^{-1}\left[f_{in}[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]+c_i[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}] {\partial} _{x_2} U_i\right] T in i [ ϕ i ] := ( L in i ) − 1 [ f in [ ϕ 1 , … , ϕ n , ψ , Γ s ] + c i [ ϕ 1 , … , ϕ n , ψ , Γ s ] ∂ x 2 U i ] has a fixed point ϕ ~ i \widetilde\phi_i ϕ i in the set B i n i : = { ϕ i ∈ H e k ( S δ ) ∣ ∥ ϕ i ∥ H k ( S δ ) ≤ δ k + 1 } . B_{in}^i:=\left\{\phi_i\in H_{\mathrm{e}}^{k}(\mathcal{S}_\delta)\;\left|\; \|\phi_i\|_{H^{k}(\mathcal{S}_\delta)}\leq\delta^{k+1}\right.\right\}. B in i := { ϕ i ∈ H e k ( S δ ) ∥ ϕ i ∥ H k ( S δ ) ≤ δ k + 1 } . Here c i [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] c_i[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}] c i [ ϕ 1 , … , ϕ n , ψ , Γ s ] is the orthogonal projection coefficient such that the following orthogonality condition holds: ( f i n [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] + c i [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] ∂ x 2 U i , ∂ x 2 U i ) L 2 ( S δ ) = 0. \left(f_{in}[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]+c_i[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}] {\partial} _{x_2} U_i, {\partial} _{x_2} U_i\right)_{L^2(\mathcal{S}_\delta)}=0. ( f in [ ϕ 1 , … , ϕ n , ψ , Γ s ] + c i [ ϕ 1 , … , ϕ n , ψ , Γ s ] ∂ x 2 U i , ∂ x 2 U i ) L 2 ( S δ ) = 0. And ϕ ~ i \widetilde\phi_i ϕ i satisfies ∥ ϕ i ~ ∥ H k ( S δ ) ≤ C ~ ( δ 1 2 ∑ j ≠ i ∥ ϕ j ∥ H k ( S δ ) + δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ ) , \left\|\widetilde{\phi_i}\right\|_{H^{k}(\mathcal{S}_\delta)}\leq\widetilde C \left(\delta^{\frac{1}{2}}\sum_{j\neq i}\left\|\phi_j\right\|_{H^{k}(\mathcal{S}_\delta)}+\delta^{\frac 14} | \log \delta|^{\frac{1}{2}}e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}\right), ϕ i H k ( S δ ) ≤ C ( δ 2 1 ∑ j = i ∥ ϕ j ∥ H k ( S δ ) + δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) ) , ∥ ϕ ~ i ∥ C 1 , α ( B i , 2 ) ≤ C ~ ( δ 1 2 ∑ j ≠ i ∥ ϕ j ∥ H k ( S δ ) + δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ ) for k ≥ 2. \left\|\widetilde\phi_i\right\|_{C^{1,\alpha}(B_{i,2})}\leq\widetilde C \left(\delta^{\frac{1}{2}}\sum_{j\neq i}\left\|\phi_j\right\|_{H^{k}(\mathcal{S}_\delta)}+\delta^{\frac 14} | \log \delta|^{\frac{1}{2}}e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}\right)\quad\text{for } k\geq 2. ϕ i C 1 , α ( B i , 2 ) ≤ C ( δ 2 1 ∑ j = i ∥ ϕ j ∥ H k ( S δ ) + δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) ) for k ≥ 2.
Proof. Firstly we prove that T i n i [ ϕ i ] ∈ B i n i \mathrm{T}^i_{in}[\phi_i]\in B^i_{in} T in i [ ϕ i ] ∈ B in i for any ϕ i ∈ B i n i \phi_i\in B^i_{in} ϕ i ∈ B in i . By the linear theory Lemma (?) for the inner problem and Lemma 2.1 ,
∥ T i n i [ ϕ i ] ∥ H k ( S δ ) ≲ ∥ f i n [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] + c i [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] ∂ x 2 U i ∥ H k − 2 ( S δ ) ≲ ∥ F ( τ , ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ) ∥ H k − 2 ( S δ ) + ∥ ψ ~ ∥ H k ( S δ ) + ∥ c i [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] ∥ . \begin{align*}
\left\|\mathrm{T}^i_{in}[\phi_i]\right\|_{H^{k}(\mathcal{S}_\delta)}&\lesssim\left\|f_{in}[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]+c_i[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}] {\partial} _{x_2} U_i\right\|_{H^{k-2}(\mathcal{S}_\delta)}\\
&\lesssim\left\|F(\boldsymbol{\tau},\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}})\right\|_{H^{k-2}(\mathcal{S}_\delta)}+\left\|\widetilde\psi\right\|_{H^{k}(\mathcal{S}_\delta)}+\left\|c_i[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]\right\|.
\end{align*} T in i [ ϕ i ] H k ( S δ ) ≲ f in [ ϕ 1 , … , ϕ n , ψ , Γ s ] + c i [ ϕ 1 , … , ϕ n , ψ , Γ s ] ∂ x 2 U i H k − 2 ( S δ ) ≲ F ( τ , ϕ 1 , … , ϕ n , ψ , Γ s ) H k − 2 ( S δ ) + ψ H k ( S δ ) + c i [ ϕ 1 , … , ϕ n , ψ , Γ s ] . Since c i [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] c_i[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}] c i [ ϕ 1 , … , ϕ n , ψ , Γ s ] is the orthogonal projection coefficient, we obtain
∣ c i [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] ∣ = ∣ ( f i n [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] , ∂ x 2 U i ) L 2 ( S δ ) ∥ ∂ x 2 U i ∥ L 2 ( S δ ) 2 ∣ ≲ ∥ f i n [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] ∥ H k − 2 ( S δ ) ∥ ∂ x 2 U i ∥ L 2 ( S δ ) ≲ ∥ f i n [ ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ] ∥ H k − 2 ( S δ ) . \begin{align*}
&|c_i[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]|=\left|\frac{\left(f_{\mathrm{in}}[\phi_1,\dots,\phi_n,\widetilde\psi,\widetilde\Gamma_{\mathrm{s}}],\partial_{x_2}U_i\right)_{L^2(\mathcal{S}_\delta)}}{\|\partial_{x_2}U_i\|_{L^2(\mathcal{S}_\delta)}^2}\right|\\
&\lesssim\left\|f_{in}[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]\right\|_{H^{k-2}(\mathcal{S}_\delta)}\left\| {\partial} _{x_2} U_i\right\|_{L^2(\mathcal{S}_\delta)}\lesssim\left\|f_{in}[\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]\right\|_{H^{k-2}(\mathcal{S}_\delta)}.
\end{align*} ∣ c i [ ϕ 1 , … , ϕ n , ψ , Γ s ] ∣ = ∥ ∂ x 2 U i ∥ L 2 ( S δ ) 2 ( f in [ ϕ 1 , … , ϕ n , ψ , Γ s ] , ∂ x 2 U i ) L 2 ( S δ ) ≲ f in [ ϕ 1 , … , ϕ n , ψ , Γ s ] H k − 2 ( S δ ) ∥ ∂ x 2 U i ∥ L 2 ( S δ ) ≲ f in [ ϕ 1 , … , ϕ n , ψ , Γ s ] H k − 2 ( S δ ) . Thus
∥ T i n i [ ϕ i ] ∥ H k ( S δ ) ≲ ∥ F ( τ , ϕ 1 , … , ϕ n , ψ ~ , Γ ~ s ) ∥ H k − 2 ( S δ ) + ∥ ψ ~ ∥ H k ( S δ ) . \left\|\mathrm{T}^i_{in}[\phi_i]\right\|_{H^{k}(\mathcal{S}_\delta)}\lesssim\left\|F(\boldsymbol{\tau},\phi_1,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}})\right\|_{H^{k-2}(\mathcal{S}_\delta)}+\left\|\widetilde\psi\right\|_{H^{k}(\mathcal{S}_\delta)}. T in i [ ϕ i ] H k ( S δ ) ≲ F ( τ , ϕ 1 , … , ϕ n , ψ , Γ s ) H k − 2 ( S δ ) + ψ H k ( S δ ) . Applying (4.8) , (4.9) , (4.10) , Proposition (?) and Proposition (?) ,
∥ T i n i [ ϕ i ] ∥ H k ( S δ ) ≲ δ 1 2 ∑ j = 1 n ∥ ϕ j ∥ H k ( S δ ) + δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ . \left\|\mathrm{T}^i_{in}[\phi_i]\right\|_{H^{k}(\mathcal{S}_\delta)}\lesssim\delta^{\frac{1}{2}}\sum_{j=1}^n\left\|\phi_j\right\|_{H^{k}(\mathcal{S}_\delta)}+\delta^{\frac 14} | \log \delta|^{\frac{1}{2}}e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}. T in i [ ϕ i ] H k ( S δ ) ≲ δ 2 1 j = 1 ∑ n ∥ ϕ j ∥ H k ( S δ ) + δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) . From the above estimate, choosing δ \delta δ small, then T i n i [ ϕ i ] ∈ B i n i \mathrm{T}^i_{in}[\phi_i]\in B^i_{in} T in i [ ϕ i ] ∈ B in i .
Next we prove T i n i \mathrm{T}^i_{in} T in i is a contraction on B i n i B^i_{in} B in i . By (4.9) , (4.10) , (4.5) and (4.17) ,
∥ T i n i [ ϕ i , 1 ] − T i n i [ ϕ i , 2 ] ∥ H k ( S δ ) ≲ ∥ f i n [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ ~ , Γ ~ s ] − f i n [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ ~ , Γ ~ s ] ∥ H k − 2 ( S δ ) + ∣ c i [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ ~ , Γ ~ s ] − c i [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ ~ , Γ ~ s ] ∣ ∥ ∂ x 2 U i ∥ H k − 2 ( S δ ) ≲ ∥ D u F ( τ , ζ , Γ ~ s ) ( ϕ i , 1 − ϕ i , 2 ) ∥ H k − 2 ( S δ ) + ∥ D Γ s F ( τ , u , ζ ) D u Γ ~ s [ ζ ] ( ϕ i , 1 − ϕ i , 2 ) ∥ H k − 2 ( S δ ) + ∥ γ ′ ( ∑ j = 1 n σ j U j ) − 1 ∥ H k ( S δ ) ∥ D ϕ i ψ ~ ( ϕ i , 1 − ϕ i , 2 ) ∥ H k ( S δ ) + ∣ c i [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ ~ , Γ ~ s ] − c i [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ ~ , Γ ~ s ] ∣ ≲ ( δ − 1 ∥ Γ ~ s ∥ H k ( R ) + ∥ u ∥ H k ( S δ ) + ∑ j = 1 n ∥ ( U j ) b c ∥ H k ( S δ ) ) ∥ ϕ i , 1 − ϕ i , 2 ∥ H k ( S δ ) + δ ∥ ϕ i , 1 − ϕ i , 2 ∥ H k ( S δ ) + δ − 1 ∥ Γ ~ s ∥ H k ( R ) ( δ 1 2 − k ∥ u ∥ H k ( S δ ) + δ 3 4 − k e − 1 − ∣ τ ∣ δ ) ∥ ϕ i , 1 − ϕ i , 2 ∥ H k ( S δ ) + ∣ c i [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ ~ , Γ ~ s ] − c i [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ ~ , Γ ~ s ] ∣ . \begin{align*}
&\left\|\mathrm{T}^i_{in}[\phi_{i,1}]-\mathrm{T}^i_{in}[\phi_{i,2}]\right\|_{H^{k}(\mathcal{S}_\delta)}\\
&\lesssim\left\|f_{in}[\phi_1,\dots,\phi_{i,1},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]-f_{in}[\phi_1,\dots,\phi_{i,2},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]\right\|_{H^{k-2}(\mathcal{S}_\delta)}\\
&\quad\;+\left|c_i[\phi_1,\dots,\phi_{i,1},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]-c_i[\phi_1,\dots,\phi_{i,2},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]\right|\left\| {\partial} _{x_2} U_i\right\|_{H^{k-2}(\mathcal{S}_\delta)}\\
&\lesssim\left\|D_u F(\boldsymbol{\tau}, \zeta, \widetilde{\Gamma}_{\mathrm{s}})(\phi_{i,1}-\phi_{i,2})\right\|_{H^{k-2}(\mathcal{S}_\delta)}+\left\|D_{\Gamma_{\mathrm{s}}} F (\boldsymbol{\tau}, u, \zeta)D_u \widetilde{\Gamma}_{\mathrm{s}}[\zeta](\phi_{i,1}-\phi_{i,2})\right\|_{H^{k-2}(\mathcal{S}_\delta)}\\
&\quad\;+\left\|\gamma^\prime\left(\sum_{j=1}^n\sigma_jU_j\right)-1\right\|_{H^{k}(\mathcal{S}_\delta)}\left\|D_{\phi_i}\widetilde\psi(\phi_{i,1}-\phi_{i,2})\right\|_{H^{k}(\mathcal{S}_\delta)}\\
&\quad\;+\left|c_i[\phi_1,\dots,\phi_{i,1},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]-c_i[\phi_1,\dots,\phi_{i,2},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]\right|\\
&\lesssim\left(\delta^{-1} \left\|\widetilde{\Gamma}_{\mathrm{s}}\right\|_{H^{k}( {\mathbb{R}})} + \|u\|_{H^{k} (\mathcal{S}_\delta)} + \sum_{j=1}^n\|(U_j)_\mathrm{bc}\|_{H^{k} (\mathcal{S}_\delta)}\right)\|\phi_{i,1}-\phi_{i,2}\|_{H^{k}(\mathcal{S}_\delta)}+\delta\|\phi_{i,1}-\phi_{i,2}\|_{H^{k}(\mathcal{S}_\delta)}\\
&\quad\;+\delta^{-1} \left\|\widetilde{\Gamma}_{\mathrm{s}}\right\|_{H^{k}( {\mathbb{R}})}\left(\delta^{\frac 12-k}\|u\|_{H^k (\mathcal{S}_\delta)} + \delta^{\frac 34-k} e^{-\frac {1-|\boldsymbol{\tau}|}\delta} \right)\|\phi_{i,1}-\phi_{i,2}\|_{H^{k}(\mathcal{S}_\delta)}\\
&\quad\;+\left|c_i[\phi_1,\dots,\phi_{i,1},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]-c_i[\phi_1,\dots,\phi_{i,2},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]\right|.
