Introduction
Dirac operators with singular interactions supported on curves in R 2 \mathbb{R}^{2} R 2 or surfaces in R 3 \mathbb{R}^{3} R 3 have received a lot of attention in the mathematical community in recent years. Such operators in n n n dimensions, with n ∈ { 2 , 3 } n \in\{2,3\} n ∈ { 2 , 3 } , are formally given by the first order matrix differential expression acting on vector functions f f f as
f ↦ − i ∑ k = 1 n α k ∂ k f + m β f + ( η I N + τ β ) δ Σ f , (1) f \mapsto-\mathrm{i}\sum_{k=1}^{n}\alpha_k \partial_k f + m\beta f + (\eta I_N+\tau\beta)\delta_{\Sigma}f, \tag*{(1)} f ↦ − i k = 1 ∑ n α k ∂ k f + m β f + ( η I N + τ β ) δ Σ f , (1)
where m ∈ R m \in\mathbb{R} m ∈ R is a mass parameter and α k , β \alpha_k,\beta α k , β denote the Pauli spin matrices if n = 2 n=2 n = 2 and the Dirac matrices if n = 3 n=3 n = 3 , see § 3.1 . Furthermore, the singular interaction term ( η I N + τ β ) δ Σ (\eta I_N+\tau\beta)\delta_{\Sigma} ( η I N + τ β ) δ Σ consists of an electrostatic interaction with strength η ∈ R \eta\in\mathbb{R} η ∈ R , a Lorentz scalar interaction with strength τ ∈ R \tau\in\mathbb{R} τ ∈ R , and δ Σ \delta_{\Sigma} δ Σ stands for the δ \delta δ -distribution supported on a compact curve or surface Σ ⊂ R n \Sigma\subset\mathbb{R}^{n} Σ ⊂ R n , n ∈ { 2 , 3 } n \in\{2,3\} n ∈ { 2 , 3 } . In order to rigorously interpret the above expression as a self-adjoint operator D η , τ D^{\eta,\tau} D η , τ in L 2 ( R n , C N ) L^{2}(\mathbb{R}^{n},\mathbb{C}^{N}) L 2 ( R n , C N ) (with N = 2 N=2 N = 2 if n = 2 n=2 n = 2 and N = 4 N=4 N = 4 if n = 3 n=3 n = 3 ), the formal singular interaction term is usually treated as a transmission condition along Σ \Sigma Σ . The first most obvious mathematical problem is to specify an appropriate operator domain ensuring the self-adjointness, which is followed by a qualitative and quantitative spectral analysis of the operator. We refer the interested reader to the review paper [12 ] for an overview on general results and methods in the treatment of Dirac operators with singular interactions, as well as for further references, and the references listed below.
In the following we will focus on a particularly interesting spectral peculiarity of Dirac operators of the above type, namely, the emergence of a new portion of essential spectrum; cf. [8 , 10 , 12 , 22 , 36 ] and, in particular, [13 , 20 ] . In the absence of interactions (η = τ = 0 \eta=\tau=0 η = τ = 0 ) and for m ≠ 0 m\ne0 m = 0 , the unperturbed operator D D D is just the free Dirac operator in R n \mathbb{R}^{n} R n with the spectral gap ( − ∣ m ∣ , ∣ m ∣ ) (-|m|,|m|) ( − ∣ m ∣ , ∣ m ∣ ) , and it was seen early that adding a singular perturbation of the above type with η 2 − τ 2 ≠ 4 \eta^{2}-\tau^{2}\ne4 η 2 − τ 2 = 4 (the so-called non-critical coupling) only produces finitely many eigenvalues in the gap, while the essential spectrum remains unchanged (which corresponds to the usual intuition that compactly supported perturbations do not affect the essential spectrum) and one has the expected Sobolev regularity of functions in the operator domain; we refer to [2 , 3 , 6 , 7 , 16 , 21 , 24 , 28 , 35 , 40 , 41 , 42 ] for a small list of related contributions. However, the sitation changes for η 2 − τ 2 = 4 \eta^{2}-\tau^{2}=4 η 2 − τ 2 = 4 : this parameter combination case is usually referred to as the critical couplings. It was first observed for n = 2 n=2 n = 2 in [13 ] that if Σ \Sigma Σ is a smooth closed loop, then
spec e s s D η , τ = ( − ∞ , − ∣ m ∣ ] ∪ { − τ η m } ∪ [ ∣ m ∣ , ∞ ) , (2) \operatorname{spec}_{\mathrm{ess}} D^{\eta,\tau}=(-\infty,-|m|]\cup\left\{-\frac{\tau}{\eta}m\right\}\cup[|m|,\infty), \tag*{(2)} spec ess D η , τ = ( − ∞ , − ∣ m ∣ ] ∪ { − η τ m } ∪ [ ∣ m ∣ , ∞ ) , (2)
i.e. a new point of essential spectrum appears. Later, it turned out that the effect of critical couplings in three dimensions is even more striking. More precisely, if Σ ⊂ R 3 \Sigma\subset\mathbb{R}^{3} Σ ⊂ R 3 is a compact smooth surface and η 2 − τ 2 = 4 \eta^{2}-\tau^{2}=4 η 2 − τ 2 = 4 , then it was observed in [20 ] that
spec e s s D η , τ = ( − ∞ , − ∣ m ∣ ] ∪ [ − τ η m − A Σ 2 ∣ η ∣ , − τ η m + A Σ 2 ∣ η ∣ ] ∪ [ ∣ m ∣ , ∞ ) , (3) \operatorname{spec}_{\mathrm{ess}} D^{\eta,\tau}=(-\infty,-|m|]\cup\left[-\frac{\tau}{\eta}m-\frac{\mathcal{A}_{\Sigma}}{2|\eta|},-\frac{\tau}{\eta}m+\frac{\mathcal{A}_{\Sigma}}{2|\eta|}\right]\cup[|m|,\infty), \tag*{(3)} spec ess D η , τ = ( − ∞ , − ∣ m ∣ ] ∪ [ − η τ m − 2∣ η ∣ A Σ , − η τ m + 2∣ η ∣ A Σ ] ∪ [ ∣ m ∣ , ∞ ) , (3)
where A Σ \mathcal{A}_{\Sigma} A Σ is a geometric parameter given by
A Σ : = max x ∈ Σ ∣ κ 1 ( x ) − κ 2 ( x ) ∣ , \mathcal{A}_{\Sigma}:=\max_{x\in\Sigma}|\kappa_{1}(x)-\kappa_{2}(x)|, A Σ := x ∈ Σ max ∣ κ 1 ( x ) − κ 2 ( x ) ∣ ,
with κ j ( x ) \kappa_j(x) κ j ( x ) being the principal curvatures of Σ \Sigma Σ at x x x . In particular, if Σ \Sigma Σ is not a union of spheres, then A Σ > 0 A_\Sigma> 0 A Σ > 0 , and one obtains an additional non-trivial interval in the essential spectrum. Moreover, in all dimensions a lack of Sobolev regularity in the operator domain was observed in the critical case.
The present paper is inspired by an enthusiastic discussion of one of the authors with the world renowned physicists Yakir Aharonov in Cetraro, Italy, in June 2024, who became very interested in the spectral effect for Dirac operators described above in (2) and (3) . However, for a physical interpretation it is necessary, according to Aharonov, to better understand the origin of the new essential spectrum and, in particular, it should be clarified if such an effect also appears in “finite boxes”. More precisely, assume that Ω ⊂ R n \Omega\subset\mathbb{R}^n Ω ⊂ R n , n ∈ { 2 , 3 } n \in\{2,3\} n ∈ { 2 , 3 } , is a bounded domain and let Σ ⊂ Ω \Sigma\subset\Omega Σ ⊂ Ω be a smooth compact hypersurface (curve for n = 2 n=2 n = 2 and surface for n = 3 n=3 n = 3 ). Impose a self-adjoint boundary condition on ∂ Ω \partial\Omega ∂ Ω and model the singular interaction term ( η I N + τ β ) δ Σ (\eta I_N+\tau\beta)\delta_\Sigma ( η I N + τ β ) δ Σ by the same transmission condition as in R n \mathbb{R}^n R n . The question is then whether the new essential spectrum also appears for the perturbed Dirac operator at critical couplings. The present paper addresses this question and provides an affirmative answer.
Let us describe our results and methods in more detail. Let Ω ⊂ R n \Omega\subset\mathbb{R}^n Ω ⊂ R n be a bounded Lipschitz domain, n ∈ { 2 , 3 } n \in\{2,3\} n ∈ { 2 , 3 } , and Σ ⊂ Ω \Sigma\subset\Omega Σ ⊂ Ω be a compact smooth hypersurface with finitely many connected components Σ 1 , … , Σ P \Sigma_1,\ldots,\Sigma_P Σ 1 , … , Σ P ; cf. § 4.1 . Choose η , τ ∈ R P \eta,\tau\in\mathbb{R}^P η , τ ∈ R P , η = ( η 1 , … , η P ) \eta=(\eta_1,\ldots,\eta_P) η = ( η 1 , … , η P ) , τ = ( τ 1 , … , τ P ) \tau=(\tau_1,\ldots,\tau_P) τ = ( τ 1 , … , τ P ) , and consider the Dirac operator A η , τ A^{\eta,\tau} A η , τ in Ω \Omega Ω with some local self-adjoint boundary conditions on ∂ Ω \partial\Omega ∂ Ω (see § 3.1 and, e.g., [4 , 5 , 11 , 15 , 29 , 36 ] ) as well as the transmission conditions
i ( α ⋅ ν Σ ) ( f p + − f p − ) + 1 2 ( η p I N + τ p β ) ( f p + + f p − ) = 0 on each Σ p , p ∈ { 1 , … , P } , \mathrm{i}(\alpha\cdot\nu_\Sigma)(f_{p+}-f_{p-})+\frac{1}{2}(\eta_p I_N+\tau_p\beta)(f_{p+}+f_{p-})=0\text{ on each }\Sigma_p,\quad p\in\{1,\ldots,P\}, i ( α ⋅ ν Σ ) ( f p + − f p − ) + 2 1 ( η p I N + τ p β ) ( f p + + f p − ) = 0 on each Σ p , p ∈ { 1 , … , P } ,
modeling the distributional perturbation term in (1) ; here f p ± f_{p\pm} f p ± are one-sided traces on the two sides of Σ p \Sigma_p Σ p (see § 4.1 ) and η p , τ p ∈ R \eta_p,\tau_p\in\mathbb{R} η p , τ p ∈ R , p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } , are viewed as coupling parameters. Due to some technical reasons we assume ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p|\ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ for all p p p (we comment on this case in Remark 4.20). As an important preparation, we show that the operator A η , τ A^{\eta,\tau} A η , τ is self-adjoint in L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^N) L 2 ( Ω , C N ) on a suitable domain (Theorem 4.8 ). This requires a preliminary study of the integral kernel of the resolvent of the unperturbed self-adjoint Dirac operator in L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^N) L 2 ( Ω , C N ) , which is presented in § 3.2 . In Section 4 we turn to our main point of interest, the spectral properties of A η , τ A^{\eta,\tau} A η , τ . For the non-critical case η p 2 − τ p 2 ≠ 4 \eta_p^2-\tau_p^2\ne4 η p 2 − τ p 2 = 4 we show that the functions in the operator domain are H 1 H^1 H 1 -regular near Σ p \Sigma_p Σ p (Corollary 4.10). If η p 2 − τ p 2 ≠ 4 \eta_p^2-\tau_p^2\ne4 η p 2 − τ p 2 = 4 for all p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } , then the functions in the operator domain are indeed H 1 H^1 H 1 -regular near each Σ p \Sigma_p Σ p , however, the inclusion dom A η , τ ⊂ H 1 ( Ω ∖ Σ , C N ) \operatorname{dom}A^{\eta,\tau}\subset H^1(\Omega\setminus\Sigma,\mathbb{C}^N) dom A η , τ ⊂ H 1 ( Ω ∖ Σ , C N ) cannot be ensured as the boundary condition at ∂ Ω \partial\Omega ∂ Ω may still require lower regularity (see Corollary 4.11 for an accurate description of the operator domain in the non-critical case). By adapting the earlier analysis for the operators in the whole space using the tools of microlocal analysis we then arrive at our main result about the essential spectrum of A η , τ A^{\eta,\tau} A η , τ (Theorem 4.19 ), which reads as follows:
spec e s s A η , τ = ⋃ p = 1 P J p , (4) \operatorname{spec}_{\mathrm{ess}} A^{\eta,\tau}=\bigcup_{p=1}^{P}J_p, \tag*{(4)} spec ess A η , τ = p = 1 ⋃ P J p , (4)
where
J p : = { ∅ , η p 2 − τ p 2 ≠ 4 , { − τ p η p m } , n = 2 , η p 2 − τ p 2 = 4 , − τ p η p m + 1 η p [ − a p 2 , a p 2 ] , n = 3 , η p 2 − τ p 2 = 4 , J_p:=\begin{cases}
\varnothing, & \eta_p^2-\tau_p^2\ne4,\\
\left\{-\frac{\tau_p}{\eta_p}m\right\}, & n=2,\ \eta_p^2-\tau_p^2=4,\\
-\frac{\tau_p}{\eta_p}m+\frac{1}{\eta_p}\left[-\frac{a_p}{2},\frac{a_p}{2}\right], & n=3,\ \eta_p^2-\tau_p^2=4,
\end{cases} J p := ⎩ ⎨ ⎧ ∅ , { − η p τ p m } , − η p τ p m + η p 1 [ − 2 a p , 2 a p ] , η p 2 − τ p 2 = 4 , n = 2 , η p 2 − τ p 2 = 4 , n = 3 , η p 2 − τ p 2 = 4 ,
and
a p : = max x ∈ Σ p ∣ κ 1 ( x ) − κ 2 ( x ) ∣ for n = 3 , a_p:=\max_{x\in\Sigma_p}\left|\kappa_1(x)-\kappa_2(x)\right|\quad\text{for }n=3, a p := x ∈ Σ p max ∣ κ 1 ( x ) − κ 2 ( x ) ∣ for n = 3 ,
with κ j ( x ) \kappa_j(x) κ j ( x ) being the principal curvatures of Σ p \Sigma_p Σ p at x x x . In other words, each connected component Σ p \Sigma_p Σ p , p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } , carrying a critical coupling produces its own contribution J p J_p J p to the essential spectrum. Note that the sets J p J_p J p depend neither on Ω \Omega Ω nor on the boundary condition on ∂ Ω \partial\Omega ∂ Ω , and that, in contrast to the whole-space setting (2) –(3) , they are completely visible in a bounded domain, in particular also for m = 0 m=0 m = 0 . The strategy for proving the main result (4) relies on characterizing the essential spectrum of A η , τ A^{\eta,\tau} A η , τ by reducing the spectral problem to the boundary space L 2 ( Σ , C N ) L^2(\Sigma,\mathbb{C}^N) L 2 ( Σ , C N ) via the abstract framework of boundary triples; cf. § 4.1 , Lemma 3.7, and Appendix § 9 . Within this setting, the problem translates into investigating the invertibility of an operator family parameterized by the corresponding Weyl function, which explicitly incorporates certain boundary integral operators of block matrix form, where the integral kernel of the unperturbed Dirac operator in Ω \Omega Ω mentioned before plays an important role. By analyzing the corresponding blocks using pseudodifferential methods (see Appendix A), computing their principal symbols in § 5.1 , and employing decoupling and perturbation techniques in § 5.2 , we are able to explicitly determine the exact intervals J p J_p J p of the essential spectrum.
Acknowledgments
The authors are grateful to Yakir Aharonov for stimulating discussions leading to the problem of the present paper. The research of Jussi Behrndt is funded in part by the Austrian Science Fund (FWF) 10.55776/P 33568-N. Konstantin Pankrashkin was partially supported by the Deutsche Forschungsgemeinschaft (German Research Foundation, DFG), project 491606144. This article is based upon work from COST Action 24122 mSPACE, supported by COST (European Cooperation in Science and Technology), www.cost.eu.
Dirac operators on bounded Lipschitz domains
In this preparatory section we introduce a class of self-adjoint Dirac operators on bounded Lipschitz domains and we construct an integral representation for their resolvents in Theorem 2.5 , which is of independent interest.
Boundary conditions and self-adjointness
For n ∈ { 2 , 3 } n\in\{2,3\} n ∈ { 2 , 3 } we set
N : = { 2 , n = 2 , 4 , n = 3. N:=\begin{cases}
2, & n=2,\\
4, & n=3.
\end{cases} N := { 2 , 4 , n = 2 , n = 3.
and let α 1 , … , α n , β \alpha_1,\ldots,\alpha_n,\beta α 1 , … , α n , β be pairwise anticommuting Hermitian N × N N \times N N × N matrices such that α k 2 = β 2 = I N \alpha_k^2=\beta^2=I_N α k 2 = β 2 = I N , where I N I_N I N is the N × N N \times N N × N identity matrix. In order to simplify some computations below we make an explicit choice based on the Pauli matrices
σ 1 : = ( 0 1 1 0 ) , σ 2 : = ( 0 − i i 0 ) , σ 3 : = ( 1 0 0 − 1 ) . \sigma_1:=\begin{pmatrix}0&1\\1&0\end{pmatrix},\qquad\sigma_2:=\begin{pmatrix}0&-\mathrm{i}\\ \mathrm{i}&0\end{pmatrix},\qquad\sigma_3:=\begin{pmatrix}1&0\\0&-1\end{pmatrix}. σ 1 := ( 0 1 1 0 ) , σ 2 := ( 0 i − i 0 ) , σ 3 := ( 1 0 0 − 1 ) .
For n = 2 n=2 n = 2 we choose α k : = σ k \alpha_k:=\sigma_k α k := σ k for k ∈ { 1 , 2 } k\in\{1,2\} k ∈ { 1 , 2 } and β : = σ 3 \beta:=\sigma_3 β := σ 3 , while for n = 3 n=3 n = 3 we set
α k : = ( 0 σ k σ k 0 ) , k ∈ { 1 , 2 , 3 } , β : = ( I 2 0 0 − I 2 ) . \alpha_k:=\begin{pmatrix}0&\sigma_k\\ \sigma_k&0\end{pmatrix},\qquad k\in\{1,2,3\},\qquad\beta:=\begin{pmatrix}I_2&0\\0&-I_2\end{pmatrix}. α k := ( 0 σ k σ k 0 ) , k ∈ { 1 , 2 , 3 } , β := ( I 2 0 0 − I 2 ) .
We will use the usual abbreviations
α ⋅ x : = ∑ k = 1 n α k x k , σ ⋅ x : = ∑ k = 1 n σ k x k for x = ( x 1 , … , x n ) ∈ R n . \alpha\cdot x:=\sum_{k=1}^{n}\alpha_kx_k,\qquad\sigma\cdot x:=\sum_{k=1}^{n}\sigma_kx_k\quad\text{for }x=(x_1,\ldots,x_n)\in\mathbb{R}^n. α ⋅ x := k = 1 ∑ n α k x k , σ ⋅ x := k = 1 ∑ n σ k x k for x = ( x 1 , … , x n ) ∈ R n .
The Dirac operator D m D_m D m with a mass m ∈ R m\in\mathbb{R} m ∈ R is a matrix differential operator acting on differentiable vector functions f : R n → C N f:\mathbb{R}^n\to\mathbb{C}^N f : R n → C N by
D m f : = − i α ⋅ ∇ f + m β f = − i ∑ k = 1 n α k ∂ k f + m β f . (5) D_mf:=-\mathrm{i}\alpha\cdot\nabla f+m\beta f=-\mathrm{i}\sum_{k=1}^{n}\alpha_k\partial_kf+m\beta f. \tag*{(5)} D m f := − i α ⋅ ∇ f + m β f = − i k = 1 ∑ n α k ∂ k f + m β f . (5)
In the sequel the mass m m m will often play no essential role in our considerations and hence we will typically denote the operator in (2.1) simply by D D D ; whenever the mass term is relevant for an argument we shall indicate it by using the subindex m m m , that is, D = D m D=D_m D = D m ; in most cases this will be done for m = 0 m=0 m = 0 .
From now on we will assume throughout this paper that Ω ⊂ R n \Omega\subset\mathbb{R}^n Ω ⊂ R n is a bounded domain (connected open set) with Lipschitz boundary and in short we will say that Ω \Omega Ω is a bounded Lipschitz domain . For s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] we consider the Sobolev-type spaces
H α s ( Ω , C N ) : = { f ∈ H s ( Ω , C N ) : D 0 f ∈ L 2 ( Ω , C N ) } , (6) H_\alpha^s(\Omega,\mathbb{C}^N):=\{f\in H^s(\Omega,\mathbb{C}^N):D_0f\in L^2(\Omega,\mathbb{C}^N)\}, \tag*{(6)} H α s ( Ω , C N ) := { f ∈ H s ( Ω , C N ) : D 0 f ∈ L 2 ( Ω , C N )} , (6)
which are complete if endowed with norms
∥ f ∥ H α s ( Ω , C N ) 2 : = ∥ f ∥ H s ( Ω , C N ) 2 + ∥ D 0 f ∥ L 2 ( Ω , C N ) 2 . (7) \|f\|_{H_\alpha^s(\Omega,\mathbb{C}^N)}^2:=\|f\|_{H^s(\Omega,\mathbb{C}^N)}^2+\|D_0f\|_{L^2(\Omega,\mathbb{C}^N)}^2. \tag*{(7)} ∥ f ∥ H α s ( Ω , C N ) 2 := ∥ f ∥ H s ( Ω , C N ) 2 + ∥ D 0 f ∥ L 2 ( Ω , C N ) 2 . (7)
Observe that for s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] and f ∈ H s ( Ω , C N ) f\in H^s(\Omega,\mathbb{C}^N) f ∈ H s ( Ω , C N ) one has D 0 f ∈ L 2 ( Ω , C N ) D_0f\in L^2(\Omega,\mathbb{C}^N) D 0 f ∈ L 2 ( Ω , C N ) if and only if D m f ∈ L 2 ( Ω , C N ) D_mf\in L^2(\Omega,\mathbb{C}^N) D m f ∈ L 2 ( Ω , C N ) for some m ∈ R m\in\mathbb{R} m ∈ R and that the norm in (2.3) is equivalent to ∥ f ∥ H s ( Ω , C N ) 2 + ∥ D m f ∥ L 2 ( Ω , C N ) 2 \|f\|_{H^s(\Omega,\mathbb{C}^N)}^2+\|D_mf\|_{L^2(\Omega,\mathbb{C}^N)}^2 ∥ f ∥ H s ( Ω , C N ) 2 + ∥ D m f ∥ L 2 ( Ω , C N ) 2 .
We start with a standard result on the Dirichlet trace map that can be found in, e.g. [12 Lemma 4.1] .
Lemma 2.1. The Dirichlet trace map C ∞ ( Ω ‾ , C N ) ∋ f ↦ f ∣ ∂ Ω ∈ L 2 ( ∂ Ω , C N ) C^\infty(\overline{\Omega},\mathbb{C}^N)\ni f\mapsto f|_{\partial\Omega}\in L^2(\partial\Omega,\mathbb{C}^N) C ∞ ( Ω , C N ) ∋ f ↦ f ∣ ∂ Ω ∈ L 2 ( ∂ Ω , C N ) admits a unique extension to a bounded linear operator
γ ∂ Ω : H α 1 2 ( Ω , C N ) ↦ L 2 ( ∂ Ω , C N ) . \gamma^{\partial\Omega}:H_\alpha^{\frac{1}{2}}(\Omega,\mathbb{C}^N)\mapsto L^2(\partial\Omega,\mathbb{C}^N). γ ∂ Ω : H α 2 1 ( Ω , C N ) ↦ L 2 ( ∂ Ω , C N ) .
In the following, let B ∈ L ∞ ( ∂ Ω , C N × N ) B\in L^\infty(\partial\Omega,\mathbb{C}^{N\times N}) B ∈ L ∞ ( ∂ Ω , C N × N ) be an a.e. defined matrix function such that the operator
A f = − i α ⋅ ∇ f + m β f , dom A = { f ∈ H α 1 2 ( Ω , C N ) : B γ ∂ Ω f = 0 on ∂ Ω } , (8) Af=-\mathrm{i}\alpha\cdot\nabla f+m\beta f,\qquad\operatorname{dom}A=\{f\in H_\alpha^{\frac{1}{2}}(\Omega,\mathbb{C}^N):B\gamma^{\partial\Omega}f=0\text{ on }\partial\Omega\}, \tag*{(8)} A f = − i α ⋅ ∇ f + m β f , dom A = { f ∈ H α 2 1 ( Ω , C N ) : B γ ∂ Ω f = 0 on ∂ Ω } , (8)
is self-adjoint in L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^N) L 2 ( Ω , C N ) . Then B \mathcal{B} B will be referred to as an elliptic boundary condition on ∂ Ω \partial\Omega ∂ Ω . We note that the resolvent of the operator A A A is compact, and hence the spectrum of A A A is purely discrete. The matrix function B \mathcal{B} B will remain fixed throughout the rest of the text.
Remark 2.2. A large choice of elliptic boundary conditions on ∂ Ω \partial\Omega ∂ Ω is available for smooth Ω \Omega Ω , which is discussed in, e.g. [4 , 5 , 11 , 15 , 29 , 36 ] , and these results were extended for non-smooth ∂ Ω \partial\Omega ∂ Ω in [12 , 19 ] . An important boundary condition suitable for any bounded Lipschitz Ω \Omega Ω is given by the choice
B = B ± : = I N ± i β α ⋅ ν ∂ Ω , (9) \mathcal{B}=\mathcal{B}_{\pm}:=I_N\pm i\boldsymbol{\beta}\boldsymbol{\alpha}\cdot\nu^{\partial\Omega}, \tag*{(9)} B = B ± := I N ± i β α ⋅ ν ∂ Ω , (9)
where ν ∂ Ω = ( ν 1 ∂ Ω , … , ν n ∂ Ω ) \nu^{\partial\Omega}=(\nu_1^{\partial\Omega},\ldots,\nu_n^{\partial\Omega}) ν ∂ Ω = ( ν 1 ∂ Ω , … , ν n ∂ Ω ) is the a.e. defined outer unit normal on ∂ Ω \partial\Omega ∂ Ω . The boundary condition B ± γ ∂ Ω f = 0 \mathcal{B}_{\pm}\gamma^{\partial\Omega}f=0 B ± γ ∂ Ω f = 0 is often referred to as MIT bag (for + + + ) and anti-MIT bag (for − - − ). We note that for this choice one has dom A ⊂ H 1 ( Ω , C N ) \operatorname{dom} A\subset H^1(\Omega,\mathbb{C}^N) dom A ⊂ H 1 ( Ω , C N ) if Ω \Omega Ω is smooth [11 , 15 , 36 ] or convex [37 ] .
For subsequent constructions we explicitly state the following assertion:
Lemma 2.3. One has
H 0 1 ( Ω , C N ) ⊂ dom A ⊂ H l o c 1 ( Ω , C N ) , (10) H_0^1(\Omega,\mathbb{C}^N)\subset\operatorname{dom} A\subset H_{\mathrm{loc}}^1(\Omega,\mathbb{C}^N), \tag*{(10)} H 0 1 ( Ω , C N ) ⊂ dom A ⊂ H loc 1 ( Ω , C N ) , (10) and
C c ∞ ( Ω , C N ) ⊂ ran ( A − z ) ( dom A ∩ C ∞ ( Ω , C N ) ) , z ∈ res A . C_c^\infty(\Omega,\mathbb{C}^N)\subset\operatorname{ran}(A-z)\bigl(\operatorname{dom} A\cap C^\infty(\Omega,\mathbb{C}^N)\bigr),\qquad z\in\operatorname{res} A. C c ∞ ( Ω , C N ) ⊂ ran ( A − z ) ( dom A ∩ C ∞ ( Ω , C N ) ) , z ∈ res A . In particular, A A A is essentially self-adjoint on
dom ∞ A : = dom A ∩ C ∞ ( Ω , C N ) . (11) \operatorname{dom}_{\infty} A:=\operatorname{dom} A\cap C^\infty(\Omega,\mathbb{C}^N). \tag*{(11)} dom ∞ A := dom A ∩ C ∞ ( Ω , C N ) . (11)
Proof. The left-hand side inclusion in (2.6) is clear, and the one on the right-hand side follows by interior elliptic regularity. For any g ∈ C c ∞ ( Ω , C N ) g\in C_c^\infty(\Omega,\mathbb{C}^N) g ∈ C c ∞ ( Ω , C N ) and z ∈ res A z\in\operatorname{res} A z ∈ res A there is a unique f ∈ dom A f\in\operatorname{dom} A f ∈ dom A with ( A − z ) f = g (A-z)f=g ( A − z ) f = g . In particular, ( D − z ) f = g (D-z)f=g ( D − z ) f = g in Ω \Omega Ω , and due to the interior elliptic regularity we obtain f ∈ C ∞ ( Ω , C N ) f\in C^\infty(\Omega,\mathbb{C}^N) f ∈ C ∞ ( Ω , C N ) . Hence, C c ∞ ( Ω , C N ) ⊂ ran ( A − z ) dom ∞ A C_c^\infty(\Omega,\mathbb{C}^N)\subset\operatorname{ran}(A-z)\operatorname{dom}_{\infty} A C c ∞ ( Ω , C N ) ⊂ ran ( A − z ) dom ∞ A , and the density of C c ∞ ( Ω , C N ) C_c^\infty(\Omega,\mathbb{C}^N) C c ∞ ( Ω , C N ) in L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^N) L 2 ( Ω , C N ) implies that dom ∞ A \operatorname{dom}_{\infty} A dom ∞ A is an operator core of A A A for any z ∈ res A z\in\operatorname{res} A z ∈ res A ; thus A A A restricted to dom ∞ A \operatorname{dom}_{\infty} A dom ∞ A is essentially self-adjoint.
Integral kernels and fundamental solutions
Recall that the Euclidean Dirac operator D D D satisfies the identity
( D − z ) ( D + z ) = ( − Δ + m 2 − z 2 ) ⊗ I N , z ∈ C , (12) (D-z)(D+z)=(-\Delta+m^2-z^2)\otimes I_N,\qquad z\in\mathbb{C}, \tag*{(12)} ( D − z ) ( D + z ) = ( − Δ + m 2 − z 2 ) ⊗ I N , z ∈ C , (12)
where Δ \Delta Δ is the Laplacian in R n \mathbb{R}^n R n . For any λ ∈ C \lambda\in\mathbb{C} λ ∈ C , a fundamental solution E λ E_\lambda E λ of − Δ − λ -\Delta-\lambda − Δ − λ is given by
E λ : R n ∖ { 0 } → C , E λ ( x ) = { 1 2 π K 0 ( − i λ ∣ x ∣ ) , if n = 2 and λ ≠ 0 , − 1 2 π log ∣ x ∣ , if n = 2 and λ = 0 , e i λ ∣ x ∣ 4 π ∣ x ∣ , if n = 3 , (13) E_\lambda:\mathbb{R}^n\setminus\{0\}\to\mathbb{C},\qquad E_\lambda(x)=
\begin{cases}
\frac{1}{2\pi}K_0(-i\sqrt{\lambda}|x|), & \text{if } n=2 \text{ and } \lambda\ne0,\\
-\frac{1}{2\pi}\log|x|, & \text{if } n=2 \text{ and } \lambda=0,\\
\frac{e^{i\sqrt{\lambda}|x|}}{4\pi|x|}, & \text{if } n=3,
\end{cases} \tag*{(13)} E λ : R n ∖ { 0 } → C , E λ ( x ) = ⎩ ⎨ ⎧ 2 π 1 K 0 ( − i λ ∣ x ∣ ) , − 2 π 1 log ∣ x ∣ , 4 π ∣ x ∣ e i λ ∣ x ∣ , if n = 2 and λ = 0 , if n = 2 and λ = 0 , if n = 3 , (13)
where the square root is fixed by ℑ w > 0 \Im\sqrt{w}>0 ℑ w > 0 for w ∈ C ∖ [ 0 , ∞ ) w\in\mathbb{C}\setminus[0,\infty) w ∈ C ∖ [ 0 , ∞ ) and w ≥ 0 \sqrt{w}\geq0 w ≥ 0 for w ∈ [ 0 , ∞ ) w\in[0,\infty) w ∈ [ 0 , ∞ ) , and K j K_j K j are the j j j -th modified Bessel functions of second kind. It follows from (12) that the function
E z : = ( D + z ) ( E z 2 − m 2 ⊗ I N ) E_z:=(D+z)(\mathcal{E}_{z^2-m^2}\otimes I_N) E z := ( D + z ) ( E z 2 − m 2 ⊗ I N )
is a fundamental solution of D − z D-z D − z for any z ∈ C z\in\mathbb{C} z ∈ C . We set
k ( z ) : = z 2 − m 2 ; k(z):=\sqrt{z^2-m^2}; k ( z ) := z 2 − m 2 ;
then a simple direct computation shows that for any x ∈ R n ∖ { 0 } x\in\mathbb{R}^n\setminus\{0\} x ∈ R n ∖ { 0 }
E z ( x ) = { 1 2 π k ( z ) K 1 ( − i k ( z ) ∣ x ∣ ) α ⋅ x ∣ x ∣ + 1 2 π K 0 ( − i k ( z ) ∣ x ∣ ) ( z I 2 + m β ) , z 2 ≠ m 2 , − 1 2 π ( z I 2 + m β ) log ∣ x ∣ + i 2 π α ⋅ x ∣ x ∣ 2 , z 2 = m 2 . (14) E_z(x)=
\begin{cases}
\frac{1}{2\pi}k(z)K_1(-ik(z)|x|)\frac{\alpha\cdot x}{|x|}+\frac{1}{2\pi}K_0(-ik(z)|x|)(zI_2+m\beta), & z^2\ne m^2,\\[6pt]
-\frac{1}{2\pi}(zI_2+m\beta)\log|x|+\frac{i}{2\pi}\frac{\alpha\cdot x}{|x|^2}, & z^2=m^2.
\end{cases} \tag*{(14)} E z ( x ) = ⎩ ⎨ ⎧ 2 π 1 k ( z ) K 1 ( − ik ( z ) ∣ x ∣ ) ∣ x ∣ α ⋅ x + 2 π 1 K 0 ( − ik ( z ) ∣ x ∣ ) ( z I 2 + m β ) , − 2 π 1 ( z I 2 + m β ) log ∣ x ∣ + 2 π i ∣ x ∣ 2 α ⋅ x , z 2 = m 2 , z 2 = m 2 . (14)
E z ( x ) = ( z I 4 + m β + ( 1 − i k ( z ) ∣ x ∣ ) i α ⋅ x ∣ x ∣ 2 ) 1 4 π ∣ x ∣ e i k ( z ) ∣ x ∣ . (15) E_z(x)=\left(zI_4+m\beta+(1-ik(z)|x|)\frac{i\alpha\cdot x}{|x|^2}\right)\frac{1}{4\pi|x|}e^{ik(z)|x|}. \tag*{(15)} E z ( x ) = ( z I 4 + m β + ( 1 − ik ( z ) ∣ x ∣ ) ∣ x ∣ 2 i α ⋅ x ) 4 π ∣ x ∣ 1 e ik ( z ) ∣ x ∣ . (15)
In particular, since E z E_z E z is a fundamental solution for D − z D-z D − z one has for all y ∈ R n y\in\mathbb{R}^n y ∈ R n and any ξ ∈ C N \xi\in\mathbb{C}^N ξ ∈ C N
( D − z ) E z ( ⋅ − y ) ξ = δ y ξ . (16) (D-z)E_z(\mathord{\cdot}-y)\xi=\delta_y\xi. \tag*{(16)} ( D − z ) E z ( ⋅ − y ) ξ = δ y ξ . (16)
Furthermore, as E z E_z E z has a singularity of order n − 1 n-1 n − 1 at x = 0 x=0 x = 0 and it is smooth for x ≠ 0 x\ne0 x = 0 , one has
E z ∈ L l o c 1 ( R n , C N × N ) . (17) E_z\in L^1_{\mathrm{loc}}(\mathbb{R}^n,\mathbb{C}^{N\times N}). \tag*{(17)} E z ∈ L loc 1 ( R n , C N × N ) . (17)
In our further considerations we will use the function
Υ z ( x ) : = E z ( x ) − E z ‾ ( − x ) ∗ , x ∈ R n ∖ { 0 } , z ∈ C . (18) \Upsilon_z(x):=E_z(x)-E_{\overline{z}}(-x)^*,\qquad x\in\mathbb{R}^n\setminus\{0\},\quad z\in\mathbb{C}. \tag*{(18)} Υ z ( x ) := E z ( x ) − E z ( − x ) ∗ , x ∈ R n ∖ { 0 } , z ∈ C . (18)
Some basic facts about Υ z \Upsilon_z Υ z are collected in the following lemma.
Lemma 2.4. Set I m : = ( − ∞ , − ∣ m ∣ ) ∪ ( ∣ m ∣ , ∞ ) I_m:=(-\infty,-|m|)\cup(|m|,\infty) I m := ( − ∞ , − ∣ m ∣ ) ∪ ( ∣ m ∣ , ∞ ) . Then, Υ z ≡ 0 \Upsilon_z\equiv0 Υ z ≡ 0 for all z ∉ I m z\notin I_m z ∈ / I m , while for z ∈ I m z\in I_m z ∈ I m one has
Υ z = 2 i ( D + z ) ( E k ( z ) 2 ⊗ I N ) , E k ( z ) 2 ( x ) = { 1 4 J 0 ( k ( z ) ∣ x ∣ ) , n = 2 , sin ( k ( z ) ∣ x ∣ ) 4 π ∣ x ∣ , n = 3 , (19) \Upsilon_z=2i(D+z)(\mathcal{E}_{k(z)^2}\otimes I_N),\qquad\mathcal{E}_{k(z)^2}(x)=
\begin{cases}
\frac{1}{4}J_0(k(z)|x|), & n=2,\\[6pt]
\frac{\sin(k(z)|x|)}{4\pi|x|}, & n=3,
\end{cases} \tag*{(19)} Υ z = 2 i ( D + z ) ( E k ( z ) 2 ⊗ I N ) , E k ( z ) 2 ( x ) = ⎩ ⎨ ⎧ 4 1 J 0 ( k ( z ) ∣ x ∣ ) , 4 π ∣ x ∣ s i n ( k ( z ) ∣ x ∣ ) , n = 2 , n = 3 , (19) where J 0 J_0 J 0 is the Bessel function of the first kind of order 0. In particular, in all cases Υ z \Upsilon_z Υ z extends to a real-analytic function on R n \mathbb{R}^n R n , which is denoted by the same symbol, and ( D − z ) Υ z = 0 (D-z)\Upsilon_z=0 ( D − z ) Υ z = 0 in R n \mathbb{R}^n R n .
Proof. First, using the convention for the complex square root one sees directly that Υ z ≡ 0 \Upsilon_z\equiv0 Υ z ≡ 0 for z ∉ I m z\notin I_m z ∈ / I m . Next, for z ∈ I m z \in I_m z ∈ I m one has k ( z ) > 0 k(z)>0 k ( z ) > 0 and thus, Υ z = 2 i ( D + z ) ( E k 2 ( z ) ⊗ I N ) \Upsilon_z=2i(D+z)(\mathfrak{E}_{k^2(z)}\otimes I_N) Υ z = 2 i ( D + z ) ( E k 2 ( z ) ⊗ I N ) . To further study this function in dimension n = 2 n=2 n = 2 , we use the identity π i J 0 ( t ) = K 0 ( − i t ) − K 0 ( i t ) \pi iJ_0(t)=K_0(-it)-K_0(it) π i J 0 ( t ) = K 0 ( − i t ) − K 0 ( i t ) for t > 0 t>0 t > 0 , see, e.g., [34 equation 10.27.9] , yielding for x ≠ 0 x\ne0 x = 0
E k 2 ( z ) ( x ) = 1 4 π i ( K 0 ( − i k ( z ) ∣ x ∣ ) − K 0 ( i k ( z ) ∣ x ∣ ) ) = 1 4 J 0 ( k ( z ) ∣ x ∣ ) = 1 4 ∑ j = 0 ∞ ( − 1 ) j ( k ( z ) ∣ x ∣ ) 2 j 4 j ( j ! ) 2 , \mathfrak{E}_{k^2(z)}(x)=\frac{1}{4\pi i}\left(K_0(-ik(z)|x|)-K_0(ik(z)|x|)\right)
=\frac{1}{4}J_0(k(z)|x|)=\frac{1}{4}\sum_{j=0}^{\infty}(-1)^j\frac{(k(z)|x|)^{2j}}{4^j(j!)^2}, E k 2 ( z ) ( x ) = 4 π i 1 ( K 0 ( − ik ( z ) ∣ x ∣ ) − K 0 ( ik ( z ) ∣ x ∣ ) ) = 4 1 J 0 ( k ( z ) ∣ x ∣ ) = 4 1 j = 0 ∑ ∞ ( − 1 ) j 4 j ( j ! ) 2 ( k ( z ) ∣ x ∣ ) 2 j ,
where in the last step J 0 J_0 J 0 was represented as power series (see, e.g., [34 equation 10.2.2] ). From this it follows that E k 2 ( z ) \mathfrak{E}_{k^2(z)} E k 2 ( z ) and thus, also Υ z \Upsilon_z Υ z can be extended to real-analytic functions on R 2 \mathbb{R}^2 R 2 .
Likewise, for n = 3 n=3 n = 3 one gets with Euler’s identity and the power series representation of sin t \sin t sin t that
E k 2 ( z ) ( x ) = sin ( k ( z ) ∣ x ∣ ) 4 π ∣ x ∣ = k ( z ) 4 π ∑ j = 0 ∞ ( − 1 ) j ( k ( z ) ∣ x ∣ ) 2 j ( 2 j + 1 ) ! . \mathfrak{E}_{k^2(z)}(x)=\frac{\sin(k(z)|x|)}{4\pi|x|}=\frac{k(z)}{4\pi}\sum_{j=0}^{\infty}(-1)^j\frac{(k(z)|x|)^{2j}}{(2j+1)!}. E k 2 ( z ) ( x ) = 4 π ∣ x ∣ sin ( k ( z ) ∣ x ∣ ) = 4 π k ( z ) j = 0 ∑ ∞ ( − 1 ) j ( 2 j + 1 )! ( k ( z ) ∣ x ∣ ) 2 j .
and therefore, E k 2 ( z ) \mathfrak{E}_{k^2(z)} E k 2 ( z ) and also Υ z \Upsilon_z Υ z can be extended to real-analytic functions on R 3 \mathbb{R}^3 R 3 .
Finally, while one clearly has ( D − z ) Υ z = 0 (D-z)\Upsilon_z=0 ( D − z ) Υ z = 0 in R n \mathbb{R}^n R n for z ∉ I m z\notin I_m z ∈ / I m , one also has for z ∈ I m z\in I_m z ∈ I m
( D − z ) Υ z = 2 i ( D − z ) ( D + z ) ( E k 2 ( z ) ⊗ I N ) = 2 i ( − Δ − k ( z ) 2 ) ( E k 2 ( z ) ⊗ I N ) = 0 in D ′ ( R n ) \begin{aligned}
(D-z)\Upsilon_z=2i(D-z)(D+z)(\mathfrak{E}_{k^2(z)}\otimes I_N) \\
&=2i(-\Delta-k(z)^2)(\mathfrak{E}_{k^2(z)}\otimes I_N)=0\quad\text{in }\mathcal{D}'(\mathbb{R}^n)
\end{aligned} ( D − z ) Υ z = 2 i ( D − z ) ( D + z ) ( E k 2 ( z ) ⊗ I N ) = 2 i ( − Δ − k ( z ) 2 ) ( E k 2 ( z ) ⊗ I N ) = 0 in D ′ ( R n )
and thus, due to the smoothness of Υ z \Upsilon_z Υ z also ( D − z ) Υ z = 0 (D-z)\Upsilon_z=0 ( D − z ) Υ z = 0 in R n \mathbb{R}^n R n . This finishes the proof.
In the next theorem we provide an integral representation of the resolvent of the self-adjoint Dirac operator A A A in (2.4) with the help of the fundamental solution E z E_z E z . The proof is inspired by [1 , 43 ] .
Theorem 2.5 (Integral kernel of the resolvent of A A A ). For any z ∈ res A z\in\operatorname{res} A z ∈ res A there exists a matrix function T z ∈ C ∞ ( Ω × Ω , C N × N ) T_z\in C^\infty(\Omega\times\Omega,\mathbb{C}^{N\times N}) T z ∈ C ∞ ( Ω × Ω , C N × N ) such that the matrix function G z G_z G z defined by
G z : { ( x , y ) ∈ Ω × Ω : x ≠ y } → C N × N , G z ( x , y ) = E z ( x − y ) + T z ( x , y ) , G_z:\{(x,y)\in\Omega\times\Omega:x\ne y\}\to\mathbb{C}^{N\times N},\qquad G_z(x,y)=E_z(x-y)+T_z(x,y), G z : {( x , y ) ∈ Ω × Ω : x = y } → C N × N , G z ( x , y ) = E z ( x − y ) + T z ( x , y ) , satisfies G z ∈ L l o c 1 ( Ω × Ω , C N × N ) G_z\in L^1_{\mathrm{loc}}(\Omega\times\Omega,\mathbb{C}^{N\times N}) G z ∈ L loc 1 ( Ω × Ω , C N × N ) and
( A − z ) − 1 f ( x ) = ∫ Ω G z ( x , y ) f ( y ) d y , x ∈ Ω , f ∈ C ∞ ( Ω , C N ) . (20) (A-z)^{-1}f(x)=\int_{\Omega}G_z(x,y)f(y)\,\mathrm{d}y,\qquad x\in\Omega,\ f\in C^\infty(\Omega,\mathbb{C}^N). \tag*{(20)} ( A − z ) − 1 f ( x ) = ∫ Ω G z ( x , y ) f ( y ) d y , x ∈ Ω , f ∈ C ∞ ( Ω , C N ) . (20) Moreover, with Υ z \Upsilon_z Υ z as in (2.14), for all z ∈ res A z\in\operatorname{res} A z ∈ res A one has
T z ( x , y ) ∗ = T z ‾ ( y , x ) − Υ z ( x − y ) ∗ , ( x , y ) ∈ Ω × Ω , (21) T_z(x,y)^*=T_{\overline{z}}(y,x)-\Upsilon_z(x-y)^*,\qquad(x,y)\in\Omega\times\Omega, \tag*{(21)} T z ( x , y ) ∗ = T z ( y , x ) − Υ z ( x − y ) ∗ , ( x , y ) ∈ Ω × Ω , (21) and
G z ( x , y ) ∗ = G z ‾ ( y , x ) , ( x , y ) ∈ Ω × Ω , x ≠ y . (22) G_z(x,y)^*=G_{\overline{z}}(y,x),\qquad(x,y)\in\Omega\times\Omega,\ x\ne y. \tag*{(22)} G z ( x , y ) ∗ = G z ( y , x ) , ( x , y ) ∈ Ω × Ω , x = y . (22) In particular, T z ( x , y ) ∗ = T z ‾ ( y , x ) T_z(x,y)^*=T_{\overline{z}}(y,x) T z ( x , y ) ∗ = T z ( y , x ) for all ( x , y ) ∈ Ω × Ω (x,y)\in\Omega\times\Omega ( x , y ) ∈ Ω × Ω if z ∈ ( res A ) ∖ I m z\in(\operatorname{res} A)\setminus I_m z ∈ ( res A ) ∖ I m .
For the proof we need some preparations. Recall that a function F : M → L 2 ( Ω , C N ) F:M\to L^2(\Omega,\mathbb{C}^N) F : M → L 2 ( Ω , C N ) is differentiable in an open and bounded set M ⊂ R n M\subset\mathbb{R}^n M ⊂ R n with respect to x j x_j x j , if there exists a function ∂ x j F : M → L 2 ( Ω , C N ) \partial_{x_j}F:M\to L^2(\Omega,\mathbb{C}^N) ∂ x j F : M → L 2 ( Ω , C N ) such that for any x ∈ M x\in M x ∈ M
lim t → 0 ∥ F ( x + t e j ) − F ( x ) t − ∂ x j F ( x ) ∥ L 2 ( Ω , C N ) = 0 , \lim_{t\to0}\left\lVert\frac{F(x+te_j)-F(x)}{t}-\partial_{x_j}F(x)\right\rVert_{L^2(\Omega,\mathbb{C}^N)}=0, t → 0 lim t F ( x + t e j ) − F ( x ) − ∂ x j F ( x ) L 2 ( Ω , C N ) = 0 ,
where e j e_j e j denotes the j j j -th canonical basis vector in R n \mathbb{R}^n R n . In the following we write C ∞ ( M , L 2 ( Ω , C N ) ) C^\infty(M,L^2(\Omega,\mathbb{C}^N)) C ∞ ( M , L 2 ( Ω , C N )) for the set of all smooth L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^N) L 2 ( Ω , C N ) -valued functions. The following simple lemma will be useful to show the differentiability of L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^N) L 2 ( Ω , C N ) -valued functions.
Lemma 2.6. Let M ⊂ R n M\subset\mathbb{R}^n M ⊂ R n be open and bounded and let f ∈ C ∞ ( M × Ω ‾ , C N ) f\in C^\infty(M\times\overline{\Omega},\mathbb{C}^N) f ∈ C ∞ ( M × Ω , C N ) . Then, the function F : M → L 2 ( Ω , C N ) F:M\to L^2(\Omega,\mathbb{C}^N) F : M → L 2 ( Ω , C N ) defined by F ( x ) : = f ( x , ⋅ ) F(x):=f(x,\cdot) F ( x ) := f ( x , ⋅ ) is differentiable and ∂ x j F ( x ) = ( ∂ x j f ) ( x , ⋅ ) \partial_{x_j}F(x)=(\partial_{x_j}f)(x,\cdot) ∂ x j F ( x ) = ( ∂ x j f ) ( x , ⋅ ) .
Proof. First, as f ∈ C ∞ ( M × Ω ‾ , C N ) f\in C^\infty(M\times\overline{\Omega},\mathbb{C}^N) f ∈ C ∞ ( M × Ω , C N ) , the map F : M → L 2 ( Ω , C N ) F:M\to L^2(\Omega,\mathbb{C}^N) F : M → L 2 ( Ω , C N ) is well-defined. Moreover, fix x ∈ M x\in M x ∈ M and choose δ > 0 \delta>0 δ > 0 with B δ ( x ) ⊂ M B_\delta(x)\subset M B δ ( x ) ⊂ M . For 0 < ∣ t ∣ < δ 0<|t|<\delta 0 < ∣ t ∣ < δ and y ∈ Ω y\in\Omega y ∈ Ω one has
f ( x + t e j , y ) − f ( x , y ) t − ( ∂ x j f ) ( x , y ) = ∫ 0 1 ( ( ∂ x j f ) ( x + s t e j , y ) − ( ∂ x j f ) ( x , y ) ) d s , \frac{f(x+te_j,y)-f(x,y)}{t}-(\partial_{x_j}f)(x,y)=\int_0^1\left((\partial_{x_j}f)(x+ste_j,y)-(\partial_{x_j}f)(x,y)\right)\,\mathrm{d}s, t f ( x + t e j , y ) − f ( x , y ) − ( ∂ x j f ) ( x , y ) = ∫ 0 1 ( ( ∂ x j f ) ( x + s t e j , y ) − ( ∂ x j f ) ( x , y ) ) d s ,
and as ∂ x j f \partial_{x_j}f ∂ x j f is uniformly continuous on the compact set B δ ( x ) ‾ × Ω ‾ \overline{B_\delta(x)}\times\overline{\Omega} B δ ( x ) × Ω the right-hand side converges to 0 uniformly in y ∈ Ω y\in\Omega y ∈ Ω , as t → 0 t\to0 t → 0 . Therefore,
lim t → 0 ∥ F ( x + t e j ) − F ( x ) t − ( ∂ x j f ) ( x , ⋅ ) ∥ L 2 ( Ω , C N ) 2 = lim t → 0 ∫ Ω ∣ f ( x + t e j , y ) − f ( x , y ) t − ( ∂ x j f ) ( x , y ) ∣ 2 d y = 0 , \lim_{t\to0}\left\lVert\frac{F(x+te_j)-F(x)}{t}-(\partial_{x_j}f)(x,\cdot)\right\rVert_{L^2(\Omega,\mathbb{C}^N)}^2
=\lim_{t\to0}\int_\Omega\left|\frac{f(x+te_j,y)-f(x,y)}{t}-(\partial_{x_j}f)(x,y)\right|^2\,\mathrm{d}y=0, t → 0 lim t F ( x + t e j ) − F ( x ) − ( ∂ x j f ) ( x , ⋅ ) L 2 ( Ω , C N ) 2 = t → 0 lim ∫ Ω t f ( x + t e j , y ) − f ( x , y ) − ( ∂ x j f ) ( x , y ) 2 d y = 0 ,
which shows that F F F is differentiable and ∂ x j F ( x ) = ( ∂ x j f ) ( x , ⋅ ) \partial_{x_j}F(x)=(\partial_{x_j}f)(x,\cdot) ∂ x j F ( x ) = ( ∂ x j f ) ( x , ⋅ ) .
Lemma 2.7. For any z ∈ res A z\in\operatorname{res}A z ∈ res A and any y ∈ Ω y\in\Omega y ∈ Ω there is a unique matrix function M y z : Ω → C N × N \mathcal{M}_y^z:\Omega\to\mathbb{C}^{N\times N} M y z : Ω → C N × N such that for any ξ ∈ C N \xi\in\mathbb{C}^N ξ ∈ C N one has M y z ( ⋅ ) ξ ∈ H 2 ( Ω , C N ) \mathcal{M}_y^z(\cdot)\xi\in H^2(\Omega,\mathbb{C}^N) M y z ( ⋅ ) ξ ∈ H 2 ( Ω , C N ) and the function
G y z : Ω ∖ { y } → C N × N , x ↦ G y z ( x ) = E z ‾ ( x − y ) + M y z ( x ) \mathcal{G}_y^z:\Omega\setminus\{y\}\to\mathbb{C}^{N\times N},\qquad x\mapsto\mathcal{G}_y^z(x)=E_{\overline{z}}(x-y)+\mathcal{M}_y^z(x) G y z : Ω ∖ { y } → C N × N , x ↦ G y z ( x ) = E z ( x − y ) + M y z ( x ) satisfies the identity ( D − z ‾ ) G y z ξ = δ y ξ (D-\overline{z})\mathcal{G}_y^z\xi=\delta_y\xi ( D − z ) G y z ξ = δ y ξ in D ′ ( Ω , C N ) \mathcal{D}'(\Omega,\mathbb{C}^N) D ′ ( Ω , C N ) and the boundary condition for A A A , i.e. for every χ ∈ C ∞ ( Ω ‾ ) \chi\in C^\infty(\overline{\Omega}) χ ∈ C ∞ ( Ω ) with χ = 1 \chi=1 χ = 1 near ∂ Ω \partial\Omega ∂ Ω and χ = 0 \chi=0 χ = 0 near y y y one has χ G y z ξ ∈ H A 1 / 2 ( Ω , C N ) \chi\mathcal{G}_y^z\xi\in H_A^{1/2}(\Omega,\mathbb{C}^N) χ G y z ξ ∈ H A 1/2 ( Ω , C N ) and B γ ∂ Ω ( χ G y z ξ ) = 0 B_{\gamma^{\partial\Omega}}(\chi\mathcal{G}_y^z\xi)=0 B γ ∂ Ω ( χ G y z ξ ) = 0 on ∂ Ω \partial\Omega ∂ Ω .
