Processing math: 100%
Research article Special Issues

An efficient spectral-Galerkin method for a new Steklov eigenvalue problem in inverse scattering

  • An efficient spectral method is proposed for a new Steklov eigenvalue problem in inverse scattering. Firstly, we establish the weak form and the associated discrete scheme by introducing an appropriate Sobolev space and a corresponding approximation space. Then, according to the Fredholm Alternative, the corresponding operator forms of weak formulation and discrete formulation are derived. After that, the error estimates of approximated eigenvalues and eigenfunctions are proved by using the spectral approximation results of completely continuous operators and the approximation properties of orthogonal projection operators. We also construct an appropriate set of basis functions in the approximation space and derive the matrix form of the discrete scheme based on the tensor product. In addition, we extend the algorithm to the circular domain. Finally, we present plenty of numerical experiments and compare them with some existing numerical methods, which validate that our algorithm is effective and high accuracy.

    Citation: Shixian Ren, Yu Zhang, Ziqiang Wang. An efficient spectral-Galerkin method for a new Steklov eigenvalue problem in inverse scattering[J]. AIMS Mathematics, 2022, 7(5): 7528-7551. doi: 10.3934/math.2022423

    Related Papers:

    [1] Liangkun Xu, Hai Bi . A multigrid discretization scheme of discontinuous Galerkin method for the Steklov-Lamé eigenproblem. AIMS Mathematics, 2023, 8(6): 14207-14231. doi: 10.3934/math.2023727
    [2] Yuanqiang Chen, Jihui Zheng, Jing An . A Legendre spectral method based on a hybrid format and its error estimation for fourth-order eigenvalue problems. AIMS Mathematics, 2024, 9(3): 7570-7588. doi: 10.3934/math.2024367
    [3] Lingling Sun, Hai Bi, Yidu Yang . A posteriori error estimates of mixed discontinuous Galerkin method for a class of Stokes eigenvalue problems. AIMS Mathematics, 2023, 8(9): 21270-21297. doi: 10.3934/math.20231084
    [4] Bo Tang, Huasheng Wang . The a posteriori error estimate in fractional differential equations using generalized Jacobi functions. AIMS Mathematics, 2023, 8(12): 29017-29041. doi: 10.3934/math.20231486
    [5] Xueyong Wang, Gang Wang, Ping Yang . Novel Pareto Z-eigenvalue inclusion intervals for tensor eigenvalue complementarity problems and its applications. AIMS Mathematics, 2024, 9(11): 30214-30229. doi: 10.3934/math.20241459
    [6] Tingting Jiang, Jiantao Jiang, Jing An . An efficient Fourier spectral method and error analysis for the fourth order problem with periodic boundary conditions and variable coefficients. AIMS Mathematics, 2023, 8(4): 9585-9601. doi: 10.3934/math.2023484
    [7] Mingyuan Cao, Yueting Yang, Chaoqian Li, Xiaowei Jiang . An accelerated conjugate gradient method for the Z-eigenvalues of symmetric tensors. AIMS Mathematics, 2023, 8(7): 15008-15023. doi: 10.3934/math.2023766
    [8] Jinhua Feng, Shixi Wang, Hai Bi, Yidu Yang . The a posteriori error estimates of the Ciarlet-Raviart mixed finite element method for the biharmonic eigenvalue problem. AIMS Mathematics, 2024, 9(2): 3332-3348. doi: 10.3934/math.2024163
    [9] Junchi Ma, Lin Chen, Xinbo Cheng . Virtual element method for the Laplacian eigenvalue problem with Neumann boundary conditions. AIMS Mathematics, 2025, 10(4): 8203-8219. doi: 10.3934/math.2025377
    [10] Panpan Liu, Hongbin Lv . An algorithm for calculating spectral radius of s-index weakly positive tensors. AIMS Mathematics, 2024, 9(1): 205-217. doi: 10.3934/math.2024012
  • An efficient spectral method is proposed for a new Steklov eigenvalue problem in inverse scattering. Firstly, we establish the weak form and the associated discrete scheme by introducing an appropriate Sobolev space and a corresponding approximation space. Then, according to the Fredholm Alternative, the corresponding operator forms of weak formulation and discrete formulation are derived. After that, the error estimates of approximated eigenvalues and eigenfunctions are proved by using the spectral approximation results of completely continuous operators and the approximation properties of orthogonal projection operators. We also construct an appropriate set of basis functions in the approximation space and derive the matrix form of the discrete scheme based on the tensor product. In addition, we extend the algorithm to the circular domain. Finally, we present plenty of numerical experiments and compare them with some existing numerical methods, which validate that our algorithm is effective and high accuracy.



    Steklov eigenvalue problems have significant physical background and wide applications in many fields of science and engineering[1,2,3,4]. Its theoretical analysis and numerical calculation have attracted the attention of many scholars, and a variety of finite element methods and spectral methods have been proposed [5,6,7,8,9,10,11,12,13], but these numerical methods are mainly based on the selfadjoint Steklov eigenvalue problem.

    Recently, a new Steklov eigenvalue problem arising from inverse scattering attracts many researchers' interest. The corresponding weak formulation of the problem is non-selfadjoint and indefinite. We can not use Lax-Milgram theorem to determine the existence and uniqueness of the solution of the corresponding resource problems, which brings some difficulties to deduce the equivalent operator form of the eigenvalue problem. In addition, in order to prove the error estimation of approximated eigenvalues and eigenfunctions, we need not only to introduce conjugate eigenvalue problem and corresponding conjugate operators, but also to prove the error estimation between these operators and their approximated operators. However, there are still some numerical methods to solve the problem. For instance, Cakoni et al. discussed the conforming finite element approximation in [14], Liu et al. studied spectral indicator method in [15], Bi et al. discussed two-grid discretizations and a local finite element scheme in [16], Zhang et al. established a multigrid correction scheme in [17], Yang et al. used non-conforming Crouzeix-Raviart element solve it [18], Xu et al. discussed an asymptotically exact a posteriori error estimator for non-selfadjoint Steklov eigenvalue problem [19], Wang et al. studied a priori and a posteriori error estimates for a virtual element method for the non-self-adjoint Steklov eigenvalue problem [20]. To our knowledge, the study of spectral method for solving this problem has not been reported. Since the spectral methods have the characteristics of spectral accuracy [21,22,23,24,25], that is to say, we only need to spend less degrees of freedom to obtain higher accurate numerical solutions. Then, it is meaningful to propose an effective spectral method for solving a new Steklov eigenvalue problem in inverse scattering.

    Hence, we shall study an effective spectral-Galerkin method for the new Steklov eigenvalue problem. Firstly, we establish the weak formulation and the associated discrete scheme by introducing an appropriate Sobolev space and a corresponding approximation space. Then, according to the Fredholm Alternative, the corresponding operator forms of weak formulation and discrete formulation are derived. After that, the error estimates of approximated eigenvalues and eigenfunctions are proved by using the spectral theory of compact operators and the approximation properties of orthogonal projection operators. We also construct an appropriate set of basis functions in the approximation space and derive the matrix form of the discrete scheme based on tensor product. In addition, we extend the algorithm to the circular domain and transform the original problem into an equivalent form under the polar coordinates. By using orthogonal polynomial approximation in the radial direction and Fourier basis function approximation in the θ direction, combining with pole conditions, we construct an appropriate approximation space and derive the corresponding matrix form of the discrete scheme. Finally, we provide plenty of numerical examples and compare them with some existing numerical methods, the numerical results confirm the effectiveness and high accuracy of our algorithm.

    We derive the weak formulation and the corresponding discrete scheme and prove the error estimations of approximation eigenvalues and eigenfunctions in next section. In §3, we propose an efficient algorithm to solve the Steklov eigenvalue problem in the square domain. In §4, we extend our algorithm to the case of circular domain. In §5, we provide several numerical examples to validate the accuracy and efficiency of our algorithm. In §6, some concluding remarks are presented.