\end{align*} T in i [ ϕ i , 1 ] − T in i [ ϕ i , 2 ] H k ( S δ ) ≲ f in [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ , Γ s ] − f in [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ , Γ s ] H k − 2 ( S δ ) + c i [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ , Γ s ] − c i [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ , Γ s ] ∥ ∂ x 2 U i ∥ H k − 2 ( S δ ) ≲ D u F ( τ , ζ , Γ s ) ( ϕ i , 1 − ϕ i , 2 ) H k − 2 ( S δ ) + D Γ s F ( τ , u , ζ ) D u Γ s [ ζ ] ( ϕ i , 1 − ϕ i , 2 ) H k − 2 ( S δ ) + γ ′ ( j = 1 ∑ n σ j U j ) − 1 H k ( S δ ) D ϕ i ψ ( ϕ i , 1 − ϕ i , 2 ) H k ( S δ ) + c i [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ , Γ s ] − c i [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ , Γ s ] ≲ ( δ − 1 Γ s H k ( R ) + ∥ u ∥ H k ( S δ ) + j = 1 ∑ n ∥ ( U j ) bc ∥ H k ( S δ ) ) ∥ ϕ i , 1 − ϕ i , 2 ∥ H k ( S δ ) + δ ∥ ϕ i , 1 − ϕ i , 2 ∥ H k ( S δ ) + δ − 1 Γ s H k ( R ) ( δ 2 1 − k ∥ u ∥ H k ( S δ ) + δ 4 3 − k e − δ 1 − ∣ τ ∣ ) ∥ ϕ i , 1 − ϕ i , 2 ∥ H k ( S δ ) + c i [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ , Γ s ] − c i [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ , Γ s ] . For ∣ c i [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ ~ , Γ ~ s ] − c i [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ ~ , Γ ~ s ] ∣ \left|c_i[\phi_1,\dots,\phi_{i,1},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]-c_i[\phi_1,\dots,\phi_{i,2},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]\right| c i [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ , Γ s ] − c i [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ , Γ s ] , we have
∣ c i [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ ~ , Γ ~ s ] − c i [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ ~ , Γ ~ s ] ∣ ≲ ∥ f i n [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ ~ , Γ ~ s ] − f i n [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ ~ , Γ ~ s ] ∥ H k − 2 ( S δ ) . \begin{align*}
&\left|c_i[\phi_1,\dots,\phi_{i,1},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]-c_i[\phi_1,\dots,\phi_{i,2},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]\right|\\
&\lesssim\left\|f_{in}[\phi_1,\dots,\phi_{i,1},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]-f_{in}[\phi_1,\dots,\phi_{i,2},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]\right\|_{H^{k-2}(\mathcal{S}_\delta)}.
\end{align*} c i [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ , Γ s ] − c i [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ , Γ s ] ≲ f in [ ϕ 1 , … , ϕ i , 1 , … , ϕ n , ψ , Γ s ] − f in [ ϕ 1 , … , ϕ i , 2 , … , ϕ n , ψ , Γ s ] H k − 2 ( S δ ) . Thus
∥ T i n i [ ϕ i , 1 ] − T i n i [ ϕ i , 2 ] ∥ H k ( S δ ) ≲ ( δ − 1 ∥ Γ ~ s ∥ H k ( R ) + ∥ u ∥ H k ( S δ ) + ∑ j = 1 n ∥ ( U j ) b c ∥ H k ( S δ ) ) ∥ ϕ i , 1 − ϕ i , 2 ∥ H k ( S δ ) + δ ∥ ϕ i , 1 − ϕ i , 2 ∥ H k ( S δ ) + δ − 1 ∥ Γ ~ s ∥ H k ( R ) ( δ 1 2 − k ∥ u ∥ H k ( S δ ) + δ 3 4 − k e − 1 − ∣ τ ∣ δ ) ∥ ϕ i , 1 − ϕ i , 2 ∥ H k ( S δ ) . \begin{align*}
&\|\mathrm{T}^i_{in}[\phi_{i,1}]-\mathrm{T}^i_{in}[\phi_{i,2}]\|_{H^{k}(\mathcal{S}_\delta)}\\
&\lesssim\left(\delta^{-1} \left\|\widetilde{\Gamma}_{\mathrm{s}}\right\|_{H^{k}( {\mathbb{R}})} + \|u\|_{H^{k} (\mathcal{S}_\delta)} + \sum_{j=1}^n\|(U_j)_\mathrm{bc}\|_{H^{k} (\mathcal{S}_\delta)}\right)\|\phi_{i,1}-\phi_{i,2}\|_{H^{k}(\mathcal{S}_\delta)}+\delta\|\phi_{i,1}-\phi_{i,2}\|_{H^{k}(\mathcal{S}_\delta)}\\
&\quad\;+\delta^{-1} \left\|\widetilde{\Gamma}_{\mathrm{s}}\right\|_{H^{k}( {\mathbb{R}})}\left(\delta^{\frac 12-k}\|u\|_{H^k (\mathcal{S}_\delta)} + \delta^{\frac 34-k} e^{-\frac {1-|\boldsymbol{\tau}|}\delta} \right)\|\phi_{i,1}-\phi_{i,2}\|_{H^{k}(\mathcal{S}_\delta)}.
\end{align*} ∥ T in i [ ϕ i , 1 ] − T in i [ ϕ i , 2 ] ∥ H k ( S δ ) ≲ ( δ − 1 Γ s H k ( R ) + ∥ u ∥ H k ( S δ ) + j = 1 ∑ n ∥ ( U j ) bc ∥ H k ( S δ ) ) ∥ ϕ i , 1 − ϕ i , 2 ∥ H k ( S δ ) + δ ∥ ϕ i , 1 − ϕ i , 2 ∥ H k ( S δ ) + δ − 1 Γ s H k ( R ) ( δ 2 1 − k ∥ u ∥ H k ( S δ ) + δ 4 3 − k e − δ 1 − ∣ τ ∣ ) ∥ ϕ i , 1 − ϕ i , 2 ∥ H k ( S δ ) . Finally, from Corollary (?) and Proposition (?) , choosing δ \delta δ small, then T i n i \mathrm{T}^i_{in} T in i is a contraction mapping on B i n i B_{in}^i B in i . Then by the contraction mapping principle, T i n i \mathrm{T}^i_{in} T in i possesses a fixed point ϕ i ~ \widetilde{\phi_i} ϕ i in B i n i B^i_{in} B in i . And we have
∥ ϕ i ~ ∥ H k ( S δ ) ≲ ∥ f i n [ ϕ 1 , … , ϕ i − 1 , 0 , ϕ i + 1 , … , ϕ n , ψ ~ , Γ ~ s ] ∥ H k − 2 ( S δ ) ≲ δ 1 2 ∑ j ≠ i ∥ ϕ j ∥ H k ( S δ ) + δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ . \begin{align*}
\left\|\widetilde{\phi_i}\right\|_{H^{k}(\mathcal{S}_\delta)}&\lesssim \left\|f_{in}[\phi_1,\dots,\phi_{i-1},0,\phi_{i+1},\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]\right\|_{H^{k-2}(\mathcal{S}_\delta)}\\
&\lesssim\delta^{\frac{1}{2}}\sum_{j\neq i}\left\|\phi_j\right\|_{H^{k}(\mathcal{S}_\delta)}+\delta^{\frac 14} | \log \delta|^{\frac{1}{2}}e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}.
\end{align*} ϕ i H k ( S δ ) ≲ f in [ ϕ 1 , … , ϕ i − 1 , 0 , ϕ i + 1 , … , ϕ n , ψ , Γ s ] H k − 2 ( S δ ) ≲ δ 2 1 j = i ∑ ∥ ϕ j ∥ H k ( S δ ) + δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) . For the C 1 , α C^{1,\alpha} C 1 , α estimate, by (3.10) , (5.5) and the above estimates,
∥ ϕ ~ i ∥ C 1 , α ( B i , 2 ) ≲ ∥ f i n [ ϕ 1 , … , ϕ ~ i , … , ϕ n , ψ ~ , Γ ~ s ] ∥ H 3 2 ( S δ ) ≲ ∥ f i n [ ϕ 1 , … , ϕ ~ i , … , ϕ n , ψ ~ , Γ ~ s ] ∥ H k − 1 2 ( S δ ) ≲ δ 1 2 ∑ j ≠ i ∥ ϕ j ∥ H k ( S δ ) + δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ , k ≥ 2. \begin{align*}
\left\|\widetilde\phi_i\right\|_{C^{1,\alpha}(B_{i,2})}&\lesssim\left\|f_{in}[\phi_1,\dots,\widetilde\phi_i,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]\right\|_{H^{\frac{3}{2}}(\mathcal{S}_\delta)}\\
&\lesssim\left\|f_{in}[\phi_1,\dots,\widetilde\phi_i,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]\right\|_{H^{k-\frac{1}{2}}(\mathcal{S}_\delta)}\\
&\lesssim\delta^{\frac{1}{2}}\sum_{j\neq i}\left\|\phi_j\right\|_{H^{k}(\mathcal{S}_\delta)}+\delta^{\frac 14} | \log \delta|^{\frac{1}{2}}e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta},\quad k\geq 2.
\end{align*} ϕ i C 1 , α ( B i , 2 ) ≲ f in [ ϕ 1 , … , ϕ i , … , ϕ n , ψ , Γ s ] H 2 3 ( S δ ) ≲ f in [ ϕ 1 , … , ϕ i , … , ϕ n , ψ , Γ s ] H k − 2 1 ( S δ ) ≲ δ 2 1 j = i ∑ ∥ ϕ j ∥ H k ( S δ ) + δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) , k ≥ 2. ◻
By a method similar to that in Remark (?) , for each i i i , we also have
∥ D ϕ j ϕ i ~ [ ϕ 1 , … , ϕ i − 1 , ϕ i + 1 , … , ϕ n ] v ∥ H k ( S δ ) ≲ δ 1 2 ∥ v ∥ H k ( S δ ) for all j ≠ i , v ∈ H k ( S δ ) , \begin{equation}
\left\|D_{\phi_j}\widetilde{\phi_i}[\phi_1,\dots,\phi_{i-1},\phi_{i+1},\dots,\phi_n]v\right\|_{H^{k}(\mathcal{S}_\delta)}\lesssim\delta^{\frac{1}{2}}\|v\|_{H^{k}(\mathcal{S}_\delta)}\quad\text{for all }j\neq i, v\in H^{k}(\mathcal{S}_\delta), \tag*{(4.19)}
\end{equation} D ϕ j ϕ i [ ϕ 1 , … , ϕ i − 1 , ϕ i + 1 , … , ϕ n ] v H k ( S δ ) ≲ δ 2 1 ∥ v ∥ H k ( S δ ) for all j = i , v ∈ H k ( S δ ) , (4.19)
where
D ϕ j ϕ i ~ [ ϕ 1 , … , ϕ i − 1 , ϕ i + 1 , … , ϕ n ] v : = d d t ϕ i ~ [ ϕ 1 , … , ϕ j + t v , … , ϕ i − 1 , ϕ i + 1 , … , ϕ n ] ∣ t = 0 D_{\phi_j}\widetilde{\phi_i}[\phi_1,\dots,\phi_{i-1},\phi_{i+1},\dots,\phi_n]v:=\frac{\mathrm{d}}{\mathrm{d}t}\widetilde{\phi_i}[\phi_1,\dots,\phi_j+tv,\dots,\phi_{i-1},\phi_{i+1},\dots,\phi_n]\bigg|_{t=0} D ϕ j ϕ i [ ϕ 1 , … , ϕ i − 1 , ϕ i + 1 , … , ϕ n ] v := d t d ϕ i [ ϕ 1 , … , ϕ j + t v , … , ϕ i − 1 , ϕ i + 1 , … , ϕ n ] t = 0
We employ the idea in [19 section 9.2] , repeatedly use Proposition 4.3 and (4.19) to construct the functions ϕ 1 , … , ϕ n \phi_1,\dots,\phi_n ϕ 1 , … , ϕ n . The main idea is to solve the inner equations one by one, regarding the functions already constructed as fixed functions in the subsequent equations.
More precisely, we first fix ( ϕ 2 , … , ϕ n ) (\phi_2,\dots,\phi_n) ( ϕ 2 , … , ϕ n ) and apply Proposition 4.3 with i = 1 i=1 i = 1 . This gives a unique fixed point
ϕ ~ 1 [ ϕ 2 , … , ϕ n ] ∈ B i n 1 . \widetilde\phi_1[\phi_2,\dots,\phi_n]\in B_{in}^1. ϕ 1 [ ϕ 2 , … , ϕ n ] ∈ B in 1 .
We then substitute this expression for ϕ 1 \phi_1 ϕ 1 into the equation for ϕ 2 \phi_2 ϕ 2 and apply Proposition 4.3 again. In this way, ϕ 2 \phi_2 ϕ 2 is determined as a function of ( ϕ 3 , … , ϕ n ) (\phi_3,\dots,\phi_n) ( ϕ 3 , … , ϕ n ) . Continuing this procedure, we successively determine ϕ i \phi_i ϕ i as a function of the remaining unknowns ( ϕ i + 1 , … , ϕ n ) (\phi_{i+1},\dots,\phi_n) ( ϕ i + 1 , … , ϕ n ) .
We next proceed to construct ϕ 2 \phi_2 ϕ 2 . For fixed ( ϕ 2 , … , ϕ n ) (\phi_2,\dots,\phi_n) ( ϕ 2 , … , ϕ n ) , the first equation has already been solved and gives ϕ ~ 1 [ ϕ 2 , … , ϕ n ] ∈ B i n 1 \widetilde\phi_1[\phi_2,\dots,\phi_n]\in B_{\mathrm{in}}^1 ϕ 1 [ ϕ 2 , … , ϕ n ] ∈ B in 1 . We substitute this expression into the second equation and regard ϕ 3 , … , ϕ n \phi_3,\dots,\phi_n ϕ 3 , … , ϕ n as fixed. Thus, the fixed-point problem for ϕ 2 \phi_2 ϕ 2 takes the form ϕ 2 = T i n 2 [ ϕ 2 ] : = ( L i n 2 ) − 1 [ f i n [ ϕ ~ 1 [ ϕ 2 , … , ϕ n ] , ϕ 2 , … , ϕ n , ψ ~ , Γ ~ s ] + c i [ ϕ ~ 1 [ ϕ 2 , … , ϕ n ] , ϕ 2 , … , ϕ n , ψ ~ , Γ ~ s ] ∂ x 2 U i ] . \begin{aligned}
&\phi_2=\mathrm{T}_{in}^2\left[\phi_2\right]\\
&:=(L_{in}^2)^{-1}\left[f_{in}[\widetilde\phi_1[\phi_2,\dots,\phi_n],\phi_2,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]+c_i[\widetilde\phi_1[\phi_2,\dots,\phi_n],\phi_2,\dots,\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}] {\partial} _{x_2} U_i\right].
\end{aligned} ϕ 2 = T in 2 [ ϕ 2 ] := ( L in 2 ) − 1 [ f in [ ϕ 1 [ ϕ 2 , … , ϕ n ] , ϕ 2 , … , ϕ n , ψ , Γ s ] + c i [ ϕ 1 [ ϕ 2 , … , ϕ n ] , ϕ 2 , … , ϕ n , ψ , Γ s ] ∂ x 2 U i ] . By Proposition 4.3 and (4.19) , for δ > 0 \delta>0 δ > 0 sufficiently small, the above map is a contraction on B i n 2 B_{in}^2 B in 2 . Therefore, there exists a unique
ϕ ~ 2 [ ϕ 3 , … , ϕ n ] ∈ B i n 2 , \widetilde\phi_2[\phi_3,\dots,\phi_n]\in B_{in}^2, ϕ 2 [ ϕ 3 , … , ϕ n ] ∈ B in 2 ,
and ϕ ~ 2 \widetilde\phi_2 ϕ 2 satisfies the estimate
∥ ϕ ~ 2 [ ϕ 3 , … , ϕ n ] ∥ H k ( S δ ) ≲ δ 1 2 ∑ j = 3 n ∥ ϕ j ∥ H k ( S δ ) + δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ , \left\|\widetilde\phi_2[\phi_3,\dots,\phi_n]\right\|_{H^k(\mathcal{S}_\delta)}\lesssim
\delta^{\frac12}
\sum_{j=3}^n\|\phi_j\|_{H^k(\mathcal{S}_\delta)}+\delta^{\frac14}|\log\delta|^{\frac12}e^{-\frac{d_m}{\delta}}+\delta^{-k}e^{-\frac{2(1-|\boldsymbol{\tau}|)}{\delta}}, ϕ 2 [ ϕ 3 , … , ϕ n ] H k ( S δ ) ≲ δ 2 1 j = 3 ∑ n ∥ ϕ j ∥ H k ( S δ ) + δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) ,
∥ ϕ ~ 2 [ ϕ 3 , … , ϕ n ] ∥ C 1 , α ( B i , 2 ) ≲ δ 1 2 ∑ j = 3 n ∥ ϕ j ∥ H k ( S δ ) + δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ for k ≥ 2. \left\|\widetilde\phi_2[\phi_3,\dots,\phi_n]\right\|_{C^{1,\alpha}(B_{i,2})}\lesssim\delta^{\frac{1}{2}}\sum_{j=3}^n\left\|\phi_j\right\|_{H^{k}(\mathcal{S}_\delta)}+\delta^{\frac 14} | \log \delta|^{\frac{1}{2}}e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}\quad\text{for } k\geq 2. ϕ 2 [ ϕ 3 , … , ϕ n ] C 1 , α ( B i , 2 ) ≲ δ 2 1 j = 3 ∑ n ∥ ϕ j ∥ H k ( S δ ) + δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) for k ≥ 2.
The construction for ϕ ~ 3 , … , ϕ ~ n \widetilde\phi_3,\dots,\widetilde\phi_n ϕ 3 , … , ϕ n is analogous.
Finally, solving the equation for ϕ n \phi_n ϕ n and successively substituting the resulting estimate back into those for ϕ n − 1 , … , ϕ 1 \phi_{n-1},\dots,\phi_1 ϕ n − 1 , … , ϕ 1 , we obtain, for all i = 1 , … , n i=1,\dots,n i = 1 , … , n ,
∥ ϕ ~ i ∥ H k ( S δ ) ≲ δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ , \begin{equation}
\left\|\widetilde\phi_i\right\|_{H^k(\mathcal{S}_\delta)}\lesssim\delta^{\frac14}|\log\delta|^{\frac12}e^{-\frac{d_m}{\delta}}+\delta^{-k}e^{-\frac{2(1-|\boldsymbol{\tau}|)}{\delta}}, \tag*{(4.20)}
\end{equation} ϕ i H k ( S δ ) ≲ δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) , (4.20)
and by Proposition (?) , Proposition (?)
∥ ψ ~ ∥ H k ( S δ ) ≲ δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ , \begin{equation}
\left\|\widetilde{\psi}\right\|_{H^{k}(\mathcal{S}_\delta)}\lesssim\delta^{\frac 14} |\log \delta|^{\frac{1}{2}}e^{-\frac {d_m}\delta}+\delta^{-k}e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}, \tag*{(4.21)}
\end{equation} ψ H k ( S δ ) ≲ δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) , (4.21)
∥ Γ ~ s ∥ H k ( R ) ≲ δ 1 − k e − 2 ( 1 − ∣ τ ∣ ) δ + δ 1 − k ∣ log δ ∣ e − 2 d m δ . \begin{equation}
\left\|\widetilde{\Gamma}_{\mathrm{s}}\right\|_{H^k( {\mathbb{R}})}\lesssim\delta^{1-k} e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{1-k}|\log \delta|e^{-\frac {2d_m}\delta}. \tag*{(4.22)}
\end{equation} Γ s H k ( R ) ≲ δ 1 − k e − δ 2 ( 1 − ∣ τ ∣ ) + δ 1 − k ∣ log δ ∣ e − δ 2 d m . (4.22)
Proof of the main result
This section is devoted to prove the main result of this paper. In § 4 , we constructed ( ϕ ~ 1 , … , ϕ ~ n , ψ ~ , Γ ~ s ) (\widetilde\phi_1,\dots,\widetilde\phi_n,\widetilde\psi,\widetilde{\Gamma}_{\mathrm{s}}) ( ϕ 1 , … , ϕ n , ψ , Γ s ) satisfying the following system
{ − Δ ϕ ~ i + γ ′ ( U i ) ϕ ~ i + f i n [ ϕ ~ 1 , … , ϕ ~ n , ψ ~ , Γ ~ s ] + c i ∂ x 2 U i = 0 , in S δ , i = 1 , … , n , − Δ ψ ~ + ψ ~ = f o u t [ ϕ ~ 1 , … , ϕ ~ n , ψ ~ , Γ ~ s ] , in S δ , ψ ~ = 0 , on ∂ S δ , ( g − α 2 ( ∂ x 1 ) 2 ) Γ ~ s + G ( τ , ϕ ~ 1 , … , ϕ ~ n , ψ ~ , Γ ~ s ) = 0 , on R . \begin{equation}
\begin{cases}
-\Delta\widetilde\phi_i+\gamma^\prime\left(U_i\right)\widetilde\phi_i+f_{in}[\widetilde\phi_1,\dots,\widetilde\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]+c_i {\partial} _{x_2} U_i=0,\;&\text{in }\mathcal{S}_\delta, i=1,\dots,n,\\
-\Delta\widetilde\psi+\widetilde\psi=f_{out}[\widetilde\phi_1,\dots,\widetilde\phi_n,\widetilde\psi,\widetilde\Gamma_{\mathrm{s}}],\;&\text{in }\mathcal{S}_\delta,\\
\widetilde\psi=0,\;&\text{on }\partial\mathcal{S}_\delta,\\
\left(g-\alpha^2({\partial_{x_1}})^2\right)\widetilde\Gamma_{\mathrm{s}}+G(\boldsymbol{\tau},\widetilde\phi_1,\dots,\widetilde\phi_n,\widetilde\psi,\widetilde\Gamma_{\mathrm{s}})=0,\;&\textrm{on }\mathbb{R}.
\end{cases} \tag*{(5.1)}
\end{equation} ⎩ ⎨ ⎧ − Δ ϕ i + γ ′ ( U i ) ϕ i + f in [ ϕ 1 , … , ϕ n , ψ , Γ s ] + c i ∂ x 2 U i = 0 , − Δ ψ + ψ = f o u t [ ϕ 1 , … , ϕ n , ψ , Γ s ] , ψ = 0 , ( g − α 2 ( ∂ x 1 ) 2 ) Γ s + G ( τ , ϕ 1 , … , ϕ n , ψ , Γ s ) = 0 , in S δ , i = 1 , … , n , in S δ , on ∂ S δ , on R . (5.1)
Therefore, it suffices to find a suitable τ ∈ P ε \boldsymbol{\tau}\in P_\varepsilon τ ∈ P ε such that
c i = c i [ ϕ ~ 1 , … , ϕ ~ n , ψ ~ , Γ ~ s ] = 0 c_i=c_i[\widetilde\phi_1,\dots,\widetilde\phi_n,\widetilde{\psi},\widetilde\Gamma_{\mathrm{s}}]=0 c i = c i [ ϕ 1 , … , ϕ n , ψ , Γ s ] = 0
for all i = 1 , … , n . i=1,\dots,n. i = 1 , … , n . Once such a τ \boldsymbol{\tau} τ is found, let
u ~ ( x ) : = ∑ i = 1 n η i ϕ ~ i ( x ) + ψ ~ ( x ) , \tilde u(x):=\sum_{i=1}^n\eta_i\widetilde\phi_i(x)+\widetilde\psi(x), u ~ ( x ) := i = 1 ∑ n η i ϕ i ( x ) + ψ ( x ) ,
where η i ( x ) : = η ( δ R ∣ x − ( 0 , τ i δ ) ∣ ) \eta_i(x):=\eta\left(\frac{\delta}{R}
\left|x-\left(0,\frac{\tau_i}{\delta}\right)\right|
\right) η i ( x ) := η ( R δ x − ( 0 , δ τ i ) ) . We obtain that ( u ~ , Γ ~ s ) (\tilde u,\widetilde\Gamma_{\mathrm{s}}) ( u ~ , Γ s ) is a solution of the original equation (?) . For this purpose, we need the following lemma.
Lemma 5.1. Let φ i : = σ i U i − σ i ( U i ) b c + η i ϕ ~ i , \varphi_i:=\sigma_iU_i-\sigma_i(U_i)_\mathrm{bc}+\eta_i\widetilde\phi_i, φ i := σ i U i − σ i ( U i ) bc + η i ϕ i , for all i = 1 , … , n i=1,\dots,n i = 1 , … , n , we have ∣ ( ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) , η i ∂ x 2 U i ) L 2 ( S δ ) ∣ ≂ 1 , \left|\left(\frac{\partial_{x_2}\varphi_i}
{1+\widetilde{\Gamma}'(\delta\,\cdot)},\eta_i {\partial} _{x_2} U_i\right)_{L^2(\mathcal{S}_\delta)}\right|\eqsim1, ( 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i , η i ∂ x 2 U i ) L 2 ( S δ ) ≂ 1 , ∣ ( ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) , η j ∂ x 2 U j ) L 2 ( S δ ) ∣ → 0 , as δ → 0 and j ≠ i , \left|\left(\frac{\partial_{x_2}\varphi_i}
{1+\widetilde{\Gamma}'(\delta\,\cdot)},\eta_j {\partial} _{x_2} U_j\right)_{L^2(\mathcal{S}_\delta)}\right|\to 0,\quad\text{as }\delta\to0\text{ and }j\neq i, ( 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i , η j ∂ x 2 U j ) L 2 ( S δ ) → 0 , as δ → 0 and j = i , where Γ ~ = ( Γ ~ 1 , Γ ~ 2 ) \widetilde{\Gamma}=(\widetilde{\Gamma}_1,\widetilde{\Gamma}_2) Γ = ( Γ 1 , Γ 2 ) (see § 2.1 )
Proof. Recall that
φ i = σ i U i − σ i ( U i ) b c + η i ϕ ~ i , i = 1 , … , n . \varphi_i=\sigma_iU_i-\sigma_i(U_i)_{\mathrm{bc}}+\eta_i\widetilde\phi_i,\qquad i=1,\dots,n. φ i = σ i U i − σ i ( U i ) bc + η i ϕ i , i = 1 , … , n . By the Sobolev trace theorem, Sobolev embedding and (4.22) , we have
∥ Γ ~ ′ ( δ ⋅ ) ∥ L ∞ ( R ) ≲ ∥ Γ ~ s ∥ H k ( R ) ≲ δ 1 − k e − 2 ( 1 − ∣ τ ∣ ) δ + δ 1 − k ∣ log δ ∣ e − 2 d m δ = o ( 1 ) , \left\|\widetilde{\Gamma}'(\delta\,\cdot)\right\|_{L^\infty(\mathbb R)}\lesssim\left\|\widetilde{\Gamma}_{\mathrm{s}}\right\|_{H^k( {\mathbb{R}})}\lesssim\delta^{1-k} e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{1-k}|\log \delta|e^{-\frac {2d_m}\delta}=o(1), Γ ′ ( δ ⋅ ) L ∞ ( R ) ≲ Γ s H k ( R ) ≲ δ 1 − k e − δ 2 ( 1 − ∣ τ ∣ ) + δ 1 − k ∣ log δ ∣ e − δ 2 d m = o ( 1 ) , and by Corollary (?) , (4.20) and (4.21) ,
∥ ∂ x 2 ( U i ) b c ∥ L 2 ( S δ ) ≲ δ 1 4 e − 1 − ∣ τ ∣ δ = o ( 1 ) , \left\|\partial_{x_2}(U_i)_{\mathrm{bc}}\right\|_{L^2(\mathcal{S}_\delta)}\lesssim \delta^{\frac 14} e^{-\frac {1 - |\tau|}\delta}=o(1), ∥ ∂ x 2 ( U i ) bc ∥ L 2 ( S δ ) ≲ δ 4 1 e − δ 1 − ∣ τ ∣ = o ( 1 ) , ∥ u ~ ∥ H k ( S δ ) ≲ δ 1 4 ∣ log δ ∣ 1 2 e − d m δ + δ − k e − 2 ( 1 − ∣ τ ∣ ) δ = o ( 1 ) , \|\tilde u\|_{H^k(\mathcal{S}_\delta)}\lesssim\delta^{\frac14}|\log\delta|^{\frac12}e^{-\frac{d_m}{\delta}}+\delta^{-k}e^{-\frac{2(1-|\boldsymbol{\tau}|)}{\delta}}=o(1), ∥ u ~ ∥ H k ( S δ ) ≲ δ 4 1 ∣ log δ ∣ 2 1 e − δ d m + δ − k e − δ 2 ( 1 − ∣ τ ∣ ) = o ( 1 ) , as δ → 0 \delta\to0 δ → 0 , uniformly for τ ∈ P ε \boldsymbol{\tau}\in P_\varepsilon τ ∈ P ε . Hence
∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) = σ i ∂ x 2 U i + r i , \frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)}=\sigma_i\partial_{x_2}U_i+r_i, 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i = σ i ∂ x 2 U i + r i , where ∥ r i ∥ L 2 ( S δ ) = o ( 1 ) \|r_i\|_{L^2(\mathcal{S}_\delta)}=o(1) ∥ r i ∥ L 2 ( S δ ) = o ( 1 ) .