Moreover,
( y ↦ M y z ξ ) ∈ C ∞ ( Ω , L 2 ( Ω , C N ) ) . (23) (y\mapsto\mathcal{M}_y^z\xi)\in C^\infty(\Omega,L^2(\Omega,\mathbb{C}^N)). \tag*{(23)} ( y ↦ M y z ξ ) ∈ C ∞ ( Ω , L 2 ( Ω , C N )) . (23)
Proof. We first prove the uniqueness. Let z ∈ res A z\in\operatorname{res}A z ∈ res A , y ∈ Ω y\in\Omega y ∈ Ω , ξ ∈ C N \xi\in\mathbb{C}^N ξ ∈ C N , and suppose that there are M y z \mathcal{M}_y^z M y z and M ~ y z \widetilde{\mathcal{M}}_y^z M y z such that the matrix functions
x ↦ G y z ( x ) = E z ‾ ( x − y ) + M y z ( x ) , x ↦ G ~ y z ( x ) = E z ‾ ( x − y ) + M ~ y z ( x ) , x ≠ y , x\mapsto\mathcal{G}_y^z(x)=E_{\overline{z}}(x-y)+\mathcal{M}_y^z(x),\qquad x\mapsto\widetilde{\mathcal{G}}_y^z(x)=E_{\overline{z}}(x-y)+\widetilde{\mathcal{M}}_y^z(x),\qquad x\ne y, x ↦ G y z ( x ) = E z ( x − y ) + M y z ( x ) , x ↦ G y z ( x ) = E z ( x − y ) + M y z ( x ) , x = y ,
satisfy the assumptions. Then ( M y z − M ~ y z ) ξ ∈ H 1 / 2 ( Ω , C N ) (\mathcal{M}_y^z-\widetilde{\mathcal{M}}_y^z)\xi\in H^{1/2}(\Omega,\mathbb{C}^N) ( M y z − M y z ) ξ ∈ H 1/2 ( Ω , C N ) and from
( D − z ‾ ) ( M y z − M ~ y z ) ξ = ( D − z ‾ ) ( G y z − G ~ y z ) ξ = δ y ξ − δ y ξ = 0 (24) (D-\overline{z})(\mathcal{M}_y^z-\widetilde{\mathcal{M}}_y^z)\xi=(D-\overline{z})(\mathcal{G}_y^z-\widetilde{\mathcal{G}}_y^z)\xi=\delta_y\xi-\delta_y\xi=0 \tag*{(24)} ( D − z ) ( M y z − M y z ) ξ = ( D − z ) ( G y z − G y z ) ξ = δ y ξ − δ y ξ = 0 (24)
we conclude D ( M y z − M ~ y z ) ξ ∈ L 2 ( Ω , C N ) D(\mathcal{M}^{z}_{y}-\widetilde{\mathcal{M}}^{z}_{y})\xi\in L^{2}(\Omega,\mathbb{C}^{N}) D ( M y z − M y z ) ξ ∈ L 2 ( Ω , C N ) and also D 0 ( M y z − M ~ y z ) ξ ∈ L 2 ( Ω , C N ) D_{0}(\mathcal{M}^{z}_{y}-\widetilde{\mathcal{M}}^{z}_{y})\xi\in L^{2}(\Omega,\mathbb{C}^{N}) D 0 ( M y z − M y z ) ξ ∈ L 2 ( Ω , C N ) . Hence we have ( M y z − M ~ y z ) ξ ∈ H α 1 2 ( Ω , C N ) (\mathcal{M}^{z}_{y}-\widetilde{\mathcal{M}}^{z}_{y})\xi\in H^{\frac{1}{2}}_{\alpha}(\Omega,\mathbb{C}^{N}) ( M y z − M y z ) ξ ∈ H α 2 1 ( Ω , C N ) ; cf. (2.2). Furthermore, for χ ∈ C ∞ ( Ω ‾ ) \chi\in C^{\infty}(\overline{\Omega}) χ ∈ C ∞ ( Ω ) with χ = 1 \chi=1 χ = 1 near ∂ Ω \partial\Omega ∂ Ω and χ = 0 \chi=0 χ = 0 near y y y we have by assumption
χ ( M y z − M ~ y z ) ξ = χ ( G y z − G ~ y z ) ξ ∈ H α 1 2 ( Ω , C N ) \chi(\mathcal{M}^{z}_{y}-\widetilde{\mathcal{M}}^{z}_{y})\xi=\chi(\mathcal{G}^{z}_{y}-\widetilde{\mathcal{G}}^{z}_{y})\xi\in H^{\frac{1}{2}}_{\alpha}(\Omega,\mathbb{C}^{N}) χ ( M y z − M y z ) ξ = χ ( G y z − G y z ) ξ ∈ H α 2 1 ( Ω , C N )
and
B γ ∂ Ω ( χ ( M y z − M ~ y z ) ξ ) = B γ ∂ Ω ( χ ( G y z − G ~ y z ) ξ ) = 0 , B_{\gamma}^{\partial\Omega}\bigl(\chi(\mathcal{M}^{z}_{y}-\widetilde{\mathcal{M}}^{z}_{y})\xi\bigr)=B_{\gamma}^{\partial\Omega}\bigl(\chi(\mathcal{G}^{z}_{y}-\widetilde{\mathcal{G}}^{z}_{y})\xi\bigr)=0, B γ ∂ Ω ( χ ( M y z − M y z ) ξ ) = B γ ∂ Ω ( χ ( G y z − G y z ) ξ ) = 0 ,
which shows B γ ∂ Ω ( ( M y z − M ~ y z ) ξ ) = 0 B_{\gamma}^{\partial\Omega}((\mathcal{M}^{z}_{y}-\widetilde{\mathcal{M}}^{z}_{y})\xi)=0 B γ ∂ Ω (( M y z − M y z ) ξ ) = 0 . Therefore, ( M y z − M ~ y z ) ξ ∈ dom A (\mathcal{M}^{z}_{y}-\widetilde{\mathcal{M}}^{z}_{y})\xi\in\operatorname{dom}A ( M y z − M y z ) ξ ∈ dom A and (2.20) implies ( A − z ‾ ) ( M y z − M ~ y z ) ξ = 0 (A-\overline{z})(\mathcal{M}^{z}_{y}-\widetilde{\mathcal{M}}^{z}_{y})\xi=0 ( A − z ) ( M y z − M y z ) ξ = 0 . Since the operator A − z ‾ A-\overline{z} A − z is injective and ξ ∈ C N \xi\in\mathbb{C}^{N} ξ ∈ C N is arbitrary, it follows that M y z = M ~ y z \mathcal{M}^{z}_{y}=\widetilde{\mathcal{M}}^{z}_{y} M y z = M y z .
Now we prove the existence. Let y 0 ∈ Ω y_{0}\in\Omega y 0 ∈ Ω and r > 0 r>0 r > 0 with B 2 r ( y 0 ) ⊂ Ω B_{2r}(y_{0})\subset\Omega B 2 r ( y 0 ) ⊂ Ω . Let ρ ∈ C ∞ ( Ω ) \rho\in C^{\infty}(\Omega) ρ ∈ C ∞ ( Ω ) such that ρ = 1 \rho=1 ρ = 1 in B r ( y 0 ) B_{r}(y_{0}) B r ( y 0 ) and supp ρ ⊂ B 2 r ( y 0 ) \operatorname{supp}\rho\subset B_{2r}(y_{0}) supp ρ ⊂ B 2 r ( y 0 ) . Let y ∈ B r ( y 0 ) y\in B_{r}(y_{0}) y ∈ B r ( y 0 ) , consider the matrix function x ↦ R ( x ) : = − i α ⋅ ∇ ρ ( x ) x\mapsto R(x):=-i\alpha\cdot\nabla\rho(x) x ↦ R ( x ) := − i α ⋅ ∇ ρ ( x ) , and note that (2.12) implies
( D − z ‾ ) ( ρ E z ‾ ( ⋅ − y ) ξ ) = ρ ( D − z ‾ ) E z ‾ ( ⋅ − y ) ξ + R E z ‾ ( ⋅ − y ) ξ = δ y ξ + R E z ‾ ( ⋅ − y ) ξ (25) \begin{aligned}
(D-\overline{z})\bigl(\rho E_{\overline{z}}(\mathord{\cdot}-y)\xi\bigr)
&=\rho(D-\overline{z})E_{\overline{z}}(\mathord{\cdot}-y)\xi+RE_{\overline{z}}(\mathord{\cdot}-y)\xi\\
&=\delta_{y}\xi+RE_{\overline{z}}(\mathord{\cdot}-y)\xi \tag*{(25)}
\end{aligned} ( D − z ) ( ρ E z ( ⋅ − y ) ξ ) = ρ ( D − z ) E z ( ⋅ − y ) ξ + R E z ( ⋅ − y ) ξ = δ y ξ + R E z ( ⋅ − y ) ξ (25)
for any ξ ∈ C N \xi\in\mathbb{C}^{N} ξ ∈ C N . Since the matrix function R R R is smooth with support in B 2 r ( y 0 ) B_{2r}(y_{0}) B 2 r ( y 0 ) and vanishes in B r ( y 0 ) B_{r}(y_{0}) B r ( y 0 ) , we conclude that the second term on the right-hand side in (2.21) belongs to L 2 ( Ω ; C N ) L^{2}(\Omega;\mathbb{C}^{N}) L 2 ( Ω ; C N ) . Therefore, there exists a matrix function F y z F^{z}_{y} F y z with
F y z ξ ∈ dom A , ( A − z ‾ ) F y z ξ = R E z ‾ ( ⋅ − y ) ξ , F^{z}_{y}\xi\in\operatorname{dom}A,\qquad(A-\overline{z})F^{z}_{y}\xi=RE_{\overline{z}}(\mathord{\cdot}-y)\xi, F y z ξ ∈ dom A , ( A − z ) F y z ξ = R E z ( ⋅ − y ) ξ ,
and (2.21) takes the form
( D − z ‾ ) [ ( ρ E z ‾ ( ⋅ − y ) − F y z ) ξ ] = δ y ξ . (D-\overline{z})\bigl[(\rho E_{\overline{z}}(\mathord{\cdot}-y)-F^{z}_{y})\xi\bigr]=\delta_{y}\xi. ( D − z ) [ ( ρ E z ( ⋅ − y ) − F y z ) ξ ] = δ y ξ .
Hence, the matrix function G y z : = ρ E z ‾ ( ⋅ − y ) − F y z \mathcal{G}^{z}_{y}:=\rho E_{\overline{z}}(\mathord{\cdot}-y)-F^{z}_{y} G y z := ρ E z ( ⋅ − y ) − F y z satisfies by construction the identity ( D − z ‾ ) G y z ξ = δ y ξ (D-\overline{z})\mathcal{G}^{z}_{y}\xi=\delta_{y}\xi ( D − z ) G y z ξ = δ y ξ . In particular, for χ ∈ C ∞ ( Ω ‾ ) \chi\in C^{\infty}(\overline{\Omega}) χ ∈ C ∞ ( Ω ) with χ = 1 \chi=1 χ = 1 near ∂ Ω \partial\Omega ∂ Ω and χ = 0 \chi=0 χ = 0 near y y y one has D ( χ G y z ξ ) ∈ L 2 ( Ω , C N ) D(\chi\mathcal{G}^{z}_{y}\xi)\in L^{2}(\Omega,\mathbb{C}^{N}) D ( χ G y z ξ ) ∈ L 2 ( Ω , C N ) and D 0 ( χ G y z ξ ) ∈ L 2 ( Ω , C N ) D_{0}(\chi\mathcal{G}^{z}_{y}\xi)\in L^{2}(\Omega,\mathbb{C}^{N}) D 0 ( χ G y z ξ ) ∈ L 2 ( Ω , C N ) . Furthermore, as χ ρ E z ‾ ( ⋅ − y ) \chi\rho E_{\overline{z}}(\mathord{\cdot}-y) χ ρ E z ( ⋅ − y ) is smooth and F y z ξ ∈ dom A ⊂ H α 1 2 ( Ω , C N ) F^{z}_{y}\xi\in\operatorname{dom}A\subset H^{\frac{1}{2}}_{\alpha}(\Omega,\mathbb{C}^{N}) F y z ξ ∈ dom A ⊂ H α 2 1 ( Ω , C N ) we get χ G y z ξ ∈ H α 1 2 ( Ω , C N ) \chi\mathcal{G}^{z}_{y}\xi\in H^{\frac{1}{2}}_{\alpha}(\Omega,\mathbb{C}^{N}) χ G y z ξ ∈ H α 2 1 ( Ω , C N ) . Next, since ρ E z ‾ ( ⋅ − y ) \rho E_{\overline{z}}(\mathord{\cdot}-y) ρ E z ( ⋅ − y ) vanishes near ∂ Ω \partial\Omega ∂ Ω , one has
B γ ∂ Ω ( χ G y z ξ ) = − B γ ∂ Ω ( χ F y z ξ ) = 0 on ∂ Ω . B_{\gamma}^{\partial\Omega}(\chi\mathcal{G}^{z}_{y}\xi)=-B_{\gamma}^{\partial\Omega}(\chi F^{z}_{y}\xi)=0\quad\text{on }\partial\Omega. B γ ∂ Ω ( χ G y z ξ ) = − B γ ∂ Ω ( χ F y z ξ ) = 0 on ∂ Ω.
By denoting M y z : = ( ρ − 1 ) E z ‾ ( ⋅ − y ) − F y z \mathcal{M}^{z}_{y}:=(\rho-1)E_{\overline{z}}(\mathord{\cdot}-y)-F^{z}_{y} M y z := ( ρ − 1 ) E z ( ⋅ − y ) − F y z we have G y z = E z ‾ ( ⋅ − y ) + M y z \mathcal{G}^{z}_{y}=E_{\overline{z}}(\mathord{\cdot}-y)+\mathcal{M}^{z}_{y} G y z = E z ( ⋅ − y ) + M y z . As ( ρ − 1 ) E z ‾ ( ⋅ − y ) (\rho-1)E_{\overline{z}}(\mathord{\cdot}-y) ( ρ − 1 ) E z ( ⋅ − y ) is smooth on Ω ‾ \overline{\Omega} Ω and F y z ξ ∈ dom A F^{z}_{y}\xi\in\operatorname{dom}A F y z ξ ∈ dom A , it follows that M y z ξ ∈ H α 1 2 ( Ω , C N ) \mathcal{M}^{z}_{y}\xi\in H^{\frac{1}{2}}_{\alpha}(\Omega,\mathbb{C}^{N}) M y z ξ ∈ H α 2 1 ( Ω , C N ) and, in particular, M y z ξ ∈ H 1 2 ( Ω , C N ) \mathcal{M}^{z}_{y}\xi\in H^{\frac{1}{2}}(\Omega,\mathbb{C}^{N}) M y z ξ ∈ H 2 1 ( Ω , C N ) .
It remains to show (2.19), which is essentially a consequence of Lemma 2.6. In fact, with R = − i α ⋅ ∇ ρ R=-i\alpha\cdot\nabla\rho R = − i α ⋅ ∇ ρ and ξ ∈ C N \xi\in\mathbb{C}^{N} ξ ∈ C N consider the function
F 1 : B r 2 ( y 0 ) → L 2 ( Ω , C N ) , F 1 ( y ) : = R E z ‾ ( ⋅ − y ) ξ , F_{1}:B_{\frac{r}{2}}(y_{0})\to L^{2}(\Omega,\mathbb{C}^{N}),\qquad F_{1}(y):=RE_{\overline{z}}(\mathord{\cdot}-y)\xi, F 1 : B 2 r ( y 0 ) → L 2 ( Ω , C N ) , F 1 ( y ) := R E z ( ⋅ − y ) ξ ,
and observe that
( x , y ) ↦ f 1 ( x , y ) : = R ( x ) E z ‾ ( x − y ) ξ ∈ C ∞ ( Ω ‾ × B r 2 ( y 0 ) ‾ , C N ) . (x,y)\mapsto f_{1}(x,y):=R(x)E_{\overline{z}}(x-y)\xi\in C^{\infty}\left(\overline{\Omega}\times\overline{B_{\frac{r}{2}}(y_{0})},\mathbb{C}^{N}\right). ( x , y ) ↦ f 1 ( x , y ) := R ( x ) E z ( x − y ) ξ ∈ C ∞ ( Ω × B 2 r ( y 0 ) , C N ) .
Then it follows from Lemma 2.6 that ∂ y i F 1 ( y ) = ( ∂ y i f 1 ) ( ⋅ , y ) = ∂ y i R E z ‾ ( ⋅ − y ) ξ \partial_{y_i}F_1(y)=(\partial_{y_i}f_1)(\cdot,y)=\partial_{y_i}R E_{\overline{z}}(\cdot-y)\xi ∂ y i F 1 ( y ) = ( ∂ y i f 1 ) ( ⋅ , y ) = ∂ y i R E z ( ⋅ − y ) ξ . By iterating this procedure one concludes ( ∂ γ F 1 ) ( y ) = ∂ y γ R E z ‾ ( ⋅ − y ) ξ ∈ L 2 ( Ω , C N ) (\partial^\gamma F_1)(y)=\partial_y^\gamma R E_{\overline{z}}(\cdot-y)\xi\in L^2(\Omega,\mathbb{C}^N) ( ∂ γ F 1 ) ( y ) = ∂ y γ R E z ( ⋅ − y ) ξ ∈ L 2 ( Ω , C N ) for any γ ∈ N 0 n \gamma\in\mathbb{N}_0^n γ ∈ N 0 n , which shows
F 1 ∈ C ∞ ( B r 2 ( y 0 ) , L 2 ( Ω , C N ) ) . (26) F_1\in C^\infty\left(B_{\frac{r}{2}}(y_0),L^2(\Omega,\mathbb{C}^N)\right). \tag*{(26)} F 1 ∈ C ∞ ( B 2 r ( y 0 ) , L 2 ( Ω , C N ) ) . (26)
Since the resolvent ( A − z ‾ ) − 1 (A-\overline{z})^{-1} ( A − z ) − 1 is bounded in L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^N) L 2 ( Ω , C N ) , one gets that
y ↦ F y z ξ = ( A − z ‾ ) − 1 F 1 ( y ) ∈ C ∞ ( B r 2 ( y 0 ) , L 2 ( Ω , C N ) ) y\mapsto F_y^z\xi=(A-\overline{z})^{-1}F_1(y)\in C^\infty\left(B_{\frac{r}{2}}(y_0),L^2(\Omega,\mathbb{C}^N)\right) y ↦ F y z ξ = ( A − z ) − 1 F 1 ( y ) ∈ C ∞ ( B 2 r ( y 0 ) , L 2 ( Ω , C N ) ) .
Next, in a similar way as in (26) , one shows for the map F 2 : B r 2 ( y 0 ) → L 2 ( Ω , C N ) F_2:B_{\frac{r}{2}}(y_0)\to L^2(\Omega,\mathbb{C}^N) F 2 : B 2 r ( y 0 ) → L 2 ( Ω , C N ) defined by F 2 ( y ) : = ( ρ − 1 ) E z ‾ ( ⋅ − y ) ξ F_2(y):=(\rho-1)E_{\overline{z}}(\cdot-y)\xi F 2 ( y ) := ( ρ − 1 ) E z ( ⋅ − y ) ξ that
F 2 ∈ C ∞ ( B r 2 ( y 0 ) , L 2 ( Ω , C N ) ) . F_2\in C^\infty\left(B_{\frac{r}{2}}(y_0),L^2(\Omega,\mathbb{C}^N)\right). F 2 ∈ C ∞ ( B 2 r ( y 0 ) , L 2 ( Ω , C N ) ) .
Therefore, taking the definition M y z = ( ρ − 1 ) E z ‾ ( ⋅ − y ) − F y z M_y^z=(\rho-1)E_{\overline{z}}(\cdot-y)-F_y^z M y z = ( ρ − 1 ) E z ( ⋅ − y ) − F y z into account, one concludes that
( y ↦ M y z ξ ) ∈ C ∞ ( B r 2 ( y 0 ) , L 2 ( Ω , C N ) ) (y\mapsto M_y^z\xi)\in C^\infty\left(B_{\frac{r}{2}}(y_0),L^2(\Omega,\mathbb{C}^N)\right) ( y ↦ M y z ξ ) ∈ C ∞ ( B 2 r ( y 0 ) , L 2 ( Ω , C N ) ) .
Since this is true for any fixed y 0 ∈ Ω y_0\in\Omega y 0 ∈ Ω , the uniqueness implies that ( y ↦ M y z ξ ) (y\mapsto M_y^z\xi) ( y ↦ M y z ξ ) is smooth in Ω \Omega Ω , i.e. that (2.19) is true. □
Lemma 2.8. For any f ∈ dom ∞ A = dom A ∩ C ∞ ( Ω , C N ) f\in\operatorname{dom}_\infty A=\operatorname{dom}A\cap C^\infty(\Omega,\mathbb{C}^N) f ∈ dom ∞ A = dom A ∩ C ∞ ( Ω , C N ) there holds
∫ Ω ( G y z ) ∗ ( x ) ( A − z ) f ( x ) d x = f ( y ) for all y ∈ Ω . \int_\Omega(\mathcal{G}_y^z)^*(x)(A-z)f(x)\,\mathrm{d}x=f(y)\qquad\text{for all }y\in\Omega. ∫ Ω ( G y z ) ∗ ( x ) ( A − z ) f ( x ) d x = f ( y ) for all y ∈ Ω. Proof. Let f ∈ dom ∞ A f\in\operatorname{dom}_\infty A f ∈ dom ∞ A and y ∈ Ω y\in\Omega y ∈ Ω . Pick an arbitrary ξ ∈ C N \xi\in\mathbb{C}^N ξ ∈ C N and set
J : = ∫ Ω ⟨ ξ , ( G y z ) ∗ ( x ) ( A − z ) f ( x ) ⟩ C N d x . J:=\int_\Omega\langle\xi,(\mathcal{G}_y^z)^*(x)(A-z)f(x)\rangle_{\mathbb{C}^N}\,\mathrm{d}x. J := ∫ Ω ⟨ ξ , ( G y z ) ∗ ( x ) ( A − z ) f ( x ) ⟩ C N d x . By linearity, it suffices to show that J = ⟨ ξ , f ( y ) ⟩ C N J=\langle\xi,f(y)\rangle_{\mathbb{C}^N} J = ⟨ ξ , f ( y ) ⟩ C N . Let r > 0 r>0 r > 0 with B 2 r ( y ) ⊂ Ω B_{2r}(y)\subset\Omega B 2 r ( y ) ⊂ Ω and ρ ∈ C ∞ ( Ω ) \rho\in C^\infty(\Omega) ρ ∈ C ∞ ( Ω ) such that ρ = 1 \rho=1 ρ = 1 in B r ( y ) B_r(y) B r ( y ) and supp ρ ⊂ B 2 r ( y ) \operatorname{supp}\rho\subset B_{2r}(y) supp ρ ⊂ B 2 r ( y ) . Then J = J 1 + J 2 J=J_1+J_2 J = J 1 + J 2 with
J 1 : = ∫ Ω ⟨ ξ , ( G y z ) ∗ ( x ) ( A − z ) ( ρ f ) ( x ) ⟩ C N d x , J_1:=\int_\Omega\langle\xi,(\mathcal{G}_y^z)^*(x)(A-z)(\rho f)(x)\rangle_{\mathbb{C}^N}\,\mathrm{d}x, J 1 := ∫ Ω ⟨ ξ , ( G y z ) ∗ ( x ) ( A − z ) ( ρ f ) ( x ) ⟩ C N d x , J 2 : = ∫ Ω ⟨ ξ , ( G y z ) ∗ ( x ) ( A − z ) [ ( 1 − ρ ) f ] ( x ) ⟩ C N d x . J_2:=\int_\Omega\langle\xi,(\mathcal{G}_y^z)^*(x)(A-z)[(1-\rho)f](x)\rangle_{\mathbb{C}^N}\,\mathrm{d}x. J 2 := ∫ Ω ⟨ ξ , ( G y z ) ∗ ( x ) ( A − z ) [( 1 − ρ ) f ] ( x ) ⟩ C N d x . Due to ρ f ∈ C ∞ ( Ω , C N ) \rho f\in C^\infty(\Omega,\mathbb{C}^N) ρ f ∈ C ∞ ( Ω , C N ) we compute J 1 J_1 J 1 as follows:
J 1 = ⟨ G y z ξ , ( D ‾ − z ) ( ρ f ) ⟩ D ′ ( Ω , C N ) , D ( Ω , C N ) = ⟨ ( D ‾ − z ‾ ) G y z ξ , ρ f ⟩ D ′ ( Ω , C N ) , D ( Ω , C N ) = ⟨ δ y ξ , ρ f ⟩ D ′ ( Ω , C N ) , D ( Ω , C N ) = ⟨ ξ , ( ρ f ) ( y ) ⟩ C N = ⟨ ξ , f ( y ) ⟩ C N , \begin{aligned}
J_1&=\langle\mathcal{G}_y^z\xi,(\overline{D}-z)(\rho f)\rangle_{\mathcal{D}'(\Omega,\mathbb{C}^N),\mathcal{D}(\Omega,\mathbb{C}^N)}\\
&=\langle(\overline{D}-\overline{z})\mathcal{G}_y^z\xi,\rho f\rangle_{\mathcal{D}'(\Omega,\mathbb{C}^N),\mathcal{D}(\Omega,\mathbb{C}^N)}\\
&=\langle\delta_y\xi,\rho f\rangle_{\mathcal{D}'(\Omega,\mathbb{C}^N),\mathcal{D}(\Omega,\mathbb{C}^N)}\\
&=\langle\xi,(\rho f)(y)\rangle_{\mathbb{C}^N}\\
&=\langle\xi,f(y)\rangle_{\mathbb{C}^N},
\end{aligned} J 1 = ⟨ G y z ξ , ( D − z ) ( ρ f ) ⟩ D ′ ( Ω , C N ) , D ( Ω , C N ) = ⟨( D − z ) G y z ξ , ρ f ⟩ D ′ ( Ω , C N ) , D ( Ω , C N ) = ⟨ δ y ξ , ρ f ⟩ D ′ ( Ω , C N ) , D ( Ω , C N ) = ⟨ ξ , ( ρ f ) ( y ) ⟩ C N = ⟨ ξ , f ( y ) ⟩ C N , and it remains to show that J 2 = 0 J_2=0 J 2 = 0 . For that, choose ρ ~ ∈ C ∞ ( Ω ‾ ) \widetilde{\rho}\in C^\infty(\overline{\Omega}) ρ ∈ C ∞ ( Ω ) such that ρ ~ = 0 \widetilde{\rho}=0 ρ = 0 on B r 2 ( y ) B_{\frac{r}{2}}(y) B 2 r ( y ) and ρ ~ = 1 \widetilde{\rho}=1 ρ = 1 on Ω ∖ B 3 r 4 ( y ) \Omega\setminus B_{\frac{3r}{4}}(y) Ω ∖ B 4 3 r ( y ) ; as 1 − ρ = 0 1-\rho=0 1 − ρ = 0 on B r ( y ) B_r(y) B r ( y ) , one has ρ ~ = 1 \widetilde{\rho}=1 ρ = 1 on supp ( 1 − ρ ) \operatorname{supp}(1-\rho) supp ( 1 − ρ ) and ∇ ρ ~ \nabla\widetilde{\rho} ∇ ρ and 1 − ρ 1-\rho 1 − ρ have disjoint support. Then
J 2 = ∫ Ω ρ ~ ( x ) G y z ( x ) ξ , ( D − z ) [ ( 1 − ρ ) f ] ( x ) ⟩ C N d x . J_2=\int_\Omega\widetilde{\rho}(x)\mathcal{G}_y^z(x)\xi,(D-z)[(1-\rho)f](x)\rangle_{\mathbb{C}^N}\,\mathrm{d}x. J 2 = ∫ Ω ρ ( x ) G y z ( x ) ξ , ( D − z ) [( 1 − ρ ) f ] ( x ) ⟩ C N d x .
It follows from Lemma 2.7 and ρ ~ = 0 \widetilde{\rho}=0 ρ = 0 near y y y (and ρ ~ = 1 \widetilde{\rho}=1 ρ = 1 near ∂ Ω \partial\Omega ∂ Ω ) that ρ ~ G y z ( ⋅ ) ξ ∈ dom A \widetilde{\rho}G^z_y(\cdot)\xi\in\operatorname{dom} A ρ G y z ( ⋅ ) ξ ∈ dom A for any ξ ∈ C N \xi\in\mathbb{C}^N ξ ∈ C N ; moreover, ( 1 − ρ ) f ∈ dom A (1-\rho)f\in\operatorname{dom} A ( 1 − ρ ) f ∈ dom A , since ( 1 − ρ ) f ∈ H α 1 2 ( Ω , C N ) (1-\rho)f\in H^{\frac{1}{2}}_{\alpha}(\Omega,\mathbb{C}^N) ( 1 − ρ ) f ∈ H α 2 1 ( Ω , C N ) and γ ∂ Ω ( ( 1 − ρ ) f ) = γ ∂ Ω f \gamma^{\partial\Omega}((1-\rho)f)=\gamma^{\partial\Omega}f γ ∂ Ω (( 1 − ρ ) f ) = γ ∂ Ω f . Thus
J 2 = ∫ Ω ⟨ ρ ~ ( x ) G y z ( x ) ξ , ( D − z ) [ ( 1 − ρ ) f ] ( x ) ⟩ C N d x = ⟨ ρ ~ G y z ξ , ( A − z ) [ ( 1 − ρ ) f ] ⟩ L 2 ( Ω , C N ) = ⟨ ( A − z ‾ ) ( ρ ~ G y z ξ ) , ( 1 − ρ ) f ⟩ L 2 ( Ω , C N ) . \begin{aligned}
J_2&=\int_{\Omega}\langle\widetilde{\rho}(x)G^z_y(x)\xi,(D-z)[(1-\rho)f](x)\rangle_{\mathbb{C}^N}\,\mathrm{d}x\\
&=\langle\widetilde{\rho}G^z_y\xi,(A-z)[(1-\rho)f]\rangle_{L^2(\Omega,\mathbb{C}^N)}\\
&=\langle(A-\overline{z})(\widetilde{\rho}G^z_y\xi),(1-\rho)f\rangle_{L^2(\Omega,\mathbb{C}^N)}.
\end{aligned} J 2 = ∫ Ω ⟨ ρ ( x ) G y z ( x ) ξ , ( D − z ) [( 1 − ρ ) f ] ( x ) ⟩ C N d x = ⟨ ρ G y z ξ , ( A − z ) [( 1 − ρ ) f ] ⟩ L 2 ( Ω , C N ) = ⟨( A − z ) ( ρ G y z ξ ) , ( 1 − ρ ) f ⟩ L 2 ( Ω , C N ) .
From ρ ~ = 0 \widetilde{\rho}=0 ρ = 0 near y y y we also have ρ ~ δ y = 0 \widetilde{\rho}\delta_y=0 ρ δ y = 0 and hence
( A − z ‾ ) ( ρ ~ G y z ξ ) = ρ ~ ( D − z ‾ ) G y z ξ + ( − i α ⋅ ∇ ρ ~ ) G y z ξ = ρ ~ δ y ξ + ( − i α ⋅ ∇ ρ ~ ) G y z ξ = ( − i α ⋅ ∇ ρ ~ ) G y z ξ . \begin{aligned}
(A-\overline{z})(\widetilde{\rho}G^z_y\xi)&=\widetilde{\rho}(D-\overline{z})G^z_y\xi+(-i\alpha\cdot\nabla\widetilde{\rho})G^z_y\xi\\
&=\widetilde{\rho}\delta_y\xi+(-i\alpha\cdot\nabla\widetilde{\rho})G^z_y\xi\\
&=(-i\alpha\cdot\nabla\widetilde{\rho})G^z_y\xi.
\end{aligned} ( A − z ) ( ρ G y z ξ ) = ρ ( D − z ) G y z ξ + ( − i α ⋅ ∇ ρ ) G y z ξ = ρ δ y ξ + ( − i α ⋅ ∇ ρ ) G y z ξ = ( − i α ⋅ ∇ ρ ) G y z ξ .
Therefore, using that the functions − i α ⋅ ∇ ρ ~ -i\alpha\cdot\nabla\widetilde{\rho} − i α ⋅ ∇ ρ and 1 − ρ 1-\rho 1 − ρ have disjoint supports we obtain J 2 = 0 J_2=0 J 2 = 0 .
Proof of Theorem 2.5. For z ∈ res A z\in\operatorname{res} A z ∈ res A and x , y ∈ Ω x,y\in\Omega x , y ∈ Ω define
G z ( x , y ) : = G x z ( y ) ∗ , x ≠ y , and T z ( x , y ) : = M x z ( y ) ∗ − Υ z ( x − y ) . G_z(x,y):=G^z_x(y)^*,\qquad x\ne y,\qquad\text{and}\qquad T_z(x,y):=M^z_x(y)^*-\Upsilon_z(x-y). G z ( x , y ) := G x z ( y ) ∗ , x = y , and T z ( x , y ) := M x z ( y ) ∗ − Υ z ( x − y ) .
with G x z G^z_x G x z and M x z M^z_x M x z from Lemma 2.7 (now viewed as functions in the y y y -variable) and Υ z \Upsilon_z Υ z from (2.14). As G x z ( y ) = E z ‾ ( y − x ) + M x z ( y ) G^z_x(y)=E_{\overline{z}}(y-x)+M^z_x(y) G x z ( y ) = E z ( y − x ) + M x z ( y ) and E z ‾ ( y − x ) ∗ = E z ( x − y ) − Υ z ( x − y ) E_{\overline{z}}(y-x)^*=E_z(x-y)-\Upsilon_z(x-y) E z ( y − x ) ∗ = E z ( x − y ) − Υ z ( x − y ) by (2.14), we conclude for all x , y ∈ Ω x,y\in\Omega x , y ∈ Ω with x ≠ y x\ne y x = y that
G z ( x , y ) = E z ( x − y ) + T z ( x , y ) . (27) G_z(x,y)=E_z(x-y)+T_z(x,y). \tag*{(27)} G z ( x , y ) = E z ( x − y ) + T z ( x , y ) . (27)
In view of (2.19), for all j , k ∈ { 1 , … , N } j,k\in\{1,\ldots,N\} j , k ∈ { 1 , … , N } and each compact ω ⊂ Ω \omega\subset\Omega ω ⊂ Ω it holds that
∫ ω ∫ Ω ∣ ( M x z ) j , k ( y ) ∣ 2 d y d x = ∫ ω ∥ ( M x z ) j , k ∥ L 2 ( Ω ) 2 d x < ∞ ; (28) \int_{\omega}\int_{\Omega}\left|(M^z_x)_{j,k}(y)\right|^2\,\mathrm{d}y\,\mathrm{d}x=\int_{\omega}\left\|(M^z_x)_{j,k}\right\|_{L^2(\Omega)}^2\,\mathrm{d}x<\infty; \tag*{(28)} ∫ ω ∫ Ω ∣ ( M x z ) j , k ( y ) ∣ 2 d y d x = ∫ ω ∥ ( M x z ) j , k ∥ L 2 ( Ω ) 2 d x < ∞ ; (28)
here we use that the continuous map ω ∋ x ↦ ( M x z ) j , k ∈ L 2 ( Ω ) \omega\ni x\mapsto(M^z_x)_{j,k}\in L^2(\Omega) ω ∋ x ↦ ( M x z ) j , k ∈ L 2 ( Ω ) defines an element of L 2 ( ω ; L 2 ( Ω ) ) ≃ L 2 ( ω × Ω ) L^2(\omega;L^2(\Omega))\simeq L^2(\omega\times\Omega) L 2 ( ω ; L 2 ( Ω )) ≃ L 2 ( ω × Ω ) , which also gives a meaning to ( x , y ) ↦ ( M x z ) j , k ( y ) (x,y)\mapsto(M^z_x)_{j,k}(y) ( x , y ) ↦ ( M x z ) j , k ( y ) as a measurable function on Ω × Ω \Omega\times\Omega Ω × Ω . As Υ z \Upsilon_z Υ z is smooth by Lemma 2.4, ( T z ) j , k ∈ L l o c 2 ( Ω × Ω ) ⊂ L l o c 1 ( Ω × Ω ) (T_z)_{j,k}\in L^2_{\mathrm{loc}}(\Omega\times\Omega)\subset L^1_{\mathrm{loc}}(\Omega\times\Omega) ( T z ) j , k ∈ L loc 2 ( Ω × Ω ) ⊂ L loc 1 ( Ω × Ω ) . Since E z ∈ L l o c 1 ( R N , C N × N ) E_z\in L^1_{\mathrm{loc}}(\mathbb{R}^N,\mathbb{C}^{N\times N}) E z ∈ L loc 1 ( R N , C N × N ) by (2.13), we conclude that G z ∈ L l o c 1 ( Ω × Ω , C N × N ) G_z\in L^1_{\mathrm{loc}}(\Omega\times\Omega,\mathbb{C}^{N\times N}) G z ∈ L loc 1 ( Ω × Ω , C N × N ) . Since C c ∞ ( Ω , C N ) ⊂ ( A − z ) dom ∞ A C^\infty_c(\Omega,\mathbb{C}^N)\subset(A-z)\operatorname{dom}_{\infty}A C c ∞ ( Ω , C N ) ⊂ ( A − z ) dom ∞ A by Lemma 2.3, one gets with Lemma 2.8
( A − z ) − 1 g ( x ) = ∫ Ω G z ( x , y ) g ( y ) d y , g ∈ C c ∞ ( Ω , C N ) , x ∈ Ω . (A-z)^{-1}g(x)=\int_{\Omega}G_z(x,y)g(y)\,\mathrm{d}y,\qquad g\in C^\infty_c(\Omega,\mathbb{C}^N),\ x\in\Omega. ( A − z ) − 1 g ( x ) = ∫ Ω G z ( x , y ) g ( y ) d y , g ∈ C c ∞ ( Ω , C N ) , x ∈ Ω.
Due to the last identity and the local integrability for any f , g ∈ C c ∞ ( Ω , C N ) f,g\in C^\infty_c(\Omega,\mathbb{C}^N) f , g ∈ C c ∞ ( Ω , C N ) one uses Fubini’s theorem to show that
∫ Ω ∫ Ω ⟨ f ( x ) , G z ( x , y ) g ( y ) ⟩ C N d y d x = ⟨ f , ( A − z ) − 1 g ⟩ L 2 ( Ω , C N ) = ⟨ ( A − z ‾ ) − 1 f , g ⟩ L 2 ( Ω , C N ) = ∫ Ω ∫ Ω ⟨ G z ‾ ( y , x ) f ( x ) , g ( y ) ⟩ C N d x d y = ∫ Ω ∫ Ω ⟨ f ( x ) , G z ‾ ( y , x ) ∗ g ( y ) ⟩ C N d x d y = ∫ Ω ∫ Ω ⟨ f ( x ) , G z ‾ ( y , x ) ∗ g ( y ) ⟩ C N d y d x , \begin{aligned}
\int_{\Omega}\int_{\Omega}\langle f(x),G_z(x,y)g(y)\rangle_{\mathbb{C}^N}\,\mathrm{d}y\,\mathrm{d}x
&=\langle f,(A-z)^{-1}g\rangle_{L^2(\Omega,\mathbb{C}^N)}\\
&=\langle(A-\overline{z})^{-1}f,g\rangle_{L^2(\Omega,\mathbb{C}^N)}\\
&=\int_{\Omega}\int_{\Omega}\langle G_{\overline{z}}(y,x)f(x),g(y)\rangle_{\mathbb{C}^N}\,\mathrm{d}x\,\mathrm{d}y\\
&=\int_{\Omega}\int_{\Omega}\langle f(x),G_{\overline{z}}(y,x)^*g(y)\rangle_{\mathbb{C}^N}\,\mathrm{d}x\,\mathrm{d}y\\
&=\int_{\Omega}\int_{\Omega}\langle f(x),G_{\overline{z}}(y,x)^*g(y)\rangle_{\mathbb{C}^N}\,\mathrm{d}y\,\mathrm{d}x,
\end{aligned} ∫ Ω ∫ Ω ⟨ f ( x ) , G z ( x , y ) g ( y ) ⟩ C N d y d x = ⟨ f , ( A − z ) − 1 g ⟩ L 2 ( Ω , C N ) = ⟨( A − z ) − 1 f , g ⟩ L 2 ( Ω , C N ) = ∫ Ω ∫ Ω ⟨ G z ( y , x ) f ( x ) , g ( y ) ⟩ C N d x d y = ∫ Ω ∫ Ω ⟨ f ( x ) , G z ( y , x ) ∗ g ( y ) ⟩ C N d x d y = ∫ Ω ∫ Ω ⟨ f ( x ) , G z ( y , x ) ∗ g ( y ) ⟩ C N d y d x ,
which yields G z ( x , y ) = G z ‾ ( y , x ) ∗ G_z(x,y)=G_{\overline{z}}(y,x)^* G z ( x , y ) = G z ( y , x ) ∗ for a.e. ( x , y ) (x,y) ( x , y ) , Moreover, in view of (2.23) and the equality E z ‾ ( y − x ) ∗ = E z ( x − y ) − Υ z ( x − y ) E_{\overline{z}}(y-x)^*=E_z(x-y)-\Upsilon_z(x-y) E z ( y − x ) ∗ = E z ( x − y ) − Υ z ( x − y ) , see (2.14), we also obtain
T z ( x , y ) = T z ‾ ( y , x ) ∗ − Υ z ( x − y ) (29) T_z(x,y)=T_{\overline{z}}(y,x)^*-\Upsilon_z(x-y) \tag*{(29)} T z ( x , y ) = T z ( y , x ) ∗ − Υ z ( x − y ) (29)
It remains to show the smoothness of T z T_z T z (which then also implies that (2.25) is valid for all ( x , y ) ∈ Ω × Ω (x,y)\in\Omega\times\Omega ( x , y ) ∈ Ω × Ω , i.e. (2.17) and hence also (2.18) hold): As this smoothness is of crucial importance for all subsequent constructions, we provide a detailed argument. To show the claim, we verify that all distributional derivatives of any component of T z T_z T z with respect to x x x - and y y y -variables are regular distributions, and then use an elliptic regularity argument to deduce the claimed smoothness.
Pick any j , k ∈ { 1 , … , N } j,k\in\{1,\ldots,N\} j , k ∈ { 1 , … , N } and set h : = ( T z ) j , k ∈ L l o c 2 ( Ω × Ω ) h:=(T_z)_{j,k}\in L^2_{\mathrm{loc}}(\Omega\times\Omega) h := ( T z ) j , k ∈ L loc 2 ( Ω × Ω ) and
H : Ω → L 2 ( Ω ) , H ( x ) : = h ( x , ⋅ ) . H:\Omega\to L^2(\Omega),\qquad H(x):=h(x,\cdot). H : Ω → L 2 ( Ω ) , H ( x ) := h ( x , ⋅ ) .
By Lemma 2.7, Lemma 2.6, and the smoothness of Υ z \Upsilon_z Υ z from Lemma 2.4, the map H H H is well-defined and H ∈ C ∞ ( Ω , L 2 ( Ω ) ) H\in C^\infty(\Omega,L^2(\Omega)) H ∈ C ∞ ( Ω , L 2 ( Ω )) . Let i ∈ { 1 , … , n } i\in\{1,\ldots,n\} i ∈ { 1 , … , n } and e i e_i e i be the i i i -th canonical basis vector in R n \mathbb{R}^n R n . Then for any φ ∈ C ∞ ( Ω × Ω ) \varphi\in C^\infty(\Omega\times\Omega) φ ∈ C ∞ ( Ω × Ω ) and Φ ∈ C ∞ ( Ω , L 2 ( Ω ) ) \Phi\in C^\infty(\Omega,L^2(\Omega)) Φ ∈ C ∞ ( Ω , L 2 ( Ω )) defined by Φ ( x ) : = φ ( x , ⋅ ) \Phi(x):=\varphi(x,\cdot) Φ ( x ) := φ ( x , ⋅ ) we have ∂ x i Φ ( x ) = ∂ x i φ ( x , ⋅ ) \partial_{x_i}\Phi(x)=\partial_{x_i}\varphi(x,\cdot) ∂ x i Φ ( x ) = ∂ x i φ ( x , ⋅ ) (see Lemma 2.6) and we compute using integration by parts (where d s \mathrm{d}s d s denotes the usual ( n − 1 ) (n-1) ( n − 1 ) -dimensional Hausdorff measure on ∂ Ω \partial\Omega ∂ Ω and ( ν ∂ Ω ) i (\nu^{\partial\Omega})_i ( ν ∂ Ω ) i is the i i i -th component of the unit normal field pointing outwards)
∫ Ω ∫ Ω ( ∂ x i H ( x ) ) ( y ) φ ( x , y ) ‾ d y d x = ∫ Ω ⟨ ∂ x i H ( x ) , Φ ( x ) ⟩ L 2 ( Ω ) d x = ∫ Ω ∂ x i ⟨ H ( x ) , Φ ( x ) ⟩ L 2 ( Ω ) d x − ∫ Ω ⟨ H ( x ) , ∂ x i Φ ( x ) ⟩ L 2 ( Ω ) d x = ∫ ∂ Ω ( ν ∂ Ω ) i ⟨ H ( x ) , Φ ( x ) ⟩ L 2 ( Ω ) d s ( x ) − ∫ Ω ⟨ H ( x ) , ∂ x i Φ ( x ) ⟩ L 2 ( Ω ) d x = − ∫ Ω ⟨ H ( x ) , ∂ x i Φ ( x ) ⟩ L 2 ( Ω ) d x = − ∫ Ω ∫ Ω h ( x , y ) ∂ x i φ ( x , y ) ‾ d y d x , (30) \begin{aligned}
\int_\Omega\int_\Omega(\partial_{x_i}H(x))(y)\overline{\varphi(x,y)}\,\mathrm{d}y\,\mathrm{d}x
&=\int_\Omega\langle\partial_{x_i}H(x),\Phi(x)\rangle_{L^2(\Omega)}\,\mathrm{d}x\\
&=\int_\Omega\partial_{x_i}\langle H(x),\Phi(x)\rangle_{L^2(\Omega)}\,\mathrm{d}x-\int_\Omega\langle H(x),\partial_{x_i}\Phi(x)\rangle_{L^2(\Omega)}\,\mathrm{d}x\\
&=\int_{\partial\Omega}(\nu^{\partial\Omega})_i\langle H(x),\Phi(x)\rangle_{L^2(\Omega)}\,\mathrm{d}s(x)-\int_\Omega\langle H(x),\partial_{x_i}\Phi(x)\rangle_{L^2(\Omega)}\,\mathrm{d}x \\
&=-\int_\Omega\langle H(x),\partial_{x_i}\Phi(x)\rangle_{L^2(\Omega)}\,\mathrm{d}x\\
&=-\int_\Omega\int_\Omega h(x,y)\overline{\partial_{x_i}\varphi(x,y)}\,\mathrm{d}y\,\mathrm{d}x, \tag*{(30)}
\end{aligned} ∫ Ω ∫ Ω ( ∂ x i H ( x )) ( y ) φ ( x , y ) d y d x = ∫ Ω ⟨ ∂ x i H ( x ) , Φ ( x ) ⟩ L 2 ( Ω ) d x = ∫ Ω ∂ x i ⟨ H ( x ) , Φ ( x ) ⟩ L 2 ( Ω ) d x − ∫ Ω ⟨ H ( x ) , ∂ x i Φ ( x ) ⟩ L 2 ( Ω ) d x = ∫ ∂ Ω ( ν ∂ Ω ) i ⟨ H ( x ) , Φ ( x ) ⟩ L 2 ( Ω ) d s ( x ) − ∫ Ω ⟨ H ( x ) , ∂ x i Φ ( x ) ⟩ L 2 ( Ω ) d x = − ∫ Ω ⟨ H ( x ) , ∂ x i Φ ( x ) ⟩ L 2 ( Ω ) d x = − ∫ Ω ∫ Ω h ( x , y ) ∂ x i φ ( x , y ) d y d x , (30)
where in the last but one step it was used that Φ ( x ) = 0 \Phi(x)=0 Φ ( x ) = 0 for all x ∈ ∂ Ω x\in\partial\Omega x ∈ ∂ Ω . Therefore, h h h admits a weak derivative with respect to x i x_i x i given by ∂ x i h ( x , y ) = ( ∂ x i H ( x ) ) ( y ) \partial_{x_i}h(x,y)=(\partial_{x_i}H(x))(y) ∂ x i h ( x , y ) = ( ∂ x i H ( x )) ( y ) . Since ∂ x i H ∈ C ∞ ( Ω , L 2 ( Ω ) ) \partial_{x_i}H\in C^\infty(\Omega,L^2(\Omega)) ∂ x i H ∈ C ∞ ( Ω , L 2 ( Ω )) one shows as in (2.24) that ∂ x i h ∈ L l o c 2 ( Ω × Ω ) \partial_{x_i}h\in L^2_{\mathrm{loc}}(\Omega\times\Omega) ∂ x i h ∈ L loc 2 ( Ω × Ω ) . By iterating this procedure we conclude ∂ x γ h ∈ L l o c 2 ( Ω × Ω ) \partial_x^\gamma h\in L^2_{\mathrm{loc}}(\Omega\times\Omega) ∂ x γ h ∈ L loc 2 ( Ω × Ω ) for any γ ∈ N 0 n \gamma\in\mathbb{N}_0^n γ ∈ N 0 n .
The preceding discussion shows that
∂ x γ ( T z ) j , k ∈ L l o c 2 ( Ω × Ω ) for all z ∈ C ∖ spec A , j , k ∈ { 1 , … , N } , γ ∈ N 0 n . \partial_x^\gamma(T_z)_{j,k}\in L^2_{\mathrm{loc}}(\Omega\times\Omega)\quad\text{for all }z\in\mathbb{C}\setminus\operatorname{spec}A,\ j,k\in\{1,\ldots,N\},\ \gamma\in\mathbb{N}_0^n. ∂ x γ ( T z ) j , k ∈ L loc 2 ( Ω × Ω ) for all z ∈ C ∖ spec A , j , k ∈ { 1 , … , N } , γ ∈ N 0 n .
At the same time, by (2.25) we have ( T z ) j , k ( x , y ) = ( T z ‾ ) k , j ( y , x ) − ( Υ z ) j , k ( x − y ) (T_z)_{j,k}(x,y)=(T_{\overline{z}})_{k,j}(y,x)-(\Upsilon_z)_{j,k}(x-y) ( T z ) j , k ( x , y ) = ( T z ) k , j ( y , x ) − ( Υ z ) j , k ( x − y ) ; as Υ z \Upsilon_z Υ z is smooth, the preceding discussion applied to T z ‾ T_{\overline{z}} T z yields
∂ y γ ( T z ) j , k ∈ L l o c 2 ( Ω × Ω ) for all z ∈ C ∖ spec A , j , k ∈ { 1 , … , N } , γ ∈ N 0 n . \partial_y^\gamma(T_z)_{j,k}\in L^2_{\mathrm{loc}}(\Omega\times\Omega)\quad\text{for all }z\in\mathbb{C}\setminus\operatorname{spec}A,\ j,k\in\{1,\ldots,N\},\ \gamma\in\mathbb{N}_0^n. ∂ y γ ( T z ) j , k ∈ L loc 2 ( Ω × Ω ) for all z ∈ C ∖ spec A , j , k ∈ { 1 , … , N } , γ ∈ N 0 n .
In particular, for any p ∈ N p\in\mathbb{N} p ∈ N and any j , k ∈ { 1 , … , N } j,k\in\{1,\ldots,N\} j , k ∈ { 1 , … , N } we have
[ ( Δ x ) p + ( Δ y ) p ] ( T z ) j , k ∈ L l o c 2 ( Ω × Ω ) , \left[(\Delta_x)^p+(\Delta_y)^p\right](T_z)_{j,k}\in L^2_{\mathrm{loc}}(\Omega\times\Omega), [ ( Δ x ) p + ( Δ y ) p ] ( T z ) j , k ∈ L loc 2 ( Ω × Ω ) ,
and, since the constant coefficient operator ( Δ x ) p + ( Δ y ) p (\Delta_x)^p + (\Delta_y)^p ( Δ x ) p + ( Δ y ) p is elliptic of order 2 p 2p 2 p in R 2 n \mathbb{R}^{2n} R 2 n , the interior elliptic regularity implies ( T z ) j , k ∈ H l o c 2 p ( Ω × Ω ) (T_z)_{j,k} \in H^{2p}_{\mathrm{loc}}(\Omega\times\Omega) ( T z ) j , k ∈ H loc 2 p ( Ω × Ω ) . As p ∈ N p \in\mathbb{N} p ∈ N is arbitrary, the Sobolev embedding theorem yields ( T z ) j , k ∈ C ∞ ( Ω × Ω ) (T_z)_{j,k} \in C^\infty(\Omega\times\Omega) ( T z ) j , k ∈ C ∞ ( Ω × Ω ) .
Finally, we provide an asymptotic expansion of the Green’s function E z E_z E z from (14) & (15) which will be useful for our further analysis. The technical proof is presented in Appendix A.
Lemma 2.9. Define the functions w ± : R n ∖ { 0 } → C N / 2 × N / 2 w^\pm:\mathbb{R}^n \setminus\{0\} \to\mathbb{C}^{N/2 \times N/2} w ± : R n ∖ { 0 } → C N /2 × N /2 by
w ± ( x ) = { i 2 π 1 x 1 ± i x 2 , n = 2 , i ( σ ⋅ x ) 4 π ∣ x ∣ 3 , n = 3 , (31) w^\pm(x) =
\begin{cases}
\dfrac{\mathrm{i}}{2\pi}\dfrac{1}{x_1 \pm\mathrm{i}x_2}, & n=2,\\
\dfrac{\mathrm{i}(\boldsymbol{\sigma}\cdot x)}{4\pi|x|^3}, & n=3,
\end{cases} \tag*{(31)} w ± ( x ) = ⎩ ⎨ ⎧ 2 π i x 1 ± i x 2 1 , 4 π ∣ x ∣ 3 i ( σ ⋅ x ) , n = 2 , n = 3 , (31) let E 0 E_0 E 0 be as in (13) , set for z ∈ C z \in\mathbb{C} z ∈ C
H z ( x ) = ( ( z + m ) E 0 ( x ) ⊗ I N / 2 w + ( x ) w − ( x ) ( z − m ) E 0 ( x ) ⊗ I N / 2 ) , x ∈ R n ∖ { 0 } , \mathcal{H}_z(x) =
\begin{pmatrix}
(z+m)E_0(x)\otimes I_{N/2} & w^+(x)\\
w^-(x) & (z-m)E_0(x)\otimes I_{N/2}
\end{pmatrix},
\qquad x\in\mathbb{R}^n\setminus\{0\}, H z ( x ) = ( ( z + m ) E 0 ( x ) ⊗ I N /2 w − ( x ) w + ( x ) ( z − m ) E 0 ( x ) ⊗ I N /2 ) , x ∈ R n ∖ { 0 } , and define with χ ∈ C c ∞ ( R n ) \chi\in C^\infty_c(\mathbb{R}^n) χ ∈ C c ∞ ( R n )
K z : = χ ( E z − H z ) . \mathcal{K}_z := \chi(E_z-\mathcal{H}_z). K z := χ ( E z − H z ) . Then, the following is true:
(a) K z ∈ L 1 ( R n , C N ) ∩ L ∞ ( R n , C N ) \mathcal{K}_z \in L^1(\mathbb{R}^n,\mathbb{C}^N)\cap L^\infty(\mathbb{R}^n,\mathbb{C}^N) K z ∈ L 1 ( R n , C N ) ∩ L ∞ ( R n , C N ) ;
(b) K z \mathcal{K}_z K z is smooth outside the origin and rapidly decreasing (with all of its derivatives) at infinity;
(c) K z \mathcal{K}_z K z has an expansion of the form
K z ( x ) ∼ ∑ ℓ ≥ 0 ( q ℓ ( x ) + p ℓ ( x ) log ∣ x ∣ ) , (32) \mathcal{K}_z(x)\sim\sum_{\ell\geq0}\left(q_\ell(x)+p_\ell(x)\log|x|\right), \tag*{(32)} K z ( x ) ∼ ℓ ≥ 0 ∑ ( q ℓ ( x ) + p ℓ ( x ) log ∣ x ∣ ) , (32) where each q ℓ q_\ell q ℓ is a homogeneous distribution of degree 3 + ℓ − n 3+\ell-n 3 + ℓ − n that is smooth outside the origin and p ℓ p_\ell p ℓ is a polynomial homogeneous of degree 3 + ℓ − n 3+\ell-n 3 + ℓ − n ;
(d) for n = 2 n=2 n = 2 the function K z \mathcal{K}_z K z is continuous and for every R > 0 R>0 R > 0 there is c > 0 c>0 c > 0 such that ∣ ∇ K z ( x ) ∣ ≤ c ( 1 + ∣ log ∣ x ∣ ∣ ) |\nabla\mathcal{K}_z(x)|\leq c(1+|\log|x||) ∣∇ K z ( x ) ∣ ≤ c ( 1 + ∣ log ∣ x ∣∣ ) for all 0 < ∣ x ∣ ≤ R 0<|x|\leq R 0 < ∣ x ∣ ≤ R ;
(e) for n = 3 n=3 n = 3 and every R > 0 R>0 R > 0 there is c > 0 c>0 c > 0 such that ∣ ∇ K z ( x ) ∣ ≤ c ∣ x ∣ − 1 |\nabla\mathcal{K}_z(x)|\leq c|x|^{-1} ∣∇ K z ( x ) ∣ ≤ c ∣ x ∣ − 1 for all 0 < ∣ x ∣ ≤ R 0<|x|\leq R 0 < ∣ x ∣ ≤ R .
The asymptotic expansion in (32) is understood in the following sense: For any L ∈ N 0 L\in\mathbb{N}_0 L ∈ N 0 there exists ℓ 0 ∈ N 0 \ell_0\in\mathbb{N}_0 ℓ 0 ∈ N 0 such that
K z ( x ) − ∑ ℓ = 0 ℓ 0 ( q ℓ ( x ) + p ℓ ( x ) log ∣ x ∣ ) ∈ C L ( R n ) . \mathcal{K}_z(x)-\sum_{\ell=0}^{\ell_0}\left(q_\ell(x)+p_\ell(x)\log|x|\right)\in C^L(\mathbb{R}^n). K z ( x ) − ℓ = 0 ∑ ℓ 0 ( q ℓ ( x ) + p ℓ ( x ) log ∣ x ∣ ) ∈ C L ( R n ) .
Dirac operator with a shell interaction on a bounded Lipschitz domain
In this section, we rigorously define the Dirac operator A η , τ A^{\eta,\tau} A η , τ with an electrostatic δ \delta δ -shell interaction of strength η \eta η and a Lorentz scalar δ \delta δ -shell interaction of strength τ \tau τ on a compact surface Σ ⊂ Ω \Sigma\subset\Omega Σ ⊂ Ω , as well as a self-adjoint boundary condition on ∂ Ω \partial\Omega ∂ Ω . First, we establish the geometric setup, function spaces, and trace operators along Σ \Sigma Σ . We then introduce the operator A η , τ A^{\eta,\tau} A η , τ by interpreting the singular interaction as a transmission condition. Finally, we formulate this setup within the abstract framework of boundary triples and Weyl functions, providing the essential foundation for the subsequent spectral analysis.
Geometric assumptions, function spaces, and trace operators
Let Σ ⊂ Ω \Sigma\subset\Omega Σ ⊂ Ω be a compact C ∞ C^\infty C ∞ -smooth hypersurface (i.e. a not necessarily connected curve for n = 2 n=2 n = 2 or surface for n = 3 n=3 n = 3 ) with finitely many connected components Σ p \Sigma_p Σ p , p ∈ { 1 , … , P } p \in\{1,\ldots,P\} p ∈ { 1 , … , P } . Due to the Jordan-Brouwer separation theorem, for each p p p one has R n ∖ Σ p = U p + ∪ U p − \mathbb{R}^n \setminus\Sigma_p = U_p^+ \cup U_p^- R n ∖ Σ p = U p + ∪ U p − , where U p + U_p^+ U p + and U p − U_p^- U p − are disjoint C ∞ C^\infty C ∞ -domains such that U p + U_p^+ U p + is bounded and U p − U_p^- U p − is unbounded. In the following we will only make use of the parts of U p ± U_p^\pm U p ± that belong to Ω \Omega Ω and we use the notation Ω p ± : = U p ± ∩ Ω \Omega_p^\pm:= U_p^\pm\cap\Omega Ω p ± := U p ± ∩ Ω . We denote by ν Σ \nu^\Sigma ν Σ the unit normal field on Σ \Sigma Σ which points towards Ω p − \Omega_p^- Ω p − on each Σ p \Sigma_p Σ p , i.e. ν Σ \nu^\Sigma ν Σ is the exterior unit normal for each Ω p + \Omega_p^+ Ω p + . For a function f ∈ L 2 ( Ω ) f \in L^2(\Omega) f ∈ L 2 ( Ω ) we use the notation f p ± f_{p\pm} f p ± for the restriction of f f f onto Ω p ± \Omega_p^\pm Ω p ± .
For s ∈ [ 0 , 1 ] s \in[0,1] s ∈ [ 0 , 1 ] consider the Sobolev-type spaces H α s ( Ω p ± , C N ) H_\alpha^s(\Omega_p^\pm,\mathbb{C}^N) H α s ( Ω p ± , C N ) defined as in (2.2) equipped with the corresponding Hilbert space norms, and let
H α s ( Ω ∖ Σ p , C N ) = H α s ( Ω p + , C N ) × H α s ( Ω p − , C N ) = { f ∈ H s ( Ω ∖ Σ p , C N ) : D 0 f ∈ D ′ ( Ω ∖ Σ p ) ∩ L 2 ( Ω , C N ) } , \begin{aligned}
H_\alpha^s(\Omega\setminus\Sigma_p,\mathbb{C}^N) = H_\alpha^s(\Omega_p^+,\mathbb{C}^N) \times H_\alpha^s(\Omega_p^-,\mathbb{C}^N) \\
&= \{f \in H^s(\Omega\setminus\Sigma_p,\mathbb{C}^N): D_0f \in\mathcal{D}'(\Omega\setminus\Sigma_p) \cap L^2(\Omega,\mathbb{C}^N)\},
\end{aligned} H α s ( Ω ∖ Σ p , C N ) = H α s ( Ω p + , C N ) × H α s ( Ω p − , C N ) = { f ∈ H s ( Ω ∖ Σ p , C N ) : D 0 f ∈ D ′ ( Ω ∖ Σ p ) ∩ L 2 ( Ω , C N )} ,
where D 0 f ∈ D ′ ( Ω ∖ Σ p ) D_0f \in\mathcal{D}'(\Omega\setminus\Sigma_p) D 0 f ∈ D ′ ( Ω ∖ Σ p ) is the distribution
C c ∞ ( Ω ∖ Σ p , C N ) ∋ φ ↦ ⟨ D 0 f , φ ⟩ D ′ ( Ω ∖ Σ p , C N ) , D ( Ω ∖ Σ p , C N ) , C_c^\infty(\Omega\setminus\Sigma_p,\mathbb{C}^N) \ni\varphi\mapsto\langle D_0f,\varphi\rangle_{\mathcal{D}'(\Omega\setminus\Sigma_p,\mathbb{C}^N),\mathcal{D}(\Omega\setminus\Sigma_p,\mathbb{C}^N)}, C c ∞ ( Ω ∖ Σ p , C N ) ∋ φ ↦ ⟨ D 0 f , φ ⟩ D ′ ( Ω ∖ Σ p , C N ) , D ( Ω ∖ Σ p , C N ) ,
and the requirement D 0 f ∈ D ′ ( Ω ∖ Σ p ) ∩ L 2 ( Ω , C N ) D_0f \in\mathcal{D}'(\Omega\setminus\Sigma_p) \cap L^2(\Omega,\mathbb{C}^N) D 0 f ∈ D ′ ( Ω ∖ Σ p ) ∩ L 2 ( Ω , C N ) means that this distribution is represented by a function g ∈ L 2 ( Ω , C N ) g \in L^2(\Omega,\mathbb{C}^N) g ∈ L 2 ( Ω , C N ) , i.e.
⟨ D 0 f , φ ⟩ D ′ ( Ω ∖ Σ p , C N ) , D ( Ω ∖ Σ p , C N ) = ∫ Ω g ⋅ φ d x , φ ∈ C c ∞ ( Ω ∖ Σ p , C N ) ; \langle D_0f,\varphi\rangle_{\mathcal{D}'(\Omega\setminus\Sigma_p,\mathbb{C}^N),\mathcal{D}(\Omega\setminus\Sigma_p,\mathbb{C}^N)}
= \int_\Omega g \cdot\varphi\,\mathrm{d}x,\qquad\varphi\in C_c^\infty(\Omega\setminus\Sigma_p,\mathbb{C}^N); ⟨ D 0 f , φ ⟩ D ′ ( Ω ∖ Σ p , C N ) , D ( Ω ∖ Σ p , C N ) = ∫ Ω g ⋅ φ d x , φ ∈ C c ∞ ( Ω ∖ Σ p , C N ) ;
we will again denote this representing function by D 0 f D_0f D 0 f , i.e. D 0 f = g D_0f=g D 0 f = g . In particular, elliptic regularity implies that f ∈ H α s ( Ω ∖ Σ p , C N ) f \in H_\alpha^s(\Omega\setminus\Sigma_p,\mathbb{C}^N) f ∈ H α s ( Ω ∖ Σ p , C N ) is locally H 1 H^1 H 1 -regular in Ω ∖ Σ p \Omega\setminus\Sigma_p Ω ∖ Σ p (but not in all of Ω \Omega Ω ). As Σ p \Sigma_p Σ p is C ∞ C^\infty C ∞ -smooth it follows from [12 Lemma 4.1 and Corollary 4.6] that there exist unique continuous trace operators
γ ± Σ p : H α s ( Ω ∖ Σ p , C N ) ⟶ H s − 1 2 ( Σ p , C N ) (33) \gamma_{\pm}^{\Sigma_p}: H_\alpha^s(\Omega\setminus\Sigma_p,\mathbb{C}^N) \longrightarrow H^{s-\frac{1}{2}}(\Sigma_p,\mathbb{C}^N) \tag*{(33)} γ ± Σ p : H α s ( Ω ∖ Σ p , C N ) ⟶ H s − 2 1 ( Σ p , C N ) (33)
such that for f = ( f p + , f p − ) ∈ H α s ( Ω ∖ Σ p , C N ) f=(f_{p+},f_{p-}) \in H_\alpha^s(\Omega\setminus\Sigma_p,\mathbb{C}^N) f = ( f p + , f p − ) ∈ H α s ( Ω ∖ Σ p , C N ) one has γ + Σ p f = f p + ∣ Σ p \gamma_+^{\Sigma_p}f=f_{p+}|_{\Sigma_p} γ + Σ p f = f p + ∣ Σ p if f p + ∈ C ∞ ( Ω p + ‾ ) f_{p+}\in C^\infty(\overline{\Omega_p^+}) f p + ∈ C ∞ ( Ω p + ) , and γ − Σ p f = f p − ∣ Σ p \gamma_-^{\Sigma_p}f=f_{p-}|_{\Sigma_p} γ − Σ p f = f p − ∣ Σ p if f p − ∈ C ∞ ( Ω p − ‾ ) f_{p-}\in C^\infty(\overline{\Omega_p^-}) f p − ∈ C ∞ ( Ω p − ) .
In the following we work with traces of functions having different Sobolev regularity properties near Σ = ⋃ p = 1 P Σ p \Sigma=\bigcup_{p=1}^P\Sigma_p Σ = ⋃ p = 1 P Σ p and near ∂ Ω \partial\Omega ∂ Ω . To take these aspects into account we first define (in the same way as above) for s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] the Sobolev-type spaces
H α s ( Ω ∖ Σ , C N ) = { f ∈ H s ( Ω ∖ Σ , C N ) : D 0 f ∈ D ′ ( Ω ∖ Σ ) ∩ L 2 ( Ω , C N ) } , (34) H_\alpha^s(\Omega\setminus\Sigma,\mathbb{C}^N)
=\{f\in H^s(\Omega\setminus\Sigma,\mathbb{C}^N):D_0f\in\mathcal{D}'(\Omega\setminus\Sigma)\cap L^2(\Omega,\mathbb{C}^N)\}, \tag*{(34)} H α s ( Ω ∖ Σ , C N ) = { f ∈ H s ( Ω ∖ Σ , C N ) : D 0 f ∈ D ′ ( Ω ∖ Σ ) ∩ L 2 ( Ω , C N )} , (34)
and the function spaces
H α s , Σ ( Ω ∖ Σ , C N ) : = { f ∈ L 2 ( Ω , C N ) : χ f ∈ H α s ( Ω ∖ Σ , C N ) for any χ ∈ C c ∞ ( Ω ) } (35) H_\alpha^{s,\Sigma}(\Omega\setminus\Sigma,\mathbb{C}^N)
:=\{f\in L^2(\Omega,\mathbb{C}^N):\chi f\in H_\alpha^s(\Omega\setminus\Sigma,\mathbb{C}^N)\text{ for any }\chi\in C_c^\infty(\Omega)\} \tag*{(35)} H α s , Σ ( Ω ∖ Σ , C N ) := { f ∈ L 2 ( Ω , C N ) : χ f ∈ H α s ( Ω ∖ Σ , C N ) for any χ ∈ C c ∞ ( Ω )} (35)
and
H α s , ∂ Ω ( Ω ∖ Σ , C N ) : = { f ∈ L 2 ( Ω , C N ) : χ f ∈ H α s ( Ω ∖ Σ , C N ) H_\alpha^{s,\partial\Omega}(\Omega\setminus\Sigma,\mathbb{C}^N)
:=\{f\in L^2(\Omega,\mathbb{C}^N):\chi f\in H_\alpha^s(\Omega\setminus\Sigma,\mathbb{C}^N) H α s , ∂ Ω ( Ω ∖ Σ , C N ) := { f ∈ L 2 ( Ω , C N ) : χ f ∈ H α s ( Ω ∖ Σ , C N )
for any χ ∈ C ∞ ( Ω ‾ ) with χ = 0 near Σ } . (36) \text{for any }\chi\in C^\infty(\overline{\Omega})\text{ with }\chi=0\text{ near }\Sigma\}. \tag*{(36)} for any χ ∈ C ∞ ( Ω ) with χ = 0 near Σ } . (36)
Roughly speaking, the functions in H α s , Σ ( Ω ∖ Σ , C N ) H^{s,\Sigma}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) H α s , Σ ( Ω ∖ Σ , C N ) are locally H 1 H^{1} H 1 -regular in Ω ∖ Σ \Omega\setminus\Sigma Ω ∖ Σ (since D 0 ( χ f ) ∈ D ′ ( Ω ∖ Σ ) ∩ L 2 ( Ω , C N ) D_{0}(\chi f) \in\mathcal{D}'(\Omega\setminus\Sigma) \cap L^{2}(\Omega,\mathbb{C}^{N}) D 0 ( χ f ) ∈ D ′ ( Ω ∖ Σ ) ∩ L 2 ( Ω , C N ) ) and H s H^{s} H s -regular near Σ \Sigma Σ . Similarly, the functions in H α s , ∂ Ω ( Ω ∖ Σ , C N ) H^{s,\partial\Omega}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) H α s , ∂ Ω ( Ω ∖ Σ , C N ) are locally H 1 H^{1} H 1 -regular in Ω ∖ Σ \Omega\setminus\Sigma Ω ∖ Σ and H s H^{s} H s -regular near ∂ Ω \partial\Omega ∂ Ω , but since for g ∈ H α s ( Ω ∖ Σ , C N ) g \in H^{s}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) g ∈ H α s ( Ω ∖ Σ , C N ) and χ ∈ C ∞ ( Ω ‾ ) \chi\in C^{\infty}(\overline{\Omega}) χ ∈ C ∞ ( Ω ) with χ = 0 \chi= 0 χ = 0 near Σ \Sigma Σ one always has χ g ∈ H α s ( Ω ) \chi g \in H^{s}_{\alpha}(\Omega) χ g ∈ H α s ( Ω ) , no statement about the regularity of f ∈ H α s , ∂ Ω ( Ω ∖ Σ , C N ) f \in H^{s,\partial\Omega}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) f ∈ H α s , ∂ Ω ( Ω ∖ Σ , C N ) near Σ \Sigma Σ can be made. Making use of the traces γ ± Σ p \gamma^{\Sigma_{p}}_{\pm} γ ± Σ p at Σ p \Sigma_{p} Σ p in (3.1) it is clear that there exist unique continuous trace operators
γ ± Σ : H α s , Σ ( Ω ∖ Σ , C N ) ↦ H s − 1 2 ( Σ , C N ) , γ ± Σ f ∣ Σ p = γ ± Σ p ( χ p f p ± ) , p ∈ { 1 , … , P } , (37) \gamma^{\Sigma}_{\pm}: H^{s,\Sigma}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) \mapsto H^{s-\frac{1}{2}}(\Sigma,\mathbb{C}^{N}), \qquad\gamma^{\Sigma}_{\pm}f|_{\Sigma_{p}} = \gamma^{\Sigma_{p}}_{\pm}(\chi_{p}f_{p\pm}), \quad p \in\{1,\ldots,P\}, \tag*{(37)} γ ± Σ : H α s , Σ ( Ω ∖ Σ , C N ) ↦ H s − 2 1 ( Σ , C N ) , γ ± Σ f ∣ Σ p = γ ± Σ p ( χ p f p ± ) , p ∈ { 1 , … , P } , (37)
where χ p ∈ C ∞ ( Ω ‾ ) \chi_{p} \in C^{\infty}(\overline{\Omega}) χ p ∈ C ∞ ( Ω ) with χ p = 1 \chi_{p}=1 χ p = 1 near Σ p \Sigma_{p} Σ p and χ p = 0 \chi_{p}=0 χ p = 0 near ⋃ j ≠ p Σ j \bigcup_{j\ne p}\Sigma_{j} ⋃ j = p Σ j . On the Lipschitz boundary ∂ Ω \partial\Omega ∂ Ω we choose s = 1 2 s=\frac{1}{2} s = 2 1 and observe that there is a unique continuous trace operator
γ ∂ Ω : H α 1 2 , ∂ Ω ( Ω ∖ Σ , C N ) ↦ L 2 ( ∂ Ω , C N ) (38) \gamma^{\partial\Omega}: H^{\frac{1}{2},\partial\Omega}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) \mapsto L^{2}(\partial\Omega,\mathbb{C}^{N}) \tag*{(38)} γ ∂ Ω : H α 2 1 , ∂ Ω ( Ω ∖ Σ , C N ) ↦ L 2 ( ∂ Ω , C N ) (38)
that acts on f ∈ H α 1 2 , ∂ Ω ( Ω ∖ Σ , C N ) f \in H^{\frac{1}{2},\partial\Omega}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) f ∈ H α 2 1 , ∂ Ω ( Ω ∖ Σ , C N ) in the same way as the trace operator in Lemma 2.1 when applied to χ ∂ Ω f \chi_{\partial\Omega}f χ ∂ Ω f , where χ ∂ Ω ∈ C ∞ ( Ω ‾ ) \chi_{\partial\Omega} \in C^{\infty}(\overline{\Omega}) χ ∂ Ω ∈ C ∞ ( Ω ) with χ ∂ Ω = 1 \chi_{\partial\Omega}=1 χ ∂ Ω = 1 near ∂ Ω \partial\Omega ∂ Ω and χ ∂ Ω = 0 \chi_{\partial\Omega}=0 χ ∂ Ω = 0 near Σ \Sigma Σ (so that χ ∂ Ω f ∈ H α 1 2 ( Ω , C N ) \chi_{\partial\Omega}f \in H^{\frac{1}{2}}_{\alpha}(\Omega,\mathbb{C}^{N}) χ ∂ Ω f ∈ H α 2 1 ( Ω , C N ) ); for the trace operator in (3.6) we use the same symbol as in Lemma 2.1 and it will be clear from the context on which space the trace operator γ ∂ Ω \gamma^{\partial\Omega} γ ∂ Ω is considered.
Sometimes we will also work with locally H 1 H^{1} H 1 -regular functions in Ω ∖ Σ \Omega\setminus\Sigma Ω ∖ Σ which are H s H^{s} H s -regular near Σ p \Sigma_{p} Σ p for some p ∈ { 1 , … , P } p \in\{1,\ldots,P\} p ∈ { 1 , … , P } and s ∈ [ 0 , 1 ] s \in[0,1] s ∈ [ 0 , 1 ] . Thus, we consider the spaces
H α s , Σ p ( Ω ∖ Σ , C N ) : = { f ∈ L 2 ( Ω , C N ) : χ f ∈ H α s ( Ω ∖ Σ p , C N ) for any χ ∈ C c ∞ ( Ω ∖ ⋃ j ≠ p Σ j ) } . (39) H^{s,\Sigma_{p}}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) := \left\{ f \in L^{2}(\Omega,\mathbb{C}^{N}) : \chi f \in H^{s}_{\alpha}(\Omega\setminus\Sigma_{p},\mathbb{C}^{N}) \text{ for any } \chi\in C^{\infty}_{c}\left(\Omega\setminus\bigcup_{j\ne p}\Sigma_{j}\right) \right\}. \tag*{(39)} H α s , Σ p ( Ω ∖ Σ , C N ) := ⎩ ⎨ ⎧ f ∈ L 2 ( Ω , C N ) : χ f ∈ H α s ( Ω ∖ Σ p , C N ) for any χ ∈ C c ∞ Ω ∖ j = p ⋃ Σ j ⎭ ⎬ ⎫ . (39)
Note that the functions in H α s , Σ p ( Ω ∖ Σ , C N ) H^{s,\Sigma_{p}}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) H α s , Σ p ( Ω ∖ Σ , C N ) are locally H 1 H^{1} H 1 -regular in Ω ∖ Σ \Omega\setminus\Sigma Ω ∖ Σ and H s H^{s} H s -regular near Σ p \Sigma_{p} Σ p , but since for g ∈ H α s ( Ω ∖ Σ , C N ) g \in H^{s}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) g ∈ H α s ( Ω ∖ Σ , C N ) and χ ∈ C c ∞ ( Ω ∖ ⋃ j ≠ p Σ j ) \chi\in C^{\infty}_{c}\left(\Omega\setminus\bigcup_{j\ne p}\Sigma_{j}\right) χ ∈ C c ∞ ( Ω ∖ ⋃ j = p Σ j ) one always has χ g ∈ H α s ( Ω ∖ Σ p , C N ) \chi g \in H^{s}_{\alpha}(\Omega\setminus\Sigma_{p},\mathbb{C}^{N}) χ g ∈ H α s ( Ω ∖ Σ p , C N ) , no statement about the regularity of f ∈ H α s , Σ p ( Ω ∖ Σ , C N ) f \in H^{s,\Sigma_{p}}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) f ∈ H α s , Σ p ( Ω ∖ Σ , C N ) near ⋃ j ≠ p Σ j \bigcup_{j\ne p}\Sigma_{j} ⋃ j = p Σ j (and ∂ Ω \partial\Omega ∂ Ω ) can be made. Observe that
H α s , Σ ( Ω ∖ Σ , C N ) = ⋂ p = 1 P H α s , Σ p ( Ω ∖ Σ , C N ) . H^{s,\Sigma}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) = \bigcap_{p=1}^{P} H^{s,\Sigma_{p}}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}). H α s , Σ ( Ω ∖ Σ , C N ) = p = 1 ⋂ P H α s , Σ p ( Ω ∖ Σ , C N ) .
For each p ∈ { 1 , … , P } p \in\{1,\ldots,P\} p ∈ { 1 , … , P } there exist unique continuous trace operators
γ ± Σ p : H α s , Σ p ( Ω ∖ Σ , C N ) ↦ H s − 1 2 ( Σ p , C N ) , (40) \gamma^{\Sigma_{p}}_{\pm}: H^{s,\Sigma_{p}}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) \mapsto H^{s-\frac{1}{2}}(\Sigma_{p},\mathbb{C}^{N}), \tag*{(40)} γ ± Σ p : H α s , Σ p ( Ω ∖ Σ , C N ) ↦ H s − 2 1 ( Σ p , C N ) , (40)
where we use the same symbol as in (3.1) to avoid further technicalities. We note for completeness that the action of (3.8) on f ∈ H α s , Σ p ( Ω ∖ Σ , C N ) f \in H^{s,\Sigma_{p}}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) f ∈ H α s , Σ p ( Ω ∖ Σ , C N ) is defined as the action of the trace operator in (3.1) applied to χ p f \chi_{p}f χ p f , where χ p ∈ C c ∞ ( Ω ) \chi_{p} \in C^{\infty}_{c}(\Omega) χ p ∈ C c ∞ ( Ω ) with χ p = 1 \chi_{p}=1 χ p = 1 near Σ p \Sigma_{p} Σ p and χ p = 0 \chi_{p}=0 χ p = 0 near ∂ Ω \partial\Omega ∂ Ω and ⋃ j ≠ p Σ j \bigcup_{j\ne p}\Sigma_{j} ⋃ j = p Σ j (so that χ p f ∈ H α s ( Ω ∖ Σ p , C N ) \chi_{p}f \in H^{s}_{\alpha}(\Omega\setminus\Sigma_{p},\mathbb{C}^{N}) χ p f ∈ H α s ( Ω ∖ Σ p , C N ) ). It is also clear that the trace operator in (3.5) applied to functions in H α s , Σ ( Ω ∖ Σ , C N ) H^{s,\Sigma}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) H α s , Σ ( Ω ∖ Σ , C N ) and restricted to Σ p \Sigma_{p} Σ p coincides with (3.8).
The next lemma is essentially a consequence of [15 Lemma 2.4] for n = 2 n=2 n = 2 and [36 Proposition 2.16] for n = 3 n=3 n = 3 .
Lemma 3.1. Let p ∈ { 1 , … , P } p \in\{1,\ldots,P\} p ∈ { 1 , … , P } and f ∈ H α 0 , Σ p ( Ω ∖ Σ , C N ) f \in H_{\alpha}^{0,\Sigma_p}(\Omega\setminus\Sigma,\mathbb{C}^{N}) f ∈ H α 0 , Σ p ( Ω ∖ Σ , C N ) . If γ ± Σ p f ∈ H 1 2 ( Σ p , C N ) \gamma_{\pm}^{\Sigma_p} f \in H^{\frac{1}{2}}(\Sigma_p,\mathbb{C}^{N}) γ ± Σ p f ∈ H 2 1 ( Σ p , C N ) , then f ∈ H α 1 , Σ p ( Ω ∖ Σ , C N ) f \in H_{\alpha}^{1,\Sigma_p}(\Omega\setminus\Sigma,\mathbb{C}^{N}) f ∈ H α 1 , Σ p ( Ω ∖ Σ , C N ) .
Finally, we introduce one more trace operator that acts on the domain of the self-adjoint Dirac operator A A A in (8) ,
γ Σ : dom A = { f ∈ H α 1 2 ( Ω , C N ) : B ∂ Ω γ ∂ Ω f = 0 } → H 1 2 ( Σ , C N ) , (41) \gamma^{\Sigma}:\operatorname{dom} A = \{f \in H_{\alpha}^{\frac{1}{2}}(\Omega,\mathbb{C}^{N}): B^{\partial\Omega}\gamma^{\partial\Omega}f=0\} \to H^{\frac{1}{2}}(\Sigma,\mathbb{C}^{N}), \tag*{(41)} γ Σ : dom A = { f ∈ H α 2 1 ( Ω , C N ) : B ∂ Ω γ ∂ Ω f = 0 } → H 2 1 ( Σ , C N ) , (41)
which is well defined since dom A ⊂ H l o c 1 ( Ω , C N ) \operatorname{dom} A \subset H_{\mathrm{loc}}^{1}(\Omega,\mathbb{C}^{N}) dom A ⊂ H loc 1 ( Ω , C N ) ; cf. Lemma 2.3.
Lemma 3.2. The trace operator (41) is surjective and bounded with respect to the graph norm of A A A .
Proof. Let U U U be a smooth open neighborhood of Σ \Sigma Σ with U ‾ ⊂ Ω \overline{U} \subset\Omega U ⊂ Ω . Then U U U is an extension domain, and each function in H 1 ( U , C N ) H^{1}(U,\mathbb{C}^{N}) H 1 ( U , C N ) can be extended to a function in H 0 1 ( Ω , C N ) ⊂ dom A H_{0}^{1}(\Omega,\mathbb{C}^{N}) \subset\operatorname{dom} A H 0 1 ( Ω , C N ) ⊂ dom A . It follows that the map
dom A ∋ f ↦ f ∣ U ∈ H 1 ( U , C N ) \operatorname{dom} A \ni f \mapsto f|_{U} \in H^{1}(U,\mathbb{C}^{N}) dom A ∋ f ↦ f ∣ U ∈ H 1 ( U , C N )
is surjective. Furthermore, the map
H 1 ( U , C N ) ∋ f ↦ f ∣ Σ ∈ H 1 2 ( Σ , C N ) H^{1}(U,\mathbb{C}^{N}) \ni f \mapsto f|_{\Sigma} \in H^{\frac{1}{2}}(\Sigma,\mathbb{C}^{N}) H 1 ( U , C N ) ∋ f ↦ f ∣ Σ ∈ H 2 1 ( Σ , C N )
is surjective due to the standard Sobolev trace theorem. Hence, the trace operator γ Σ \gamma^{\Sigma} γ Σ in (41) is surjective being the composition of these two maps.
In order to show the boundedness let us pick a function χ ∈ C ∞ ( Ω ) \chi\in C^{\infty}(\Omega) χ ∈ C ∞ ( Ω ) with χ = 1 \chi=1 χ = 1 on U U U , then for any f ∈ dom A f \in\operatorname{dom} A f ∈ dom A one has χ f ∈ H 0 1 ( Ω , C N ) \chi f \in H_{0}^{1}(\Omega,\mathbb{C}^{N}) χ f ∈ H 0 1 ( Ω , C N ) and hence the 0 0 0 -extension onto R n \mathbb{R}^{n} R n satisfies χ f ~ ∈ H 1 ( R n , C N ) \widetilde{\chi f} \in H^{1}(\mathbb{R}^{n},\mathbb{C}^{N}) χ f ∈ H 1 ( R n , C N ) . Due to the Sobolev trace theorem there is a c 1 > 0 c_{1}>0 c 1 > 0 such that
∥ γ Σ g ∥ H 1 2 ( Σ , C N ) ≤ c 1 ∥ g ∥ H 1 ( U , C N ) for all g ∈ H 1 ( U , C N ) . \|\gamma^{\Sigma}g\|_{H^{\frac{1}{2}}(\Sigma,\mathbb{C}^{N})} \le c_{1}\|g\|_{H^{1}(U,\mathbb{C}^{N})}\text{ for all }g \in H^{1}(U,\mathbb{C}^{N}). ∥ γ Σ g ∥ H 2 1 ( Σ , C N ) ≤ c 1 ∥ g ∥ H 1 ( U , C N ) for all g ∈ H 1 ( U , C N ) .
The uniform ellipticity of the Dirac operator in R n \mathbb{R}^{n} R n implies that there is a c 2 > 0 c_{2}>0 c 2 > 0 such that for any h ∈ H 1 ( R n , C N ) h \in H^{1}(\mathbb{R}^{n},\mathbb{C}^{N}) h ∈ H 1 ( R n , C N ) one has
∥ h ∥ H 1 ( R n , C N ) ≤ c 2 ( ∥ h ∥ L 2 ( R n , C N ) + ∥ D h ∥ L 2 ( R n , C N ) ) . \|h\|_{H^{1}(\mathbb{R}^{n},\mathbb{C}^{N})} \le c_{2}\left(\|h\|_{L^{2}(\mathbb{R}^{n},\mathbb{C}^{N})}+\|Dh\|_{L^{2}(\mathbb{R}^{n},\mathbb{C}^{N})}\right). ∥ h ∥ H 1 ( R n , C N ) ≤ c 2 ( ∥ h ∥ L 2 ( R n , C N ) + ∥ D h ∥ L 2 ( R n , C N ) ) .
Then for any f ∈ dom A f \in\operatorname{dom} A f ∈ dom A one has
∥ γ Σ f ∥ H 1 2 ( Σ , C N ) ≤ c 1 ∥ f ∣ U ∥ H 1 ( U , C N ) ≤ c 1 ∥ χ f ~ ∥ H 1 ( R n , C N ) ≤ c 1 c 2 ( ∥ χ f ~ ∥ L 2 ( R n , C N ) + ∥ D ( χ f ~ ) ∥ L 2 ( R n , C N ) ) = c 1 c 2 ( ∥ χ f ∥ L 2 ( Ω , C N ) + ∥ D ( χ f ) ∥ L 2 ( Ω , C N ) ) = c 1 c 2 ( ∥ χ f ∥ L 2 ( Ω , C N ) + ∥ χ D f − i ( α ⋅ ∇ χ ) f ∥ L 2 ( Ω , C N ) ) ≤ c 1 c 2 c 3 ( ∥ f ∥ L 2 ( Ω , C N ) + ∥ D f ∥ L 2 ( Ω , C N ) ) \begin{aligned}
\|\gamma^{\Sigma}f\|_{H^{\frac{1}{2}}(\Sigma,\mathbb{C}^{N})}
&\le c_{1}\|f|_{U}\|_{H^{1}(U,\mathbb{C}^{N})}
\le c_{1}\|\widetilde{\chi f}\|_{H^{1}(\mathbb{R}^{n},\mathbb{C}^{N})}\\
&\le c_{1}c_{2}\left(\|\widetilde{\chi f}\|_{L^{2}(\mathbb{R}^{n},\mathbb{C}^{N})}+\|D(\widetilde{\chi f})\|_{L^{2}(\mathbb{R}^{n},\mathbb{C}^{N})}\right)\\
&=c_{1}c_{2}\left(\|\chi f\|_{L^{2}(\Omega,\mathbb{C}^{N})}+\|D(\chi f)\|_{L^{2}(\Omega,\mathbb{C}^{N})}\right)\\
&=c_{1}c_{2}\left(\|\chi f\|_{L^{2}(\Omega,\mathbb{C}^{N})}+\|\chi Df-i(\alpha\cdot\nabla\chi)f\|_{L^{2}(\Omega,\mathbb{C}^{N})}\right)\\
&\le c_{1}c_{2}c_{3}\left(\|f\|_{L^{2}(\Omega,\mathbb{C}^{N})}+\|Df\|_{L^{2}(\Omega,\mathbb{C}^{N})}\right)
\end{aligned} ∥ γ Σ f ∥ H 2 1 ( Σ , C N ) ≤ c 1 ∥ f ∣ U ∥ H 1 ( U , C N ) ≤ c 1 ∥ χ f ∥ H 1 ( R n , C N ) ≤ c 1 c 2 ( ∥ χ f ∥ L 2 ( R n , C N ) + ∥ D ( χ f ) ∥ L 2 ( R n , C N ) ) = c 1 c 2 ( ∥ χ f ∥ L 2 ( Ω , C N ) + ∥ D ( χ f ) ∥ L 2 ( Ω , C N ) ) = c 1 c 2 ( ∥ χ f ∥ L 2 ( Ω , C N ) + ∥ χD f − i ( α ⋅ ∇ χ ) f ∥ L 2 ( Ω , C N ) ) ≤ c 1 c 2 c 3 ( ∥ f ∥ L 2 ( Ω , C N ) + ∥ D f ∥ L 2 ( Ω , C N ) )
for c 3 : = ∥ χ ∥ ∞ + ∥ α ⋅ ∇ χ ∥ ∞ c_{3}:=\|\chi\|_{\infty}+\|\alpha\cdot\nabla\chi\|_{\infty} c 3 := ∥ χ ∥ ∞ + ∥ α ⋅ ∇ χ ∥ ∞ , which gives the sought claim. □
The operator A η , τ A^{\eta,\tau} A η , τ
Let Σ \Sigma Σ and Σ p \Sigma_p Σ p , p ∈ { 1 , … , P } p \in\{1,\ldots,P\} p ∈ { 1 , … , P } , be as in the beginning of the previous subsection, and let η , τ : Σ → R \eta,\tau:\Sigma\to\mathbb{R} η , τ : Σ → R be functions that are constant on each Σ p \Sigma_p Σ p . We denote
η p : = η ∣ Σ p ∈ R and τ p : = τ ∣ Σ p ∈ R , p ∈ { 1 , … , P } . \eta_p:=\eta|_{\Sigma_p}\in\mathbb{R}\qquad\text{and}\qquad\tau_p:=\tau|_{\Sigma_p}\in\mathbb{R},\quad p\in\{1,\ldots,P\}. η p := η ∣ Σ p ∈ R and τ p := τ ∣ Σ p ∈ R , p ∈ { 1 , … , P } .
Our next goal is to rigorously define a Dirac operator in Ω \Omega Ω with the same elliptic boundary condition B ∂ Ω γ f = 0 B_{\partial\Omega}^{\gamma}f=0 B ∂ Ω γ f = 0 on ∂ Ω \partial\Omega ∂ Ω as A A A in (8) , an electrostatic δ \delta δ -shell interaction of strength η \eta η and a Lorentz scalar δ \delta δ -shell interaction of strength τ \tau τ , i.e. a Dirac operator acting as
f ↦ D f in D ′ ( Ω ∖ Σ , C N ) ∩ L 2 ( Ω , C N ) f\mapsto Df\text{ in }D'(\Omega\setminus\Sigma,\mathbb{C}^{N})\cap L^2(\Omega,\mathbb{C}^{N}) f ↦ D f in D ′ ( Ω ∖ Σ , C N ) ∩ L 2 ( Ω , C N )
on the functions f f f satisfying the same boundary condition on ∂ Ω \partial\Omega ∂ Ω as A A A and the transmission condition
i ( α ⋅ ν Σ ) ( f p + − f p − ) + 1 2 ( η p I N + τ p β ) ( f p + + f p − ) = 0 on each Σ p . i(\alpha\cdot\nu^\Sigma)(f_{p+}-f_{p-})+\frac{1}{2}(\eta_p I_N+\tau_p\beta)(f_{p+}+f_{p-})=0\text{ on each }\Sigma_p. i ( α ⋅ ν Σ ) ( f p + − f p − ) + 2 1 ( η p I N + τ p β ) ( f p + + f p − ) = 0 on each Σ p .
Definition 3.3. The operator A η , τ A^{\eta,\tau} A η , τ in L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^{N}) L 2 ( Ω , C N ) is defined as
A η , τ f = D f , A^{\eta,\tau}f=Df, A η , τ f = D f ,
dom A η , τ = { f ∈ H α 0 , Σ ( Ω ∖ Σ , C N ) ∩ H α 1 2 , ∂ Ω ( Ω ∖ Σ , C N ) : B ∂ Ω γ f = 0 on ∂ Ω , \operatorname{dom}A^{\eta,\tau}=\left\{f\in H_{\alpha}^{0,\Sigma}(\Omega\setminus\Sigma,\mathbb{C}^{N})\cap H_{\alpha}^{\frac{1}{2},\partial\Omega}(\Omega\setminus\Sigma,\mathbb{C}^{N}):B_{\partial\Omega}^{\gamma}f=0\text{ on }\partial\Omega,
\right. dom A η , τ = { f ∈ H α 0 , Σ ( Ω ∖ Σ , C N ) ∩ H α 2 1 , ∂ Ω ( Ω ∖ Σ , C N ) : B ∂ Ω γ f = 0 on ∂ Ω ,
i ( α ⋅ ν Σ ) ( γ + Σ f − γ − Σ f ) + 1 2 ( η I N + τ β ) ( γ + Σ f + γ − Σ f ) = 0 on Σ } , (42) \left.
i(\alpha\cdot\nu^\Sigma)(\gamma_+^\Sigma f-\gamma_-^\Sigma f)+\frac{1}{2}(\eta I_N+\tau\beta)(\gamma_+^\Sigma f+\gamma_-^\Sigma f)=0\text{ on }\Sigma\right\}, \tag*{(42)} i ( α ⋅ ν Σ ) ( γ + Σ f − γ − Σ f ) + 2 1 ( η I N + τ β ) ( γ + Σ f + γ − Σ f ) = 0 on Σ } , (42)
where γ ± Σ \gamma_\pm^\Sigma γ ± Σ and γ ∂ Ω \gamma^{\partial\Omega} γ ∂ Ω are the trace operators in (37) (for s = 0 s=0 s = 0 ) and (38) , respectively.
It will turn out in Theorem 4.8 with the help of extension theory methods for symmetric operators (see Appendix B) that the Dirac operator A η , τ A^{\eta,\tau} A η , τ is self-adjoint in L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^{N}) L 2 ( Ω , C N ) . For these considerations it is useful to introduce the minimal and maximal Dirac operators defined by
A min f = D f , dom A min = H 0 1 ( Ω ∖ Σ , C N ) , A_{\min}f=Df,\qquad\operatorname{dom}A_{\min}=H_0^1(\Omega\setminus\Sigma,\mathbb{C}^{N}), A m i n f = D f , dom A m i n = H 0 1 ( Ω ∖ Σ , C N ) ,
and
A max f = D f , dom A max = H α 0 ( Ω ∖ Σ , C N ) . A_{\max}f=Df,\qquad\operatorname{dom}A_{\max}=H_\alpha^{0}(\Omega\setminus\Sigma,\mathbb{C}^{N}). A m a x f = D f , dom A m a x = H α 0 ( Ω ∖ Σ , C N ) .
It can be shown in the same way as in [10 Proposition 3.1] that A min A_{\min} A m i n is a densely defined closed symmetric operator in L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^{N}) L 2 ( Ω , C N ) such that A min ∗ = A max A_{\min}^*=A_{\max} A m i n ∗ = A m a x holds. Besides the minimal and maximal operator we shall make use of the restriction S S S of the self-adjoint Dirac operator A A A onto the kernel of the trace map γ Σ \gamma^\Sigma γ Σ in (41) , which is given by
S f = D f , dom S = { f ∈ dom A : γ Σ f = 0 } = dom A ∣ ker γ Σ . (43) Sf=Df,\qquad\operatorname{dom}S=\{f\in\operatorname{dom}A:\gamma^\Sigma f=0\}=\left.\operatorname{dom}A\right|_{\ker\gamma^\Sigma}. \tag*{(43)} S f = D f , dom S = { f ∈ dom A : γ Σ f = 0 } = dom A ∣ k e r γ Σ . (43)
It follows from Lemma 3.2 that S S S is a densely defined closed symmetric operator in L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^{N}) L 2 ( Ω , C N ) , and it is also clear that the following inclusions hold
A min ⊂ S ⊂ A = A ∗ ⊂ S ∗ ⊂ A min ∗ = A max . (44) A_{\min}\subset S\subset A=A^*\subset S^*\subset A_{\min}^*=A_{\max}. \tag*{(44)} A m i n ⊂ S ⊂ A = A ∗ ⊂ S ∗ ⊂ A m i n ∗ = A m a x . (44)
Furthermore, it is not difficult to see that also the inclusion S ⊂ A η , τ S\subset A^{\eta,\tau} S ⊂ A η , τ holds. In the next lemma we specify the domain of the operator S ∗ S^* S ∗ .
Lemma 3.4. The adjoint of the operator S S S in (43) is given by
dom S ∗ = { f ∈ H α 0 , Σ ( Ω ∖ Σ , C N ) ∩ H α 1 2 , ∂ Ω ( Ω ∖ Σ , C N ) : B γ ∂ Ω f = 0 on ∂ Ω } , (45) \operatorname{dom} S^* = \left\{ f \in H^{0,\Sigma}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) \cap H^{\frac{1}{2},\partial\Omega}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) : B_{\gamma}^{\partial\Omega}f = 0\ \text{on }\partial\Omega\right\}, \tag*{(45)} dom S ∗ = { f ∈ H α 0 , Σ ( Ω ∖ Σ , C N ) ∩ H α 2 1 , ∂ Ω ( Ω ∖ Σ , C N ) : B γ ∂ Ω f = 0 on ∂ Ω } , (45) and, in particular, the operator A η , τ A^{\eta,\tau} A η , τ in Definition 3.3 is a restriction of S ∗ S^* S ∗ .
Proof. It is clear from (44) that S ∗ S^* S ∗ acts as D D D and that dom S ∗ \operatorname{dom} S^* dom S ∗ is contained in dom A max = H α 0 ( Ω ∖ Σ , C N ) \operatorname{dom} A_{\max} = H^{0}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) dom A m a x = H α 0 ( Ω ∖ Σ , C N ) .
Let us first show the inclusion ⊂ \subset ⊂ for the operator domain in (45) . Consider an arbitrary f ∈ dom S ∗ f \in\operatorname{dom} S^* f ∈ dom S ∗ . We shall often use that for χ ∈ C ∞ ( Ω ‾ ) \chi\in C^\infty(\overline{\Omega}) χ ∈ C ∞ ( Ω ) one has
D ( χ f ) = χ D f − i ( α ⋅ ∇ χ ) f ∈ D ′ ( Ω ∖ Σ , C N ) ∩ L 2 ( Ω , C N ) . (46) D(\chi f) = \chi Df - \mathrm{i}(\alpha\cdot\nabla\chi)f \in\mathcal{D}'(\Omega\setminus\Sigma,\mathbb{C}^{N}) \cap L^2(\Omega,\mathbb{C}^{N}). \tag*{(46)} D ( χ f ) = χD f − i ( α ⋅ ∇ χ ) f ∈ D ′ ( Ω ∖ Σ , C N ) ∩ L 2 ( Ω , C N ) . (46)
In particular, choosing χ ∈ C ∞ ( Ω ‾ ) \chi\in C^\infty(\overline{\Omega}) χ ∈ C ∞ ( Ω ) in (46) shows χ f ∈ H α 0 ( Ω ∖ Σ , C N ) \chi f \in H^{0}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) χ f ∈ H α 0 ( Ω ∖ Σ , C N ) and hence f ∈ H α 0 , Σ ( Ω ∖ Σ , C N ) f \in H^{0,\Sigma}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) f ∈ H α 0 , Σ ( Ω ∖ Σ , C N ) . Now let χ ∈ C ∞ ( Ω ‾ ) \chi\in C^\infty(\overline{\Omega}) χ ∈ C ∞ ( Ω ) with χ = 0 \chi= 0 χ = 0 near Σ \Sigma Σ . Using (46) we compute for g ∈ dom A g \in\operatorname{dom} A g ∈ dom A
⟨ D ( χ f ) , g ⟩ L 2 ( Ω , C N ) = ⟨ χ D f − i ( α ⋅ ∇ χ ) f , g ⟩ L 2 ( Ω , C N ) = ⟨ D f , χ ‾ g ⟩ L 2 ( Ω , C N ) + ⟨ f , i ( α ⋅ ∇ χ ‾ ) g ⟩ L 2 ( Ω , C N ) . (47) \left\langle D(\chi f),g\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}
= \left\langle\chi Df-\mathrm{i}(\alpha\cdot\nabla\chi)f,g\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}
= \left\langle Df,\overline{\chi}g\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}
+\left\langle f,\mathrm{i}(\alpha\cdot\nabla\overline{\chi})g\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}. \tag*{(47)} ⟨ D ( χ f ) , g ⟩ L 2 ( Ω , C N ) = ⟨ χD f − i ( α ⋅ ∇ χ ) f , g ⟩ L 2 ( Ω , C N ) = ⟨ D f , χ g ⟩ L 2 ( Ω , C N ) + ⟨ f , i ( α ⋅ ∇ χ ) g ⟩ L 2 ( Ω , C N ) . (47)
We further have χ ‾ g ∈ dom S \overline{\chi}g \in\operatorname{dom} S χ g ∈ dom S , hence,
⟨ D f , χ ‾ g ⟩ L 2 ( Ω , C N ) = ⟨ S ∗ f , χ ‾ g ⟩ L 2 ( Ω , C N ) = ⟨ f , S ( χ ‾ g ) ⟩ L 2 ( Ω , C N ) = ⟨ f , D ( χ ‾ g ) ⟩ L 2 ( Ω , C N ) = ⟨ f , χ ‾ D g − i ( α ⋅ ∇ χ ‾ ) g ⟩ L 2 ( Ω , C N ) = ⟨ χ f , A g ⟩ L 2 ( Ω , C N ) − ⟨ f , i ( α ⋅ ∇ χ ‾ ) g ⟩ L 2 ( Ω , C N ) , \left\langle Df,\overline{\chi}g\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}
= \left\langle S^*f,\overline{\chi}g\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}
= \left\langle f,S(\overline{\chi}g)\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}
= \left\langle f,D(\overline{\chi}g)\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}
= \left\langle f,\overline{\chi}Dg-\mathrm{i}(\alpha\cdot\nabla\overline{\chi})g\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}
= \left\langle\chi f,Ag\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}
-\left\langle f,\mathrm{i}(\alpha\cdot\nabla\overline{\chi})g\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}, ⟨ D f , χ g ⟩ L 2 ( Ω , C N ) = ⟨ S ∗ f , χ g ⟩ L 2 ( Ω , C N ) = ⟨ f , S ( χ g ) ⟩ L 2 ( Ω , C N ) = ⟨ f , D ( χ g ) ⟩ L 2 ( Ω , C N ) = ⟨ f , χ D g − i ( α ⋅ ∇ χ ) g ⟩ L 2 ( Ω , C N ) = ⟨ χ f , A g ⟩ L 2 ( Ω , C N ) − ⟨ f , i ( α ⋅ ∇ χ ) g ⟩ L 2 ( Ω , C N ) ,
and the substitution into (47) yields
⟨ D ( χ f ) , g ⟩ L 2 ( Ω , C N ) = ⟨ χ f , A g ⟩ L 2 ( Ω , C N ) . \left\langle D(\chi f),g\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}
= \left\langle\chi f,Ag\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}. ⟨ D ( χ f ) , g ⟩ L 2 ( Ω , C N ) = ⟨ χ f , A g ⟩ L 2 ( Ω , C N ) .
As g ∈ dom A g \in\operatorname{dom} A g ∈ dom A was arbitrary, this means χ f ∈ dom A ∗ = dom A \chi f \in\operatorname{dom} A^* = \operatorname{dom} A χ f ∈ dom A ∗ = dom A , i.e.
χ f ∈ H α 1 2 ( Ω , C N ) and B γ ∂ Ω ( χ f ) = 0 on ∂ Ω . \chi f \in H^{\frac{1}{2}}_{\alpha}(\Omega,\mathbb{C}^{N})
\quad\text{and}\quad
B_{\gamma}^{\partial\Omega}(\chi f)=0\ \text{on }\partial\Omega. χ f ∈ H α 2 1 ( Ω , C N ) and B γ ∂ Ω ( χ f ) = 0 on ∂ Ω.
Since this holds for any χ ∈ C ∞ ( Ω ‾ ) \chi\in C^\infty(\overline{\Omega}) χ ∈ C ∞ ( Ω ) with χ = 0 \chi= 0 χ = 0 near Σ \Sigma Σ , one obtains f ∈ H α 1 2 , ∂ Ω ( Ω ∖ Σ , C N ) f \in H^{\frac{1}{2},\partial\Omega}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^{N}) f ∈ H α 2 1 , ∂ Ω ( Ω ∖ Σ , C N ) ; cf. (36) . The identity B γ ∂ Ω ( χ f ) = 0 B_{\gamma}^{\partial\Omega}(\chi f)=0 B γ ∂ Ω ( χ f ) = 0 with χ ∈ C ∞ ( Ω ‾ ) \chi\in C^\infty(\overline{\Omega}) χ ∈ C ∞ ( Ω ) such that χ = 1 \chi= 1 χ = 1 near ∂ Ω \partial\Omega ∂ Ω gives the required boundary condition for f f f on ∂ Ω \partial\Omega ∂ Ω . The inclusion ⊂ \subset ⊂ in (45) is now proved.
It remains to show the inclusion ⊃ \supset ⊃ in (45) . Let f f f be an arbitrary function from the set on the right-hand side of (45) . Let χ ∈ C c ∞ ( Ω ) \chi\in C^\infty_{\mathrm{c}}(\Omega) χ ∈ C c ∞ ( Ω ) with χ = 1 \chi= 1 χ = 1 near Σ \Sigma Σ . We represent
f = χ f + ( 1 − χ ) f f = \chi f + (1-\chi)f f = χ f + ( 1 − χ ) f
and show that both summands are in dom S ∗ \operatorname{dom} S^* dom S ∗ . Note first that ( 1 − χ ) f ∈ H α 1 2 ( Ω , C N ) (1-\chi)f \in H^{\frac{1}{2}}_{\alpha}(\Omega,\mathbb{C}^{N}) ( 1 − χ ) f ∈ H α 2 1 ( Ω , C N ) and ( 1 − χ ) f = f (1-\chi)f=f ( 1 − χ ) f = f near ∂ Ω \partial\Omega ∂ Ω , and thus we have ( 1 − χ ) f ∈ dom A ⊂ dom S ∗ (1-\chi)f \in\operatorname{dom} A \subset\operatorname{dom} S^* ( 1 − χ ) f ∈ dom A ⊂ dom S ∗ . Hence, it suffices to show χ f ∈ dom S ∗ \chi f \in\operatorname{dom} S^* χ f ∈ dom S ∗ . For this fix some ρ ∈ C c ∞ ( Ω ) \rho\in C^\infty_{\mathrm{c}}(\Omega) ρ ∈ C c ∞ ( Ω ) with ρ = 1 \rho=1 ρ = 1 on an open neighborhood of supp χ \operatorname{supp}\chi supp χ . Then, due to the locality of D D D for any g ∈ dom S g \in\operatorname{dom} S g ∈ dom S one obtains
⟨ χ f , S g ⟩ L 2 ( Ω , C N ) = ⟨ χ f , D g ⟩ L 2 ( Ω , C N ) = ⟨ χ f , D ( ρ g ) ⟩ L 2 ( Ω , C N ) . \left\langle\chi f,Sg\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}
= \left\langle\chi f,Dg\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}
= \left\langle\chi f,D(\rho g)\right\rangle_{L^2(\Omega,\mathbb{C}^{N})}. ⟨ χ f , S g ⟩ L 2 ( Ω , C N ) = ⟨ χ f , D g ⟩ L 2 ( Ω , C N ) = ⟨ χ f , D ( ρ g ) ⟩ L 2 ( Ω , C N ) .
We have ρ g ∈ H 0 1 ( Ω ∖ Σ , C N ) = dom A min \rho g \in H^1_0(\Omega\setminus\Sigma,\mathbb{C}^N)=\operatorname{dom} A_{\min} ρ g ∈ H 0 1 ( Ω ∖ Σ , C N ) = dom A m i n . Due to χ f ∈ H α 0 ( Ω ∖ Σ , C N ) = dom A max \chi f \in H^0_\alpha(\Omega\setminus\Sigma,\mathbb{C}^N)=\operatorname{dom} A_{\max} χ f ∈ H α 0 ( Ω ∖ Σ , C N ) = dom A m a x and A max = A min ∗ A_{\max}=A_{\min}^* A m a x = A m i n ∗ we have
⟨ χ f , D ( ρ g ) ⟩ L 2 ( Ω , C N ) = ⟨ χ f , A min ( ρ g ) ⟩ L 2 ( Ω , C N ) = ⟨ A max ( χ f ) , ρ g ⟩ L 2 ( Ω , C N ) = ⟨ D ( χ f ) , ρ g ⟩ L 2 ( Ω , C N ) = ⟨ D ( χ f ) , g ⟩ L 2 ( Ω , C N ) , \langle\chi f,D(\rho g)\rangle_{L^2(\Omega,\mathbb{C}^N)}
=\langle\chi f,A_{\min}(\rho g)\rangle_{L^2(\Omega,\mathbb{C}^N)}
=\langle A_{\max}(\chi f),\rho g\rangle_{L^2(\Omega,\mathbb{C}^N)}
=\langle D(\chi f),\rho g\rangle_{L^2(\Omega,\mathbb{C}^N)}
=\langle D(\chi f),g\rangle_{L^2(\Omega,\mathbb{C}^N)}, ⟨ χ f , D ( ρ g ) ⟩ L 2 ( Ω , C N ) = ⟨ χ f , A m i n ( ρ g ) ⟩ L 2 ( Ω , C N ) = ⟨ A m a x ( χ f ) , ρ g ⟩ L 2 ( Ω , C N ) = ⟨ D ( χ f ) , ρ g ⟩ L 2 ( Ω , C N ) = ⟨ D ( χ f ) , g ⟩ L 2 ( Ω , C N ) ,
which yields ⟨ χ f , S g ⟩ L 2 ( Ω , C N ) = ⟨ D ( χ f ) , g ⟩ L 2 ( Ω , C N ) \langle\chi f,Sg\rangle_{L^2(\Omega,\mathbb{C}^N)}=\langle D(\chi f),g\rangle_{L^2(\Omega,\mathbb{C}^N)} ⟨ χ f , S g ⟩ L 2 ( Ω , C N ) = ⟨ D ( χ f ) , g ⟩ L 2 ( Ω , C N ) for any g ∈ dom S g\in\operatorname{dom}S g ∈ dom S and shows χ f ∈ dom S ∗ \chi f\in\operatorname{dom}S^* χ f ∈ dom S ∗ . This completes the proof. □
Definition using boundary triples
In this subsection we use the abstract concept of boundary triples from Appendix § 9 to study the operator A η , τ A^{\eta,\tau} A η , τ in Definition 3.3. More precisely, we use the construction of a boundary triple from Theorem B.4 by starting with the operator Q : = A Q:=A Q := A and its symmetric restriction S S S . Then the operator A η , τ A^{\eta,\tau} A η , τ can be represented as in Theorem B.2 . The following considerations can be viewed as a generalization of the treatment in [13 ] and, in particular, we refer the reader there for a more detailed exposition for the case n = 2 n=2 n = 2 .
Recall that Σ p \Sigma_p Σ p , p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } , denote the connected components of the compact C ∞ C^\infty C ∞ -smooth hypersurface Σ \Sigma Σ on which the transmission conditions are imposed. For each p p p we choose a classical elliptic (scalar) pseudodifferential operator Λ p \Lambda_p Λ p of order 1 2 \frac{1}{2} 2 1 on Σ p \Sigma_p Σ p , and we additionally require Λ p \Lambda_p Λ p to be injective and symmetric in L 2 ( Σ p ) L^2(\Sigma_p) L 2 ( Σ p ) ; a specific choice of Λ p \Lambda_p Λ p will be made later. Then
Λ : = ⨁ p = 1 P Λ p : H s ( Σ ) → H s − 1 2 ( Σ ) (48) \Lambda:=\bigoplus_{p=1}^{P}\Lambda_p:H^s(\Sigma)\to H^{s-\frac{1}{2}}(\Sigma) \tag*{(48)} Λ := p = 1 ⨁ P Λ p : H s ( Σ ) → H s − 2 1 ( Σ ) (48)
is an isomorphism for any s ∈ R s\in\mathbb{R} s ∈ R , and we denote its natural extension Λ ⊗ I N \Lambda\otimes I_N Λ ⊗ I N on C N \mathbb{C}^N C N -valued functions by the same symbol. Due to Lemma 3.2 the map
γ : = Λ γ Σ : dom A → L 2 ( Σ , C N ) \gamma:=\Lambda\gamma^\Sigma:\operatorname{dom}A\to L^2(\Sigma,\mathbb{C}^N) γ := Λ γ Σ : dom A → L 2 ( Σ , C N )
is surjective and bounded with respect to the graph norm. Since ker γ = ker γ Σ \ker\gamma=\ker\gamma^\Sigma ker γ = ker γ Σ it is clear from (43) that
S f = D f , dom S = { f ∈ dom A : γ f = 0 } = dom A ∣ ker γ . (49) Sf=Df,\qquad\operatorname{dom}S=\{f\in\operatorname{dom}A:\gamma f=0\}=\operatorname{dom}A|_{\ker\gamma}. \tag*{(49)} S f = D f , dom S = { f ∈ dom A : γ f = 0 } = dom A ∣ k e r γ . (49)
Following (B.2) we now define
Γ z : = ( γ ( A − z ‾ ) − 1 ) ∗ , z ∈ res A . \Gamma_z:=\bigl(\gamma(A-\overline{z})^{-1}\bigr)^*,\qquad z\in\operatorname{res}A. Γ z := ( γ ( A − z ) − 1 ) ∗ , z ∈ res A .
In order to determine Γ z \Gamma_z Γ z more explicitly consider first φ ∈ C ∞ ( Σ , C N ) \varphi\in C^\infty(\Sigma,\mathbb{C}^N) φ ∈ C ∞ ( Σ , C N ) and f ∈ C c ∞ ( Ω , C N ) f\in C^\infty_c(\Omega,\mathbb{C}^N) f ∈ C c ∞ ( Ω , C N ) , and note that
[ γ Σ ( A − z ‾ ) − 1 f ] ( x ) = ∫ Ω G z ‾ ( x , y ) f ( y ) d y for any x ∈ Σ [\gamma^\Sigma(A-\overline{z})^{-1}f](x)=\int_\Omega G_{\overline{z}}(x,y)f(y)\,dy\quad\text{for any }x\in\Sigma [ γ Σ ( A − z ) − 1 f ] ( x ) = ∫ Ω G z ( x , y ) f ( y ) d y for any x ∈ Σ
by virtue of Theorem 2.5 . We have Λ φ ∈ C ∞ ( Σ , C N ) \Lambda\varphi\in C^\infty(\Sigma,\mathbb{C}^N) Λ φ ∈ C ∞ ( Σ , C N ) , and due to the properties of the integral kernel G z G_z G z stated in Theorem 2.5 , the function
Σ × Ω ∋ ( x , y ) ↦ G z ‾ ( x , y ) ∈ C N × N \Sigma\times\Omega\ni(x,y)\mapsto G_{\overline{z}}(x,y)\in\mathbb{C}^{N\times N} Σ × Ω ∋ ( x , y ) ↦ G z ( x , y ) ∈ C N × N
is integrable on Σ × supp f \Sigma\times\operatorname{supp} f Σ × supp f . Using (2.18) and Fubini’s theorem we compute
⟨ φ , γ ( A − z ‾ ) − 1 f ⟩ L 2 ( Σ , C N ) = ⟨ Λ φ , γ Σ ( A − z ‾ ) − 1 f ⟩ L 2 ( Σ , C N ) = ∫ Σ ⟨ ( Λ φ ) ( x ) , ∫ Ω G z ‾ ( x , y ) f ( y ) d y ⟩ C N d s ( x ) = ∫ Σ ∫ Ω ⟨ G z ‾ ( x , y ) ∗ ( Λ φ ) ( x ) , f ( y ) ⟩ C N d y d s ( x ) = ∫ Ω ∫ Σ ⟨ G z ( y , x ) ( Λ φ ) ( x ) , f ( y ) ⟩ C N d s ( x ) d y = ∫ Ω ⟨ ∫ Σ G z ( y , x ) ( Λ φ ) ( x ) d s ( x ) , f ( y ) ⟩ C N d y = ⟨ Φ z Λ φ , f ⟩ L 2 ( Ω , C N ) . \begin{aligned}
\langle\varphi,\gamma(A-\overline{z})^{-1}f\rangle_{L^{2}(\Sigma,\mathbb{C}^{N})}
&=\langle\Lambda\varphi,\gamma^{\Sigma}(A-\overline{z})^{-1}f\rangle_{L^{2}(\Sigma,\mathbb{C}^{N})}\\
&=\int_{\Sigma}\left\langle(\Lambda\varphi)(x),\int_{\Omega}G_{\overline{z}}(x,y)f(y)\,\mathrm{d}y\right\rangle_{\mathbb{C}^{N}}\mathrm{d}s(x)\\
&=\int_{\Sigma}\int_{\Omega}\left\langle G_{\overline{z}}(x,y)^{*}(\Lambda\varphi)(x),f(y)\right\rangle_{\mathbb{C}^{N}}\mathrm{d}y\,\mathrm{d}s(x)\\
&=\int_{\Omega}\int_{\Sigma}\left\langle G_{z}(y,x)(\Lambda\varphi)(x),f(y)\right\rangle_{\mathbb{C}^{N}}\mathrm{d}s(x)\,\mathrm{d}y\\
&=\int_{\Omega}\left\langle\int_{\Sigma}G_{z}(y,x)(\Lambda\varphi)(x)\,\mathrm{d}s(x),f(y)\right\rangle_{\mathbb{C}^{N}}\mathrm{d}y\\
&=\langle\Phi_{z}\Lambda\varphi,f\rangle_{L^{2}(\Omega,\mathbb{C}^{N})}.
\end{aligned} ⟨ φ , γ ( A − z ) − 1 f ⟩ L 2 ( Σ , C N ) = ⟨ Λ φ , γ Σ ( A − z ) − 1 f ⟩ L 2 ( Σ , C N ) = ∫ Σ ⟨ ( Λ φ ) ( x ) , ∫ Ω G z ( x , y ) f ( y ) d y ⟩ C N d s ( x ) = ∫ Σ ∫ Ω ⟨ G z ( x , y ) ∗ ( Λ φ ) ( x ) , f ( y ) ⟩ C N d y d s ( x ) = ∫ Ω ∫ Σ ⟨ G z ( y , x ) ( Λ φ ) ( x ) , f ( y ) ⟩ C N d s ( x ) d y = ∫ Ω ⟨ ∫ Σ G z ( y , x ) ( Λ φ ) ( x ) d s ( x ) , f ( y ) ⟩ C N d y = ⟨ Φ z Λ φ , f ⟩ L 2 ( Ω , C N ) .
where the mapping Φ z \Phi_z Φ z is defined by
Φ z ψ ( x ) : = ∫ Σ G z ( x , y ) ψ ( y ) d s ( y ) , ψ ∈ C ∞ ( Σ , C N ) , x ∈ Ω ∖ Σ . (50) \Phi_z\psi(x):=\int_{\Sigma}G_z(x,y)\psi(y)\,\mathrm{d}s(y),\qquad\psi\in C^\infty(\Sigma,\mathbb{C}^{N}),\quad x\in\Omega\setminus\Sigma. \tag*{(50)} Φ z ψ ( x ) := ∫ Σ G z ( x , y ) ψ ( y ) d s ( y ) , ψ ∈ C ∞ ( Σ , C N ) , x ∈ Ω ∖ Σ. (50)
As C ∞ ( Ω , C N ) C^\infty(\Omega,\mathbb{C}^{N}) C ∞ ( Ω , C N ) is dense in L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^{N}) L 2 ( Ω , C N ) , the resulting identity
⟨ φ , γ ( A − z ‾ ) − 1 f ⟩ L 2 ( Σ , C N ) = ⟨ Φ z Λ φ , f ⟩ L 2 ( Ω , C N ) \langle\varphi,\gamma(A-\overline{z})^{-1}f\rangle_{L^{2}(\Sigma,\mathbb{C}^{N})}
=\langle\Phi_z\Lambda\varphi,f\rangle_{L^{2}(\Omega,\mathbb{C}^{N})} ⟨ φ , γ ( A − z ) − 1 f ⟩ L 2 ( Σ , C N ) = ⟨ Φ z Λ φ , f ⟩ L 2 ( Ω , C N )
holds for all φ ∈ C ∞ ( Σ , C N ) \varphi\in C^\infty(\Sigma,\mathbb{C}^{N}) φ ∈ C ∞ ( Σ , C N ) and f ∈ L 2 ( Ω , C N ) f\in L^2(\Omega,\mathbb{C}^{N}) f ∈ L 2 ( Ω , C N ) , and hence we conclude
Γ z φ = Φ z Λ φ , φ ∈ C ∞ ( Σ , C N ) . (51) \Gamma_z\varphi=\Phi_z\Lambda\varphi,\qquad\varphi\in C^\infty(\Sigma,\mathbb{C}^{N}). \tag*{(51)} Γ z φ = Φ z Λ φ , φ ∈ C ∞ ( Σ , C N ) . (51)
By the abstract theory (see Appendix § 9 ), the map Γ z \Gamma_z Γ z extends uniquely to a bounded linear operator L 2 ( Σ , C N ) → ker ( S ∗ − z ) ⊂ dom S ∗ L^2(\Sigma,\mathbb{C}^{N})\to\ker(S^*-z)\subset\operatorname{dom}S^* L 2 ( Σ , C N ) → ker ( S ∗ − z ) ⊂ dom S ∗ equipped with the graph norm. Using the fact that Λ : L 2 ( Σ , C N ) → H − 1 / 2 ( Σ , C N ) \Lambda:L^2(\Sigma,\mathbb{C}^{N})\to H^{-1/2}(\Sigma,\mathbb{C}^{N}) Λ : L 2 ( Σ , C N ) → H − 1/2 ( Σ , C N ) is an isomorphism, we conclude that the mapping Φ z \Phi_z Φ z in (3.18) defined on C ∞ ( Σ , C N ) C^\infty(\Sigma,\mathbb{C}^{N}) C ∞ ( Σ , C N ) extends uniquely to a bounded linear operator
Φ z : H − 1 / 2 ( Σ , C N ) → L 2 ( Ω , C N ) , ran Φ z ⊂ ker ( S ∗ − z ) . \Phi_z:H^{-1/2}(\Sigma,\mathbb{C}^{N})\to L^2(\Omega,\mathbb{C}^{N}),\qquad\operatorname{ran}\Phi_z\subset\ker(S^*-z). Φ z : H − 1/2 ( Σ , C N ) → L 2 ( Ω , C N ) , ran Φ z ⊂ ker ( S ∗ − z ) .
As the L 2 L^2 L 2 -norm and the graph norm of S ∗ S^* S ∗ are equivalent on ker ( S ∗ − z ) \ker(S^*-z) ker ( S ∗ − z ) we can also view Φ z \Phi_z Φ z as a bounded linear operator in the context Φ z : H − 1 / 2 ( Σ , C N ) → H α 0 ( Ω ∖ Σ , C N ) \Phi_z:H^{-1/2}(\Sigma,\mathbb{C}^{N})\to H^0_\alpha(\Omega\setminus\Sigma,\mathbb{C}^{N}) Φ z : H − 1/2 ( Σ , C N ) → H α 0 ( Ω ∖ Σ , C N ) , where dom S ∗ ⊂ dom A max = H α 0 ( Ω ∖ Σ , C N ) \operatorname{dom}S^*\subset\operatorname{dom}A_{\max}=H^0_\alpha(\Omega\setminus\Sigma,\mathbb{C}^{N}) dom S ∗ ⊂ dom A m a x = H α 0 ( Ω ∖ Σ , C N ) ; cf. (3.12). We conclude that the identity (3.19) extends to
Γ z φ = Φ z Λ φ , φ ∈ L 2 ( Σ , C N ) . (52) \Gamma_z\varphi=\Phi_z\Lambda\varphi,\qquad\varphi\in L^2(\Sigma,\mathbb{C}^{N}). \tag*{(52)} Γ z φ = Φ z Λ φ , φ ∈ L 2 ( Σ , C N ) . (52)
Similarly, considering Φ z \Phi_z Φ z as a bounded operator from L 2 ( Σ , C N ) L^2(\Sigma,\mathbb{C}^{N}) L 2 ( Σ , C N ) to L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^{N}) L 2 ( Ω , C N ) we conclude that its adjoint Φ z ∗ \Phi_z^* Φ z ∗ given by
Φ z ∗ f ( x ) = ∫ Ω G z ‾ ( x , y ) f ( y ) d y , f ∈ C c ∞ ( Ω , C N ) , x ∈ Σ , \Phi_z^*f(x)=\int_{\Omega}G_{\overline{z}}(x,y)f(y)\,\mathrm{d}y,\qquad f\in C^\infty_c(\Omega,\mathbb{C}^{N}),\quad x\in\Sigma, Φ z ∗ f ( x ) = ∫ Ω G z ( x , y ) f ( y ) d y , f ∈ C c ∞ ( Ω , C N ) , x ∈ Σ ,
extends uniquely to a bounded linear operator
Φ z ∗ : L 2 ( Ω , C N ) → H 1 / 2 ( Σ , C N ) . \Phi_z^*:L^2(\Omega,\mathbb{C}^{N})\to H^{1/2}(\Sigma,\mathbb{C}^{N}). Φ z ∗ : L 2 ( Ω , C N ) → H 1/2 ( Σ , C N ) .
Following the construction in Appendix § 9 we note that for z ∈ res A z\in\operatorname{res}A z ∈ res A each function f ∈ dom S ∗ f\in\operatorname{dom}S^* f ∈ dom S ∗ can be uniquely represented in the form f = f z + Φ z Λ ξ f=f_z+\Phi_z\Lambda\xi f = f z + Φ z Λ ξ with f z ∈ dom A f_z\in\operatorname{dom}A f z ∈ dom A and ξ ∈ L 2 ( Σ , C N ) \xi\in L^2(\Sigma,\mathbb{C}^{N}) ξ ∈ L 2 ( Σ , C N ) (where ξ \xi ξ does not depend on z z z ). In order to construct a boundary triple with the help of Theorem B.4 it is necessary to determine ξ \xi ξ directly from f f f . This will be based on the fact that the function Φ z Λ ξ \Phi_z\Lambda\xi Φ z Λ ξ has a jump along Σ \Sigma Σ , while f z f_z f z has no jump along Σ \Sigma Σ (as f z ∈ dom A ⊂ H l o c 1 ( Ω , C N ) f_z \in\operatorname{dom} A \subset H^1_{\mathrm{loc}}(\Omega,\mathbb{C}^N) f z ∈ dom A ⊂ H loc 1 ( Ω , C N ) ).
Lemma 3.5. For z ∈ res A z \in\operatorname{res} A z ∈ res A the following is true:
(a) The singular integral operator C z : C ∞ ( Σ , C N ) → C ∞ ( Σ , C N ) C_z : C^\infty(\Sigma,\mathbb{C}^N) \to C^\infty(\Sigma,\mathbb{C}^N) C z : C ∞ ( Σ , C N ) → C ∞ ( Σ , C N ) ,
C z φ ( x ) = p.v. ∫ Σ G z ( x , y ) φ ( y ) d s ( y ) , x ∈ Σ , φ ∈ C ∞ ( Σ , C N ) , (53) C_z\varphi(x)=\operatorname{p.v.}\int_\Sigma G_z(x,y)\varphi(y)\,\mathrm{d}s(y), \qquad x\in\Sigma,\quad\varphi\in C^\infty(\Sigma,\mathbb{C}^N), \tag*{(53)} C z φ ( x ) = p.v. ∫ Σ G z ( x , y ) φ ( y ) d s ( y ) , x ∈ Σ , φ ∈ C ∞ ( Σ , C N ) , (53) where p. v. stands for the Cauchy principal value, extends uniquely to a bounded linear operator in H r ( Σ , C N ) H^r(\Sigma,\mathbb{C}^N) H r ( Σ , C N ) for any r ∈ R r\in\mathbb{R} r ∈ R .
(b) For any φ ∈ H − 1 2 ( Σ , C N ) \varphi\in H^{- \frac{1}{2}}(\Sigma,\mathbb{C}^N) φ ∈ H − 2 1 ( Σ , C N ) one has
γ ± Σ Φ z φ = ∓ i 2 ( α ⋅ ν Σ ) φ + C z φ . (54) \gamma_\pm^\Sigma\Phi_z\varphi=\mp\frac{\mathrm{i}}{2}(\alpha\cdot\nu_\Sigma)\varphi+C_z\varphi. \tag*{(54)} γ ± Σ Φ z φ = ∓ 2 i ( α ⋅ ν Σ ) φ + C z φ . (54)
In Section 4.1 we will analyze the operator C z C_z C z further with pseudodifferential techniques, see Corollary 4.6 . For the construction of the boundary triple used to define A η , τ A^{\eta,\tau} A η , τ the results of the preceding lemma are sufficient.
Proof of Lemma 3.5. (a) Choose R > 0 R>0 R > 0 such that Ω ⊂ B R ( 0 ) \Omega\subset B_R(0) Ω ⊂ B R ( 0 ) and χ ∈ C ∞ ( R n ) \chi\in C^\infty(\mathbb{R}^n) χ ∈ C ∞ ( R n ) such that χ ≡ 1 \chi\equiv1 χ ≡ 1 on B 2 R ( 0 ) B_{2R}(0) B 2 R ( 0 ) . Note that for the function H z \mathcal{H}_z H z in Lemma 2.9 one has H z − H 0 = z E 0 ⊗ I N \mathcal{H}_z-\mathcal{H}_0=z\mathcal{E}_0\otimes I_N H z − H 0 = z E 0 ⊗ I N with E 0 \mathcal{E}_0 E 0 defined by (2.9). Hence,
E z ( x ) − E 0 ( x ) = E z ( x ) − H z ( x ) − ( E 0 ( x ) − H 0 ( x ) ) + z E 0 ( x ) ⊗ I N , x ∈ R n ∖ { 0 } , E_z(x)-E_0(x)=E_z(x)-\mathcal{H}_z(x)-\bigl(E_0(x)-\mathcal{H}_0(x)\bigr)+z\mathcal{E}_0(x)\otimes I_N,\qquad x\in\mathbb{R}^n\setminus\{0\}, E z ( x ) − E 0 ( x ) = E z ( x ) − H z ( x ) − ( E 0 ( x ) − H 0 ( x ) ) + z E 0 ( x ) ⊗ I N , x ∈ R n ∖ { 0 } ,
where E 0 E_0 E 0 is defined by (2.10) & (2.11). Therefore, using Theorem 2.5 and the function K z \mathcal{K}_z K z from Lemma 2.9 one can write for x , y ∈ Σ x,y\in\Sigma x , y ∈ Σ , x ≠ y x\ne y x = y ,
G z ( x , y ) = E 0 ( x − y ) + χ ( x − y ) ( E z ( x − y ) − E 0 ( x − y ) ) + T z ( x , y ) = E 0 ( x − y ) + K z ( x − y ) − K 0 ( x − y ) + z E 0 ( x − y ) ⊗ I N + T z ( x , y ) . (55) \begin{aligned}
G_z(x,y)&=E_0(x-y)+\chi(x-y)\bigl(E_z(x-y)-E_0(x-y)\bigr)+T_z(x,y)\\
&=E_0(x-y)+\mathcal{K}_z(x-y)-\mathcal{K}_0(x-y)+z\mathcal{E}_0(x-y)\otimes I_N+T_z(x,y). \tag*{(55)}
\end{aligned} G z ( x , y ) = E 0 ( x − y ) + χ ( x − y ) ( E z ( x − y ) − E 0 ( x − y ) ) + T z ( x , y ) = E 0 ( x − y ) + K z ( x − y ) − K 0 ( x − y ) + z E 0 ( x − y ) ⊗ I N + T z ( x , y ) . (55)
Based on the above decomposition of G z G_z G z we consider the operator
C ~ φ ( x ) = p.v. ∫ Σ E 0 ( x − y ) φ ( y ) d s ( y ) , x ∈ Σ , φ ∈ C ∞ ( Σ , C N ) , (56) \widetilde{C}\varphi(x)=\operatorname{p.v.}\int_\Sigma E_0(x-y)\varphi(y)\,\mathrm{d}s(y),\qquad x\in\Sigma,\quad\varphi\in C^\infty(\Sigma,\mathbb{C}^N), \tag*{(56)} C φ ( x ) = p.v. ∫ Σ E 0 ( x − y ) φ ( y ) d s ( y ) , x ∈ Σ , φ ∈ C ∞ ( Σ , C N ) , (56)
and the integral operators acting on φ ∈ C ∞ ( Σ , C N ) \varphi\in C^\infty(\Sigma,\mathbb{C}^N) φ ∈ C ∞ ( Σ , C N ) evaluated at x ∈ Σ x\in\Sigma x ∈ Σ as
K Σ , z φ ( x ) = ∫ Σ K z ( x − y ) φ ( y ) d s ( y ) , K_{\Sigma,z}\varphi(x)=\int_\Sigma K_z(x-y)\varphi(y)\,\mathrm{d}s(y), K Σ , z φ ( x ) = ∫ Σ K z ( x − y ) φ ( y ) d s ( y ) ,
K Σ , 0 φ ( x ) = ∫ Σ K 0 ( x − y ) φ ( y ) d s ( y ) , K_{\Sigma,0}\varphi(x)=\int_\Sigma K_0(x-y)\varphi(y)\,\mathrm{d}s(y), K Σ , 0 φ ( x ) = ∫ Σ K 0 ( x − y ) φ ( y ) d s ( y ) ,
S Σ φ ( x ) = ∫ Σ E 0 ( x − y ) φ ( y ) d s ( y ) , S_\Sigma\varphi(x)=\int_\Sigma E_0(x-y)\varphi(y)\,\mathrm{d}s(y), S Σ φ ( x ) = ∫ Σ E 0 ( x − y ) φ ( y ) d s ( y ) ,
T ~ Σ , z φ ( x ) = ∫ Σ T z ( x , y ) φ ( y ) d s ( y ) . (57) \widetilde{T}_{\Sigma,z}\varphi(x)=\int_\Sigma T_z(x,y)\varphi(y)\,\mathrm{d}s(y). \tag*{(57)} T Σ , z φ ( x ) = ∫ Σ T z ( x , y ) φ ( y ) d s ( y ) . (57)
It is known that C ~ \widetilde{C} C extends to a bounded linear operator in H r ( Σ , C N ) H^r(\Sigma,\mathbb{C}^N) H r ( Σ , C N ) for any r ∈ R r\in\mathbb{R} r ∈ R ; for n = 2 n=2 n = 2 this is stated in [13 Proposition 3.3] , while for n = 3 n=3 n = 3 one has by [18 Theorem 4.1] that C ~ \widetilde{C} C is an h h h -pseudodifferential operator of order 0 in the sense of [18 Definition 2.1] and hence, by [51 Remark after Theorem 14.2] this implies the claimed continuity. Next, consider the map
K ~ z φ ( x ) : = ∫ R n K z ( x − y ) φ ( y ) d y , x ∈ R n , φ ∈ C c ∞ ( R n ) . \widetilde{K}_z\varphi(x):=\int_{\mathbb{R}^n}K_z(x-y)\varphi(y)\,\mathrm{d}y,\qquad x\in\mathbb{R}^n,\quad\varphi\in C^\infty_{\mathrm{c}}(\mathbb{R}^n). K z φ ( x ) := ∫ R n K z ( x − y ) φ ( y ) d y , x ∈ R n , φ ∈ C c ∞ ( R n ) .
In the following, we use the notations for classical pseudodifferential operators introduced in Appendix A. Taking the expansion of K z K_z K z from Lemma 2.9 into account, one can directly apply [48 Chapter 7, Proposition 2.8] which shows that K ~ z ∈ Op S N − 3 ( R n ) \widetilde{K}_z \in\operatorname{Op} S_N^{-3}(\mathbb{R}^n) K z ∈ Op S N − 3 ( R n ) . Hence, following the standard construction of restricting an integral operator in R n \mathbb{R}^n R n to Σ \Sigma Σ one gets K Σ , z ∈ Op S N − 2 ( Σ ) K_{\Sigma,z} \in\operatorname{Op} S_N^{-2}(\Sigma) K Σ , z ∈ Op S N − 2 ( Σ ) , see the considerations leading to [48 Chapter 7, equation (11.4)] ; in particular, the maps K Σ , z K_{\Sigma,z} K Σ , z can be extended to a bounded operator K Σ , z : H r ( Σ , C N ) → H r + 2 ( Σ , C N ) K_{\Sigma,z}: H^r(\Sigma,\mathbb{C}^N) \to H^{r+2}(\Sigma,\mathbb{C}^N) K Σ , z : H r ( Σ , C N ) → H r + 2 ( Σ , C N ) for any r ∈ R r \in\mathbb{R} r ∈ R and thus, K Σ , z K_{\Sigma,z} K Σ , z gives rise to a bounded operator in H r ( Σ , C N ) H^r(\Sigma,\mathbb{C}^N) H r ( Σ , C N ) . In the same way one shows that K Σ , 0 ∈ Op S N − 2 ( Σ ) K_{\Sigma,0} \in\operatorname{Op} S_N^{-2}(\Sigma) K Σ , 0 ∈ Op S N − 2 ( Σ ) and that K Σ , 0 K_{\Sigma,0} K Σ , 0 can be extended to a bounded operator in H r ( Σ , C N ) H^r(\Sigma,\mathbb{C}^N) H r ( Σ , C N ) for any r ∈ R r \in\mathbb{R} r ∈ R . Moreover, it is known from [48 Chapter 7, Proposition 11.2] that S Σ ∈ Op S 1 − 1 ( Σ ) S_\Sigma\in\operatorname{Op} S_1^{-1}(\Sigma) S Σ ∈ Op S 1 − 1 ( Σ ) and thus, it can also be extended to a bounded operator in H r ( Σ , C N ) H^r(\Sigma,\mathbb{C}^N) H r ( Σ , C N ) , r ∈ R r \in\mathbb{R} r ∈ R . Finally, as T z T_z T z is smooth (see Theorem 2.5 ), one has T ~ Σ , z ∈ Op S N − ∞ ( Σ ) \widetilde{T}_{\Sigma,z} \in\operatorname{Op} S_N^{-\infty}(\Sigma) T Σ , z ∈ Op S N − ∞ ( Σ ) , implying that also T ~ Σ , z \widetilde{T}_{\Sigma,z} T Σ , z can be extended to a bounded operator in H r ( Σ , C N ) H^r(\Sigma,\mathbb{C}^N) H r ( Σ , C N ) , r ∈ R r \in\mathbb{R} r ∈ R . Thus, it follows from (3.23) and (3.24) that
C z = C ~ + K Σ , z − K Σ , 0 + z S Σ ⊗ I N + T ~ Σ , z C_z = \widetilde{C} + K_{\Sigma,z} - K_{\Sigma,0} + zS_\Sigma\otimes I_N + \widetilde{T}_{\Sigma,z} C z = C + K Σ , z − K Σ , 0 + z S Σ ⊗ I N + T Σ , z
can be extended to a bounded operator in H r ( Σ , C N ) H^r(\Sigma,\mathbb{C}^N) H r ( Σ , C N ) for any r ∈ R r \in\mathbb{R} r ∈ R . In particular, for any φ ∈ C ∞ ( Σ , C N ) \varphi\in C^\infty(\Sigma,\mathbb{C}^N) φ ∈ C ∞ ( Σ , C N ) one has C z φ ∈ ⋂ r ∈ R H r ( Σ , C N ) = C ∞ ( Σ , C N ) C_z\varphi\in\bigcap_{r\in\mathbb{R}} H^r(\Sigma,\mathbb{C}^N) = C^\infty(\Sigma,\mathbb{C}^N) C z φ ∈ ⋂ r ∈ R H r ( Σ , C N ) = C ∞ ( Σ , C N ) by the Sobolev embedding theorem, and thus C z C_z C z maps C ∞ ( Σ , C N ) C^\infty(\Sigma,\mathbb{C}^N) C ∞ ( Σ , C N ) into itself.
(b) We show (3.22) for φ ∈ C ∞ ( Σ , C N ) \varphi\in C^\infty(\Sigma,\mathbb{C}^N) φ ∈ C ∞ ( Σ , C N ) , the claim for φ ∈ H − 1 2 ( Σ , C N ) \varphi\in H^{-\frac{1}{2}}(\Sigma,\mathbb{C}^N) φ ∈ H − 2 1 ( Σ , C N ) follows then by density and continuity. Again, the decomposition of G z G_z G z in (3.23) and the operators defined in (3.24) and (3.25) will play an important role. Define the operator
Φ ~ φ ( x ) = ∫ Σ E 0 ( x − y ) φ ( y ) d s ( y ) , x ∈ Ω ∖ Σ , φ ∈ C ∞ ( Σ , C N ) . \widetilde{\Phi}\varphi(x) = \int_\Sigma E_0(x-y)\varphi(y)\,\mathrm{d}s(y), \qquad x \in\Omega\setminus\Sigma,\quad\varphi\in C^\infty(\Sigma,\mathbb{C}^N). Φ φ ( x ) = ∫ Σ E 0 ( x − y ) φ ( y ) d s ( y ) , x ∈ Ω ∖ Σ , φ ∈ C ∞ ( Σ , C N ) .
Then, by [12 Proposition 4.4] one has
γ ± Σ Φ ~ φ = ∓ i 2 ( α ⋅ ν Σ ) φ + C ~ φ . (58) \gamma_\pm^\Sigma\widetilde{\Phi}\varphi= \mp\frac{\mathrm{i}}{2}(\alpha\cdot\nu^\Sigma)\varphi+ \widetilde{C}\varphi. \tag*{(58)} γ ± Σ Φ φ = ∓ 2 i ( α ⋅ ν Σ ) φ + C φ . (58)
Next, consider the functions f 1 , f 2 f_1, f_2 f 1 , f 2 defined for x ∈ Ω ∖ Σ x \in\Omega\setminus\Sigma x ∈ Ω ∖ Σ by
f 1 ( x ) : = ∫ Σ ( K z ( x − y ) − K 0 ( x − y ) + T z ( x , y ) ) φ ( y ) d s ( y ) , f_1(x) := \int_\Sigma\left(K_z(x-y)-K_0(x-y)+T_z(x,y)\right)\varphi(y)\,\mathrm{d}s(y), f 1 ( x ) := ∫ Σ ( K z ( x − y ) − K 0 ( x − y ) + T z ( x , y ) ) φ ( y ) d s ( y ) ,
f 2 ( x ) : = z ∫ Σ E 0 ( x − y ) φ ( y ) d s ( y ) . f_2(x) := z\int_\Sigma E_0(x-y)\varphi(y)\,\mathrm{d}s(y). f 2 ( x ) := z ∫ Σ E 0 ( x − y ) φ ( y ) d s ( y ) .
Since K z − K 0 K_z-K_0 K z − K 0 is bounded on R n \mathbb{R}^n R n and continuous on R n ∖ { 0 } \mathbb{R}^n \setminus\{0\} R n ∖ { 0 } by Lemma 2.9, since single points have surface measure zero on Σ \Sigma Σ , and since T z T_z T z is smooth by Theorem 2.5 , one gets with dominated convergence that f 1 f_1 f 1 can be extended to a continuous function f ~ 1 \widetilde{f}_1 f 1 in Ω \Omega Ω satisfying
f ~ 1 ( x ) = ∫ Σ ( K z ( x − y ) − K 0 ( x − y ) + T z ( x , y ) ) φ ( y ) d s ( y ) = ( K Σ , z φ − K Σ , 0 φ + T ~ Σ , z φ ) ( x ) , x ∈ Σ . \begin{aligned}
\widetilde{f}_1(x) = \int_\Sigma\left(K_z(x-y)-K_0(x-y)+T_z(x,y)\right)\varphi(y)\,\mathrm{d}s(y) \\
&= \left(K_{\Sigma,z}\varphi-K_{\Sigma,0}\varphi+\widetilde{T}_{\Sigma,z}\varphi\right)(x), \qquad x \in\Sigma.
\end{aligned} f 1 ( x ) = ∫ Σ ( K z ( x − y ) − K 0 ( x − y ) + T z ( x , y ) ) φ ( y ) d s ( y ) = ( K Σ , z φ − K Σ , 0 φ + T Σ , z φ ) ( x ) , x ∈ Σ.
Furthermore, by the gradient estimates in Lemma 2.9 (d) & (e) one has, for 0 < ∣ x ∣ ≤ 2 R 0 < |x| \le2R 0 < ∣ x ∣ ≤ 2 R ,
∣ ∇ ( K z − K 0 ) ( x ) ∣ ≤ c ( 1 + ∣ log ∣ x ∣ ∣ ) , n = 2 , |\nabla(K_z-K_0)(x)| \le c(1+|\log|x||), \qquad n=2, ∣∇ ( K z − K 0 ) ( x ) ∣ ≤ c ( 1 + ∣ log ∣ x ∣∣ ) , n = 2 ,
∣ ∇ ( K z − K 0 ) ( x ) ∣ ≤ c ∣ x ∣ − 1 , n = 3. |\nabla(K_z-K_0)(x)| \le c|x|^{-1}, \qquad n=3. ∣∇ ( K z − K 0 ) ( x ) ∣ ≤ c ∣ x ∣ − 1 , n = 3.
Since for any r > 0 r>0 r > 0 we have
sup ∣ x ∣ < r ∫ Σ ( 1 + ∣ log ∣ x − y ∣ ∣ ) d s ( y ) < ∞ , n = 2 , \sup_{|x|<r}\int_{\Sigma}(1+|\log|x-y||)\,ds(y)<\infty,\qquad n=2, ∣ x ∣ < r sup ∫ Σ ( 1 + ∣ log ∣ x − y ∣∣ ) d s ( y ) < ∞ , n = 2 ,
sup ∣ x ∣ < r ∫ Σ ∣ x − y ∣ − 1 d s ( y ) < ∞ , n = 3 , \sup_{|x|<r}\int_{\Sigma}|x-y|^{-1}\,ds(y)<\infty,\qquad n=3, ∣ x ∣ < r sup ∫ Σ ∣ x − y ∣ − 1 d s ( y ) < ∞ , n = 3 ,
and T z T_z T z is smooth, the differentiation under the integral sign shows that ∇ f 1 \nabla f_1 ∇ f 1 is bounded on Ω 0 ∖ Σ \Omega_0\setminus\Sigma Ω 0 ∖ Σ for any open set Ω 0 \Omega_0 Ω 0 with Σ ⊂ Ω 0 \Sigma\subset\Omega_0 Σ ⊂ Ω 0 and Ω 0 ‾ ⊂ Ω \overline{\Omega_0}\subset\Omega Ω 0 ⊂ Ω . Hence, f 1 f_1 f 1 belongs to H 1 H^1 H 1 on both sides of Σ \Sigma Σ in a neighborhood of Σ \Sigma Σ , and, therefore, its trace γ ± Σ f 1 \gamma^\Sigma_\pm f_1 γ ± Σ f 1 coincide with f ~ 1 ∣ Σ \widetilde{f}_1|_\Sigma f 1 ∣ Σ . Moreover, by [32 equation (7.8)] one has
γ ± Σ f 2 = z ( S Σ ⊗ I N ) φ . \gamma^\Sigma_\pm f_2=z(S_\Sigma\otimes I_N)\varphi. γ ± Σ f 2 = z ( S Σ ⊗ I N ) φ .
Hence, we conclude
γ ± Σ Φ z φ = γ ± Σ ( Φ ~ φ + f 1 + f 2 ) = ∓ i 2 ( α ⋅ ν Σ ) φ + C ~ φ + K Σ , z φ − K Σ , 0 φ + T ~ Σ , z φ + z ( S Σ ⊗ I N ) φ = ∓ i 2 ( α ⋅ ν Σ ) φ + C z φ , \begin{aligned}
\gamma^\Sigma_\pm\Phi_z\varphi
&=\gamma^\Sigma_\pm(\widetilde{\Phi}\varphi+f_1+f_2)\\
&=\mp\frac{i}{2}(\alpha\cdot\nu^\Sigma)\varphi+\widetilde{C}\varphi+K_{\Sigma,z}\varphi-K_{\Sigma,0}\varphi+\widetilde{T}_{\Sigma,z}\varphi+z(S_\Sigma\otimes I_N)\varphi\\
&=\mp\frac{i}{2}(\alpha\cdot\nu^\Sigma)\varphi+C_z\varphi,
\end{aligned} γ ± Σ Φ z φ = γ ± Σ ( Φ φ + f 1 + f 2 ) = ∓ 2 i ( α ⋅ ν Σ ) φ + C φ + K Σ , z φ − K Σ , 0 φ + T Σ , z φ + z ( S Σ ⊗ I N ) φ = ∓ 2 i ( α ⋅ ν Σ ) φ + C z φ ,
where in the last step again (55) was used. This is the claimed formula. □
Now we use Theorem B.4 to construct an explicit boundary triple in the present situation. The following assertion is the first step for the spectral study of A η , τ A^{\eta,\tau} A η , τ . We provide a short proof of Lemma 3.6 and refer the reader to [13 Proof of Proposition 3.5] for more details (in the case n = 2 n=2 n = 2 , Σ = Σ 1 \Sigma=\Sigma_1 Σ = Σ 1 and Ω = R 2 \Omega=\mathbb{R}^2 Ω = R 2 ).
Lemma 3.6. Fix some ζ ∈ R ∩ res A \zeta\in\mathbb{R}\cap\operatorname{res} A ζ ∈ R ∩ res A , let S ∗ S^* S ∗ be the adjoint of the densely defined closed symmetric operator S = A ∣ ker γ S=A|_{\ker\gamma} S = A ∣ k e r γ from Lemma 3.4, and consider the mappings
Π 0 , Π 1 : dom S ∗ → L 2 ( Σ , C N ) \Pi_0,\Pi_1:\operatorname{dom}S^*\to L^2(\Sigma,\mathbb{C}^N) Π 0 , Π 1 : dom S ∗ → L 2 ( Σ , C N ) defined by
Π 0 f : = i Λ − 1 ( α ⋅ ν Σ ) ( γ + Σ f − γ − Σ f ) , \Pi_0f:=i\Lambda^{-1}(\alpha\cdot\nu^\Sigma)(\gamma^\Sigma_+f-\gamma^\Sigma_-f), Π 0 f := i Λ − 1 ( α ⋅ ν Σ ) ( γ + Σ f − γ − Σ f ) , Π 1 f : = Λ [ 1 2 ( γ + Σ f + γ − Σ f ) − C ζ Λ Π 0 f ] . \Pi_1f:=\Lambda\left[\frac{1}{2}(\gamma^\Sigma_+f+\gamma^\Sigma_-f)-C_\zeta\Lambda\Pi_0f\right]. Π 1 f := Λ [ 2 1 ( γ + Σ f + γ − Σ f ) − C ζ Λ Π 0 f ] . Then ( L 2 ( Σ , C N ) , Π 0 , Π 1 ) (L^2(\Sigma,\mathbb{C}^N),\Pi_0,\Pi_1) ( L 2 ( Σ , C N ) , Π 0 , Π 1 ) is a boundary triple for S ∗ S^* S ∗ such that A = S ∗ ∣ ker Π 0 A=S^*|_{\ker\Pi_0} A = S ∗ ∣ k e r Π 0 . For z ∈ res A z\in\operatorname{res} A z ∈ res A the associated γ \gamma γ -field Γ z \Gamma_z Γ z is given by (52) and the associated Weyl function M z M_z M z is given by
M z : = Λ ( C z − C ζ ) Λ . (59) M_z:=\Lambda(C_z-C_\zeta)\Lambda. \tag*{(59)} M z := Λ ( C z − C ζ ) Λ. (59)
Proof. Essentially the assertions follow from Theorem B.4 for the special case ζ ∈ R \zeta\in\mathbb{R} ζ ∈ R . More precisely, write f ∈ dom S ∗ f\in\operatorname{dom}S^* f ∈ dom S ∗ in the form
f = f ζ + Γ ζ ξ = f ζ + Φ ζ Λ ξ , f ζ ∈ dom A , ξ ∈ L 2 ( Σ , C N ) , f=f_\zeta+\Gamma_\zeta\xi=f_\zeta+\Phi_\zeta\Lambda\xi,\qquad f_\zeta\in\operatorname{dom}A,\quad\xi\in L^2(\Sigma,\mathbb{C}^N), f = f ζ + Γ ζ ξ = f ζ + Φ ζ Λ ξ , f ζ ∈ dom A , ξ ∈ L 2 ( Σ , C N ) ,
see (B.3) and (52) . Due to f ζ ∈ dom A ⊂ H l o c 1 ( Ω , C N ) f_\zeta\in\operatorname{dom}A\subset H^1_{\mathrm{loc}}(\Omega,\mathbb{C}^N) f ζ ∈ dom A ⊂ H loc 1 ( Ω , C N ) we have γ + Σ f ζ − γ − Σ f ζ = 0 \gamma^\Sigma_+f_\zeta-\gamma^\Sigma_-f_\zeta=0 γ + Σ f ζ − γ − Σ f ζ = 0 . Therefore, Lemma 3.5(b) yields
γ + Σ f − γ − Σ f = − i ( α ⋅ ν Σ ) Λ ξ , \gamma^\Sigma_+f-\gamma^\Sigma_-f=-i(\alpha\cdot\nu^\Sigma)\Lambda\xi, γ + Σ f − γ − Σ f = − i ( α ⋅ ν Σ ) Λ ξ ,
and hence
ξ = i Λ − 1 ( α ⋅ ν Σ ) ( γ + Σ f − γ − Σ f ) , \xi= \mathrm{i}\Lambda^{-1}(\alpha\cdot\nu^\Sigma)(\gamma_+^\Sigma f-\gamma_-^\Sigma f), ξ = i Λ − 1 ( α ⋅ ν Σ ) ( γ + Σ f − γ − Σ f ) ,
which yields in the above expression for Π 0 \Pi_0 Π 0 according to Theorem B.4 . Furthermore, as ζ = ζ ˉ \zeta=\bar{\zeta} ζ = ζ ˉ Theorem B.4 gives
Π 1 f = γ f ζ = Λ γ + Σ f ζ = 1 2 Λ ( γ + Σ + γ − Σ ) f ζ = 1 2 Λ ( γ + Σ + γ − Σ ) ( f − Φ ζ Λ ξ ) = 1 2 Λ ( γ + Σ + γ − Σ ) ( f − Φ ζ Λ Π 0 f ) , \begin{aligned}
\Pi_1 f=\gamma f_\zeta=\Lambda\gamma_+^\Sigma f_\zeta=\frac{1}{2}\Lambda(\gamma_+^\Sigma+\gamma_-^\Sigma)f_\zeta\\
&=\frac{1}{2}\Lambda(\gamma_+^\Sigma+\gamma_-^\Sigma)(f-\Phi_\zeta\Lambda\xi)=\frac{1}{2}\Lambda(\gamma_+^\Sigma+\gamma_-^\Sigma)(f-\Phi_\zeta\Lambda\Pi_0f),
\end{aligned} Π 1 f = γ f ζ = Λ γ + Σ f ζ = 2 1 Λ ( γ + Σ + γ − Σ ) f ζ = 2 1 Λ ( γ + Σ + γ − Σ ) ( f − Φ ζ Λ ξ ) = 2 1 Λ ( γ + Σ + γ − Σ ) ( f − Φ ζ Λ Π 0 f ) ,
and another application of Lemma 3.5 (b) gives the required expression for Π 1 \Pi_1 Π 1 . Now Theorem B.4 implies that ( L 2 ( Σ , C N ) , Π 0 , Π 1 ) (L^2(\Sigma,\mathbb{C}^N),\Pi_0,\Pi_1) ( L 2 ( Σ , C N ) , Π 0 , Π 1 ) is a boundary triple for S ∗ S^* S ∗ such that A = S ∗ ∣ ker Π 0 A=S^*|_{\ker\Pi_0} A = S ∗ ∣ k e r Π 0 and the associated γ \gamma γ -field Γ z \Gamma_z Γ z is given by (3.20).
For the corresponding Weyl function M z M_z M z we have
M z = γ ( Γ z − Γ ζ ) = 1 2 Λ ( γ + Σ + γ − Σ ) ( Φ z − Φ ζ ) Λ , M_z=\gamma(\Gamma_z-\Gamma_\zeta)=\frac{1}{2}\Lambda(\gamma_+^\Sigma+\gamma_-^\Sigma)(\Phi_z-\Phi_\zeta)\Lambda, M z = γ ( Γ z − Γ ζ ) = 2 1 Λ ( γ + Σ + γ − Σ ) ( Φ z − Φ ζ ) Λ ,
and the result follows from Lemma 3.5 (b).
Using the above construction, we can finally represent the operator A η , τ A^{\eta,\tau} A η , τ in Definition 3.3 via the boundary triple ( L 2 ( Σ , C N ) , Π 0 , Π 1 ) (L^2(\Sigma,\mathbb{C}^N),\Pi_0,\Pi_1) ( L 2 ( Σ , C N ) , Π 0 , Π 1 ) from Lemma 3.6.
Lemma 3.7. Assume that ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p|\ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ for all p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } and define
Θ η , τ : = − Λ [ 1 η 2 − τ 2 ( η I N − τ β ) + C ζ ] Λ , (60) \Theta^{\eta,\tau}:=-\Lambda\left[\frac{1}{\eta^2-\tau^2}(\eta I_N-\tau\beta)+C_\zeta\right]\Lambda, \tag*{(60)} Θ η , τ := − Λ [ η 2 − τ 2 1 ( η I N − τ β ) + C ζ ] Λ , (60) which is viewed as an operator on the maximal domain dom Θ η , τ \operatorname{dom}\Theta^{\eta,\tau} dom Θ η , τ in L 2 ( Σ , C N ) L^2(\Sigma,\mathbb{C}^N) L 2 ( Σ , C N ) . Then
dom A η , τ = { f ∈ dom S ∗ : Π 0 f ∈ dom Θ η , τ , Π 1 f = Θ η , τ Π 0 f } , \operatorname{dom}A^{\eta,\tau}=\{f\in\operatorname{dom}S^*:\Pi_0f\in\operatorname{dom}\Theta^{\eta,\tau},\ \Pi_1f=\Theta^{\eta,\tau}\Pi_0f\}, dom A η , τ = { f ∈ dom S ∗ : Π 0 f ∈ dom Θ η , τ , Π 1 f = Θ η , τ Π 0 f } , i.e. the operator A η , τ A^{\eta,\tau} A η , τ is the restriction of S ∗ S^* S ∗ on the functions f f f satisfying
Π 0 f ∈ dom Θ η , τ and Π 1 f = Θ η , τ Π 0 f . \Pi_0f\in\operatorname{dom}\Theta^{\eta,\tau}\quad\text{and}\quad\Pi_1f=\Theta^{\eta,\tau}\Pi_0f. Π 0 f ∈ dom Θ η , τ and Π 1 f = Θ η , τ Π 0 f .