    Throughout this article, a notation ab is used to mean that aCb, where C is a positive constant independent of any function or any discretization parameters.

    Denote by Hs(D) and Hs(D) the usual Sobolev spaces with integer order s in D and on D, respectively. In particular, H0(D)=L2(D) and H0(D)=L2(D). The norm in Hs(D) and Hs(D) are expressed as ψs,D and ψs,D, separately.

    Consider the following Steklov eigenvalue problem:

    Δψ+k2β(x)ψ=0, in D, (2.1)
    ψν+λψ=0, on D, (2.2)

    where DR2 (or R3) is a bounded polygon with Lipschitz boundary D, ν is the unit outward normal on D. Let k be the wavenumber and β(x)=β1(x)+iβ2(x)k be the index of refraction that is a bounded complex value function with β1(x)>0 and β2(x)0.

    The weak formulation of the Eqs (2.1) and (2.2) is to find (λ,ψ)C×H1(D), ψ0 such that

    A(ψ,ϕ)=λB(ψ,ϕ),ϕH1(D), (2.3)

    where

    A(ψ,ϕ)=(ψ,ϕ)(k2β(x)ψ,ϕ),(ψ,ϕ)=Dψ¯ϕdx,B(ψ,ϕ)=Dψ¯ϕds.

    Refering to [15], we know that A(,) satisfies Garding's inequality, i.e., there exist constants K< and α0>0 such that

    Re{A(ϕ,ϕ)}+Kϕ20,Dα0ϕ21,D,ϕH1(D).

    Let K be a positive constant which is large enough, and the sesquilinear form is defined as follow:

    ˜A(ψ,ϕ):=A(ψ,ϕ)+K(ψ,ϕ)=(ψ,ϕ)k2(βψ,ϕ)+K(ψ,ϕ),ψ,ϕH1(D),

    then it is easy to verify that ˜A is H1(D)-elliptic (see [15]).

    We first focus on the case of D=Id(d=2,3) with I=(1,1). Denote by Lp(t) the Legendre polynomial of degree p. Let VN=span{L0(t),L1(t),,LN(t)}, then the approximation space XN=(VN)d.

    The spectral-Galerkin approximation for the eigenvalue problem (2.3) reads: Find (λN,ψN)C×XN, ψN0 such that

    A(ψN,ϕN)=λNB(ψN,ϕN),ϕNXN. (2.4)

    We first consider the following source problem associated with (2.3): Given gH12(D), find wH1(D) such that

    A(w,ϕ)=B(g,ϕ),ϕH1(D), (2.5)

    and the approximation source problem associated with (2.4): Find wNXN such that

    A(wN,ϕN)=B(g,ϕN),ϕNXN. (2.6)

    Further, we introduce Neumann eigenvalue problem as follows:

    Δw+k2β(x)w=0inD, (2.7)
    wν=0onD. (2.8)

    When k2 is not a Neumann eigenvalue of (2.7) and (2.8), from the Fredholm Alternative (see, e.g., Section 5.3 of [26] or Lemma 1 in [18]), we know that for gH12(D), there exists a unique solution wH1(D) of (2.5) such that

    w1,DCg12,D. (2.9)

    Define an operator A:H12(D)H1(D) by

    A(Ag,ϕ)=B(g,ϕ),ϕH1(D), (2.10)

    and a Neumann-to-Dirichlet mapping Γ:H12(D)H12(D) by

    Γg=Ag|D.

    We can similarly define a discrete operator AN:H12(D)XN such that

    A(ANg,ϕN)=B(g,ϕN),ϕNXN, (2.11)

    and a discrete Neumann-to-Dirichlet operator ΓN:H12(D)XBN satisfies

    ΓNg=ANg|D,

    where XBN=XN|D. Then from (2.9), we obtain Ag1,Dg12,D.

    Thus, the equivalent operator forms of (2.3) and (2.4) are:

    Aψ=1λψ,ANψN=1λNψN, (2.12)
    Γψ=1λψ,ΓNψN=1λNψN. (2.13)

    In this paper, we always assume that k2 is not a Neumann eigenvalue of (2.7) and (2.8).

    Consider the dual problem of (2.3): Find (λ,ψ)C×H1(D), ψ0 such that

    A(ϕ,ψ)=¯λB(ϕ,ψ),ϕH1(D). (2.14)

    It's clearly that the primal and dual eigenvalues satisfy the relation: λ=¯λ.

    The discrete variational formulation associated with (2.14) is given by: Find (λN,ψN)C×XN, ψN0 such that

    A(ϕN,ψN)=¯λNB(ϕN,ψN),ϕNXN. (2.15)

    Likewise, the primal and dual eigenvalues satisfy the relation: λN=¯λN.

    Similarly, define the dual operators A:H12(D)H1(D) and AN:H12(D)XN such that

    A(ϕ,Ag)=B(ϕ,g),ϕH1(D), (2.16)
    A(ϕN,ANg)=B(ϕN,g),ϕNXN. (2.17)

    The corresponding Neumann-to-Dirichlet and discrete Neumann-to-Dirichlet dual operators are defined as follows:

    Γ:H12(D)H12(D),ΓN:H12(D)XBN.

    Define the H1 projection operators Π1N:H1(D)XN and Π1N:H1(D)XN by

    A(wΠ1Nw,ϕN)=0,ϕNXN. (2.18)
    A(ϕN,wΠ1Nw)=0,ϕNXN. (2.19)

    Then for any gH12(D), we have

    A(ANgΠ1N(Ag),ϕN)=A(ANgAg+AgΠ1N(Ag),ϕN)=A(ANgAg,ϕN)+A(AgΠ1N(Ag),ϕN)=0,ϕNXN.

    Since the above equation admits a unique solution, we have AN=Π1NA. Similarly, we can obtain AN=Π1NA.

    There holds the following regular results, which will be used in the following theoretical analysis.

    Lemma 1. If gL2(D), then AgH1+κ2(D) and

    Ag1+κ2,DCg0,D, (2.20)

    if gH12(D), then AgH1+κ(D) and

    Ag1+κ,DCg12,D, (2.21)

    where κ=1 when the largest inner angle θ of D satisfying θ<π, and κ<πθ which can be arbitrarily close to πθ when θ>π.

    Proof. see [27].

    For the dual problem, we have the same regular results as the corresponding source problem. Defining an H1-projection operator P1N:H1(D)XN by

    ((wP1Nw),ϕN)+(wP1Nw,ϕN)=0,ϕNXN.

    According to Theorem 8.4 in [28], we have the following lemma:

    Lemma 2. For any wHs(D) with s1 and sl0,

    wP1Nwl,DNlsws,D. (2.22)

    Denote

    ηN=supgH12(D),g12,D=1infϕNXNAgϕN1,D,ηN=supgH12(D),g12,D=1infϕNXNAgϕN1,D.

    It is obvious that

    limNηN=limNηN=0.

    Lemma 3. Let w be the solution of (2.5), if wHs(D)(s2), there hold:

    wΠ1Nw1,DN1sws,D, (2.23)
    wΠ1Nw1,DN1sws,D, (2.24)
    wΠ1Nw12,DNsws,D. (2.25)

    Proof. According to the definition of the projection operator Π1N, we obtain

    A(wΠ1Nw,ϕN)=0,ϕNXN.

    From Theorem 3.1 in [15] and the inequality (2.22), we derive that

    wΠ1Nw1,DinfϕNXNwϕN1,DwP1Nw1,DN1sws,D.

    Similarly, we can arrive at

    wΠ1Nw1,DN1sws,D.

    From Lemma 2.2 in [16], for any ϕNXN we derive that

    wΠ1Nw12,D=supgH12(D),g0|B(wΠ1Nw,g)|g12,DsupgH12(D),g0wΠ1Nw1,DAgϕN1,Dg12,D.

    Taking ϕN=Π1NAg and using (2.21) and (2.23) we obtain that

    wΠ1Nw12,DsupgH12(D),g0wΠ1Nw1,DAgΠ1NAg1,Dg12,DN1sws,DN1Ag2,Dg12,DNsws,D.

    The proof is completed.

    Consider the following auxiliary problem: Find ξfH1(D), such that

    A(ϕ,ξf)=(ϕ,f),ϕH1(D). (2.26)

    Referring to [16], we obtain the following result.

    Lemma 4. If fL2(D), then there exists a unique solution ξfH1+κ(D) to (2.26) and

    ξf1+κ,Df0,D, (2.27)

    where the principle to determine κ see Lemma 1.

    Remark 1. If the regularity results, Lemmas 1 and 4 hold in the case of DR3, we can prove the analysis and conclusion in this paper for DR3. However, there are no such good results in R3 (see, e.g., Remark 2.1 in [29] and Remark 1 in [18]). Our analysis is still valid in the case of DR3, but the conclusions need minor modifications.

    Lemma 5. For wH1(D), we have

    wΠ1Nw0,DN1swΠ1Nw1,D. (2.28)

    Proof. According to the definition of the projection Π1N, for any ξHs(D)(s2) we have

    A(ϕ,ξΠ1Nξ)=0,ϕXN.

    From Lemma 2.4 in [16], we can obtain that

    wΠ1Nw20,D=A(wΠ1Nw,ξwΠ1NwΠ1NξwΠ1Nw)wΠ1Nw1,DξwΠ1NwΠ1NξwΠ1Nw1,D.

    From the inequalities (2.24) and (2.27), we have

    ξwΠ1NwΠ1NξwΠ1Nw1,DN1sξwΠ1Nws,DN1swΠ1Nw0,D.

    Thus, we obtain

    wΠ1Nw0,DN1swΠ1Nw1,D.

    This ends our proof.

    According to the classical spectral approximation theory: when limNΓΓNH12(D)H12(D)=0, we can obtain the error estimates of eigenvalue problem.

    Lemma 6. If s2, then limNΓΓNH12(D)H12(D)=0 and Γ is a compact operator.

    Proof. From Lemma 2.6 in [16], for any ϕNXN we have

    (ΓΓN)φ12,DsupgH12(D),g0(AAN)φ1,DAgϕN1,Dg12,D.

    Taking ϕN=Π1NAg, from (2.23) and (2.21) we derive that

    (ΓΓN)φ12,DsupgH12(D),g0(AAN)φ1,DAgΠ1NAg1,Dg12,DsupgH12(D),g0(AAN)φ1,DN1Ag2,Dg12,DN1(AAN)φ1,D.

    From the above inequality, together with (2.23) and (2.9) we obtain that

    ΓΓNH12(D)H12(D)=supφH12(D),φ0(ΓΓN)φ12,Dφ12,DsupφH12(D),φ0N1(AAN)φ1,Dφ12,DsupφH12(D),φ0N1Aφ1,Dφ12,DN1.

    Since s2, then we obtain limNΓΓNH12(D)H12(D)=0. Note that ΓN is a finite rank operator. Then, Γ is a compact operator. This ends our proof.

    Let λ be the p-th eigenvalue of (2.3) with the algebraic multiplicity h and the ascent α. Since ΓN converges to Γ, there are h eigenvalues λq,N(q=δ,δ+1,δ+2,,δ+h1) of (2.4) converging to λ. Let M(λ) be the generalized eigenfunction space of (2.3) associated with the eigenvalue λ, and MN(λ) be the generalized eigenfunction space of (2.4) associated with the eigenvalue λq,N(q=δ,δ+1,δ+2,,δ+h1). As for the dual problems (2.14) and (2.15), we can also define M(λ) and MN(λ).

    Theorem 1. Assume that ψN is the eigenfunction approximation of (2.4), then there exists an eigenfunction ψM(λ) of (2.3) corresponding λ such that

    ψψN12,DNsα, (2.29)
    ψψN1,DN22sα+Nsα+N1s, (2.30)
    ψψN0,DN22sα+Nsα, (2.31)

    and

    |λλN|N22sα. (2.32)

    Proof. From the Theorem 7.3 and Theorem 7.4 in [30] we know that there exists an eigenfunction ψ corresponding to λ and

    |λλN|{δ+h1τ,q=δ|B((ΓΓN)φτ,φq)|+(ΓΓN)|M(λ)12,D(ΓΓN)|M(λ)12,D}1α, (2.33)
    ψψN12,D(ΓΓN)|M(λ)1α12,D, (2.34)

    where φδ,,φδ+h1 are the any basis for M(λ) and φδ,,φδ+h1 are the dual basis in M(λ).

    From Theorem 2.1 in [16], combining (2.23) and (2.24) we derive that

    |B((ΓΓN)φτ,φq)|(AAN)φτ1,DAφqΠ1NAφq1,DN22sAφτsAφqsN22s. (2.35)

    We derive from Lemma 6, (2.22) and (2.23) that

    (ΓΓN)|M(λ)12,D=supfM(λ),f12,D=1(ΓΓN)f12,DNs. (2.36)

    Similarly, we have

    (ΓΓN)|M(λ)12,DNs. (2.37)

    Combining the inequalities (2.34) and (2.36), we obtain (2.29). From (2.33), (2.35), (2.36) and (2.37), we obtain (2.32).

    From Theorem 2.1 in [16], and the inequalities (2.9), (2.32), (2.29) and (2.23), we deduce

    ψNψ1,D(λλN)Aψ1,D+λNA(ψψN)1,D+λN(AAN)ψN1,D|λλN|+ψNψ12,D+(Π1NAA)(ψNψ)1,D+(Π1NAA)ψ1,DN22sα+Nsα+Nsα+N1sAψs,DN22sα+Nsα+N1s.

    From Theorem 2.1 in [16], and the inequalities (2.28), (2.9), (2.32), (2.29) and (2.23), we obtain

    ψNψ0,D(λλN)Aψ0,D+λNA(ψNψ)0,D+λN(ANA)ψN0,D(λλN)Aψ0,D+λNA(ψNψ)0,D+N1s(Π1NAA)ψN1,D|λNλ|+ψNψ12,D+N1s[(Π1NAA)(ψNψ)1,D+(Π1NAA)ψ1,D]N22sα+Nsα+N1s[Nsα+N1sAψs,D]N22sα+Nsα.

    This ends our proof.

    For the dual problems (2.14) and (2.15), we can also obtain the corresponding conclusion as Theorem 1.

    In this section, we shall present an efficient algorithm to solve the discrete scheme (2.4). We first construct a group of basis functions of the approximation space XN. Denote by Lp(t) the Legendre polynomial of degree p. Let ˆφl(t)=Ll(t)Ll+2(t),l=0,1,,N2, ˆφN1(t)=L0(t),ˆφN(t)=L1(t). Then the approximation space XN=span{ˆφl(x1)ˆφj(x2):l,j=0,1,,N}.

    Let alj=11ˆφjˆφldt, blj=11ˆφjˆφldt, clj=ˆφj(1)ˆφl(1), dlj=ˆφj(1)ˆφl(1). By utilizing the orthogonal properties of the Legendre polynomials, we obtain that

    (1)

    {alj=4l+6,blj=22l+1+22l+5,l=j;blj=bjl=22l+5,j=l+2;alj=blj=clj=dlj=0,otherwise,

    where l,j=0,1,,N2.

    (2)

    {blj=bjl=2,l+j=N1;b1N=bN1=23;alj=ajl=blj=bjl=clj=cjl=dlj=djl=0,otherwise,

    where l=N1,N,j=0,1,,N2.

    (3)

    {aNN=bN1N1=2;bNN=23;cNN=dNN=cN1N1=dN1N1=1;cN1N=cNN1=1;dN1N=dNN1=1;alj=blj=0,otherwise,

    where l=N1,N,j=N1,N.

    Next, we shall derive the matrix form based on the tensor product of the discrete scheme (2.4). For simplicity, we only consider the case of d=2. It can be derived similarly for the case of d=3.

    Let ψN(x1,x2)=Nl,j=0ψljˆφl(x1)ˆφj(x2) and

    Ψ=(ψ00ψ01ψ0Nψ10ψ11ψ1NψN0ψN1ψNN).

    We denote by ¯Ψ a column vectors with (N+1)2 elements, which is consist of the N+1 columns of Ψ. Let ϕN(x1,x2)=ˆφm(x1)ˆφn(x2),m,n=0,1,,N, then we derive that

    ΩψNϕNdx1dx2=Nl,j=01111[ˆφl(x1)ˆφj(x2)][ˆφm(x1)ˆφn(x2)]dx1dx2ψlj=Nl,j=01111ˆφl(x1)ˆφm(x1)ˆφj(x2)ˆφn(x2)+ˆφl(x1)ˆφm(x1)ˆφj(x2)ˆφn(x2)dx1dx2ψlj=Nl,j=0[11ˆφl(x1)ˆφm(x1)dx111ˆφj(x2)ˆφn(x2)dx2+11ˆφl(x1)ˆφm(x1)dx111ˆφj(x2)ˆφn(x2)dx2]ψlj=Nl,j=0(amlbnj+anjbml)ψlj=ˆA(m,:)ΨˆB(n,:)T+ˆB(m,:)ΨˆA(n,:)T=ˆB(n,:)ˆA(m,:)¯Ψ+ˆA(n,:)ˆB(m,:)¯Ψ,
    DψNϕNds=Nl,j=0{11[ˆφl(x1)ˆφj(1)ˆφm(x1)ˆφn(1)+ˆφl(x1)ˆφj(1)ˆφm(x1)ˆφn(1)]dx1+11[ˆφl(1)ˆφj(x2)ˆφm(1)ˆφn(x2)+ˆφl(1)ˆφj(x2)ˆφm(1)ˆφn(x2)]dx2}ψlj=Nl,j=0[ˆφj(1)ˆφn(1)+ˆφj(1)ˆφn(1)]11ˆφl(x1)ˆφm(x1)dx1ψlj+Nl,j=0[ˆφl(1)ˆφm(1)+ˆφl(1)ˆφm(1)]11ˆφj(x2)ˆφn(x2)dx2ψlj=Nl,j=0(cnj+dnj)bmlψlj+Nl,j=0(cml+dml)bnjψlj=ˆB(m,:)Ψ(ˆC+ˆD)(n,:)T+(ˆC+ˆD)(m,:)ΨˆB(n,:)T=(ˆC+ˆD)(n,:)ˆB(m,:)¯Ψ+ˆB(n,:)(ˆC+ˆD)(m,:)¯Ψ,

    where ˆA=(alj)Nl,j=0, ˆB=(blj)Nl,j=0, ˆC=(clj)Nl,j=0, ˆD=(dlj)Nl,j=0. ˆA(m,:) indicates the m-th row of the matrix ˆA, ˆB(m,:), ˆC(m,:), ˆD(m,:) are similar to ˆA(m,:). is a notation of tensor product of matrix, then ˆAˆB=(aljˆB)Nl,j=0.

    Let ξμ,wμ(μ=0,1,,N1) be the Labatto-Gauss points and weights, respectively. Then we have

    Dβ(x1,x2)ψNϕNdx1dx2=Nl,j=01111β(x1,x2)ˆφl(x1)ˆφj(x2)ˆφm(x1)ˆφn(x2)dx1dx2ψlj=Nl,j=0N1μ,σ=0β(ξμ,ξσ)ˆφl(ξμ)ˆφj(ξσ)ˆφm(ξμ)ˆφn(ξσ)ωμωσψlj=ˆP¯Ψ,

    where

    ˆP=(p0000p0100p0N00p000Np010Np0N0Np1000p1100p1N00p100Np110Np1N0NpN000pN100pNN00pN00NpN10NpNN0Np00N0p01N0p0NN0p00NNp01NNp0NNNp10N0p11N0p1NN0p10NNp11NNp1NNNpN0N0pN1N0pNNN0pN0NNpN1NNpNNNN),
    pmlnj=N1μ,σ=0β(ξμ,ξσ)ˆφl(ξμ)ˆφm(ξμ)ˆφj(ξσ)ˆφn(ξσ)ωμωσ.

    From the above deduce, we obtain the matrix form of the discrete scheme (2.4) as follows:

    (ˆBˆA+ˆAˆBk2ˆP)¯Ψ=λN((ˆC+ˆD)ˆB+ˆB(ˆC+ˆD))¯Ψ. (3.1)

    Note that the matrix ˆP can also be written as matrix form base on the tensor product and the stiffness matrix and mass matrix in (3.1) are all sparse when β is a constant, so we can solve (3.1) efficiently.

    We extend the above algorithm to a circular domain. Utilizing polar coordinate transformation x=(x1,x2)=(rcosθ,rsinθ), the functions ψ(x) and β(x) are represented as ˜ψ(r,θ)=ψ(rcosθ,rcosθ) and ˜β(r,θ)=β(rcosθ,rcosθ). Let Lv=1rr(rvr)+1r22vθ2, then the Eqs (2.1) and (2.2) can be rewritten as follows:

    L˜ψ(r,θ)+k2˜β(r,θ)˜ψ(r,θ)=0, in Ω=[0,R)×[0,2π), (4.1)
    ˜ψr(R,θ)=λ˜ψ(R,θ),θ[0,2π),˜ψperiodic in θ. (4.2)

    Then from (2.3) we derive that the weak form of (4.1) and (4.2) is as follows:

    a(˜ψ,˜ϕ)=λb(˜ψ,˜ϕ), (4.3)

    where

    a(˜ψ,˜ϕ)=Ωrr˜ψrˉ˜ϕdrdθ+Ω1rθ˜ψθˉ˜ϕdrdθΩk2˜β˜ψˉ˜ϕrdrdθ,b(˜ψ,˜ϕ)=2π0R˜ψ(R,θ)ˉ˜ϕ(R,θ)dθ.

    Since ˜ψ is 2π periodic in θ, by using the expansion of the Fourier basis function we have

    ˜ψ(r,θ)=|˜m|=0u˜m(r)ei˜mθ. (4.4)

    Substituting the expansion (4.4) into (4.3), and taking ˜ϕ(r,θ)=v˜n(r)ei˜nθ, we derive that

    Ωr|˜m|=0u˜mv˜nei˜mθei˜nθdrdθ+Ω|˜m|=0˜m˜nru˜mv˜nei˜mθei˜nθdrdθΩ|˜m|=0k2˜β(r,θ)u˜mv˜nei˜mθei˜nθrdrdθ=λ2π0|˜m|=0Ru˜m(R)v˜n(R)ei˜mθei˜nθdθ. (4.5)

    In order to make (4.5) well posed, we need to introduce the following polar condition (see [22])

    ˜mu˜m(r)|r=0=0. (4.6)

    Let r=R2(t+1), t(1,1), w˜m(t)=u˜m(r), z˜n(t)=v˜n(r), γ(t,θ)=˜β(r,θ), then (4.5) can be rewritten as follows:

    |˜m|=02π011(t+1)w˜mz˜nei˜mθei˜nθdtdθ+|˜m|=02π011˜m˜n(t+1)w˜mz˜nei˜mθei˜nθdtdθ|˜m|=0R24k22π011(t+1)γw˜mz˜nei˜mθei˜nθdtdθ=λ|˜m|=02π0Rw˜m(1)z˜n(1)ei˜mθei˜nθdθ. (4.7)

    Define the weighted Sobolev space:

    H1˜m(Ω)={w˜m(t)ei˜mθ:11(t+1)|w˜m|2+˜m2t+1|w˜m|2dt<,˜mw˜m(1)=0},

    which is endowed with the norm as follows:

    w˜m1,˜m,Ω=[11(t+1)|w˜m|2+˜m2t+1|w˜m|2dt]12.

    Then the weak formulation of (4.5) is to find w˜m(t)ei˜mθH1˜m(Ω), λC such that

    |˜m|=02π011(t+1)w˜mz˜nei˜mθei˜nθdtdθ+|˜m|=02π011˜m˜n(t+1)w˜mz˜nei˜mθei˜nθdtdθ|˜m|=0R24k22π011(t+1)γw˜mz˜nei˜mθei˜nθdtdθ=λ|˜m|=02π0Rw˜m(1)z˜n(1)ei˜mθei˜nθdθ, (4.8)

    for z˜n(t)ei˜nθH1˜n(Ω).

    Denote by PN the space of polynomials of degree less than or equal to N, and set XMN=M|˜m|=0H1˜m(Ω)P˜mN, where P˜mN={pNei˜mθ:pNPN}.

    Then the discrete form of (4.8) is to find w˜mNei˜mθXMN, λNC such that

    M|˜m|=02π011(t+1)w˜mNz˜nNei˜mθei˜nθdtdθ+M|˜m|=02π011˜m˜n(t+1)w˜mNz˜nNei˜mθei˜nθdtdθR24k2M|˜m|=02π011(t+1)γw˜mNz˜nNei˜mθei˜nθdtdθ=λNRM|˜m|=02π0w˜mN(1)z˜nN(1)ei˜mθei˜nθdθ, (4.9)

    for z˜nNei˜nθXMN.

    Let

    φ˜i(t)=L˜i(t)L˜i+2(t),˜i=0N2,φN1(t)=t+12,φN(t)=1t2.

    It is clear that

    H10(Ω)P0N=span{φ0˜i:φ0˜i=φ˜i,˜i=0,1,,N},H1˜m(Ω)P˜mN=span{φ˜m˜iei˜mθ:φ˜m˜i=φ˜i,˜i=0,1,,N1},(|˜m|0).

    Setting

    f˜j˜i˜n˜m=2π011(t+1)(φ˜m˜i)(φ˜n˜j)ei(˜m˜n)θdtdθ,g˜j˜i˜n˜m=2π011˜m˜nt+1φ˜m˜iφ˜n˜jei(˜m˜n)θdtdθ,q˜j˜i˜n˜m=2π011(t+1)γ(t,θ)φ˜m˜iφ˜n˜jei(˜m˜n)θdtdθ,h˜j˜i˜n˜m=2π0φ˜m˜i(1)φ˜n˜j(1)ei(˜m˜n)θdθ.

    Let w˜mN=Nsign(|˜m|)˜i=0w˜m˜iφ˜m˜i(|˜m|=0,,M), z˜nN=φ˜n˜j(|˜n|=0,,M,˜j=0,,Nsign(|˜n|)), and inserting the expressions into (4.9), we obtain that

    M|˜m|=0Nsign(|˜m|)˜i=0f˜j˜i˜n˜mw˜m˜i+M|˜m|=0Nsign(|˜m|)˜i=0g˜j˜i˜n˜mw˜m˜iR24k2M|˜m|=0Nsign(|˜m|)˜i=0q˜j˜i˜n˜mw˜m˜i=λNRM|˜m|=0Nsign(|˜m|)˜i=0h˜j˜i˜n˜mw˜m˜i. (4.10)

    Then the corresponding matrix form of (4.10) can be derived as follows:

    (F+GR24k2Q)¯W=λNRH¯W,

    where

    F=(f˜j˜i˜n˜m),G=(g˜j˜i˜n˜m),Q=(q˜j˜i˜n˜m),H=(h˜j˜i˜n˜m),W=(w˜m˜i),

    ¯W is a column vector with N(2M+1)+1 elements, which is consist of the column of W. By using the orthogonality of the Fourier system ei˜mθ, we obtain that

    {f˜j˜i˜n˜m=0,g˜j˜i˜n˜m=0,h˜j˜i˜n˜m=0,˜m˜n;f˜j˜i˜n˜m=2π11(t+1)(φ˜m˜i)(φ˜m˜j)dt,˜m=˜n;g˜j˜i˜n˜m=2π11˜m2t+1φ˜m˜iφ˜m˜jdt,˜m=˜n;h˜j˜i˜n˜m=2πφ˜m˜i(1)φ˜m˜j(1),˜m=˜n.

    It's obvious that the matrices F, G, and H are diagonal block matrices.

    Let t˜s,w˜s(˜s=0,1,,N1) be the Gauss-Lobatto points and the weights and θ˜l,ˆω˜l(˜l=0,1,,M11) be the Fourier points and the weights, respectively. Then we have

    q˜j˜i˜n˜m=N1˜s=0M11˜l=0(t˜s+1)γ(t˜s,θ˜l)φ˜i(t˜s)φ˜j(t˜s)ei(˜m˜n)θ˜lω˜sˆω˜l.

    We shall present some numerical experiments to validate the effectiveness of the algorithm. We carry out our programs in Matlab 2018a.

    Example 1. Consider the problem (2.3) with k=1 and β(x)=4 on the domain D=(22,22)2. The approximation eigenvalue λiN(i=1,2,3,4) for different N are shown in Table 1.

    Table 1.  The approximation eigenvalue λiN(i=1,2,3,4) for different N on the square.
    N λ1N λ2N λ3N λ4N
    10 2.202507126351584 0.2122521695447584 0.2122521695447588 0.9080560857539495
    15 2.202507126351587 0.2122521695447581 0.2122521695447585 0.9080560857539485
    20 2.202507126351584 0.2122521695447591 0.2122521695447593 0.9080560857539485
    25 2.202507126351590 0.2122521695447577 0.2122521695447578 0.9080560857539492

     | Show Table
    DownLoad: CSV

    We observe from Table 1 that the approximation eigenvalues reach at least fourteen-digit accuracy with N20. For comparison, we list in Table 2 the numerical results obtained by Multigrid Correction Scheme in [17]. However, the numerical results reported in Table 2 have at most six-digit accuracy despite utilizing a great quantity of degrees of freedom.

    Table 2.  The eigenvalue approximations of (2.3) obtained by Multigrid Correction Scheme and direct method(square: β(x)=4).
    h λc1,h λc2,h λc3,h λc4,h
    2512 2.20250138679 0.21225453108 0.21225510721 0.90806663225
    21024 2.20250569143 0.21225275994 0.21225290397 0.90805872239
    h λ1,h λ2,h λ3,h λ4,h
    2512 2.20250138680 0.21225453108 0.21225510721 0.90806663225
    21024 2.20250569144 0.21225275992 0.21225290395 0.90805872238

     | Show Table
    DownLoad: CSV

    Next, we choose the numerical solutions of N=40 as reference solutions, the corresponding error figures of the approximate eigenvalues λiN(i=1,2,3,4) with different N are listed in Figure 1. We know from Figure 1 that the approximation eigenvalues converge gradually with the increase of N.

    Figure 1.  Errors curves between approximation solutions and the reference solution.

    Example 2. We take k=1, β(x)=4+4i and D=(22,22)2 as our second example. The approximation eigenvalue λiN(i=1,2,3,4) of complex Steklov eigenvalues with largest imaginary parts on the square domain are shown in Table 3.

    Table 3.  The approximation eigenvalue λiN(i=1,2,3,4) for different N on the square.
    N λ1N λ2N λ3N λ4N
    10 0.6865518312509367 0.3430465448633202 0.3430465448633193 0.9501102467115710
    +2.495293986064900i +0.8507465067996103i +0.8507465067996082i +0.5400967091477815i
    15 0.6865518312509322 0.3430465448633182 0.3430465448633182 0.9501102467115667
    +2.495293986064902i +0.8507465067996055i +0.8507465067996055i +0.5400967091477780i
    20 0.6865518312509390 0.3430465448633176 0.3430465448633159 0.9501102467115643
    +2.495293986064897i +0.8507465067996055i +0.8507465067996060i +0.5400967091477757i
    25 0.6865518312509413 0.3430465448633178 0.3430465448633153 0.9501102467115623
    +2.495293986064906i +0.8507465067996077i +0.8507465067996044i +0.5400967091477753i

     | Show Table
    DownLoad: CSV

    We can see from Table 3 that the complex Steklov eigenvalues reach at least thirteen-digit accuracy with N20. The numerical solutions obtained by Multigrid Correction Scheme of [17] in Table 4 have at most seven-digit accuracy despite utilizing a great quantity of degrees of freedom.

    Table 4.  The eigenvalue approximations of (2.3) obtained by Multigrid Correction Scheme and direct method(square: β(x)=4+4i).
    h λc1,h λc2,h λc3,h λc6,h
    2512 0.6865580791 0.3430478705 0.3430446279 0.9501192972
    +2.49529459i +0.85074449i +0.85074328i +0.54009581i
    21024 0.6865533933 0.3430468763 0.3430460656 0.9501125093
    +2.49529414i +0.850746i +0.8507457i +0.54009649i
    h λ1,h λ2,h λ3,h λ4,h
    2512 0.6865580791 0.3430478705 0.3430446278 0.9501192972
    +2.49529459i +0.850744489i +0.8507432795i +0.540095814i
    21024 - - - -

     | Show Table
    DownLoad: CSV

    Similarly, we also choose the numerical solutions of N=40 as reference solutions, the corresponding error figures of the approximate eigenvalues λiN(i=1,2,3,4) with different N are listed in Figure 2. We observe from Figure 2 that the approximation eigenvalues also converge gradually with the increase of N.

    Figure 2.  Errors curves between approximation solutions and the reference solution.

    Example 3. When β(x) is a variable coefficient, we take k=1, β(x1,x2)=[(x1+x2)2+1]+(x1x2)2i and D=(22,22)2. The approximation eigenvalue λiN(i=1,2,3,4) of complex Steklov eigenvalues with largest imaginary parts on the square domain are shown in Table 5.

    Table 5.  The approximation eigenvalue λiN(i=1,2,3,4) for different N on the square.
    N λ1N λ2N λ3N λ4N
    10 0.7865907038296051 0.5176843446683526 1.221875087800295 0.6465164817971206
    +0.1535916213694205i +0.1276411469661209i +0.07766877009084490i +0.01723355206635077i
    15 0.7865907038296497 0.5176843446683336 1.221875087800307 0.6465164817971384
    +0.1535916213689529i +0.1276411469658645i +0.07766877009061117i +0.01723355206591358i
    20 0.7865907038296479 0.5176843446683319 1.221875087800304 0.6465164817971389
    +0.1535916213689515i +0.1276411469658636i +0.07766877009060889i +0.01723355206591307i
    25 0.7865907038296479 0.5176843446683351 1.221875087800305 0.6465164817971382
    +0.1535916213689520i +0.1276411469658643i +0.07766877009060946i +0.01723355206591316i

     | Show Table
    DownLoad: CSV

    We can see from Table 5 that the first four of complex Steklov eigenvalues reach at least thirteen-digit accuracy with N20. Again, we choose the numerical solutions of N=40 as reference solutions, the corresponding error figures of the approximate eigenvalues λiN(i=1,2,3,4) with different N are listed in Figure 3. From Figure 3, we can see that the approximation eigenvalues also converge gradually with the increase of N.

    Figure 3.  Errors curves between approximation solutions and the reference solution.

    Example 4. We consider the problem (2.3) in a three-dimensional case, where we take k=2, β(x)=1+i and ¯D=[0,1]3. The approximation eigenvalue λiN(i=1,2,3,4) of complex Steklov eigenvalues with largest imaginary parts on ¯D=[0,1]3 are shown in Table 6.

    Table 6.  The approximation eigenvalue λiN(i=1,2,3,4) for different N on ¯D=[0,1]3.
    N λ1N λ2N λ3N λ4N
    5 0.6408976921157513 0.6930457356247105 0.6930457356247079 0.6930457356247073
    +0.8336283636488195i +0.4147709839135792i +0.4147709839135787i +0.4147709839135795i
    10 0.6408976931254834 0.6930457440435756 0.6930457440435721 0.6930457440435676
    +0.8336283608134208i +0.4147710214947442i +0.4147710214947438i +0.4147710214947377i
    15 0.6408976931254811 0.6930457440435714 0.6930457440435694 0.6930457440435678
    +0.8336283608134180i +0.4147710214947426i +0.4147710214947404i +0.4147710214947374i
    20 0.6408976931254825 0.6930457440435693 0.6930457440435679 0.6930457440435653
    +0.8336283608134184i +0.4147710214947407i +0.4147710214947393i +0.4147710214947370i

     | Show Table
    DownLoad: CSV

    We can see from Table 6 that the first four of complex Steklov eigenvalues reach at least thirteen-digit accuracy with N15.

    Example 5. Consider the problem (2.3) with k=1 in the unit disk centered at (0,0) with radius R=1. We choose the index of refraction β(x)=4. The three largest Steklov eigenvalues for different N and different M are listed in Tables 79, respectively.

    Table 7.  Numerical eigenvalues to λ1 for different N and M in the circle.
    N M=4 M=6 M=8 M=10
    10 5.151840642736440 5.151840642736451 5.151840642736432 5.151840642736449
    15 5.151840642736440 5.151840642736460 5.151840642736455 5.151840642736447
    20 5.151840642736454 5.151840642736441 5.151840642736452 5.151840642736441
    25 5.151840642736454 5.151840642736440 5.151840642736422 5.151840642736442
    30 5.151840642736442 5.151840642736443 5.151840642736438 5.151840642736472

     | Show Table
    DownLoad: CSV
    Table 8.  Numerical eigenvalues to λ2 for different N and M in the circle.
    N M=4 M=6 M=8 M=10
    10 0.2235784688644089 0.2235784688644088 0.2235784688644088 0.2235784688644094
    15 0.2235784688644086 0.2235784688644084 0.2235784688644087 0.2235784688644084
    20 0.2235784688644089 0.2235784688644082 0.2235784688644086 0.2235784688644084
    25 0.2235784688644082 0.2235784688644083 0.2235784688644086 0.2235784688644084
    30 0.2235784688644093 0.2235784688644087 0.2235784688644088 0.2235784688644090

     | Show Table
    DownLoad: CSV
    Table 9.  Numerical eigenvalues to λ3 for different N and M in the circle.
    N M=4 M=6 M=8 M=10
    10 0.2235784688644083 0.2235784688644084 0.2235784688644087 0.2235784688644091
    15 0.2235784688644081 0.2235784688644082 0.2235784688644086 0.2235784688644080
    20 0.2235784688644087 0.2235784688644075 0.2235784688644081 0.2235784688644083
    25 0.2235784688644081 0.2235784688644079 0.2235784688644085 0.2235784688644084
    30 0.2235784688644089 0.2235784688644086 0.2235784688644083 0.2235784688644088

     | Show Table
    DownLoad: CSV

    We observe from Tables 79 that approximation eigenvalues to λ1, λ2, λ3 reach at least fourteen-digit accuracy with N15 and M4. For comparison, we list the numerical results of [15] in Table 10. The numerical results to λ1, λ2, λ3 reported in Table 10 have at most three-digit accuracy despite utilizing a great quantity of degrees of freedom.