For j = i j=i j = i , we therefore obtain
( ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) , η i ∂ x 2 U i ) L 2 ( S δ ) = ( σ i ∂ x 2 U i , η i ∂ x 2 U i ) L 2 ( S δ ) + ( r i , η i ∂ x 2 U i ) L 2 ( S δ ) . \begin{align*}
\left(\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)},\eta_i\partial_{x_2}U_i\right)_{L^2(\mathcal{S}_\delta)}=\left(\sigma_i\partial_{x_2}U_i,\eta_i\partial_{x_2}U_i\right)_{L^2(\mathcal{S}_\delta)}+\left(r_i,\eta_i\partial_{x_2}U_i\right)_{L^2(\mathcal{S}_\delta)}.
\end{align*} ( 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i , η i ∂ x 2 U i ) L 2 ( S δ ) = ( σ i ∂ x 2 U i , η i ∂ x 2 U i ) L 2 ( S δ ) + ( r i , η i ∂ x 2 U i ) L 2 ( S δ ) . Since η i ≡ 1 \eta_i\equiv1 η i ≡ 1 in a neighborhood of the i i i -th concentration point and ∂ x 2 U i \partial_{x_2}U_i ∂ x 2 U i is exponentially localized, we have
( σ i ∂ x 2 U i , η i ∂ x 2 U i ) L 2 ( S δ ) = σ i ∥ ∂ x 2 U i ∥ L 2 ( S δ ) 2 + o ( 1 ) = O ( 1 ) . \left(\sigma_i\partial_{x_2}U_i,\eta_i\partial_{x_2}U_i\right)_{L^2(\mathcal{S}_\delta)}=\sigma_i\|\partial_{x_2}U_i\|_{L^2(\mathcal{S}_\delta)}^2+o(1)=O(1). ( σ i ∂ x 2 U i , η i ∂ x 2 U i ) L 2 ( S δ ) = σ i ∥ ∂ x 2 U i ∥ L 2 ( S δ ) 2 + o ( 1 ) = O ( 1 ) . On the other hand, by Cauchy-Schwarz inequality,
∣ ( r i , η i ∂ x 2 U i ) L 2 ( S δ ) ∣ ≤ ∥ r i ∥ L 2 ( S δ ) ∥ ∂ x 2 U i ∥ L 2 ( S δ ) = o ( 1 ) . \left|\left(r_i,\eta_i\partial_{x_2}U_i\right)_{L^2(\mathcal{S}_\delta)}\right|\leq\|r_i\|_{L^2(\mathcal{S}_\delta)}\|\partial_{x_2}U_i\|_{L^2(\mathcal{S}_\delta)}=o(1). ( r i , η i ∂ x 2 U i ) L 2 ( S δ ) ≤ ∥ r i ∥ L 2 ( S δ ) ∥ ∂ x 2 U i ∥ L 2 ( S δ ) = o ( 1 ) . It follows that
∣ ( ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) , η i ∂ x 2 U i ) L 2 ( S δ ) ∣ ≂ 1. \left|\left(\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)},\eta_i\partial_{x_2}U_i\right)_{L^2(\mathcal{S}_\delta)}\right|\eqsim 1. ( 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i , η i ∂ x 2 U i ) L 2 ( S δ ) ≂ 1. It remains to consider j ≠ i j\neq i j = i . In this case,
( ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) , η j ∂ x 2 U j ) L 2 ( S δ ) = ( σ i ∂ x 2 U i , η j ∂ x 2 U j ) L 2 ( S δ ) + ( r i , η j ∂ x 2 U j ) L 2 ( S δ ) . \left(\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)},\eta_j\partial_{x_2}U_j\right)_{L^2(\mathcal{S}_\delta)}=\left(\sigma_i\partial_{x_2}U_i,\eta_j\partial_{x_2}U_j\right)_{L^2(\mathcal{S}_\delta)}+\left(r_i,\eta_j\partial_{x_2}U_j\right)_{L^2(\mathcal{S}_\delta)}. ( 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i , η j ∂ x 2 U j ) L 2 ( S δ ) = ( σ i ∂ x 2 U i , η j ∂ x 2 U j ) L 2 ( S δ ) + ( r i , η j ∂ x 2 U j ) L 2 ( S δ ) . The second term tends to zero by the estimate on r i r_i r i and Cauchy-Schwarz inequality. For the first term, Lemma 2.1 and (4.7) imply
∣ ( σ i ∂ x 2 U i , η j ∂ x 2 U j ) L 2 ( S δ ) ∣ ≲ δ − 1 2 e − d m δ → 0 , j ≠ i . \begin{equation}
\left|\left(\sigma_i\partial_{x_2}U_i,\eta_j\partial_{x_2}U_j\right)_{L^2(\mathcal{S}_\delta)}\right|\lesssim\delta^{-\frac 12}e^{-\frac{d_m}\delta}\to0,\quad j\neq i. \tag*{(5.2)}
\end{equation} ( σ i ∂ x 2 U i , η j ∂ x 2 U j ) L 2 ( S δ ) ≲ δ − 2 1 e − δ d m → 0 , j = i . (5.2) Therefore,
∣ ( ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) , η j ∂ x 2 U j ) L 2 ( S δ ) ∣ → 0 , j ≠ i . \left|\left(\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)},\eta_j\partial_{x_2}U_j\right)_{L^2(\mathcal{S}_\delta)}\right|\to0,\quad j\neq i. ( 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i , η j ∂ x 2 U j ) L 2 ( S δ ) → 0 , j = i . ◻
By (5.1) , we obtain
− Δ u ~ + γ ′ ( ∑ i = 1 n σ i U i ) u ~ + F ( τ , u ~ , Γ ~ s ) + ∑ i = 1 n c i η i ∂ x 2 U i = 0. \begin{align*}
-\Delta\tilde u+\gamma^\prime\left(\textstyle\sum_{i=1}^n\sigma_iU_i\right)\tilde u+F(\boldsymbol{\tau},\tilde u,\widetilde\Gamma_{\mathrm{s}})+\sum_{i=1}^nc_i\eta_i {\partial} _{x_2} U_i=0.
\end{align*} − Δ u ~ + γ ′ ( ∑ i = 1 n σ i U i ) u ~ + F ( τ , u ~ , Γ s ) + i = 1 ∑ n c i η i ∂ x 2 U i = 0.
Define
N τ ( u ~ , Γ ~ s ) = − Δ u ~ + γ ′ ( ∑ i = 1 n σ i U i ) u ~ + F ( τ , u ~ , Γ ~ s ) , \begin{equation}
\mathcal{N}_{\boldsymbol{\tau}}(\tilde u,\widetilde\Gamma_{\mathrm{s}})=-\Delta\tilde u+\gamma^\prime\left(\textstyle\sum_{i=1}^n\sigma_iU_i\right)\tilde u+F(\boldsymbol{\tau},\tilde u,\widetilde\Gamma_{\mathrm{s}}), \tag*{(5.3)}
\end{equation} N τ ( u ~ , Γ s ) = − Δ u ~ + γ ′ ( ∑ i = 1 n σ i U i ) u ~ + F ( τ , u ~ , Γ s ) , (5.3)
a i j = ( ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) , η j ∂ x 2 U j ) L 2 ( S δ ) , a_{ij}=\left(\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)},\eta_j {\partial} _{x_2} U_j\right)_{L^2(\mathcal{S}_\delta)}, a ij = ( 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i , η j ∂ x 2 U j ) L 2 ( S δ ) ,
then
( ( N τ ( u ~ , Γ ~ s ) , ∂ x 2 φ 1 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) ( N τ ( u ~ , Γ ~ s ) , ∂ x 2 φ 2 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) ⋮ ( N τ ( u ~ , Γ ~ s ) , ∂ x 2 φ n 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) ) + ( a 11 a 12 ⋯ a 1 n a 21 a 22 ⋯ a 2 n ⋮ ⋮ ⋱ ⋮ a n 1 a n 2 ⋯ a n n ) ( c 1 c 2 ⋮ c n ) = 0. \begin{equation}
\begin{pmatrix}
\left(\mathcal{N}_{\boldsymbol{\tau}}(\tilde u,\widetilde\Gamma_{\mathrm{s}}),\frac{\partial_{x_2}\varphi_1}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}\\
\left(\mathcal{N}_{\boldsymbol{\tau}}(\tilde u,\widetilde\Gamma_{\mathrm{s}}),\frac{\partial_{x_2}\varphi_2}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}\\
\vdots\\
\left(\mathcal{N}_{\boldsymbol{\tau}}(\tilde u,\widetilde\Gamma_{\mathrm{s}}),\frac{\partial_{x_2}\varphi_n}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}
\end{pmatrix}+\begin{pmatrix}
a_{11} & a_{12} & \cdots & a_{1n}\\
a_{21} & a_{22} & \cdots & a_{2n}\\
\vdots & \vdots & \ddots & \vdots\\
a_{n1} & a_{n2} & \cdots & a_{nn}
\end{pmatrix}\begin{pmatrix}
c_1\\
c_2\\
\vdots\\
c_n
\end{pmatrix}=0. \tag*{(5.4)}
\end{equation} ( N τ ( u ~ , Γ s ) , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ 1 ) L 2 ( S δ ) ( N τ ( u ~ , Γ s ) , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ 2 ) L 2 ( S δ ) ⋮ ( N τ ( u ~ , Γ s ) , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ n ) L 2 ( S δ ) + a 11 a 21 ⋮ a n 1 a 12 a 22 ⋮ a n 2 ⋯ ⋯ ⋱ ⋯ a 1 n a 2 n ⋮ a nn c 1 c 2 ⋮ c n = 0. (5.4)
By Lemma 5.1 , for sufficiently small δ \delta δ , the matrix ( a i j ) 1 ≤ i , j ≤ n (a_{ij})_{1\leq i,j\leq n} ( a ij ) 1 ≤ i , j ≤ n is strictly diagonally dominant, i.e.,
∣ a i i ∣ > ∑ j ≠ i ∣ a i j ∣ , i = 1 , … , n . |a_{ii}|>\sum_{j\neq i}|a_{ij}|,
\quad i=1,\dots,n. ∣ a ii ∣ > j = i ∑ ∣ a ij ∣ , i = 1 , … , n .
Hence ( a i j ) 1 ≤ i , j ≤ n (a_{ij})_{1\leq i,j\leq n} ( a ij ) 1 ≤ i , j ≤ n is invertible. Then, from (5.4) we obtain
( c 1 c 2 ⋮ c n ) = 0 ⟺ ( ( N τ ( u ~ , Γ ~ s ) , ∂ x 2 φ 1 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) ( N τ ( u ~ , Γ ~ s ) , ∂ x 2 φ 2 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) ⋮ ( N τ ( u ~ , Γ ~ s ) , ∂ x 2 φ n 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) ) = 0. \begin{pmatrix}
c_1\\
c_2\\
\vdots\\
c_n
\end{pmatrix}=0\quad\Longleftrightarrow\quad\begin{pmatrix}
\left(\mathcal{N}_{\boldsymbol{\tau}}(\tilde u,\widetilde\Gamma_{\mathrm{s}}),\frac{\partial_{x_2}\varphi_1}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}\\
\left(\mathcal{N}_{\boldsymbol{\tau}}(\tilde u,\widetilde\Gamma_{\mathrm{s}}),\frac{\partial_{x_2}\varphi_2}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}\\
\vdots\\
\left(\mathcal{N}_{\boldsymbol{\tau}}(\tilde u,\widetilde\Gamma_{\mathrm{s}}),\frac{\partial_{x_2}\varphi_n}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}
\end{pmatrix}=0. c 1 c 2 ⋮ c n = 0 ⟺ ( N τ ( u ~ , Γ s ) , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ 1 ) L 2 ( S δ ) ( N τ ( u ~ , Γ s ) , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ 2 ) L 2 ( S δ ) ⋮ ( N τ ( u ~ , Γ s ) , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ n ) L 2 ( S δ ) = 0.
Thus, the required conditions
c i = 0 , i = 1 , … , n , c_i=0,\quad i=1,\dots,n, c i = 0 , i = 1 , … , n ,
are equivalent to the reduced system
b i ( τ ) : = ( N τ ( u ~ , Γ ~ s ) , ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) = 0 , i = 1 , … , n . b_i(\boldsymbol{\tau}):=\left(\mathcal{N}_{\boldsymbol{\tau}}(\tilde u,\widetilde\Gamma_{\mathrm{s}}),\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}=0,
\quad i=1,\dots,n. b i ( τ ) := ( N τ ( u ~ , Γ s ) , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i ) L 2 ( S δ ) = 0 , i = 1 , … , n .
Since k 0 ≥ 3 k_0\geq3 k 0 ≥ 3 , by [1 Lemma 3.8] , for any integer 2 ≤ l ≤ k 0 2\leq l\leq k_0 2 ≤ l ≤ k 0 , both F ( τ , u , Γ s ) F(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}}) F ( τ , u , Γ s ) and G ( τ , u , Γ s ) G(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}}) G ( τ , u , Γ s ) belong to H e l − 1 ( S δ ) H_{\mathrm{e}}^{l-1}(\mathcal{S}_\delta) H e l − 1 ( S δ ) , then let l = 3 l=3 l = 3 , using Sobolev embedding,
F ( τ , u , Γ s ) ∈ H 2 ( S δ ) , ∥ F ( τ , u , Γ s ) ∥ L ∞ ( S δ ) ≲ ∥ F ( τ , u , Γ s ) ∥ H 2 ( S δ ) . \begin{equation}
F(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}})\in H^2(\mathcal{S}_\delta),\qquad \|F(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}})\|_{L^\infty(\mathcal{S}_\delta)}\lesssim\|F(\boldsymbol{\tau},u,\Gamma_{\mathrm{s}})\|_{H^2(\mathcal{S}_\delta)}. \tag*{(5.5)}
\end{equation} F ( τ , u , Γ s ) ∈ H 2 ( S δ ) , ∥ F ( τ , u , Γ s ) ∥ L ∞ ( S δ ) ≲ ∥ F ( τ , u , Γ s ) ∥ H 2 ( S δ ) . (5.5)
Hence we may choose k = 2 k=2 k = 2 throughout the proof below.
Proof of Theorem 1.1 . The above argument shows that it suffices to find τ ∈ P ε \boldsymbol{\tau} \in P_\varepsilon τ ∈ P ε such that b i ( τ ) = 0 b_i(\boldsymbol{\tau}) = 0 b i ( τ ) = 0 for all i = 1 , … , n i=1,\dots,n i = 1 , … , n .
Substituting the expression (5.3) of N τ \mathcal{N}_{\boldsymbol{\tau}} N τ into b i ( τ ) b_i(\boldsymbol{\tau}) b i ( τ ) , we obtain b i ( τ ) = ( − Δ ( ω τ + u ~ ) + ∣ 1 + Γ ′ ( δ ⋅ ) ∣ 2 γ ( ω τ + u ~ ) , ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) = ( − Δ ( σ i U i − σ i ( U i ) b c + η i ϕ ~ i ) , ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) + ( ∣ 1 + Γ ~ ′ ( δ ⋅ ) ∣ 2 γ ( σ i U i − σ i ( U i ) b c + η i ϕ ~ i ) , ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) + ( ∣ 1 + Γ ~ ′ ( δ ⋅ ) ∣ 2 ( γ ( ω τ + u ~ ) − γ ( σ i U i − σ i ( U i ) b c + η i ϕ ~ i ) ) − ∑ j ≠ i σ j γ ( U j ) , ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) + ( ∑ j ≠ i σ j ( U j ) b c − ∑ j ≠ i Δ ( η j ϕ ~ j ) − Δ ψ ~ , ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) = : I 1 + I 2 + I 3 + I 4 . \begin{aligned}
&b_i(\boldsymbol{\tau})=\left(-\Delta(\omega_{\boldsymbol{\tau}}+\tilde u)+\left|1+\Gamma^\prime(\delta \,\cdot\,)\right|^2\gamma(\omega_{\boldsymbol{\tau}}+\tilde u),\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}\\
&=\left(-\Delta(\sigma_iU_i-\sigma_i(U_i)_\mathrm{bc}+\eta_i\widetilde\phi_i),\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}\\
&\quad\,+\left(\left|1+\widetilde{\Gamma}^\prime(\delta \,\cdot\,)\right|^2\gamma(\sigma_iU_i-\sigma_i(U_i)_\mathrm{bc}+\eta_i\widetilde\phi_i) ,\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}\\
&\quad\,+\left(\left|1+\widetilde{\Gamma}^\prime(\delta \,\cdot\,)\right|^2\left(\gamma(\omega_{\boldsymbol{\tau}}+\tilde u)-\gamma(\sigma_iU_i-\sigma_i(U_i)_\mathrm{bc}+\eta_i\widetilde\phi_i)\right)-\sum_{j\neq i}\sigma_j\gamma(U_j) ,\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}\\
&\quad\,+\left(\sum_{j\neq i}\sigma_j(U_j)_\mathrm{bc}-\sum_{j\neq i}\Delta(\eta_j\widetilde\phi_j)-\Delta\widetilde\psi,\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}\\
&=:I_1+I_2+I_3+I_4.
\end{aligned} b i ( τ ) = ( − Δ ( ω τ + u ~ ) + ∣ 1 + Γ ′ ( δ ⋅ ) ∣ 2 γ ( ω τ + u ~ ) , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i ) L 2 ( S δ ) = ( − Δ ( σ i U i − σ i ( U i ) bc + η i ϕ i ) , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i ) L 2 ( S δ ) + ( 1 + Γ ′ ( δ ⋅ ) 2 γ ( σ i U i − σ i ( U i ) bc + η i ϕ i ) , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i ) L 2 ( S δ ) + 1 + Γ ′ ( δ ⋅ ) 2 ( γ ( ω τ + u ~ ) − γ ( σ i U i − σ i ( U i ) bc + η i ϕ i ) ) − j = i ∑ σ j γ ( U j ) , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i L 2 ( S δ ) + j = i ∑ σ j ( U j ) bc − j = i ∑ Δ ( η j ϕ j ) − Δ ψ , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i L 2 ( S δ ) =: I 1 + I 2 + I 3 + I 4 .
For I 1 I_1 I 1 and I 2 I_2 I 2 , we use the method in Section 5 of [1 ] . By the definition of φ i \varphi_i φ i and the boundary correction operator, we have
I 1 = ( − Δ φ i , ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) . I_1=\left(-\Delta\varphi_i,\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}. I 1 = ( − Δ φ i , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i ) L 2 ( S δ ) . Let
y = x + δ − 1 Γ ~ ( δ x ) y=x+\delta^{-1}\widetilde{\Gamma}(\delta x) y = x + δ − 1 Γ ( δ x ) and denote the corresponding image domain by Ω ( τ ) = ( i d + δ − 1 Γ ~ ( δ ⋅ ) ) ( S δ ) \Omega(\boldsymbol{\tau})=(\mathrm{id}+\delta^{-1}\widetilde{\Gamma}(\delta\,\cdot))(\mathcal{S}_\delta) Ω ( τ ) = ( id + δ − 1 Γ ( δ ⋅ )) ( S δ ) . Let
Ψ i ( y ) : = φ i ∘ ( id + δ − 1 Γ ~ ( δ ⋅ ) ) − 1 , \Psi_i(y):=\varphi_i\circ
\left(\operatorname{id}
+\delta^{-1}\widetilde{\Gamma}(\delta\,\cdot)\right)^{-1}, Ψ i ( y ) := φ i ∘ ( id + δ − 1 Γ ( δ ⋅ ) ) − 1 , then
∂ y 2 Ψ i = ( ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) ) ∘ ( id + δ − 1 Γ ~ ( δ ⋅ ) ) − 1 . \partial_{y_2}\Psi_i=\left(\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)\circ
\left(\operatorname{id}
+\delta^{-1}\widetilde{\Gamma}(\delta\,\cdot)\right)^{-1}. ∂ y 2 Ψ i = ( 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i ) ∘ ( id + δ − 1 Γ ( δ ⋅ ) ) − 1 . Hence, by the conformal invariance of the Laplacian,
I 1 = ∫ Ω ( τ ) ( − Δ y Ψ i ) ∂ y 2 Ψ i d y . \begin{equation}
I_1=\int_{\Omega(\boldsymbol{\tau})}(-\Delta_y\Psi_i)\partial_{y_2}\Psi_i\,dy. \tag*{(5.6)}
\end{equation} I 1 = ∫ Ω ( τ ) ( − Δ y Ψ i ) ∂ y 2 Ψ i d y . (5.6) Since Ψ i = 0 \Psi_i=0 Ψ i = 0 on ∂ Ω ( τ ) \partial\Omega(\boldsymbol{\tau}) ∂ Ω ( τ ) , an integration by parts gives
I 1 = − ∫ Ω ( τ ) Δ y Ψ i ∂ y 2 Ψ i d y = ∫ Ω ( τ ) ∇ ∂ y 2 Ψ i ⋅ ∇ Ψ i d y − ∫ ∂ Ω ( τ ) ∂ y 2 Ψ i ∇ Ψ i ⋅ N d S = 1 2 ∫ ∂ Ω ( τ ) ∣ ∇ Ψ i ∣ 2 N 2 d S − ∫ ∂ Ω ( τ ) ∂ y 2 Ψ i ∇ Ψ i ⋅ N d S = − 1 2 ∫ ∂ Ω ( τ ) ∣ ∇ Ψ i ∣ 2 N 2 d S , \begin{align*}
I_1=-\int_{\Omega(\boldsymbol{\tau})}\Delta_y\Psi_i\,\partial_{y_2}\Psi_i\,dy&=\int_{\Omega(\boldsymbol{\tau})}\nabla\partial_{y_2}\Psi_i\cdot\nabla\Psi_i\,dy-\int_{\partial\Omega(\boldsymbol{\tau})}\partial_{y_2}\Psi_i\,\nabla\Psi_i\cdot N\,dS\\
&=\frac{1}{2}\int_{\partial\Omega(\boldsymbol{\tau})}|\nabla\Psi_i|^2N_2\,dS-\int_{\partial\Omega(\boldsymbol{\tau})}\partial_{y_2}\Psi_i\,\nabla\Psi_i\cdot N\,dS\\
&=-\frac{1}{2}\int_{\partial\Omega(\boldsymbol{\tau})}|\nabla\Psi_i|^2N_2\,dS,
\end{align*} I 1 = − ∫ Ω ( τ ) Δ y Ψ i ∂ y 2 Ψ i d y = ∫ Ω ( τ ) ∇ ∂ y 2 Ψ i ⋅ ∇ Ψ i d y − ∫ ∂ Ω ( τ ) ∂ y 2 Ψ i ∇ Ψ i ⋅ N d S = 2 1 ∫ ∂ Ω ( τ ) ∣∇ Ψ i ∣ 2 N 2 d S − ∫ ∂ Ω ( τ ) ∂ y 2 Ψ i ∇ Ψ i ⋅ N d S = − 2 1 ∫ ∂ Ω ( τ ) ∣∇ Ψ i ∣ 2 N 2 d S , where
N = ( N 1 , N 2 ) = ± 1 ∣ 1 + Γ ~ ′ ( δ ⋅ ) ∣ ( − ∂ x 1 Γ ~ 2 ( δ ⋅ ) , 1 + ∂ x 2 Γ ~ 2 ( δ ⋅ ) ) ∘ ( id + δ − 1 Γ ~ ( δ ⋅ ) ) − 1 N=(N_1,N_2)=\pm\frac{1}{\left|1+\widetilde{\Gamma}'(\delta\,\cdot)\right|}\left(
-\partial_{x_1}\widetilde{\Gamma}_2(\delta\,\cdot),1+\partial_{x_2}\widetilde{\Gamma}_2(\delta\,\cdot)
\right)\circ\left(\operatorname{id}
+\delta^{-1}\widetilde{\Gamma}(\delta\,\cdot)\right)^{-1} N = ( N 1 , N 2 ) = ± 1 + Γ ′ ( δ ⋅ ) 1 ( − ∂ x 1 Γ 2 ( δ ⋅ ) , 1 + ∂ x 2 Γ 2 ( δ ⋅ ) ) ∘ ( id + δ − 1 Γ ( δ ⋅ ) ) − 1 denotes the outward unit normal vector and the sign is chosen according to the upper or lower component of ∂ Ω ~ ( τ ) \partial\tilde{\Omega}(\boldsymbol{\tau}) ∂ Ω ~ ( τ ) respectively. Moreover, the boundary arc-length element is given by
d S = ∣ 1 + Γ ~ ′ ( δ ⋅ ) ∣ ∘ ( id + δ − 1 Γ ~ ( δ ⋅ ) ) − 1 d x 1 . dS=\left|1+\widetilde{\Gamma}'(\delta\,\cdot)\right|\circ\left(\operatorname{id}
+\delta^{-1}\widetilde{\Gamma}(\delta\,\cdot)\right)^{-1}\,dx_1. d S = 1 + Γ ′ ( δ ⋅ ) ∘ ( id + δ − 1 Γ ( δ ⋅ ) ) − 1 d x 1 . Using the expression for the outward unit normal vector and the arc-length element, we can rewrite I 1 I_1 I 1 as an integral over the boundary of S δ \mathcal{S}_\delta S δ . Since φ i ∣ ∂ S δ = 0 \varphi_i|_{\partial\mathcal{S}_\delta}=0 φ i ∣ ∂ S δ = 0 , tangential derivatives do not appear in the boundary integrals. Reversing the coordinate change, we obtain
I 1 = − 1 2 ∫ R 1 + ∂ x 2 Γ ~ 2 ( δ x 1 , 1 ) ∣ 1 + Γ ~ ′ ( δ x 1 , 1 ) ∣ 2 ∣ ∂ x 2 φ i ( x 1 , 1 δ ) ∣ 2 d x 1 + 1 2 ∫ R 1 + ∂ x 2 Γ ~ 2 ( δ x 1 , − 1 ) ∣ 1 + Γ ~ ′ ( δ x 1 , − 1 ) ∣ 2 ∣ ∂ x 2 φ i ( x 1 , − 1 δ ) ∣ 2 d x 1 . \begin{equation}
\begin{aligned}
I_1=&-\frac{1}{2}\displaystyle\int_{\mathbb R}\frac{1+\partial_{x_2}\widetilde{\Gamma}_2(\delta x_1,1)}{\left|1+\widetilde{\Gamma}'(\delta x_1,1)\right|^2}\left|\partial_{x_2}\varphi_i\left(x_1,\frac{1}{\delta}\right)\right|^2\,dx_1\\
&+\frac{1}{2}\displaystyle\int_{\mathbb R}\frac{1+\partial_{x_2}\widetilde{\Gamma}_2(\delta x_1,-1)}{\left|1+\widetilde{\Gamma}'(\delta x_1,-1)\right|^2}\left|\partial_{x_2}\varphi_i\left(x_1,-\frac{1}{\delta}\right)\right|^2\,dx_1.
