Fast Higher-Order Numerical Schemes for Two-Dimensional Nonlinear Riesz Space Fractional Diffusion Equations
doi: 10.11916/j.issn.1005-9113.25041
Shuyu Yue , Tao Wang
School of Mathematical Sciences, University of Jinan, Jinan 250022, China
Funds: Sponsored by the National Natural Science Foundation of China(Grant No.11801216),Natural Science Foundation of Shandong Province(Grant No.ZR2017BA010).
Abstract
Fast fourth-order linearized alternating direction implicit (ADI )and locally one-dimensional (LOD)schemes are presented to solve two-dimensional nonlinear riesz space fractional diffusion equations (RSFDEs). The proposed schemes employ the Crank-Nicolson method for temporal discretization,a quartic approximation for the Riesz derivative,and an explicit treatment for linearizing the nonlinear term. Through a rigorous discrete energy analysis,we derive explicit error estimates,proving that both schemes achieve second-order accuracy in time and fourth-order accuracy in space. The ADI and LOD methods are employed to decompose two-dimensional problems into a collection of symmetric positive definite (SPD)Toeplitz systems,for which the fast sine transform can be utilized to mitigate computational complexity. The computational treatment of these subsystems is further conducted using a preconditioned conjugate gradient (PCG )approach that leverages sine transforms. Theoretical analysis demonstrates that the preconditioned matrix can be decomposed into the identity matrix,a matrix of small norm,and a low-rank matrix. The effectiveness and efficiency of the scheme are validated through numerical examples.
0 Introduction
Considerable attention has been focused on space fractional differential equations(SFDEs)in recent decades due to their extensive applications in diverse fields,such as fluid mechanics[1],physics[2],and finance[3-4]. However,the nonlocality inherent in fractional calculus means that closed-form solutions to these equations are rarely attainable,making the development of robust numerical methods essential.
Consequently,numerous numerical schemes have been proposed for solving fractional diffusion equations[5-13]. To date,two principal methodologies have been widely adopted for the numerical approximation of Riesz fractional derivatives. The first approach,introduced by Ortigueira[ 14],is based on the fractional centered difference scheme. This method was later applied by Çelik and Duman[15]to riesz space fractional diffusion equations(RSFDEs),demonstrating that it achieves second-order convergence.The second major methodology originates from the work of Meerschaert and Dadjeran[5],who proposed a shifted Grünwald-Letnikov approximation for the Riemann-Liouville(RL)derivative. This foundation subsequently facilitated the development of the weighted shifted Grünwald-Letnikov difference(WSGD)scheme for solving RSFDEs[16]. Further advancing this line of research,Zhou et al.[17]constructed a quasi-compact discretization scheme with third-order accuracy. More recently,building on the WSGD approach,Hao et al.[18]introduced a higher-order scheme attaining fourth-order precision.
However,dense coefficient matrices are generated in the discretized systems because of the non-local nature of fractional derivatives,thereby imposing heavy computational and memory burdens on solving the resulting linear systems. With the development of numerical methods,several robust preconditioners have been proposed to tackle this issue in Toeplitz systems. For instance,Yu et al.[19]formulated preconditioned conjugate gradient(PCG)methods based on T. Chan[20]and R. Chan's[21]circulant preconditioners to address the challenges of solving Riesz distributed-order nonlinear SFDEs.Meanwhile,utilizing the approximate inverse preconditioning technique based on the inverse of R.Chan's[21]circulant matrix,Tang et al.[22]developed an efficient second-order numerical method for solving SFDEs with variable coefficients. The preconditioning approach incorporating the τ-preconditioner has emerged as a promising method for solving SFDEs[23-25].
When it comes to modeling and simulating multidimensional systems,the numerical discretization of partial differential equations,which is fundamental to the process,proves to be computationally expensive. By reducing multidimensional partial differential equations to a sequence of one-dimensional problems,alternating direction implicit(ADI)and locally one-dimensional(LOD)methods offer a simplified approach[26-30]. Recently,Tang and Huang[31]employed the ADI method to decompose the discretized linear system into multiple subsystems,which yielded first-order accuracy in space. They then combined the generalized minium residual(GMRES)method with the constructed large language model-augmented semantic digital twins(LSDTs)τ-preconditioner to solve these subsystems. Chen et al.[32]developed a second-order LOD multigrid method for multidimensional RSFDEs. Combining the backward differentiation formula(BDF)method with a fourth-order fractional compact operator,Hu and Cao[33]constructed a high-accuracy ADI scheme for2D nonlinear RSFDEs. However,the development of a fast algorithm was not addressed. The scarcity of high-order fast algorithms based on the ADI and LOD methods for multidimensional RSFDEs has motivated our research in this direction.
In this study,we investigate the following 2D nonlinear RSFDEs:
{∂u∂t=Kβ∂βu∂|x|β+Kγ∂γu∂|y|γ+f(u)+p,(x,y)∈𝛺,t∈(0,T]u(x,y,0)=φ(x,y),(x,y)∈𝛺u(x,y,t)=0,(x,y)∈∂𝛺,t∈(0,T]
(1)
where p and φ are sufficiently smooth in their respective domains. β,γ ∈(1,2)and Ω=(0,L1)×(0,L2),L1,L2and T are positive constants. Kβ,Kγ>0,f(u)is a continuous function and satisfies the Lipschitz condition. Moreover,∂βu∂|x|β and ∂γu∂|y|γ denote the Riesz fractional derivative[23,25].
In this study,fourth-order linearized ADI and LOD schemes are proposed to solve 2D nonlinear problem(1). Both proposed schemes employ the Crank-Nicolson method for temporal discretization and a fourth-order approximation[18]for the Riesz derivative,along with explicit linearization of the nonlinear terms. Both schemes are proven to achieve second-order temporal accuracy and fourth-order spatial accuracy,as established through explicit error estimates derived from a discrete energy analysis. The ADI and LOD methods are employed to decompose two-dimensional problems into a collection of symmetric positive definite(SPD)Toeplitz systems,for which the fast sine transform can be utilized to mitigate computational complexity. The solution of these subsystems is achieved through the application of the PCG method preconditioned by the τ-matrix approximation. In addition,theoretical analysis confirms that the τ-preconditioner matrix exhibits eigenvalues tightly clustered around unity,while numerical results substantiate the extended τ-preconditioner's remarkable efficacy in accelerating convergence for two-dimensional problems.
The study is structured as follows. In Section 1,linearized fourth-order ADI and LOD schemes are developed for the nonlinear RSFDEs. The PCG method with a τ-preconditioner is presented in Section 2 to accelerate the solution of discrete linear systems,and the spectral characteristics of the preconditioned matrices are analyzed. In Section 3,numerical examples are conducted to demonstrate the convergence behavior and computational efficiency of the fast numerical schemes incorporating preconditioners. A summary of key findings is offered in Section 4.