Proof. From the expression for the boundary maps Π 0 , Π 1 \Pi_0,\Pi_1 Π 0 , Π 1 for any f ∈ dom S ∗ f\in\operatorname{dom}S^* f ∈ dom S ∗ one obtains
γ + Σ f − γ − Σ f = − i ( α ⋅ ν Σ ) Λ Π 0 f and 1 2 ( γ + Σ f + γ − Σ f ) = Λ − 1 Π 1 f + C ζ Λ Π 0 f . \gamma_+^\Sigma f-\gamma_-^\Sigma f=-\mathrm{i}(\alpha\cdot\nu^\Sigma)\Lambda\Pi_0f
\quad\text{and}\quad
\frac{1}{2}(\gamma_+^\Sigma f+\gamma_-^\Sigma f)=\Lambda^{-1}\Pi_1f+C_\zeta\Lambda\Pi_0f. γ + Σ f − γ − Σ f = − i ( α ⋅ ν Σ ) Λ Π 0 f and 2 1 ( γ + Σ f + γ − Σ f ) = Λ − 1 Π 1 f + C ζ Λ Π 0 f .
The direct substitution into the boundary condition at Σ \Sigma Σ for A η , τ A^{\eta,\tau} A η , τ in Definition 3.3 shows that f ∈ dom A η , τ f\in\operatorname{dom}A^{\eta,\tau} f ∈ dom A η , τ if and only if
Λ Π 0 f + ( η I N + τ β ) [ Λ − 1 Π 1 + C ζ Λ Π 0 ] f = 0 , \Lambda\Pi_0f+(\eta I_N+\tau\beta)[\Lambda^{-1}\Pi_1+C_\zeta\Lambda\Pi_0]f=0, Λ Π 0 f + ( η I N + τ β ) [ Λ − 1 Π 1 + C ζ Λ Π 0 ] f = 0 ,
which can be equivalently rewritten as
( η I N + τ β ) Λ − 1 Π 1 f = − ( I N + ( η I N + τ β ) C ζ ) Λ Π 0 f . (\eta I_N+\tau\beta)\Lambda^{-1}\Pi_1f=-(I_N+(\eta I_N+\tau\beta)C_\zeta)\Lambda\Pi_0f. ( η I N + τ β ) Λ − 1 Π 1 f = − ( I N + ( η I N + τ β ) C ζ ) Λ Π 0 f .
The assumption ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p|\ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ for all p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } ensures that the piecewise constant matrix function η I N + τ β : Σ → C N × N \eta I_N+\tau\beta:\Sigma\to\mathbb{C}^{N\times N} η I N + τ β : Σ → C N × N is invertible with
( η I N + τ β ) − 1 = 1 η 2 − τ 2 ( η I N − τ β ) , (\eta I_N+\tau\beta)^{-1}=\frac{1}{\eta^2-\tau^2}(\eta I_N-\tau\beta), ( η I N + τ β ) − 1 = η 2 − τ 2 1 ( η I N − τ β ) ,
which gives the required representation. □ \square □
Spectral properties of shell interactions
The main results of this section are Theorem 4.8 and Theorem 4.19, where we show the self-adjointness and determine the essential spectrum of the operator A η , τ A^{\eta,\tau} A η , τ in Definition 3.3. Moreover, if the interaction is non-critical on one part Σ p \Sigma_p Σ p , p ∈ { 1 , … , P } p \in\{1,\ldots,P\} p ∈ { 1 , … , P } , i.e. η p 2 − τ p 2 ≠ 4 \eta_p^2-\tau_p^2 \ne4 η p 2 − τ p 2 = 4 , then it will turn out in Corollary 4.10 that the functions in dom A η , τ \operatorname{dom} A^{\eta,\tau} dom A η , τ are H 1 H^1 H 1 -regular near Σ p \Sigma_p Σ p . For both purposes, we will rewrite the parameter Θ η , τ \Theta^{\eta,\tau} Θ η , τ given by (3.28) as the sum of operators acting on Σ p \Sigma_p Σ p and a perturbation term, followed by a more detailed analysis. In the next subsection, which is of preparatory nature, we first investigate the involved maps as pseudodifferential operators; see Appendix A for the associated notations and basic results.
Separating connected components
Let us make a specific choice of the operator Λ p ∈ Op S 1 1 2 ( Σ p ) \Lambda_p \in\operatorname{Op} S^{\frac{1}{2}}_1(\Sigma_p) Λ p ∈ Op S 1 2 1 ( Σ p ) in (3.16) that will be used throughout this section. For this define the operators S p S_p S p over Σ p \Sigma_p Σ p by
S p φ ( x ) = ∫ Σ p E 0 ( x − y ) φ ( y ) d s ( y ) , x ∈ Σ p , φ ∈ C ∞ ( Σ p ) , (61) S_p\varphi(x)=\int_{\Sigma_p} E_0(x-y)\varphi(y)\,\mathrm{d}s(y),\qquad x\in\Sigma_p,\quad\varphi\in C^\infty(\Sigma_p), \tag*{(61)} S p φ ( x ) = ∫ Σ p E 0 ( x − y ) φ ( y ) d s ( y ) , x ∈ Σ p , φ ∈ C ∞ ( Σ p ) , (61)
where E 0 E_0 E 0 is given by (2.9). It is known that S p S_p S p gives rise to a compact self-adjoint operator in L 2 ( Σ p ) L^2(\Sigma_p) L 2 ( Σ p ) . We denote the lower bound of the spectrum by m p : = min spec S p ∈ ( − ∞ , 0 ] m_p:=\min\operatorname{spec}S_p\in(-\infty,0] m p := min spec S p ∈ ( − ∞ , 0 ] , the spectral projection corresponding to the spectrum of S p S_p S p in [ m p , 0 ] [m_p,0] [ m p , 0 ] by π p \pi_p π p , and the corresponding spectral subspace by H p − = ran π p \mathcal{H}_p^-=\operatorname{ran}\pi_p H p − = ran π p .
Lemma 4.1. For S p S_p S p in (4.1) and m p , π p , H p − m_p,\pi_p,\mathcal{H}_p^- m p , π p , H p − the following is true.
(a) S p ∈ Op S 1 − 1 ( Σ p ) S_p\in\operatorname{Op}S^{-1}_1(\Sigma_p) S p ∈ Op S 1 − 1 ( Σ p )
(b) dim H p − < ∞ \dim\mathcal{H}_p^-<\infty dim H p − < ∞ and π p ∈ Op S 1 − ∞ ( Σ p ) \pi_p\in\operatorname{Op}S^{-\infty}_1(\Sigma_p) π p ∈ Op S 1 − ∞ ( Σ p )
(c) The operator
Λ p : = ( S p − ( m p − 1 ) π p ) − 1 2 ∈ Op S 1 1 2 ( Σ p ) (62) \Lambda_p:=\left(S_p-(m_p-1)\pi_p\right)^{-\frac{1}{2}}\in\operatorname{Op}S^{\frac{1}{2}}_1(\Sigma_p) \tag*{(62)} Λ p := ( S p − ( m p − 1 ) π p ) − 2 1 ∈ Op S 1 2 1 ( Σ p ) (62)
is elliptic, and, as an operator in L 2 ( Σ p ) L^2(\Sigma_p) L 2 ( Σ p ) , it is positive and self-adjoint with dom Λ p = H 1 ( Σ p ) \operatorname{dom}\Lambda_p=H^1(\Sigma_p) dom Λ p = H 1 ( Σ p ) ; in particular, Λ p \Lambda_p Λ p is injective and symmetric in L 2 ( Σ p ) L^2(\Sigma_p) L 2 ( Σ p ) . Its principal symbol is
σ Λ p ( x , ξ ) = 2 ∣ ξ ∣ . (63) \sigma_{\Lambda_p}(x,\xi)=\sqrt{2|\xi|}. \tag*{(63)} σ Λ p ( x , ξ ) = 2∣ ξ ∣ . (63)
(d) Λ p S p Λ p − I = ( m p − 1 ) Λ p 2 π p ∈ Op S 1 − ∞ ( Σ p ) \Lambda_pS_p\Lambda_p-I=(m_p-1)\Lambda_p^2\pi_p\in\operatorname{Op}S^{-\infty}_1(\Sigma_p) Λ p S p Λ p − I = ( m p − 1 ) Λ p 2 π p ∈ Op S 1 − ∞ ( Σ p )
Proof. First, recall that S p S_p S p is the classical single layer boundary integral operator on Σ p \Sigma_p Σ p . In particular, one has (a) and S p S_p S p is elliptic, self-adjoint, and its principal symbol is σ S p ( x , ξ ) = 1 2 ∣ ξ ∣ \sigma_{S_p}(x,\xi)=\frac{1}{2|\xi|} σ S p ( x , ξ ) = 2∣ ξ ∣ 1 , see, e.g., [30 Section 10.2.2 and Section 11.4.1] and [46 Theorem 6.22] . Moreover, again by [46 Theorem 6.22] the space H p − = ran π p \mathcal{H}_p^-=\operatorname{ran}\pi_p H p − = ran π p has finite dimension, which is the first claim in (b). Let ( φ j p ) j = 1 L ⊂ L 2 ( Σ p ) (\varphi_j^p)_{j=1}^L\subset L^2(\Sigma_p) ( φ j p ) j = 1 L ⊂ L 2 ( Σ p ) be an orthonormal basis of H p − \mathcal{H}_p^- H p − . Then, by elliptic regularity φ j p ∈ C ∞ ( Σ p ) \varphi_j^p\in C^\infty(\Sigma_p) φ j p ∈ C ∞ ( Σ p ) , j ∈ { 1 , … , L } j\in\{1,\ldots,L\} j ∈ { 1 , … , L } . Thus, with
Π p ( x , y ) = ∑ j = 1 L φ j p ( x ) φ j p ( y ) ‾ ∈ C ∞ ( Σ p × Σ p ) \Pi_p(x,y)=\sum_{j=1}^{L}\varphi_j^p(x)\overline{\varphi_j^p(y)}\in C^\infty(\Sigma_p\times\Sigma_p) Π p ( x , y ) = j = 1 ∑ L φ j p ( x ) φ j p ( y ) ∈ C ∞ ( Σ p × Σ p )
one can write
π p ψ = ∑ j = 1 L ( ψ , φ j p ) L 2 ( Σ p ) φ j p = ∫ Σ p Π p ( ⋅ , y ) ψ ( y ) d s ( y ) , \pi_p\psi=\sum_{j=1}^{L}(\psi,\varphi_j^p)_{L^2(\Sigma_p)}\varphi_j^p=\int_{\Sigma_p}\Pi_p(\cdot,y)\psi(y)\,\mathrm{d}s(y), π p ψ = j = 1 ∑ L ( ψ , φ j p ) L 2 ( Σ p ) φ j p = ∫ Σ p Π p ( ⋅ , y ) ψ ( y ) d s ( y ) ,
which implies the second claim in (b), that is, π p ∈ Op S 1 − ∞ ( Σ p ) \pi_p\in\operatorname{Op}S_1^{-\infty}(\Sigma_p) π p ∈ Op S 1 − ∞ ( Σ p ) . Next we verify (c). As m p m_p m p is the lower bound of the spectrum of S p S_p S p , the operator B p : = S p − ( m p − 1 ) π p ∈ Op S 1 − 1 ( Σ p ) B_p:=S_p-(m_p-1)\pi_p\in\operatorname{Op}S_1^{-1}(\Sigma_p) B p := S p − ( m p − 1 ) π p ∈ Op S 1 − 1 ( Σ p ) is self-adjoint and positive: on H p − \mathcal{H}_p^- H p − its spectrum is contained in [ 1 , 1 − m p ] [1,1-m_p] [ 1 , 1 − m p ] , and on the orthogonal complement of H p − \mathcal{H}_p^- H p − it coincides with S p S_p S p , which is positive there. Moreover, B p B_p B p is elliptic by (a) and (b), and hence its inverse B p − 1 B_p^{-1} B p − 1 is a positive and self-adjoint elliptic operator in Op S 1 1 ( Σ p ) \operatorname{Op}S_1^1(\Sigma_p) Op S 1 1 ( Σ p ) with dom B p = H 1 ( Σ p ) \operatorname{dom}B_p=H^1(\Sigma_p) dom B p = H 1 ( Σ p ) . Therefore, by Seeley’s theorem on complex powers of elliptic operators [45 ] ,
Λ p = ( S p − ( m p − 1 ) π p ) − 1 2 = ( B p − 1 ) 1 2 ∈ Op S 1 2 1 ( Σ p ) \Lambda_p=(S_p-(m_p-1)\pi_p)^{-\frac{1}{2}}=(B_p^{-1})^{\frac{1}{2}}\in\operatorname{Op}S_{\frac{1}{2}}^1(\Sigma_p) Λ p = ( S p − ( m p − 1 ) π p ) − 2 1 = ( B p − 1 ) 2 1 ∈ Op S 2 1 1 ( Σ p )
is elliptic, and it is positive and self-adjoint in L 2 ( Σ p ) L^2(\Sigma_p) L 2 ( Σ p ) with dom Λ p = H 1 2 ( Σ p ) \operatorname{dom}\Lambda_p=H^{\frac{1}{2}}(\Sigma_p) dom Λ p = H 2 1 ( Σ p ) . Since σ S p ( x , ξ ) = 1 2 ∣ ξ ∣ \sigma_{S_p}(x,\xi)=\frac{1}{2|\xi|} σ S p ( x , ξ ) = 2∣ ξ ∣ 1 , we conclude that the principal symbol of Λ p \Lambda_p Λ p is
σ Λ p ( x , ξ ) = 2 ∣ ξ ∣ . \sigma_{\Lambda_p}(x,\xi)=\sqrt{2|\xi|}. σ Λ p ( x , ξ ) = 2∣ ξ ∣ .
Finally, to see (d) observe that, since S p S_p S p and π p \pi_p π p commute,
Λ p S p Λ p = ( S p − ( m p − 1 ) π p ) − 1 2 S p ( S p − ( m p − 1 ) π p ) − 1 2 = ( S p − ( m p − 1 ) π p ) − 1 S p = I + ( S p − ( m p − 1 ) π p ) − 1 ( m p − 1 ) π p = I + ( m p − 1 ) Λ p 2 π p . (64) \begin{aligned}
\Lambda_pS_p\Lambda_p&=(S_p-(m_p-1)\pi_p)^{-\frac{1}{2}}S_p(S_p-(m_p-1)\pi_p)^{-\frac{1}{2}}\\
&=(S_p-(m_p-1)\pi_p)^{-1}S_p\\
&=I+(S_p-(m_p-1)\pi_p)^{-1}(m_p-1)\pi_p\\
&=I+(m_p-1)\Lambda_p^2\pi_p. \tag*{(64)}
\end{aligned} Λ p S p Λ p = ( S p − ( m p − 1 ) π p ) − 2 1 S p ( S p − ( m p − 1 ) π p ) − 2 1 = ( S p − ( m p − 1 ) π p ) − 1 S p = I + ( S p − ( m p − 1 ) π p ) − 1 ( m p − 1 ) π p = I + ( m p − 1 ) Λ p 2 π p . (64)
and since π p ∈ Op S 1 − ∞ ( Σ p ) \pi_p\in\operatorname{Op}S_1^{-\infty}(\Sigma_p) π p ∈ Op S 1 − ∞ ( Σ p ) by (b) it follows that (d) holds. □
In our further analysis also the singular integral operators W p ± W_p^\pm W p ± defined by
W p ± φ ( x ) = p . v . ∫ Σ p w ± ( x − y ) φ ( y ) d s ( y ) , x ∈ Σ p , φ ∈ C ∞ ( Σ p ) , (65) W_p^\pm\varphi(x)=\mathop{\mathrm{p.v.}}\int_{\Sigma_p}w^\pm(x-y)\varphi(y)\,\mathrm{d}s(y),\quad x\in\Sigma_p,\quad\varphi\in C^\infty(\Sigma_p), \tag*{(65)} W p ± φ ( x ) = p.v. ∫ Σ p w ± ( x − y ) φ ( y ) d s ( y ) , x ∈ Σ p , φ ∈ C ∞ ( Σ p ) , (65)
where p. v. stands for the Cauchy principal value and
w ± ( x ) = { i 2 π 1 x 1 ± i x 2 , n = 2 , i ( σ ⋅ x ) 4 π ∣ x ∣ 3 , n = 3 , w^\pm(x)=
\begin{cases}
\frac{\mathrm{i}}{2\pi}\frac{1}{x_1\pm\mathrm{i}x_2},&n=2,\\[6pt]
\frac{\mathrm{i}(\boldsymbol{\sigma}\cdot x)}{4\pi|x|^3},&n=3,
\end{cases} w ± ( x ) = ⎩ ⎨ ⎧ 2 π i x 1 ± i x 2 1 , 4 π ∣ x ∣ 3 i ( σ ⋅ x ) , n = 2 , n = 3 ,
will be important; cf. Lemma 2.9. In the next lemma, we study their pseudodifferential properties.
For the three-dimensional case some further geometrical quantities are necessary. Consider for x ∈ Σ p x\in\Sigma_p x ∈ Σ p the shape operator
V x : = d ν Σ ∣ x : T x Σ p → T x Σ p . (66) V_x:=\left.\mathrm{d}\nu^\Sigma\right|_x:T_x\Sigma_p\to T_x\Sigma_p. \tag*{(66)} V x := d ν Σ x : T x Σ p → T x Σ p . (66)
Its eigenvalues κ 1 ( x ) \kappa_1(x) κ 1 ( x ) and κ 2 ( x ) \kappa_2(x) κ 2 ( x ) are the principal curvatures of Σ p \Sigma_p Σ p in x x x . As V x V_x V x is self-adjoint with respect to the scalar product inherited from R 3 \mathbb{R}^3 R 3 , one may find an orthonormal basis ( e 1 , e 2 ) (e_1,e_2) ( e 1 , e 2 ) of T x Σ p T_x\Sigma_p T x Σ p (we omit its dependence on x x x ) such that V x e j = κ j ( x ) e j V_xe_j=\kappa_j(x)e_j V x e j = κ j ( x ) e j for j ∈ { 1 , 2 } j\in\{1,2\} j ∈ { 1 , 2 } and ( e 1 , e 2 , ν Σ ( x ) ) (e_1,e_2,\nu^\Sigma(x)) ( e 1 , e 2 , ν Σ ( x )) is a positively oriented basis in R 3 \mathbb{R}^3 R 3 . With these notations we introduce the following special parametrizations of Σ p \Sigma_p Σ p .
Definition 4.2. Let Σ p \Sigma_p Σ p , p ∈ { 1 , … , P } p \in\{1,\ldots,P\} p ∈ { 1 , … , P } , be one of the connected components of Σ \Sigma Σ .
(a) Let n = 2 n=2 n = 2 and ℓ p \ell_p ℓ p be the length of Σ p \Sigma_p Σ p . We call an arc-length parametrization γ : ( − ℓ p 2 , ℓ p 2 ) → Σ p \gamma:\left(-\frac{\ell_p}{2},\frac{\ell_p}{2}\right)\to\Sigma_p γ : ( − 2 ℓ p , 2 ℓ p ) → Σ p (i.e. ∣ γ ′ ( ⋅ ) ∣ = 1 |\gamma'(\cdot)|=1 ∣ γ ′ ( ⋅ ) ∣ = 1 ) of Σ p \Sigma_p Σ p properly oriented , if γ ′ ( t ) \gamma'(t) γ ′ ( t ) is obtained from the normal ν Σ ( γ ( t ) ) \nu^\Sigma(\gamma(t)) ν Σ ( γ ( t )) by the anti-clockwise rotation by π 2 \frac{\pi}{2} 2 π for each t t t , i.e. ν Σ ( γ ( t ) ) = ( γ 2 ′ ( t ) , − γ 1 ′ ( t ) ) \nu^\Sigma(\gamma(t))=(\gamma_2'(t),-\gamma_1'(t)) ν Σ ( γ ( t )) = ( γ 2 ′ ( t ) , − γ 1 ′ ( t )) for all t ∈ ( − ℓ p 2 , ℓ p 2 ) t\in\left(-\frac{\ell_p}{2},\frac{\ell_p}{2}\right) t ∈ ( − 2 ℓ p , 2 ℓ p ) .
(b) Let n = 3 n=3 n = 3 , x ∈ Σ p x\in\Sigma_p x ∈ Σ p , and e 1 , e 2 e_1,e_2 e 1 , e 2 be as above. We say that a local chart φ : R 2 ⊃ U → V ⊂ Σ p \varphi:\mathbb{R}^2\supset U\to V\subset\Sigma_p φ : R 2 ⊃ U → V ⊂ Σ p corresponds to normal coordinates near x x x , if 0 ∈ U 0\in U 0 ∈ U , φ ( 0 ) = x \varphi(0)=x φ ( 0 ) = x , and ∂ j φ ( 0 ) = e j \partial_j\varphi(0)=e_j ∂ j φ ( 0 ) = e j for j ∈ { 1 , 2 } j\in\{1,2\} j ∈ { 1 , 2 } .
In the next lemma the principal symbols of W p ± W_p^\pm W p ± are specified.
Lemma 4.3. One has W p ± ∈ Op S N 2 0 ( Σ p ) W_p^\pm\in\operatorname{Op}S^0_{\frac{N}{2}}(\Sigma_p) W p ± ∈ Op S 2 N 0 ( Σ p ) and the following is true.
(a) For n = 2 n=2 n = 2 the principal symbols of W p ± W_p^\pm W p ± at x ∈ Σ p x\in\Sigma_p x ∈ Σ p with respect to any properly oriented arc-length parametrization (see Definition 4.2 (a)) are
σ W p + ( x , ξ ) = − i 2 μ ( x ) ‾ sgn ξ , σ W p − ( x , ξ ) = i 2 μ ( x ) sgn ξ , (67) \sigma_{W_p^+}(x,\xi)=-\frac{\mathrm{i}}{2}\overline{\mu(x)}\operatorname{sgn}\xi,\qquad\sigma_{W_p^-}(x,\xi)=\frac{\mathrm{i}}{2}\mu(x)\operatorname{sgn}\xi, \tag*{(67)} σ W p + ( x , ξ ) = − 2 i μ ( x ) sgn ξ , σ W p − ( x , ξ ) = 2 i μ ( x ) sgn ξ , (67) where μ : Σ p → C \mu:\Sigma_p\to\mathbb{C} μ : Σ p → C is the standard complexification of ν Σ \nu^\Sigma ν Σ given by
μ ( x ) : = ν 1 Σ ( x ) + i ν 2 Σ ( x ) , x ∈ Σ p . (68) \mu(x):=\nu_1^\Sigma(x)+\mathrm{i}\nu_2^\Sigma(x),\qquad x\in\Sigma_p. \tag*{(68)} μ ( x ) := ν 1 Σ ( x ) + i ν 2 Σ ( x ) , x ∈ Σ p . (68) (b) For n = 3 n=3 n = 3 the principal symbols of W p ± W_p^\pm W p ± at x ∈ Σ p x\in\Sigma_p x ∈ Σ p with respect to normal coordinates near x x x (see Definition 4.2 (b)) are
σ W p ± ( x , ξ ) = 1 2 σ ⋅ ( ξ 1 e 1 + ξ 2 e 2 ∣ ξ ∣ ) . (69) \sigma_{W_p^\pm}(x,\xi)=\frac{1}{2}\sigma\cdot\left(\frac{\xi_1e_1+\xi_2e_2}{|\xi|}\right). \tag*{(69)} σ W p ± ( x , ξ ) = 2 1 σ ⋅ ( ∣ ξ ∣ ξ 1 e 1 + ξ 2 e 2 ) . (69)
Proof. We will compute the principal symbols of W p ± W_p^\pm W p ± ; the fact that W p ± ∈ Op S N 2 0 ( Σ p ) W_p^\pm\in\operatorname{Op}S^0_{\frac{N}{2}}(\Sigma_p) W p ± ∈ Op S 2 N 0 ( Σ p ) will follow as a byproduct of these considerations.
(a) In the following it is convenient to complexify R 2 \mathbb{R}^2 R 2 by ( x 1 , x 2 ) ≃ x 1 + i x 2 (x_1,x_2)\simeq x_1+\mathrm{i}x_2 ( x 1 , x 2 ) ≃ x 1 + i x 2 . Let x ∈ Σ p x\in\Sigma_p x ∈ Σ p and γ : I → Σ p \gamma:I\to\Sigma_p γ : I → Σ p with I = ( − ℓ p 2 , ℓ p 2 ) I=\left(-\frac{\ell_p}{2},\frac{\ell_p}{2}\right) I = ( − 2 ℓ p , 2 ℓ p ) be a properly oriented complexified arc-length parametrization of Σ p \Sigma_p Σ p with γ ( 0 ) = x \gamma(0)=x γ ( 0 ) = x . Due to the arc-length and orientation conditions one has for all $t\in I that
1 γ ′ ( t ) ‾ = γ ′ ( t ) and γ ′ ( t ) = i μ ( γ ( t ) ) . \frac{1}{\overline{\gamma'(t)}}=\gamma'(t)\quad\text{and}\quad\gamma'(t)=\mathrm{i}\mu(\gamma(t)). γ ′ ( t ) 1 = γ ′ ( t ) and γ ′ ( t ) = i μ ( γ ( t )) .
Let J : = ( − ℓ p 4 , ℓ p 4 ) J:=\left(-\frac{\ell_p}{4},\frac{\ell_p}{4}\right) J := ( − 4 ℓ p , 4 ℓ p ) and V : = γ ( J ) V:=\gamma(J) V := γ ( J ) , which is an open neighborhood of x x x in Σ p \Sigma_p Σ p , pick ψ ∈ C c ∞ ( V ) \psi\in C^\infty_c(V) ψ ∈ C c ∞ ( V ) , and let f ∈ C ∞ ( Σ p ) f\in C^\infty(\Sigma_p) f ∈ C ∞ ( Σ p ) . Then for any s ∈ I s\in I s ∈ I one has
[ W p + ( ψ f ) ] ( γ ( s ) ) = i 2 π p.v. ∫ I 1 γ ( s ) − γ ( t ) ‾ ( ψ f ) ( γ ( t ) ) d t = i 2 π p.v. ∫ I a ( s , t ) ( ψ f ) ( γ ( t ) ) s − t d t \begin{aligned}
[W_p^+(\psi f)](\gamma(s))=\frac{\mathrm{i}}{2\pi}\operatorname{p.v.}\int_I\frac{1}{\overline{\gamma(s)-\gamma(t)}}(\psi f)(\gamma(t))\,\mathrm{d}t \\
&=\frac{\mathrm{i}}{2\pi}\operatorname{p.v.}\int_Ia(s,t)\frac{(\psi f)(\gamma(t))}{s-t}\,\mathrm{d}t
\end{aligned} [ W p + ( ψ f )] ( γ ( s )) = 2 π i p.v. ∫ I γ ( s ) − γ ( t ) 1 ( ψ f ) ( γ ( t )) d t = 2 π i p.v. ∫ I a ( s , t ) s − t ( ψ f ) ( γ ( t )) d t
with the function a : I × I → C a:I\times I\to\mathbb{C} a : I × I → C defined by
a ( s , t ) : = { s − t γ ( s ) − γ ( t ) ‾ , s ≠ t , 1 γ ′ ( s ) ‾ , s = t , = { s − t γ ( s ) − γ ( t ) ‾ , s ≠ t , − i μ ( γ ( s ) ) ‾ , s = t . a(s,t):=
\begin{cases}
\dfrac{s-t}{\overline{\gamma(s)-\gamma(t)}}, & s\ne t,\\[6pt]
\dfrac{1}{\overline{\gamma'(s)}}, & s=t,
\end{cases}
=
\begin{cases}
\dfrac{s-t}{\overline{\gamma(s)-\gamma(t)}}, & s\ne t,\\[6pt]
-\mathrm{i}\overline{\mu(\gamma(s))}, & s=t.
\end{cases} a ( s , t ) := ⎩ ⎨ ⎧ γ ( s ) − γ ( t ) s − t , γ ′ ( s ) 1 , s = t , s = t , = ⎩ ⎨ ⎧ γ ( s ) − γ ( t ) s − t , − i μ ( γ ( s )) , s = t , s = t .
Hence, we can further rewrite
[ W p + ( ψ f ) ] ( γ ( s ) ) = 1 2 π μ ‾ ( γ ( s ) ) p.v. ∫ I ( ψ f ) ( γ ( t ) ) s − t d t + ∫ I b ( s , t ) ( ψ f ) ( γ ( t ) ) d t (70) [W_p^+(\psi f)](\gamma(s)) = \frac{1}{2\pi}\overline{\mu}(\gamma(s))\,\operatorname{p.v.}\int_I \frac{(\psi f)(\gamma(t))}{s-t}\,\mathrm{d}t+\int_I b(s,t)(\psi f)(\gamma(t))\,\mathrm{d}t \tag*{(70)} [ W p + ( ψ f )] ( γ ( s )) = 2 π 1 μ ( γ ( s )) p.v. ∫ I s − t ( ψ f ) ( γ ( t )) d t + ∫ I b ( s , t ) ( ψ f ) ( γ ( t )) d t (70)
with
b ( s , t ) : = i 2 π { a ( s , t ) − a ( s , s ) s − t , s ≠ t , − ∂ 2 a ( s , s ) , s = t . b(s,t):=\frac{\mathrm{i}}{2\pi}\begin{cases}\dfrac{a(s,t)-a(s,s)}{s-t},&s\ne t,\\-\partial_2a(s,s),&s=t.\end{cases} b ( s , t ) := 2 π i ⎩ ⎨ ⎧ s − t a ( s , t ) − a ( s , s ) , − ∂ 2 a ( s , s ) , s = t , s = t .
As γ \gamma γ is injective on I I I , γ ′ ≠ 0 \gamma'\ne0 γ ′ = 0 , and γ ( s ) \gamma(s) γ ( s ) stays away from γ ( J ‾ ) \gamma(\overline{J}) γ ( J ) as s → ± ℓ p 2 s\to\pm\frac{\ell_p}{2} s → ± 2 ℓ p , one has a ∈ C ∞ ( I × J ) a\in C^\infty(I\times J) a ∈ C ∞ ( I × J ) . Hence, b ∈ C ∞ ( I × J ) b\in C^\infty(I\times J) b ∈ C ∞ ( I × J ) and thus, as ψ ∘ γ \psi\circ\gamma ψ ∘ γ has compact support in J J J , the second summand on the right hand side in (70) defines an operator in Op S 1 − ∞ ( Σ p ) \operatorname{Op}S_1^{-\infty}(\Sigma_p) Op S 1 − ∞ ( Σ p ) .
Furthermore, the first summand in (70) can be written as the convolution
1 2 π μ ‾ ( γ ( s ) ) p.v. ∫ I ( ψ f ) ( γ ( t ) ) s − t d t = 1 2 π μ ‾ ( γ ( s ) ) ( p.v. 1 s ) ∗ ( ( ψ f ) ∘ γ ) . \frac{1}{2\pi}\overline{\mu}(\gamma(s))\,\operatorname{p.v.}\int_I\frac{(\psi f)(\gamma(t))}{s-t}\,\mathrm{d}t=\frac{1}{2\pi}\overline{\mu}(\gamma(s))\left(\operatorname{p.v.}\frac{1}{s}\right)\ast((\psi f)\circ\gamma). 2 π 1 μ ( γ ( s )) p.v. ∫ I s − t ( ψ f ) ( γ ( t )) d t = 2 π 1 μ ( γ ( s )) ( p.v. s 1 ) ∗ (( ψ f ) ∘ γ ) .
Recall that the Fourier transform of the distribution p.v. 1 s \operatorname{p.v.}\frac{1}{s} p.v. s 1 on R \mathbb{R} R is given by
p.v. 1 s ^ = − i π sgn ξ . \widehat{\operatorname{p.v.}\frac{1}{s}}=-\mathrm{i}\pi\operatorname{sgn}\xi. p.v. s 1 = − i π sgn ξ .
Therefore, we conclude that the principal symbol of W p + W_p^+ W p + at x x x (corresponding to γ ( 0 ) \gamma(0) γ ( 0 ) ) is
σ W p + ( x , ξ ) = 1 2 π μ ‾ ( x ) p.v. 1 s ^ = − i 2 μ ‾ ( x ) sgn ξ . \sigma_{W_p^+}(x,\xi)=\frac{1}{2\pi}\overline{\mu}(x)\widehat{\operatorname{p.v.}\frac{1}{s}}=-\frac{\mathrm{i}}{2}\overline{\mu}(x)\operatorname{sgn}\xi. σ W p + ( x , ξ ) = 2 π 1 μ ( x ) p.v. s 1 = − 2 i μ ( x ) sgn ξ .
From this and W p − f ‾ = − W p + f ‾ \overline{W_p^-f}=-W_p^+\overline{f} W p − f = − W p + f we obtain that the principal symbol of W p − W_p^- W p − at x x x is σ W p − ( x , ξ ) = − σ W p + ( x , − ξ ) \sigma_{W_p^-}(x,\xi)=-\sigma_{W_p^+}(x,-\xi) σ W p − ( x , ξ ) = − σ W p + ( x , − ξ ) . Hence, all claims in (a) are shown.
(b) is shown in [20 Theorem 8] . It should be noted that in [20 Remark 6] there was a sign mistake: the curvatures were defined with the opposite sign, while all the subsequent computations actually used the definition corresponding to (66) . □
As a direct corollary of the expressions (67) and (69) we obtain the following facts.
Corollary 4.4. W p + W p − − 1 4 I , W p − W p + − 1 4 I ∈ Op S 1 − 1 ( Σ p ) W_p^+W_p^--\frac{1}{4}I,W_p^-W_p^+-\frac{1}{4}I\in\operatorname{Op}S_1^{-1}(\Sigma_p) W p + W p − − 4 1 I , W p − W p + − 4 1 I ∈ Op S 1 − 1 ( Σ p ) .
In order to compute the essential spectrum of A η , τ A^{\eta,\tau} A η , τ for η p 2 − τ p 2 = 4 \eta_p^2-\tau_p^2=4 η p 2 − τ p 2 = 4 we need the following finer microlocal properties of its blocks. Recall that for n = 3 n=3 n = 3 the operators W p ± W_p^\pm W p ± are equal, which allows us to set
W p : = W p + = W p − , W_p:=W_p^+=W_p^-, W p := W p + = W p − ,
and that normal coordinates near x x x were introduced in Definition 4.2 (b).
Lemma 4.5. For the singular integral operators W p ± W_p^\pm W p ± the following is true:
(a) Let n = 2 n=2 n = 2 and μ \mu μ be the complexification of ν Σ \nu^\Sigma ν Σ as in (68) . Then, it holds that
( W p + μ ) 2 = − 1 4 I = ( μ ‾ W p − ) 2 (71) (W_p^+\mu)^2=-\frac{1}{4}I=(\overline{\mu}W_p^-)^2 \tag*{(71)} ( W p + μ ) 2 = − 4 1 I = ( μ W p − ) 2 (71) W p + μ + μ ‾ W p − , W p + W p − − 1 4 I , W p − W p + − 1 4 I ∈ Op S 1 − ∞ ( Σ p ) . (72) W_p^+\mu+\overline{\mu}W_p^-,\quad W_p^+W_p^--\frac{1}{4}I,\quad W_p^-W_p^+-\frac{1}{4}I\in\operatorname{Op}S_1^{-\infty}(\Sigma_p). \tag*{(72)} W p + μ + μ W p − , W p + W p − − 4 1 I , W p − W p + − 4 1 I ∈ Op S 1 − ∞ ( Σ p ) . (72)
(b) Let n = 3 n = 3 n = 3 . Then, it holds that
( ( σ ⋅ ν Σ ) W p ) 2 = ( W p ( σ ⋅ ν Σ ) ) 2 = − 1 4 I 2 . (73) \left((\sigma\cdot\nu^{\Sigma})W_p\right)^2=\left(W_p(\sigma\cdot\nu^{\Sigma})\right)^2=-\frac{1}{4}I_2. \tag*{(73)} ( ( σ ⋅ ν Σ ) W p ) 2 = ( W p ( σ ⋅ ν Σ ) ) 2 = − 4 1 I 2 . (73)
Furthermore, the operator
Z p : = W p ( σ ⋅ ν Σ ) + ( σ ⋅ ν Σ ) W p (74) Z_p:=W_p(\sigma\cdot\nu^{\Sigma})+(\sigma\cdot\nu^{\Sigma})W_p \tag*{(74)} Z p := W p ( σ ⋅ ν Σ ) + ( σ ⋅ ν Σ ) W p (74)
belongs to Op S 1 − 1 ( Σ p ) \operatorname{Op}S_1^{-1}(\Sigma_p) Op S 1 − 1 ( Σ p ) , and for any x ∈ Σ p x\in\Sigma_p x ∈ Σ p its principal symbol with respect to normal coordinates near x x x is given by
σ Z p ( x , ξ ) = 1 2 ( κ 1 ( x ) − κ 2 ( x ) ) ( σ ⋅ ν Σ ( x ) ) ξ 1 ξ 2 ∣ ξ ∣ 3 , (75) \sigma_{Z_p}(x,\xi)=\frac{1}{2}\left(\kappa_1(x)-\kappa_2(x)\right)(\sigma\cdot\nu^{\Sigma}(x))\frac{\xi_1\xi_2}{\lvert\xi\rvert^3}, \tag*{(75)} σ Z p ( x , ξ ) = 2 1 ( κ 1 ( x ) − κ 2 ( x ) ) ( σ ⋅ ν Σ ( x )) ∣ ξ ∣ 3 ξ 1 ξ 2 , (75)
where κ 1 ( x ) \kappa_1(x) κ 1 ( x ) and κ 2 ( x ) \kappa_2(x) κ 2 ( x ) are the principal curvatures of Σ p \Sigma_p Σ p in x x x .
Proof. (a) Let γ : I → Σ p \gamma:I\to\Sigma_p γ : I → Σ p with I = ( − ℓ p 2 , ℓ p 2 ) I=\left(-\frac{\ell_p}{2},\frac{\ell_p}{2}\right) I = ( − 2 ℓ p , 2 ℓ p ) be a properly oriented complexified arc-length parametrization of Σ p \Sigma_p Σ p . Note that the operator T : = i W p + μ T:=\mathrm{i}W_p^+\mu T := i W p + μ acts on φ ∈ C ∞ ( Σ p ) \varphi\in C^\infty(\Sigma_p) φ ∈ C ∞ ( Σ p ) as
T φ ( γ ( s ) ) = i 2 π p.v. ∫ I γ 1 ′ ( t ) + i γ 2 ′ ( t ) γ 1 ( s ) + i γ 2 ( s ) − ( γ 1 ( t ) + i γ 2 ( t ) ) φ ( γ ( t ) ) d t = i 2 π p.v. ∫ Σ ~ p φ ~ ( ζ ) z − ζ d ζ , T\varphi(\gamma(s))=\frac{\mathrm{i}}{2\pi}\operatorname{p.v.}\int_I\frac{\gamma_1'(t)+\mathrm{i}\gamma_2'(t)}{\gamma_1(s)+\mathrm{i}\gamma_2(s)-(\gamma_1(t)+\mathrm{i}\gamma_2(t))}\varphi(\gamma(t))\,\mathrm{d}t=\frac{\mathrm{i}}{2\pi}\operatorname{p.v.}\int_{\widetilde{\Sigma}_p}\frac{\widetilde{\varphi}(\zeta)}{z-\zeta}\,\mathrm{d}\zeta, T φ ( γ ( s )) = 2 π i p.v. ∫ I γ 1 ( s ) + i γ 2 ( s ) − ( γ 1 ( t ) + i γ 2 ( t )) γ 1 ′ ( t ) + i γ 2 ′ ( t ) φ ( γ ( t )) d t = 2 π i p.v. ∫ Σ p z − ζ φ ( ζ ) d ζ ,
where Σ ~ p \widetilde{\Sigma}_p Σ p , φ ~ \widetilde{\varphi} φ , and z z z are the complexified versions of Σ p \Sigma_p Σ p , φ \varphi φ , and γ ( s ) \gamma(s) γ ( s ) , respectively, i.e.
Σ ~ p = { x 1 + i x 2 : ( x 1 , x 2 ) ∈ Σ p } , φ ~ ( x 1 + i x 2 ) = φ ( x 1 , x 2 ) , z = γ 1 ( s ) + i γ 2 ( s ) . \widetilde{\Sigma}_p=\{x_1+\mathrm{i}x_2:(x_1,x_2)\in\Sigma_p\},\qquad\widetilde{\varphi}(x_1+\mathrm{i}x_2)=\varphi(x_1,x_2),\qquad z=\gamma_1(s)+\mathrm{i}\gamma_2(s). Σ p = { x 1 + i x 2 : ( x 1 , x 2 ) ∈ Σ p } , φ ( x 1 + i x 2 ) = φ ( x 1 , x 2 ) , z = γ 1 ( s ) + i γ 2 ( s ) .
Hence, T T T is the well-known Hilbert transform on Σ p \Sigma_p Σ p (i.e. the direct value of the respective Cauchy integral), see, e.g., [33 §12] , which is known to satisfy 4 T 2 = I 4T^2=I 4 T 2 = I , see e.g. [33 §27] . This gives the first identity in (71) . We further note that μ ‾ W p − \overline{\mu}W_p^- μ W p − is the adjoint of W p + μ W_p^+\mu W p + μ in L 2 ( Σ p ) L^2(\Sigma_p) L 2 ( Σ p ) , which gives the second identity in (71) . The operator K : = W p + μ + μ ‾ W p − K:=W_p^+\mu+\overline{\mu}W_p^- K := W p + μ + μ W p − is known (modulo constant factors) as the Kerzman-Stein operator: it has a smooth integral kernel, see, e.g., [14 Section 5] , which gives the inclusion K ∈ Op S 1 − ∞ ( Σ p ) K\in\operatorname{Op}S_1^{-\infty}(\Sigma_p) K ∈ Op S 1 − ∞ ( Σ p ) . Finally we have
W p + W p − = ( W p + μ ) ( μ ‾ W p − ) = W p + μ ( − W p + μ + K ) = xref‘(4.11)‘ 1 4 I + W p + μ K , W_p^+W_p^-=(W_p^+\mu)(\overline{\mu}W_p^-)=W_p^+\mu(-W_p^+\mu+K)\mathrel{\overset{\text{\scriptsize{xref}`(4.11)`}}{=}}\frac{1}{4}I+W_p^+\mu K, W p + W p − = ( W p + μ ) ( μ W p − ) = W p + μ ( − W p + μ + K ) = xref ‘(4.11)‘ 4 1 I + W p + μ K ,
W p − W p + = μ ( μ ‾ W p − ) ( W p + μ ) μ ‾ = μ ( − W p + μ + K ) ( W p + μ ) μ ‾ = xref‘(4.11)‘ 1 4 I + μ K W p + , W_p^-W_p^+=\mu(\overline{\mu}W_p^-)(W_p^+\mu)\overline{\mu}=\mu(-W_p^+\mu+K)(W_p^+\mu)\overline{\mu}\mathrel{\overset{\text{\scriptsize{xref}`(4.11)`}}{=}}\frac{1}{4}I+\mu K W_p^+, W p − W p + = μ ( μ W p − ) ( W p + μ ) μ = μ ( − W p + μ + K ) ( W p + μ ) μ = xref ‘(4.11)‘ 4 1 I + μ K W p + ,
while W p + μ K , μ K W p + ∈ Op S 1 − ∞ ( Σ p ) W_p^+\mu K,\mu K W_p^+\in\operatorname{Op}S_1^{-\infty}(\Sigma_p) W p + μ K , μ K W p + ∈ Op S 1 − ∞ ( Σ p ) due to W p + , μ ∈ Op S 1 0 ( Σ p ) W_p^+,\mu\in\operatorname{Op}S_1^0(\Sigma_p) W p + , μ ∈ Op S 1 0 ( Σ p ) and K ∈ Op S 1 − ∞ ( Σ p ) K\in\operatorname{Op}S_1^{-\infty}(\Sigma_p) K ∈ Op S 1 − ∞ ( Σ p ) . This shows (72) .
(b) For the relation (73) see [20 Section 2, equation (11)] , and the expression (75) is derived in [20 Theorem 8] ; our situation corresponds to λ = m \lambda=m λ = m in [20 ] . □
In the following corollary we state the pseudodifferential properties of the operator C z C_z C z defined by (53) . Let S Σ S_\Sigma S Σ and W Σ ± W_\Sigma^\pm W Σ ± be the integral operators that are defined as in (61) and (65) with the integral over Σ \Sigma Σ instead of Σ p \Sigma_p Σ p , that is,
S Σ φ ( x ) = ∫ Σ E 0 ( x − y ) φ ( y ) d s ( y ) , x ∈ Σ , φ ∈ C ∞ ( Σ ) , (76) S_\Sigma\varphi(x)=\int_\Sigma E_0(x-y)\varphi(y)\,\mathrm{d}s(y),\qquad x\in\Sigma,\quad\varphi\in C^\infty(\Sigma), \tag*{(76)} S Σ φ ( x ) = ∫ Σ E 0 ( x − y ) φ ( y ) d s ( y ) , x ∈ Σ , φ ∈ C ∞ ( Σ ) , (76)
and
W Σ ± φ ( x ) = p.v. ∫ Σ w ± ( x − y ) φ ( y ) d s ( y ) , x ∈ Σ , φ ∈ C ∞ ( Σ ) . (77) W_\Sigma^\pm\varphi(x)=\operatorname{p.v.}\int_\Sigma w^\pm(x-y)\varphi(y)\,\mathrm{d}s(y),\qquad x\in\Sigma,\quad\varphi\in C^\infty(\Sigma). \tag*{(77)} W Σ ± φ ( x ) = p.v. ∫ Σ w ± ( x − y ) φ ( y ) d s ( y ) , x ∈ Σ , φ ∈ C ∞ ( Σ ) . (77)
It is important to note that one has
S Σ = ⨁ p = 1 P S p mod Op S 1 − ∞ ( Σ ) and W Σ ± = ⨁ p = 1 P W p ± mod Op S N / 2 − ∞ ( Σ ) , (78) S_{\Sigma}=\bigoplus_{p=1}^{P}S_p\ \operatorname{mod}\ \operatorname{Op}S_1^{-\infty}(\Sigma)\quad\text{and}\quad W_{\Sigma}^{\pm}=\bigoplus_{p=1}^{P}W_p^{\pm}\ \operatorname{mod}\ \operatorname{Op}S_{N/2}^{-\infty}(\Sigma), \tag*{(78)} S Σ = p = 1 ⨁ P S p mod Op S 1 − ∞ ( Σ ) and W Σ ± = p = 1 ⨁ P W p ± mod Op S N /2 − ∞ ( Σ ) , (78)
since for p ≠ q p\ne q p = q operators of the form (with E 0 , w ± ∈ C ∞ ( R n ∖ { 0 } ) \mathcal{E}_0,w^{\pm}\in C^{\infty}(\mathbb{R}^{n}\setminus\{0\}) E 0 , w ± ∈ C ∞ ( R n ∖ { 0 }) )
S p , q ψ ( x ) = ∫ Σ p E 0 ( x − y ) ψ ( y ) d s ( y ) , x ∈ Σ q , ψ ∈ C ∞ ( Σ p ) , S_{p,q}\psi(x)=\int_{\Sigma_p}\mathcal{E}_0(x-y)\psi(y)\,\mathrm{d}s(y),\quad x\in\Sigma_q,\quad\psi\in C^{\infty}(\Sigma_p), S p , q ψ ( x ) = ∫ Σ p E 0 ( x − y ) ψ ( y ) d s ( y ) , x ∈ Σ q , ψ ∈ C ∞ ( Σ p ) ,
W p , q ± ψ ( x ) = p.v. ∫ Σ p w ± ( x − y ) ψ ( y ) d s ( y ) , x ∈ Σ q , ψ ∈ C ∞ ( Σ p ) , W_{p,q}^{\pm}\psi(x)=\operatorname{p.v.}\int_{\Sigma_p}w^{\pm}(x-y)\psi(y)\,\mathrm{d}s(y),\quad x\in\Sigma_q,\quad\psi\in C^{\infty}(\Sigma_p), W p , q ± ψ ( x ) = p.v. ∫ Σ p w ± ( x − y ) ψ ( y ) d s ( y ) , x ∈ Σ q , ψ ∈ C ∞ ( Σ p ) ,
are smoothing. Hence, it follows from Lemma 4.1 and Lemma 4.3 that S Σ ∈ Op S 1 − 1 ( Σ ) S_{\Sigma}\in\operatorname{Op}S_1^{-1}(\Sigma) S Σ ∈ Op S 1 − 1 ( Σ ) and W Σ ± ∈ Op S N / 2 0 ( Σ ) W_{\Sigma}^{\pm}\in\operatorname{Op}S_{N/2}^{0}(\Sigma) W Σ ± ∈ Op S N /2 0 ( Σ ) . With these notations we have the following result:
Corollary 4.6. Let z ∈ res A z\in\operatorname{res}A z ∈ res A . Then C z ∈ Op S N 0 ( Σ ) C_z\in\operatorname{Op}S_N^{0}(\Sigma) C z ∈ Op S N 0 ( Σ ) and
C z = ( ( z + m ) S Σ ⊗ I N / 2 W Σ + W Σ − ( z − m ) S Σ ⊗ I N / 2 ) mod Op S N − 2 ( Σ ) . (79) C_z=
\begin{pmatrix}
(z+m)S_{\Sigma}\otimes I_{N/2} & W_{\Sigma}^{+}\\
W_{\Sigma}^{-} & (z-m)S_{\Sigma}\otimes I_{N/2}
\end{pmatrix}
\quad\operatorname{mod}\ \operatorname{Op}S_N^{-2}(\Sigma). \tag*{(79)} C z = ( ( z + m ) S Σ ⊗ I N /2 W Σ − W Σ + ( z − m ) S Σ ⊗ I N /2 ) mod Op S N − 2 ( Σ ) . (79)
Proof. Using the notation in Theorem 2.5 and Lemma 2.9 one has for x , y ∈ Σ x,y\in\Sigma x , y ∈ Σ , x ≠ y x\ne y x = y ,
G z ( x , y ) = E z ( x − y ) + T z ( x , y ) = H z ( x − y ) + χ ( x − y ) ( E z ( x − y ) − H z ( x − y ) ) + T z ( x , y ) = H z ( x − y ) + K z ( x − y ) + T z ( x , y ) , \begin{aligned}
G_z(x,y)=E_z(x-y)+T_z(x,y) \\
&=H_z(x-y)+\chi(x-y)\bigl(E_z(x-y)-H_z(x-y)\bigr)+T_z(x,y) \\
&=H_z(x-y)+K_z(x-y)+T_z(x,y),
\end{aligned} G z ( x , y ) = E z ( x − y ) + T z ( x , y ) = H z ( x − y ) + χ ( x − y ) ( E z ( x − y ) − H z ( x − y ) ) + T z ( x , y ) = H z ( x − y ) + K z ( x − y ) + T z ( x , y ) ,
where χ ∈ C c ∞ ( R n ) \chi\in C_c^{\infty}(\mathbb{R}^{n}) χ ∈ C c ∞ ( R n ) is chosen such that χ ( x − y ) = 1 \chi(x-y)=1 χ ( x − y ) = 1 for x , y ∈ Σ x,y\in\Sigma x , y ∈ Σ . Let K Σ , z K_{\Sigma,z} K Σ , z and T ~ Σ , z \widetilde{T}_{\Sigma,z} T Σ , z be defined as in (57) . Then it follows that
C z = ( ( z + m ) S Σ ⊗ I N / 2 W Σ + W Σ − ( z − m ) S Σ ⊗ I N / 2 ) + K Σ , z + T ~ Σ , z . C_z=
\begin{pmatrix}
(z+m)S_{\Sigma}\otimes I_{N/2} & W_{\Sigma}^{+}\\
W_{\Sigma}^{-} & (z-m)S_{\Sigma}\otimes I_{N/2}
\end{pmatrix}
+K_{\Sigma,z}+\widetilde{T}_{\Sigma,z}. C z = ( ( z + m ) S Σ ⊗ I N /2 W Σ − W Σ + ( z − m ) S Σ ⊗ I N /2 ) + K Σ , z + T Σ , z .
As it was shown in the proof of Lemma 3.5 one has K Σ , z ∈ Op S N − 2 ( Σ ) K_{\Sigma,z}\in\operatorname{Op}S_N^{-2}(\Sigma) K Σ , z ∈ Op S N − 2 ( Σ ) and T ~ Σ , z ∈ Op S N − ∞ ( Σ ) \widetilde{T}_{\Sigma,z}\in\operatorname{Op}S_N^{-\infty}(\Sigma) T Σ , z ∈ Op S N − ∞ ( Σ ) . Since S Σ ∈ Op S 1 − 1 ( Σ ) S_{\Sigma}\in\operatorname{Op}S_1^{-1}(\Sigma) S Σ ∈ Op S 1 − 1 ( Σ ) and W Σ ± ∈ Op S N / 2 0 ( Σ ) W_{\Sigma}^{\pm}\in\operatorname{Op}S_{N/2}^{0}(\Sigma) W Σ ± ∈ Op S N /2 0 ( Σ ) , this shows the claim. □
Recall that Θ η , τ \Theta^{\eta,\tau} Θ η , τ is given by (60) and that M z M_z M z denotes the Weyl function associated with the boundary triple ( L 2 ( Σ , C N ) , Π 0 , Π 1 ) (L^2(\Sigma,\mathbb{C}^{N}),\Pi_0,\Pi_1) ( L 2 ( Σ , C N ) , Π 0 , Π 1 ) , see (59) . Next, we provide a decomposition of
Θ η , τ − M z = − Λ [ 1 η 2 − τ 2 ( η I N − τ β ) + C z ] Λ (80) \Theta^{\eta,\tau}-M_z=-\Lambda\left[\frac{1}{\eta^2-\tau^2}(\eta I_N-\tau\beta)+C_z\right]\Lambda \tag*{(80)} Θ η , τ − M z = − Λ [ η 2 − τ 2 1 ( η I N − τ β ) + C z ] Λ (80)
into parts, which will be easier to analyze in the following. Recall that Λ p ∈ Op S 1 / 2 1 / 2 ( Σ p ) \Lambda_p\in\operatorname{Op}S_{1/2}^{1/2}(\Sigma_p) Λ p ∈ Op S 1/2 1/2 ( Σ p ) and W p ± ∈ Op S N / 2 0 ( Σ p ) W_p^{\pm}\in\operatorname{Op}S_{N/2}^{0}(\Sigma_p) W p ± ∈ Op S N /2 0 ( Σ p ) are defined by (62) and (65) , respectively. For p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } such that ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p|\ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ we consider the operators
L p : = − ( ( 1 η p + τ p Λ p 2 + m ) ⊗ I N / 2 Λ p W p + Λ p Λ p W p − Λ p ( 1 η p − τ p Λ p 2 − m ) ⊗ I N / 2 ) ∈ Op S N 1 ( Σ p ) (81) L_p:=-\begin{pmatrix}
\left(\frac{1}{\eta_p+\tau_p}\Lambda_p^2+m\right)\otimes I_{N/2} & \Lambda_pW_p^{+}\Lambda_p\\
\Lambda_pW_p^{-}\Lambda_p & \left(\frac{1}{\eta_p-\tau_p}\Lambda_p^2-m\right)\otimes I_{N/2}
\end{pmatrix}
\in\operatorname{Op}S_N^{1}(\Sigma_p) \tag*{(81)} L p := − ( η p + τ p 1 Λ p 2 + m ) ⊗ I N /2 Λ p W p − Λ p Λ p W p + Λ p ( η p − τ p 1 Λ p 2 − m ) ⊗ I N /2 ∈ Op S N 1 ( Σ p ) (81)
and have the following result.