\end{aligned} \tag*{(5.7)}
\end{equation} I 1 = − 2 1 ∫ R 1 + Γ ′ ( δ x 1 , 1 ) 2 1 + ∂ x 2 Γ 2 ( δ x 1 , 1 ) ∂ x 2 φ i ( x 1 , δ 1 ) 2 d x 1 + 2 1 ∫ R 1 + Γ ′ ( δ x 1 , − 1 ) 2 1 + ∂ x 2 Γ 2 ( δ x 1 , − 1 ) ∂ x 2 φ i ( x 1 , − δ 1 ) 2 d x 1 . (5.7) For the upper boundary, using Lemma 2.1 , Corollary (?) and (4.20) , we obtain
∫ R ∣ ∂ x 2 φ i ( x 1 , 1 δ ) ∣ 2 d x 1 ≂ δ 1 / 2 e − 2 ( 1 − τ i ) δ . \int_{\mathbb R}\left|\partial_{x_2}\varphi_i\left(x_1,\frac{1}{\delta}\right)\right|^2\,dx_1\eqsim\delta^{1/2}e^{-\frac {2(1 - \tau_i)}\delta}. ∫ R ∂ x 2 φ i ( x 1 , δ 1 ) 2 d x 1 ≂ δ 1/2 e − δ 2 ( 1 − τ i ) . Similarly, for the lower boundary,
∫ R ∣ ∂ x 2 φ i ( x 1 , − 1 δ ) ∣ 2 d x 1 ≂ δ 1 / 2 e − 2 ( 1 + τ i ) δ . \int_{\mathbb R}\left|\partial_{x_2}\varphi_i\left(x_1,-\frac{1}{\delta}\right)\right|^2\,dx_1\eqsim\delta^{1/2}e^{-\frac {2(1 +\tau_i)}\delta}. ∫ R ∂ x 2 φ i ( x 1 , − δ 1 ) 2 d x 1 ≂ δ 1/2 e − δ 2 ( 1 + τ i ) . From the proof of Lemma 5.1 ,
∣ 1 + Γ ~ ′ ∣ = 1 + O ( δ − 1 e − 2 ( 1 − ∣ τ ∣ ) δ + δ − 1 ∣ log δ ∣ e − 2 d m δ ) , \left|1+\widetilde{\Gamma}'\right|=1+O(\delta^{-1} e^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}+\delta^{-1}|\log \delta|e^{-\frac {2d_m}\delta}), 1 + Γ ′ = 1 + O ( δ − 1 e − δ 2 ( 1 − ∣ τ ∣ ) + δ − 1 ∣ log δ ∣ e − δ 2 d m ) , the factors involving Γ ~ \widetilde{\Gamma} Γ in (5.7) contribute only to the lower-order terms. Hence
I 1 = C 1 δ 1 2 ( e − 2 ( 1 + τ i ) δ − e − 2 ( 1 − τ i ) δ ) + O ( δ − 1 2 e − 3 ( 1 − ∣ τ ∣ ) δ + δ − 1 2 ∣ log δ ∣ e − 2 d m + 2 ( 1 − ∣ τ ∣ ) δ ) , \begin{equation}
I_1=C_1\delta^{\frac 12}\left(e^{-\frac {2(1 +\tau_i)}\delta}-e^{-\frac{2(1-\tau_i)}\delta}\right)+O\left(\delta^{-\frac 12}e^{-\frac {3(1-|\boldsymbol{\tau}|)}\delta}+\delta^{-\frac 12}|\log \delta|e^{-\frac {2d_m+2(1-|\boldsymbol{\tau}|)}\delta}\right), \tag*{(5.8)}
\end{equation} I 1 = C 1 δ 2 1 ( e − δ 2 ( 1 + τ i ) − e − δ 2 ( 1 − τ i ) ) + O ( δ − 2 1 e − δ 3 ( 1 − ∣ τ ∣ ) + δ − 2 1 ∣ log δ ∣ e − δ 2 d m + 2 ( 1 − ∣ τ ∣ ) ) , (5.8) where C 1 C_1 C 1 is a positive constant.
We next consider I 2 I_2 I 2 , which can be handled directly by writing γ \gamma γ as a derivative. Analogously to (5.6) , we have
I 2 = ∫ Ω ( τ ) γ ( Ψ i ) ∂ y 2 Ψ i d y . I_2=\int_{\Omega(\boldsymbol{\tau})}\gamma(\Psi_i)\partial_{y_2}\Psi_i\,dy. I 2 = ∫ Ω ( τ ) γ ( Ψ i ) ∂ y 2 Ψ i d y . Let G \mathcal{G} G be an antiderivative of γ \gamma γ satisfying G ( 0 ) = 0 \mathcal{G}(0)=0 G ( 0 ) = 0 , that is,
G ′ ( s ) = γ ( s ) . \mathcal{G}'(s)=\gamma(s). G ′ ( s ) = γ ( s ) . Then
I 2 = ∫ Ω ( τ ) γ ( Ψ i ) ∂ y 2 Ψ i d y = ∫ Ω ( τ ) ∂ y 2 G ( Ψ i ) d y = ∫ ∂ Ω ( τ ) G ( Ψ i ) N 2 d S . \begin{align*}
I_2&=\int_{\Omega(\boldsymbol{\tau})}\gamma(\Psi_i)\partial_{y_2}\Psi_i\,dy=\int_{\Omega(\boldsymbol{\tau})}\partial_{y_2}\mathcal{G}(\Psi_i)\,dy=\int_{\partial\Omega(\boldsymbol{\tau})}\mathcal{G}(\Psi_i)N_2\,dS.
\end{align*} I 2 = ∫ Ω ( τ ) γ ( Ψ i ) ∂ y 2 Ψ i d y = ∫ Ω ( τ ) ∂ y 2 G ( Ψ i ) d y = ∫ ∂ Ω ( τ ) G ( Ψ i ) N 2 d S . Since Ψ i = 0 \Psi_i=0 Ψ i = 0 on ∂ Ω ( τ ) \partial\Omega(\boldsymbol{\tau}) ∂ Ω ( τ ) and G ( 0 ) = 0 \mathcal{G}(0)=0 G ( 0 ) = 0 , it follows that
I 2 = ∫ ∂ Ω ( τ ) G ( Ψ i ) N 2 d S = 0. \begin{equation}
I_2=\int_{\partial\Omega(\boldsymbol{\tau})}\mathcal{G}(\Psi_i)N_2\,dS=0. \tag*{(5.9)}
\end{equation} I 2 = ∫ ∂ Ω ( τ ) G ( Ψ i ) N 2 d S = 0. (5.9) For I 3 I_3 I 3 , by (4.22) ,
I 3 ≂ ( γ ( ω τ + u ~ ) − γ ( σ i U i − σ i ( U i ) b c + η i ϕ ~ i ) − ∑ j ≠ i σ j γ ( U j ) , ∂ x 2 φ i ) L 2 ( S δ ) . I_3\eqsim\left(\gamma(\omega_{\boldsymbol{\tau}}+\tilde u)-\gamma(\sigma_iU_i-\sigma_i(U_i)_\mathrm{bc}+\eta_i\widetilde\phi_i)-\sum_{j\neq i}\sigma_j\gamma(U_j) ,\partial_{x_2}\varphi_i\right)_{L^2(\mathcal{S}_\delta)}. I 3 ≂ γ ( ω τ + u ~ ) − γ ( σ i U i − σ i ( U i ) bc + η i ϕ i ) − j = i ∑ σ j γ ( U j ) , ∂ x 2 φ i L 2 ( S δ ) . Since
γ ( ω τ + u ~ ) − γ ( σ i U i − σ i ( U i ) b c + η i ϕ ~ i ) − ∑ j ≠ i σ j γ ( U j ) ≂ ∑ j ≠ i σ j ( γ ′ ( U i ) − 1 ) U j − ∑ j ≠ i σ j ( U j ) b c + ∑ j ≠ i σ j η j ϕ ~ j + ψ ~ , in B i , 2 , \begin{align*}
&\gamma(\omega_{\boldsymbol{\tau}}+\tilde u)-\gamma(\sigma_iU_i-\sigma_i(U_i)_{\mathrm{bc}}+\eta_i\widetilde\phi_i)-\sum_{j\neq i}\sigma_j\gamma(U_j)\\
&\eqsim\sum_{j\neq i}\sigma_j(\gamma'(U_i)-1)U_j-\sum_{j\neq i}\sigma_j(U_j)_{\mathrm{bc}}+\sum_{j\neq i}\sigma_j\eta_j\widetilde\phi_j+\widetilde\psi,\qquad\text{in }B_{i,2},
\end{align*} γ ( ω τ + u ~ ) − γ ( σ i U i − σ i ( U i ) bc + η i ϕ i ) − j = i ∑ σ j γ ( U j ) ≂ j = i ∑ σ j ( γ ′ ( U i ) − 1 ) U j − j = i ∑ σ j ( U j ) bc + j = i ∑ σ j η j ϕ j + ψ , in B i , 2 , by Lemma 2.1 , (5.2) , Assumption assumption:A and (2.20) , we have
∫ S δ σ i ∂ x 2 U i ( γ ′ ( U i ) − 1 ) σ j U j d x = C γ ′ ′ ( 0 ) σ i σ j ∫ S δ ∂ x 2 ( U i 2 ) U j d x = − C 2 σ i σ j sgn ( τ i − τ j ) δ − 1 2 e − ∣ τ i − τ j ∣ δ , \begin{equation*}
\begin{aligned}
\int_{\mathcal{S}_\delta}\sigma_i\partial_{x_2}U_i(\gamma'(U_i)-1)\sigma_jU_j\,dx&=C\gamma''(0)\sigma_i\sigma_j\int_{\mathcal{S}_\delta}\partial_{x_2}(U_i^2)U_j\,dx\\
&=-C_2\sigma_i\sigma_j\operatorname{sgn}(\tau_i-\tau_j)\delta^{-\frac12}e^{-\frac{|\tau_i-\tau_j|}{\delta}},
\end{aligned}
\end{equation*} ∫ S δ σ i ∂ x 2 U i ( γ ′ ( U i ) − 1 ) σ j U j d x = C γ ′′ ( 0 ) σ i σ j ∫ S δ ∂ x 2 ( U i 2 ) U j d x = − C 2 σ i σ j sgn ( τ i − τ j ) δ − 2 1 e − δ ∣ τ i − τ j ∣ , where C C C and C 2 C_2 C 2 are positive constants.
For the remaining terms, by the argument in Corollary (?) , (5.2) and (4.22) , we obtain
∫ S δ ( γ ′ ( U i ) − 1 ) U j ∂ x 2 ( U i ) b c d x + ∫ S δ ( γ ′ ( U i ) − 1 ) U j ∂ x 2 ( η i ϕ ~ i ) d x − ∫ S δ ( U j ) b c ∂ x 2 φ i d x ≲ ∣ ( ∂ x 2 ( U i ) b c , ( γ ′ ( U i ) − 1 ) U j ) L 2 ( S δ ) ∣ + ∫ B i , 2 ∣ U j ∣ ∣ ∂ x 2 ( η i ϕ ~ i ) ∣ d x + ∣ ( ( U j ) b c , ∂ x 2 U i ) L 2 ( S δ ) ∣ ≲ δ 3 4 ∣ log δ ∣ 1 2 e − d m + ∣ τ i − τ j ∣ − 2 R δ + δ − 3 2 e − 2 ( 1 − ∣ τ ∣ ) + ∣ τ i − τ j ∣ − 2 R δ + δ − 5 2 e − 2 − ∣ τ i ∣ − ∣ τ j ∣ δ . \begin{align*}
&\int_{\mathcal{S}_\delta}(\gamma^\prime(U_i)-1)U_j {\partial} _{x_2}(U_i)_\mathrm{bc}\,dx+\int_{\mathcal{S}_\delta}(\gamma^\prime(U_i)-1)U_j {\partial} _{x_2}(\eta_i\widetilde\phi_i)\,dx-\int_{\mathcal{S}_\delta}(U_j)_\mathrm{bc}\partial_{x_2}\varphi_i \,dx\\
&\lesssim\left|\left(\partial_{x_2}(U_i)_\mathrm{bc},(\gamma^\prime(U_i)-1)U_j\right)_{L^2(\mathcal{S}_\delta)}\right|+\int_{B_{i,2}}|U_j|| {\partial} _{x_2}(\eta_i\widetilde\phi_i)|\,dx+\left|\left((U_j)_\mathrm{bc},\partial_{x_2}U_i\right)_{L^2(\mathcal{S}_\delta)}\right|\\
&\lesssim \delta^{\frac34}|\log\delta|^{\frac12}e^{-\frac{d_m+|\tau_i-\tau_j|-2R}{\delta}}+\delta^{-\frac{3}{2}}e^{-\frac{2(1-|\boldsymbol{\tau}|)+|\tau_i-\tau_j|-2R}{\delta}}+\delta^{-\frac{5}{2}}{\mathrm e}^{-\frac{2-|\tau_i|-|\tau_j|}{\delta}}.
\end{align*} ∫ S δ ( γ ′ ( U i ) − 1 ) U j ∂ x 2 ( U i ) bc d x + ∫ S δ ( γ ′ ( U i ) − 1 ) U j ∂ x 2 ( η i ϕ i ) d x − ∫ S δ ( U j ) bc ∂ x 2 φ i d x ≲ ( ∂ x 2 ( U i ) bc , ( γ ′ ( U i ) − 1 ) U j ) L 2 ( S δ ) + ∫ B i , 2 ∣ U j ∣∣ ∂ x 2 ( η i ϕ i ) ∣ d x + ( ( U j ) bc , ∂ x 2 U i ) L 2 ( S δ ) ≲ δ 4 3 ∣ log δ ∣ 2 1 e − δ d m + ∣ τ i − τ j ∣ − 2 R + δ − 2 3 e − δ 2 ( 1 − ∣ τ ∣ ) + ∣ τ i − τ j ∣ − 2 R + δ − 2 5 e − δ 2 − ∣ τ i ∣ − ∣ τ j ∣ . On the other hand, since the supports of η i \eta_i η i and η j \eta_j η j are disjoint whenever i ≠ j i\neq j i = j , by (4.20) we have
( η j ϕ ~ j , ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) = O ( δ 3 4 ∣ log δ ∣ 1 2 e − d m + ∣ τ i − τ j ∣ − 2 R δ + δ − 3 2 e − 2 ( 1 − ∣ τ ∣ ) + ∣ τ i − τ j ∣ − 2 R δ ) . \begin{align*}
\left(\eta_j\widetilde\phi_j,\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}=O\left(\delta^{\frac34}|\log\delta|^{\frac12}e^{-\frac{d_m+|\tau_i-\tau_j|-2R}{\delta}}+\delta^{-\frac{3}{2}}e^{-\frac{2(1-|\boldsymbol{\tau}|)+|\tau_i-\tau_j|-2R}{\delta}}\right).
\end{align*} ( η j ϕ j , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i ) L 2 ( S δ ) = O ( δ 4 3 ∣ log δ ∣ 2 1 e − δ d m + ∣ τ i − τ j ∣ − 2 R + δ − 2 3 e − δ 2 ( 1 − ∣ τ ∣ ) + ∣ τ i − τ j ∣ − 2 R ) . Finally, using the pointwise estimate for ψ ~ \widetilde\psi ψ in Proposition (?) ,
( ψ ~ , ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) = O ( δ − 3 / 2 e − R 2 δ e − d m δ + δ − k − 3 / 2 e − R 2 δ e − 2 ( 1 − ∣ τ ∣ ) δ ) . \begin{align*}
\left(\widetilde\psi,\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}=O\left(\delta^{-3/2}{\mathrm e}^{-\frac{R}{2\delta}}{\mathrm e}^{-\frac {d_m}\delta}+\delta^{-k-3/2}{\mathrm e}^{-\frac{R}{2\delta}}{\mathrm e}^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}\right).
