
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
[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.
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)]=12π∫R2f(x)e−iω⋅xdx |
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)=12π∫R2ˆf(ω)eiω⋅xdω. |
And ‖⋅‖p denotes the norm of the Sobolev space Hp(R2) with p≥0 defined by
‖f‖p=(∫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 ℓ=(ℓ1,ℓ2), ℓi(i=1,2) being non-negative integers. Set |ℓ|1=ℓ1+ℓ2 and ℓ⋅x=ℓ1x1+ℓ2x2. The normalized 2-dimensional Hermite function is defined by
Hℓ(x)=Hℓ1(x1)Hℓ2(x2), | (2.2) |
where
H0(xi)≡π−1/4exp(−(1/2)x2i),H1(xi)≡π−1/4√2xiexp(−(1/2)x2i),Hℓi+1(xi)=√2ℓi+1xiHℓi(xi)−√ℓiℓi+1Hℓi−1(xi),ℓi≥1. | (2.3) |
We know that the Fourier transform of the Hermite functions can be given as[28]:
ˆHℓ(ω)=(−i)|ℓ|1Hℓ(ω). | (2.4) |
The set of Hermite functions satisfy the orthogonality relations
∫R2Hℓ(x)Hm(x)dx=δℓ,m. | (2.5) |
The Hermite expansion of a function f∈L2(R2) is as
f(x)=∞∑|ℓ|1=0fℓHℓ(x), | (2.6) |
with the Fourier-Hermite coefficients
fℓ=∫R2f(x)Hℓ(x)dx,|ℓ|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=∂α1x1∂α2x2f. 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
‖f‖s,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),p≥2 and we only know its approximate function gδ on Λ such that
‖gδ−g‖0,Λ≤δ, | (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
limδ→0‖fδ−g‖r,Λ=0,r≥1. | (2.11) |
For any vector →f={fℓ}∞|ℓ|1=0∈l2, if we let
H→f:=∞∑|ℓ|=0fℓHℓ(x), | (3.1) |
then the process of constructing an approximation function fδ from data gδ can be transformed to solving the following equations
H→f=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:
R→f:=H−1F−1[e|ω|^H→f(ω)]. | (3.3) |
It is obvious that R is unbounded self-adjoint strictly positive definite operator. Then we choose
fδn=H→fδn, | (3.4) |
as the approximation of g, where →fδn is determined by the following implicit iteration process
→fδ0=0,→fδk=→fδk−1−(H∗H+βkR2)−1H∗(H→fδk−1−gδ),k=1,2,…,n, | (3.5) |
where βk>0 are properly chosen real numbers. For reference [27], the positive number
σn:=n∑k=11βk | (3.6) |
plays the role of the regularization parameter and we will chosen it as the solution of the nonlinear equation
d(σn):=‖H→fδn−gδ‖0,Λ=Cδ, | (3.7) |
with a constant C≥1. If we let T=HR−1, then →fδn possesses the representation [27]
→fδn=R−1sn(T∗T)T∗gδwithsn(λ)=1λ(1−n∏k=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=‖T‖2 and the corresponding residual function rn(λ):=1−λsn(λ) obey the properties
sn(λ)≤σn,λsn(λ)≤1,λrn(λ)≤σ−1n,rn(λ)≤1. | (3.9) |
From above lemma, we can deduce the following results.
Lemma 3.2.
√λsn(λ)≤√σn,√λrn(λ)≤√σ−1n. | (3.10) |
Proof. For λ≤σ−1,
√λsn(λ)≤√λσn≤√σn | (3.11) |
and
√λrn(λ)≤√λ≤√σ−1n. | (3.12) |
Moreover, for λ≥σ−1,
√λsn(λ)=√λλλsn(λ)≤√λλ≤√σn | (3.13) |
and
√λrn(λ)≤√λλλrn(λ)≤σ−1n√λ≤√σ−1n. | (3.14) |
Owing to g∈Hp(R), we suppose that
‖g‖p≤E, | (4.1) |
where E is a constant. Set the vector →g contains all Fourier-Hermite coefficients of g, i.e.,
g(x)=(H→g)(x),∀x∈R2. | (4.2) |
Let
→gN=PN→gandgN=H→gN. | (4.3) |
We define the vector →fn as
→fn=R−1sn(T∗T)T∗gN, | (4.4) |
then we have
H(→fδn−→fn)=Tsn(T∗T)T∗(gδ−gN), | (4.5) |
H(→g−→fn)=Trn(T∗T)R→gN, | (4.6) |
gδ−H→fδn=rn(TT∗)gδ, | (4.7) |
R(→fδn−→fn)=sn(T∗T)T∗(gδ−gN), | (4.8) |
R(→g−→fn)=rn(T∗T)R→gN. | (4.9) |
In our further analysis, we shall make use of the following lemmas.