Proposition 4.7. Let Λ = ⨁ p = 1 P Λ p \Lambda=\bigoplus_{p=1}^{P}\Lambda_p Λ = ⨁ p = 1 P Λ p with the special choice of Λ p \Lambda_p Λ p from (62) , assume that ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p|\ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ for all p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } , and let L p L_p L p be as above. Then, for any z ∈ res A z\in\operatorname{res}A z ∈ res A one has
Θ η , τ − M z = ⨁ p = 1 P ( L p − z ) mod Op S N − 1 ( Σ ) . (82) \Theta^{\eta,\tau}-M_z=\bigoplus_{p=1}^{P}(L_p-z)\ \operatorname{mod}\ \operatorname{Op}S_N^{-1}(\Sigma). \tag*{(82)} Θ η , τ − M z = p = 1 ⨁ P ( L p − z ) mod Op S N − 1 ( Σ ) . (82) In particular, each L p L_p L p viewed as an operator in L 2 ( Σ p , C N ) L^2(\Sigma_p,\mathbb{C}^N) L 2 ( Σ p , C N ) on its maximal domain is self-adjoint, H 1 ( Σ p , C N ) H^1(\Sigma_p,\mathbb{C}^N) H 1 ( Σ p , C N ) is a core for L p L_p L p , and ⨁ p = 1 P L p \bigoplus_{p=1}^{P}L_p ⨁ p = 1 P L p defined on dom Θ η , τ \operatorname{dom}\Theta^{\eta,\tau} dom Θ η , τ is self-adjoint in L 2 ( Σ , C N ) L^2(\Sigma,\mathbb{C}^N) L 2 ( Σ , C N ) .
Proof. Let z ∈ res A z\in\operatorname{res}A z ∈ res A be fixed. Using the block representation from (79) and the fact that Λ ∈ Op S N 1 2 ( Σ ) \Lambda\in\operatorname{Op}S_N^{\frac{1}{2}}(\Sigma) Λ ∈ Op S N 2 1 ( Σ ) one gets that
Θ η , τ − M z = − Λ [ 1 η 2 − τ 2 ( η I N − τ β ) + C z ] Λ = − Λ ( ( 1 η + τ + ( z + m ) S Σ ) ⊗ I N 2 W Σ + W Σ − ( 1 η − τ + ( z − m ) S Σ ) ⊗ I N 2 ) Λ mod Op S N − 1 ( Σ ) . (83) \begin{aligned}
\Theta^{\eta,\tau}-M_z=-\Lambda\left[\frac{1}{\eta^2-\tau^2}(\eta I_N-\tau\beta)+C_z\right]\Lambda\\
&=-\Lambda\begin{pmatrix}
\left(\dfrac{1}{\eta+\tau}+(z+m)S_\Sigma\right)\otimes I_{\frac{N}{2}} & W_\Sigma^+\\
W_\Sigma^- & \left(\dfrac{1}{\eta-\tau}+(z-m)S_\Sigma\right)\otimes I_{\frac{N}{2}}
\end{pmatrix}\Lambda\ \operatorname{mod}\ \operatorname{Op}S_N^{-1}(\Sigma). \tag*{(83)}
\end{aligned} Θ η , τ − M z = − Λ [ η 2 − τ 2 1 ( η I N − τ β ) + C z ] Λ = − Λ ( η + τ 1 + ( z + m ) S Σ ) ⊗ I 2 N W Σ − W Σ + ( η − τ 1 + ( z − m ) S Σ ) ⊗ I 2 N Λ mod Op S N − 1 ( Σ ) . (83)
According to (78) we have
S Σ = ⨁ p = 1 P S p mod Op S 1 − ∞ ( Σ ) and W Σ ± = ⨁ p = 1 P W p ± mod Op S N 2 − ∞ ( Σ ) , (84) S_\Sigma=\bigoplus_{p=1}^{P}S_p\ \operatorname{mod}\ \operatorname{Op}S_1^{-\infty}(\Sigma)
\qquad\text{and}\qquad
W_\Sigma^\pm=\bigoplus_{p=1}^{P}W_p^\pm\ \operatorname{mod}\ \operatorname{Op}S_{\frac{N}{2}}^{-\infty}(\Sigma), \tag*{(84)} S Σ = p = 1 ⨁ P S p mod Op S 1 − ∞ ( Σ ) and W Σ ± = p = 1 ⨁ P W p ± mod Op S 2 N − ∞ ( Σ ) , (84)
and in view of (83) this implies the decomposition
Θ η , τ − M z = ⨁ p = 1 P Q p , z mod Op S N − 1 ( Σ ) (85) \Theta^{\eta,\tau}-M_z=\bigoplus_{p=1}^{P}Q_{p,z}\ \operatorname{mod}\ \operatorname{Op}S_N^{-1}(\Sigma) \tag*{(85)} Θ η , τ − M z = p = 1 ⨁ P Q p , z mod Op S N − 1 ( Σ ) (85)
with
Q p , z : = − Λ p ( ( 1 η p + τ p + ( z + m ) S p ) ⊗ I N 2 W p + W p − ( 1 η p − τ p + ( z − m ) S p ) ⊗ I N 2 ) Λ p = − ( ( 1 η p + τ p Λ p 2 + ( z + m ) Λ p S p Λ p ) ⊗ I N 2 Λ p W p + Λ p Λ p W p − Λ p ( 1 η p − τ p Λ p 2 + ( z − m ) Λ p S p Λ p ) ⊗ I N 2 ) . \begin{aligned}
Q_{p,z}:=-\Lambda_p
\begin{pmatrix}
\left(\dfrac{1}{\eta_p+\tau_p}+(z+m)S_p\right)\otimes I_{\frac{N}{2}} & W_p^+\\
W_p^- & \left(\dfrac{1}{\eta_p-\tau_p}+(z-m)S_p\right)\otimes I_{\frac{N}{2}}
\end{pmatrix}
\Lambda_p \\
&=-\begin{pmatrix}
\left(\dfrac{1}{\eta_p+\tau_p}\Lambda_p^2+(z+m)\Lambda_pS_p\Lambda_p\right)\otimes I_{\frac{N}{2}} & \Lambda_pW_p^+\Lambda_p\\
\Lambda_pW_p^-\Lambda_p & \left(\dfrac{1}{\eta_p-\tau_p}\Lambda_p^2+(z-m)\Lambda_pS_p\Lambda_p\right)\otimes I_{\frac{N}{2}}
\end{pmatrix}.
\end{aligned} Q p , z := − Λ p ( η p + τ p 1 + ( z + m ) S p ) ⊗ I 2 N W p − W p + ( η p − τ p 1 + ( z − m ) S p ) ⊗ I 2 N Λ p = − ( η p + τ p 1 Λ p 2 + ( z + m ) Λ p S p Λ p ) ⊗ I 2 N Λ p W p − Λ p Λ p W p + Λ p ( η p − τ p 1 Λ p 2 + ( z − m ) Λ p S p Λ p ) ⊗ I 2 N .
Since Λ p S p Λ p = I + ( m p − 1 ) Λ p 2 π p \Lambda_pS_p\Lambda_p=I+(m_p-1)\Lambda_p^2\pi_p Λ p S p Λ p = I + ( m p − 1 ) Λ p 2 π p and ( m p − 1 ) Λ p 2 π p ∈ Op S 1 − ∞ ( Σ p ) (m_p-1)\Lambda_p^2\pi_p\in\operatorname{Op}S_1^{-\infty}(\Sigma_p) ( m p − 1 ) Λ p 2 π p ∈ Op S 1 − ∞ ( Σ p ) by Lemma 4.1 (d), one finds that
Q p , z = L p − z mod Op S N − ∞ ( Σ p ) . Q_{p,z}=L_p-z\ \operatorname{mod}\ \operatorname{Op}S_N^{-\infty}(\Sigma_p). Q p , z = L p − z mod Op S N − ∞ ( Σ p ) .
By combining this with (85) one arrives at (82) . Finally, L p L_p L p is a symmetric first order pseudodifferential operator on Σ p \Sigma_p Σ p , and hence it is self-adjoint on the maximal domain due to Friedrichs lemma, see the dicussion in Appendix A. Since, by the same reason, the adjoint of L p L_p L p defined on C ∞ ( Σ p , C N ) C^\infty(\Sigma_p,\mathbb{C}^N) C ∞ ( Σ p , C N ) is the maximal realization of L p L_p L p , the self-adjointness of the latter implies that C ∞ ( Σ p , C N ) C^\infty(\Sigma_p,\mathbb{C}^N) C ∞ ( Σ p , C N ) , and hence also H 1 ( Σ p , C N ) H^1(\Sigma_p,\mathbb{C}^N) H 1 ( Σ p , C N ) , is a core for L p L_p L p . Therefore, ⨁ p = 1 P L p \bigoplus_{p=1}^{P}L_p ⨁ p = 1 P L p is self-adjoint in L 2 ( Σ , C N ) L^2(\Sigma,\mathbb{C}^N) L 2 ( Σ , C N ) on its maximal domain, which coincides, as M z M_z M z and all operators in Op S N − 1 ( Σ ) \operatorname{Op} S_N^{-1}(\Sigma) Op S N − 1 ( Σ ) are bounded in L 2 ( Σ , C N ) L^2(\Sigma,\mathbb{C}^N) L 2 ( Σ , C N ) , with dom Θ η , τ \operatorname{dom}\Theta^{\eta,\tau} dom Θ η , τ by (4.22). This finishes the proof.
Self-adjointness and essential spectrum
First, making use of Lemma 3.7, the abstract theory of boundary triples from Appendix B, and the properties of Θ η , τ \Theta^{\eta,\tau} Θ η , τ shown in Section 4.1, we prove the self-adjointness of A η , τ A^{\eta,\tau} A η , τ defined in Definition 3.3 and relate its essential spectrum to the essential spectra of the operators L p L_p L p defined in (4.21). Throughout this subsection we assume that we always have the special choice of Λ p \Lambda_p Λ p from Lemma 4.1.
Theorem 4.8. Assume that ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p| \ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ for all p ∈ { 1 , … , P } p \in\{1,\ldots,P\} p ∈ { 1 , … , P } , let Γ z \Gamma_z Γ z and M z M_z M z , z ∈ res A z \in\operatorname{res} A z ∈ res A , be given by (3.20) and (3.27), respectively, and let Θ η , τ \Theta^{\eta,\tau} Θ η , τ be as in Lemma 3.7. Then the operator A η , τ A^{\eta,\tau} A η , τ in Definition 3.3 is self-adjoint in L 2 ( Ω , C N ) L^2(\Omega,\mathbb{C}^N) L 2 ( Ω , C N ) and the following holds:
(a) For any z ∈ res A ∩ res A η , τ z \in\operatorname{res} A \cap\operatorname{res} A^{\eta,\tau} z ∈ res A ∩ res A η , τ the operator
Θ η , τ − M z = − Λ [ 1 η 2 − τ 2 ( η I N − τ β ) + C z ] Λ \Theta^{\eta,\tau}-M_z=-\Lambda\left[\frac{1}{\eta^2-\tau^2}(\eta I_N-\tau\beta)+C_z\right]\Lambda Θ η , τ − M z = − Λ [ η 2 − τ 2 1 ( η I N − τ β ) + C z ] Λ defined on the maximal domain dom Θ η , τ \operatorname{dom}\Theta^{\eta,\tau} dom Θ η , τ is bijective in L 2 ( Σ , C N ) L^2(\Sigma,\mathbb{C}^N) L 2 ( Σ , C N ) , and the resolvent formula
( A η , τ − z ) − 1 = ( A − z ) − 1 + Γ z ( Θ η , τ − M z ) − 1 Γ z ˉ ∗ (A^{\eta,\tau}-z)^{-1}=(A-z)^{-1}+\Gamma_z(\Theta^{\eta,\tau}-M_z)^{-1}\Gamma_{\bar z}^{*} ( A η , τ − z ) − 1 = ( A − z ) − 1 + Γ z ( Θ η , τ − M z ) − 1 Γ z ˉ ∗ holds.
(b) For any z ∈ res A z \in\operatorname{res} A z ∈ res A one has
z ∈ spec A η , τ if and only if 0 ∈ spec ( Θ η , τ − M z ) z\in\operatorname{spec} A^{\eta,\tau}\quad\text{if and only if}\quad0\in\operatorname{spec}(\Theta^{\eta,\tau}-M_z) z ∈ spec A η , τ if and only if 0 ∈ spec ( Θ η , τ − M z ) and
z ∈ spec e s s A η , τ if and only if 0 ∈ spec e s s ( Θ η , τ − M z ) , z\in\operatorname{spec}_{\mathrm{ess}} A^{\eta,\tau}\quad\text{if and only if}\quad0\in\operatorname{spec}_{\mathrm{ess}}(\Theta^{\eta,\tau}-M_z), z ∈ spec ess A η , τ if and only if 0 ∈ spec ess ( Θ η , τ − M z ) , and the equality dim ker ( A η , τ − z ) = dim ker ( Θ η , τ − M z ) \dim\ker(A^{\eta,\tau}-z)=\dim\ker(\Theta^{\eta,\tau}-M_z) dim ker ( A η , τ − z ) = dim ker ( Θ η , τ − M z ) holds. In particular,
spec e s s A η , τ ∩ res A = [ ⋃ p = 1 P spec e s s L p ] ∩ res A , (86) \operatorname{spec}_{\mathrm{ess}} A^{\eta,\tau}\cap\operatorname{res} A=
\left[\bigcup_{p=1}^{P}\operatorname{spec}_{\mathrm{ess}}L_p\right]\cap\operatorname{res} A, \tag*{(86)} spec ess A η , τ ∩ res A = [ p = 1 ⋃ P spec ess L p ] ∩ res A , (86) where the operator L p L_p L p is defined as in (4.21).
Proof. Recall that A η , τ A^{\eta,\tau} A η , τ is the restriction of S ∗ S^* S ∗ given by (3.13) on the functions f f f satisfying
Π 0 f ∈ dom Θ η , τ and Π 1 f = Θ η , τ Π 0 f , \Pi_0f\in\operatorname{dom}\Theta^{\eta,\tau}\quad\text{and}\quad\Pi_1f=\Theta^{\eta,\tau}\Pi_0f, Π 0 f ∈ dom Θ η , τ and Π 1 f = Θ η , τ Π 0 f ,
where ( L 2 ( Σ , C N ) , Π 0 , Π 1 ) (L^2(\Sigma,\mathbb{C}^N),\Pi_0,\Pi_1) ( L 2 ( Σ , C N ) , Π 0 , Π 1 ) is the boundary triple in Lemma 3.6; cf. Lemma 3.7. Next, note that Λ ∈ Op S N 1 / 2 ( Σ ) \Lambda\in\operatorname{Op}S_N^{1/2}(\Sigma) Λ ∈ Op S N 1/2 ( Σ ) by construction, and C ζ ∈ Op S N 0 ( Σ ) C_\zeta\in\operatorname{Op}S_N^0(\Sigma) C ζ ∈ Op S N 0 ( Σ ) by Corollary 4.6. Moreover, since ζ ∈ R \zeta\in\mathbb{R} ζ ∈ R , the operator C ζ C_\zeta C ζ is symmetric in L 2 ( Σ , C N ) L^2(\Sigma,\mathbb{C}^N) L 2 ( Σ , C N ) by (2.18). Therefore, Θ η , τ \Theta^{\eta,\tau} Θ η , τ is a symmetric first order pseudodifferential operator on Σ \Sigma Σ , and hence it is self-adjoint on the maximal domain due to Friedrichs’ lemma, see the discussion in Appendix A. Therefore, A η , τ A^{\eta,\tau} A η , τ is self-adjoint by Theorem B.2. The claims in (a) and the relations of the spectra of A η , τ A^{\eta,\tau} A η , τ and Θ η , τ − M z \Theta^{\eta,\tau}-M_z Θ η , τ − M z follow by using Theorem B.2 with the specific expressions obtained in Lemma 3.6 and Lemma 3.7. Finally, as any operator in Op S N − 1 ( Σ ) \operatorname{Op}S_N^{-1}(\Sigma) Op S N − 1 ( Σ ) gives rise to a bounded operator from L 2 ( Σ , C N ) L^2(\Sigma,\mathbb{C}^N) L 2 ( Σ , C N ) to H 1 ( Σ , C N ) H^1(\Sigma,\mathbb{C}^N) H 1 ( Σ , C N ) and hence, taking Rellich’s embedding theorem into account, to a compact operator in L 2 ( Σ , C N ) L^2(\Sigma,\mathbb{C}^N) L 2 ( Σ , C N ) , one concludes from (82) that
spec e s s ( Θ η , τ − M z ) = ⋃ p = 1 P spec e s s ( L p − z ) . \operatorname{spec}_{\mathrm{ess}}(\Theta^{\eta,\tau}-M_z)=\bigcup_{p=1}^{P}\operatorname{spec}_{\mathrm{ess}}(L_p-z). spec ess ( Θ η , τ − M z ) = p = 1 ⋃ P spec ess ( L p − z ) .
This yields (4.26) and finishes the proof.
Next, we consider the non-critical case η p 2 − τ p 2 ≠ 4 \eta_p^2-\tau_p^2\ne4 η p 2 − τ p 2 = 4 and ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p|\ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ . Here, it turns out that there is no contribution to the essential spectrum.
Lemma 4.9. Let p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } and assume that η p 2 − τ p 2 ≠ 4 \eta_p^2-\tau_p^2\ne4 η p 2 − τ p 2 = 4 and ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p|\ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ . Then, L p L_p L p is a first order elliptic operator. In particular, when L p L_p L p is viewed as a self-adjoint operator in L 2 ( Σ p , C N ) L^2(\Sigma_p,\mathbb{C}^N) L 2 ( Σ p , C N ) , then dom L p = H 1 ( Σ p , C N ) \operatorname{dom}L_p=H^1(\Sigma_p,\mathbb{C}^N) dom L p = H 1 ( Σ p , C N ) and spec e s s L p = ∅ \operatorname{spec}_{\mathrm{ess}}L_p=\varnothing spec ess L p = ∅ .
Proof. By definition L p L_p L p is of first order, so we simply need to show that its principal symbol is invertible for ∣ ξ ∣ |\xi| ∣ ξ ∣ sufficiently large. For any choice of local coordinates near x ∈ Σ p x\in\Sigma_p x ∈ Σ p the principal symbol σ L p ( x , ξ ) \sigma_{L_p}(x,\xi) σ L p ( x , ξ ) of L p L_p L p is
σ L p ( x , ξ ) = − ( 1 η p + τ p I N 2 σ W p + ( x , ξ ) σ W p − ( x , ξ ) 1 η p − τ p I N 2 ) σ Λ p ( x , ξ ) 2 , \sigma_{L_p}(x,\xi)=-\begin{pmatrix}
\frac{1}{\eta_p+\tau_p}I_{\frac{N}{2}} & \sigma_{W_p^+}(x,\xi)\\
\sigma_{W_p^-}(x,\xi) & \frac{1}{\eta_p-\tau_p}I_{\frac{N}{2}}
\end{pmatrix}\sigma_{\Lambda_p}(x,\xi)^2, σ L p ( x , ξ ) = − ( η p + τ p 1 I 2 N σ W p − ( x , ξ ) σ W p + ( x , ξ ) η p − τ p 1 I 2 N ) σ Λ p ( x , ξ ) 2 ,
and we recall that σ Λ p ( x , ξ ) 2 = 2 ∣ ξ ∣ \sigma_{\Lambda_p}(x,\xi)^2=2|\xi| σ Λ p ( x , ξ ) 2 = 2∣ ξ ∣ by construction, see Lemma 4.1 (c).
Consider the case n = 2 n=2 n = 2 , then N = 2 N=2 N = 2 . Assume that the local coordinates are given by a properly oriented arc-length parametrization (see Definition 4.2 (a)). Then by Lemma 4.3 we obtain
σ L p ( x , ξ ) = − 2 ∣ ξ ∣ ( 1 η p + τ p − i 2 μ ( x ) sgn ξ i 2 μ ( x ) sgn ξ 1 η p − τ p ) \sigma_{L_p}(x,\xi)=-2|\xi|\begin{pmatrix}
\frac{1}{\eta_p+\tau_p} & -\frac{\mathrm{i}}{2}\mu(x)\operatorname{sgn}\xi\\
\frac{\mathrm{i}}{2}\mu(x)\operatorname{sgn}\xi& \frac{1}{\eta_p-\tau_p}
\end{pmatrix} σ L p ( x , ξ ) = − 2∣ ξ ∣ ( η p + τ p 1 2 i μ ( x ) sgn ξ − 2 i μ ( x ) sgn ξ η p − τ p 1 )
and thus,
det σ L p ( x , ξ ) = 4 ξ 2 ( 1 η p 2 − τ p 2 − 1 4 ) . \det\sigma_{L_p}(x,\xi)=4\xi^2\left(\frac{1}{\eta_p^2-\tau_p^2}-\frac{1}{4}\right). det σ L p ( x , ξ ) = 4 ξ 2 ( η p 2 − τ p 2 1 − 4 1 ) .
Since η p 2 − τ p 2 ≠ 4 \eta_p^2-\tau_p^2\ne4 η p 2 − τ p 2 = 4 by assumption, the determinant is non-zero for all ξ ≠ 0 \xi\ne0 ξ = 0 , implying that σ L p ( x , ξ ) \sigma_{L_p}(x,\xi) σ L p ( x , ξ ) is invertible for ξ ≠ 0 \xi\ne0 ξ = 0 and therefore, L p L_p L p is elliptic in the case n = 2 n=2 n = 2 .
Now let n = 3 n=3 n = 3 , then N = 4 N=4 N = 4 . Using normal coordinates near x x x (see Definition 4.2 (b)) we arrive at
σ L p ( x , ξ ) = − 2 ∣ ξ ∣ ( 1 η p + τ p I 2 1 2 σ ⋅ ( ξ 1 e 1 + ξ 2 e 2 ∣ ξ ∣ ) 1 2 σ ⋅ ( ξ 1 e 1 + ξ 2 e 2 ∣ ξ ∣ ) 1 η p − τ p I 2 ) . \sigma_{L_p}(x,\xi)=-2|\xi|\begin{pmatrix}
\frac{1}{\eta_p+\tau_p}I_2 & \frac{1}{2}\sigma\cdot\left(\frac{\xi_1e_1+\xi_2e_2}{|\xi|}\right)\\
\frac{1}{2}\sigma\cdot\left(\frac{\xi_1e_1+\xi_2e_2}{|\xi|}\right) & \frac{1}{\eta_p-\tau_p}I_2
\end{pmatrix}. σ L p ( x , ξ ) = − 2∣ ξ ∣ η p + τ p 1 I 2 2 1 σ ⋅ ( ∣ ξ ∣ ξ 1 e 1 + ξ 2 e 2 ) 2 1 σ ⋅ ( ∣ ξ ∣ ξ 1 e 1 + ξ 2 e 2 ) η p − τ p 1 I 2 .
Therefore, for ξ ≠ 0 \xi\ne0 ξ = 0 one has
σ L p ( x , ξ ) ⋅ 1 2 ∣ ξ ∣ ( 1 η p − τ p I 2 − 1 2 σ ⋅ ( ξ 1 e 1 + ξ 2 e 2 ∣ ξ ∣ ) − 1 2 σ ⋅ ( ξ 1 e 1 + ξ 2 e 2 ∣ ξ ∣ ) 1 η p + τ p I 2 ) = ( 1 4 − 1 η p 2 − τ p 2 ) I 4 , (87) \sigma_{L_p}(x,\xi)\cdot\frac{1}{2|\xi|}\begin{pmatrix}
\frac{1}{\eta_p-\tau_p}I_2 & -\frac{1}{2}\sigma\cdot\left(\frac{\xi_1e_1+\xi_2e_2}{|\xi|}\right)\\
-\frac{1}{2}\sigma\cdot\left(\frac{\xi_1e_1+\xi_2e_2}{|\xi|}\right) & \frac{1}{\eta_p+\tau_p}I_2
\end{pmatrix}
=\left(\frac{1}{4}-\frac{1}{\eta_p^2-\tau_p^2}\right)I_4, \tag*{(87)} σ L p ( x , ξ ) ⋅ 2∣ ξ ∣ 1 η p − τ p 1 I 2 − 2 1 σ ⋅ ( ∣ ξ ∣ ξ 1 e 1 + ξ 2 e 2 ) − 2 1 σ ⋅ ( ∣ ξ ∣ ξ 1 e 1 + ξ 2 e 2 ) η p + τ p 1 I 2 = ( 4 1 − η p 2 − τ p 2 1 ) I 4 , (87)
which shows that σ L p ( x , ξ ) \sigma_{L_p}(x,\xi) σ L p ( x , ξ ) is invertible for all ξ ≠ 0 \xi\ne0 ξ = 0 , if η p 2 − τ p 2 ≠ 4 \eta_p^2-\tau_p^2 \ne4 η p 2 − τ p 2 = 4 . Hence, L p L_p L p is also elliptic in the case n = 3 n=3 n = 3 .
Finally, as L p L_p L p is a first-order elliptic operator we have dom L p = H 1 ( Σ p , C N ) \operatorname{dom} L_p=H^1(\Sigma_p,\mathbb{C}^N) dom L p = H 1 ( Σ p , C N ) , when L p L_p L p is viewed as a self-adjoint operator in L 2 ( Σ p , C N ) L^2(\Sigma_p,\mathbb{C}^N) L 2 ( Σ p , C N ) . Furthermore, it follows that the resolvent of L p L_p L p is compact and thus spec e s s L p = ∅ \operatorname{spec}_{\mathrm{ess}} L_p=\varnothing spec ess L p = ∅ .
As an immediate consequence of the ellipticity of L p L_p L p in the non-critical case one gets an improved Sobolev regularity near Σ p \Sigma_p Σ p for functions belonging to dom A η , τ \operatorname{dom} A^{\eta,\tau} dom A η , τ . Recall that the space H α 1 , Σ p ( Ω ∖ Σ , C N ) H^{1,\Sigma_p}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^N) H α 1 , Σ p ( Ω ∖ Σ , C N ) is defined by (3.7).
Corollary 4.10. Let p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } and assume that η p 2 − τ p 2 ≠ 4 \eta_p^2-\tau_p^2\ne4 η p 2 − τ p 2 = 4 and ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p|\ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ . Then
dom A η , τ ⊂ H α 1 , Σ p ( Ω ∖ Σ , C N ) . \operatorname{dom} A^{\eta,\tau}\subset H^{1,\Sigma_p}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^N). dom A η , τ ⊂ H α 1 , Σ p ( Ω ∖ Σ , C N ) .
Proof. Let f ∈ dom A η , τ f\in\operatorname{dom} A^{\eta,\tau} f ∈ dom A η , τ . Then Π 0 f ∈ dom Θ η , τ \Pi_0f\in\operatorname{dom}\Theta^{\eta,\tau} Π 0 f ∈ dom Θ η , τ . According to Proposition 4.7 one has
dom Θ η , τ = ⨁ p = 1 P dom L p , \operatorname{dom}\Theta^{\eta,\tau}=\bigoplus_{p=1}^{P}\operatorname{dom}L_p, dom Θ η , τ = p = 1 ⨁ P dom L p ,
which implies
( Π 0 f ) ∣ Σ p = i Λ p − 1 ( α ⋅ ν Σ ) ( γ + Σ f − γ − Σ f ) ∣ Σ p ∈ dom L p = H 1 ( Σ p , C N ) , (\Pi_0f)|_{\Sigma_p}=i\Lambda_p^{-1}(\alpha\cdot\nu^{\Sigma})(\gamma_+^{\Sigma}f-\gamma_-^{\Sigma}f)|_{\Sigma_p}\in\operatorname{dom}L_p=H^1(\Sigma_p,\mathbb{C}^N), ( Π 0 f ) ∣ Σ p = i Λ p − 1 ( α ⋅ ν Σ ) ( γ + Σ f − γ − Σ f ) ∣ Σ p ∈ dom L p = H 1 ( Σ p , C N ) ,
where the last equality holds by Lemma 4.9. This yields ( γ + Σ f − γ − Σ f ) ∣ Σ p ∈ H 1 2 ( Σ p , C N ) (\gamma_+^{\Sigma}f-\gamma_-^{\Sigma}f)|_{\Sigma_p}\in H^{\frac12}(\Sigma_p,\mathbb{C}^N) ( γ + Σ f − γ − Σ f ) ∣ Σ p ∈ H 2 1 ( Σ p , C N ) . Next, note that the assumption ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p|\ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ implies that the constant matrix η p I N + τ p β \eta_p I_N+\tau_p\beta η p I N + τ p β is invertible. Hence, due to the transmission condition for f ∈ dom A η , τ f\in\operatorname{dom} A^{\eta,\tau} f ∈ dom A η , τ one gets
( γ + Σ f + γ − Σ f ) ∣ Σ p = − 2 ( η p I N + τ p β ) − 1 i ( α ⋅ ν Σ ) ( γ + Σ f − γ − Σ f ) ∣ Σ p ∈ H 1 2 ( Σ p , C N ) . (\gamma_+^{\Sigma}f+\gamma_-^{\Sigma}f)|_{\Sigma_p}=-2(\eta_p I_N+\tau_p\beta)^{-1}i(\alpha\cdot\nu^{\Sigma})(\gamma_+^{\Sigma}f-\gamma_-^{\Sigma}f)|_{\Sigma_p}\in H^{\frac12}(\Sigma_p,\mathbb{C}^N). ( γ + Σ f + γ − Σ f ) ∣ Σ p = − 2 ( η p I N + τ p β ) − 1 i ( α ⋅ ν Σ ) ( γ + Σ f − γ − Σ f ) ∣ Σ p ∈ H 2 1 ( Σ p , C N ) .
Overall, we obtain ( γ ± Σ f ) ∣ Σ p ∈ H 1 2 ( Σ p , C N ) (\gamma_{\pm}^{\Sigma}f)|_{\Sigma_p}\in H^{\frac12}(\Sigma_p,\mathbb{C}^N) ( γ ± Σ f ) ∣ Σ p ∈ H 2 1 ( Σ p , C N ) , and then f ∈ H α 1 , Σ p ( Ω ∖ Σ , C N ) f\in H^{1,\Sigma_p}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^N) f ∈ H α 1 , Σ p ( Ω ∖ Σ , C N ) by Lemma 3.1.
Corollary 4.11. If η p 2 − τ p 2 ≠ 4 \eta_p^2-\tau_p^2\ne4 η p 2 − τ p 2 = 4 and ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p|\ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ for all p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } , then
dom A η , τ = { f ∈ H α 1 , Σ ( Ω ∖ Σ , C N ) ∩ H α 1 2 , ∂ Ω ( Ω ∖ Σ , C N ) : B γ ∂ Ω f = 0 on ∂ Ω , \operatorname{dom} A^{\eta,\tau}=\left\{f\in H^{1,\Sigma}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^N)\cap H^{\frac12,\partial\Omega}_{\alpha}(\Omega\setminus\Sigma,\mathbb{C}^N):B\gamma_{\partial\Omega}f=0\ \text{on }\partial\Omega,
\right. dom A η , τ = { f ∈ H α 1 , Σ ( Ω ∖ Σ , C N ) ∩ H α 2 1 , ∂ Ω ( Ω ∖ Σ , C N ) : B γ ∂ Ω f = 0 on ∂ Ω , i ( α ⋅ ν Σ ) ( γ + Σ f − γ − Σ f ) + 1 2 ( η I N + τ β ) ( γ + Σ f + γ − Σ f ) = 0 on Σ } . (88) \left.i(\alpha\cdot\nu^{\Sigma})(\gamma_+^{\Sigma}f-\gamma_-^{\Sigma}f)+\frac12(\eta I_N+\tau\beta)(\gamma_+^{\Sigma}f+\gamma_-^{\Sigma}f)=0\ \text{on }\Sigma\right\}. \tag*{(88)} i ( α ⋅ ν Σ ) ( γ + Σ f − γ − Σ f ) + 2 1 ( η I N + τ β ) ( γ + Σ f + γ − Σ f ) = 0 on Σ } . (88)
Now, we turn our focus to the critical case , i.e. when η p 2 − τ p 2 = 4 \eta_p^2-\tau_p^2=4 η p 2 − τ p 2 = 4 . The following result on block operator matrices will be used below. It is an adapted formulation of [49 Theorem 2.2.18 and Theorem 2.4.7] . Here and below, we understand the essential spectrum of a closed, but not necessarily self-adjoint operator T T T to be the set of all λ ∈ C \lambda\in\mathbb{C} λ ∈ C such that T − λ T-\lambda T − λ is not a Fredholm operator.
Proposition 4.12 (Schur complement). Let H \mathcal{H} H be a Hilbert space and A A A , B B B , C C C , D D D be densely defined and closable operators in H \mathcal{H} H such that D D D has a bounded inverse, dom D ⊂ dom B \operatorname{dom}D\subset\operatorname{dom}B dom D ⊂ dom B , dom A ⊂ dom C \operatorname{dom}A\subset\operatorname{dom}C dom A ⊂ dom C , and D − 1 C D^{-1}C D − 1 C is bounded on dom C \operatorname{dom}C dom C . Consider the linear operator Q Q Q in H × H \mathcal{H}\times\mathcal{H} H × H ,
Q = ( A B C D ) , dom Q = dom A × dom D , Q=\begin{pmatrix}A&B\\C&D\end{pmatrix},\qquad\operatorname{dom}Q=\operatorname{dom}A\times\operatorname{dom}D, Q = ( A C B D ) , dom Q = dom A × dom D ,
and its Schur complement X : = A − B D − 1 C X := A - BD^{-1}C X := A − B D − 1 C defined on dom A \operatorname{dom} A dom A , which is viewed as a linear operator in H \mathcal{H} H . Then, the following is true:
(a) Q Q Q is closed (closable) if and only if X X X is closed (closable, respectively), and in this case one has
Q ‾ = ( I B D − 1 0 I ) ( X ‾ 0 0 D ) ( I 0 D − 1 C I ) . \overline{Q} =
\begin{pmatrix}
I & BD^{-1} \\
0 & I
\end{pmatrix}
\begin{pmatrix}
\overline{X} & 0 \\
0 & D
\end{pmatrix}
\begin{pmatrix}
I & 0 \\
D^{-1}C & I
\end{pmatrix}. Q = ( I 0 B D − 1 I ) ( X 0 0 D ) ( I D − 1 C 0 I ) .
(b) If Q Q Q is closable, then 0 ∈ spec e s s Q ‾ 0 \in\operatorname{spec}_{\mathrm{ess}} \overline{Q} 0 ∈ spec ess Q if and only if 0 ∈ spec e s s X ‾ 0 \in\operatorname{spec}_{\mathrm{ess}} \overline{X} 0 ∈ spec ess X .
Lemma 4.13. Let p ∈ { 1 , … , P } p \in\{1,\ldots,P\} p ∈ { 1 , … , P } , assume that η p 2 − τ p 2 = 4 \eta_p^2 - \tau_p^2 = 4 η p 2 − τ p 2 = 4 , and let L p L_p L p be the self-adjoint operator in L 2 ( Σ p , C N ) L^2(\Sigma_p,\mathbb{C}^N) L 2 ( Σ p , C N ) in (4.21). Then
J p = spec e s s L p = { { − τ p η p m } , n = 2 , − τ p η p m + 1 η p [ − a p 2 , a p 2 ] , n = 3 , (89) J_p = \operatorname{spec}_{\mathrm{ess}} L_p =
\begin{cases}
\left\{-\dfrac{\tau_p}{\eta_p}m\right\}, & n=2, \\[6pt]
-\dfrac{\tau_p}{\eta_p}m + \dfrac{1}{\eta_p}\left[-\dfrac{a_p}{2},\dfrac{a_p}{2}\right], & n=3,
\end{cases} \tag*{(89)} J p = spec ess L p = ⎩ ⎨ ⎧ { − η p τ p m } , − η p τ p m + η p 1 [ − 2 a p , 2 a p ] , n = 2 , n = 3 , (89) where, for n = 3 n=3 n = 3 , a p : = max x ∈ Σ p ∣ κ 1 ( x ) − κ 2 ( x ) ∣ a_p := \max_{x \in\Sigma_p} |\kappa_1(x)-\kappa_2(x)| a p := max x ∈ Σ p ∣ κ 1 ( x ) − κ 2 ( x ) ∣ and κ 1 ( x ) \kappa_1(x) κ 1 ( x ) , κ 2 ( x ) \kappa_2(x) κ 2 ( x ) are the principal curvatures of Σ p \Sigma_p Σ p in x x x .
Proof. Throughout the proof, if not declared differently, all operators will be considered to be considered to be defined on their maximal domains in the respective L 2 L^2 L 2 -spaces. Let z ∈ C z \in\mathbb{C} z ∈ C and note that z ∈ spec e s s L p z \in\operatorname{spec}_{\mathrm{ess}} L_p z ∈ spec ess L p if and only if 0 ∈ spec e s s ( z − L p ) 0 \in\operatorname{spec}_{\mathrm{ess}}(z-L_p) 0 ∈ spec ess ( z − L p ) . As the essential spectrum is invariant under compact perturbations, for any Π ∈ Op S 1 − ∞ ( Σ p ) \Pi\in\operatorname{Op} S_1^{-\infty}(\Sigma_p) Π ∈ Op S 1 − ∞ ( Σ p ) , the condition 0 ∈ spec e s s ( z − L p ) 0 \in\operatorname{spec}_{\mathrm{ess}}(z-L_p) 0 ∈ spec ess ( z − L p ) is equivalent to
0 ∈ spec e s s ( ( 1 η p + τ p Λ p 2 + ( z + m ) ) ⊗ I N 2 Λ p W p + Λ p Λ p W p − Λ p ( 1 η p − τ p Λ p 2 + ( z − m ) + Π ) ⊗ I N 2 ) . (90) 0 \in\operatorname{spec}_{\mathrm{ess}}
\begin{pmatrix}
\left(\dfrac{1}{\eta_p+\tau_p}\Lambda_p^2+(z+m)\right)\otimes I_{\frac{N}{2}} & \Lambda_p W_p^+\Lambda_p \\
\Lambda_p W_p^-\Lambda_p & \left(\dfrac{1}{\eta_p-\tau_p}\Lambda_p^2+(z-m)+\Pi\right)\otimes I_{\frac{N}{2}}
\end{pmatrix}. \tag*{(90)} 0 ∈ spec ess ( η p + τ p 1 Λ p 2 + ( z + m ) ) ⊗ I 2 N Λ p W p − Λ p Λ p W p + Λ p ( η p − τ p 1 Λ p 2 + ( z − m ) + Π ) ⊗ I 2 N . (90)
In particular, as Λ p \Lambda_p Λ p has compact resolvent its spectrum is discrete and the same is true for Λ p 2 \Lambda_p^2 Λ p 2 . Thus, one can choose
Π : = the orthogonal projector on ker ( 1 η p − τ p Λ p 2 + ( z − m ) ) in L 2 ( Σ p ) , \Pi:= \text{the orthogonal projector on } \ker\left(\dfrac{1}{\eta_p-\tau_p}\Lambda_p^2+(z-m)\right)\text{ in }L^2(\Sigma_p), Π := the orthogonal projector on ker ( η p − τ p 1 Λ p 2 + ( z − m ) ) in L 2 ( Σ p ) ,
which is a finite-dimensional projector with smooth integral kernel; cf. the proof of Lemma 4.1 (b) for a similar argument. Since Π \Pi Π is a spectral projection of the positive operator Λ p 2 \Lambda_p^2 Λ p 2 , it commutes with Λ p \Lambda_p Λ p and Λ p − 1 \Lambda_p^{-1} Λ p − 1 . This choice of Π \Pi Π guarantees that the bottom right block in (4.29) has a bounded inverse and, as Λ p \Lambda_p Λ p is elliptic,
( 1 η p − τ p Λ p 2 + ( z − m ) + Π ) − 1 ∈ Op S 1 − 1 ( Σ p ) . (91) \left(\dfrac{1}{\eta_p-\tau_p}\Lambda_p^2+(z-m)+\Pi\right)^{-1}\in\operatorname{Op} S_1^{-1}(\Sigma_p). \tag*{(91)} ( η p − τ p 1 Λ p 2 + ( z − m ) + Π ) − 1 ∈ Op S 1 − 1 ( Σ p ) . (91)
Here and below we write a a a instead of a ⊗ I N 2 a \otimes I_{\frac{N}{2}} a ⊗ I 2 N for better readability. Consider the block operator matrix
L p ( z ) : = ( 1 η p + τ p Λ p 2 + ( z + m ) Λ p W p + Λ p Λ p W p − Λ p 1 η p − τ p Λ p 2 + ( z − m ) + Π ) , \mathcal{L}_p(z) := \begin{pmatrix} \dfrac{1}{\eta_p+\tau_p}\Lambda_p^2+(z+m) & \Lambda_p W_p^+\Lambda_p \\ \Lambda_p W_p^-\Lambda_p & \dfrac{1}{\eta_p-\tau_p}\Lambda_p^2+(z-m)+\Pi\end{pmatrix}, L p ( z ) := η p + τ p 1 Λ p 2 + ( z + m ) Λ p W p − Λ p Λ p W p + Λ p η p − τ p 1 Λ p 2 + ( z − m ) + Π ,
dom L p ( z ) : = H 1 ( Σ p , C N ) . \operatorname{dom}\mathcal{L}_p(z) := H^1(\Sigma_p,\mathbb{C}^N). dom L p ( z ) := H 1 ( Σ p , C N ) .
The operator L p ( z ) \mathcal{L}_p(z) L p ( z ) is a restriction of the closed operator
z − L p + ( 0 0 0 Π ) , z-L_p+\begin{pmatrix}0&0\\0&\Pi\end{pmatrix}, z − L p + ( 0 0 0 Π ) ,
where L p L_p L p is defined on its maximal domain; in particular, L p ( z ) \mathcal{L}_p(z) L p ( z ) is closable. Since H 1 ( Σ p , C N ) H^1(\Sigma_p,\mathbb{C}^N) H 1 ( Σ p , C N ) is a core for L p L_p L p by Proposition 4.7 and Π \Pi Π is bounded, we conclude that
L p ( z ) ‾ = z − L p + ( 0 0 0 Π ) . \overline{\mathcal{L}_p(z)}=z-L_p+\begin{pmatrix}0&0\\0&\Pi\end{pmatrix}. L p ( z ) = z − L p + ( 0 0 0 Π ) .
As Λ p W p − Λ p ∈ Op S N / 2 1 ( Σ p ) \Lambda_pW_p^-\Lambda_p\in\operatorname{Op}S^{1}_{N/2}(\Sigma_p) Λ p W p − Λ p ∈ Op S N /2 1 ( Σ p ) , one obtains with (91) that
[ 1 η p − τ p Λ p 2 + ( z − m ) + Π ] − 1 Λ p W p − Λ p ∈ Op S N / 2 0 ( Σ p ) \left[\frac{1}{\eta_p-\tau_p}\Lambda_p^2+(z-m)+\Pi\right]^{-1}\Lambda_pW_p^-\Lambda_p\in\operatorname{Op}S^{0}_{N/2}(\Sigma_p) [ η p − τ p 1 Λ p 2 + ( z − m ) + Π ] − 1 Λ p W p − Λ p ∈ Op S N /2 0 ( Σ p )
and hence, this operator is bounded in L 2 ( Σ p , C N / 2 ) L^2(\Sigma_p,\mathbb{C}^{N/2}) L 2 ( Σ p , C N /2 ) . In particular, for the block operator matrix L p ( z ) \mathcal{L}_p(z) L p ( z ) all assumptions in Proposition 4.12 are fulfilled and because of L p ( z ) ‾ = z − L p mod Op S N − ∞ ( Σ p ) \overline{\mathcal{L}_p(z)}=z-L_p\ \operatorname{mod}\ \operatorname{Op}S^{-\infty}_{N}(\Sigma_p) L p ( z ) = z − L p mod Op S N − ∞ ( Σ p ) one has that
0 ∈ spec e s s ( z − L p ) ⟺ 0 ∈ spec e s s ( X p ( z ) ‾ ) , (92) 0\in\operatorname{spec}_{\mathrm{ess}}(z-L_p)\Longleftrightarrow0\in\operatorname{spec}_{\mathrm{ess}}(\overline{X_p(z)}), \tag*{(92)} 0 ∈ spec ess ( z − L p ) ⟺ 0 ∈ spec ess ( X p ( z ) ) , (92)
where X p ( z ) X_p(z) X p ( z ) is the Schur complement associated with L p ( z ) \mathcal{L}_p(z) L p ( z ) ,
X p ( z ) = 1 η p + τ p Λ p 2 + ( z + m ) − Λ p W p + Λ p [ 1 η p − τ p Λ p 2 + ( z − m ) + Π ] − 1 Λ p W p − Λ p , X_p(z)=\frac{1}{\eta_p+\tau_p}\Lambda_p^2+(z+m)-\Lambda_pW_p^+\Lambda_p\left[\frac{1}{\eta_p-\tau_p}\Lambda_p^2+(z-m)+\Pi\right]^{-1}\Lambda_pW_p^-\Lambda_p, X p ( z ) = η p + τ p 1 Λ p 2 + ( z + m ) − Λ p W p + Λ p [ η p − τ p 1 Λ p 2 + ( z − m ) + Π ] − 1 Λ p W p − Λ p ,
dom X p ( z ) = H 1 ( Σ p , C N / 2 ) . \operatorname{dom}X_p(z)=H^1(\Sigma_p,\mathbb{C}^{N/2}). dom X p ( z ) = H 1 ( Σ p , C N /2 ) .
In the following, we will consider X p ( z ) X_p(z) X p ( z ) also as an element in Op S N / 2 1 ( Σ p ) \operatorname{Op}S^1_{N/2}(\Sigma_p) Op S N /2 1 ( Σ p ) . Since Π \Pi Π commutes with Λ p \Lambda_p Λ p we have
Λ p [ 1 η p − τ p Λ p 2 + ( z − m ) + Π ] − 1 Λ p = ( η p − τ p ) [ I + T ] − 1 \Lambda_p\left[\frac{1}{\eta_p-\tau_p}\Lambda_p^2+(z-m)+\Pi\right]^{-1}\Lambda_p=(\eta_p-\tau_p)[I+T]^{-1} Λ p [ η p − τ p 1 Λ p 2 + ( z − m ) + Π ] − 1 Λ p = ( η p − τ p ) [ I + T ] − 1
with
T = ( η p − τ p ) ( z − m ) Λ p − 2 + ( η p − τ p ) Π Λ p − 2 ∈ Op S N / 2 − 1 ( Σ p ) T=(\eta_p-\tau_p)(z-m)\Lambda_p^{-2}+(\eta_p-\tau_p)\Pi\Lambda_p^{-2}\in\operatorname{Op}S^{-1}_{N/2}(\Sigma_p) T = ( η p − τ p ) ( z − m ) Λ p − 2 + ( η p − τ p ) Π Λ p − 2 ∈ Op S N /2 − 1 ( Σ p )
since Λ p − 2 ∈ Op S N / 2 − 1 ( Σ p ) \Lambda_p^{-2}\in\operatorname{Op}S^{-1}_{N/2}(\Sigma_p) Λ p − 2 ∈ Op S N /2 − 1 ( Σ p ) and Π Λ p − 2 ∈ Op S N / 2 − ∞ ( Σ p ) \Pi\Lambda_p^{-2}\in\operatorname{Op}S^{-\infty}_{N/2}(\Sigma_p) Π Λ p − 2 ∈ Op S N /2 − ∞ ( Σ p ) . Using ( I + T ) − 1 = I − T + T 2 ( 1 + T ) − 1 (I+T)^{-1}=I-T+T^2(1+T)^{-1} ( I + T ) − 1 = I − T + T 2 ( 1 + T ) − 1 , T 2 ∈ Op S N / 2 − 2 ( Σ p ) T^2\in\operatorname{Op}S^{-2}_{N/2}(\Sigma_p) T 2 ∈ Op S N /2 − 2 ( Σ p ) , and ( 1 + T ) − 1 ∈ Op S N / 2 0 ( Σ p ) (1+T)^{-1}\in\operatorname{Op}S^0_{N/2}(\Sigma_p) ( 1 + T ) − 1 ∈ Op S N /2 0 ( Σ p ) we arrive at
Λ p [ 1 η p − τ p Λ p 2 + ( z − m ) + Π ] − 1 Λ p = ( η p − τ p ) − ( η p − τ p ) 2 ( z − m ) Λ p − 2 mod Op S N / 2 − 2 ( Σ p ) . \Lambda_p\left[\frac{1}{\eta_p-\tau_p}\Lambda_p^2+(z-m)+\Pi\right]^{-1}\Lambda_p=(\eta_p-\tau_p)-(\eta_p-\tau_p)^2(z-m)\Lambda_p^{-2}\ \operatorname{mod}\ \operatorname{Op}S^{-2}_{N/2}(\Sigma_p). Λ p [ η p − τ p 1 Λ p 2 + ( z − m ) + Π ] − 1 Λ p = ( η p − τ p ) − ( η p − τ p ) 2 ( z − m ) Λ p − 2 mod Op S N /2 − 2 ( Σ p ) .
Plugging this into the expression for the Schur complement, one gets
X p ( z ) = 1 η p + τ p Λ p 2 + ( z + m ) − ( η p − τ p ) Λ p W p + W p − Λ p + ( η p − τ p ) 2 ( z − m ) Λ p W p + Λ p − 2 W p − Λ p m o d Op S N / 2 − 1 2 ( Σ p ) . X_p(z)=\frac{1}{\eta_p+\tau_p}\Lambda_p^2+(z+m)-(\eta_p-\tau_p)\Lambda_pW_p^+W_p^-\Lambda_p+(\eta_p-\tau_p)^2(z-m)\Lambda_pW_p^+\Lambda_p^{-2}W_p^-\Lambda_p\ \mathop{\mathrm{mod}}\ \operatorname{Op}S^{-\frac{1}{2}}_{N/2}(\Sigma_p). X p ( z ) = η p + τ p 1 Λ p 2 + ( z + m ) − ( η p − τ p ) Λ p W p + W p − Λ p + ( η p − τ p ) 2 ( z − m ) Λ p W p + Λ p − 2 W p − Λ p mod Op S N /2 − 2 1 ( Σ p ) .
As Λ p \Lambda_p Λ p is a scalar operator, we have W p ± Λ p = Λ p W p ± W_p^\pm\Lambda_p=\Lambda_pW_p^\pm W p ± Λ p = Λ p W p ± m o d Op S N / 2 − 1 2 ( Σ p ) \mathop{\mathrm{mod}}\ \operatorname{Op}S^{-\frac{1}{2}}_{N/2}(\Sigma_p) mod Op S N /2 − 2 1 ( Σ p ) (see Appendix A). Hence,
Λ p W p + Λ p − 2 W p − Λ p = W p + Λ p − 1 W p − Λ p m o d Op S N / 2 − 1 ( Σ p ) = W p + W p − m o d Op S N / 2 − 1 ( Σ p ) = 1 4 m o d Op S N / 2 − 1 ( Σ p ) , \begin{aligned}
\Lambda_pW_p^+\Lambda_p^{-2}W_p^-\Lambda_p=W_p^+\Lambda_p^{-1}W_p^-\Lambda_p\ \mathop{\mathrm{mod}}\ \operatorname{Op}S^{-1}_{N/2}(\Sigma_p) \\
&=W_p^+W_p^-\ \mathop{\mathrm{mod}}\ \operatorname{Op}S^{-1}_{N/2}(\Sigma_p) \\
&=\frac{1}{4}\ \mathop{\mathrm{mod}}\ \operatorname{Op}S^{-1}_{N/2}(\Sigma_p),
\end{aligned} Λ p W p + Λ p − 2 W p − Λ p = W p + Λ p − 1 W p − Λ p mod Op S N /2 − 1 ( Σ p ) = W p + W p − mod Op S N /2 − 1 ( Σ p ) = 4 1 mod Op S N /2 − 1 ( Σ p ) ,
where in the last step Corollary 4.4 was employed. Therefore, we conclude
X p ( z ) = X ~ p ( z ) m o d Op S N / 2 − 1 ( Σ p ) . X_p(z)=\widetilde{X}_p(z)\ \mathop{\mathrm{mod}}\ \operatorname{Op}S^{-1}_{N/2}(\Sigma_p). X p ( z ) = X p ( z ) mod Op S N /2 − 1 ( Σ p ) .
where
X ~ p ( z ) : = 1 η p + τ p Λ p 2 + ( z + m ) − ( η p − τ p ) Λ p W p + W p − Λ p + ( η p − τ p ) 2 ( z − m ) 4 = ( z + m ) + ( η p − τ p ) 2 ( z − m ) 4 + ( η p − τ p ) Λ p ( 1 4 − W p + W p − ) Λ p ; \begin{aligned}
\widetilde{X}_p(z):=\frac{1}{\eta_p+\tau_p}\Lambda_p^2+(z+m)-(\eta_p-\tau_p)\Lambda_pW_p^+W_p^-\Lambda_p+\frac{(\eta_p-\tau_p)^2(z-m)}{4} \\
&=(z+m)+\frac{(\eta_p-\tau_p)^2(z-m)}{4}+(\eta_p-\tau_p)\Lambda_p\left(\frac{1}{4}-W_p^+W_p^-\right)\Lambda_p;
\end{aligned} X p ( z ) := η p + τ p 1 Λ p 2 + ( z + m ) − ( η p − τ p ) Λ p W p + W p − Λ p + 4 ( η p − τ p ) 2 ( z − m ) = ( z + m ) + 4 ( η p − τ p ) 2 ( z − m ) + ( η p − τ p ) Λ p ( 4 1 − W p + W p − ) Λ p ;
in the last simplification the condition η p 2 − τ p 2 = 4 \eta_p^2-\tau_p^2=4 η p 2 − τ p 2 = 4 was used and X ~ p ( z ) \widetilde{X}_p(z) X p ( z ) is viewed as the maximal realization of the associated pseudodifferential operator in L 2 ( Σ p , C N 2 ) L^2(\Sigma_p,\mathbb{C}^{\frac{N}{2}}) L 2 ( Σ p , C 2 N ) defined on dom X ~ p ( z ) = dom X p ( z ) \operatorname{dom}\widetilde{X}_p(z)=\operatorname{dom}X_p(z) dom X p ( z ) = dom X p ( z ) . In particular, as operators in Op S N / 2 − 1 ( Σ p ) \operatorname{Op}S^{-1}_{N/2}(\Sigma_p) Op S N /2 − 1 ( Σ p ) are bounded from L 2 ( Σ p , C N 2 ) L^2(\Sigma_p,\mathbb{C}^{\frac{N}{2}}) L 2 ( Σ p , C 2 N ) to H 1 ( Σ p , C N 2 ) H^1(\Sigma_p,\mathbb{C}^{\frac{N}{2}}) H 1 ( Σ p , C 2 N ) and hence, by Rellich’s embedding theorem, compact in L 2 ( Σ p , C N 2 ) L^2(\Sigma_p,\mathbb{C}^{\frac{N}{2}}) L 2 ( Σ p , C 2 N ) , one has
0 ∈ spec e s s X p ( z ) ⟺ 0 ∈ spec e s s X ~ p ( z ) . (93) 0\in\operatorname{spec}_{\mathrm{ess}}X_p(z)\Longleftrightarrow0\in\operatorname{spec}_{\mathrm{ess}}\widetilde{X}_p(z). \tag*{(93)} 0 ∈ spec ess X p ( z ) ⟺ 0 ∈ spec ess X p ( z ) . (93)
A simple algebra using again η p 2 − τ p 2 = 4 \eta_p^2-\tau_p^2=4 η p 2 − τ p 2 = 4 gives
( z + m ) + ( η p − τ p ) 2 ( z − m ) 4 = η p z + τ p m 2 ( η p − τ p ) , (z+m)+\frac{(\eta_p-\tau_p)^2(z-m)}{4}=\frac{\eta_p z+\tau_p m}{2}(\eta_p-\tau_p), ( z + m ) + 4 ( η p − τ p ) 2 ( z − m ) = 2 η p z + τ p m ( η p − τ p ) ,
and thus,
0 ∈ spec e s s X ~ p ( z ) ⟺ η p z + τ p m ∈ spec e s s ( 2 Λ p ( W p + W p − − 1 4 ) Λ p ) . 0\in\operatorname{spec}_{\mathrm{ess}}\widetilde{X}_p(z)\Longleftrightarrow\eta_pz+\tau_pm\in\operatorname{spec}_{\mathrm{ess}}\left(2\Lambda_p\left(W_p^+W_p^--\frac{1}{4}\right)\Lambda_p\right). 0 ∈ spec ess X p ( z ) ⟺ η p z + τ p m ∈ spec ess ( 2 Λ p ( W p + W p − − 4 1 ) Λ p ) .
By comparing with (93) and then with (92) we arrive at
spec e s s L p = − τ p η p m + 2 η p spec e s s R with R : = Λ p ( W p + W p − − 1 4 ) Λ p ; (94) \operatorname{spec}_{\mathrm{ess}}L_p=-\frac{\tau_p}{\eta_p}m+\frac{2}{\eta_p}\operatorname{spec}_{\mathrm{ess}}R\quad\text{with}\quad R:=\Lambda_p\left(W_p^+W_p^--\frac{1}{4}\right)\Lambda_p; \tag*{(94)} spec ess L p = − η p τ p m + η p 2 spec ess R with R := Λ p ( W p + W p − − 4 1 ) Λ p ; (94)
observe that η p ≠ 0 \eta_p\ne0 η p = 0 as we have η p 2 − τ p 2 = 4 \eta_p^2-\tau_p^2=4 η p 2 − τ p 2 = 4 . Now we consider the cases n = 2 n=2 n = 2 and n = 3 n=3 n = 3 separately.
(a) For n = 2 n=2 n = 2 one has by Lemma 4.5 (a) that W p + W p − = 1 4 m o d Op S 1 − ∞ ( Σ p ) W_p^+W_p^-=\frac{1}{4}\mod\operatorname{Op}S_1^{-\infty}(\Sigma_p) W p + W p − = 4 1 mod Op S 1 − ∞ ( Σ p ) . It follows that R ∈ Op S 1 − ∞ ( Σ p ) R\in\operatorname{Op}S_1^{-\infty}(\Sigma_p) R ∈ Op S 1 − ∞ ( Σ p ) . Since all operators in Op S 1 − ∞ ( Σ p ) \operatorname{Op}S_1^{-\infty}(\Sigma_p) Op S 1 − ∞ ( Σ p ) are compact in L 2 ( Σ p ) L^2(\Sigma_p) L 2 ( Σ p ) , we conclude spec e s s R = { 0 } \operatorname{spec}_{\mathrm{ess}}R=\{0\} spec ess R = { 0 } , and (94) yields
spec e s s L p = { − τ p η p m } . \operatorname{spec}_{\mathrm{ess}}L_p=\left\{-\frac{\tau_p}{\eta_p}m\right\}. spec ess L p = { − η p τ p m } .
(b) Consider now n = 3 n=3 n = 3 . Recall that in this case W p + = W p − = : W p W_p^+=W_p^-=:\,W_p W p + = W p − =: W p . By Lemma 4.5 (b) we have
R = Λ p ( W p 2 − 1 4 ) Λ p = Λ p ( ( σ ⋅ ν Σ ) 2 W p 2 + ( σ ⋅ ν Σ ) W p ( σ ⋅ ν Σ ) W p ) Λ p = Λ p ( σ ⋅ ν Σ ) [ ( σ ⋅ ν Σ ) W p + W p ( σ ⋅ ν Σ ) ] W p Λ p = Λ p ( σ ⋅ ν Σ ) Z p W p Λ p , \begin{aligned}
R=\Lambda_p\left(W_p^2-\frac{1}{4}\right)\Lambda_p=\Lambda_p\left((\sigma\cdot\nu^\Sigma)^2W_p^2+(\sigma\cdot\nu^\Sigma)W_p(\sigma\cdot\nu^\Sigma)W_p\right)\Lambda_p \\
&=\Lambda_p(\sigma\cdot\nu^\Sigma)\left[(\sigma\cdot\nu^\Sigma)W_p+W_p(\sigma\cdot\nu^\Sigma)\right]W_p\Lambda_p \\
&=\Lambda_p(\sigma\cdot\nu^\Sigma)Z_pW_p\Lambda_p,
\end{aligned} R = Λ p ( W p 2 − 4 1 ) Λ p = Λ p ( ( σ ⋅ ν Σ ) 2 W p 2 + ( σ ⋅ ν Σ ) W p ( σ ⋅ ν Σ ) W p ) Λ p = Λ p ( σ ⋅ ν Σ ) [ ( σ ⋅ ν Σ ) W p + W p ( σ ⋅ ν Σ ) ] W p Λ p = Λ p ( σ ⋅ ν Σ ) Z p W p Λ p ,
where in the last step the definition of Z p Z_p Z p from (74) was used.
Let x ∈ Σ p x\in\Sigma_p x ∈ Σ p be fixed. Using Lemma 4.3 (b), Lemma 4.5 (b), and Lemma 4.1 (c) one gets for the principal symbol of R R R with respect to normal coordinates near x x x (see Definition 4.2 (b))
σ R ( x , ξ ) = σ Λ p ( x , ξ ) σ σ ⋅ ν Σ ( x , ξ ) σ Z p ( x , ξ ) σ W p ( x , ξ ) σ Λ p ( x , ξ ) = 2 ∣ ξ ∣ ( σ ⋅ ν Σ ( x ) ) 1 2 ( κ 1 ( x ) − κ 2 ( x ) ) ( σ ⋅ ν Σ ( x ) ) ξ 1 ξ 2 ∣ ξ ∣ 3 1 2 σ ⋅ ( ξ 1 e 1 + ξ 2 e 2 ∣ ξ ∣ ) 2 ∣ ξ ∣ = 1 2 ( κ 1 ( x ) − κ 2 ( x ) ) ξ 1 ξ 2 ∣ ξ ∣ 3 σ ⋅ ( ξ 1 e 1 + ξ 2 e 2 ) . \begin{aligned}
\sigma_R(x,\xi)=\sigma_{\Lambda_p}(x,\xi)\sigma_{\sigma\cdot\nu^\Sigma}(x,\xi)\sigma_{Z_p}(x,\xi)\sigma_{W_p}(x,\xi)\sigma_{\Lambda_p}(x,\xi) \\
&=\sqrt{2|\xi|}(\sigma\cdot\nu^\Sigma(x))\frac{1}{2}\bigl(\kappa_1(x)-\kappa_2(x)\bigr)(\sigma\cdot\nu^\Sigma(x))\frac{\xi_1\xi_2}{|\xi|^3}\frac{1}{2}\sigma\cdot\left(\frac{\xi_1e_1+\xi_2e_2}{|\xi|}\right)\sqrt{2|\xi|} \\
&=\frac{1}{2}\bigl(\kappa_1(x)-\kappa_2(x)\bigr)\frac{\xi_1\xi_2}{|\xi|^3}\sigma\cdot(\xi_1e_1+\xi_2e_2).
\end{aligned} σ R ( x , ξ ) = σ Λ p ( x , ξ ) σ σ ⋅ ν Σ ( x , ξ ) σ Z p ( x , ξ ) σ W p ( x , ξ ) σ Λ p ( x , ξ ) = 2∣ ξ ∣ ( σ ⋅ ν Σ ( x )) 2 1 ( κ 1 ( x ) − κ 2 ( x ) ) ( σ ⋅ ν Σ ( x )) ∣ ξ ∣ 3 ξ 1 ξ 2 2 1 σ ⋅ ( ∣ ξ ∣ ξ 1 e 1 + ξ 2 e 2 ) 2∣ ξ ∣ = 2 1 ( κ 1 ( x ) − κ 2 ( x ) ) ∣ ξ ∣ 3 ξ 1 ξ 2 σ ⋅ ( ξ 1 e 1 + ξ 2 e 2 ) .
Recall that spec ( σ ⋅ a ) = { − ∣ a ∣ , ∣ a ∣ } \operatorname{spec}(\sigma\cdot a)=\{-|a|,|a|\} spec ( σ ⋅ a ) = { − ∣ a ∣ , ∣ a ∣ } for any a ∈ R 3 a\in\mathbb{R}^3 a ∈ R 3 . Thus, using ∣ ξ 1 e 1 + ξ 2 e 2 ∣ = ∣ ξ ∣ |\xi_1e_1+\xi_2e_2|=|\xi| ∣ ξ 1 e 1 + ξ 2 e 2 ∣ = ∣ ξ ∣ we arrive at
spec σ R ( x , ξ ) = { − ∣ κ 1 ( x ) − κ 2 ( x ) ∣ 2 ∣ ξ 1 ξ 2 ∣ ∣ ξ ∣ 2 , ∣ κ 1 ( x ) − κ 2 ( x ) ∣ 2 ∣ ξ 1 ξ 2 ∣ ∣ ξ ∣ 2 } . \operatorname{spec}\sigma_R(x,\xi)=\left\{-\frac{|\kappa_1(x)-\kappa_2(x)|}{2}\frac{|\xi_1\xi_2|}{|\xi|^2},\frac{|\kappa_1(x)-\kappa_2(x)|}{2}\frac{|\xi_1\xi_2|}{|\xi|^2}\right\}. spec σ R ( x , ξ ) = { − 2 ∣ κ 1 ( x ) − κ 2 ( x ) ∣ ∣ ξ ∣ 2 ∣ ξ 1 ξ 2 ∣ , 2 ∣ κ 1 ( x ) − κ 2 ( x ) ∣ ∣ ξ ∣ 2 ∣ ξ 1 ξ 2 ∣ } .
Since the range of the function ( ξ 1 , ξ 2 ) ↦ ∣ ξ 1 ξ 2 ∣ ∣ ξ ∣ 2 (\xi_1,\xi_2)\mapsto\frac{|\xi_1\xi_2|}{|\xi|^2} ( ξ 1 , ξ 2 ) ↦ ∣ ξ ∣ 2 ∣ ξ 1 ξ 2 ∣ , ξ ≠ 0 \xi\ne0 ξ = 0 , is [ 0 , 1 2 ] [0,\frac{1}{2}] [ 0 , 2 1 ] the usual rule for zeroth order pseudodifferential operators (see, e.g., [20 Lemma 11] ) gives
spec e s s R = ⋃ x ∈ Σ p ⋃ ξ ≠ 0 spec σ R ( x , ξ ) = [ − a p 4 , a p 4 ] . \operatorname{spec}_{\mathrm{ess}}R=\bigcup_{x\in\Sigma_p}\bigcup_{\xi\ne0}\operatorname{spec}\sigma_R(x,\xi)=\left[-\frac{a_p}{4},\frac{a_p}{4}\right]. spec ess R = x ∈ Σ p ⋃ ξ = 0 ⋃ spec σ R ( x , ξ ) = [ − 4 a p , 4 a p ] .
with a p = max x ∈ Σ p ∣ κ 1 ( x ) − κ 2 ( x ) ∣ a_p=\max_{x\in\Sigma_p}|\kappa_1(x)-\kappa_2(x)| a p = max x ∈ Σ p ∣ κ 1 ( x ) − κ 2 ( x ) ∣ . The substitution into (94) completes the proof. □
By combining Proposition 4.7 with Lemmas 4.9 and 4.13 we arrive at the following intermediate result.
Corollary 4.14. If ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p|\ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ for all p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } , then
spec e s s A η , τ ∩ res A = ( ⋃ p : η p 2 − τ p 2 = 4 J p ) ∩ res A (95) \operatorname{spec}_{\mathrm{ess}}A^{\eta,\tau}\cap\operatorname{res}A=\left(\bigcup_{p:\,\eta_p^2-\tau_p^2=4}J_p\right)\cap\operatorname{res}A \tag*{(95)} spec ess A η , τ ∩ res A = p : η p 2 − τ p 2 = 4 ⋃ J p ∩ res A (95) with the sets J p J_p J p given by (89) .
In order to complete the analysis of the essential spectrum, we have to show that (95) remains valid if the set res A \operatorname{res}A res A is removed on both sides. More precisely, we have to take care of the following remaining aspects: (I) it is still unclear if an eigenvalue in spec A \operatorname{spec} A spec A lying outside of J p J_p J p can belong to spec e s s A η , τ \operatorname{spec}_{\mathrm{ess}} A^{\eta,\tau} spec ess A η , τ ;
(II) even if η p 2 − τ p 2 = 4 \eta_p^2-\tau_p^2=4 η p 2 − τ p 2 = 4 for some p p p , the right-hand side of (4.34) can be empty, if J p J_p J p degenerates to a point which coincides with an eigenvalue of A A A .
The open point (I) will be answered in Lemma 4.16 and (II) will be subject of Lemma 4.18. Let us first recall the following abstract result [44 Theorem XIII.13] .
Proposition 4.15 (Meromorphic Fredholm theorem). Let H \mathcal{H} H be a Hilbert space and let U ⊂ C U\subset\mathbb{C} U ⊂ C be a connected open set. Let Z ⊂ U Z\subset U Z ⊂ U be a subset having no accumulation points in U U U and K : U ∖ Z → L ( H ) K:U\setminus Z\to\mathcal{L}(\mathcal{H}) K : U ∖ Z → L ( H ) be an analytic operator function which is meromorphic in U U U and such that
the operator K ( z ) K(z) K ( z ) is compact for any z ∈ U ∖ Z z\in U\setminus Z z ∈ U ∖ Z ,
at each z 0 ∈ Z z_0\in Z z 0 ∈ Z all coefficients of the principal part of the Laurent series of K ( z ) K(z) K ( z ) at z 0 z_0 z 0 are finite-dimensional operators.
Then either
(a) 0 ∈ spec ( I + K ( z ) ) 0\in\operatorname{spec}(I+K(z)) 0 ∈ spec ( I + K ( z )) for all z ∈ U ∖ Z z\in U\setminus Z z ∈ U ∖ Z ,
or
(b) there is a subset Z ′ ⊂ U Z'\subset U Z ′ ⊂ U without accumulation points in U U U such that the bounded inverse ( I + K ( z ) ) − 1 (I+K(z))^{-1} ( I + K ( z ) ) − 1 exists for all z ∈ U ∖ ( Z ∪ Z ′ ) z\in U\setminus(Z\cup Z') z ∈ U ∖ ( Z ∪ Z ′ ) and extends to an operator function analytic in U ∖ Z ′ U\setminus Z' U ∖ Z ′ and meromorphic in U U U , and for each z 0 ∈ Z ′ z_0\in Z' z 0 ∈ Z ′ all coefficients of the principal part of the Laurent series of ( I + K ( z ) ) − 1 (I+K(z))^{-1} ( I + K ( z ) ) − 1 at z 0 z_0 z 0 are finite-dimensional operators.
Lemma 4.16. If ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p|\ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ for all p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } , then
spec e s s A η , τ ⊂ ⋃ p : η p 2 − τ p 2 = 4 J p (96) \operatorname{spec}_{\mathrm{ess}} A^{\eta,\tau}\subset\bigcup_{p:\,\eta_p^2-\tau_p^2=4}J_p \tag*{(96)} spec ess A η , τ ⊂ p : η p 2 − τ p 2 = 4 ⋃ J p (96) with the sets J p J_p J p given by (4.28).
Proof. Denote the set on the right-hand side of (4.35) by J J J and note that the sought assertion is a direct consequence of the following claim:
The resolvent of A η , τ is meromorphic in C ∖ J and the coefficients of the principal part of the Laurent series of ( A η , τ − z ) − 1 at each pole are finite-dimensional operators. (97) \begin{gathered}
\text{The resolvent of }A^{\eta,\tau}\text{ is meromorphic in }\mathbb{C}\setminus J\text{ and the}\\
\text{coefficients of the principal part of the Laurent series of}\\
(A^{\eta,\tau}-z)^{-1}\text{ at each pole are finite-dimensional operators.} \tag*{(97)}
\end{gathered} The resolvent of A η , τ is meromorphic in C ∖ J and the coefficients of the principal part of the Laurent series of ( A η , τ − z ) − 1 at each pole are finite-dimensional operators. (97)
In fact, (4.36) means that the spectrum of A η , τ A^{\eta,\tau} A η , τ in C ∖ J \mathbb{C}\setminus J C ∖ J is purely discrete. Note also that due to the self-adjointness of A η , τ A^{\eta,\tau} A η , τ each pole of the resolvent is necessarily of first order. Recall that we have the resolvent formula
( A η , τ − z ) − 1 = ( A − z ) − 1 + Γ z ( Θ η , τ − M z ) − 1 Γ z ˉ ∗ , z ∈ res A η , τ ∩ res A , (98) (A^{\eta,\tau}-z)^{-1}=(A-z)^{-1}+\Gamma_z(\Theta^{\eta,\tau}-M_z)^{-1}\Gamma_{\bar z}^{*},\qquad z\in\operatorname{res} A^{\eta,\tau}\cap\operatorname{res} A, \tag*{(98)} ( A η , τ − z ) − 1 = ( A − z ) − 1 + Γ z ( Θ η , τ − M z ) − 1 Γ z ˉ ∗ , z ∈ res A η , τ ∩ res A , (98)
with Γ z \Gamma_z Γ z , M z M_z M z , and Θ η , τ \Theta^{\eta,\tau} Θ η , τ given by (3.20), (3.27), and (3.28), respectively. By abstract considerations the maps
z ↦ ( A − z ) − 1 , z ↦ Γ z , z ↦ M z z\mapsto(A-z)^{-1},\qquad z\mapsto\Gamma_z,\qquad z\mapsto M_z z ↦ ( A − z ) − 1 , z ↦ Γ z , z ↦ M z
are meromorphic in C \mathbb{C} C having only first-order poles with finite-dimensional residues, see Remark B.5. Furthermore, in view of the representation (4.22) we have
Θ η , τ − M z = L − z + T ( z ) , \Theta^{\eta,\tau}-M_z=L-z+T(z), Θ η , τ − M z = L − z + T ( z ) ,
where L : = ⨁ p = 1 P L p L:=\bigoplus_{p=1}^{P}L_p L := ⨁ p = 1 P L p is self-adjoint in L 2 ( Σ , C N ) L^2(\Sigma,\mathbb{C}^N) L 2 ( Σ , C N ) and
T : res A ∋ z ↦ T ( z ) ∈ L ( L 2 ( Σ , C N ) ) T:\operatorname{res} A\ni z\mapsto T(z)\in\mathcal{L}(L^2(\Sigma,\mathbb{C}^N)) T : res A ∋ z ↦ T ( z ) ∈ L ( L 2 ( Σ , C N ))
is a meromorphic operator-valued function having only first-order poles with finite-dimensional residues and such that for any z z z the value T ( z ) ∈ Op S N − 1 ( Σ ) T(z)\in\operatorname{Op}S_N^{-1}(\Sigma) T ( z ) ∈ Op S N − 1 ( Σ ) gives rise to a bounded operator from L 2 ( Σ , C N ) L^2(\Sigma,\mathbb{C}^N) L 2 ( Σ , C N ) to H 1 ( Σ , C N ) H^1(\Sigma,\mathbb{C}^N) H 1 ( Σ , C N ) and thus, it is a compact operator in L 2 ( Σ , C N ) L^2(\Sigma,\mathbb{C}^N) L 2 ( Σ , C N ) . Due to Lemma 4.9 and Lemma 4.13 one has spec e s s L = J \operatorname{spec}_{\mathrm{ess}}L=J spec ess L = J . Therefore, the resolvent z ↦ ( L − z ) − 1 z\mapsto(L-z)^{-1} z ↦ ( L − z ) − 1 is meromorphic in C ∖ J \mathbb{C}\setminus J C ∖ J with only first-order poles and finite-dimensional residues. We represent
Θ η , τ − M z = ( I + K ( z ) ) ( L − z ) , K ( z ) : = T ( z ) ( L − z ) − 1 , \Theta^{\eta,\tau}-M_z=(I+K(z))(L-z),\qquad K(z):=T(z)(L-z)^{-1}, Θ η , τ − M z = ( I + K ( z )) ( L − z ) , K ( z ) := T ( z ) ( L − z ) − 1 ,
and note that all K ( z ) K(z) K ( z ) are compact operators and satisfy the assumptions of the meromorphic Fredholm theorem (Proposition 4.15) on the domain U : = C ∖ J U:=\mathbb{C}\setminus J U := C ∖ J . The case (a) in Proposition 4.15 can be excluded as the operators Θ η , τ − M z \Theta^{\eta,\tau}-M_z Θ η , τ − M z and L − z L-z L − z are boundedly invertible for z ∈ C ∖ R z\in\mathbb{C}\setminus\mathbb{R} z ∈ C ∖ R ; for Θ η , τ − M z \Theta^{\eta,\tau}-M_z Θ η , τ − M z this follows from Theorem 4.8 (a), while for L − z L-z L − z this is clear due to the self-adjointness of L L L . It follows that z ↦ ( I + K ( z ) ) − 1 z\mapsto(I+K(z))^{-1} z ↦ ( I + K ( z ) ) − 1 is meromorphic in C ∖ J \mathbb{C}\setminus J C ∖ J with finite-dimensional coefficients in the principal parts of the Laurent series at the poles. Then
( Θ η , τ − M z ) − 1 = ( L − z ) − 1 ( I + K ( z ) ) − 1 (\Theta^{\eta,\tau}-M_z)^{-1}=(L-z)^{-1}(I+K(z))^{-1} ( Θ η , τ − M z ) − 1 = ( L − z ) − 1 ( I + K ( z ) ) − 1
has the same property, and the substitution into the resolvent formula (4.37) proves the required claim (4.36). □
Recall the following abstract convergence result [50 Satz 9.24] .
Proposition 4.17. Let T j T_j T j and T T T be self-adjoint operators in a Hilbert space such that T j → T T_j\to T T j → T , j → ∞ j\to\infty j → ∞ , in the norm resolvent sense. If λ j ∈ spec e s s T j \lambda_j\in\operatorname{spec}_{\mathrm{ess}}T_j λ j ∈ spec ess T j with λ j → λ \lambda_j\to\lambda λ j → λ , j → ∞ j\to\infty j → ∞ , then λ ∈ spec e s s T \lambda\in\operatorname{spec}_{\mathrm{ess}}T λ ∈ spec ess T .
Lemma 4.18. Let p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } and assume that η p 2 − τ p 2 = 4 \eta_p^2-\tau_p^2=4 η p 2 − τ p 2 = 4 . Then − τ p η p m ∈ spec e s s A η , τ -\frac{\tau_p}{\eta_p}m\in\operatorname{spec}_{\mathrm{ess}}A^{\eta,\tau} − η p τ p m ∈ spec ess A η , τ .
Proof. In the case that − τ p η p m ∈ res A -\frac{\tau_p}{\eta_p}m\in\operatorname{res}A − η p τ p m ∈ res A the claim follows from Corollary 4.14, since − τ p η p ∈ J p -\frac{\tau_p}{\eta_p}\in J_p − η p τ p ∈ J p by Lemma 4.13. The objective of the next steps (a)–(c) is to cover the case − τ p η p m ∈ spec A -\frac{\tau_p}{\eta_p}m\in\operatorname{spec}A − η p τ p m ∈ spec A .
(a) In the following, we denote by L p L_p L p the maximal realization as an operator in L 2 ( Σ p , C N ) L^2(\Sigma_p,\mathbb{C}^N) L 2 ( Σ p , C N ) associated with the pseudodifferential expression defined in (4.21). We start by proving the following auxiliary statement:
The operators L p L_p L p depend continuously on ( η p , τ p ) (\eta_p,\tau_p) ( η p , τ p ) on the set Ξ : = { ( η p , τ p ) : η p 2 − τ p 2 = 4 } \Xi:=\{(\eta_p,\tau_p):\eta_p^2-\tau_p^2=4\} Ξ := {( η p , τ p ) : η p 2 − τ p 2 = 4 } with respect to the norm resolvent topology.
The operators L p depend continuously on ( η p , τ p ) on the set Ξ : = { ( η p , τ p ) : η p 2 − τ p 2 = 4 } with respect to the norm resolvent topology. (99) \text{The operators }L_p\text{ depend continuously on }(\eta_p,\tau_p)\text{ on the set }\Xi:=\{(\eta_p,\tau_p):\eta_p^2-\tau_p^2=4\}\text{ with respect to the norm resolvent topology.} \tag*{(99)} The operators L p depend continuously on ( η p , τ p ) on the set Ξ := {( η p , τ p ) : η p 2 − τ p 2 = 4 } with respect to the norm resolvent topology. (99)
It is sufficient to consider the case m = 0 m=0 m = 0 (as adding a fixed bounded operator does not influence the norm resolvent convergence). Consider the block operator matrix (with a ⊗ I a\otimes I a ⊗ I shortened as a a a )
L p : = ( 1 η p + τ p Λ p 2 + i Λ p W p + Λ p Λ p W p − Λ p 1 η p − τ p Λ p 2 + i ) , dom L p = H 1 ( Σ p , C N ) . \mathcal{L}_p:=\begin{pmatrix}\dfrac{1}{\eta_p+\tau_p}\Lambda_p^2+i & \Lambda_pW_p^+\Lambda_p \\ \Lambda_pW_p^-\Lambda_p & \dfrac{1}{\eta_p-\tau_p}\Lambda_p^2+i\end{pmatrix},\qquad\operatorname{dom}\mathcal{L}_p=H^1(\Sigma_p,\mathbb{C}^N). L p := η p + τ p 1 Λ p 2 + i Λ p W p − Λ p Λ p W p + Λ p η p − τ p 1 Λ p 2 + i , dom L p = H 1 ( Σ p , C N ) .
As in the proof of Lemma 4.13 (see the considerations on the operator L p ( z ) \mathcal{L}_p(z) L p ( z ) there) one sees that L p \mathcal{L}_p L p is closable, L p ‾ = i − L p \overline{\mathcal{L}_p}=i-L_p L p = i − L p (since H 1 ( Σ p , C N ) H^1(\Sigma_p,\mathbb{C}^N) H 1 ( Σ p , C N ) is a core for L p L_p L p by Proposition 4.7), and L p \mathcal{L}_p L p satisfies all assumptions in Proposition 4.12. For better readability we denote the right lower entry in L p \mathcal{L}_p L p by D p D_p D p . Then we conclude
L p ‾ = ( i − L p ) = ( I Λ p W p + Λ p D p − 1 0 I ) ( X ( η p , τ p ) ‾ 0 0 D p ) ( I 0 D p − 1 Λ p W p − Λ p I ) , (100) \overline{\mathcal{L}_p}=(i-L_p)=\begin{pmatrix}I&\Lambda_pW_p^+\Lambda_pD_p^{-1}\\0&I\end{pmatrix}\begin{pmatrix}\overline{X(\eta_p,\tau_p)}&0\\0&D_p\end{pmatrix}\begin{pmatrix}I&0\\D_p^{-1}\Lambda_pW_p^-\Lambda_p&I\end{pmatrix}, \tag*{(100)} L p = ( i − L p ) = ( I 0 Λ p W p + Λ p D p − 1 I ) ( X ( η p , τ p ) 0 0 D p ) ( I D p − 1 Λ p W p − Λ p 0 I ) , (100)
where X ( η p , τ p ) X(\eta_p,\tau_p) X ( η p , τ p ) is the Schur complement associated with L p \mathcal{L}_p L p defined as in Proposition 4.12,
X ( η p , τ p ) = 1 η p + τ p Λ p 2 + i − Λ p W p + Λ p D p − 1 Λ p W p − Λ p , X(\eta_p,\tau_p)=\frac{1}{\eta_p+\tau_p}\Lambda_p^2+i-\Lambda_pW_p^+\Lambda_pD_p^{-1}\Lambda_pW_p^-\Lambda_p, X ( η p , τ p ) = η p + τ p 1 Λ p 2 + i − Λ p W p + Λ p D p − 1 Λ p W p − Λ p ,
dom X ( η p , τ p ) = H 1 ( Σ p , C N / 2 ) . \operatorname{dom}X(\eta_p,\tau_p)=H^1(\Sigma_p,\mathbb{C}^{N/2}). dom X ( η p , τ p ) = H 1 ( Σ p , C N /2 ) .
Note that X ( η p , τ p ) ∈ Op S N / 2 0 ( Σ p ) X(\eta_p,\tau_p)\in\operatorname{Op}S^0_{N/2}(\Sigma_p) X ( η p , τ p ) ∈ Op S N /2 0 ( Σ p ) , since its principal symbol of order one is
2 ∣ ξ ∣ ( 1 η p + τ p − η p − τ p 4 ) = 0 2|\xi|\left(\frac{1}{\eta_p+\tau_p}-\frac{\eta_p-\tau_p}{4}\right)=0 2∣ ξ ∣ ( η p + τ p 1 − 4 η p − τ p ) = 0
by Corollary 4.4 and η p 2 − τ p 2 = 4 \eta_p^2-\tau_p^2=4 η p 2 − τ p 2 = 4 ; cf. the considerations in the proof of Lemma 4.13 leading to (4.33). Hence, X ‾ ( η p , τ p ) \overline{X}(\eta_p,\tau_p) X ( η p , τ p ) is the bounded extension of X ( η p , τ p ) X(\eta_p,\tau_p) X ( η p , τ p ) to L 2 ( Σ p , C N / 2 ) L^2(\Sigma_p,\mathbb{C}^{N/2}) L 2 ( Σ p , C N /2 ) . With a computation one sees that
X ‾ ( η p , τ p ) = 1 η p + τ p Λ p 2 + i − Λ p W p + Λ p ( 1 η p − τ p Λ p 2 + i ) − 1 Λ p W p − Λ p = i + η p − τ p 4 Λ p 2 − ( η p − τ p ) Λ p W p + Λ p ( Λ p 2 + i ( η p − τ p ) ) − 1 Λ p W p − Λ p = i + η p − τ p 4 Λ p 2 − ( η p − τ p ) Λ p W p + [ I − i ( η p − τ p ) ( Λ p 2 + i ( η p − τ p ) ) − 1 ] W p − Λ p = i + ( η p − τ p ) Λ p ( 1 4 − W p + W p − ) Λ p + i ( η p − τ p ) 2 Λ p W p + ( Λ p 2 + i ( η p − τ p ) ) − 1 W p − Λ p = i + ( η p − τ p ) K 1 + i ( η p − τ p ) 2 K 2 ( η p , τ p ) , \begin{aligned}
\overline{X}(\eta_p,\tau_p)&=\frac{1}{\eta_p+\tau_p}\Lambda_p^2+i-\Lambda_pW_p^+\Lambda_p\left(\frac{1}{\eta_p-\tau_p}\Lambda_p^2+i\right)^{-1}\Lambda_pW_p^-\Lambda_p\\
&=i+\frac{\eta_p-\tau_p}{4}\Lambda_p^2-(\eta_p-\tau_p)\Lambda_pW_p^+\Lambda_p(\Lambda_p^2+i(\eta_p-\tau_p))^{-1}\Lambda_pW_p^-\Lambda_p\\
&=i+\frac{\eta_p-\tau_p}{4}\Lambda_p^2-(\eta_p-\tau_p)\Lambda_pW_p^+\left[I-i(\eta_p-\tau_p)(\Lambda_p^2+i(\eta_p-\tau_p))^{-1}\right]W_p^-\Lambda_p\\
&=i+(\eta_p-\tau_p)\Lambda_p\left(\frac{1}{4}-W_p^+W_p^-\right)\Lambda_p\\
&\quad+i(\eta_p-\tau_p)^2\Lambda_pW_p^+(\Lambda_p^2+i(\eta_p-\tau_p))^{-1}W_p^-\Lambda_p\\
&=i+(\eta_p-\tau_p)K_1+i(\eta_p-\tau_p)^2K_2(\eta_p,\tau_p),
\end{aligned} X ( η p , τ p ) = η p + τ p 1 Λ p 2 + i − Λ p W p + Λ p ( η p − τ p 1 Λ p 2 + i ) − 1 Λ p W p − Λ p = i + 4 η p − τ p Λ p 2 − ( η p − τ p ) Λ p W p + Λ p ( Λ p 2 + i ( η p − τ p ) ) − 1 Λ p W p − Λ p = i + 4 η p − τ p Λ p 2 − ( η p − τ p ) Λ p W p + [ I − i ( η p − τ p ) ( Λ p 2 + i ( η p − τ p ) ) − 1 ] W p − Λ p = i + ( η p − τ p ) Λ p ( 4 1 − W p + W p − ) Λ p + i ( η p − τ p ) 2 Λ p W p + ( Λ p 2 + i ( η p − τ p ) ) − 1 W p − Λ p = i + ( η p − τ p ) K 1 + i ( η p − τ p ) 2 K 2 ( η p , τ p ) ,
where we have set
K 1 : = Λ p ( 1 4 − W p + W p − ) Λ p K_1:=\Lambda_p\left(\frac{1}{4}-W_p^+W_p^-\right)\Lambda_p K 1 := Λ p ( 4 1 − W p + W p − ) Λ p
and
K 2 ( η p , τ p ) : = Λ p W p + ( Λ p 2 + i ( η p − τ p ) ) − 1 W p − Λ p . K_2(\eta_p,\tau_p):=\Lambda_pW_p^+(\Lambda_p^2+i(\eta_p-\tau_p))^{-1}W_p^-\Lambda_p. K 2 ( η p , τ p ) := Λ p W p + ( Λ p 2 + i ( η p − τ p ) ) − 1 W p − Λ p .
We first remark that X ( η p , τ p ) X(\eta_p,\tau_p) X ( η p , τ p ) always has a bounded inverse, as with W p + = ( W p − ) ∗ W_p^+=(W_p^-)^* W p + = ( W p − ) ∗ and
ℑ D p − 1 = ℑ ( 1 η p − τ p Λ p 2 + i ) − 1 ≤ 0 \Im D_p^{-1}=\Im\left(\frac{1}{\eta_p-\tau_p}\Lambda_p^2+\mathrm{i}\right)^{-1}\leq0 ℑ D p − 1 = ℑ ( η p − τ p 1 Λ p 2 + i ) − 1 ≤ 0
one finds that
ℑ X ( η p , τ p ) ‾ = ℑ i − ℑ [ Λ p W p + Λ p D p − 1 Λ p W p − Λ p ] = 1 − Λ p W p + Λ p [ ℑ D p − 1 ] Λ p W p − Λ p ≥ 1 \begin{aligned}
\Im\overline{X(\eta_p,\tau_p)}
&=\Im\mathrm{i}-\Im\left[\Lambda_pW_p^+\Lambda_pD_p^{-1}\Lambda_pW_p^-\Lambda_p\right]\\
&=1-\Lambda_pW_p^+\Lambda_p\left[\Im D_p^{-1}\right]\Lambda_pW_p^-\Lambda_p\geq1
\end{aligned} ℑ X ( η p , τ p ) = ℑ i − ℑ [ Λ p W p + Λ p D p − 1 Λ p W p − Λ p ] = 1 − Λ p W p + Λ p [ ℑ D p − 1 ] Λ p W p − Λ p ≥ 1
in the sense of quadratic forms. In particular, zero does not belong to the closure of the numerical range of X ( η p , τ p ) ‾ \overline{X(\eta_p,\tau_p)} X ( η p , τ p ) and thus, as X ( η p , τ p ) ‾ \overline{X(\eta_p,\tau_p)} X ( η p , τ p ) is bounded, it is also boundedly invertible. The fixed operator K 1 K_1 K 1 belongs to Op S N 2 0 ( Σ p ) \operatorname{Op}S^0_{\frac{N}{2}}(\Sigma_p) Op S 2 N 0 ( Σ p ) by Corollary 4.4, hence, it is bounded in L 2 ( Σ p , C N 2 ) L^2(\Sigma_p,\mathbb{C}^{\frac{N}{2}}) L 2 ( Σ p , C 2 N ) . On the other hand, for any s ∈ R s\in\mathbb{R} s ∈ R one has fixed bounded maps
Λ p : H s ( Σ p ) → H s − 1 2 ( Σ p ) , W p ± : H s ( Σ p , C N 2 ) → H s ( Σ p , C N 2 ) , \Lambda_p:H^s(\Sigma_p)\to H^{s-\frac{1}{2}}(\Sigma_p),\qquad W_p^\pm:H^s\left(\Sigma_p,\mathbb{C}^{\frac{N}{2}}\right)\to H^s\left(\Sigma_p,\mathbb{C}^{\frac{N}{2}}\right), Λ p : H s ( Σ p ) → H s − 2 1 ( Σ p ) , W p ± : H s ( Σ p , C 2 N ) → H s ( Σ p , C 2 N ) ,
and the continuous map
R ∋ λ ↦ ( Λ p 2 + i λ ) − 1 ∈ L ( H s ( Σ p ) , H s + 1 ( Σ p ) ) , \mathbb{R}\ni\lambda\mapsto(\Lambda_p^2+\mathrm{i}\lambda)^{-1}\in\mathcal{L}\left(H^s(\Sigma_p),H^{s+1}(\Sigma_p)\right), R ∋ λ ↦ ( Λ p 2 + i λ ) − 1 ∈ L ( H s ( Σ p ) , H s + 1 ( Σ p ) ) ,
which shows that the operators D p − 1 D_p^{-1} D p − 1 , Λ p W p + Λ p D p − 1 \Lambda_pW_p^+\Lambda_pD_p^{-1} Λ p W p + Λ p D p − 1 , D p − 1 Λ p W p − Λ p D_p^{-1}\Lambda_pW_p^-\Lambda_p D p − 1 Λ p W p − Λ p , and K 2 ( η p , τ p ) K_2(\eta_p,\tau_p) K 2 ( η p , τ p ) depend continuously on ( η p , τ p ) ∈ Ξ (\eta_p,\tau_p)\in\Xi ( η p , τ p ) ∈ Ξ in L ( L 2 ( Σ p , C N 2 ) ) \mathcal{L}\left(L^2\left(\Sigma_p,\mathbb{C}^{\frac{N}{2}}\right)\right) L ( L 2 ( Σ p , C 2 N ) ) . Then, the same applies to X ( η p , τ p ) ‾ \overline{X(\eta_p,\tau_p)} X ( η p , τ p ) and hence also to X ( η p , τ p ) ‾ − 1 \overline{X(\eta_p,\tau_p)}^{-1} X ( η p , τ p ) − 1 . To complete the argument note that (4.39) implies
( i − L p ) − 1 = ( I 0 − D p − 1 Λ p W p − Λ p I ) ( X ( η p , τ p ) ‾ − 1 0 0 D p − 1 ) ( I − Λ p W p + Λ p D p − 1 0 I ) , (101) (\mathrm{i}-L_p)^{-1}=
\begin{pmatrix}
I&0\\
-D_p^{-1}\Lambda_pW_p^-\Lambda_p&I
\end{pmatrix}
\begin{pmatrix}
\overline{X(\eta_p,\tau_p)}^{-1}&0\\
0&D_p^{-1}
\end{pmatrix}
\begin{pmatrix}
I&-\Lambda_pW_p^+\Lambda_pD_p^{-1}\\
0&I
\end{pmatrix}, \tag*{(101)} ( i − L p ) − 1 = ( I − D p − 1 Λ p W p − Λ p 0 I ) ( X ( η p , τ p ) − 1 0 0 D p − 1 ) ( I 0 − Λ p W p + Λ p D p − 1 I ) , (101)
and the above considerations show that all blocks in (4.40) depend continuously on ( η p , τ p ) ∈ Ξ (\eta_p,\tau_p)\in\Xi ( η p , τ p ) ∈ Ξ in L ( L 2 ( Σ p , C N 2 ) ) \mathcal{L}\left(L^2\left(\Sigma_p,\mathbb{C}^{\frac{N}{2}}\right)\right) L ( L 2 ( Σ p , C 2 N ) ) . This yields the claim (4.38).
(b) Assume that m ≠ 0 m\neq0 m = 0 and − τ p η p m ∈ spec A -\frac{\tau_p}{\eta_p}m\in\operatorname{spec}A − η p τ p m ∈ spec A . For ( η p ′ , τ p ′ ) ∈ R 2 (\eta_p',\tau_p')\in\mathbb{R}^2 ( η p ′ , τ p ′ ) ∈ R 2 with ( η p ′ ) 2 − ( τ p ′ ) 2 = 4 (\eta_p')^2-(\tau_p')^2=4 ( η p ′ ) 2 − ( τ p ′ ) 2 = 4 and ( η p ′ , τ p ′ ) ≠ ( η p , τ p ) (\eta_p',\tau_p')\neq(\eta_p,\tau_p) ( η p ′ , τ p ′ ) = ( η p , τ p ) define locally constant functions η ~ , τ ~ : Σ → R \widetilde{\eta},\widetilde{\tau}:\Sigma\to\mathbb{R} η , τ : Σ → R by
η ~ = η on Σ ∖ Σ p , τ ~ = τ on Σ ∖ Σ p , \widetilde{\eta}=\eta\ \text{on }\Sigma\setminus\Sigma_p,\qquad\widetilde{\tau}=\tau\ \text{on }\Sigma\setminus\Sigma_p, η = η on Σ ∖ Σ p , τ = τ on Σ ∖ Σ p ,
η ~ = η p ′ on Σ p , τ ~ = τ p ′ on Σ p , \widetilde{\eta}=\eta_p'\ \text{on }\Sigma_p,\qquad\widetilde{\tau}=\tau_p'\ \text{on }\Sigma_p, η = η p ′ on Σ p , τ = τ p ′ on Σ p ,
and consider the associated operator A η ~ , τ ~ A^{\widetilde{\eta},\widetilde{\tau}} A η , τ . It follows from the definitions (3.28) and (4.21) and from Proposition 4.7 (with z = ζ z=\zeta z = ζ , so that M ζ = 0 M_\zeta=0 M ζ = 0 ) that
Θ η , τ = ⨁ p = 1 P L p + B \Theta^{\eta,\tau}=\bigoplus_{p=1}^{P}L_p+B Θ η , τ = p = 1 ⨁ P L p + B
holds with some bounded operator B B B independent of ( η , τ ) (\eta,\tau) ( η , τ ) . Hence, part (a) implies that Θ η ~ , τ ~ \Theta^{\widetilde{\eta},\widetilde{\tau}} Θ η , τ converges to Θ η , τ \Theta^{\eta,\tau} Θ η , τ in the norm resolvent sense for ( η p ′ , τ p ′ ) → ( η p , τ p ) (\eta_p',\tau_p')\to(\eta_p,\tau_p) ( η p ′ , τ p ′ ) → ( η p , τ p ) . Lemma B.3 implies the norm resolvent convergence of A η ~ , τ ~ A^{\widetilde{\eta},\widetilde{\tau}} A η , τ to A η , τ A^{\eta,\tau} A η , τ in the same regime. Note that the map ( η p ′ , τ p ′ ) ↦ τ p ′ η p ′ (\eta_p',\tau_p')\mapsto\frac{\tau_p'}{\eta_p'} ( η p ′ , τ p ′ ) ↦ η p ′ τ p ′ is injective on each of the two connected component of Ξ \Xi Ξ defined in (4.38). As m ≠ 0 m\neq0 m = 0 , it follows that λ ′ : = − m τ p ′ η p ′ ≠ − m τ p η p \lambda' := -m\frac{\tau_p'}{\eta_p'}\neq-m\frac{\tau_p}{\eta_p} λ ′ := − m η p ′ τ p ′ = − m η p τ p for all ( η p ′ , τ p ′ ) ≠ ( η p , τ p ) (\eta_p',\tau_p')\neq(\eta_p,\tau_p) ( η p ′ , τ p ′ ) = ( η p , τ p ) sufficiently close to ( η p , τ p ) (\eta_p,\tau_p) ( η p , τ p ) , and since spec A \operatorname{spec}A spec A consists of isolated points, also λ ′ ∉ spec A \lambda'\notin\operatorname{spec}A λ ′ ∈ / spec A for all such ( η p ′ , τ p ′ ) (\eta_p',\tau_p') ( η p ′ , τ p ′ ) . As in the first step of the proof, Corollary 4.14 applied to A η ~ , τ ~ A^{\widetilde{\eta},\widetilde{\tau}} A η , τ shows that λ ′ ∈ spec e s s A η ′ , τ ′ \lambda' \in\operatorname{spec}_{\mathrm{ess}} A^{\eta',\tau'} λ ′ ∈ spec ess A η ′ , τ ′ for all ( η p ′ , τ p ′ ) (\eta'_p,\tau'_p) ( η p ′ , τ p ′ ) sufficiently close to ( η p , τ p ) (\eta_p,\tau_p) ( η p , τ p ) . Then Proposition 4.17 yields
− τ p η p m = lim ( η p ′ , τ p ′ ) → ( η p , τ p ) λ ′ ∈ spec e s s A η , τ . -\frac{\tau_p}{\eta_p}m=\lim_{(\eta'_p,\tau'_p)\to(\eta_p,\tau_p)}\lambda'\in\operatorname{spec}_{\mathrm{ess}} A^{\eta,\tau}. − η p τ p m = ( η p ′ , τ p ′ ) → ( η p , τ p ) lim λ ′ ∈ spec ess A η , τ .
(c) Consider the remaining case m = 0 m=0 m = 0 and − m τ p η p = 0 ∈ spec A -m\frac{\tau_p}{\eta_p}=0\in\operatorname{spec} A − m η p τ p = 0 ∈ spec A . We have obviously A η , τ + M β → A η , τ A^{\eta,\tau}+M\beta\to A^{\eta,\tau} A η , τ + M β → A η , τ , as M → 0 M\to0 M → 0 , in the norm resolvent sense. For all M ≠ 0 M\ne0 M = 0 we have − τ p η p M ∈ spec e s s ( A η , τ + M β ) -\frac{\tau_p}{\eta_p}M\in\operatorname{spec}_{\mathrm{ess}}(A^{\eta,\tau}+M\beta) − η p τ p M ∈ spec ess ( A η , τ + M β ) by the first step of the proof if − τ p η p M ∈ res ( A + M β ) -\frac{\tau_p}{\eta_p}M\in\operatorname{res}(A+M\beta) − η p τ p M ∈ res ( A + M β ) , and by (b) if − τ p η p M ∈ spec ( A + M β ) -\frac{\tau_p}{\eta_p}M\in\operatorname{spec}(A+M\beta) − η p τ p M ∈ spec ( A + M β ) ; with the help of Proposition 4.17 we obtain
0 = lim M → 0 , M ≠ 0 − τ p η p M ∈ spec e s s A η , τ . 0=\lim_{M\to0,M\ne0}-\frac{\tau_p}{\eta_p}M\in\operatorname{spec}_{\mathrm{ess}} A^{\eta,\tau}. 0 = M → 0 , M = 0 lim − η p τ p M ∈ spec ess A η , τ .
We summarize all the constructions to arrive at the main result of the paper:
Theorem 4.19 (Essential spectrum of shell interactions). Assume that ∣ η p ∣ ≠ ∣ τ p ∣ |\eta_p|\ne|\tau_p| ∣ η p ∣ = ∣ τ p ∣ for all p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } . Then
spec e s s A η , τ = ⋃ p = 1 P J p , (102) \operatorname{spec}_{\mathrm{ess}} A^{\eta,\tau}=\bigcup_{p=1}^{P}J_p, \tag*{(102)} spec ess A η , τ = p = 1 ⋃ P J p , (102) where
J p : = { ∅ , η p 2 − τ p 2 ≠ 4 , { − τ p η p m } , n = 2 , η p 2 − τ p 2 = 4 , − τ p η p m + 1 η p [ − a p 2 , a p 2 ] , n = 3 , η p 2 − τ p 2 = 4 , J_p:=
\begin{cases}
\varnothing, & \eta_p^2-\tau_p^2\ne4,\\
\left\{-\frac{\tau_p}{\eta_p}m\right\}, & n=2,\ \eta_p^2-\tau_p^2=4,\\
-\frac{\tau_p}{\eta_p}m+\frac{1}{\eta_p}\left[-\frac{a_p}{2},\frac{a_p}{2}\right], & n=3,\ \eta_p^2-\tau_p^2=4,
\end{cases} J p := ⎩ ⎨ ⎧ ∅ , { − η p τ p m } , − η p τ p m + η p 1 [ − 2 a p , 2 a p ] , η p 2 − τ p 2 = 4 , n = 2 , η p 2 − τ p 2 = 4 , n = 3 , η p 2 − τ p 2 = 4 , where a p : = max x ∈ Σ p ∣ κ 1 ( x ) − κ 2 ( x ) ∣ a_p:=\max_{x\in\Sigma_p}|\kappa_1(x)-\kappa_2(x)| a p := max x ∈ Σ p ∣ κ 1 ( x ) − κ 2 ( x ) ∣ and κ 1 ( x ) \kappa_1(x) κ 1 ( x ) , κ 2 ( x ) \kappa_2(x) κ 2 ( x ) are the principal curvatures of Σ p \Sigma_p Σ p in x x x .
Proof. The inclusion ⊂ \subset ⊂ in (102) holds by Lemma 4.16, and it remains to show the reverse inclusion.
Let p ∈ { 1 , … , P } p\in\{1,\ldots,P\} p ∈ { 1 , … , P } with η p 2 − τ p 2 = 4 \eta_p^2-\tau_p^2=4 η p 2 − τ p 2 = 4 . If J p = { − τ p η p m } J_p=\left\{-\frac{\tau_p}{\eta_p}m\right\} J p = { − η p τ p m } , then J p ⊂ spec e s s A η , τ J_p\subset\operatorname{spec}_{\mathrm{ess}} A^{\eta,\tau} J p ⊂ spec ess A η , τ by Lemma 4.18. Assume now that J p J_p J p has a non-empty interior, then by Corollary 4.14 it holds that J p ′ : = J p ∩ res A ⊂ spec e s s A η , τ J'_p:=J_p\cap\operatorname{res} A\subset\operatorname{spec}_{\mathrm{ess}} A^{\eta,\tau} J p ′ := J p ∩ res A ⊂ spec ess A η , τ . The essential spectrum of A η , τ A^{\eta,\tau} A η , τ is a closed subset of R \mathbb{R} R and the spectrum of A A A is discrete, therefore J p ′ J'_p J p ′ is dense in J p J_p J p and we conclude J p = J p ′ ‾ ⊂ spec e s s A η , τ J_p=\overline{J'_p}\subset\operatorname{spec}_{\mathrm{ess}} A^{\eta,\tau} J p = J p ′ ⊂ spec ess A η , τ .
Remark 4.20. The case ∣ η p ∣ = ∣ τ p ∣ |\eta_p|=|\tau_p| ∣ η p ∣ = ∣ τ p ∣ was only excluded for technical reasons, namely in order to ensure the invertibility of η I N + τ β \eta I_N+\tau\beta η I N + τ β in the proof of Lemma 3.7. It can still be covered by introducing additional projectors similarly to [13 Proposition 4.3] , and one can show that the result of Theorem 4.19 still holds with J p = ∅ J_p=\varnothing J p = ∅ for ∣ η p ∣ = ∣ τ p ∣ |\eta_p|=|\tau_p| ∣ η p ∣ = ∣ τ p ∣ .
Remark 4.21. Throughout this paper the interaction strengths are constant on each connected component Σ p \Sigma_p Σ p . If η \eta η and τ \tau τ are allowed to vary along Σ \Sigma Σ , then a non-degenerate interval of essential spectrum can appear already for n = 2 n=2 n = 2 . This was shown in [17 ] for the Landau-Dirac operator in R 2 \mathbb{R}^2 R 2 and smooth coefficients η \eta η and τ \tau τ on a closed curve Σ \Sigma Σ satisfying η 2 − τ 2 = 4 \eta^2-\tau^2=4 η 2 − τ 2 = 4 on Σ \Sigma Σ : in this case the essential spectrum consists of the Landau-Dirac levels and the range of the function − m τ η -m\frac{\tau}{\eta} − m η τ on Σ \Sigma Σ . It is also pointed out in [17 ] that an adaptation of the present approach should give the analogous result on bounded domains. In fact, if n = 2 n=2 n = 2 and η , τ ∈ C ∞ ( Σ p ) \eta,\tau\in C^\infty(\Sigma_p) η , τ ∈ C ∞ ( Σ p ) with η p 2 − τ p 2 = 4 \eta_p^2-\tau_p^2=4 η p 2 − τ p 2 = 4 on Σ p \Sigma_p Σ p , then in the proof of Lemma 4.13 the constants 1 η p ± τ p \frac{1}{\eta_p\pm\tau_p} η p ± τ p 1 have to be replaced by the corresponding multiplication operators, which commute with W p ± W_p^\pm W p ± up to operators with smooth integral kernels, and one find that X p ( z ) ‾ \overline{X_p(z)} X p ( z ) coincides, up to a compact operator, with the operator of multiplication by the function 1 2 ( η − τ ) ( η z + τ m ) \frac{1}{2}(\eta-\tau)(\eta z+\tau m) 2 1 ( η − τ ) ( η z + τ m ) . This indicates that for n = 2 n=2 n = 2 the set J p J_p J p in Theorem 4.19 has to be replaced by the range of − m τ η -m\frac{\tau}{\eta} − m η τ on Σ p \Sigma_p Σ p , which is a non-degenerate interval if m ≠ 0 m\ne0 m = 0 and η , τ \eta,\tau η , τ are not constant on Σ p \Sigma_p Σ p . For n = 3 n=3 n = 3 additional work is needed, as the commutators of W p W_p W p with non-constant multiplication operators are not smoothing.
Appendix A. Notations on pseudodifferential operators and proof of Lemma 2.9
In the following, let M ⊂ R n M\subset\mathbb{R}^n M ⊂ R n be a smooth, closed, and compact hypersurface in R n \mathbb{R}^n R n , i.e. a curve in R 2 \mathbb{R}^2 R 2 or a surface in R 3 \mathbb{R}^3 R 3 . We use the following convention for the Fourier transform in R r \mathbb{R}^r R r . For f f f in the Schwartz class, it is given by
f ^ ( ξ ) = ∫ R r e − i ⟨ x , ξ ⟩ f ( x ) d x . \widehat{f}(\xi)=\int_{\mathbb{R}^r}e^{-i\langle x,\xi\rangle}f(x)\,\mathrm{d}x. f ( ξ ) = ∫ R r e − i ⟨ x , ξ ⟩ f ( x ) d x .
We denote by Op S q k ( M ) \operatorname{Op}S_q^k(M) Op S q k ( M ) the class of classical pseudodifferential operators of order ≤ k \le k ≤ k acting on sections of M × C q M\times\mathbb{C}^q M × C q defined as in [48 Chapter 7] ; we refer to the latter reference for basic definitions and properties of pseudodifferential operators on manifolds. Moreover, we set Op S q − ∞ ( M ) : = ⋂ ℓ ∈ R Op S q ℓ ( M ) \operatorname{Op}S_q^{-\infty}(M):=\bigcap_{\ell\in\mathbb{R}}\operatorname{Op}S_q^\ell(M) Op S q − ∞ ( M ) := ⋂ ℓ ∈ R Op S q ℓ ( M ) . For two pseudodifferential operators A , B ∈ Op S q k ( M ) A,B\in\operatorname{Op}S_q^k(M) A , B ∈ Op S q k ( M ) and ℓ < k \ell<k ℓ < k we write A = B mod Op S q ℓ ( M ) A=B\ \operatorname{mod}\ \operatorname{Op}S_q^\ell(M) A = B mod Op S q ℓ ( M ) , if A − B ∈ Op S q ℓ ( M ) A-B\in\operatorname{Op}S_q^\ell(M) A − B ∈ Op S q ℓ ( M ) . Furthermore, we will write σ A ( x , ξ ) \sigma_A(x,\xi) σ A ( x , ξ ) for the principal symbol of A ∈ Op S q k ( M ) A\in\operatorname{Op}S_q^k(M) A ∈ Op S q k ( M ) .
In the following we recall some properties of pseudodifferential operators on M M M that will be used frequently in the paper. Let A ∈ Op S q k ( M ) A\in\operatorname{Op}S_q^k(M) A ∈ Op S q k ( M ) . Then, for any s ∈ R s\in\mathbb{R} s ∈ R the map A A A extends uniquely to a bounded operator A : H s ( M , C q ) → H s − k ( M , C q ) A:H^s(M,\mathbb{C}^q)\to H^{s-k}(M,\mathbb{C}^q) A : H s ( M , C q ) → H s − k ( M , C q ) , see [48 Chapter 7, Proposition 5.5 and Section 10] . In particular, if k < 0 k<0 k < 0 , then A A A gives rise to a compact operator in L 2 ( M , C q ) L^2(M,\mathbb{C}^q) L 2 ( M , C q ) by Rellich’s embedding theorem. If k > 0 k>0 k > 0 , then it is a consequence of Friedrichs’ lemma, see e.g. [47 Chapter II, Proposition 7.4] , that A : H k ( M , C q ) → L 2 ( M , C q ) A:H^k(M,\mathbb{C}^q)\to L^2(M,\mathbb{C}^q) A : H k ( M , C q ) → L 2 ( M , C q ) viewed as an operator in L 2 ( M , C q ) L^2(M,\mathbb{C}^q) L 2 ( M , C q ) is closable and its closure is defined on the maximal domain in L 2 ( M , C q ) L^2(M,\mathbb{C}^q) L 2 ( M , C q ) ; in particular, if A A A is symmetric with respect to the inner product in L 2 ( M , C q ) L^2(M,\mathbb{C}^q) L 2 ( M , C q ) , then its maximal realization is self-adjoint in L 2 ( M , C q ) L^2(M,\mathbb{C}^q) L 2 ( M , C q ) . If B ∈ Op S q ℓ ( M ) B\in\operatorname{Op}S_q^\ell(M) B ∈ Op S q ℓ ( M ) is another pseudodifferential operator, then by [48 Chapter 7, Proposition 3.3 and Section 10] one has A B ∈ Op S q k + ℓ ( M ) AB\in\operatorname{Op}S_q^{k+\ell}(M) A B ∈ Op S q k + ℓ ( M ) and its principal symbol is σ A B ( x , ξ ) = σ A ( x , ξ ) σ B ( x , ξ ) \sigma_{AB}(x,\xi)=\sigma_A(x,\xi)\sigma_B(x,\xi) σ A B ( x , ξ ) = σ A ( x , ξ ) σ B ( x , ξ ) . If C ∈ Op S q k ( M ) C\in\operatorname{Op}S_q^k(M) C ∈ Op S q k ( M ) , i.e. if C C C is a scalar operator, then ( C ⊗ I q ) B − B ( C ⊗ I q ) ∈ Op S q k + ℓ − 1 ( M ) (C\otimes I_q)B-B(C\otimes I_q)\in\operatorname{Op}S_q^{k+\ell-1}(M) ( C ⊗ I q ) B − B ( C ⊗ I q ) ∈ Op S q k + ℓ − 1 ( M ) , see [48 Chapter 7, equation (3.24) and Section 10] .
Important examples of classical pseudodifferential operators are multiplication operators associated with smooth functions, which belong to Op S q 0 ( M ) \operatorname{Op}S_q^0(M) Op S q 0 ( M ) , and integral operators with smooth integral kernels, which are elements in Op S q − ∞ ( M ) : = ⋂ ℓ ∈ R Op S q ℓ ( M ) \operatorname{Op}S_q^{-\infty}(M):=\bigcap_{\ell\in\mathbb{R}}\operatorname{Op}S_q^\ell(M) Op S q − ∞ ( M ) := ⋂ ℓ ∈ R Op S q ℓ ( M ) .
Finally, we present the proof of Lemma 2.9, which will lead to the fact that certain integral operators can be viewed as pseudodifferential operators.
Proof of Lemma 2.9. Using the expression for E z E_z E z from (14) and (15) , one sees that
E z ( x ) = ( ( z + m ) E z 2 − m 2 ( x ) ⊗ I N 2 w z + ( x ) w z − ( x ) ( z − m ) E z 2 − m 2 ( x ) ⊗ I N 2 ) , x ∈ R n ∖ { 0 } , E_z(x)=\begin{pmatrix}
(z+m)E_{z^2-m^2}(x)\otimes I_{\frac{N}{2}} & w_z^+(x)\\
w_z^-(x) & (z-m)E_{z^2-m^2}(x)\otimes I_{\frac{N}{2}}
\end{pmatrix},\qquad x\in\mathbb{R}^n\setminus\{0\}, E z ( x ) = ( ( z + m ) E z 2 − m 2 ( x ) ⊗ I 2 N w z − ( x ) w z + ( x ) ( z − m ) E z 2 − m 2 ( x ) ⊗ I 2 N ) , x ∈ R n ∖ { 0 } ,
where E z 2 − m 2 E_{z^2-m^2} E z 2 − m 2 is given by (13) and in the case n = 2 n=2 n = 2 we have set
w z ± ( x ) = { 1 2 π k ( z ) ∣ x ∣ K 1 ( − i k ( z ) ∣ x ∣ ) 1 x 1 ± i x 2 , z 2 ≠ m 2 , i 2 π 1 x 1 ± i x 2 , z 2 = m 2 , w_z^\pm(x)=
\begin{cases}
\frac{1}{2\pi}k(z)|x|K_1(-\mathrm{i}k(z)|x|)\frac{1}{x_1\pm\mathrm{i}x_2}, & z^2\ne m^2,\\[6pt]
\frac{\mathrm{i}}{2\pi}\frac{1}{x_1\pm\mathrm{i}x_2}, & z^2=m^2,
\end{cases} w z ± ( x ) = ⎩ ⎨ ⎧ 2 π 1 k ( z ) ∣ x ∣ K 1 ( − i k ( z ) ∣ x ∣ ) x 1 ± i x 2 1 , 2 π i x 1 ± i x 2 1 , z 2 = m 2 , z 2 = m 2 ,
whereas in the case n = 3 n=3 n = 3 we have set
w z ± ( x ) = ( 1 − i k ( z ) ∣ x ∣ ) i ( σ ⋅ x ) 4 π ∣ x ∣ 3 e i k ( z ) ∣ x ∣ . w_z^\pm(x)=(1-\mathrm{i}k(z)|x|)\frac{\mathrm{i}(\sigma\cdot x)}{4\pi|x|^3}e^{\mathrm{i}k(z)|x|}. w z ± ( x ) = ( 1 − i k ( z ) ∣ x ∣ ) 4 π ∣ x ∣ 3 i ( σ ⋅ x ) e i k ( z ) ∣ x ∣ .
In the remaining part of the proof we distinguish between the two cases n = 2 n=2 n = 2 and n = 3 n=3 n = 3 . Let us first consider n = 2 n=2 n = 2 . Using the representations
K 1 ( t ) = 1 t + t g 1 ( t 2 ) log t + t g 2 ( t 2 ) , K 0 ( t ) = − log t + g 3 ( t 2 ) + t 2 g 4 ( t 2 ) log t (103) K_1(t)=\frac{1}{t}+tg_1(t^2)\log t+tg_2(t^2),\qquad K_0(t)=-\log t+g_3(t^2)+t^2g_4(t^2)\log t \tag*{(103)} K 1 ( t ) = t 1 + t g 1 ( t 2 ) log t + t g 2 ( t 2 ) , K 0 ( t ) = − log t + g 3 ( t 2 ) + t 2 g 4 ( t 2 ) log t (103)
with some analytic functions g 1 , g 2 , g 3 , g 4 g_1,g_2,g_3,g_4 g 1 , g 2 , g 3 , g 4 , see the proof of [13 Lemma 3.3] , one has
E z ( x ) = ( − z + m 2 π log ∣ x ∣ w z + ( x ) w z − ( x ) − z − m 2 π log ∣ x ∣ ) + F z ( x ) , x ∈ R n ∖ { 0 } , (104) E_z(x)=
\begin{pmatrix}
-\dfrac{z+m}{2\pi}\log|x| & w_z^+(x)\\
w_z^-(x) & -\dfrac{z-m}{2\pi}\log|x|
\end{pmatrix}
+F_z(x),\qquad x\in\mathbb{R}^n\setminus\{0\}, \tag*{(104)} E z ( x ) = − 2 π z + m log ∣ x ∣ w z − ( x ) w z + ( x ) − 2 π z − m log ∣ x ∣ + F z ( x ) , x ∈ R n ∖ { 0 } , (104)
where F z F_z F z is a matrix-valued function with components
F z [ 1 , 1 ] ( x ) = z + m 2 π [ − log ( − i k ( z ) ) + g 3 ( − k ( z ) 2 ∣ x ∣ 2 ) − k ( z ) 2 ∣ x ∣ 2 g 4 ( − k ( z ) 2 ∣ x ∣ 2 ) log ( − i k ( z ) ∣ x ∣ ) ] , F_z[1,1](x)=\frac{z+m}{2\pi}\left[-\log(-\mathrm{i}k(z))+g_3(-k(z)^2|x|^2)-k(z)^2|x|^2g_4(-k(z)^2|x|^2)\log(-\mathrm{i}k(z)|x|)\right], F z [ 1 , 1 ] ( x ) = 2 π z + m [ − log ( − i k ( z )) + g 3 ( − k ( z ) 2 ∣ x ∣ 2 ) − k ( z ) 2 ∣ x ∣ 2 g 4 ( − k ( z ) 2 ∣ x ∣ 2 ) log ( − i k ( z ) ∣ x ∣ ) ] ,
F z [ 1 , 2 ] ( x ) = − i k ( z ) 2 2 π ( x 1 − i x 2 ) [ g 1 ( − k ( z ) 2 ∣ x ∣ 2 ) log ( − i k ( z ) ∣ x ∣ ) + g 2 ( − k ( z ) 2 ∣ x ∣ 2 ) ] , F_z[1,2](x)=-\frac{\mathrm{i}k(z)^2}{2\pi}(x_1-\mathrm{i}x_2)\left[g_1(-k(z)^2|x|^2)\log(-\mathrm{i}k(z)|x|)+g_2(-k(z)^2|x|^2)\right], F z [ 1 , 2 ] ( x ) = − 2 π i k ( z ) 2 ( x 1 − i x 2 ) [ g 1 ( − k ( z ) 2 ∣ x ∣ 2 ) log ( − i k ( z ) ∣ x ∣ ) + g 2 ( − k ( z ) 2 ∣ x ∣ 2 ) ] ,
F z [ 2 , 1 ] ( x ) = − i k ( z ) 2 2 π ( x 1 + i x 2 ) [ g 1 ( − k ( z ) 2 ∣ x ∣ 2 ) log ( − i k ( z ) ∣ x ∣ ) + g 2 ( − k ( z ) 2 ∣ x ∣ 2 ) ] , F_z[2,1](x)=-\frac{\mathrm{i}k(z)^2}{2\pi}(x_1+\mathrm{i}x_2)\left[g_1(-k(z)^2|x|^2)\log(-\mathrm{i}k(z)|x|)+g_2(-k(z)^2|x|^2)\right], F z [ 2 , 1 ] ( x ) = − 2 π i k ( z ) 2 ( x 1 + i x 2 ) [ g 1 ( − k ( z ) 2 ∣ x ∣ 2 ) log ( − i k ( z ) ∣ x ∣ ) + g 2 ( − k ( z ) 2 ∣ x ∣ 2 ) ] ,
F z [ 2 , 2 ] ( x ) = z − m 2 π [ − log ( − i k ( z ) ) + g 3 ( − k ( z ) 2 ∣ x ∣ 2 ) − k ( z ) 2 ∣ x ∣ 2 g 4 ( − k ( z ) 2 ∣ x ∣ 2 ) log ( − i k ( z ) ∣ x ∣ ) ] , F_z[2,2](x)=\frac{z-m}{2\pi}\left[-\log(-\mathrm{i}k(z))+g_3(-k(z)^2|x|^2)-k(z)^2|x|^2g_4(-k(z)^2|x|^2)\log(-\mathrm{i}k(z)|x|)\right], F z [ 2 , 2 ] ( x ) = 2 π z − m [ − log ( − i k ( z )) + g 3 ( − k ( z ) 2 ∣ x ∣ 2 ) − k ( z ) 2 ∣ x ∣ 2 g 4 ( − k ( z ) 2 ∣ x ∣ 2 ) log ( − i k ( z ) ∣ x ∣ ) ] ,
if z 2 ≠ m 2 z^2\ne m^2 z 2 = m 2 , and F z = 0 F_z=0 F z = 0 , if z 2 = m 2 z^2=m^2 z 2 = m 2 . Since g 1 , g 2 , g 3 , g 4 g_1,g_2,g_3,g_4 g 1 , g 2 , g 3 , g 4 are analytic, F z F_z F z is smooth outside the origin and it has no singularity at 0 0 0 (note that ∣ x ∣ 2 log ∣ x ∣ → 0 |x|^2\log|x|\to0 ∣ x ∣ 2 log ∣ x ∣ → 0 and ( x 1 ± i x 2 ) log ∣ x ∣ → 0 (x_1\pm\mathrm{i}x_2)\log|x|\to0 ( x 1 ± i x 2 ) log ∣ x ∣ → 0 as x → 0 x\to0 x → 0 ). In particular, K z = χ F z ∈ L 1 ( R n ) ∩ C ( R n ) K_z=\chi F_z\in L^1(\mathbb{R}^n)\cap C(\mathbb{R}^n) K z = χ F z ∈ L 1 ( R n ) ∩ C ( R n ) , which shows (a) and the first claim in (d). Moreover, as ∇ log ∣ x ∣ = x / ∣ x ∣ 2 \nabla\log|x|=x/|x|^2 ∇ log ∣ x ∣ = x /∣ x ∣ 2 , the explicit form of F z F_z F z shows
∣ ∇ F z ( x ) ∣ ≤ c ( 1 + ∣ log ∣ x ∣ ∣ ) |\nabla F_z(x)|\le c(1+|\log|x||) ∣∇ F z ( x ) ∣ ≤ c ( 1 + ∣ log ∣ x ∣∣ )
for 0 < ∣ x ∣ ≤ R 0<|x|\le R 0 < ∣ x ∣ ≤ R , and the same bound holds for ∇ K z = ( ∇ χ ) F z + χ ∇ F z \nabla K_z=(\nabla\chi)F_z+\chi\nabla F_z ∇ K z = ( ∇ χ ) F z + χ ∇ F z ; this finishes the proof of (d). Moreover, as χ \chi χ is compactly supported, K z K_z K z is rapidly decreasing (with all of its derivatives) at infinity, i.e. (b) is true. It remains to show that K z K_z K z satisfies (32) ; here and below, smooth contributions to K z K_z K z can always be absorbed into the C L C^L C L -remainder in the expansion. We show that the term χ F z [ 1 , 2 ] \chi F_z[1,2] χ F z [ 1 , 2 ] has such an expansion, the other terms can be treated in a similar way. Note first that
f ( x ) : = − i k ( z ) 2 2 π ( x 1 − i x 2 ) [ g 1 ( − k ( z ) 2 ∣ x ∣ 2 ) log ( − i k ( z ) ) + g 2 ( − k ( z ) 2 ∣ x ∣ 2 ) ] f(x):=-\frac{\mathrm{i}k(z)^2}{2\pi}(x_1-\mathrm{i}x_2)\left[g_1\left(-k(z)^2|x|^2\right)\log\left(-\mathrm{i}k(z)\right)+g_2\left(-k(z)^2|x|^2\right)\right] f ( x ) := − 2 π i k ( z ) 2 ( x 1 − i x 2 ) [ g 1 ( − k ( z ) 2 ∣ x ∣ 2 ) log ( − i k ( z ) ) + g 2 ( − k ( z ) 2 ∣ x ∣ 2 ) ]
is smooth. Hence, using the power series representation g 1 ( t ) = ∑ j ≥ 0 a j t j g_1(t)=\sum_{j\geq0}a_jt^j g 1 ( t ) = ∑ j ≥ 0 a j t j of the analytic function g 1 g_1 g 1 , one finds that
F z [ 1 , 2 ] ( x ) = − i k ( z ) 2 2 π ∑ j ≥ 0 a j ( − k ( z ) 2 ) j ( x 1 − i x 2 ) ∣ x ∣ 2 j log ∣ x ∣ + f ( x ) = ∑ j ≥ 0 p ~ 2 j + 1 ( x ) log ∣ x ∣ + f ( x ) , (105) \begin{aligned}
F_z[1,2](x)&=-\frac{\mathrm{i}k(z)^2}{2\pi}\sum_{j\geq0}a_j\left(-k(z)^2\right)^j(x_1-\mathrm{i}x_2)|x|^{2j}\log|x|+f(x)\\
&=\sum_{j\geq0}\widetilde{p}_{2j+1}(x)\log|x|+f(x), \tag*{(105)}
\end{aligned} F z [ 1 , 2 ] ( x ) = − 2 π i k ( z ) 2 j ≥ 0 ∑ a j ( − k ( z ) 2 ) j ( x 1 − i x 2 ) ∣ x ∣ 2 j log ∣ x ∣ + f ( x ) = j ≥ 0 ∑ p 2 j + 1 ( x ) log ∣ x ∣ + f ( x ) , (105)
where each
p ~ 2 j + 1 ( x ) : = i 2 π a j ( − k ( z ) 2 ) j + 1 ( x 1 − i x 2 ) ∣ x ∣ 2 j , j ∈ N 0 , \widetilde{p}_{2j+1}(x):=\frac{\mathrm{i}}{2\pi}a_j\left(-k(z)^2\right)^{j+1}(x_1-\mathrm{i}x_2)|x|^{2j},\qquad j\in\mathbb{N}_0, p 2 j + 1 ( x ) := 2 π i a j ( − k ( z ) 2 ) j + 1 ( x 1 − i x 2 ) ∣ x ∣ 2 j , j ∈ N 0 ,
is a homogeneous polynomial of order 2 j + 1 2j+1 2 j + 1 . To conclude the asymptotic expansion in (2.28), fix L ∈ N L\in\mathbb{N} L ∈ N and note that one has the Taylor series expansion
χ ( x ) = ∑ ∣ γ ∣ ≤ L c γ x γ + ∑ ∣ γ ∣ = L + 1 x γ χ ~ γ ( x ) , \chi(x)=\sum_{|\gamma|\leq L}c_\gamma x^\gamma+\sum_{|\gamma|=L+1}x^\gamma\widetilde{\chi}_\gamma(x), χ ( x ) = ∣ γ ∣ ≤ L ∑ c γ x γ + ∣ γ ∣ = L + 1 ∑ x γ χ γ ( x ) ,
where the sums are over multi-indices γ ∈ N 0 2 \gamma\in\mathbb{N}_0^2 γ ∈ N 0 2 , c γ c_\gamma c γ are suitable constants, and χ ~ γ \widetilde{\chi}_\gamma χ γ is smooth. Hence, we can write
χ ( x ) F z [ 1 , 2 ] ( x ) = ( ∑ ∣ γ ∣ ≤ L c γ x γ + ∑ ∣ γ ∣ = L + 1 x γ χ ~ γ ( x ) ) ⋅ ( ∑ j ≥ 0 p ~ 2 j + 1 ( x ) log ∣ x ∣ ) + χ ( x ) f ( x ) = ∑ ℓ = 0 L p ℓ ( x ) log ∣ x ∣ + ∑ ∣ γ ∣ ≤ L ∣ γ ∣ + 2 j + 1 > L + 1 c γ x γ p ~ 2 j + 1 ( x ) log ∣ x ∣ + ∑ ∣ γ ∣ = L + 1 x γ χ ~ γ ( x ) log ∣ x ∣ ∑ j ≥ 0 p ~ 2 j + 1 ( x ) + χ ( x ) f ( x ) . \begin{aligned}
\chi(x)F_z[1,2](x)
={}&\left(\sum_{|\gamma|\leq L}c_\gamma x^\gamma+\sum_{|\gamma|=L+1}x^\gamma\widetilde{\chi}_\gamma(x)\right)\cdot\left(\sum_{j\geq0}\widetilde{p}_{2j+1}(x)\log|x|\right)+\chi(x)f(x)\\
={}&\sum_{\ell=0}^{L}p_\ell(x)\log|x|+\sum_{\substack{|\gamma|\leq L\\|\gamma|+2j+1>L+1}}c_\gamma x^\gamma\widetilde{p}_{2j+1}(x)\log|x|\\
&+\sum_{|\gamma|=L+1}x^\gamma\widetilde{\chi}_\gamma(x)\log|x|\sum_{j\geq0}\widetilde{p}_{2j+1}(x)+\chi(x)f(x).
\end{aligned} χ ( x ) F z [ 1 , 2 ] ( x ) = = ∣ γ ∣ ≤ L ∑ c γ x γ + ∣ γ ∣ = L + 1 ∑ x γ χ γ ( x ) ⋅ ( j ≥ 0 ∑ p 2 j + 1 ( x ) log ∣ x ∣ ) + χ ( x ) f ( x ) ℓ = 0 ∑ L p ℓ ( x ) log ∣ x ∣ + ∣ γ ∣ ≤ L ∣ γ ∣ + 2 j + 1 > L + 1 ∑ c γ x γ p 2 j + 1 ( x ) log ∣ x ∣ + ∣ γ ∣ = L + 1 ∑ x γ χ γ ( x ) log ∣ x ∣ j ≥ 0 ∑ p 2 j + 1 ( x ) + χ ( x ) f ( x ) .
with
p ℓ ( x ) = ∑ ∣ γ ∣ + 2 j + 1 = ℓ + 1 c γ x γ p ~ 2 j + 1 ( x ) , ℓ ∈ { 0 , … , L } , p_\ell(x)=\sum_{|\gamma|+2j+1=\ell+1}c_\gamma x^\gamma\widetilde{p}_{2j+1}(x),\qquad\ell\in\{0,\ldots,L\}, p ℓ ( x ) = ∣ γ ∣ + 2 j + 1 = ℓ + 1 ∑ c γ x γ p 2 j + 1 ( x ) , ℓ ∈ { 0 , … , L } ,
being a homogeneous polynomial of order ℓ + 1 \ell+1 ℓ + 1 . Since x γ log ∣ x ∣ ∈ C L ( R 2 ) x^\gamma\log|x|\in C^L(\mathbb{R}^2) x γ log ∣ x ∣ ∈ C L ( R 2 ) for ∣ γ ∣ > L |\gamma|>L ∣ γ ∣ > L , one concludes that
χ ( x ) F z [ 1 , 2 ] ( x ) − ∑ ℓ = 0 L p ℓ ( x ) log ∣ x ∣ ∈ C L ( R 2 ) , \chi(x)F_z[1,2](x)-\sum_{\ell=0}^{L}p_\ell(x)\log|x|\in C^L(\mathbb{R}^2), χ ( x ) F z [ 1 , 2 ] ( x ) − ℓ = 0 ∑ L p ℓ ( x ) log ∣ x ∣ ∈ C L ( R 2 ) ,
which yields (2.28) with q ℓ ≡ 0 q_\ell\equiv0 q ℓ ≡ 0 for all ℓ \ell ℓ ; note that the smooth function χ f \chi f χ f contributes only to the C L C^L C L -remainders. Thus, also (c) is true for n = 2 n=2 n = 2 .
In a similar way, one has for n = 3 n=3 n = 3
E z ( x ) = ( z + m 4 π ∣ x ∣ ⊗ I 2 w + ( x ) w − ( x ) z − m 4 π ∣ x ∣ ⊗ I 2 ) + F z ( x ) , x ∈ R 3 ∖ { 0 } . (106) E_z(x)=
\begin{pmatrix}
\dfrac{z+m}{4\pi|x|}\otimes I_2 & w^+(x)\\
w^-(x) & \dfrac{z-m}{4\pi|x|}\otimes I_2
\end{pmatrix}
+F_z(x),\qquad x\in\mathbb{R}^3\setminus\{0\}. \tag*{(106)} E z ( x ) = 4 π ∣ x ∣ z + m ⊗ I 2 w − ( x ) w + ( x ) 4 π ∣ x ∣ z − m ⊗ I 2 + F z ( x ) , x ∈ R 3 ∖ { 0 } . (106)
where F z F_z F z is a block matrix function
F z ( x ) = ( F z [ 1 , 1 ] ( x ) F z [ 1 , 2 ] ( x ) F z [ 2 , 1 ] ( x ) F z [ 2 , 2 ] ( x ) ) F_z(x)=\begin{pmatrix} F_z[1,1](x) & F_z[1,2](x) \\ F_z[2,1](x) & F_z[2,2](x) \end{pmatrix} F z ( x ) = ( F z [ 1 , 1 ] ( x ) F z [ 2 , 1 ] ( x ) F z [ 1 , 2 ] ( x ) F z [ 2 , 2 ] ( x ) )
with blocks F z [ j , k ] ∈ C 2 × 2 F_z[j,k]\in\mathbb{C}^{2\times2} F z [ j , k ] ∈ C 2 × 2 given by
F z [ 1 , 1 ] ( x ) = z + m 4 π ∣ x ∣ ( e i k ( z ) ∣ x ∣ − 1 ) ⊗ I 2 , F_z[1,1](x)=\frac{z+m}{4\pi|x|}\left(e^{\mathrm{i}k(z)|x|}-1\right)\otimes I_2, F z [ 1 , 1 ] ( x ) = 4 π ∣ x ∣ z + m ( e i k ( z ) ∣ x ∣ − 1 ) ⊗ I 2 ,
F z [ 1 , 2 ] ( x ) = F z [ 2 , 1 ] ( x ) = i ( σ ⋅ x ) 4 π ∣ x ∣ 3 ( ( 1 − i k ( z ) ∣ x ∣ ) e i k ( z ) ∣ x ∣ − 1 ) , F_z[1,2](x)=F_z[2,1](x)=\frac{\mathrm{i}(\boldsymbol{\sigma}\cdot x)}{4\pi|x|^3}\left((1-\mathrm{i}k(z)|x|)e^{\mathrm{i}k(z)|x|}-1\right), F z [ 1 , 2 ] ( x ) = F z [ 2 , 1 ] ( x ) = 4 π ∣ x ∣ 3 i ( σ ⋅ x ) ( ( 1 − i k ( z ) ∣ x ∣ ) e i k ( z ) ∣ x ∣ − 1 ) ,
F z [ 2 , 2 ] ( x ) = z − m 4 π ∣ x ∣ ( e i k ( z ) ∣ x ∣ − 1 ) ⊗ I 2 . F_z[2,2](x)=\frac{z-m}{4\pi|x|}\left(e^{\mathrm{i}k(z)|x|}-1\right)\otimes I_2. F z [ 2 , 2 ] ( x ) = 4 π ∣ x ∣ z − m ( e i k ( z ) ∣ x ∣ − 1 ) ⊗ I 2 .
Taking the power series representation of the exponential function into account, one finds first that
F z [ 1 , 1 ] ( x ) = z + m 4 π ∣ x ∣ ( e i k ( z ) ∣ x ∣ − 1 ) ⊗ I 2 = z + m 4 π ( ∑ ℓ = 0 ∞ ( i k ( z ) ) ℓ + 1 ( ℓ + 1 ) ! ∣ x ∣ ℓ ) ⊗ I 2 . F_z[1,1](x)=\frac{z+m}{4\pi|x|}\left(e^{\mathrm{i}k(z)|x|}-1\right)\otimes I_2=\frac{z+m}{4\pi}\left(\sum_{\ell=0}^{\infty}\frac{(\mathrm{i}k(z))^{\ell+1}}{(\ell+1)!}|x|^\ell\right)\otimes I_2. F z [ 1 , 1 ] ( x ) = 4 π ∣ x ∣ z + m ( e i k ( z ) ∣ x ∣ − 1 ) ⊗ I 2 = 4 π z + m ( ℓ = 0 ∑ ∞ ( ℓ + 1 )! ( i k ( z ) ) ℓ + 1 ∣ x ∣ ℓ ) ⊗ I 2 .
Note that each term in the above sum is a homogeneous function (and hence, a homogeneous distribution) of degree ℓ \ell ℓ that is smooth outside the origin. A similar expansion holds for F z [ 2 , 2 ] F_z[2,2] F z [ 2 , 2 ] . Using again the power series representation of the exponential function, one obtains
F z [ 1 , 2 ] ( x ) = ∑ ℓ = 0 ∞ i ℓ + 3 ( 1 ( ℓ + 2 ) ! − 1 ( ℓ + 1 ) ! ) k ( z ) ℓ + 2 ∣ x ∣ ℓ − 1 ( σ ⋅ x ) 4 π . F_z[1,2](x)=\sum_{\ell=0}^{\infty}\mathrm{i}^{\ell+3}\left(\frac{1}{(\ell+2)!}-\frac{1}{(\ell+1)!}\right)k(z)^{\ell+2}|x|^{\ell-1}\frac{(\boldsymbol{\sigma}\cdot x)}{4\pi}. F z [ 1 , 2 ] ( x ) = ℓ = 0 ∑ ∞ i ℓ + 3 ( ( ℓ + 2 )! 1 − ( ℓ + 1 )! 1 ) k ( z ) ℓ + 2 ∣ x ∣ ℓ − 1 4 π ( σ ⋅ x ) .
Again, each term in the above sum is a homogeneous function (and hence, a homogeneous distribution) of degree ℓ \ell ℓ that is smooth outside the origin. In particular, the term with ℓ = 0 \ell=0 ℓ = 0 in the last sum equals i k ( z ) 2 8 π ( σ ⋅ x ) ∣ x ∣ \frac{\mathrm{i}k(z)^2}{8\pi}\frac{(\boldsymbol{\sigma}\cdot x)}{|x|} 8 π i k ( z ) 2 ∣ x ∣ ( σ ⋅ x ) , which is bounded, homogeneous of degree 0 0 0 , discontinuous at the origin if k ( z ) ≠ 0 k(z)\ne0 k ( z ) = 0 , and whose gradient is bounded by c ∣ x ∣ − 1 c|x|^{-1} c ∣ x ∣ − 1 . All other terms, i.e., ∣ x ∣ ℓ |x|^\ell ∣ x ∣ ℓ for ℓ ≥ 0 \ell\ge0 ℓ ≥ 0 in the diagonal blocks and ∣ x ∣ ℓ − 1 ( σ ⋅ x ) |x|^{\ell-1}(\boldsymbol{\sigma}\cdot x) ∣ x ∣ ℓ − 1 ( σ ⋅ x ) for ℓ ≥ 1 \ell\ge1 ℓ ≥ 1 in the off-diagonal blocks, are continuous on R 3 \mathbb{R}^3 R 3 with gradients bounded by c ℓ R ℓ − 1 c_\ell R^{\ell-1} c ℓ R ℓ − 1 on B R ( 0 ) ∖ { 0 } B_R(0)\setminus\{0\} B R ( 0 ) ∖ { 0 } , and their coefficients decay factorially; hence these series and the series of their gradients converge uniformly on B R ( 0 ) ∖ { 0 } B_R(0)\setminus\{0\} B R ( 0 ) ∖ { 0 } . Consequently, F z F_z F z is bounded near 0 0 0 , K z = χ F z ∈ L 1 ( R n ) ∩ L ∞ ( R n ) K_z=\chi F_z\in L^1(\mathbb{R}^n)\cap L^\infty(\mathbb{R}^n) K z = χ F z ∈ L 1 ( R n ) ∩ L ∞ ( R n ) , and ∣ ∇ K z ( x ) ∣ ≤ c ∣ x ∣ − 1 |\nabla K_z(x)|\le c|x|^{-1} ∣∇ K z ( x ) ∣ ≤ c ∣ x ∣ − 1 for 0 < ∣ x ∣ ≤ R 0<|x|\le R 0 < ∣ x ∣ ≤ R . This function is smooth outside the origin and rapidly decreasing (with all of its derivatives) at infinity. This shows (a), (b), and (e). Moreover, using the Taylor series expansion of χ \chi χ and the last two displayed formulas, one finds via a Cauchy product that K z K_z K z satisfies (2.28) (with p ℓ ≡ 0 p_\ell\equiv0 p ℓ ≡ 0 for all ℓ \ell ℓ ), i.e. (c) is true. □
A brief introduction to boundary triples and Weyl functions
Let us recall some basic facts about boundary triples in a minimalistic form strictly adapted to the needs of the present paper. We follow [23 Section 1] , where the proofs of all statements can be found. We also refer the interested reader to the original papers [25 , 26 ] and the monographs [9 , 27 ] for a more detailed discussion.
General overview
Let S S S be a closed densely defined symmetric operator in a Hilbert space H \mathcal{H} H .
Definition B.1. A boundary triple for S ∗ S^{*} S ∗ is a triple ( G , Π 0 , Π 1 ) (\mathcal{G},\Pi_{0},\Pi_{1}) ( G , Π 0 , Π 1 ) consisting of a Hilbert space G \mathcal{G} G and two linear maps Π 0 , Π 1 : dom S ∗ → G \Pi_{0},\Pi_{1}:\operatorname{dom}S^{*}\to\mathcal{G} Π 0 , Π 1 : dom S ∗ → G satisfying the following two conditions:
(i) For all f , g ∈ dom S ∗ f,g\in\operatorname{dom}S^{*} f , g ∈ dom S ∗ it holds that
⟨ S ∗ f , g ⟩ H − ⟨ f , S ∗ g ⟩ H = ⟨ Π 1 f , Π 0 g ⟩ G − ⟨ Π 0 f , Π 1 g ⟩ G . \langle S^{*}f,g\rangle_{\mathcal{H}}-\langle f,S^{*}g\rangle_{\mathcal{H}}=\langle\Pi_{1}f,\Pi_{0}g\rangle_{\mathcal{G}}-\langle\Pi_{0}f,\Pi_{1}g\rangle_{\mathcal{G}}. ⟨ S ∗ f , g ⟩ H − ⟨ f , S ∗ g ⟩ H = ⟨ Π 1 f , Π 0 g ⟩ G − ⟨ Π 0 f , Π 1 g ⟩ G . (ii) The map dom S ∗ ∋ f ↦ ( Π 0 f , Π 1 f ) ∈ G × G \operatorname{dom}S^{*}\ni f\mapsto(\Pi_{0}f,\Pi_{1}f)\in\mathcal{G}\times\mathcal{G} dom S ∗ ∋ f ↦ ( Π 0 f , Π 1 f ) ∈ G × G is surjective.
A boundary triple for S ∗ S^{*} S ∗ exists if and only if S S S admits self-adjoint extensions in H \mathcal{H} H . From now on we assume that this is satisfied, and we pick a boundary triple ( G , Π 0 , Π 1 ) (\mathcal{G},\Pi_{0},\Pi_{1}) ( G , Π 0 , Π 1 ) . Then a number of additional associated objects arise naturally. First, the operator
Q : = S ∗ ∣ ker Π 0 Q:=S^{*}\big|_{\ker\Pi_{0}} Q := S ∗ k e r Π 0
turns out to be self-adjoint, and for any z ∈ res Q z\in\operatorname{res}Q z ∈ res Q one has the direct sum decomposition
dom S ∗ = dom Q ∔ ker ( S ∗ − z ) = ker Π 0 ∔ ker ( S ∗ − z ) , (107) \operatorname{dom}S^{*}=\operatorname{dom}Q\dotplus\ker(S^{*}-z)=\ker\Pi_{0}\dotplus\ker(S^{*}-z), \tag*{(107)} dom S ∗ = dom Q ∔ ker ( S ∗ − z ) = ker Π 0 ∔ ker ( S ∗ − z ) , (107)
showing that the restriction Π 0 : ker ( S ∗ − z ) → G \Pi_{0}:\ker(S^{*}-z)\to\mathcal{G} Π 0 : ker ( S ∗ − z ) → G is bijective. The γ \gamma γ -field z ↦ Γ z z\mapsto\Gamma_{z} z ↦ Γ z and the Weyl function z ↦ M z z\mapsto M_{z} z ↦ M z associated with the chosen boundary triple are the holomorphic maps defined by
res Q ∋ z ↦ Γ z : = Π 0 ∣ ker ( S ∗ − z ) − 1 ∈ L ( G , H ) , \operatorname{res}Q\ni z\mapsto\Gamma_{z}:=\left.\Pi_{0}\right|_{\ker(S^{*}-z)}^{-1}\in\mathcal{L}(\mathcal{G},\mathcal{H}), res Q ∋ z ↦ Γ z := Π 0 ∣ k e r ( S ∗ − z ) − 1 ∈ L ( G , H ) ,
res Q ∋ z ↦ M z : = Π 1 Γ z ∈ L ( G ) . \operatorname{res}Q\ni z\mapsto M_{z}:=\Pi_{1}\Gamma_{z}\in\mathcal{L}(\mathcal{G}). res Q ∋ z ↦ M z := Π 1 Γ z ∈ L ( G ) .
In addition, one may consider Γ z \Gamma_{z} Γ z as a bounded linear operator from G \mathcal{G} G to dom S ∗ \operatorname{dom}S^{*} dom S ∗ endowed with the graph norm, because ran Γ z ⊂ ker ( S ∗ − z ) \operatorname{ran}\Gamma_{z}\subset\ker(S^{*}-z) ran Γ z ⊂ ker ( S ∗ − z ) and the graph norm on ker ( S ∗ − z ) \ker(S^{*}-z) ker ( S ∗ − z ) is equivalent to the norm of H \mathcal{H} H . The adjoints of Γ z \Gamma_{z} Γ z and M z M_{z} M z are given by
Γ z ∗ = Π 1 ( Q − z ‾ ) − 1 , M z ∗ = M z ‾ . \Gamma_{z}^{*}=\Pi_{1}(Q-\overline{z})^{-1},\qquad M_{z}^{*}=M_{\overline{z}}. Γ z ∗ = Π 1 ( Q − z ) − 1 , M z ∗ = M z .
The following theorem describes the main application of boundary triples.
Theorem B.2. For any linear operator Θ \Theta Θ in G \mathcal{G} G , denote by Q Θ Q_{\Theta} Q Θ the restriction of S ∗ S^{*} S ∗ on the domain
dom Q Θ = { f ∈ dom S ∗ : Π 0 f ∈ dom Θ , Π 1 f = Θ Π 0 f } . \operatorname{dom}Q_{\Theta}=\{f\in\operatorname{dom}S^{*}:\Pi_{0}f\in\operatorname{dom}\Theta,\ \Pi_{1}f=\Theta\Pi_{0}f\}. dom Q Θ = { f ∈ dom S ∗ : Π 0 f ∈ dom Θ , Π 1 f = Θ Π 0 f } . Then, the following is true:
(a) The operator Q Θ Q_{\Theta} Q Θ is self-adjoint in H \mathcal{H} H if and only if the operator Θ \Theta Θ is self-adjoint in G \mathcal{G} G .
If this assumption is fulfilled, then for any z ∈ res Q z\in\operatorname{res}Q z ∈ res Q the following assertions are true:
(b) z ∈ spec Q Θ z\in\operatorname{spec}Q_{\Theta} z ∈ spec Q Θ if and only if 0 ∈ spec ( Θ − M z ) 0\in\operatorname{spec}(\Theta-M_{z}) 0 ∈ spec ( Θ − M z ) .
(c) ker ( Q Θ − z ) = Γ z ker ( Θ − M z ) \ker(Q_{\Theta}-z)=\Gamma_{z}\ker(\Theta-M_{z}) ker ( Q Θ − z ) = Γ z ker ( Θ − M z ) . In particular, z ∈ spec p Q Θ z\in\operatorname{spec}_{p}Q_{\Theta} z ∈ spec p Q Θ if and only if 0 ∈ spec p ( Θ − M z ) 0\in\operatorname{spec}_{p}(\Theta-M_{z}) 0 ∈ spec p ( Θ − M z ) with the same multiplicities.
(d) z ∈ spec e s s Q Θ z\in\operatorname{spec}_{\mathrm{ess}}Q_{\Theta} z ∈ spec ess Q Θ if and only if 0 ∈ spec e s s ( Θ − M z ) 0\in\operatorname{spec}_{\mathrm{ess}}(\Theta-M_{z}) 0 ∈ spec ess ( Θ − M z ) .
Moreover, for all z ∈ res Q Θ ∩ res Q z\in\operatorname{res}Q_{\Theta}\cap\operatorname{res}Q z ∈ res Q Θ ∩ res Q one has the resolvent formula
( Q Θ − z ) − 1 = ( Q − z ) − 1 + Γ z ( Θ − M z ) − 1 Γ z ‾ ∗ . (Q_{\Theta}-z)^{-1}=(Q-z)^{-1}+\Gamma_{z}(\Theta-M_{z})^{-1}\Gamma_{\overline{z}}^{*}. ( Q Θ − z ) − 1 = ( Q − z ) − 1 + Γ z ( Θ − M z ) − 1 Γ z ∗ .
It seems that the following simple result is not explicitly stated anywhere in the required form, so we provide a short complete proof:
Lemma B.3. If Θ j , Θ \Theta_j,\Theta Θ j , Θ are self-adjoint operators in G \mathcal{G} G such that Θ j → Θ \Theta_j \to\Theta Θ j → Θ , j → ∞ j \to\infty j → ∞ , in the norm resolvent sense, then Q Θ j → Q Θ Q_{\Theta_j} \to Q_\Theta Q Θ j → Q Θ , j → ∞ j \to\infty j → ∞ , in the norm resolvent sense.
Proof. Let us fix some z ∈ C ∖ R z \in\mathbb{C} \setminus\mathbb{R} z ∈ C ∖ R . In view of the resolvent formula in Theorem B.2 it is sufficient to show that ( Θ j − M z ) − 1 (\Theta_j-M_z)^{-1} ( Θ j − M z ) − 1 converges in the norm to ( Θ − M z ) − 1 (\Theta-M_z)^{-1} ( Θ − M z ) − 1 as j → ∞ j \to\infty j → ∞ . Recall that Θ j − M z \Theta_j-M_z Θ j − M z and Θ − M z \Theta-M_z Θ − M z have bounded inverses by Theorem B.2 .
Note first that ( Θ − M z ) − 1 − ( Θ + i ) − 1 = ( Θ − M z ) − 1 ( M z + i ) ( Θ + i ) − 1 (\Theta-M_z)^{-1}-(\Theta+i)^{-1}=(\Theta-M_z)^{-1}(M_z+i)(\Theta+i)^{-1} ( Θ − M z ) − 1 − ( Θ + i ) − 1 = ( Θ − M z ) − 1 ( M z + i ) ( Θ + i ) − 1 , and hence
( Θ − M z ) − 1 ( I − ( M z + i ) ( Θ + i ) − 1 ) = ( Θ + i ) − 1 . (\Theta-M_z)^{-1}\left(I-(M_z+i)(\Theta+i)^{-1}\right)=(\Theta+i)^{-1}. ( Θ − M z ) − 1 ( I − ( M z + i ) ( Θ + i ) − 1 ) = ( Θ + i ) − 1 . It follows from this that the operator T : = I − ( M z + i ) ( Θ + i ) − 1 T:=I-(M_z+i)(\Theta+i)^{-1} T := I − ( M z + i ) ( Θ + i ) − 1 has an everywhere defined bounded inverse, and
( Θ − M z ) − 1 = ( Θ + i ) − 1 T − 1 . (\Theta-M_z)^{-1}=(\Theta+i)^{-1}T^{-1}. ( Θ − M z ) − 1 = ( Θ + i ) − 1 T − 1 . The same conclusion and identity hold if one replaces the operators Θ \Theta Θ and T T T by Θ j \Theta_j Θ j and T j : = I − ( M z + i ) ( Θ j + i ) − 1 T_j:=I-(M_z+i)(\Theta_j+i)^{-1} T j := I − ( M z + i ) ( Θ j + i ) − 1 , respectively. Our assumption Θ j → Θ \Theta_j \to\Theta Θ j → Θ , j → ∞ j \to\infty j → ∞ , in the norm resolvent sense implies T j → T T_j \to T T j → T , j → ∞ j \to\infty j → ∞ , in the operator norm. As both T T T and T j T_j T j have bounded inverses, one also has
T j − 1 = ( T + ( T j − T ) ) − 1 = ( I + T − 1 ( T j − T ) ) − 1 T − 1 → T − 1 , j → ∞ , T_j^{-1}=\left(T+(T_j-T)\right)^{-1}=\left(I+T^{-1}(T_j-T)\right)^{-1}T^{-1}\to T^{-1},\qquad j\to\infty, T j − 1 = ( T + ( T j − T ) ) − 1 = ( I + T − 1 ( T j − T ) ) − 1 T − 1 → T − 1 , j → ∞ , in the operator norm. Now it follows that
( Θ j − M z ) − 1 = ( Θ j + i ) − 1 T j − 1 → ( Θ + i ) − 1 T − 1 = ( Θ − M z ) − 1 , j → ∞ , (\Theta_j-M_z)^{-1}=(\Theta_j+i)^{-1}T_j^{-1}\to(\Theta+i)^{-1}T^{-1}=(\Theta-M_z)^{-1},\qquad j\to\infty, ( Θ j − M z ) − 1 = ( Θ j + i ) − 1 T j − 1 → ( Θ + i ) − 1 T − 1 = ( Θ − M z ) − 1 , j → ∞ , which yields the claim. □
For further aspects on the dependence of Q Θ Q_\Theta Q Θ on Θ \Theta Θ see, e.g. [9 Section 2.8] , [27 Section 8.5.2] , and [31 ] .
Construction of boundary triples using abstract trace maps
Finally, we recall an approach to the construction of boundary triples using abstract trace maps as initially proposed in [38 , 39 ] , see also [23 Section 1.4.2] .
Let Q Q Q be a self-adjoint operator in the Hilbert space H \mathcal{H} H . Let G \mathcal{G} G be another Hilbert space and
γ : dom Q → G \gamma:\operatorname{dom}Q\to\mathcal{G} γ : dom Q → G
be a surjective linear operator which is bounded with respect to the graph norm of Q Q Q and such that ker γ \ker\gamma ker γ is a dense subspace of H \mathcal{H} H (such γ \gamma γ are often referred to as abstract trace maps). Then
S : = Q ∣ ker γ S:=Q|_{\ker\gamma} S := Q ∣ k e r γ
is a closed densely defined symmetric operator in H \mathcal{H} H . For z ∈ res Q z\in\operatorname{res}Q z ∈ res Q define
Γ z : = ( γ ( Q − z ˉ ) − 1 ) ∗ ∈ L ( G , H ) , (108) \Gamma_z:=\left(\gamma(Q-\bar{z})^{-1}\right)^*\in\mathcal{L}(\mathcal{G},\mathcal{H}), \tag*{(108)} Γ z := ( γ ( Q − z ˉ ) − 1 ) ∗ ∈ L ( G , H ) , (108)
then
ran Γ z = ker ( S ∗ − z ) , dom S ∗ = dom Q ∔ ran Γ z , \operatorname{ran}\Gamma_z=\ker(S^*-z),\qquad\operatorname{dom}S^*=\operatorname{dom}Q\mathbin{\dotplus}\operatorname{ran}\Gamma_z, ran Γ z = ker ( S ∗ − z ) , dom S ∗ = dom Q ∔ ran Γ z ,
i.e. each f ∈ dom S ∗ f\in\operatorname{dom}S^* f ∈ dom S ∗ is uniquely represented as
f = f z + Γ z ξ , f z ∈ dom Q , ξ ∈ G , (109) f=f_z+\Gamma_z\xi,\qquad f_z\in\operatorname{dom}Q,\quad\xi\in\mathcal{G}, \tag*{(109)} f = f z + Γ z ξ , f z ∈ dom Q , ξ ∈ G , (109)
and one can show that ξ \xi ξ in the above representation is independent of z z z .
Theorem B.4. Pick an arbitrary ζ ∈ res Q \zeta\in\operatorname{res} Q ζ ∈ res Q and define Π 0 , Π 1 : dom S ∗ → G \Pi_0, \Pi_1 : \operatorname{dom} S^* \to\mathcal{G} Π 0 , Π 1 : dom S ∗ → G by
Π 0 f : = ξ and Π 1 f : = 1 2 γ ( f ζ + f ζ ‾ ) . \Pi_0 f := \xi\qquad\text{and} \qquad\Pi_1 f := \frac{1}{2}\gamma(f_{\zeta}+f_{\overline{\zeta}}). Π 0 f := ξ and Π 1 f := 2 1 γ ( f ζ + f ζ ) . Then ( G , Π 0 , Π 1 ) (\mathcal{G}, \Pi_0, \Pi_1) ( G , Π 0 , Π 1 ) is a boundary triple for S ∗ S^* S ∗ such that Q = S ∗ ∣ ker Π 0 Q = S^*|_{\ker\Pi_0} Q = S ∗ ∣ k e r Π 0 . Moreover, the associated γ \gamma γ -field Γ z \Gamma_z Γ z is given by (B.2) and the associated Weyl function M z M_z M z is
M z = γ ( Γ z − 1 2 ( Γ ζ + Γ ζ ‾ ) ) . (110) M_z = \gamma\left(\Gamma_z-\frac{1}{2}(\Gamma_{\zeta}+\Gamma_{\overline{\zeta}})\right). \tag*{(110)} M z = γ ( Γ z − 2 1 ( Γ ζ + Γ ζ ) ) . (110)
Remark B.5. If the operator Q Q Q has purely discrete spectrum, then its resolvent z ↦ ( Q − z ) − 1 z \mapsto(Q-z)^{-1} z ↦ ( Q − z ) − 1 is meromorphic in C \mathbb{C} C , and the only possible singularities are poles of first order at the eigenvalues of Q Q Q with finite-dimensional residues equal to the negatives of the orthogonal projectors on the respective eigensubspaces. From the construction of the maps Γ z \Gamma_z Γ z and and M z M_z M z in (B.2) and (B.4) it follows that the maps
z ↦ Γ z ∈ L ( G , H ) , z ↦ Γ z ‾ ∗ ∈ L ( H , G ) , z ↦ M z ∈ L ( G ) z \mapsto\Gamma_z \in L(\mathcal{G},\mathcal{H}), \qquad z \mapsto\Gamma_{\overline{z}}^* \in L(\mathcal{H},\mathcal{G}), \qquad z \mapsto M_z \in L(\mathcal{G}) z ↦ Γ z ∈ L ( G , H ) , z ↦ Γ z ∗ ∈ L ( H , G ) , z ↦ M z ∈ L ( G )
are also meromorphic in C \mathbb{C} C and only have poles of first order with finite-dimensional residues.
References
(J. Behrndt) Institut für Angewandte Mathematik, Technische Universität Graz, Steyrergasse 30, 8010 Graz, Austria Email address: behrndt@tugraz.at
(B. Benhellal) National Higher School of Mathematics, Scientific and Technology Hub of Sidi Abdellah, P.O. Box 75, 16093 Algiers, Algeria Email address: badreddine.benhellal@nhsm.edu.dz
(M. Holzmann) Institut für Angewandte Mathematik, Technische Universität Graz, Steyrergasse 30, 8010 Graz, Austria Email address: holzmann@math.tugraz.at
(K. Pankrashkin) Carl von Ossietzky Universität Oldenburg, Fakultät V – Mathematik und Naturwissenschaften, Institut für Mathematik, Ammerländer Heerstr. 114–118, 26129 Oldenburg, Germany Email address: konstantin.pankrashkin@uol.de