1 Linearized Numerical Scheme
1.1 Derivation of the Numerical Scheme
Take positive integers M1,M2 and N. Denote Δx=L1/(M1+1),Δy=L2/(M2+1) and Δt=T/N,Let[[k0,k1]]={k∣k0⩽k⩽k1,and k∈N},xi=iΔx(i∈[[0,M1+1]]),yj= iΔy(j ∈ [[0,M2+1]]),tk= kΔt (k ∈ [[0,N]]),Ω¯h{(xi,yi)| i ∈[[0,M1+ 1]],j ∈ [[0,M2+ 1]]},𝛺h=𝛺¯h∩𝛺,∂𝛺h=𝛺¯h∩∂𝛺,ω={(i,j)∣(xi,yj)∈ 𝛺h},∂ω={(i,j)∣(xi,yj)∈∂𝛺h},ω¯= ω∪∂ω,ΩΔt={t k ∣0≤k ≤ N},Vh={u ∣ u={ui,j ∣(i,j)∈ ω¯}}.
Given any discrete function u ∈ Vh,we define
  ui,jk+12=12(ui,jk+1+ui,jk),δtui,jk+12=1Δt(ui,jk+1−ui,jk)
Hxβui,j={d2βui−1,j+(1−2d2β)ui,j+d2βui+1,j,i∈[[1,M1]],j∈[[0,M2+1]]ui,j,i=0,M1+1,j∈[[0,M2+1]]
Hyγui,j={d2γui,j−1+(1−2d2γ)ui,j+d2γui,j+1,j∈[[1,M2]],i∈[[0,M1+1]]ui,j,j=0,M2+1,i∈[[0,M1+1]]
where
d2β=−β2+β+424∈(112,16)
d2γ=−γ2+γ+424∈(112,16)
Define
Ui,jn=u(xi,yj,tn),pi,jn=p(xi,yj,tn)pi,jn+12=12(pi,jn+1+pi,jn),(i,j)∈ω¯,n∈[[0,N−1]] 
Let
{w^0(α)=α2+3α+212g0(α)w^1(α)=α2+3α+212g1(α)+4−α26g0(α)w^k(α)=α2+3α+212gk(α)+4−α26gk−1(α)+α2−3α+212gk−2(α)
where k ≥ 2,gk(α)=(−1)k(αk).
Lemma 1​[18]    Assuming α​ ∈ (1,2) , the coefficients w^k(α)​satisfy
{w^0(α)>0,w^1(α)⩽0,w^k(α)⩾0,k⩾3∑k=0∞w^k(α)=0,∑k=0mw^k(α)⩽0,m⩾2w^0(α)+w^2(α)⩾0
Let
Wxβu=∂βu∂|x|β,Wyγu=∂γu∂|y|γ
Define the following difference operator:
δxβui,jn=−12cos⁡(βπ/2)Δx−β.(∑k=0i+1w^k(β)ui−k+1,jn+∑k=0M1−i+2w^k(β)ui+k+1,jn)
δyγui,jn=−12cos⁡(γπ/2)Δy−γ.(∑k=0j+1w^k(γ)ui,j−k+1n+∑k=0M2−j+2w^k(γ)ui,j+k+1n)
Lemma 2​[34]   For any u​(x​,y​,·) ∈ C3​[0,T​] ,we have
12(∂u(xi,yj,tn+1)∂t+∂u(xi,yj,tn)∂t)=δtUi,jn+12+O(Δt2)
Denote
ℓ 4 + α ( R ) = { f ∣ f ∈ L 1 ( R ) ∫ − ∞ ∞ ( 1 + | k | ) 4 + α | F ( k ) | d k < ∞ } , α = β , γ
where F(k)is represented as the Fourier transformation of f(t).
Lemma 3​[18]    Let u​(·,y​,t​) ∈ℓ4+β​( ℝ​) , and u​(x​,·,t​) ∈ ℓ4+γ​( ℝ​) , we have
HxβWxβu(xi,yj,tn)=δxβUi,jn+O(Δx4)
HyγWyγu(xi,yj,tn)=δyγUi,jn+O(Δy4)
Lemma 4​    Let f​(u​) ∈ C2​(R​) , then we have
{f(Un+1)+f(Un)2=3f(Un)−f(Un−1)2+O(Δt2),n⩾1f(U1)+f(U0)2=f(U0)+O(Δt)
Proof:​ By employing a Taylor series expansion,the correctness of the formula can be directly validated.
Building upon the foregoing lemmas,we now proceed to derive high-order numerical schemes for solving problem(1),under the assumption that the solution u(x,y,t)adheres to the conditions stipulated in Lemmas 2 and 3. Considering the main equation in problem(1)at(xi,yj,tk),and averaging two adjacent time layers,we have
12(∂u(xi,yj,tn+1)∂t+∂u(xi,yj,tn)∂t)=12Kβ(Wxβu(xi,yj,tn+1)+Wxβu(xi,yj,tn))+12Kγ(Wyγu(xi,yj,tn+1)+Wyγu(xi,yj,tn))+12(f(uijn+1)+f(uijn))+pijn+12,(i,j)∈ω,n∈[[0,N−1]]
(2)
ApplyingHxβHyγ to Eq.(2)and using Lemmas 2-4,we have
HxβHyγδtUi,jn+12=KβHyγδxβUi,jn+12+KγHxβδyγUi,jn+12+32HxβHyγf(Ui,jn)−12HxβHyγf(Ui,jn−1)+HxβHyγpi,jn+12+(R1)i,jn+12,(i,j)∈ω,n∈[[0,N−1]]
(3)
where Ui,j−1=Ui,j0,and
{(R1)i,jn+12=O(Δt2+Δx4+Δy4),n⩾1(R1)i,j12=O(Δt+Δx4+Δy4)
Adding KβKγΔt24δxβδyγδtUi,jn+12 to both sides of Eq.(3)gives
HxβHyγδtUi,jn+12+KβKγΔt24δxβδyγδtUi,jn+12=KβHyγδxβUi,jn+12+KγHxβδyγUi,jn+12+32HxβHyγf(Ui,jn)−12HxβHyγf(Ui,jn−1)+HxβHyγpi,jn+12+Ri,jn+12,(i,j)∈ω,n∈[[0,N−1]]
(4)
where
{Ri,jn+12=O(Δt2+Δx4+Δy4),n⩾1Ri,j12=O(Δt+Δx4+Δy4)
(5)
By omitting the small terms in Eq.(4)and substituting the exact solution Ui,jn with an approximated solution Ui,jn,the following linearized numerical scheme for solving problem(1)is obtained
{HxβHyγδtui,jn+12+KβKγΔt24δxβδyγδtui,jn+12=KβHyγδxβui,jn+12+KγHxβδyγui,jn+12+HxβHyγgi,jn+12(i,j)∈ω,n∈[[0,N−1]]ui,j0=φ(xi,yj),(i,j)∈ωui,jn=0,(i,j)∈∂ω,n∈[[0,N−1]]
(6)
where gi,jn+12=32f(ui,jn)−12f(ui,jn−1)+pi,jn+12.
Multiplying both sides of the main equation in Eq.(6)by Δt yields the resulting form
(Hxβ−KβΔt2δxβ)(Hyγ−KγΔt2δyγ)ui,jn+1=(Hxβ+KβΔt2δxβ)(Hyγ+KγΔt2δyγ)ui,jn+ΔtHxβHyγgi,jn+12
By introducing the intermediate variable vi,j∗,we have the following linearized ADI scheme:
{(Hxβ−Δt2δxβ)vi,j∗=(Hxβ+Δt2δxβ)⋅(Hyγ+Δt2δyγ)ui,jn+ΔtHxβHyγgi,jn+12(Hyγ−Δt2δyγ)ui,jn+1=vi,j∗
(7)
We now develop a linearized LOD scheme.Adding small term
KβKγΔt24δxβδyγ(32f(Ui,jn)−12f(Ui,jn−1)+pi,jn+12)
to both sides of Eq.(4)yields
HxβHyγδtUi,jn+12+KβKγΔt24δxβδyγδtUi,jn+12=KβHyγδxβUi,jn+12+KγHxβδyγUi,jn+12+32HxβHyγf(Ui,jn)−12HxβHyγf(Ui,jn−1)+HxβHyγpi,jn+12+KβKγΔt24δxβδyγ.
(8)
(32f(Ui,jn)−12f(Ui,jn−1)+pi,jn+12)+R¯i,jn+12,(i,j)∈ω,n∈[[0,N−1]]
where
{R¯ijn+12=O(Δt2+Δx4+Δy4),n⩾1R¯ij12=O(Δt+Δx4+Δy4)
(9)
By omitting the small terms in Eq.(8)and substituting the exact solution Ui,jnwith an approximated solutionUi,jn,the following linearized numerical scheme for solving problem(1)is obtained:
{HxβHyγδtui,jn+12+KβKγΔt24δxβδyγδtui,jn+12=KβHyγδxβui,jn+12+KγHxβδyγui,jn+12+HxβHyγgi,jn+12+KβKγΔt24δxβδyγgi,jn+12(i,j)∈ω,n∈[[0,N−1]]ui,j0=φ(xi,yj),(i,j)∈ωui,jn=0,(i,j)∈∂ω,n∈[[0,N−1]]
(10)
We multiply both sides of the main equation in Eq.(10)by Δt and reformulate it as
(Hxβ−KβΔt2δxβ)(Hyγ−KγΔt2δyγ)ui,jn+1=(Hxβ+KβΔt2δxβ)(Hyγ+KγΔt2δyγ)ui,jn+ΔtHxβHyγgi,jn+12+KβKγΔt34δxβδyγgi,jn+12
By introducing the intermediate variable ui,j∗,we have the following linearized LOD scheme:
{(Hxβ−KβΔt2δxβ)ui,j∗=(Hxβ+KβΔt2δxβ)ui,jn+Δt2(Hxβ+KβΔt2δxβ)gi,jn+12(Hyγ−KγΔt2δyγ)ui,jn+1=(Hyγ+KγΔt2δyγ)ui,j∗+Δt2(Hyγ−KγΔt2δyγ)gi,jn+12
(11)
Let
μβ=KβΔt4Δxβcos⁡(βπ/2),μγ=KγΔt4Δyγcos⁡(γπ/2)
and we assume that |μβ|and |μγ|share a common upper bound,i.e.,
max { | μ β | , | μ γ | } < μ
(12)
where μ is a real number.
Let
sk(α,±)={−2d2α±2μαw^1(α),k=0d2α±μα(w^0(α)+w^2(α)),|k|=1±μαw^k+1(α),|k|∈[[2,M−1]]
(13)
where(M,α)=(M1,β)or(M2,γ).
Define
Bα±=I+S±(α),α=β,γ
(14)
where
S±(α)=[s0(α,±)s−1(α,±)⋯s2−M(α,±)s1−M(α,±)s1(α,±)s0(α,±)⋯s3−M(α,±)s2−M(α,±)⋮⋮⋮⋮sM−2(α,±)sM−3(α,±)⋯s0(α,±)s−1(α,±)sM−1(α,±)sM−2(α,±)⋯s1(α,±)s0(α,±)]
(15)
where(M,α)=(M1 ,β)or(M2,γ).
Clearly,Bα±and S±(α)±are the symmetric Toeplitz(ST)matrices.
We define the following M-order symmetric tridiagonal matrices:
C(α)=tridiag(d2α,1−2d2α,d2α)
where(M,α)=(M1,β)or(M2 ,γ).
Define
un=[ui,jn]M1×M2,gn+12=[gi,jn+12]M1×M2
U―n=Bβ−[[Bγ−]T[un]T]T=[U¯i,jn]M1×M2
Gn+12=C(β)[[C(γ)]T[gn+12]T]T=[Gi,jn+12]M1×M2
U―Δx,jn=[U¯1,jn,U¯2,jn,⋯,U¯M1,jn]T
GΔx,jn+12=[G1,jn+12,G2,jn+12,⋯,GM1,jn+12]T
vΔx,j∗=[u1,j∗,u2,j∗,⋯,uM1,j∗]T
vi,Δy∗=[ui,1∗,ui,2∗,⋯,ui,M2∗]T
uΔx,jn=[u1,jn,u2,jn,⋯,uM1,jn]T
ui,Δyn+1=[ui,1n+1,ui,2n+1,⋯,ui,M2n+1]T
uΔx,j∗=[u1,j∗,u2,j∗,⋯,uM1,j∗]T
ui,Δy∗=[ui,1∗,ui,2∗,⋯,ui,M2∗]T
gΔx,jn+12=[g1,jn+12,g2,jn+12⋯,gM1,jn+12]T
gi,Δyn+12=[gi,1n+12,gi,2n+12⋯,gi,M22n+12]T
We then derive the matrix form of the linearized ADI scheme as follows:
{Bβ+vΔx,j∗=U―Δx,jn+ΔtGΔx,jn+12,j=1,2,⋯,M2Bγ+ui,Δyn+1=vi,Δy∗,i=1,2,⋯,M1
(16)
Let
σΔx,j=[w^2(β)ρ0,jnw^3(β)ρ0,jn⋮w^M1(β)ρ0,jnw^M1+1(β)ρ0,jn]+[w^M1+1(β)ρM1+1,jnw^M1(β)ρM1+1,jn⋮w^3(β)ρM1+1,jnw^2(β)ρM1+1,jn]+[w^0(β)ρ0,jn0⋮0w^0(β)ρM1+1,jn]+[d2βϱ0,jn0⋮0d2βϱM1+1,jn]
σi,Δy=[w^2(γ)ξi,0nw^3(γ)ξi,0n⋮w^M2(γ)ξi,0nw^M2+1(γ)ξi,0n]+[w^M2+1(γ)ξi,M2+1nw^M2(γ)ξi,M2+1n⋮w^3(γ)ξi,M2+1nw^2(γ)ξi,M2+1n]+[w^0(γ)ξi,0n0⋮0w^0(γ)ξi,M2+1n]+[d2γζi,0n0⋮0d2γζi,M2+1n]
where
ρk,jn=−μβ(Δt2gk,jn+12+uk,jn+uk,j∗)
ϱk,jn=Δt2gk,jn+12+uk,jn−uk,j∗,k=0,M1+1
ξi,kn=μγ(Δt2gi,kn+12−ui,kn+1−ui,k∗)
ζi,kn=Δt2gi,kn+12−ui,kn+1+ui,k∗,k=0,M2+1
Then,we arrive at the following matrix form of the linearized LOD scheme:
{Bβ+uΔx,j∗=Bβ−uΔx,jn+Δt2Bβ−gΔx,jn+12+σΔx,j,j=0,1,⋯,M2+1Bγ+ui,Δyn+1=Bγ−ui,Δy∗+Δt2Bγ+gi,Δyn+12+σi,Δy,i=1,2,⋯,M1
(17)
1.2 Error Estimate
Let Let V∘h={u∣u∈Vh;if(i,j)∈ ∂ω,ui,j= 0}.Given any discrete functions u, v ∈ V∘h,we define(u,v)= ΔxΔy∑i=1M1∑j=1M2ui,jvi,j,‖u‖=(u,u)
SinceHxβandHyγare positive definite and self-adjoint,we haveHxβ=(Qx)2andHyγ=(Qy)2 [18].
Lemma 5​[18]   For any grid function u​ ∈V∘h​, we have
19‖u‖2⩽(HxβHyγu,u)=(QxQyu,QxQyu)⩽‖u‖2,(δxβu,u)⩽0,(δyγu,u)⩽0
According to the definitions of δxβ and δyγ,it is straightforward to show that both-δxβ and-δyγ are symmetric positive definite operators. Therefore,there exist fractional symmetric positive definite difference operators Λx and Λy such that the following conclusions hold[18]:
(δxβu,u)=−(Λxu,Λxu)⩽0
(δyγu,u)=−(Λyu,Λyu)⩽0
(δxβδyγu,u)=(ΛxΛyu,ΛxΛyu)
Lemma 6​    For any grid function u​ ∈ V∘h​ , we have
((Hyγδxβ+Hxβδyγ)u,u)⩽0
Proof:​ By Lemma 5, we get
(Hyγδxβu,u)=(δxβQyu,Qyu)⩽0
  (Hxβδyγu,u)=(δyγQxu,Qxu)⩽0
The proof is completed.
Lemma 7​[30]    Assume that {ςn​} and {ηn​} are nonnegative sequences, and that the sequence {θn​}satisfies
ϕ0⩾0,θ0⩽ϕ0θn⩽ϕ0+∑l=0n−1ηl+∑l=0n−1ςlθl,n⩾1
Then it holds that
θn⩽(ϕ0+∑l=0n−1ηl)exp⁡(∑l=0n−1ςl),n⩾1
We begin with an analysis of the stability and convergence of the linearized ADI scheme in Eq.(7).Let ei,jn=Ui,jn−ui,jn,by subtracting the master equation in Eq.(6)from Eq.(4),we get
HxβHyγδtei,jn+12+KβKγΔt24δxβδyγδtei,jn+12=KβHyγδxβei,jn+12+KγHxβδyγei,jn+12+32HxβHyγ(f(Ui,jn)−f(ui,jn))−12HxβHyγ(f(Ui,jn−1)−f(ui,jn−1))+Ri,jn+12,(i,j)∈ω,n∈[[0,N−1]]
(18)
Theorem 1    ​Let  ei,jn​be the solution of Eq.(18)and Δt​ < 1 / (18(L​ +12​)) , we have
‖en‖⩽c(Δt2+Δx4+Δy4),n∈[[1,N]]
where c is a constant independent of Δx,Δy and Δt,and L is the Lipschitz constant of the function f(u).
Proof:​ Forming the inner product of Eq.(18)with en+12​ gives
(HxβHyγδten+12,en+12)+KβKγΔt24.(δxβδyγδten+12,en+12)=(KβHyγδxβen+12,en+12)+(KγHxβδyγen+12,en+12)+(32HxβHyγ(f(Un)−f(un)),en+12)−(12HxβHyγ(f(Un−1)−f(un−1)),en+12)+(Rn+12,en+12),(i,j)∈ω,n∈[[0,N−1]]
(19)
Let En=(HxβHyγen,en)=‖QxQyen‖2,Fn=(δxβδyγen,en)=‖ΛxΛyen‖2,we have
(HxβHyγδten+12,en+12)=12Δt(En+1−En)
(20a)
(δxβδyγδten+12,en+12)=12Δt(Fn+1−Fn)
(20b)
When n ≥ 1,an application of Lemma 5,coupled with the Lipschitz condition and the Schwarz inequality,leads to
(32HxβHyγ(f(Un)−f(un)),en+12)⩽(32HxβHyγ|f(Un)−f(un)|,|en+12|)⩽3L2(HxβHyγ|en|,|en+12|)⩽3L4((HxβHyγ|en|,|en+1|)+(HxβHyγ|en|,|en|))⩽9L8‖en‖2+3L8‖en+1‖2
(21)
By the same reasoning,we obtain
−(12HxβHyγ(f(Un−1)−f(un−1)),en+12)⩽L2(HxβHyγ|en−1|,|en+12|)⩽L4((HxβHyγ|en−1|,|en+1|)+(HxβHyγ|en−1|,|en|))⩽L8(2‖en−1‖2+‖en‖2+‖en+1‖2)
(22)
Inserting Eq.(20)and inequalities(21)-(22)into Eq.(19),and using Lemma 6,yields
En+1+KβKγΔt24Fn+1⩽En+KβKγΔt24Fn+LΔt2‖en−1‖2+5LΔt2‖en‖2+LΔt‖en+1‖2+2Δt(Rn+12,en+12)
Therefore,for all n ≥ 2,
En+KβKγΔt24Fn⩽E1+KβKγΔt2F14+LΔt2∑m=1n−1‖em−1‖2+5LΔt2∑m=1n−1‖em‖2+LΔt∑m=1n−1‖em+1‖2+2Δt∑m=1n−1(Rm+12,em+12)⩽E1+KβKγΔt2F14+4LΔt∑m=1n−1‖em‖2+LΔt‖en‖2+Δt∑m=1n−1(‖Rm+12‖2+‖em+12‖2)⩽E1+KβKγΔt2F14+4LΔt∑m=1n−1‖em‖2+LΔt‖en‖2+Δt2∑m=1n−1(‖em+1‖2+‖em‖2)+Δt∑m=1n−1‖Rm+12‖2⩽E1+KβKγΔt2F14+(4L+1)Δt∑m=1n−1‖em‖2+(L+12)Δt‖en‖2+Δt∑m=1n−1‖Rm+12‖2
(23)
Upon taking the inner product of Eq.(18)with e12 for the case n= 0,and noting the identity f(U0)-f(u0)= 0,we establish that
(HxβHyγδte12,e12)+(KβKγΔt24δxβδyγδte12,e12)=(KβHyγδxβe12,e12)+(KγHxβδyγe12,e12)+(R12,e12),(i,j)∈ω
(24)
Using Eq.(20)and Lemma 6,we obtain
E1+KβKγΔt24F1⩽2Δt(R12,e1)⩽2Δt‖R12‖‖e1‖
(25)
It follows from Lemma 5,the definition of E1,and inequality(25)that
‖e1‖⩽18Δt‖R12‖
(26)
Substituting inequality(26)into(25),we obtain
E1+KβKγΔt24F1⩽36Δt2‖R12‖2
(27)
Inserting inequality(27)into(23)yields
En+KβKγΔt24Fn⩽36Δt2‖R12‖2+(4L+1).
Δt∑m=1n−1‖em‖2+(L+12)Δt‖en‖2+Δt∑m=1n−1‖Rm+12‖2
Using Lemma 5 and the definition of En,we get
(1−9(L+12)Δt)‖en‖2⩽324Δt2‖R12‖2+
9(4L+1)Δt∑m=1n−1‖em‖2+9Δt∑m=1n−1‖Rm+12‖2
When Δt < 1/(18(L+12)),
‖en‖2⩽648Δt2‖R12‖2+18(4L+1).
Δt∑m=1n−1‖em‖2+18Δt∑m=1n−1‖Rm+12‖2
By Lemma 7,we get
‖en‖2⩽exp⁡(18(4L+1)T)(648Δt2‖R12‖2+18Δt∑m=1n−1‖Rm+12‖2)
From Eqs.(5)and(26),we obtain the required result immediately.
Remark 1​    The stability of the linearized ADI scheme (7) can be readily established in a similar manner.
In what follows,we focus exclusively on the error estimates of the LOD scheme for linear equations. We consider the linear version of problem(1),which is formulated by discarding the nonlinear term. For this linear equation,the compact LOD scheme is given by
{HxβHyγδtui,jn+12+KβKγΔt24δxβδyγδtui,jn+12=KβHyγδxβui,jn+12+KγHxβδyγui,jn+12+HxβHyγpi,jn+12+KβKγΔt24δxβδyγpi,jn+12,(i,j)∈ω,n∈[[0,N−1]]ui,j0=φ(xi,yj),(i,j)∈ωui,jn=0,(i,j)∈∂ω,n∈[[0,N−1]]
(28)
Let ei,jn=Ui,jn−ui,jn,by subtracting the master equation in Eq.(28)from Eq.(8)and neglecting the nonlinear terms,we get
(29)
where= O(Δt2+ Δx4+ Δy4),n ≥ 0.
Theorem 2​    Let ei,jn​ be the solution of Eq.(29)and Δt​ <19​, we have
‖en‖⩽c(Δt2+Δx4+Δy4),n∈[[1,N]]
where c is a constant independent of Δx,Δy and Δt.
Proof:​ By forming the inner product of Eq.(29)with en+12​, we get
Let En= (HxβHyγen,en)=‖QxQyen‖2,Fn=(δxβδyγen,en)=‖ΛxΛyen‖2. By Lemma 6,we obtain
Therefore,for all n ≥ 1,
By Lemma 5 and the definitions of Enand Fn,we get
When Δt< 1/ 9,
By Lemma 7,we get
From the definition of ,the needed result follows immediately.
Remark 2​    The stability of the LOD scheme(28) can be easily proved through an analogous approach.
2 PCG Method
The CG method is a well-known and efficient technique for solving linear systems(16)and(17).However,experiments indicate that the coefficient matrix has a widely distributed eigenvalue spectrum,which significantly degrades the convergence rate of the standard CG method. To address this issue,this section introduces a PCG approach that utilizes theτ-preconditioner to accelerate convergence for these systems. Next,we introduce a τ-preconditioner derived from the sine transform.
For any Toeplitz matrix
SM=[s0s−1⋯s2−Ms1−Ms1s0⋯s3−Ms2−M⋮⋮⋮⋮sM−2sM−3⋯s0s−1sM−1sM−2⋯s1s0]
the generating function of {SM}M=1∞can be defined as
p(θ)=∑k=−∞∞skexp⁡(ikθ)
where p(θ)is in the Wiener class if and only if ∑ k = − ∞ ∞ | s k | < ∞ [25]. In particular,if sk=s−k,then SM is symmetric,and we denote it as S~M. Let HM be a Hankel matrix with antidiagonal entries
[s2,s3,⋯,sM−1,0,0,0,sM−1,⋯,s3,s2]
Then,the τ-matrix τ(S~M)can be described by τ(S~M)= S~M-HM[25]. Hence,we propose the τ-preconditioners of the linear systems(16)and(17)as
τ(Bα+)=Bα+−HM
(30)
The dominant computational overhead of PCG when applied to linear systems(16)and(17)arises from evaluating Bα−v and (τ(Bα+))−1v. Leveraging the Toeplitz property of Bα− and the diagonalizability of(τ(Bα+))−1,these matrix-vector multiplications achieve an optimal complexity of O(Mlog(M))(M1=M2= M)[25].
Lemma 8​[35]   If TM​ is a ST matrix with the first column being [t0,t1,⋯,tM−1]T​. Then, the eigenvalues of τ​(TM​) can be given by
λm=t0+2∑k=1M−1(tkcos⁡(πmkM+1)),m∈[[1,M]]
Lemma 9   ​ τ​(Bα+​) in Eq. ( 30) is SPD, and ‖ ( τ ( B α + ) ) − 1 ‖ 2 < 3 ​,(M​,α​) = (M1​ ,β​) or (M2​ ,γ​).
Proof:​ Since Bα+​is an ST matrix and τ​(Bα+​) is defined by Eq. ( 30 ), it follows that τ​(Bα+​) is a symmetric matrix. We apply Lemma 8 and Lemma 1while taking into account μα​< 0 and d2α∈(112,16)​to obtain
λm(τ(Bα+))=1−2d2α+2d2αcos⁡(πmM+1)+2μα(w^1(α)+(w^0(α)+w^2(α))cos⁡(πmM+1)+∑k=3Mw^k(α)cos⁡(πm(k−1)M+1))⩾1−2d2α+2d2αcos⁡(πmM+1)+2μα(w^1(α)+(w^0(α)+w^2(α))+∑k=3Mw^k(α))=1−2d2α+2d2αcos⁡(πmM+1)+2μα∑k=0Mw^k(α)>1−4d2α>13
Therefore,the matrix τ(Bα+)is SPD,and we can further obtain that
‖ ( τ ( B α + ) ) − 1 ‖ 2 = max 1 ⩽ m ⩽ M | λ m ( ( τ ( B α + ) ) − 1 ) | = 1 min 1 ⩽ m ⩽ M λ m ( τ ( B α + ) ) < 3
This completes the proof.
Lemma 10​    Let S+(α)​ be defined by Eq. ( 15).Under the assumption (12), the generating functions of {S+(α)}M=1∞​( (M​,α​) = (M1​ ,β​) or (M2​ ,γ​) ) are in the Wiener class.
Proof:​ Applying Lemma 1, and noting that d2α∈(112,16)​, we deduce that
∑ k = − ∞ ∞ | s k ( α , + ) | = | 2 μ α w ^ 1 ( α ) − 2 d 2 α | + 2 | μ α ( w ^ 0 ( α ) + w ^ 2 ( α ) ) + d 2 α | + 2 ∑ k = 3 ∞ | μ α w ^ k ( α ) | ⩽ | 2 μ α w ^ 1 ( α ) | + | 2 d 2 α | + 2 | μ α ( w ^ 0 ( α ) + w ^ 2 ( α ) ) | + 2 | d 2 α | + 2 | μ α | ∑ k = 3 ∞ | w ^ k ( α ) | = 4 d 2 α + 4 μ α w ^ 1 ( α ) < ∞
This completes the proof.
Lemma 11​[35]    If the generating function of {S+(α)}M=1∞​is in the Wiener class, then for any ε​ > 0,there exists Mα​> 0 such that for all M​ > Mα​ ,
S+(α)−τ(S+(α))=UM′+VM′
where ‖ U M ′ ‖ 2 < ε 3 and rank(VM′)⩽Mα.
Theorem 3​    Let Bα+​and τ​(Bα+​) be defined by Eqs. ( 14 ) and ( 30 ), respectively. Under the assumption ( 12), and then for any ε​ > 0, there exists Mα​> 0 such that for all M​ >Mα​ ,
(τ(Bα+))−1Bα+−I=UM+VM
where ‖ U M ‖ 2 < ε and rank(VM)≤ Mα,(M,α)=(M1,β)or(M2,γ).
Proof:  ​ From the definition of Bα+​and Lemmas10-11, we get
Bα+−τ(Bα+)=I+S+(α)−τ(I+S+(α))=S+(α)−τ(S+(α))=UM′+VM′
(31)
where
‖ U M ′ ‖ 2 < ε 3 , rank ( V M ′ ) ⩽ M α
(32)
Multiplying both sides of Eq.(31)by(τ(Bα+))−1,we have
(τ(Bα+))−1Bα+−I=(τ(Bα+))−1UM′+(τ(Bα+))−1VM′=UM+VM
where UM=(τ(Bα+))−1UM′ and VM=(τ(Bα+))−1VM′.
Applying Lemma 9 and inequality(32),we have
‖ U M ‖ 2 ⩽ ‖ ( τ ( B α + ) ) − 1 ‖ 2 ‖ U M ′ ‖ 2 < ε rank ( V M ) = rank ( V M ′ ) ⩽ M α
This completes the proof.
According to the Weyl's theorem,Theorem 3indicates the spectrum of (τ(Bα+))−1Bα+ is clustered around 1[36]. This typically leads to a fast convergence rate for the PCG method in solving linear systems(16)and(17)[24-25,36].
3 Numerical Results
A series of numerical tests are performed to evaluate the method's effectiveness. We set the stopping criterion to be ‖r(k)‖2/‖r(0)‖2< 10-7.As shown in Tables 1-8,the acronyms “CG-LOD”and “CG-ADI” denote the base CG method for the linearized LOD and ADI schemes,respectively.“PCG-LOD(τ)” and “ PCG-ADI(τ)” represent our proposed PCG method with the τ-preconditioner.“ PCG-LOD(S/ T)” and “ PCG-ADI(S/ T)”indicate the PCG methods with Strang's[37]and T.Chan's[20]circulant preconditioners,respectively.“Time(s)” denotes the total CPU time in seconds.“ Iter ” represents the average number of iterations across all time steps for solving linear systems in both directions. Let Δx= Δy= h. Define
Eerror =ΔxΔy∑i=1M1∑j=1M2(Ui,j−ui,j)2
Rconvergence- Δt=log2⁡[Eerror (h,Δt)Eerror (h,Δt/2)]
Rconvergence-h =log2⁡[Eerror (h,Δt)Eerror (h/2,Δt)]
Table 1 Temporal errors and convergence rates of the LOD scheme (11)with h= 1/ 200
Table 2 Spatial errors and convergence rates of the LOD scheme (11)with Δt= 1/ 10 000
Table 3 Temporal errors and convergence rates of the ADI scheme (7)with h= 1/ 200
Table 4 Spatial errors and convergence rates of the ADI scheme (7)with Δt= 1/ 10 000
Example 1 Consider the problem(1)with Kβ= Kγ= 1,L1= L2= T= 1,f(u)= sin(u),
p(x,y,t)=exp⁡(−t)[12cos⁡(βπ/2)y4(1−y)4𝛷(x,β)+12cos⁡(γπ/2)x4(1−x)4⋅𝛷(y,γ)]−exp⁡(−t)(x(1−x)y(1−y))4−sin⁡(u)
and
𝛷(z,α)=𝛤(5)𝛤(5−α)[z4−α+(1−z)4−α]−4𝛤(6)𝛤(6−α)[z5−α+(1−z)5−α]+6𝛤(7)𝛤(7−α)[z6−α+(1−z)6−α]−4𝛤(8)𝛤(8−α)[z7−α+(1−z)7−α]+𝛤(9)𝛤(9−α)[z8−α+(1−z)8−α](z,α)=(x,β) or (y,γ)
The exact solution is
u(x,y,t)=exp⁡(−t)(x(1−x)y(1−y))4
Tables 1-4 describe the numerical outcomes for solving Example 1,showing the errors and convergence rates for the linearized ADI scheme(7)and LOD scheme(11). It can be seen that both schemes achieve second-order temporal convergence and fourth-order spatial convergence.
Table 5 Comparisons of PCG-LOD(τ)method and the CG-LOD method with Δt= 2h
Table 6 Comparisons of PCG-ADI(τ)method and the CG-ADI method with Δt= 2h
In Tables 5-6,we compare the performance of our PCG-LOD(τ)and PCG-ADI(τ)methods with those of the CG-LOD and CG-ADI methods. It is evident that PCG-LOD(τ)and PCG-ADI(τ)achieve both a lower iteration count and reduced computational time. For instance,when β= γ= 1.9,h= 1/ 1 000,the CG-LOD method achieves an accuracy of 1.167 21×10-9at the cost of an average of246. 25 iterations and 18 705.9 s. In contrast,PCG-LOD(τ)achieves the same precision with only 4.00 iterations on average and a computation time of1 145.0 s. The results demonstrate that the proposed methods markedly enhance computational efficiency.
Existing circulant preconditioners have been developed for solving the linear systems(16)and(17)with the ST coefficient matrices[20,37]. As indicated in Tables 7-8,our PCG-LOD(τ)and PCG-ADI(τ)methods outperform the existing methods in terms of average iteration count,including PCG-LOD(S/ T )and PCG-ADI( S/ T).
The spectra of several preconditioned matrices are presented in Figs. 1 and 2. As clearly shown,matrices(τ(Bβ+))−1Bβ+and (τ(Bγ+))−1Bγ+exhibit significantly greater spectral concentration compared with matrices(S(Bβ+))−1Bβ+,(T(Bβ+))−1Bβ+,(S(Bγ+))−1Bγ+and(T(Bγ+))−1Bγ+. The average iteration numbers for various preconditioning methods,presented in Tables 7-8,lend further support to this conclusion.
Table 7 Performance of different PCG methods with Δt= 2h
Table 8 Performance of different PCG methods with Δt= 2h
Fig. 1 The spectrum of Bβ+,(τ(Bβ+))−1Bβ+,(S(Bβ+))−1Bβ+ and (τ(Bβ+))−1Bβ+with h= Δt= 1/ 26 and β=γ= 1.9 for Example 1
Fig. 2 The spectrum of Bγ+,(τ(Bγ+))−1Bγ+,(S(Bγ+))−1Bγ+ and (T(Bγ+))−1Bγ+with h= Δt= 1/ 26and β=γ= 1.9 for Example 1
4 Conclusions
Fourth-order linearized ADI and LOD methods are derived herein to solve 2D nonlinear RSFDEs numerically. Discrete energy analysis provides explicit error estimates that establish the second-order temporal and fourth-order spatial accuracy of both schemes.Moreover,the PCG method,which incorporates a τ-preconditioner,is employed to solve these discrete linear systems. It has been theoretically proved that the preconditioned matrix can be decomposed into the sum of an identity matrix,a small-norm matrix,and a low-rank matrix. As demonstrated by numerical tests,both methods with τ-preconditioner outperform existing methods in both iteration efficiency and computational cost. Furthermore,preliminary evidence suggests that the proposed τ-preconditioner may also be applicable to the ADI scheme developed by Hu and Cao[33],which constitutes a pertinent subject for our further investigation.
Conflicts of Interest
The authors declare no potential conflicts of interest.
Fig. 1 The spectrum of Bβ+,(τ(Bβ+))−1Bβ+,(S(Bβ+))−1Bβ+ and (τ(Bβ+))−1Bβ+with h= Δt= 1/ 26 and β=γ= 1.9 for Example 1
Fig. 2 The spectrum of Bγ+,(τ(Bγ+))−1Bγ+,(S(Bγ+))−1Bγ+ and (T(Bγ+))−1Bγ+with h= Δt= 1/ 26and β=γ= 1.9 for Example 1
Table 1 Temporal errors and convergence rates of the LOD scheme (11)with h= 1/ 200
Table 2 Spatial errors and convergence rates of the LOD scheme (11)with Δt= 1/ 10 000
Table 3 Temporal errors and convergence rates of the ADI scheme (7)with h= 1/ 200
Table 4 Spatial errors and convergence rates of the ADI scheme (7)with Δt= 1/ 10 000
Table 5 Comparisons of PCG-LOD(τ)method and the CG-LOD method with Δt= 2h
Table 6 Comparisons of PCG-ADI(τ)method and the CG-ADI method with Δt= 2h
Table 7 Performance of different PCG methods with Δt= 2h
Table 8 Performance of different PCG methods with Δt= 2h
Metzler R, Nonnenmacher T F. Fractional relaxation processes and fractional rheological models for the description of a class of viscoelastic materials. International Journal of Plasticity, 2003, 19 (7): 941-959. DOI: 10. 1016/S0749-6419(02)00087-6.
Benson D A, Wheatcraft S W, Meerschaert M M. Application of a fractional advection-dispersion equation. Water Resources Research, 2000, 36 (6): 1403-1412. DOI: 10.1029/2000WR900031.
Raberto M, Scalas E, Mainardi F. Waiting-times and returns in high-frequency financial data: an empirical study. Physica A: Statistical Mechanics & Its Applications, 2002, 314 (1-4): 749-755. DOI: 10. 1016/S0378-4371 (02) 01048-8.
Sabatelli L, Keating S, Dudley J, et al. Waiting time distributions in financial markets. The European Physical Journal B-Condensed Matter and Complex Systems, 2002, 27: 273-275. DOI: 10.1140/epjb/e20020151.
Meerschaert M M, Tadjeran C. Finite difference approximation for fractional advection-dispersion flow equations. Journal of Computational and Applied Mathematics, 2004, 172(1): 65-77. DOI: 10.1016/j.cam.2004.01.033.
Langlands T A M, Henry B I. The accuracy and stability of an implicit solution method for the fractional diffusion equation. Journal of Computational Physics, 2005, 205(2): 719-736. DOI: 10.1016/j.jcp.2004.11.025.
Meerschaert M M, Scheffler H P, Tadjeran C. Finite difference methods for two-dimensional fractional dispersion equation. Journal of Computational Physics, 2006, 211(1): 249-261. DOI: 10.1016/j.jcp.2005.05.017.
Liu F, Anh V, Turner I. Numerical solution of the space fractional Fokker-Planck equation. Journal of Computational and Applied Mathematics, 2004, 166(1): 209-219. DOI: 10.1016/j.cam.2003.09.028.
Ervin V J, Heuer N, Roop J P. Numerical approximation of a time dependent nonlinear space-fractional diffusion equation. SIAM Journal on Numerical Analysis, 2007, 45(2): 572-591. DOI: 10.1137/050642757.
Deng W. Finite element method for the space and time fractional Fokker-Planck equation. SIAM Journal on Numerical Analysis, 2008, 47 (1): 204-226. DOI: 10.1137/080714130.
Fu H, Sun Y, Wang H, et al. Stability and convergence of a Crank-Nicolson finite volume method for space fractional diffusion equations. Applied Numerical Mathematics, 2019, 139: 38-51. DOI: 10.1016/j.apnum.2019.01.005.
Hejazi H, Moroney T, Liu F. Stability and convergence of a finite volume method for the space fractional advection-dispersion equation. Journal of Computational & Applied Mathematics, 2014, 255: 684-697. DOI: 10.1016/j.cam.2013.06.039.
Simmons A, Yang Q, Moroney T. A finite volume method for two-sided fractional diffusion equations on non-uniform meshes. Journal of Computational Physics, 2017, 335: 747-759. DOI: 10.1016/j.jcp.2017.01.061.
Ortigueira M D. Riesz potential operators and inverses via fractional centred derivatives. International Journal of Mathematics and Mathematical Sciences, 2006, 2006(1): 048391. DOI: 10. 1155/IJMMS/2006/48391.
Çelik C, Duman M. Crank-Nicolson method for the fractional diffusion equation with the Riesz fractional derivative. Journal of Computational Physics, 2012, 231(4): 1743-1750. DOI: 10.1016/j.jcp.2011.11.008.
Tian W, Zhou H, Deng W. A class of second order difference approximations for solving space fractional diffusion equations. Mathematics of Computation, 2015, 84(294): 1703-1727. DOI: 10.1090/s0025-5718-2015-02917-2.
Zhou H, Tian W Y, Deng W H. Quasi-compact finite difference schemes for space fractional diffusion equations. Journal of Scientific Computing, 2013, 56: 45-66. DOI: 10.1007/s10915-012-9661-0.
Hao Z P, Sun Z Z, Cao W R. A fourth-order approximation of fractional derivatives with its applications. Journal of Computational Physics, 2015, 281: 787-805. DOI: 10.1016/j.jcp.2014.10.053.
Yu J W, Zhang C H, Huang X, et al. A class of preconditioner for solving the Riesz distributed-order nonlinear space-fractional diffusion equations. Japan Journal of Industrial and Applied Mathematics, 2023, 40: 537-562. DOI: 10.1007/s13160-022-00551-w.
Chan T F. An optimal circulant preconditioner for Toeplitz systems. SIAM Journal on Scientific and Statistical Computing, 1988, 9(4): 766-771. DOI: 10.1137/0909051.
Chan R H. Circulant preconditioners for Hermitian Toeplitz systems. SIAM Journal on Matrix Analysis and Applications, 1989, 10: 876-896. DOI: 10.1137/0610039.
Tang S P, Yang A L, Zhou J L, et al. R. Chan's circulant-based approximate inverse preconditioning iterative method for solving second-order space fractional advection-dispersion equations with variable coefficients. Computational and Applied Mathematics, 2024, 43(2): article number 97. DOI: 10.1007/s40314-024-02592-y.
Zhang C H, Yu J W, Wang X. A fast second-order scheme for nonlinear Riesz space fractional diffusion equations. Numerical Algorithms, 2023, 92: 1813-1836. DOI: 10.1007/s11075-022-01367-y.
Lin X L, Huang X, Ng M K, et al. A τ-preconditioner for a non-symmetric linear system arising from multidimensional Riemann-Liouville fractional diffusion equation. Numerical Algorithms, 2023, 92: 795-813. DOI: 10.1007/s11075-022-01342-7.
Huang Y Y, Qu W, Lel S L. On τ-preconditioner for a novel fourth-order difference scheme of two-dimensional Riesz space-fractional diffusion equations. Computers & Mathematics with Applications, 2023, 145: 124-140. DOI: 10.1016/j.camwa.2023.06.015.
Tadjeran C, Meerschaert M M. A second-order accurate numerical method for the two-dimensional fractional diffusion equation. Journal of Computational Physics, 2007, 220 (2): 813-823. DOI: 10.1016/j.jcp.2006.05.030.
Wang H, Wang K. An O (N log 2N) alternating-direction finite difference method for two-dimensional fractional diffusion equations. Journal of Computational Physics, 2011, 230(21): 7830-7839. DOI: 10.1016/j.jcp.2011.07.003.
Wang H, Du N. Fast alternating-direction finite difference methods for three-dimensional space-fractional diffusion equations. Journal of Computational Physics, 2014, 258: 305-318. DOI: 10.1016/j.jcp.2013.10.040.
Qin J, Wang T. A compact locally one-dimensional finite difference method for nonhomogeneous parabolic differential equations. International Journal for Numerical Methods in Biomedical Engineering, 2011, 27(1): 128-142. DOI: 10.1002/cnm.1299.
Wang T, Wang Y M. A higher-order compact LOD method and its extrapolations for nonhomogeneous parabolic differential equations. Applied Mathematics & Computation, 2014, 237: 512-530. DOI: 10.1016/j.amc.2014.03.139.
Tang S P, Huang Y M. A fast ADI based matrix splitting preconditioning method for the high dimensional space fractional diffusion equations in conservative form. Computers and Mathematics with Applications, 2023, 144: 210-220. DOI: 10.1016/j.camwa.2023.05.028.
Chen M, Wang Y, Cheng X, et al. Second-order LOD multigrid method for multidimensional Riesz fractional diffusion equation. BIT Numerical Mathematics, 2014, 54: 623-647. DOI: 10.1007/s10543-014-0477-1.
Hu D, Cao X. A fourth-order compact ADI scheme for two-dimensional Riesz space fractional nonlinear reaction-diffusion equation. International Journal of Computer Mathematics, 2014, 97(9): 1928-1948. DOI: 10.1080/00207160.2019.1671587.
Sun Z Z. The Method of Order Reduction and Its Application to the Numerical Solutions of Partial Differential Equations. Beijing: Science Press, 2009.
Bini D, Benedetto F. A new preconditioner for the parallel solution of positive definite Toeplitz systems. Proceedings of the Second Annual ACM Symposium on Parallel Algorithms and Architectures, New York: ACM, 1990: 220-223. DOI: 10.1145/97444.97688.
Jin X. Preconditioning Techniques for Toeplitz Systems. Beijing: Higher Education Press, 2010.
Strang G. A proposal for Toeplitz matrix calculations. Studies in Applied Mathematics, 1986, 74(2): 171-176.

LINKS