Lemma 4.1. If the condition (4.1) holds, then
‖g−gN‖≤N−pEand‖R→gN‖l2≤CNE, | (4.10) |
where
CN=max(1,eNNp). | (4.11) |
Proof. From (2.1) and (3.3), we can obtain
‖g−gN‖2=∫|ω|>N|ˆg(ω)|2dω≤N−2p∫|ω|>N(1+|ω|2)p|ˆg|2dω≤N−2p‖g‖2p | (4.12) |
and
‖R→gN‖l2=∫|ω|≤Ne2|ω||ˆg(ω)|2dω=∫|ω|≤Ne2|ω|(1+|ω|2)p(1+|ω|2)p|ˆg(ω)|2dω≤max(1,e2NN2p)‖g‖2p. | (4.13) |
Lemma 4.2. If the condition (4.1) holds, we have
‖H(→fδn−→gN)‖0,Λ≤(C+1)δ+N−pE, | (4.14) |
‖R(→fδn−→gN)‖l2≤√σn(δ+N−pE)+CNE | (4.15) |
and
‖H→fδn−gδ‖0,Λ≤δ+N−pE+√σ−1nCNE. | (4.16) |
Proof. From (2.10), (3.7), (4.10) and the triangle inequality
‖H(→fδn−→gN)‖0,Λ≤‖H→fδn−gδ‖0,Λ+‖gδ−g‖0,Λ+‖g−gN‖0,Λ≤(C+1)δ+N−pE. | (4.17) |
And by using the triangle inequality, (2.10), (3.10) and (4.8)–(4.10)
‖R(→fδn−→gN)‖l2≤‖R(→fδn−→fn)‖l2+‖R(→fn−→gN)‖l2=‖sn(T∗T)T∗(gδ−gN)‖l2+‖rn(T∗T)R→gN‖l2≤√σn‖(gδ−gN)‖0,Λ+‖R→gN‖l2≤√σn(δ+N−pE)+CNE. | (4.18) |
Moreover, in terms of the triangle inequality, (2.10), (3.9), (3.10) and (4.7), we have
‖H→fδn−gδ‖0,Λ=‖rn(TT∗)gδ‖0,Λ≤‖rn(TT∗)(gδ−g)‖0,Λ+‖rn(TT∗)(g−gN)‖0,Λ+‖rn(TT∗)gN‖0,Λ≤δ+‖g−gN‖0,Λ+‖rn(TT∗)T‖⋅‖R→gN‖≤δ+N−pE+√σ−1nCNE. | (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<ϵ≤ϵ0 and 0≤j≤s
‖f‖j,Ω≤K(ϵ‖f‖s,Ω+ϵ−j/(s−j)‖f‖0,Ω). | (4.20) |
Lemma 4.4. Suppose that the vector sequence →hδn={hδℓ}∞|ℓ|1=0 satisfies
‖H→hδ‖0,Λ≤k1δ,‖R→hδ‖l2≤k2ek3δ−1pδ,δ→0, | (4.21) |
then for any Ω⊆Λ satisfying the cone condition, there exists a constant M
‖H→hδ‖p,Ω≤M. | (4.22) |
Proof. It is easy to deduce that there exist a constant δ0 such that
ek3δ−1p>kp3δ,∀δ<δ0. | (4.23) |
And for simplicity, we prove the theorem with δ<δ0. Let
N0=k3δ−1p, | (4.24) |
and we have
‖H→hδ‖p,Ω≤‖H(PN0→hδ)‖p,Ω+‖H[(I−PN0)→hδ]‖p,Ω=I1+I2. | (4.25) |
By Parseval's formula, we can see that the second term I2 satisfies
‖H[(I−PN0)→hδ]‖2p,Ω≤‖H[(I−PN0)→hδ]‖2=∫|ω|>N0(1+|ω|2)p|^H→hδ(ω)|2dω=∫|ω|>N0(1+|ω|2)pe2|ω||e2|ω|^H→hδ(ω)|2dω≤(N0+1)2pe2N0∫|ω|>N0|e2|ω|^H→hδ(ω)|2dω≤N2p0e2(N0−1)‖R→hδ‖2l2≤e−2k2p31δ2⋅k22δ2=e−2k2p3k22. | (4.26) |
Hence
I2≤2e−1kp3k2. | (4.27) |
So all we need is to prove there exist a constant M1 such that
I1<M1,δ→0. | (4.28) |
Note that
‖H(PN0→hδ)‖0,Ω≤‖H→hδ‖0,Ω+‖H[(I−PN0)→hδ]‖0,Ω | (4.29) |
and
‖H(I−PN0)→hδ‖0,Ω≤‖H[(I−PN0)→hδ]‖2=∫|ω|>N0|^H→hδ(ω)|2dω=∫|ω|>N01e2|ω||e|ω|^H→hδ(ω)|2dω≤1e2N0‖R→hδ‖2l2≤k22δ2. | (4.30) |
Therefore
‖H(PN0→hδ)‖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(PN0→hδ)‖q≥2k2kp3(δikp3)p−qp,δi→0. | (4.32) |
If not, ∃ˉq for any δ→0
‖H(PN0→hδ)‖ˉq<2k2kp3(δkp3)p−ˉqp, | (4.33) |
then (4.28) can be derived by 4.29 and Lemma 4.3 with ϵ=(δikp3)ˉq−pp, s=ˉq and j=p. Then
∫|ω|<N0(N0∑k=0|ω|kk!)2|^H→hδi(ω)|2dω=∫|ω|<N0(N0∑k=0|ω|kk!)2(1+|ω|2)N0(1+|ω|2)N0|^H→hδi(ω)|2dω≥∫|ω|<N0(N0∑k=0|ω|kk!)2(1+|ω|)2N0(1+|ω|2)N0|^H→hδi(ω)|2dω≥∫|ω|<N0(N0−1∑k=0|ω|kk!)2|ω|2N0(1+|ω|2)N0|^H→hδi(ω)|2dω≥∫|ω|<N0(N0−1∑k=0N0kk!)2N2N00(1+|ω|2)N0|^H→hδi(ω)|2dω≥(N0−1∑k=0N0kk!)2N2N00‖H(PN0→hδi)‖2N0≥k22(N0∑k=0(k3δ−1pi)kk!)2δ2i. | (4.34) |
Therefore
∫R2e2|ω||^H→hδi(ω)|2dω=limδi→0∫|ω|<N0(δi)(N0(δi)∑k=0|ω|kk!)2|^H→hδi(ω)|2dω≥k22(N0∑k=0(k3δ−1pi)kk!)2δ2i=k22limδi→0e2k3δ−1piδ2i. | (4.35) |
So there exists a ˉδ such that
‖R→hˉδ‖2l2=∫R2e2|ω||^Hhˉδ(ω)|2dω>∫R2e2|ω||^Hhˉδ(ω)|2dω≥k22e2k3ˆδ−1pˉδ2, | (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 Ω⊆Λ satisfying the cone condition and 0<j≤p,
‖fδn−g‖j,Ω=O(δp−jp). | (4.37) |
Proof. Let
N0=(2E(C−1)δ)1p. | (4.38) |
Then from Lemma 4.2, we have
‖H(→fδn−→gN0)‖0,Λ≤3C+12δ, | (4.39) |
‖R(→fδn−→gN0)‖l2≤C+12e(2EC−1)1pδ−1pδ. | (4.40) |
Thus, by using Lemma 4.4, there exists a constant M
‖H(→fδn−→gN0)‖0,Ω≤M. | (4.41) |
Then
‖H→fδn−g‖p,Ω≤‖H(→fδn−→gN0)‖0,Ω+‖g−gN‖p,Ω≤‖H(→fδn−→gN0)‖0,Ω+‖g‖p≤M+E. | (4.42) |
Moreover, by using (2.10), (3.7) and the triangle inequality
‖H→fδn−g‖0,Ω≤‖H→fδn−gδ‖0,Ω+‖gδ−g‖0,Ω≤‖H→fδn−gδ‖0,Λ+‖gδ−g‖0,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 xi∈Λ(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)),(0≤j1,j2≤n) being the Hermite-Gauss type interpolation points and ρ(j) are the corresponding Hermite-Gauss weights. For f,g∈L2(R2), we define the discrete inner product
⟨f,g⟩n:=n∑j1=1n∑j2=1ρ(j)f(σ(j))¯g(σ(j)), |
and
˘Hℓ(ω):=(−1)|ℓ|1e|ω|Hℓ(ω). |
Let
Hn=span{H(0,0)(x),H(1,0)(x),H(1,1)(x),…,H(n,n−1)(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|ℓ|1+1,|k|1+1=∑mi=1Hℓ(xi)Hk(xi),R|ℓ|1+1,|k|1+1=⟨˘Hℓ,˘Hk⟩n,Hi,|ℓ|1+1=Hℓ(xi),i=1,2,…,m;|ℓ|1,|k|1=0,1,…n. |
With these preparations, the discrete form of the implicit iteration method can be given as
fδ0=0,fk=fk−1−(HTH+βkR)−1HT(Hfδk−1−gδ),k=1,2,…,n. | (5.1) |
Suppose that
(m∑i=1(gδ(xi)−g(xi)2)1/2≤δ. | (5.2) |
Similar to what is done in [27], we take β1=1,βk=qk−1β1 with some q<1 and choose n as the first integer for which
‖Hfδn−gδ‖≤C1δ<‖Hfδk−gδ‖,0≤k<n, | (5.3) |
and then adjust the parameter βn such that
C2δ≤‖Hfδn−gδ‖≤C1δ, | (5.4) |
where C1,C2 are two constants that obey 1≤C1≤C2.
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+x22≤1} is a disk and scatter nodes are given as Figure 2a. We choose the exact function as g(x)=(x21+x22−2)3 and set ϵ=0.01. The numerical results are exhibited in Figures 2c–3h.
Example 6.2. [18] Let Λ={(x1,x2)|0≤x1≤1,0≤x2≤π} is a rectangle and scatter nodes are given as Figure 3i. We choose the exact function as g(x)=(x21+x22−2)3 and set ϵ=0.01. We have shown the numerical results in Figures 3k–4h.
Example 6.3. [19] Let Λ={(x1,x2)|−1≤x1≤3,−1≤x2≤3}, 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+1−exp(x2)). We have shown the numerical results in Figures 4i–5h with ϵ=0.01.
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(x1⋅x2) 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
![]() |
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 |