Loading [MathJax]/jax/output/SVG/jax.js
Research article

Numerical differentiation for two-dimensional scattered data on arbitrary domain base on Hermite extension with an implicit iteration process

  • Received: 19 August 2021 Revised: 13 December 2021 Accepted: 04 January 2022 Published: 14 January 2022
  • MSC : 47A52, 65D25

  • In this paper, we develop a method for numerical differentiation of two-dimensional scattered input data on arbitrary domain. A Hermite extension approach is used to realize the approximation and a modified implicit iteration method is proposed to stabilize the approximation process. For functions with various smooth conditions, the numerical solution process of the method is uniform. The error estimates are obtained and numerical results show that the new method is effective. The advantage of the method is that it can solve the problem in any domain.

    Citation: Benxue Gong, Zhenyu Zhao, Tiao Bian, Yingmei Wang. Numerical differentiation for two-dimensional scattered data on arbitrary domain base on Hermite extension with an implicit iteration process[J]. AIMS Mathematics, 2022, 7(4): 5991-6015. doi: 10.3934/math.2022334

    Related Papers:

    [1] Yu Xu, Youjun Deng, Dong Wei . Numerical solution of forward and inverse problems of heat conduction in multi-layered media. AIMS Mathematics, 2025, 10(3): 6144-6167. doi: 10.3934/math.2025280
    [2] Fan Yang, Qianchao Wang, Xiaoxiao Li . A fractional Landweber iterative regularization method for stable analytic continuation. AIMS Mathematics, 2021, 6(1): 404-419. doi: 10.3934/math.2021025
    [3] Dun-Gang Li, Fan Yang, Ping Fan, Xiao-Xiao Li, Can-Yun Huang . Landweber iterative regularization method for reconstructing the unknown source of the modified Helmholtz equation. AIMS Mathematics, 2021, 6(9): 10327-10342. doi: 10.3934/math.2021598
    [4] Zhongdi Cen, Jian Huang, Aimin Xu, Anbo Le . An iterative method for solving a PDE with free boundary arising from pricing corporate bond with credit rating migration. AIMS Mathematics, 2023, 8(2): 3286-3302. doi: 10.3934/math.2023169
    [5] Imo Kalu Agwu, Umar Ishtiaq, Naeem Saleem, Donatus Ikechi Igbokwe, Fahd Jarad . Equivalence of novel IH-implicit fixed point algorithms for a general class of contractive maps. AIMS Mathematics, 2023, 8(1): 841-872. doi: 10.3934/math.2023041
    [6] Xiaopeng Yi, Chongyang Liu, Huey Tyng Cheong, Kok Lay Teo, Song Wang . A third-order numerical method for solving fractional ordinary differential equations. AIMS Mathematics, 2024, 9(8): 21125-21143. doi: 10.3934/math.20241026
    [7] Christelle Dleuna Nyoumbi, Antoine Tambue . A fitted finite volume method for stochastic optimal control problems in finance. AIMS Mathematics, 2021, 6(4): 3053-3079. doi: 10.3934/math.2021186
    [8] Xu Li, Rui-Feng Li . Shift-splitting iteration methods for a class of large sparse linear matrix equations. AIMS Mathematics, 2021, 6(4): 4105-4118. doi: 10.3934/math.2021243
    [9] Shubham Kumar Mittal, Sunil Panday, Lorentz Jäntschi, Liviu C. Bolunduţ . Two novel efficient memory-based multi-point iterative methods for solving nonlinear equations. AIMS Mathematics, 2025, 10(3): 5421-5443. doi: 10.3934/math.2025250
    [10] Mouhamad Al Sayed Ali, Miloud Sadkane . Acceleration of implicit schemes for large systems of nonlinear differential-algebraic equations. AIMS Mathematics, 2020, 5(1): 603-618. doi: 10.3934/math.2020040
  • In this paper, we develop a method for numerical differentiation of two-dimensional scattered input data on arbitrary domain. A Hermite extension approach is used to realize the approximation and a modified implicit iteration method is proposed to stabilize the approximation process. For functions with various smooth conditions, the numerical solution process of the method is uniform. The error estimates are obtained and numerical results show that the new method is effective. The advantage of the method is that it can solve the problem in any domain.



    In the application of many mathematical physics problems, we need to estimate derivatives of an unknown function from given noisy data. It turns out to be an ill-posed problem, which means, the small errors in the measurement data can induce huge errors in its computed derivatives. Many methods and techniques have been proposed regarding this topic [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17]. According to type of regularization techniques, these methods can be classified into finite difference methods, mollification methods, differentiation by integration method and Tikhonov methods. Most of these methods are for one-dimensional case, but only a few are for high-dimensional cases [18,19,20]. These methods of dealing with two-dimensional problems are basically aimed at the regular region, and there are few methods that can deal with the numerical differentiation problem on the irregular domain. For example, if the data are given at scatter points of the domain in Figure 1, numerical implementation of most existing methods is difficult.

    Figure 1.  Irregular region.

    By the extension theorem [21], any function on sub-domain of R2 can be extended to R2 without losing its smoothness. There are many work of extension problem has been developed since Whitney's seminal work [22,23,24,25]. And it is well known that functions with high smoothness can be approximated very precisely by their Hermite expansion. So in this paper, we consider a Hermite extension method to deal with the numerical differentiation problem on the irregular domain. That is to say, we take a function defined on any domain as part of a function on R2. For the one-dimensional case, the numerical differentiation method along the line of this method has been given in [26]. The numerical process of extension is usually unstable, so regularization technology is needed. As an alternative regularization method, we use a modified implicit iteration method in this paper. Compared with Tikhonov method which is used in [17], the implicit iteration method can select larger regularization parameters in each iteration, which makes the calculation process more stable. In [27], Jin has given the detailed theory and numerical implementation of implicit iteration method in Hilbert scales. But the application of the method to our problem still needs some improvement. The existing implicit iterative method can only deal with the problem of finite smoothness, and the numerical implementation of the algorithm is inconsistent for different smoothness. In this paper, we will present a modified form of implicit iteration method. It can deal with numerical differentiation of functions with any smoothness and the solution process is uniform.

    This paper is organized as follows: In the next section, we introduce some preliminary materials. In Section 3, we describe the modified implicit iteration method for numerical differentiation problem and give some auxiliary results. The convergence estimate of the approximation solution can be founded in Section 4. In Section 5, some numerical examples are given. Conclusion is then given in the final Section 6.

    In this section, we introduce some notations and preliminaries that will be used throughout the paper.

    Let x=(x1,x2), |x|=x21+x22. Let

    ˆf(ω)=F[f(x)]=12πR2f(x)eiωxdx

    be the 2-dimensional Fourier transform of the function f(x)L2(R2). The corresponding inverse Fourier transform of the function ˆf(ω) is

    f(x)=12πR2ˆf(ω)eiωxdω.

    And p denotes the norm of the Sobolev space Hp(R2) with p0 defined by

    fp=(R2(1+|ω|2)p|ˆf(ω)|2dω)12. (2.1)

    Particularly, for p=0 we can recover the L2(R2) norm, i.e.,

    f=(R2|ˆf(ω)|2dω)12.

    Let =(1,2), i(i=1,2) being non-negative integers. Set ||1=1+2 and x=1x1+2x2. The normalized 2-dimensional Hermite function is defined by

    H(x)=H1(x1)H2(x2), (2.2)

    where

    H0(xi)π1/4exp((1/2)x2i),H1(xi)π1/42xiexp((1/2)x2i),Hi+1(xi)=2i+1xiHi(xi)ii+1Hi1(xi),i1. (2.3)

    We know that the Fourier transform of the Hermite functions can be given as[28]:

    ˆH(ω)=(i)||1H(ω). (2.4)

    The set of Hermite functions satisfy the orthogonality relations

    R2H(x)Hm(x)dx=δ,m. (2.5)

    The Hermite expansion of a function fL2(R2) is as

    f(x)=||1=0fH(x), (2.6)

    with the Fourier-Hermite coefficients

    f=R2f(x)H(x)dx,||1=0,1,. (2.7)

    Suppose that Λ is a subdomain of R2. For any two-tuples α=(α1,α2)N2,|α|1=α1+α2. The notation j stands for xj and Dαf=α1x1α2x2f. For any positive integer q, we define Dqf=:{Dαf:|α|1=q} and

    |Dqf|=(|α|1=q|Dαf|2)1/2. (2.8)

    The norm s,Λ in Sobolev space Hs(Λ) is defined as

    fs,Λ:=(Λ|f|2+|Dsf|2dx)1/2, (2.9)

    where s=0 and 0,Λ denotes the L2(Λ) norm.

    Suppose that g(x)Hp(R),p2 and we only know its approximate function gδ on Λ such that

    gδg0,Λδ, (2.10)

    where δ>0 is a given constant called the error level. Our problem is to calculate approximate derivatives of g on Λ from the noisy data gδ, or, equivalently, to construct a function fδ(x) from gδ(x) which is close to g(x) in the sense that

    limδ0fδgr,Λ=0,r1. (2.11)

    For any vector f={f}||1=0l2, if we let

    Hf:=||=0fH(x), (3.1)

    then the process of constructing an approximation function fδ from data gδ can be transformed to solving the following equations

    Hf=gδ. (3.2)

    In this paper, we present an modified implicit iteration method to solve the above equations. For this purpose, we introduce the following operator:

    Rf:=H1F1[e|ω|^Hf(ω)]. (3.3)

    It is obvious that R is unbounded self-adjoint strictly positive definite operator. Then we choose

    fδn=Hfδn, (3.4)

    as the approximation of g, where fδn is determined by the following implicit iteration process

    fδ0=0,fδk=fδk1(HH+βkR2)1H(Hfδk1gδ),k=1,2,,n, (3.5)

    where βk>0 are properly chosen real numbers. For reference [27], the positive number

    σn:=nk=11βk (3.6)

    plays the role of the regularization parameter and we will chosen it as the solution of the nonlinear equation

    d(σn):=Hfδngδ0,Λ=Cδ, (3.7)

    with a constant C1. If we let T=HR1, then fδn possesses the representation [27]

    fδn=R1sn(TT)Tgδwithsn(λ)=1λ(1nk=1βkλ+βk). (3.8)

    Remark 3.1. If we use the operator B=(|α|1=qDα) with some a constant q instead of R, then we return to the framework in [27] and the convergence results can be obtained accordingly. When p and q satisfy a certain relation, the result is order optimal. It should be noticed that for large q, the numerical process of the method is difficult. We will point that the method is always order optimal when we use the operator R and the numerical process is uniform for any p.

    The following lemma holds for sn(λ).

    Lemma 3.1. [27] The function sn:(0,c](0,) with c=T2 and the corresponding residual function rn(λ):=1λsn(λ) obey the properties

    sn(λ)σn,λsn(λ)1,λrn(λ)σ1n,rn(λ)1. (3.9)

    From above lemma, we can deduce the following results.

    Lemma 3.2.

    λsn(λ)σn,λrn(λ)σ1n. (3.10)

    Proof. For λσ1,

    λsn(λ)λσnσn (3.11)

    and

    λrn(λ)λσ1n. (3.12)

    Moreover, for λσ1,

    λsn(λ)=λλλsn(λ)λλσn (3.13)

    and

    λrn(λ)λλλrn(λ)σ1nλσ1n. (3.14)

    Owing to gHp(R), we suppose that

    gpE, (4.1)

    where E is a constant. Set the vector g contains all Fourier-Hermite coefficients of g, i.e.,

    g(x)=(Hg)(x),xR2. (4.2)

    Let

    gN=PNgandgN=HgN. (4.3)

    We define the vector fn as

    fn=R1sn(TT)TgN, (4.4)

    then we have

    H(fδnfn)=Tsn(TT)T(gδgN), (4.5)
    H(gfn)=Trn(TT)RgN, (4.6)
    gδHfδn=rn(TT)gδ, (4.7)
    R(fδnfn)=sn(TT)T(gδgN), (4.8)
    R(gfn)=rn(TT)RgN. (4.9)

    In our further analysis, we shall make use of the following lemmas.

    Lemma 4.1. If the condition (4.1) holds, then

    ggNNpEandRgNl2CNE, (4.10)

    where

    CN=max(1,eNNp). (4.11)

    Proof. From (2.1) and (3.3), we can obtain

    ggN2=|ω|>N|ˆg(ω)|2dωN2p|ω|>N(1+|ω|2)p|ˆg|2dωN2pg2p (4.12)

    and

    RgNl2=|ω|Ne2|ω||ˆg(ω)|2dω=|ω|Ne2|ω|(1+|ω|2)p(1+|ω|2)p|ˆg(ω)|2dωmax(1,e2NN2p)g2p. (4.13)

    Lemma 4.2. If the condition (4.1) holds, we have

    H(fδngN)0,Λ(C+1)δ+NpE, (4.14)
    R(fδngN)l2σn(δ+NpE)+CNE (4.15)

    and

    Hfδngδ0,Λδ+NpE+σ1nCNE. (4.16)

    Proof. From (2.10), (3.7), (4.10) and the triangle inequality

    H(fδngN)0,ΛHfδngδ0,Λ+gδg0,Λ+ggN0,Λ(C+1)δ+NpE. (4.17)

    And by using the triangle inequality, (2.10), (3.10) and (4.8)–(4.10)

    R(fδngN)l2R(fδnfn)l2+R(fngN)l2=sn(TT)T(gδgN)l2+rn(TT)RgNl2σn(gδgN)0,Λ+RgNl2σn(δ+NpE)+CNE. (4.18)

    Moreover, in terms of the triangle inequality, (2.10), (3.9), (3.10) and (4.7), we have

    Hfδngδ0,Λ=rn(TT)gδ0,Λrn(TT)(gδg)0,Λ+rn(TT)(ggN)0,Λ+rn(TT)gN0,Λδ+ggN0,Λ+rn(TT)TRgNδ+NpE+σ1nCNE. (4.19)

    Lemma 4.3. [21] Let Ω be a domain in R2 satisfying the cone condition. There exists a constant K depending on ϵ0 and j, s, such that for any 0<ϵϵ0 and 0js

    fj,ΩK(ϵfs,Ω+ϵj/(sj)f0,Ω). (4.20)

    Lemma 4.4. Suppose that the vector sequence hδn={hδ}||1=0 satisfies

    Hhδ0,Λk1δ,Rhδl2k2ek3δ1pδ,δ0, (4.21)

    then for any ΩΛ satisfying the cone condition, there exists a constant M

    Hhδp,ΩM. (4.22)

    Proof. It is easy to deduce that there exist a constant δ0 such that

    ek3δ1p>kp3δ,δ<δ0. (4.23)

    And for simplicity, we prove the theorem with δ<δ0. Let

    N0=k3δ1p, (4.24)

    and we have

    Hhδp,ΩH(PN0hδ)p,Ω+H[(IPN0)hδ]p,Ω=I1+I2. (4.25)

    By Parseval's formula, we can see that the second term I2 satisfies

    H[(IPN0)hδ]2p,ΩH[(IPN0)hδ]2=|ω|>N0(1+|ω|2)p|^Hhδ(ω)|2dω=|ω|>N0(1+|ω|2)pe2|ω||e2|ω|^Hhδ(ω)|2dω(N0+1)2pe2N0|ω|>N0|e2|ω|^Hhδ(ω)|2dωN2p0e2(N01)Rhδ2l2e2k2p31δ2k22δ2=e2k2p3k22. (4.26)

    Hence

    I22e1kp3k2. (4.27)

    So all we need is to prove there exist a constant M1 such that

    I1<M1,δ0. (4.28)

    Note that

    H(PN0hδ)0,ΩHhδ0,Ω+H[(IPN0)hδ]0,Ω (4.29)

    and

    H(IPN0)hδ0,ΩH[(IPN0)hδ]2=|ω|>N0|^Hhδ(ω)|2dω=|ω|>N01e2|ω||e|ω|^Hhδ(ω)|2dω1e2N0Rhδ2l2k22δ2. (4.30)

    Therefore

    H(PN0hδ)0,Ω(k1+2k2)δ. (4.31)

    Now we prove (4.28) by using reduction to absurdity, if (4.28) does not hold, then for any q>p there exist a sequence δi such that

    H(PN0hδ)q2k2kp3(δikp3)pqp,δi0. (4.32)

    If not, ˉq for any δ0

    H(PN0hδ)ˉq<2k2kp3(δkp3)pˉqp, (4.33)

    then (4.28) can be derived by 4.29 and Lemma 4.3 with ϵ=(δikp3)ˉqpp, s=ˉq and j=p. Then

    |ω|<N0(N0k=0|ω|kk!)2|^Hhδi(ω)|2dω=|ω|<N0(N0k=0|ω|kk!)2(1+|ω|2)N0(1+|ω|2)N0|^Hhδi(ω)|2dω|ω|<N0(N0k=0|ω|kk!)2(1+|ω|)2N0(1+|ω|2)N0|^Hhδi(ω)|2dω|ω|<N0(N01k=0|ω|kk!)2|ω|2N0(1+|ω|2)N0|^Hhδi(ω)|2dω|ω|<N0(N01k=0N0kk!)2N2N00(1+|ω|2)N0|^Hhδi(ω)|2dω(N01k=0N0kk!)2N2N00H(PN0hδi)2N0k22(N0k=0(k3δ1pi)kk!)2δ2i. (4.34)

    Therefore

    R2e2|ω||^Hhδi(ω)|2dω=limδi0|ω|<N0(δi)(N0(δi)k=0|ω|kk!)2|^Hhδi(ω)|2dωk22(N0k=0(k3δ1pi)kk!)2δ2i=k22limδi0e2k3δ1piδ2i. (4.35)

    So there exists a ˉδ such that

    Rhˉδ2l2=R2e2|ω||^Hhˉδ(ω)|2dω>R2e2|ω||^Hhˉδ(ω)|2dωk22e2k3ˆδ1pˉδ2, (4.36)

    which contradicts the assumptions of the Lemma.

    Theorem 4.1. Suppose that the conditions (2.10) and (4.1) hold, fδn is defined by (3.4) and (3.7) then for any ΩΛ satisfying the cone condition and 0<jp,

    fδngj,Ω=O(δpjp). (4.37)

    Proof. Let

    N0=(2E(C1)δ)1p. (4.38)

    Then from Lemma 4.2, we have

    H(fδngN0)0,Λ3C+12δ, (4.39)
    R(fδngN0)l2C+12e(2EC1)1pδ1pδ. (4.40)

    Thus, by using Lemma 4.4, there exists a constant M

    H(fδngN0)0,ΩM. (4.41)

    Then

    Hfδngp,ΩH(fδngN0)0,Ω+ggNp,ΩH(fδngN0)0,Ω+gpM+E. (4.42)

    Moreover, by using (2.10), (3.7) and the triangle inequality

    Hfδng0,ΩHfδngδ0,Ω+gδg0,ΩHfδngδ0,Λ+gδg0,Λ(C+1)δ. (4.43)

    The assertion of theorem follows from (4.42), (4.43) and Lemma 4.3.

    The data are usually given at scatter points in practical applications. Let xiΛ(i=1,2,,m) be the given points and

    gδ=(gδ(x1),gδ(x2),,gδ(xm))T

    be the noisy data vector. Let σ(j)=(σ(j1),σ(j2)),(0j1,j2n) being the Hermite-Gauss type interpolation points and ρ(j) are the corresponding Hermite-Gauss weights. For f,gL2(R2), we define the discrete inner product

    f,gn:=nj1=1nj2=1ρ(j)f(σ(j))¯g(σ(j)),

    and

    ˘H(ω):=(1)||1e|ω|H(ω).

    Let

    Hn=span{H(0,0)(x),H(1,0)(x),H(1,1)(x),,H(n,n1)(x),H(n,n)(x)},

    then we give the matrices A(n+1)2×(n+1)2, R(n+1)2×(n+1)2, Hm×(n+1)2 as

    A||1+1,|k|1+1=mi=1H(xi)Hk(xi),R||1+1,|k|1+1=˘H,˘Hkn,Hi,||1+1=H(xi),i=1,2,,m;||1,|k|1=0,1,n.

    With these preparations, the discrete form of the implicit iteration method can be given as

    fδ0=0,fk=fk1(HTH+βkR)1HT(Hfδk1gδ),k=1,2,,n. (5.1)

    Suppose that

    (mi=1(gδ(xi)g(xi)2)1/2δ. (5.2)

    Similar to what is done in [27], we take β1=1,βk=qk1β1 with some q<1 and choose n as the first integer for which

    HfδngδC1δ<Hfδkgδ,0k<n, (5.3)

    and then adjust the parameter βn such that

    C2δHfδngδC1δ, (5.4)

    where C1,C2 are two constants that obey 1C1C2.

    Remark 5.1. It should be noted that the Hermite-Gauss points are only used to calculate the matrix R, regardless of the location of the noisy data.

    In this section, we give some numerical tests to verify the effect of the new method. All tests are realized on Windows 10 system with Memory 16GB, CPU Intel(R) Core(TM)i7-8550U by using Matlab 2017b. Let x=(x1,x2,,xm)T and the perturbed data are generated by

    gδ(x)=g(x)+randn(size(x))ϵ, (6.1)

    where ϵ is the error level and randn(size()) is Matlab functions. In all cases we choose the parameter n=64, q=1/2 and C=1.01. We have tested these parameters with other values, and the results are similar. In order to adapt to the characteristic of Hermite function approximation, the scaling factor [29] is used in the numerical processing.

    Example 6.1. [18] Let Λ={(x1,x2)|x21+x221} is a disk and scatter nodes are given as Figure 2a. We choose the exact function as g(x)=(x21+x222)3 and set ϵ=0.01. The numerical results are exhibited in Figures 2c3h.

    Figure 2.  Example 1.
    Figure 3.  Example 2.

    Example 6.2. [18] Let Λ={(x1,x2)|0x11,0x2π} is a rectangle and scatter nodes are given as Figure 3i. We choose the exact function as g(x)=(x21+x222)3 and set ϵ=0.01. We have shown the numerical results in Figures 3k4h.

    Figure 4.  Example 3.

    Example 6.3. [19] Let Λ={(x1,x2)|1x13,1x23}, the data are given at the equidistant nodes whose sampling step was 0.1×0.1. The exact function is chosen as g(x)=sin(12x21+14x22+3)cos(2x1+1exp(x2)). We have shown the numerical results in Figures 4i5h with ϵ=0.01.

    Figure 5.  Example 4.

    Example 6.4. Now we let Λ is a irregular domain and scatter nodes are given as Figure 5i. We choose the exact function as g(x)=cos(x1x2) and set ϵ=0.01. The numerical approximations and corresponding errors are shown in Figures 5k–6h.

    All the above numerical results show that the proposed method is effective.

    In this paper, we present a Hermite extension method with an implicit iteration process for numerical differentiation of two-dimensional functions. Because the method can directly deal with the data given on any domain, it is more convenient than other methods in practical application. The theoretical results show that the convergence rates of the method is self-adaptive.

    The authors declared that they have no conflicts of interest to this work. We declare that we do not have any commercial or associative interest that represents a conflict of interest in connection with the work submitted.



    [1] A. Ramm, A. Smirnova, On stable numerical differentiation, Math. Comp., 70 (2001), 1131–1153. https://dx.doi.org/10.1090/S0025-5718-01-01307-2 doi: 10.1090/S0025-5718-01-01307-2
    [2] R. S. Anderssen, P. Bloomfield, Numerical differentiation procedures for non-exact data, Numer. Math., 22 (1974), 157–182. https://dx.doi.org/10.1007/BF01436965 doi: 10.1007/BF01436965
    [3] S. Lu, V. Naumova, S. V. Pereverzev, Legendre polynomials as a recommended basis for numerical differentiation in the presence of stochastic white noise, J. Inverse Ill-Posed Probl., 21 (2013), 193–216. https://dx.doi.org/10.1515/jip-2012-0050 doi: 10.1515/jip-2012-0050
    [4] H. N. Mhaskar, V. Naumova, S. V. Pereverzyev, Filtered Legendre expansion method for numerical differentiation at the boundary point with application to blood glucose predictions, Appl. Math. Comput., 224 (2013), 835–847. https://dx.doi.org/10.1016/j.amc.2013.09.015 doi: 10.1016/j.amc.2013.09.015
    [5] D. A. Murio, C. E. Mejía, S. Zhan, Discrete mollification and automatic numerical differentiation, Comput. Math. Appl., 35 (1998), 1–16. https://dx.doi.org/10.1016/S0898-1221(98)00001-7 doi: 10.1016/S0898-1221(98)00001-7
    [6] Z. Y. Zhao, A truncated Legendre spectral method for solving numerical differentiation, Int. J. Comput. Math., 87 (2010), 3209–3217. https://dx.doi.org/10.1080/00207160902974404 doi: 10.1080/00207160902974404
    [7] Z. Y. Zhao, J. F. Liu, Hermite spectral and pseudospectral methods for numerical differentiation, Appl. Numer. Math., 61 (2011), 1322–1330. https://dx.doi.org/10.1016/j.apnum.2011.09.006 doi: 10.1016/j.apnum.2011.09.006
    [8] G. W. Groetsch, Lanczo's generalized derivative, Am. Math. Mon., 105 (1998), 320–326. https://dx.doi.org/10.2307/2589707 doi: 10.2307/2589707
    [9] Z. W. Wang, R. S. Wen, Numerical differentiation for high orders by an integration method, J. Comput. Appl. Math., 234 (2010), 941–948. https://dx.doi.org/10.1016/j.cam.2010.01.056 doi: 10.1016/j.cam.2010.01.056
    [10] B. Q. Chen, Z. Y. Zhao, Z. Li, Z. H. Meng, Numerical differentiation by a Fourier extension method with super-order regularization, Appl. Math. Comput., 334 (2018), 1–10. https://dx.doi.org/10.1016/j.amc.2018.04.005 doi: 10.1016/j.amc.2018.04.005
    [11] J. Cullum, Numerical differentiation and regularization, SIAM J. Numer. Anal., 8 (1971), 254–265. https://dx.doi.org/10.1137/0708026 doi: 10.1137/0708026
    [12] M. Hanke, O. Scherzer, Inverse problems light: Numerical differentiation, Am. Math. Mon., 108 (2001), 512–521. https://dx.doi.org/10.2307/2695705 doi: 10.2307/2695705
    [13] B. Hu, S. Lu, Numerical differentiation by a Tikhonov regularization method based on the discrete cosine transform, Appl. Anal., 91 (2012), 719–736. https://dx.doi.org/10.1080/00036811.2011.598862 doi: 10.1080/00036811.2011.598862
    [14] T. Wei, Y. C. Hon, Y. B. Wang, Reconstruction of numerical derivatives from scattered noisy data, Inverse Probl., 21 (2005), 657. https://dx.doi.org/10.1088/0266-5611/21/2/013 doi: 10.1088/0266-5611/21/2/013
    [15] T. Wei, Y. C. Hon, Numerical differentiation by radial basis functions approximation, Adv. Comput. Math., 27 (2007), 247–272. https://dx.doi.org/10.1007/s10444-005-9001-0 doi: 10.1007/s10444-005-9001-0
    [16] S. Lu, S. V. Pereverzev, Numerical differentiation from a viewpoint of regularization theory, Math. Comp., 75 (2006), 1853–1870. https://dx.doi.org/10.1090/S0025-5718-06-01857-6 doi: 10.1090/S0025-5718-06-01857-6
    [17] Z. H. Meng, Z. Y. Zhao, D. Mei, Y. X. Zhou, Numerical differentiation for two-dimensional functions by a Fourier extension method, Inverse Probl. Sci. En., 28 (2020), 126–143. https://dx.doi.org/10.1080/17415977.2019.1661410 doi: 10.1080/17415977.2019.1661410
    [18] Y. B. Wang, T. Wei, Numerical differentiation for two-dimensional scattered data, J. Math. Anal. Appl., 312 (2005), 121–137. https://dx.doi.org/10.1016/j.jmaa.2005.03.025 doi: 10.1016/j.jmaa.2005.03.025
    [19] S. Riachy, M. Mboup, J. Richard, Multivariate numerical differentiation, J. Comput. Appl. Math., 236 (2011), 1069–1089. https://dx.doi.org/10.1016/j.cam.2011.07.031 doi: 10.1016/j.cam.2011.07.031
    [20] S. Qiu, Z. W. Wang, A. Xie, Multivariate numerical derivative by solving an inverse heat source problem, Inverse Probl. Sci. En., 26 (2018), 1178–1197. https://dx.doi.org/10.1080/17415977.2017.1386187 doi: 10.1080/17415977.2017.1386187
    [21] R. Adams, Sobolev spaces, New York: Academic press, 1975.
    [22] H. Whitney, Analytic extensions of differentiable functions defined in closed sets, Trans. Amer. Math. Soc., 36 (1934), 63–89. https://dx.doi.org/10.2307/1989708 doi: 10.2307/1989708
    [23] C. Fefferman, Interpolation and extrapolation of smooth functions by linear operators, Rev. Mat. Iberoam., 21 (2005), 313–348. https://dx.doi.org/10.4171/RMI/424 doi: 10.4171/RMI/424
    [24] C. Fefferman, Whitney's extension problems and interpolation of data, Bull. Amer. Math. Soc., 46 (2008), 207–220. https://dx.doi.org/10.1090/S0273-0979-08-01240-8 doi: 10.1090/S0273-0979-08-01240-8
    [25] C. Fefferman, A. Israel, G. K. Luli, Sobolev extension by linear operators, J. Amer. Math. Soc., 27 (2014), 69–145. https://dx.doi.org/10.1090/S0894-0347-2013-00763-8 doi: 10.1090/S0894-0347-2013-00763-8
    [26] Z. Y. Zhao, A Hermite extension method for numerical differentiation, Appl. Numer. Math., 159 (2021), 46–60. https://dx.doi.org/10.1016/j.apnum.2020.08.016 doi: 10.1016/j.apnum.2020.08.016
    [27] Q. N. Jin, U. Tautenhahn, Implicit iteration methods in Hilbert scales under general smoothness conditions, Inverse Probl., 27 (2011), 045012. https://dx.doi.org/10.1088/0266-5611/27/4/045012 doi: 10.1088/0266-5611/27/4/045012
    [28] S. Thangavelu, Lectures on Hermite and Laguerre expansions, New Jersey: Princeton University Press, 1993. https://dx.doi.org/10.1515/9780691213927
    [29] T. Tao, The Hermite spectral method for gaussian-type functions, SIAM J. Sci. Comput., 14 (1993), 594–606. https://dx.doi.org/10.1137/0914038 doi: 10.1137/0914038
  • This article has been cited by:

    1. Bin Zheng, Junfeng Liu, Zhenyu Zhao, Zhihong Dou, Benxue Gong, A Generalized Iterated Tikhonov Method in the Fourier Domain for Determining the Unknown Source of the Time-Fractional Diffusion Equation, 2024, 16, 2073-8994, 864, 10.3390/sym16070864
  • 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(2256) PDF downloads(58) Cited by(1)

Figures and Tables

Figures(5)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog