Next Article in Journal
Mach Fronts in Random Media with Fractal and Hurst Effects
Next Article in Special Issue
A Mixed Element Algorithm Based on the Modified L1 Crank–Nicolson Scheme for a Nonlinear Fourth-Order Fractional Diffusion-Wave Model
Previous Article in Journal
Solvability of a New q-Differential Equation Related to q-Differential Inequality of a Special Type of Analytic Functions
Previous Article in Special Issue
Novel Numerical Investigations of Fuzzy Cauchy Reaction–Diffusion Models via Generalized Fuzzy Fractional Derivative Operators
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

A Fast Preconditioned Semi-Implicit Difference Scheme for Strongly Nonlinear Space-Fractional Diffusion Equations

1
Institute of Mathematics, School of Economic Mathematics, Southwestern University of Finance and Economics, Chengdu 611130, China
2
Fundamental Department, Shanghai Customs College, Shanghai 201204, China
3
School of Mathematics, Chengdu Normal University, Chengdu 611130, China
4
School of Mathematical Sciences, Sichuan Normal University, Chengdu 610068, China
5
Faculty of Computer Science, Free University of Bozen-Bolzano, 39100 Bolzano, Italy
*
Authors to whom correspondence should be addressed.
Fractal Fract. 2021, 5(4), 230; https://doi.org/10.3390/fractalfract5040230
Submission received: 26 October 2021 / Revised: 11 November 2021 / Accepted: 13 November 2021 / Published: 18 November 2021
(This article belongs to the Special Issue Novel Numerical Solutions of Fractional PDEs)

Abstract

:
In this paper, we propose a semi-implicit difference scheme for solving one-dimensional nonlinear space-fractional diffusion equations. The method is first-order accurate in time and second-order accurate in space. It uses a fractional central difference formula and the backward Euler method to approximate its space and time derivatives, respectively. Stability and convergence properties of the proposed scheme are proved with the help of a discrete Grönwall inequality. Moreover, we extend the method to the solution of two-dimensional nonlinear models. A fast matrix-free implementation based on preconditioned Krylov subspace methods is presented for solving the discretized linear systems. The resulting fast preconditioned semi-implicit difference scheme reduces the memory requirement of conventional semi-implicit difference schemes from O ( N s 2 ) to O ( N s ) and the computational complexity from O ( N s 3 ) to O ( N s log N s ) in each iterative step, where N s is the number of space grid points. Experiments with two numerical examples are shown to support the theoretical findings and to illustrate the efficiency of our proposed method.

1. Introduction

In this paper, we are interested in developing efficient numerical solutions of strongly nonlinear space-fractional diffusion equations (NSFDEs) of the following form:
u ( x , t ) t = a ( u ) α u ( x , t ) | x | α + f ( x , t ) , ( x , t ) ( 0 , L x ) × ( 0 , T ] , u ( x , 0 ) = ϕ ( x ) , x [ 0 , L x ] , u ( 0 , t ) = φ ( t ) , u ( L x , t ) = ψ ( t ) , t [ 0 , T ] ,
where a ( u ) , f ( x , t ) , ϕ ( x ) , φ ( t ) , ψ ( t ) are known functions. Function u ( x , t ) may represent, for example, the concentration of a particle plume undergoing anomalous diffusion. The diffusion coefficient a ( u ) is assumed to be positive, and the forcing function f ( x , t ) represents the source or sink term [1,2]. In addition, we assume that Equation (1) has a smooth solution u ( x , t ) , and that there exist real positive numbers ϵ 1 , ϵ 2 , ϵ , a 0 , a 1 and L satisfying the following:
0 < a 0 a ( u ( x , t ) + ϵ 1 ) a 1 ,
| a ( u ( x , t ) + ϵ 1 ) a ( u ( x , t ) + ϵ 2 ) | L | ϵ 1 ϵ 2 | ,
where | ϵ 1 | ϵ , | ϵ 2 | ϵ , and ( x , t ) [ 0 , L x ] × [ 0 , T ] . Moreover, the symbol α | x | α denotes the Riesz fractional derivative operator [3,4,5] for 1 < α 2 , defined as follows:
α u ( x , t ) | x | α = 1 2 cos α π 2 Γ ( 2 α ) 2 x 2 | x ξ | 1 α u ( ξ , t ) d ξ ,
where Γ ( · ) is the Gamma function.
In the past few decades, a significant amount of research work has been devoted to the analysis and solution of fractional partial differential equations (fPDEs) especially for modeling anomalous diffusion [6,7], subdiffusion and and superdiffusion processes that can be observed in many real physical systems, such as heat transfer problems and wave analysis [8,9]. Space fractional equations are obtained by replacing the traditional space derivative with a fractional derivative in the integer-order diffusion equation and can describe heterogeneous substances with memory and genetic properties very effectively [10]. Closed-form analytic solutions are available only for limited classes of fPDEs. For solving more general equations, the only viable approach is to use discretization methods such as finite difference [11,12,13,14,15], finite element [16,17], finite volume [18,19] and spectral [20] methods.
The Riemann–Liouville fractional derivative is often used to model the dynamics of discrete systems with long-range interaction. Several studies have applied the weighted and shifted Grünwald difference (WSGD) formula to approximate the Riemann–Liouville derivative; see [14,21,22] for details. By using the WSGD formula for two-sided fractional derivatives combining the compact technique, Hao et al. presented a fourth-order difference approximation for the space fractional derivatives [23]. In addition, Çelik and Duman used a fractional centered difference operator to approximate the Riesz fractional derivative with second-order accuracy and to solve the Riesz SFDE (RSFDE) on a finite domain [24]. Later, Yang et al. [25] used the Diethelm method [15] to derive a new O ( Δ x 3 α ) approximation for discretizing the Riesz fractional derivatives of order α ( 1 , 2 ) , where Δ x is the step size. Then, a new finite difference scheme for solving the RSFDE was obtained by discretizing the first-order time derivative using the Crank–Nicolson (CN) method. In contrast with the classical diffusion operator Δ , Riesz fractional derivative [3,5] is a special linear combination of left- and right-sided Riemann–Liouville fractional derivatives. By exploiting this property, Ding and Li [4] established a novel class of high-order numerical algorithms for Riesz derivatives through constructing new generating functions. In 2018, Lin et al. [26] studied the CN temporal discretization with various high-order spatial difference schemes for RSFDEs with variable diffusion coefficients and gave the unconditional stability and convergence analysis for temporally second-order and spatially jth-order ( j = 1 , 2 , 3 , 4 ) difference schemes for such equations with variable coefficients. In 2019, Lin et al. [27] studied the CN alternative direction implicit (CN-ADI) method for two-dimensional RSFDEs with nonseparable coefficients. They showed under mild assumptions the unconditional stability of the CN-ADI method in discrete 2 -norm and the consistency of cross perturbation terms arising from the CN-ADI method. We point the reader to [3,12,13,28] for other recent works on the numerical solution of fPDEs with the Riesz fractional derivative.
Due to their non-local nature, numerical approximations of space fractional operators generally lead to dense matrices. Conventional direct solution methods for linear systems based on variants of the Gaussian elimination algorithm may not be affordable to use because of their high computational complexity and large memory costs. The quest for fast algorithms for solving linear systems originated in the numerical discretization of fPDEs has become a topic of great interest in the recent literature on numerical methods for fPDEs. Lin et al. [29] proposed a splitting preconditioner for fast solution of Toeplitz-like linear systems arising from one- and two-dimensional time-dependent RSFDEs with variable diffusion coefficients [27]. Some other preconditioners are being applied also to solving the discretized (non-)linear systems from the solution of linear and non-linear problems; see, for example, [1,2,30,31] and the references therein.
In this paper, we study the efficient numerical solution of strongly NSFDEs because they are more suitable than linear models to describe some difficult physical processes, such as fractional distillation in nonlinear space, and when the diffusion coefficient is related to the concentration of a particle plume [31]. However, to the best of our knowledge, only a few studies on fast numerical schemes for nonlinear or semi-linear problems have been published in the literature; see, for example, refs. [32,33,34]. Our main contribution to the subject can be summarized as follows: first, for this class of strongly NSFDEs, we propose a semi-implicit difference scheme that can avoid solving the discretized nonlinear systems [35], and analyze its stability and convergence properties. Then, we extend the method to the solution of the two-dimensional formulation of Equation (1). In order to compute an efficient iterative solution of the discretized systems, a suitable preconditioning technique is developed and presented in the paper. The complexity and memory requirement of the resulting semi-implicit difference scheme are fairly lower, compared to the fully implicit difference scheme that requires to solve the discretized nonlinear systems, motivating our choice of the semi-implicit difference scheme for Equation (1) [36]. Moreover, unlike the traditional (semi-)implicit difference schemes, our proposed semi-implicit difference scheme can be implemented as a matrix-free method because it does not store any of the coefficient matrices involved in the whole solution process.
The rest of this article is organized as follows. In Section 2, we propose a semi-implicit difference scheme for Equation (1), and then we analyze its stability and convergence properties. In Section 3, we extend our method to the solution of the two-dimensional formulation of Equation (1). A fast matrix-free preconditioned iterative algorithm for solving the discretized linear system arising in the analysis of the two-dimensional case of Equation (1) is developed, and it is presented in Section 4. In Section 5, we report on numerical experiments to illustrate the efficiency of the proposed semi-implicit difference method. Some conclusions arising from this work are drawn in Section 6.

2. The Semi-Implicit Difference Scheme and Its Stability and Convergence Analysis

In this section, we propose a semi-implicit difference method to approximate the solution of Equation (1). The proposed method uses a fractional central difference formula to discretize the Riesz fractional derivative; it is stable and achieves first-order convergence in time and second-order convergence in space, respectively.

2.1. The Semi-Implicit Difference Scheme

We introduce the following notation. Let M , N be positive integers, h = L x / N , x i = i h , t j = j Δ t and Δ t = T / M . We consider the sets of grid points Ω h : = x i | 0 i N , Ω Δ t : = t j | 0 j M , Ω h Δ t : = { ( x i , t j ) | 0 i N , 0 j M } , and let v = { v i j | 0 i N , 0 j M } be a grid function defined on Ω h Δ t . To simplify the notation, we write Equation (1) at the generic points ( x i , t j ) as follows:
u ( x i , t j ) t = a u ( x i , t j ) α u ( x i , t j ) | x | α + f ( x i , t j ) , 1 i N 1 , 1 j M ,
or, equivalently, as follows:
u ( x i , t j ) t a u ( x i , t j 1 ) α u ( x i , t j ) | x | α = F ( x i , t j ) , 1 i N 1 , 1 j M ,
where F ( x i , t j ) = f ( x i , t j ) + a ( u ( x i , t j ) ) a ( u ( x i , t j 1 ) ) α u ( x i , t j ) | x | α . Additionally, we introduce a grid function U defined as follows:
U i j : = u ( x i , t j ) , f i j : = f ( x i , t j ) , 0 i N , 0 j M .
By these notations, the partial derivative of u ( x , t ) with respect to t evaluated at point ( x i , t j ) is expressed as follows:
u ( x i , t j ) t = u ( x i , t j ) u ( x i , t j 1 ) Δ t + R i , j ( 1 ) U i j U i j 1 Δ t + O ( Δ t ) ,
and the Riesz space fractional derivative as follows:
α u ( x i , t j ) | x | α = h α Δ h α u ( x i , t j ) + R i , j ( 2 ) δ h α U i j + O ( h 2 ) ,
where Δ h α U i j = k = i N i g k ( α ) U i k j and g k ( α ) = ( 1 ) k Γ ( α + 1 ) Γ ( α / 2 k + 1 ) Γ ( α / 2 + k + 1 ) ; refer to [24] for details. For brevity, we denote the following:
R i j = R i , j ( 1 ) R i , j ( 2 ) + a ( u ( x i , t j ) ) a ( u ( x i , t j 1 ) ) α u ( x i , t j ) | x | α .
Finally, Equation (1) at grid points ( x i , t j ) is written as follows:
U i j U i j 1 Δ t a ( U i j 1 ) δ h α U i j = f i j + R ^ i j , 1 i N 1 , 1 j M ,
where the terms { R ^ i j } are small and satisfy the inequality as follows:
| R ^ i j | c 1 ( Δ t + h 2 ) , 1 i N 1 , 1 j M .
Since the quantities { R ^ i j } are small, we can omit them. By imposing the initial-boundary value conditions,
U i 0 = ϕ ( x j ) , 0 i N ,
U 0 j = φ ( t j ) , U N j = ψ ( t j ) , 1 j M ,
the following semi-implicit difference scheme is obtained:
u i j u i j 1 Δ t a ( u i j 1 ) δ h α u i j = f i j , 1 i N 1 , 1 j M , u i 0 = ϕ ( x j ) , 0 i N , u 0 j = φ ( t j ) , u N j = ψ ( t j ) , 1 j M ,
that requires to solve the discretized linear system at each time step.

2.2. Stability and Convergence Analysis

In this subsection, we study the stability and convergence properties of the numerical scheme given by Equation (15). Let u i j be the approximate solution of Equation (15), and introduce the error function as follows:
e i j = U i j u i j , i = 0 , , N , j = 0 , , M ,
defined as
e i j e i j 1 Δ t a ( u i j 1 ) δ h α e i j = a ( U i j 1 ) a ( u i j 1 ) δ h α U i j + R ^ i j , 1 i N 1 , 1 j M , e i 0 = 0 , 0 i N , e 0 j = 0 , e N j = 0 , 1 j M ,
where | R ^ i j | c 2 ( Δ t + h 2 ) . Denote E j = [ e 1 j , e 2 j , , e N 1 j ] , and assume the following:
E j = | e j | = max 1 N 1 | e j | ( 0 j M , 1 j N 1 ) .
To prove the stability and assess the convergence order of the scheme (15), we need some preliminary theoretical results given in [24,35,37] that are recalled below.
Lemma 1.
Let
g k ( α ) = ( 1 ) k Γ ( α + 1 ) Γ ( α / 2 k + 1 ) Γ ( α / 2 + k + 1 )
be the coefficients of the centered finite difference approximation (9) for k = 0 , ± 1 , ± 2 , , and α > 1 . Then, we have the following:
(1) 
g 0 ( α ) 0 and g k ( α ) = g k ( α ) 0 for all | k | 1 ;
(2) 
g k ( α ) = 0 and k = i N , k 0 i g k ( α ) g 0 .
Lemma 2.
(Discrete Grönwall inequality) Let { F k | k 0 } be a sequence of non-negative numbers and let Δ t > 0 be such that the following is true:
F k + 1 ( 1 + c Δ t ) F k + ( Δ t ) g , k = 0 , 1 , 2 , ,
where both c and g are non-negative scalars. Then, the following inequality holds:
F k e c k Δ t F 0 + g c
with k = 0 , 1 , .
At this stage, the following result on the numerical stability of the proposed one dimensional semi-implicit difference scheme can be proved:
Theorem 1.
Suppose that { u i j | 0 j M , 0 i N } is the solution of the semi-implicit difference scheme (15), and define the following:
C 1 = c 1 L M α exp ( L M α T ) , M α = max 0 x L , 1 t T α u ( x , t ) | x | α .
Then, if Δ t and h satisfy the following condition
Δ t ϵ 2 C 1 , h ϵ 2 C 1 ,
the semi-implicit difference scheme (15) is stable, and the following inequality holds:
e k C 1 ( Δ t + h 2 ) , 0 k n .
Proof. 
According to Equation (15), we know that e k C 1 ( Δ t + h 2 ) holds for k = 0 . Suppose that the inequality (22) holds for 0 k l , when Δ t and h satisfy the condition (21) and
e k ϵ , 0 k l .
Therefore, the following holds:
a 0 a ( u i j ) a 1 , | a ( U i j ) a ( u i j ) | L | e i j | , 1 i N 1 , 0 j l .
From Equation (17), we have the following:
e i j e i j 1 = Δ t a ( u i j 1 ) δ h α e i j + Δ t a ( U i j 1 ) a ( u i j 1 ) δ h α U i j + Δ t R ^ i j ,
where δ h α e i j = h ( α ) k = i N i g k ( α ) e i k j . Thus we obtain the following:
e i j e i j 1 = Δ t h α a ( u i j 1 ) k = i N , k 0 i g k ( α ) e i k j + g 0 ( α ) e i j + Δ t a ( U i j 1 ) a ( u i j 1 ) δ h α U i j + Δ t R ^ i j .
So, we rewrite Equation (17) to obtain the following:
1 + μ g 0 ( α ) a ( u i j 1 ) e i j = μ a ( u i j 1 ) k = i N , k 0 i g k ( α ) e i k j + e i j 1 + Δ t a ( U i j 1 ) a ( u i j 1 ) δ h α U i j + Δ t R ˜ i j ,
where μ = Δ t / h α . If we take infinite norms on both sides of Equation (25), and we set i = i obtained so that
1 + μ g 0 a ( u i j 1 ) e j μ a ( u i j 1 ) g 0 ( α ) e j + e j 1 + Δ t L M α e j 1 + c 1 Δ t ( Δ t + h 2 )
for 1 i N 1 , 1 j l + 1 . Thus, we have the following:
e j = ( 1 + Δ t L M α ) e j 1 + c 1 Δ t ( Δ t + h 2 ) , 1 j l + 1 .
By applying the discrete Grönwall inequality (cf. Lemma 2) to F k = e l 1 , c = L M α and g = c 1 ( Δ t + h 2 ) , we obtain the inequality as follows:
e l + 1 = exp L M α ( l + 1 ) Δ t · e 0 + c 1 ( Δ t + h 2 ) L M α c 1 L M α exp ( L M α T ) ( Δ t + h 2 ) C 1 ( Δ t + h 2 ) .
The above proof shows that the semi-implicit difference scheme (15) is stable in the discrete L -norm.    □

3. Extensions to the Two-Dimensional Problem

In this section, we extend our method to the solution of two-dimensional problems. For this purpose, let N x , N y be positive integers and define a spatial partition of nodes h = L / N , x i = i h x ( i = 1 , 2 , , N x ) , y j = j h y ( j = 1 , 2 , , N y ) with step sizes h x = L x N x , h y = L y N y in the x-axis and y-axis, respectively, and a uniform time partition t k = k Δ t of the interval [ 0 , T ] with time step Δ t = T / M . We consider the sets of grid points Ω h : = { ( x i , y j ) | 0 i N x , 0 j N y } , Ω Δ t : = { t k | 0 k M } , Ω h Δ t : = { ( x i , y j , t k ) | 0 i N x , 0 j N y , 0 k M } , and let u = { u i , j k | 0 i N x , 0 j N y , 0 k M } be a grid function defined on Ω h Δ t .
We consider the following two-dimensional problem:
u ( x , y , t ) t = a ( u ) α u ( x , y , t ) | x | α + α u ( x , y , t ) | y | α + f ( x , y , t ) , ( x , y , t ) Ω × ( 0 , T ] , u ( x , y , 0 ) = ϕ ( x , y ) , ( x , y ) Ω , u ( x , y , t ) = 0 , ( x , y , t ) Ω × ( 0 , T ] .
where Ω = [ 0 , L x ] × [ 0 , L y ] . Upon evaluating Equation (26) at the points ( x i , y j , t k ) , we obtain the following discrete equation:
u ( x i , y j , t k ) t = a ( u ( x i , y j , t k ) ) u ( x i , y j , t k ) | x | α + u ( x i , y j , t k ) | y | α + f ( x i , y j , t k ) , 1 i N x 1 , 1 j N y 1 , 1 k M ,
which can be written equivalently as follows:
u ( x i , y j , t k ) t a ( u ( x i , y j , t k 1 ) ) α u ( x i , y j , t k ) | x | α + α u ( x i , y j , t k ) | y | α = a ( u ( x i , y j , t k ) ) a ( u ( x i , y j , t k 1 ) ) α u ( x i , y j , t k ) | x | α + α u ( x i , y j , t k ) | y | α + f ( x i , y j , t k ) .
Let U be the grid function defined as follows:
U i , j k : = u ( x i , y j , t k ) , f i , j k : = f ( x i , y j , t k ) , 0 i N x , 0 j N y , 0 k M .
Thus, the partial derivative of t at the point ( x i , y j , t k ) is as follows:
u ( x i , y j , t k ) t = u ( x i , y j , t k ) u ( x i , y j , t k 1 ) Δ t + R i , j , k ( 1 ) = u i , j k u i , j k 1 Δ t + O ( Δ t ) ,
and
α u ( x i , y j , t k ) | x | α = h x α Δ h x α u ( x i , y j , t k ) + R i , j , k ( 2 ) δ h x α U i j + O ( h x 2 ) ,
α u ( x i , y j , t k ) | y | α = h y α Δ h y α u ( x i , y j , t k ) + R i , j , k ( 3 ) δ h y α U i j + O ( h y 2 ) ,
where Δ h x α U i , j k = = i N x i g l ( α ) U i , j k , Δ h y α U i , j k = = j N y j g ( α ) U i , j k and g ( α ) = ( 1 ) Γ ( α + 1 ) Γ ( α / 2 + 1 ) Γ ( α / 2 + + 1 ) .
We denote for simplicity the following:
R i , j , k = R i , j , k ( 1 ) a ( u ( x i , y j , t k 1 ) ) R i , j , k ( 2 ) a ( u ( x i , y j , t k 1 ) ) R i , j , k ( 3 ) + R ,
where R = a ( u ( x i , y j , t k ) ) a ( u ( x i , y j , t k 1 ) ) α u ( x i , y j , t k ) | x | α + α u ( x i , y j , t k ) | y | α , so that the equation can be rewritten as follows:
U i , j k U i , j k 1 Δ t a ( U i , j k 1 ) δ h x α U i , j k + δ h y α U i , j k = f i , j k + R i , j , k ,
with 1 i N x 1 , 1 j N y 1 , 1 k M .
Observe that the terms R i , j , k are small. Omitting the quantities R i , j , k in the previous expression, and imposing the initial-boundary value conditions,
U i , j 0 = ϕ ( x i , y j ) , 0 i N x , 0 j N y ; U 0 , j k = U N x , j k = φ ( t k ) , U i , 0 k = U i , N y k = φ ( t k ) , 0 k M ,
the following semi-implicit difference scheme is obtained, where by u i , j k we denote the approximate solution of Equation (31) at the points ( x i , y j , t k ) :
u i , j k u i , j k 1 Δ t a ( u i , j k 1 ) δ h x α u i , j k + δ h y α u i , j k = f i , j k , 1 i N x 1 , 1 i N y 1 , 1 k M u i , j 0 = ϕ ( x i , y j ) , 1 i N x 1 , 1 i N y 1 , u 0 , j k = u N x , j k = φ ( t k ) , u i , 0 k = u i , N y k = ψ ( t k ) , 1 k M .
Below, we analyze stability and convergence properties of the two-dimensional semi-implicit difference scheme given by Equation (32). We define the following:
e i , j k = U i , j k u i , j k , i = 0 , , N x , j = 1 , , N y , k = 0 , , M ,
the error satisfying the following equations:
e i , j k e i , j k 1 Δ t a ( u i , j k 1 ) δ h x α e i , j k + δ h y α e i , j k = a ( U i , j k 1 ) a ( u i , j k 1 ) δ h x α U i , j k + δ h y α U i , j k + R i , j , k , 1 i N x 1 , 1 j N y 1 , 1 k M , e i , j 0 = 0 , 0 i N x , 0 j N y , e 0 , j k = e N x , j k = 0 , e i , 0 k = e i , N y k = 0 , 1 k M .
By a similar proof to Theorem 1, the following result is established about the numerical stability of the semi-implicit difference scheme given by Equation (26).
Theorem 2.
Suppose that { u i , j k | 0 i N x , 0 j N y , 0 k M } is the solution of the semi-implicit difference scheme (25), and define the following:
C 2 = c 2 L M α exp ( L M α T ) , M α = max 0 x , y L , 1 t T α u ( x , t ) | x | α + α u ( x , y , t ) | y | α .
Then, when constants Δ t , h ( = h x = h y ) satisfy the following condition,
Δ t ϵ 2 C 2 , h ϵ 2 C 2 ,
the semi-implicit difference scheme (32) is stable and as follows:
e k C 2 ( Δ t + h 2 ) , 0 k M .

4. Fast Implementation of the Semi-Implicit Difference Scheme

In this section, we present a fast matrix-free implementation of the proposed one-dimensional and two-dimensional semi-implicit difference schemes that use the preconditioned biconjugate gradients stabilized (BiCGSTAB) method [38] to iteratively solve the nonsymmetric linear system with a Toeplitz-like structure arising from the discretization. Such an implementation enables us to reduce both the algorithmic complexity and the memory requirements for the solution significantly.
First, we analyze the properties of the coefficient matrices of the discretized linear systems.

4.1. One-Dimensional Coefficient Matrix

By rewriting Equation (15) in the following form,
u j u j 1 Δ t = 1 h α D ( j 1 ) A u j + 1 h α D ( j 1 ) v j 1 + f j ,
where v j 1 = [ g 1 ( α ) u 0 j 1 + g ( N 1 ) ( α ) u N j , g 2 ( α ) u 0 j 1 + g ( N 2 ) ( α ) u N j , , g N 1 ( α ) u 0 j 1 + g 1 ( α ) u N j ] , we derive the following matrix form of the semi-implicit difference scheme:
I Δ t h α D ( j 1 ) A u j = u j 1 + Δ t F j ,
where I is an identity matrix of suitable order, F j = 1 h α D ( j 1 ) v j 1 + f j , u j = [ u 1 j , u 2 j , , u N j ] ,
D ( j 1 ) = a ( u 1 j 1 ) a ( u 2 j 1 ) a ( u N 1 j 1 ) R ( N 1 ) × ( N 1 ) ,
is a diagonal matrix and
A = g 0 ( α ) g 1 ( α ) g ( N 2 ) ( α ) g 1 ( α ) g 0 ( α ) g ( N 3 ) ( α ) g N 2 ( α ) g N 3 ( α ) g 0 ( α )
is a symmetric negative definite Toeplitz matrix [3,35], which can be represented by a vector containing the elements of the first column, i.e., its storage requirement is of O ( N ) .

4.2. Two-Dimensional Coefficient Matrix

The structure of the two-dimensional coefficient matrix is more complex than in the one-dimensional case. By writing Equation (26) in the following equivalent form
u k u k 1 Δ t = D ˜ ( k 1 ) W x ( α ) I y + I x W y ( α ) u k + f i , j k ,
in which u k = [ u 11 k , , u N x 1 , 1 k , u 1 , 2 k , , u N x 1 , 2 k , , u 1 , N y 1 k , , u N x 1 , N y 1 k ] , we can derive the discretized linear system as follows:
I Δ t h α D ˜ ( k 1 ) A ˜ u k = u k 1 + Δ t f k ,
where A ˜ = W x ( α ) I y + W y ( α ) I x , I q ( q { x , y } ) is also an identity matrix of suitable order and f k = [ f 11 k , , f N x 1 , 1 k , f 12 k , , f N x 1 , 2 k , , f 1 , N y 1 k , , f N x 1 , N y 1 k ] and
D ˜ ( k 1 ) = a ( u 11 k 1 ) a ( u N x 1 , 1 k 1 ) a ( u 1 , N y 1 k 1 ) a ( u N x 1 , N y 1 k 1 ) .
is a diagonal matrix. In order to simplify the calculation, we choose h = h x = h y (i.e., N x = N y = N ) and thus W x α = W y ( α ) = W ( α ) = A .
According to Lemma 2, W ( α ) is a symmetric Toeplitz matrix of order ( N 1 ) . We convert W ( α ) into a τ -matrix defined as follows:
τ ( W ( α ) ) = W ( α ) H N 1 ,
where H N 1 is a Hankel matrix with constant entries along antidiagonals, namely g 2 ( α ) , g 3 ( α ) , , g N 2 ( α ) , 0 , 0 , 0 , g N 2 ( α ) , , g 3 ( α ) , g 2 ( α ) .
Notice that the τ -matrix can be diagonalized by the sine transform matrix S N 1 [39] as follows:
τ ( W ( α ) ) = S N 1 Λ S N 1 ,
where Λ is the diagonal matrix consisting of the eigenvalues of τ ( W ( α ) ) . The entries of S N 1 are given by the following:
S N 1 i , j = 2 N sin π i j N , 1 i , j N 1 .
A simple calculation shows that S N 1 is a symmetric orthogonal matrix, that is,
S N 1 1 = S N 1 = S N 1 .
Upon replacing W ( α ) with τ ( W ( α ) ) in the expression of A ˜ , and using the properties of the Kronecker product, we derive the following:
P = τ ( W ( α ) ) I + I τ ( W ( α ) ) = S N 1 I ( Λ I ) + ( I Λ ) S N 1 I .
Finally, we introduce matrices A and M defined as follows:
A = I Δ t D ˜ ( k 1 ) A ˜ and P τ = I Δ t d ( k 1 ) P ,
where constant d ( k 1 ) is defined as follows:
d ( k 1 ) : = a ( u 11 k 1 ) + + a ( u 1 , N 1 k 1 ) + + a ( u N 1 , 1 k 1 ) + + a ( u N 1 , N 1 k 1 ) ( N 1 ) 2 > 0 ,
due to the condition (2).
Under these assumptions, we can prove the following property of the preconditioner P τ that guarantees the invertibility of P τ :
Proposition 1.
The preconditioner P τ defined in (43) is symmetric positive definite.
Proof. 
From Equation (41) and Lemma 3.2 in [30], we know that all the eigenvalues of τ ( W ( α ) ) are negative as are those of matrix W ( α ) ; that is, the matrix P defined in (42) is negative definite. Recalling the definition of the preconditioner P τ given by Equation (43), and that Δ t , d ( k 1 ) > 0 , we can conclude that the preconditioner P τ is symmetric positive definite. □
Henceforth, for clarity, we rewrite Equation (40) as the following linear system:
A u = b ,
where the right-hand side vector is b = u k 1 + Δ t f k . Since the coefficient matrix A is non-symmetric, we can solve it by the preconditioned BiCGSTAB method described below: (Algorithm 1).
Algorithm 1 The preconditioned BiCGSTAB method.
1:
x 0 is an initial guess; r 0 = b A u 0 ;
2:
r ¯ 0 is an arbitrary vector, such that
3:
( r ¯ 0 , r 0 ) 0 , e.g., r ¯ 0 = r 0 ; ρ 0 = α = ω 0 = 1 ;
4:
v 0 = p 0 = 0 ;
5:
for i = 1 , 2 , , do
6:
     ρ i = ( r ¯ 0 , r i 1 ) ; β = ( ρ i / ρ i 1 ) ( α / ω i 1 ) ;
7:
     p i = r i 1 + β ( p i 1 ω i 1 v i 1 ) ;
8:
    Solve y from P τ y = p i ;
9:
     v i = A y
10:
     α = ρ i / ( r ¯ 0 , v i ) ;
11:
     s = r i 1 α v i ;
12:
    Solve z from P τ z = s ;
13:
     t = A z ;
14:
     ω i = ( P τ 1 t , P τ 1 s ) / ( P τ 1 t , P τ 1 t ) ;
15:
     u i = u i 1 + α y + ω i z ;
16:
    if u i is accurate enough then quit;
17:
     r i = s ω i t
18:
end for 
At each iteration of the preconditioned BiCGSTAB algorithm, two matrix-vector products need to be computed, i.e., A v and P τ 1 v , where v represents an arbitrary vector. The matrix–vector product
A v = ( I Δ t D ˜ ( k 1 ) A ˜ ) v = v Δ t D ˜ ( k 1 ) ( A ˜ v ) ,
is calculated as follows:
A ˜ v = ( W ( α ) I + I W ( α ) ) v = ( W ( α ) I ) v + ( I W ( α ) ) v = W ( α ) · V + V · W ( α ) ,
where V = [ v 1 , v 2 , , v N 1 ] , v i R N 1 and v = v e c ( V ) . Obviously, this problem is transformed into a matrix–vector operation with matrix W ( α ) and vector v. On the one hand, the calculation W ( α ) · V is expressed as W ( α ) · [ v 1 , v 2 , , v N 1 ] one by one and costs O ( n log n ) arithmetic operations, where n = ( N 1 ) 2 . On the other hand, the operation V · W ( α ) is computed as W ( α ) · V , and thus, it can be carried out at the same with O ( n log n ) complexity. Similarly, since matrix P can be diagonalized, the preconditioner P τ 1 can be diagonalized too, so the inverse of P τ exists. The preconditioning operation P τ 1 v is as follows:
P τ 1 v = I Δ t d ( k 1 ) P 1 v = ( S N 1 I ) I Δ t d ( k 1 ) ( Λ I + I Λ ) 1 ( S N 1 I ) v .
In conclusion, each iteration of the preconditioned BiCGSTAB iterative algorithm to solve Equation (45) requires O ( n log n ) arithmetic operations in total and can be implemented as a matrix-free manner.
It is worth noting that, although the preconditioner P τ 1 is well defined according to Proposition 1, due to the presence of nonlinear diffusion coefficients and the nonsymmetry of the discretized linear systems (40), it is not simple to derive theoretically the eigenvalue distribution of the preconditioned matrix [33]. However, in the next section, we provide numerical evidence of the favorable clustering of the eigenvalues of the preconditioned matrices for some problems.

5. Numerical Experiments

In this section, two numerical experiments are presented to illustrate the efficiency of the proposed preconditioned semi-implicit difference scheme, and to support our theoretical findings. All experiments are performed in MATLAB R2019b on Intel(R) Core(TM) i5-10210U CPU @ 1.60∼2.11 GHz and 8.00 GB of RAM. On occasion, one can reduce non-homogeneous Dirichlet boundary conditions to homogeneous Dirichlet boundary conditions by using the following transformation:
u ( x , t ) = W ( x , t ) + V ( x , t ) ,
where V ( x , t ) = ϕ ( t ) + [ Φ ( t ) φ ( t ) ] x L x and W ( x , t ) is an unknown function such that W ( 0 , t ) = 0 , W ( L x , t ) = 0 .
Example 1.
For the first problem, we consider Equation (1) with L x = T = 1 , a ( u ) = u 2 , ϕ ( x ) = x 2 ( 1 x ) 2 , φ ( t ) = ψ ( t ) = 0 , and the source term f ( x , t ) is computed as follows:
f ( x , t ) = α ( 1 + t ) α 1 x 2 ( 1 x ) 2 + a ( u ) ( 1 + t ) α 2 cos ( π α / 2 ) { Γ ( 3 ) Γ ( 3 α ) x 2 α + ( 1 x ) 2 α 2 Γ ( 4 ) Γ ( 4 α ) x 3 α + ( 1 x ) 3 α + Γ ( 5 ) Γ ( 5 α ) x 4 α + ( 1 x ) 4 α } .
This problem can be viewed as a modification of [1,24], and its exact solution is u ( x , t ) = ( 1 + t ) α x 2 ( 1 x ) 2 . Denoting by u j n the numerical solution, we define the L -norm errors and the L 2 -norm errors as follows:
E ( Δ t , h ) = max 1 j M u ( i h , j Δ t ) u i j , E 2 ( Δ t , h ) = max 1 j M u ( i h , j Δ t ) u i j ,
respectively, and the space and time convergence orders as follows:
r a t e 1 t = log 2 E ( 2 Δ t , h ) E ( Δ t , h ) , r a t e 1 x = log 2 E ( 4 h 2 , 2 h ) E ( h 2 , h ) , r a t e 2 t = log 2 E 2 ( 2 Δ t , h ) E 2 ( Δ t , h ) , r a t e 2 x = log 2 E 2 ( 4 h 2 , 2 h ) E 2 ( h 2 , h ) .
The data reported in Table 1 and Table 2 refer to the convergence results obtained by the fast preconditioned iterative algorithm and by a direct method (i.e., the backslash “∖” operator in MATLAB) for solving the non-symmetric discretized linear system, respectively. Because of the small size of the linear systems presented in Table 2, the cost of direct solvers is affordable [16]. Table 1 presents the maximum errors and the corresponding convergence orders in time of the semi-implicit difference scheme when N = 512 , for α = 1.20 , 1.50 , 1.90 , respectively. The numerical results show a convergence order approximately equal to 1 in time for the semi-implicit difference scheme. Meanwhile, the experiments reported in Table 2 confirm that the semi-implicit difference scheme is convergent with second-order accuracy in space. These results are consistent with the theoretical analysis presented in Section 2.
Example 2.
For the second problem, we consider the two-dimensional nonlinear Riesz space-fractional diffusion problem, which is modified from [34] as follows:
u ( x , y , t ) t = a ( u ) u ( x , y , t ) α | x | α + u ( x , y , t ) α | y | α + f ( x , y , t ) , ( x , y ) Ω , 0 < t 1 u ( 0 , y , t ) = u ( 1 , y , t ) = u ( x , 0 , t ) = u ( x , 1 , t ) = 0 , 0 t 1 u ( x , y , 0 ) = x 3 ( 1 x ) 3 y 3 ( 1 y ) 3 , ( x , y ) Ω ,
where Ω = ( 0 , 1 ) × ( 0 , 1 ) , a ( u ) = | u | + 10 , and
f ( x , y , t ) = u ( x , y , t ) + a ( u ) 2 cos ( α π 2 ) e t { y 3 ( 1 y ) 3 [ g 1 ( x ) 3 g 2 ( x ) + 3 g 3 ( x ) g 4 ( x ) ] + x 3 ( 1 x ) 3 [ g 1 ( y ) 3 g 2 ( y ) + 3 g 3 ( y ) g 4 ( y ) ] }
where
g i ( x ) = Γ ( i + 3 ) Γ ( i + 3 α ) x i + 2 α + ( 1 x ) i + 2 α , i = 1 , 2 , 3 , 4 .
The exact solution for this problem is u ( x , y , t ) = e t x 3 ( 1 x ) 3 y 3 ( 1 y ) 3 .
The data reported in Table 3 and Table 4 are convergence results obtained with the preconditioned iterative algorithm. Table 3 presents the maximum errors and the corresponding convergence orders in time of the semi-implicit difference scheme when N = 32 , for α = 1.20 , 1.50 , 1.90 , respectively. The numerical results exhibit a convergence order approximately equal to 1 in time for the semi-implicit scheme. Meanwhile, Table 4 shows that the semi-implicit scheme is convergent with second-order accuracy in space. These results are consistent with the theoretical analysis presented in Section 3.
Table 5 shows comparative experiments between a direct method and two fast iterative algorithms for the solution of the discretized linear system at each step of the implicit difference scheme for α = 1.20 , 1.60 , 1.70 , 1.90 . It presents the total number of iterations (Iter) and the solution time (CPU in seconds) when M = N 2 . Obviously, we can see from the table that the two iterative algorithms are significantly faster than a direct algorithm, motivating our quest for fast solvers. As seen from Figure 1, the proposed preconditioner P τ can be indeed efficient to accelerate the convergence of the preconditioned BiCGSTAB method in terms of the clustering eigenvalues, i.e., most eigenvalues of the preconditioned matrices increasingly gather around 1 when we take α 2 .
Since it is proved that the convergence order of the proposed semi-implicit different scheme is O ( Δ t + h 2 ) , the Richardson extrapolation can be used to improve the temporal convergence order. For the sake of clarity, we do the following:
(1)
We apply the numerical scheme (24) with time step Δ t , and we compute the solution U i j n ( h x , h y , Δ t ) . According to the error analysis of the method, the following holds:
u ( x i , y j , t n ) U i , j n ( h x , h y , Δ t ) = O ( h x 2 + h y 2 ) + c Δ t + O ( Δ t 2 ) ,
(2)
Another run with the scheme (24) is carried out using Δ t 2 , and a new solution U i j 2 n h x , h y , Δ t 2 is computed on the fine temporal grid. According to the error analysis, we have the following:
u ( x i , y j , t n ) U i , j 2 n h x , h y , Δ t 2 = O ( h x 2 + h y 2 ) + c Δ t 2 + O Δ t 2 2 ,
(3)
We compute ( 5.6 ) × 2 ( 5.5 ) and define U ˜ i , j n = 2 U i , j 2 n h x , h y , Δ t 2 U i , j n ( h x , h y , Δ t 2 ) :
u ( x i , y j , t n ) U ˜ i , j n = O ( h x 2 + h y 2 + Δ t 2 ) .
This means that Richardson extrapolation indeed can improve the temporal convergence order of the proposed semi-implicit difference scheme from 1 to 2. However, it is always sensitive to the parameter selection used for solving the model problem and thus, it is numerically unstable (we omit the specific numerical results here).

6. Concluding Remarks

In this paper, a fast preconditioned semi-implicit difference scheme for a one-dimensional strongly NSFDE is first presented, which is also extended to solve a two-dimensional NSFDE. In addition, the stability and convergence properties of the new numerical scheme are proved. At the same time, a fast preconditioned iterative algorithm is designed to solve the discretized linear systems on each time step efficiently and rapidly without necessarily storing the coefficient matrix of the linear system at each time level. In the future work, we will design the fast numerical methods and explore their stability and convergence analyses for solving a class of strongly NSFDEs. Moreover, due to the nonlocality of the Riesz fractional derivative, developing the fast matrix-free iterative methods accelerated by suitable preconditioners (e.g., the recent parallel-in-time preconditioners [36,40]) for solving the discretized linear systems should be also very meaningful.

Author Contributions

Funding acquisition, X.-M.G. and H.L.; investigation, Y.-Y.H. and Y.G.; methodology, X.-M.G., Y.G. and B.C.; project administration, X.-M.G., H.L. and B.C.; software, Y.-L.Z.; supervision, X.-M.G. and Y.-L.Z.; visualization, Y.-Y.H.; writing—original draft, Y.-Y.H.; writing—review and editing, Y.-Y.H., X.-M.G., Y.-L.Z., H.L., Y.G. and B.C. All authors have read and agreed to the published version of the manuscript.

Funding

This research was supported by NSFC (Nos. 11801463 and 12101089) and the Applied Basic Research Program of Sichuan Province (No. 2020YJ0007). The last author is a member of the Gruppo Nazionale per il Calcolo Scientifico (GNCS) of the Istituto Nazionale di Alta Matematica (INdAM) and this work was partially supported by INdAM-GNCS under Progetti di Ricerca 2020.

Data Availability Statement

The MATLAB codes for numerical experiments will be available by corresponding authors’ emails.

Acknowledgments

The authors would like to thank anonymous referees for their many valuable comments and suggestions for improving the presentation of this work.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Moroney, T.; Yang, Q. A banded preconditioner for the two-sided, nonlinear space-fractional diffusion equation. Comput. Math. Appl. 2013, 66, 659–667. [Google Scholar] [CrossRef] [Green Version]
  2. Moroney, T.; Yang, Q. Efficient solution of two-sided nonlinear space-fractional diffusion equations using fast Poisson preconditioners. J. Comput. Phys. 2013, 246, 304–317. [Google Scholar] [CrossRef] [Green Version]
  3. Yang, Q.; Liu, F.; Turner, I. Numerical methods for fractional partial differential equations with Riesz space fractional derivatives. Appl. Math. Model. 2010, 34, 200–218. [Google Scholar] [CrossRef] [Green Version]
  4. Ding, H. High-order numerical algorithms for Riesz derivatives via constructing new generating functions. J. Sci. Comput. 2017, 71, 759–784. [Google Scholar] [CrossRef] [Green Version]
  5. Cai, M.; Li, C. On Riesz derivative. Fract. Calc. Appl. Anal. 2019, 22, 287–301. [Google Scholar] [CrossRef]
  6. Metzler, R.; Klafter, J. The random walk’s guide to anomalous diffusion: A fractional dynamics approach. Phys. Rep. 2000, 399, 1–77. [Google Scholar] [CrossRef]
  7. Metzler, R.; Klafter, J. The restaurant at the end of the random walk: Recent developments in the description of anomalous transport by franctional dynamics. J. Phys. A Math. Gen. 2004, 37, R161. [Google Scholar] [CrossRef]
  8. Schneider, W.R.; Wyss, W. Fractional diffusion and wave equations. J. Math. Phys. 1989, 30, 134–144. [Google Scholar] [CrossRef]
  9. Wyss, W. The fractional diffusion equation. J. Math. Phys. 1986, 27, 2782–2785. [Google Scholar] [CrossRef]
  10. Kilbas, A.A.; Srivastava, H.M.; Trujillo, J.J. Theory and Applications of Fractional Differential Equations; Elsevier: Amsterdam, The Netherlands, 2006. [Google Scholar]
  11. Ail, U.; Sohail, M.; Abdullah, F.A. An efficient numerical scheme for variable-order fractional sub-diffusion equation. Symmetry 2020, 13, 1437. [Google Scholar] [CrossRef]
  12. Ilic, M.; Liu, F.; Turner, I.; Anh, V. Numerical approximation of a fractional-in-space diffusion equation, I. Fract. Calc. Appl. Anal. 2005, 8, 323–341. [Google Scholar]
  13. Ilic, M.; Liu, F.; Turner, I.; Anh, V. Numerical approximation of a fractional-in-space diffusion equation (II)–with nonhomogeneous boundary conditions. Fract. Calc. Appl. Anal. 2006, 9, 333–349. [Google Scholar]
  14. Sousa, E.; Li, C. A weighted finite difference method for the fractional diffusion equation based on the Riemann-Liouville derivative. Appl. Numer. Math. 2015, 90, 22–37. [Google Scholar] [CrossRef] [Green Version]
  15. Diethelm, K. An algorithm for the numerical solution of differential equations of fractional order. Electron. Trans. Numer. Anal. 1997, 5, 1–6. [Google Scholar]
  16. Li, M.; Gu, X.-M.; Huang, C.; Fei, M.; Zhang, G. A fast linearized conservative finite element method for the strongly coupled nonlinear fractional Schrödinger equations. J. Comput. Phys. 2018, 358, 256–282. [Google Scholar] [CrossRef]
  17. Zhao, X.; Hu, X.; Cai, W.; Karniadakis, G.E. Adaptive finite element method for fractional differential equations using hierarchical matrices. Comput. Methods Appl. Mech. Eng. 2017, 325, 56–76. [Google Scholar] [CrossRef] [Green Version]
  18. Fu, H.; Liu, H.; Zheng, X. A preconditioned fast finite volume method for distributed-order diffusion equation and applications. East. Asian J. Appl. Math. 2019, 9, 28–44. [Google Scholar] [CrossRef]
  19. Simmons, A.; Yang, Q.; Moroney, T. A finite volume method for two-sided fractional diffusion equations on non-uniform meshes. J. Comput. Phys. 2017, 335, 747–759. [Google Scholar] [CrossRef] [Green Version]
  20. Zeng, F.; Liu, F.; Li, C.; Burrage, K.; Turner, I.; Anh, V. A Crank-Nicolson ADI spectral method for a two-dimensional Riesz space fractional nonlinear reaction-diffusion equation. SIAM J. Numer. Anal. 2014, 52, 2599–2622. [Google Scholar] [CrossRef] [Green Version]
  21. Tian, W.Y.; Zhou, H.; Deng, W. A class of second order difference approximations for solving space fractional diffusion equations. Math. Comp. 2015, 84, 1703–1727. [Google Scholar] [CrossRef] [Green Version]
  22. Li, C.; Deng, W. A new family of difference schemes for space fractional advection diffusion equation. Adv. Appl. Math. Mech. 2017, 9, 282–306. [Google Scholar] [CrossRef] [Green Version]
  23. Hao, Z.-P.; Sun, Z.-Z.; Cao, W.-R. A fourth-order approximation of fractional derivatives with its applications. J. Comput. Phys. 2015, 281, 787–805. [Google Scholar] [CrossRef]
  24. Çelik, C.; Duman, M. Crank-Nicolson method for the fractional diffusion equation with the Riesz fractional derivative. J. Comput. Phys. 2012, 231, 1743–1750. [Google Scholar] [CrossRef]
  25. Yang, J.; Li, Z.; Yan, Y. A new numerical method for solving Riesz space-fractional diffusion equation. Math. Numer. Sin. 2019, 41, 170–190. (In Chinese) [Google Scholar]
  26. Lin, X.-L.; Ng, M.K.; Sun, H.-W. Stability and convergence analysis of finite difference schemes for time-dependent space-fractional diffusion equations with variable diffusion coefficients. J. Sci. Comput. 2018, 75, 1102–1127. [Google Scholar] [CrossRef]
  27. Lin, X.-L.; Ng, M.K.; Sun, H.-W. Crank-Nicolson alternative direction implicit method for space-fractional diffusion equations with nonseparable coefficients. SIAM J. Numer. Anal. 2019, 57, 997–1019. [Google Scholar] [CrossRef] [Green Version]
  28. Farquhar, N.E.; Moroney, T.J.; Yang, Q.; Turner, I.W. GPU accelerated algorithms for computing matrix function vector products with applications to exponential integrators and fractional diffusion. SIAM J. Sci. Comput. 2016, 38, C127–C149. [Google Scholar] [CrossRef]
  29. Lin, X.-L.; Ng, M.K.; Sun, H.-W. A splitting preconditioner for Toeplitz-like linear systems arising from fractional diffusion equations. SIAM J. Matrix Anal. Appl. 2017, 38, 1580–1614. [Google Scholar] [CrossRef]
  30. Huang, X.; Lin, X.-L.; Ng, M.K.; Sun, H.-W. Spectral analysis for preconditioning of multi-dimensional Riesz fractional diffusion equations. arXiv 2021, arXiv:2102.01371. [Google Scholar]
  31. Simmons, A.; Yang, Q.; Moroney, T. A preconditioned numerical solver for stiff nonlinear reaction-diffusion equations with fractional Laplacians that avoids dense matrices. J. Comput. Phys. 2015, 287, 254–268. [Google Scholar] [CrossRef] [Green Version]
  32. Bertaccini, D.; Durastante, F. Efficient Preconditioner Updates for Semilinear Space-Time Fractional Reaction-Diffusion Equations. In Structured Matrices in Numerical Linear Algebra; Bini, D., Di Benedetto, F., Tyrtyshnikov, E., Van Barel, M., Eds.; Springer INdAM Series; Springer: Cham, Switzerland, 2019; Volume 30, pp. 285–302. [Google Scholar]
  33. Huang, X.; Sun, H.-W. A preconditioner based on sine transform for two-dimensional semi-linear Riesz space fractional diffusion equations in convex domains. Appl. Numer. Math. 2021, 169, 289–302. [Google Scholar] [CrossRef]
  34. Jian, H.-Y.; Huang, T.-Z.; Gu, X.-M.; Zhao, Y.-L. Fast compact implicit integration factor method with non-uniform meshes for the two-dimensional nonlinear Riesz space-fractional reaction-diffusion equation. Appl. Numer. Math. 2020, 156, 346–363. [Google Scholar] [CrossRef]
  35. Sun, Z.-Z. Numerical Solutions of Partial Differential Equations, 2nd ed.; Science Press: Beijing, China, 2012. (In Chinese) [Google Scholar]
  36. Zhao, Y.-L.; Zhu, P.-Y.; Gu, X.-M.; Zhao, X.-L.; Jian, H.-Y. A preconditioning technique for all-at-once system from the nonlinear tempered fractional diffusion equation. J. Sci. Comput. 2020, 83, 10. [Google Scholar] [CrossRef] [Green Version]
  37. Zhou, Y. Application of Discrete Functional Analysis to the Finite Difference Methods; International Academic Publishers: Beijing, China, 1990. [Google Scholar]
  38. Van der Vorst, H.A. Bi-CGSTAB: A fast and smoothly converging variant of Bi-CG for the solution of nonsymmetric linear systems. SIAM J. Sci. Stat. Comput. 1992, 13, 631–644. [Google Scholar] [CrossRef]
  39. Bini, D.; Di Benedetto, F. A new preconditioner for the parallel solution of positive definite Toeplitz systems. In Proceedings of the Second Annual ACM Symposium on Parallel Algorithms and Architectures, SPAA ’90, Island of Crete, Greece, 2–6 July 1990; AMC: New York, NY, USA, 1990; pp. 220–223. [Google Scholar] [CrossRef]
  40. Gu, X.-M.; Zhao, Y.-L.; Zhao, X.-L.; Carpentieri, B.; Huang, Y.-Y. A note on parallel preconditioning for the all-at-once solution of Riesz fractional diffusion equations. Numer. Math. Theor. Meth. Appl. 2021, 14, 893–919. [Google Scholar]
Figure 1. The eigenvalue distributions of both original and preconditioned matrices with N = 2 6 , M = N 2 , first line: α = 1.2 ; second line: α = 1.5 ; third line: α = 1.9 ).
Figure 1. The eigenvalue distributions of both original and preconditioned matrices with N = 2 6 , M = N 2 , first line: α = 1.2 ; second line: α = 1.5 ; third line: α = 1.9 ).
Fractalfract 05 00230 g001
Table 1. The errors and temporal convergence orders when N = 512 .
Table 1. The errors and temporal convergence orders when N = 512 .
α M    E ( Δ t , h )     rate 1 t     E 2 ( Δ t , h )     rate 2 t
81.6125e − 038.6746e − 04
167.9547e − 041.01944.2894e − 041.0160
1.20323.9507e − 041.00972.1329e − 041.0080
641.9688e − 041.00481.0635e − 041.0040
1289.8288e − 051.00225.3109e − 051.0018
84.7362e − 032.6365e − 03
162.3256e − 031.02611.2984e − 031.0219
1.50321.1525e − 031.01286.4439e − 041.0107
645.7374e − 041.00633.2102e − 041.0053
1282.8628e − 041.00301.6023e − 041.0025
81.2993e − 027.4538e − 03
166.3112e − 031.04173.6319e − 031.0373
1.90323.1120e − 031.02011.7935e − 031.0179
641.5455e − 031.00988.9131e − 041.0088
1287.7025e − 041.00474.4437e − 041.0042
Table 2. The errors and spatial convergence orders when M = N 2 .
Table 2. The errors and spatial convergence orders when M = N 2 .
α N    E ( Δ t , h )     rate 1 x     E 2 ( Δ t , h )     rate 2 x
41.1988e − 036.3875e − 04
82.9694e − 042.01331.5674e − 042.0269
1.20167.4054e − 052.00353.9116e − 052.0025
321.8502e − 052.00099.7746e − 062.0006
644.6248e − 062.00022.4434e − 062.0001
43.5560e − 031.9379e − 03
88.7005e − 042.04894.7737e − 042.0213
1.50162.1633e − 042.01201.1889e − 042.0055
325.4010e − 052.00302.9695e − 052.0013
641.3498e − 052.00077.4219e − 062.0004
41.0923e − 026.1965e − 03
82.6106e − 032.06491.5174e − 031.9285
1.90166.4524e − 042.01653.7556e − 041.9873
321.6085e − 042.00419.3655e − 051.9954
644.0184e − 052.00102.3399e − 051.9992
Table 3. The errors and temporal convergence orders when N = 32 .
Table 3. The errors and temporal convergence orders when N = 32 .
α M    E ( Δ t , h )     rate 1 t     E 2 ( Δ t , h )     rate 2 t
1.2041.8489e − 056.3044e − 06
89.4453e − 060.96903.2208e − 060.9689
164.7734e − 060.98461.6277e − 060.9846
322.3998e − 060.99218.1828e − 070.9922
1.5041.8488e − 056.3044e − 06
89.4452e − 060.96893.2207e − 060.9690
164.7732e − 060.98461.6276e − 060.9846
322.3996e − 060.99228.1822e − 070.9922
1.9041.8489e − 056.3045e − 06
89.4456e − 060.96903.2208e − 060.9690
164.7737e − 060.98451.6278e − 060.9845
322.4001e − 060.99208.1838e − 070.9921
Table 4. The errors and spatial convergence orders when M = N 2 .
Table 4. The errors and spatial convergence orders when M = N 2 .
α N    E ( Δ t , h )     rate 1 x     E 2 ( Δ t , h )     rate 2 x
1.2044.7734e − 061.6181e − 06
81.2036e − 061.98774.1036e − 071.9793
163.0226e − 071.99431.0299e − 071.9943
327.6388e − 081.98442.6021e − 081.9855
1.5044.7732e − 061.6180e − 06
81.2034e − 061.98784.1030e − 071.9795
163.0205e − 071.99431.0298e − 071.9943
327.6178e − 081.98732.5954e − 081.9883
1.9044.7738e − 061.6182e − 06
81.2039e − 061.98744.1046e − 071.9791
163.0264e − 071.99201.0314e − 071.9926
327.6769e − 081.97902.6120e − 081.9814
Table 5. Total number of iterations and CPU time (in seconds) when M = N 2 .
Table 5. Total number of iterations and CPU time (in seconds) when M = N 2 .
α NDirectI P τ
CPUIterCPUIterCPU
160.02615.00.0414.00.021
321.63524.00.3504.00.094
1.2064130.98237.02.3545.00.532
12823,599.05950.441.2075.05.248
256out of memory59.9844.7155.038.040
160.02418.00.0444.00.023
321.61433.00.4794.00.094
1.6064117.90559.43.7644.00.430
12824,804.13795.178.6854.84.962
256out of memory154.3706.8245.038.935
160.02019.00.0503.00.016
321.56236.00.5224.00.096
1.7064751.89664.04.0654.00.428
12832,105.020109.489.9024.04.185
256out of memory188.2858.8534.030.393
160.02120.00.0513.00.017
321.68943.70.6363.00.071
1.9064263.74275.34.7854.00.426
12822,645.044144.7387.9914.04.211
256out of memory273.05633.4584.030.324
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Huang, Y.-Y.; Gu, X.-M.; Gong, Y.; Li, H.; Zhao, Y.-L.; Carpentieri, B. A Fast Preconditioned Semi-Implicit Difference Scheme for Strongly Nonlinear Space-Fractional Diffusion Equations. Fractal Fract. 2021, 5, 230. https://doi.org/10.3390/fractalfract5040230

AMA Style

Huang Y-Y, Gu X-M, Gong Y, Li H, Zhao Y-L, Carpentieri B. A Fast Preconditioned Semi-Implicit Difference Scheme for Strongly Nonlinear Space-Fractional Diffusion Equations. Fractal and Fractional. 2021; 5(4):230. https://doi.org/10.3390/fractalfract5040230

Chicago/Turabian Style

Huang, Yu-Yun, Xian-Ming Gu, Yi Gong, Hu Li, Yong-Liang Zhao, and Bruno Carpentieri. 2021. "A Fast Preconditioned Semi-Implicit Difference Scheme for Strongly Nonlinear Space-Fractional Diffusion Equations" Fractal and Fractional 5, no. 4: 230. https://doi.org/10.3390/fractalfract5040230

Article Metrics

Back to TopTop