\end{align*} ( ψ , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i ) L 2 ( S δ ) = O ( δ − 3/2 e − 2 δ R e − δ d m + δ − k − 3/2 e − 2 δ R e − δ 2 ( 1 − ∣ τ ∣ ) ) . Therefore,
I 3 = − C 2 ∑ j ≠ i σ i σ j s g n ( τ i − τ j ) δ − 1 / 2 e − ∣ τ i − τ j ∣ δ + O ( e − R 4 δ e − d m δ + e − R 4 δ e − 2 ( 1 − ∣ τ ∣ ) δ ) + O ( ∑ j ≠ i ( e − d m + ∣ τ i − τ j ∣ − 2 R δ + e − 2 ( 1 − ∣ τ ∣ ) + ∣ τ i − τ j ∣ − R δ + δ − 5 2 e − 2 − ∣ τ i ∣ − ∣ τ j ∣ δ ) ) . \begin{equation}
\begin{aligned}
I_3&=-C_2\sum_{j\neq i}\sigma_i\sigma_jsgn(\tau_i-\tau_j)\delta^{-1/2}{\mathrm e}^{-\frac{|\tau_i-\tau_j|}{\delta}}+O\left({\mathrm e}^{-\frac{R}{4\delta}}{\mathrm e}^{-\frac {d_m}\delta}+{\mathrm e}^{-\frac{R}{4\delta}}{\mathrm e}^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}\right)\\
&\quad\,+O\left(\sum_{j\neq i}\left({\mathrm e}^{-\frac{d_m+|\tau_i-\tau_j|-2R}{\delta}}+{\mathrm e}^{-\frac{2(1-|\boldsymbol{\tau}|)+|\tau_i-\tau_j|-R}{\delta}}+\delta^{-\frac{5}{2}}{\mathrm e}^{-\frac{2-|\tau_i|-|\tau_j|}{\delta}}\right)\right).
\end{aligned} \tag*{(5.10)}
\end{equation} I 3 = − C 2 j = i ∑ σ i σ j s g n ( τ i − τ j ) δ − 1/2 e − δ ∣ τ i − τ j ∣ + O ( e − 4 δ R e − δ d m + e − 4 δ R e − δ 2 ( 1 − ∣ τ ∣ ) ) + O j = i ∑ ( e − δ d m + ∣ τ i − τ j ∣ − 2 R + e − δ 2 ( 1 − ∣ τ ∣ ) + ∣ τ i − τ j ∣ − R + δ − 2 5 e − δ 2 − ∣ τ i ∣ − ∣ τ j ∣ ) . (5.10) For I 4 I_4 I 4 , analogously to I 3 I_3 I 3 ,
I 4 = ( ∑ j ≠ i σ j ( U j ) b c − ∑ j ≠ i Δ ( η j ϕ ~ j ) − Δ ψ ~ , ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) ≲ ∑ j ≠ i ∫ S δ ∣ ( U j ) b c ∣ ∣ ∂ x 2 U i ∣ d x − ( ψ ~ − f o u t [ ϕ ~ 1 , … , ϕ ~ n , ψ ~ , Γ ~ s ] , ∂ x 2 φ i 1 + Γ ~ ′ ( δ ⋅ ) ) L 2 ( S δ ) + ∑ j ≠ i ( δ 3 4 ∣ log δ ∣ 1 2 e − d m + ∣ τ i − τ j ∣ − 2 R δ + δ − 3 2 e − 2 ( 1 − ∣ τ ∣ ) + ∣ τ i − τ j ∣ − 2 R δ ) ≲ ∑ j ≠ i ( δ − 1 2 e − 2 − τ i − τ j δ + δ − 1 2 e − 2 + τ i + τ j δ ) + δ − 1 / 2 e − R 2 δ e − d m δ + δ − k − 1 / 2 e − R 2 δ e − 2 ( 1 − ∣ τ ∣ ) δ + ∑ j ≠ i ( δ 3 4 ∣ log δ ∣ 1 2 e − d m + ∣ τ i − τ j ∣ − 2 R δ + δ − 3 2 e − 2 ( 1 − ∣ τ ∣ ) + ∣ τ i − τ j ∣ − 2 R δ ) , \begin{equation}
\begin{aligned}
I_4&=\left(\sum_{j\neq i}\sigma_j(U_j)_\mathrm{bc}-\sum_{j\neq i}\Delta(\eta_j\widetilde\phi_j)-\Delta\widetilde\psi,\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}\\
&\lesssim\sum_{j\neq i}\int_{\mathcal{S}_\delta}|(U_j)_\mathrm{bc}||\partial_{x_2}U_i| \,dx-\left(\widetilde\psi-f_{out}[\widetilde\phi_1,\dots,\widetilde\phi_n,\widetilde\psi,\widetilde\Gamma_{\mathrm{s}}],\frac{\partial_{x_2}\varphi_i}{1+\widetilde{\Gamma}'(\delta\,\cdot)}\right)_{L^2(\mathcal{S}_\delta)}\\
&\quad\,+\sum_{j\neq i}\left(\delta^{\frac34}|\log\delta|^{\frac12}e^{-\frac{d_m+|\tau_i-\tau_j|-2R}{\delta}}+\delta^{-\frac{3}{2}}e^{-\frac{2(1-|\boldsymbol{\tau}|)+|\tau_i-\tau_j|-2R}{\delta}}\right)\\
&\lesssim\sum_{j\neq i}\left(\delta^{-\frac{1}{2}}e^{-\frac {2-\tau_i-\tau_j}\delta}+\delta^{-\frac{1}{2}}e^{-\frac {2+\tau_i+\tau_j}\delta}\right)+\delta^{-1/2}{\mathrm e}^{-\frac{R}{2\delta}}{\mathrm e}^{-\frac {d_m}\delta}+\delta^{-k-1/2}{\mathrm e}^{-\frac{R}{2\delta}}{\mathrm e}^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}\\
&\quad\,+\sum_{j\neq i}\left(\delta^{\frac34}|\log\delta|^{\frac12}e^{-\frac{d_m+|\tau_i-\tau_j|-2R}{\delta}}+\delta^{-\frac{3}{2}}e^{-\frac{2(1-|\boldsymbol{\tau}|)+|\tau_i-\tau_j|-2R}{\delta}}\right),
\end{aligned} \tag*{(5.11)}
\end{equation} I 4 = j = i ∑ σ j ( U j ) bc − j = i ∑ Δ ( η j ϕ j ) − Δ ψ , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i L 2 ( S δ ) ≲ j = i ∑ ∫ S δ ∣ ( U j ) bc ∣∣ ∂ x 2 U i ∣ d x − ( ψ − f o u t [ ϕ 1 , … , ϕ n , ψ , Γ s ] , 1 + Γ ′ ( δ ⋅ ) ∂ x 2 φ i ) L 2 ( S δ ) + j = i ∑ ( δ 4 3 ∣ log δ ∣ 2 1 e − δ d m + ∣ τ i − τ j ∣ − 2 R + δ − 2 3 e − δ 2 ( 1 − ∣ τ ∣ ) + ∣ τ i − τ j ∣ − 2 R ) ≲ j = i ∑ ( δ − 2 1 e − δ 2 − τ i − τ j + δ − 2 1 e − δ 2 + τ i + τ j ) + δ − 1/2 e − 2 δ R e − δ d m + δ − k − 1/2 e − 2 δ R e − δ 2 ( 1 − ∣ τ ∣ ) + j = i ∑ ( δ 4 3 ∣ log δ ∣ 2 1 e − δ d m + ∣ τ i − τ j ∣ − 2 R + δ − 2 3 e − δ 2 ( 1 − ∣ τ ∣ ) + ∣ τ i − τ j ∣ − 2 R ) , (5.11) here we used the pointwise estimate for f o u t f_{out} f o u t in Proposition (?) .
Combining (5.8) , (5.9) , (5.10) and (5.11) , we obtain
b i ( τ ) = C 1 δ 1 2 ( e − 2 ( 1 + τ i ) δ − e − 2 ( 1 − τ i ) δ ) − C 2 ∑ j ≠ i σ i σ j s g n ( τ i − τ j ) δ − 1 / 2 e − ∣ τ i − τ j ∣ δ + O ( ∑ j ≠ i ( e − d m + ∣ τ i − τ j ∣ − R δ + e − 2 ( 1 − ∣ τ ∣ ) + ∣ τ i − τ j ∣ − R δ ) + δ − 1 2 e − 3 ( 1 − ∣ τ ∣ ) δ + e − d m + 2 ( 1 − ∣ τ ∣ ) δ ) + O ( e − R 4 δ e − d m δ + e − R 4 δ e − 2 ( 1 − ∣ τ ∣ ) δ ) . \begin{equation*}
\begin{aligned}
b_i(\boldsymbol{\tau})=&C_1\delta^{\frac 12}\left({\mathrm e}^{-\frac {2(1 +\tau_i)}\delta}-{\mathrm e}^{-\frac {2(1 - \tau_i)}\delta}\right)-C_2\sum_{j\neq i}\sigma_i\sigma_jsgn(\tau_i-\tau_j)\delta^{-1/2}{\mathrm e}^{-\frac{|\tau_i-\tau_j|}{\delta}}\\
&+O\left(\sum_{j\neq i}\left({\mathrm e}^{-\frac{d_m+|\tau_i-\tau_j|-R}{\delta}}+{\mathrm e}^{-\frac{2(1-|\boldsymbol{\tau}|)+|\tau_i-\tau_j|-R}{\delta}}\right)+\delta^{-\frac{1}{2}}{\mathrm e}^{-\frac{3(1-|\boldsymbol{\tau}|)}\delta}+{\mathrm e}^{-\frac{d_m+2(1-|\boldsymbol{\tau}|)}\delta}\right)\\
&+O\left({\mathrm e}^{-\frac{R}{4\delta}}{\mathrm e}^{-\frac {d_m}\delta}+{\mathrm e}^{-\frac{R}{4\delta}}{\mathrm e}^{-\frac {2(1-|\boldsymbol{\tau}|)}\delta}\right).
\end{aligned}
\end{equation*} b i ( τ ) = C 1 δ 2 1 ( e − δ 2 ( 1 + τ i ) − e − δ 2 ( 1 − τ i ) ) − C 2 j = i ∑ σ i σ j s g n ( τ i − τ j ) δ − 1/2 e − δ ∣ τ i − τ j ∣ + O j = i ∑ ( e − δ d m + ∣ τ i − τ j ∣ − R + e − δ 2 ( 1 − ∣ τ ∣ ) + ∣ τ i − τ j ∣ − R ) + δ − 2 1 e − δ 3 ( 1 − ∣ τ ∣ ) + e − δ d m + 2 ( 1 − ∣ τ ∣ ) + O ( e − 4 δ R e − δ d m + e − 4 δ R e − δ 2 ( 1 − ∣ τ ∣ ) ) . For simplicity, we assume that τ ∈ P ε \boldsymbol{\tau}\in P_{\varepsilon} τ ∈ P ε (here ε < 1 n \varepsilon<\frac{1}{n} ε < n 1 ) satisfing
− 1 < τ n < τ n − 1 < ⋯ < τ 2 < τ 1 < 1 -1<\tau_n<\tau_{n-1}<\cdots<\tau_{2}<\tau_{1}<1 − 1 < τ n < τ n − 1 < ⋯ < τ 2 < τ 1 < 1 and
2 ( 1 − τ 1 ) = 2 ( τ n + 1 ) + o ( 1 ) = τ k − τ k + 1 + o ( 1 ) , for k = 1 , … , n − 1 , as δ → 0. \begin{equation}
2(1-\tau_1)=2(\tau_n+1)+o(1)=\tau_{k}-\tau_{k+1}+o(1), \quad\text{for }k=1,\dots,n-1,\text{ as }\delta\to 0. \tag*{(5.12)}
\end{equation} 2 ( 1 − τ 1 ) = 2 ( τ n + 1 ) + o ( 1 ) = τ k − τ k + 1 + o ( 1 ) , for k = 1 , … , n − 1 , as δ → 0. (5.12) Indeed, 2 ( τ n + 1 ) ≈ 2 ( 1 − τ 1 ) ≈ τ k − τ k + 1 = 2 n + o ( 1 ) 2(\tau_n+1)\approx2(1-\tau_1)\approx\tau_{k}-\tau_{k+1}=\frac{2}{n}+o(1) 2 ( τ n + 1 ) ≈ 2 ( 1 − τ 1 ) ≈ τ k − τ k + 1 = n 2 + o ( 1 ) for all k = 1 , … , n − 1 k=1,\dots,n-1 k = 1 , … , n − 1 , as δ → 0 \delta\to 0 δ → 0 . Then we have
b 1 ( τ ) = 0 ⟺ C ~ 1 δ − 1 e 2 ( 1 − τ 1 ) − ( τ 1 − τ 2 ) δ − 1 + O ( δ − 1 e − 1 n δ + e − R 2 δ ) = 0 , b 2 ( τ ) = 0 ⟺ C ~ 2 e ( τ 1 − τ 2 ) − ( τ 2 − τ 3 ) δ − 1 + O ( e − 1 n δ + e − R 2 δ ) = 0 , ⋮ b j ( τ ) = 0 ⟺ C ~ j e ( τ j − 1 − τ j ) − ( τ j − τ j + 1 ) δ − 1 + O ( e − 1 n δ + e − R 2 δ ) = 0 , ⋮ b n − 1 ( τ ) = 0 ⟺ C ~ n − 1 e ( τ n − 2 − τ n − 1 ) − ( τ n − 1 − τ n ) δ − 1 + O ( e − 1 n δ + e − R 2 δ ) = 0 , b n ( τ ) = 0 ⟺ C ~ n δ − 1 e 2 ( 1 + τ n ) − ( τ n − 1 − τ n ) δ − 1 + O ( δ − 1 e − 1 n δ + e − R 2 δ ) = 0 , \begin{align*}
b_1(\boldsymbol{\tau})=0&\;\iff\widetilde C_1\delta^{-1}e^{\frac{2(1-\tau_1)-(\tau_1-\tau_2)}{\delta}}-1+O\left(\delta^{-1}{\mathrm e}^{-\frac{1}{n\delta}}+{\mathrm e}^{-\frac{R}{2\delta}}\right)=0,\\
b_2(\boldsymbol{\tau})=0&\;\iff\widetilde C_2e^{\frac{(\tau_1-\tau_2)-(\tau_2-\tau_3)}{\delta}}-1+O\left(e^{-\frac{1}{n\delta}}+{\mathrm e}^{-\frac{R}{2\delta}}\right)=0,\\
&\;\;\,\,\,\,\vdots\\
b_j(\boldsymbol{\tau})=0&\;\iff\widetilde C_je^{\frac{(\tau_{j-1}-\tau_{j})-(\tau_{j}-\tau_{j+1})}{\delta}}-1+O\left(e^{-\frac{1}{n\delta}}+{\mathrm e}^{-\frac{R}{2\delta}}\right)=0,\\
&\;\;\,\,\,\,\vdots\\
b_{n-1}(\boldsymbol{\tau})=0&\;\iff\widetilde C_{n-1}e^{\frac{(\tau_{n-2}-\tau_{n-1})-(\tau_{n-1}-\tau_{n})}{\delta}}-1+O\left(e^{-\frac{1}{n\delta}}+{\mathrm e}^{-\frac{R}{2\delta}}\right)=0,\\
b_n(\boldsymbol{\tau})=0&\;\iff\widetilde C_n\delta^{-1}e^{\frac{2(1+\tau_n)-(\tau_{n-1}-\tau_{n})}{\delta}}-1+O\left(\delta^{-1}{\mathrm e}^{-\frac{1}{n\delta}}+{\mathrm e}^{-\frac{R}{2\delta}}\right)=0,
\end{align*} b 1 ( τ ) = 0 b 2 ( τ ) = 0 b j ( τ ) = 0 b n − 1 ( τ ) = 0 b n ( τ ) = 0 ⟺ C 1 δ − 1 e δ 2 ( 1 − τ 1 ) − ( τ 1 − τ 2 ) − 1 + O ( δ − 1 e − n δ 1 + e − 2 δ R ) = 0 , ⟺ C 2 e δ ( τ 1 − τ 2 ) − ( τ 2 − τ 3 ) − 1 + O ( e − n δ 1 + e − 2 δ R ) = 0 , ⋮ ⟺ C j e δ ( τ j − 1 − τ j ) − ( τ j − τ j + 1 ) − 1 + O ( e − n δ 1 + e − 2 δ R ) = 0 , ⋮ ⟺ C n − 1 e δ ( τ n − 2 − τ n − 1 ) − ( τ n − 1 − τ n ) − 1 + O ( e − n δ 1 + e − 2 δ R ) = 0 , ⟺ C n δ − 1 e δ 2 ( 1 + τ n ) − ( τ n − 1 − τ n ) − 1 + O ( δ − 1 e − n δ 1 + e − 2 δ R ) = 0 , where C ~ i > 0 \widetilde C_i>0 C i > 0 are bounded for all i = 1 , … , n i=1,\dots,n i = 1 , … , n , and we have used the fact that σ i σ j = − 1 \sigma_i\sigma_j=-1 σ i σ j = − 1 for adjacent indices i i i and j j j .
We now prove that for small δ \delta δ , there exists τ ∈ P ε \boldsymbol{\tau}\in P_\varepsilon τ ∈ P ε (ε < 1 n \varepsilon<\frac{1}{n} ε < n 1 ), which is the solution to the following system of equations
{ τ 2 − 3 τ 1 = − 2 + δ log δ − δ log C ~ 1 + O ( δ − 1 e − 1 n δ + e − R 2 δ ) τ 1 − 2 τ 2 + τ 3 = − δ log C ~ 2 + O ( e − 1 n δ + e − R 2 δ ) ⋯ τ j − 1 − 2 τ j + τ j + 1 = − δ log C ~ j + O ( e − 1 n δ + e − R 2 δ ) ⋯ τ n − 2 − 2 τ n − 1 + τ n = − δ log C ~ n − 1 + O ( e − 1 n δ + e − R 2 δ ) 3 τ n − τ n − 1 = − 2 + δ log δ − δ log C ~ n + O ( δ − 1 e − 1 n δ + e − R 2 δ ) . \begin{equation}
\begin{cases}
\tau_2-3\tau_1&=-2+\delta\log\delta-\delta\log\widetilde C_1+O\left(\delta^{-1}{\mathrm e}^{-\frac{1}{n\delta}}+{\mathrm e}^{-\frac{R}{2\delta}}\right)\\
\tau_1-2\tau_2+\tau_3&=-\delta\log\widetilde C_2+O\left(e^{-\frac{1}{n\delta}}+{\mathrm e}^{-\frac{R}{2\delta}}\right)\\
&\cdots\\
\tau_{j-1}-2\tau_{j}+\tau_{j+1}&=-\delta\log\widetilde C_j+O\left(e^{-\frac{1}{n\delta}}+{\mathrm e}^{-\frac{R}{2\delta}}\right)\\
&\cdots\\
\tau_{n-2}-2\tau_{n-1}+\tau_{n}&=-\delta\log\widetilde C_{n-1}+O\left(e^{-\frac{1}{n\delta}}+{\mathrm e}^{-\frac{R}{2\delta}}\right)\\
3\tau_n-\tau_{n-1}&=-2+\delta\log\delta-\delta\log\widetilde C_n+O\left(\delta^{-1}{\mathrm e}^{-\frac{1}{n\delta}}+{\mathrm e}^{-\frac{R}{2\delta}}\right).
\end{cases} \tag*{(5.13)}
\end{equation} ⎩ ⎨ ⎧ τ 2 − 3 τ 1 τ 1 − 2 τ 2 + τ 3 τ j − 1 − 2 τ j + τ j + 1 τ n − 2 − 2 τ n − 1 + τ n 3 τ n − τ n − 1 = − 2 + δ log δ − δ log C 1 + O ( δ − 1 e − n δ 1 + e − 2 δ R ) = − δ log C 2 + O ( e − n δ 1 + e − 2 δ R ) ⋯ = − δ log C j + O ( e − n δ 1 + e − 2 δ R ) ⋯ = − δ log C n − 1 + O ( e − n δ 1 + e − 2 δ R ) = − 2 + δ log δ − δ log C n + O ( δ − 1 e − n δ 1 + e − 2 δ R ) . (5.13) To prove the solvability of the above system of equations, we consider we consider a matrix A n A_n A n of the form
A n = ( − 3 1 − 1 3 ) for n = 2 , A n = ( − 3 1 0 ⋯ 0 1 − 2 1 ⋱ 0 ⋮ ⋱ ⋱ ⋱ 0 0 ⋯ 1 − 2 1 0 ⋯ 0 − 1 3 ) for n ≥ 3. A_n=\begin{pmatrix}
-3&1\\
-1&3
\end{pmatrix}\quad\text{for }n=2,\qquad A_n=\begin{pmatrix}
-3 & 1 & 0 & \cdots & 0\\
1 & -2 & 1 & \ddots & 0\\
\vdots & \ddots & \ddots & \ddots & 0\\
0 & \cdots & 1 & -2 & 1\\
0 & \cdots & 0 & -1 & 3
\end{pmatrix}\quad\text{for }n\geq3. A n = ( − 3 − 1 1 3 ) for n = 2 , A n = − 3 1 ⋮ 0 0 1 − 2 ⋱ ⋯ ⋯ 0 1 ⋱ 1 0 ⋯ ⋱ ⋱ − 2 − 1 0 0 0 1 3 for n ≥ 3. Since A n A_n A n is an irreducible, weakly diagonally dominant matrix and is strictly diagonally dominant in at least one row, it follows that A n A_n A n is nonsingular. Then system (5.13) is solvable, and a direct calculation shows that solution τ ~ = ( τ ~ 1 , … , τ ~ n ) \tilde{\boldsymbol{\tau}}=(\tilde\tau_1,\dots,\tilde\tau_n) τ ~ = ( τ ~ 1 , … , τ ~ n ) satisfies
{ τ ~ 1 = n − 1 n + O ( δ log δ ) τ ~ 2 = n − 3 n + O ( δ log δ ) ⋯ τ ~ j = n − 2 j + 1 n + O ( δ log δ ) ⋯ τ ~ n − 1 = 3 − n n + O ( δ log δ ) τ ~ n = 1 − n n + O ( δ log δ ) as δ → 0. \begin{equation}
\begin{cases}
\tilde\tau_1=\frac{n-1}{n}+O\left(\delta\log\delta\right)\\
\tilde\tau_2=\frac{n-3}{n}+O\left(\delta\log\delta\right)\\
\quad\cdots\\
\tilde\tau_{j}=\frac{n-2j+1}{n}+O\left(\delta\log\delta\right)\\
\quad\cdots\\
\tilde\tau_{n-1}=\frac{3-n}{n}+O\left(\delta\log\delta\right)\\
\tilde\tau_n=\frac{1-n}{n}+O\left(\delta\log\delta\right)
\end{cases}\quad\text{as }\delta\to0. \tag*{(5.14)}
\end{equation} ⎩ ⎨ ⎧ τ ~ 1 = n n − 1 + O ( δ log δ ) τ ~ 2 = n n − 3 + O ( δ log δ ) ⋯ τ ~ j = n n − 2 j + 1 + O ( δ log δ ) ⋯ τ ~ n − 1 = n 3 − n + O ( δ log δ ) τ ~ n = n 1 − n + O ( δ log δ ) as δ → 0. (5.14) Then ( τ ~ , u ~ ( τ ~ ) , Γ ~ s ( τ ~ ) ) (\tilde{\boldsymbol{\tau}}, \tilde u(\tilde{\boldsymbol{\tau}}), \widetilde{\Gamma}_{\mathrm{s}}(\tilde{\boldsymbol{\tau}})) ( τ ~ , u ~ ( τ ~ ) , Γ s ( τ ~ )) is a solution to (?) , thus corresponds to a solution to the stationary capillary-gravity wave problem.
The stream function is given by
Ψ = ( ω τ ~ + u ~ ( τ ~ ) ) ∘ ( 1 δ ( i d + Γ ~ ( τ ~ ) ) − 1 ) , \begin{equation*}
\Psi = \left(\omega_{\tilde{\boldsymbol{\tau}}}+ \tilde u (\tilde{\boldsymbol{\tau}})\right) \circ \left( \tfrac 1\delta \left( id + \widetilde \Gamma (\tilde{\boldsymbol{\tau}}) \right)^{-1}\right),
\end{equation*} Ψ = ( ω τ ~ + u ~ ( τ ~ ) ) ∘ ( δ 1 ( i d + Γ ( τ ~ ) ) − 1 ) , defined on
Ω = ( i d + Γ ~ ( τ ~ ) ) ( { ( x 1 , x 2 ) ∈ R 2 ∣ ∣ x 2 ∣ < 1 } ) . \Omega = \left(id + \widetilde \Gamma (\tilde{\boldsymbol{\tau}}) \right) \left(\{(x_1,x_2)\in {\mathbb{R}}^2\;|\;|x_2|< 1\}\right). Ω = ( i d + Γ ( τ ~ ) ) ( {( x 1 , x 2 ) ∈ R 2 ∣ ∣ x 2 ∣ < 1 } ) . For simplicity of presentation, let
v ~ ( ⋅ / δ ) : = Ψ − ω τ ~ ( ⋅ / δ ) . \widetilde v(\cdot/\delta):=\Psi-\omega_{\tilde{\boldsymbol{\tau}}}(\cdot/\delta). v ( ⋅ / δ ) := Ψ − ω τ ~ ( ⋅ / δ ) . And from the construction in § 2.1 , we have
ζ δ = Γ ~ 2 ∘ ( i d + Γ ~ 1 ( ⋅ , 1 ) ) − 1 . \zeta_\delta=\widetilde\Gamma_2\circ\left(id+\widetilde\Gamma_1(\cdot,1)\right)^{-1}. ζ δ = Γ 2 ∘ ( i d + Γ 1 ( ⋅ , 1 ) ) − 1 . Then by (4.20) -(4.22) and scaling property,
∥ v ~ ( ⋅ / δ ) ∥ H 2 ( Ω ) ≲ δ − 1 ∥ v ~ ∥ H 2 ( Ω / δ ) ≲ δ − 1 ( ∑ i = 1 n ∥ ϕ ~ i ∥ H 2 ( S δ ) + ∥ ψ ~ ∥ H 2 ( S δ ) + ∥ Γ ~ s ∥ H 2 ( R ) ) ≲ δ β exp ( − 2 n δ ) , \begin{align*}
&\left\|\widetilde v(\cdot/\delta)\right\|_{H^2(\Omega)}\lesssim\delta^{-1}\left\|\widetilde v\right\|_{H^2(\Omega/\delta)}\\
&\lesssim\delta^{-1}\left(\sum_{i=1}^n\left\|\widetilde\phi_i\right\|_{H^2(\mathcal{S}_\delta)}+\left\|\widetilde{\psi}\right\|_{H^2(\mathcal{S}_\delta)}+\left\|\widetilde{\Gamma}_{\mathrm{s}}\right\|_{H^2( {\mathbb{R}})}\right)\lesssim\delta^{\beta}\exp\left(-\frac{2}{n\delta}\right),
\end{align*} ∥ v ( ⋅ / δ ) ∥ H 2 ( Ω ) ≲ δ − 1 ∥ v ∥ H 2 ( Ω/ δ ) ≲ δ − 1 ( i = 1 ∑ n ϕ i H 2 ( S δ ) + ψ H 2 ( S δ ) + Γ s H 2 ( R ) ) ≲ δ β exp ( − n δ 2 ) , ∥ ζ δ ∥ H 2 ( R ) ≲ ∥ Γ ~ s ∥ H 2 ( R ) ≲ δ β ∣ log δ ∣ exp ( − 2 n δ ) , \begin{align*}
\|\zeta_\delta\|_{H^2(\mathbb R)}\lesssim\left\|\widetilde{\Gamma}_{\mathrm{s}}\right\|_{H^2( {\mathbb{R}})}\lesssim\delta^{\beta}|\log\delta|\exp\left(-\frac{2}{n\delta}\right),
\end{align*} ∥ ζ δ ∥ H 2 ( R ) ≲ Γ s H 2 ( R ) ≲ δ β ∣ log δ ∣ exp ( − n δ 2 ) , where constant β \beta β is determined by (5.14) . This completes the proof of the main theorem. ◻
Acknowledgements
Y. Wang is partially supported by NSF of China (No. 12471110). Y. Zheng is supported by NSF of China (No. 12671246).