    Table 10.  The largest six Steklov eigenvalues for the circle β(x)=4.
    h 1st 2nd 3rd 4th 5th 6th
    0.2341 5.016606 0.206380 0.205917 1.294039 1.294339 2.561531
    0.1208 5.116979 0.219175 0.219048 1.275370 1.275440 2.494866
    0.0613 5.143045 0.222469 0.222436 1.270670 1.270687 2.478245
    0.0309 5.149636 0.223301 0.223292 1.269493 1.269497 2.474088
    0.0155 5.151289 0.223509 0.223507 1.269198 1.269199 2.473049

     | Show Table
    DownLoad: CSV

    Example 6. When β(x) is complex, we take β(x)=4+4i, k=1 and Ω is a unit disk centered at (0,0) with radius R=1. The numerical results of the first three complex Steklov eigenvalues with largest imaginary parts in a unit circle are shown in Tables 1113, respectively.

    Table 11.  Numerical eigenvalues to λ1 for different N and M in the circle.
    N M=4 M=6 M=8 M=10
    10 0.3205059883274507 0.3205059883274481 0.3205059883274489 0.3205059883274535
    +3.124689326318511i +3.124689326318512i +3.124689326318508i +3.124689326318517i
    15 0.3205059883274459 0.3205059883274499 0.3205059883274493 0.3205059883274489
    +3.124689326318510i +3.124689326318511i +3.124689326318511i +3.124689326318497i
    20 0.3205059883274491 0.3205059883274469 0.3205059883274485 0.3205059883274508
    +3.124689326318506i +3.124689326318513i +3.124689326318509i +3.124689326318506i
    25 0.3205059883274496 0.3205059883274490 0.3205059883274446 0.3205059883274471
    +3.124689326318505i +3.124689326318504i +3.124689326318509i +3.124689326318505i
    30 0.3205059883274480 0.3205059883274495 0.3205059883274517 0.3205059883274509
    +3.124689326318509i +3.124689326318511i +3.124689326318508i +3.124689326318516i

     | Show Table
    DownLoad: CSV
    Table 12.  Numerical eigenvalues to λ2 for different N and M in the circle.
    N M=4 M=6 M=8 M=10
    10 0.1368609477039202 0.1368609477039205 0.1368609477039209 0.1368609477039210
    +1.396737494788579i +1.396737494788578i +1.396737494788579i +1.396737494788577i
    15 0.1368609477039197 0.1368609477039195 0.1368609477039205 0.1368609477039195
    +1.396737494788579i +1.396737494788580i +1.396737494788578i +1.396737494788576i
    20 0.1368609477039198 0.1368609477039204 0.1368609477039212 0.1368609477039199
    +1.396737494788577i +1.396737494788578i +1.396737494788577i +1.396737494788578i
    25 0.1368609477039201 0.1368609477039194 0.1368609477039191 0.1368609477039208
    +1.396737494788579i +1.396737494788577i +1.396737494788578i +1.396737494788579i
    30 0.1368609477039196 0.1368609477039200 0.1368609477039198 0.1368609477039203
    +1.396737494788580i +1.396737494788577i +1.396737494788579i +1.396737494788581i

     | Show Table
    DownLoad: CSV
    Table 13.  Numerical eigenvalues to λ3 for different N and M in the circle.
    N M=4 M=6 M=8 M=10
    10 0.1368609477039208 0.1368609477039200 0.1368609477039202 0.1368609477039211
    +1.396737494788581i +1.396737494788579i +1.396737494788579i +1.396737494788579i
    15 0.1368609477039203 0.1368609477039200 0.1368609477039197 0.1368609477039205
    +1.396737494788579i +1.396737494788581i +1.396737494788578i +1.396737494788579i
    20 0.1368609477039197 0.1368609477039202 0.1368609477039205 0.1368609477039204
    +1.396737494788578i +1.396737494788579i +1.396737494788578i +1.396737494788580i
    25 0.1368609477039201 0.1368609477039195 0.1368609477039200 0.1368609477039204
    +1.396737494788580i +1.396737494788581i +1.396737494788578i +1.396737494788582i
    30 0.1368609477039198 0.1368609477039193 0.1368609477039201 0.1368609477039199
    +1.396737494788581i +1.39673749478858i +1.396737494788581i +1.396737494788582i

     | Show Table
    DownLoad: CSV

    We observe from Tables 1113 that approximation eigenvalues to λ1, λ2, λ3 achieve at least thirteen-digit accuracy with N15 and M4. The results in Table 14 are obtained in [15], which are comparison to the results in Tables 11–13. The numerical eigenvalues to λ1, λ2, λ3 reported in Table 14 have at most four-digit accuracy despite utilizing a great quantity of degrees of freedom.

    Table 14.  Table Eigenvalues for the circle β(x)=4+4i.
    h 1st 2nd 3rd 4th
    0.2341 0.298121 0.134181 0.133990 1.371155
    +3.131620i +1.375387i +1.374565i +0.786327i
    0.1208 0.314981 0.136106 0.136049 1.357526
    +3.126494i +1.391267i +1.391044i +0.790318i
    0.0613 0.319127 0.136650 0.136666 1.354126
    +3.125146i +1.395302i +1.395359i +0.79135i
    0.0310 0.320161 0.136812 0.136808 1.353338
    +3.124804i +1.396392i +1.396378i +0.791628i
    0.0155 0.320420 0.136849 0.136848 1.353145
    +3.124718i +1.396651i +1.396647i +0.791701i

     | Show Table
    DownLoad: CSV

    In this paper, we propose an efficient spectral method for solving a new Steklov eigenvalue problem. First, we give an efficient spectral Galerkin approximation for the new Steklov eigenvalue problem in rectangular domain, and prove the error estimates of approximation eigenvalues and eigenfunctions. Secondly, we derive the matrix form of discrete scheme based on tensor product, and analyze the sparsity of mass matrix and stiffness matrix. In addition, we give an efficient spectral Galerkin approximation for the new Steklov eigenvalue problem in circular domain. Finally, we present ample numerical examples which validate the effectiveness and high accuracy of the algorithm.

    The method proposed in this paper can be extended to some more complex problems, such as transmission eigenvalue problem, electromagnetic eigenvalue problem, nonlinear eigenvalue problem and so on, which will be our goal in the future.

    The authors would like to thank the editor and the referees for helpful comments and suggestions. This work is supported by the National Natural Science Foundation of China (No. 11961009), Guizhou Provincial Graduate Education Innovation Program (No. YJSCXJH [2020] 097) and the Scientific Research Foundation of Guizhou University of Finance and Economics(No. 2020XYB10), the Project for Young Talents Growth of Guizhou Provincial Department of Education under (Grant No. KY[2022]179).

    The authors declare that they have no competing interests.



