Convergence and supercloseness of a finite element method for a two-parameter singularly perturbed problem on Shishkin triangular mesh

https://doi.org/10.1016/j.amc.2021.126753Get rights and content

Highlights

  • We consider a two-dimension singularly perturbed elliptic problem with two parameters.

  • We combine the characteristics of the mesh and the layers to choose technical tools for error analysis.

  • Using some important integral inequalities, we prove the convergence and supercloseness of finite element method on Shishkin triangular mesh.

Abstract

We consider a singularly perturbed elliptic problem with two parameters in two dimensions. Using linear finite element method on a Shishkin triangular mesh, we prove the uniform convergence and supercloseness in an energy norm. Some integral inequalities play an important role in our analysis. Numerical tests verify our theoretical results.

Introduction

We discuss the following singularly perturbed problem in the interval Θ:=(0,1)×(0,1)ε1Δz+ε2b(x)zx+c(x)z=f(x,y),z|Θ=0,under the condition 0<ε1,ε21 and for all x[0,1] there arec(x)λ>0,b(x)ϕ>0,and12ε2b(x)+c(x)ψ>0,where ϕ, λ, ψ are positive constants. Assume b, c and f sufficiently smooth. Moreover, assume that the funciton f satisfies the compatibility conditionsf(0,0)=f(0,1)=f(1,1)=f(1,0)=0.From these assumptions and [1] we obtain that problem (1) has a unique solution zC3,α0(Θ¯) for 0<α0<1.

Due to the influence of small parameters in the equation, boundary layers or inner layers often appear in singularly perturbed problems. Unfortunately, classical numerical methods are useless for these problems because the solution changes so quickly within the layer. In recent years, layer-adapted mesh has been developed to analyze singular perturbed problems. And one of the most popular layer-adapted meshes is Shishkin mesh, which is simple in structure and easy to analyze.

In general, one issue of numerical methods for singular perturbation problems is to obtain uniform convergence independent of perturbation parameters. It is worth noting that with the development of research, researchers have also discovered the importance of supercloseness, which is a useful tool to improve numerical solutions by postprocessing. For singularly perturbed problems with one parameter, there are plenty of supercloseness results; see [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13] and their bibliography. But, few studies have been done for singular perturbation problems with two parameters; see [14], [15].

In this article, for two-dimensional singular perturbation problems with two parameters, we prove the convergence and supercloseness of finite element method on Shishkin triangular mesh. Compared with the one-parameter problems, layers of two-parameter problems are more complex, as makes the analysis more difficult. The analysis shows that there are four corner layers, two exponential boundary layers and two parabolic layers in the solution of problem (1). Their properties are different and closely related to small parameters ε1, ε2 and mesh properties. Therefore, we combine the characteristics of the mesh and the layers to choose technical tools for error analysis. At the same time, some integral inequalities on Shishkin triangular mesh are also of great significance in error analysis. Finally, we obtain the main theoretical results of this paper.

The organization of this article is as follows. In Section 2, we present some information about the exact solution and some necessary assumptions. In Section 3, the Shishkin triangle mesh is built, and some information about the mesh is given. Besides, we provide the variational formulation and finite element method for the problem (1). In Section 4, the error results of exact solution and its interpolation in L norm and energy norm are proved. We present some important integral inequalities in Section 5, while in Section 6 our main theoretical results on supercloseness and uniform convergence are given. In the last section, we illustrate and support the theoretical results with numerical examples.

For nonnegative integers k, 1p, and any measurable subset G of Θ, a standard notation for Banach spaces Lp(G) is given, and we denote by Wk,p(G) the Sobolev space and Hk(G)=Wk,2(G) on G. We write by ·G the norm of ·L2(G) and write by |·|1,G the seminorm of ·H1(G), respectively. The notation (·,·)G stands for the scalar product in L2(G). Furthermore, we will omit the G in the norm and seminorm when it is Θ. Throughout this paper, C represents a general constant independent of the perturbation parameters ε1, ε2 and the number of mesh points.

Section snippets

The decomposition of solution and its prior information

The layers of two-parameter problems are more complex than ones of one-parameter problems. In order to better analyze the exponential layer with faster change speed, similar to [16], we introduce two constants that determine the width of the exponential boundary layersχ0=ε22bM2+4ε1λε2bM2ε1,χ1=ε22ϕ2+4ε1λ+ε2ϕ2ε1,where bMb(x) for all x[0,1]. From [16] we have the following inequalities about χ0 and χ1:max{ε1χ1,χ01}C(ε112+ε2),χ0χ1,ε2(ε1χ1)12Cε212,ε2χ0ϕ1cL.

Remark 1

It is noted that the values of

Shishkin triangular mesh

Based on Assumption 1, we take p=1/2. To construct the Shishkin triangle mesh we define the following appropriate transition parametersσi:=min{14,2τχilnN}i=0,1,σy:=min{14,τδε1lnN},where NN,N8, is multiple of 4, τ is the parameter selected by the user and δ is the parameter defined in Assumption 1.

Next, we takeσi:=2τχilnNi=0,1,σy:=τδε1lnN,which are the usual transition points used in Shishkin mesh for this kind of problem. Then similar to [16, (21)], we can build our Shishkin triangular mesh TN

Interpolation error

First, from [19, Lemma 3.2], we have the following lemma about anisotropic interpolation estimates.

Lemma 1

Assume that gW2,p(Θ) and gI is the standard nodal linear interpolation of g on Shishkin triangular mesh TN. ThenggILp(Ki,jr)Cl+m=2hx,ilhy,jm2gxlymLp(Ki,jr),(ggI)xLp(Ki,jr)Cl+m=1hx,ilhy,jm2gxl+1ymLp(Ki,jr),(ggI)yLp(Ki,jr)Cl+m=1hx,ilhy,jm2gxlym+1Lp(Ki,jr),where p(1,], l and m are nonnegative integers, Ki,jrTN with i,j=0,,N1 and r=1,2.

Lemma 2

Let Assumption 1 be true and EI

Integral inequalities

We will give some interpolation integral inequalities that are essential for convergence and supercloseness analysis.

First, the estimates of diffusion terms rely on the lemma as follows(see [7, Lemma 2.1]).

Lemma 6

Assume that gC3(Θ¯) and gI is the standard nodal linear interpolation of g on Shishkin triangular TN. Then we can obtain thatSi,j(ggI)yvyNdxdyCl+m=2hx,ilhy,jm3gxlym+1L(Si,j)vxNL1(Si,j),Qi,j(ggI)xvxNdxdyCl+m=2hx,ilhy,jm3gxl+1ymL(Qi,j)vxNL1(Qi,j),where l and m are

Supercloseness and uniform convergence

In order to estimate the bilinear form, integration by parts is used to obtaina(zzI,vN)=ε1((zzI),vN)+ε2(b(zzI)x,vN)+(c(zzI),vN)=ε1((zzI),vN)ε2(zzI,bvxN)+((cε2bx)(zzI),vN).The estimates of the three terms at the right will be given in the following three lemmas.

Lemma 8

Let zIVN be the standard nodal linear interpolation of the solution z of (1) and zNVN be the numerical approximation of z. Let Assumption 1 hold true, then one hasε1|((zzI),vN)Θs|C(ε11/2N3/2+(ε2+ε11/2)1/2N5/2)vNE,ε1|

Numerical experiments

In this section, we will give some numerical examples to verify the theoretical results obtained in the previous section.

Our test problem is given byε1Δz+ε2(2x)zx+z=f(x,y)inΘ,z=0onΘ,where the right end item is chosen such that exact solution meetsz(x,y)=(1ey/ε1)(1e(1y)/ε1)(1eχ0x)(1eχ1(1x))4.And from (5) we haveχ1=ε22+4ε1+ε22ε1,χ0=ε22+ε1ε2ε1.

In our tests, we choose δ=1/4,τ=5/2,N=23,,29. In order to satisfy conditions (14) and (11), there is a perturbation parameter range M(ε1,ε2)={

Acknowledgments

We thank the anonymous referees for his/her valuable comments and suggestions that led us to improve this paper.

References (22)

  • S. Xie et al.

    Supercloseness result of higher order FEM/LDG coupled method for solving singularly perturbed problem on S-type mesh

    Abstr. Appl. Anal., pages Art. ID 260840

    (2014)
  • Cited by (1)

    Funding: The work of the author is supported by NSFC grant 11771257 and 11601251.

    View full text