    [1] S. Bergman, M. Schiffer, Kernel functions and elliptic differential equations in mathematical physics, New York: Academic Press, 1953.
    [2] C. Conca, J. Planchard, M. Vanninathanm, Fluid and periodic structures, Paris: John Wiley & Sons Inc, 1995.
    [3] D. Bucur, I. Ionescu, Asymptotic analysis and scaling of friction parameters, Z. angew. Math. Phys., 57 (2006), 1042–1056. https://doi.org/10.1007/s00033-006-0070-9 doi: 10.1007/s00033-006-0070-9
    [4] J. Canavati, A. Minzoni, A discontinuous Steklov problem with an application to water waves, J. Math. Anal. Appl., 69 (1979), 540–558. https://doi.org/10.1016/0022-247X(79)90165-3 doi: 10.1016/0022-247X(79)90165-3
    [5] A. Andreev, T. Todorov, Isoparametric finite-element approximation of a Steklov eigenvalue problem, IMA J. Numer. Anal., 24 (2004), 309–322. https://doi.org/10.1093/imanum/24.2.309 doi: 10.1093/imanum/24.2.309
    [6] Y. Yang, Q. Li, S. Li, Nonconforming finite element approximations of the Steklov eigenvalue problem, Appl. Numer. Math., 59 (2009), 2388–2401. https://doi.org/10.1016/j.apnum.2009.04.005 doi: 10.1016/j.apnum.2009.04.005
    [7] M. Armentano, C. Padra, A posteriori error estimates for the Steklov eigenvalue problem, Appl. Numer. Math., 58 (2008), 593–601. https://doi.org/10.1016/j.apnum.2007.01.011 doi: 10.1016/j.apnum.2007.01.011
    [8] J. An, H. Bi, Z. Luo, A highly efficient spectral-Galerkin method based on tensor product for fourth-order Steklov equation with boundary eigenvalue, J. Inequal. Appl., 2016 (2016), 211. https://doi.org/10.1186/s13660-016-1158-1 doi: 10.1186/s13660-016-1158-1
    [9] Q. Li, Y. Yang, A two-grid discretization scheme for the Steklov eigenvalue problem, J. Appl. Math. Comput., 36 (2011), 129–139. https://doi.org/10.1007/s12190-010-0392-9 doi: 10.1007/s12190-010-0392-9
    [10] H. Bi, Y. Yang, A two-grid method of the non-conforming Crouzeix-Raviart element for the Steklov eigenvalue problem, Appl. Math. Comput., 217 (2011), 9669–9678. https://doi.org/10.1016/j.amc.2011.04.051 doi: 10.1016/j.amc.2011.04.051
    [11] L. Cao, L. Zhang, W. Allegretto, Y. Lin, Multiscale asymptotic method for Steklov eigenvalue equations in composite media, SIAM J. Numer. Anal., 51 (2013), 273–296. https://doi.org/10.1137/110850876 doi: 10.1137/110850876
    [12] F. Lepe, D. Mora, G. Rivera, I. Velásquez, A virtual element method for the Steklov eigenvalue problem allowing small edges, J. Sci. Comput., 88 (2021), 44. https://doi.org/10.1007/s10915-021-01555-3 doi: 10.1007/s10915-021-01555-3
    [13] S. Domínguez, Steklov eigenvalues for the Lamé operator in linear elasticity, J. Comput. App. Math., 394 (2021), 113558. https://doi.org/10.1016/j.cam.2021.113558 doi: 10.1016/j.cam.2021.113558
    [14] F. Cakoni, D. Colton, S. Meng, P. Monk, Stekloff eigenvalues in inverse scattering, SIAM J. Appl. Math., 76 (2016), 1737–1763. https://doi.org/10.1137/16M1058704 doi: 10.1137/16M1058704
    [15] J. Liu, J. Sun, T. Turner, Spectral indicator method for a non-selfadjoint Steklov eigenvalue problem, J. Sci. Comput., 79 (2019), 1814–1831. https://doi.org/10.1007/s10915-019-00913-6 doi: 10.1007/s10915-019-00913-6
    [16] H. Bi, Y. Zhang, Y. Yang, Two-grid discretizations and a local finite element scheme for a non-selfadjoint Stekloff eigenvalue problem, Comput. Math. Appl., 79 (2020), 1895–1913. https://doi.org/10.1016/j.camwa.2018.08.047 doi: 10.1016/j.camwa.2018.08.047
    [17] Y. Zhang, H. Bi, Y. Yang, A multigrid correction scheme for a new Steklov eigenvalue problem in inverse scattering, Int. J. Comput. Math., 97 (2020), 1412–1430. https://doi.org/10.1080/00207160.2019.1622686 doi: 10.1080/00207160.2019.1622686
    [18] Y. Yang, Y. Zhang, H. Bi, Non-coforming Crouzeix-Raviart element approximation for Stekloff eigenvalues in inverse scattering, Adv. Comput. Math., 46 (2020), 81. https://doi.org/10.1007/s10444-020-09818-7 doi: 10.1007/s10444-020-09818-7
    [19] F. Xu, M. Yue, Q. Huang, H. Ma, An asymptotically exact a posteriori error estimator for non-selfadjoint Steklov eigenvalue problem, Appl. Numer. Math., 156 (2020), 210–227. https://doi.org/10.1016/j.apnum.2020.04.020 doi: 10.1016/j.apnum.2020.04.020
    [20] G. Wang, J. Meng, Y. Wang, L. Mei, A priori and a posteriori error estimates for a virtual element method for the non-self-adjoint Steklov eigenvalue problem, IMA J. Numer. Anal., in press. https://doi.org/10.1093/imanum/drab079
    [21] J. Shen, Efficient spectral-Galerkin methods III: polar and cylindrical geometries, SIAM J. Sci. Comput., 18 (1997), 1583–1604. https://doi.org/10.1137/S1064827595295301 doi: 10.1137/S1064827595295301
    [22] T. Tan, J. An, Spectral Galerkin approximation and rigorous error analysis for the Steklov eigenvalue problem in circular domain, Math. Method. Appl. Sci., 41 (2018), 3764–3778. https://doi.org/10.1002/mma.4863 doi: 10.1002/mma.4863
    [23] J. An, An efficient Legendre-Galerkin spectral approximation for steklov eigenvalue problem, Scientia Sinica Mathematica, 45 (2015), 83–92. https://doi.org/10.1360/012014-64 doi: 10.1360/012014-64
    [24] B. Guo, Z. Wang, Z. Wan, D. Chu, Second order Jacobi approximation with applications to fourth-order differential equations, Appl. Numer. Math., 55 (2005), 480–520. https://doi.org/10.1016/j.apnum.2005.01.002 doi: 10.1016/j.apnum.2005.01.002
    [25] X. Yu, B. Guo, Spectral method for fourth-order problems on quadrilaterals, J. Sci. Comput., 66 (2016), 477–503. https://doi.org/10.1007/s10915-015-0031-6 doi: 10.1007/s10915-015-0031-6
    [26] G. Hsiao, W. Wendland, Boundary integral equations, Berlin: Springer-Verlag, 2008. https://doi.org/10.1007/978-3-540-68545-6
    [27] M. Dauge, Elliptic boundary value problems on corner domains: smoothness and asymptotics of solutions, Berlin: Springer-Verlag, 1988. https://doi.org/10.1007/BFb0086682
    [28] J. Shen, T. Tang, L. Wang, Spectral methods, Heidelberg: Springer, 2011. https://doi.org/10.1007/978-3-540-71041-7
    [29] E. Garau, P. Morin, Convergence and quasi-optimality of adaptive FEM for Steklov eigenvalue problems, IMA J. Numer. Anal., 31 (2011), 914–946. https://doi.org/10.1093/imanum/drp055 doi: 10.1093/imanum/drp055
    [30] I. Babuska, J. Osborn, Eigenvalue problem, In: Handbook of numerical analysis, North-Holand: Elsevier Science Publishers, 1991,641–787. https://doi.org/10.1016/S1570-8659(05)80042-0
  • This article has been cited by:

    1. Jiali Xie, Hai Bi, A multigrid discretization scheme based on the shifted inverse iteration for the Steklov eigenvalue problem in inverse scattering, 2023, 21, 2391-5455, 10.1515/math-2022-0607
  • Reader Comments
  • © 2022 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(1900) PDF downloads(103) Cited by(1)

Figures and Tables

Figures(3)  /  Tables(14)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog