Natural Science Journal of Xiangtan University 2008 have any of several connotations found in the literature.Two closely related ones are: (a)the minimum error criterion,i.e.an optimal mesh reduces the error to a minimum for a pre— scribed number of meshes. (b)the maximum efficiency criterion,i.e.an optimal mesh achieves a specified error tolerance with the minimum number of meshes.merate Our discussion is partitioned into sections on the following topics.In Sect.2,we demonstrate the con— struction of an optimal mesh for controlling the derivative error for quadratic data function using linear inter- polant and extend these idea to deal with general function belonging to C2.In Sect.3,we use this mesh gen— eration to finite element method,a iterative method to move mesh by estiatimng derivative error is designed. In Sect.4,the numerical example showed our theory and algorithm is effective. 2 Optimal meshes for minimizing the derivative error In this section,we first introduce the linear transform for simplifying the derivative error function for linear approximation to a nondegenerate quadratic date function based on interpolation at the verti— ces of a general mesh.We can write the data function as 厂(z)=ax。+bx+c, where a,6,c are constants.Let p(z)be the linear interpolant of厂(z)on the element of T一[z 1,zi], then p(x卜1)=f(x 1),p(xf)=f(x ). Let function E(z)be the interpolation error function, E(z)一厂(z)一P(z)=ax。+d1z+d2. We have E(x 1)=E(xf)=0,E(z)is a quadratic function and its level curve form a conics with a center z .Let£=E(x ),then d1=一2ax 。It is shown that the error at a displacement dx from the center can be expressed as E(z)一E(x +dx)=e+a(dx)。. We define the error in the derivative estimate based on p(z), (1) EG(z)=l P (z)一/(z)l=l(p(z)一厂(z)) l=l E (z)l=l 2adx 1. By(1)we have EG(z )=0.Introduce a coordinate transformation 一2adz,三一2ax,主 一2ax , (2) then E (z)=l 2adx l—l d主l= (主), E(z)=£+口(dz)。一£+ ( )= (主). ‘士a (3) (4) Let be the image of T under the transformation of(2). It was pointed out that the interpolation error vanishes at the vertices.For a given error tolerance ,we have Ec(z) ,z∈T.Let 2adz=d = ,name1y the derivative error at the vertices on t is the tolerant error.By(2),t=[主 一r,主 十r], and this element has the maximal length among all element meeting the error tolerance.We can con— struct meshes by using this element to cover the region such that the number of element is not larger than the number required by any other elements that meets the same tolerance for the derivative error. It can be inverse transformed by(2)to the desired optimal meshes for derivative error tolerance based on the maximal criterion. As we now seek to extent the optimal mesh construction to more general data function,we ob— serve that locally a C。functionbehaves as a quadratic given by its Taylor expansion, 厂(z+dz)≈lr(z)(dz)。+/(z)(dz)+厂(z)+。(dz)。, (5) 维普资讯 http://www.cqvip.com No.2 YANG Yin et a1 Moving Mesh Method For Minimizing DeriVatiVe Error 3 where厂,( ), ( )are the first and sec。nd derivatiVe。f厂(z)eValuated at ・ On elemen 丁'le 痧( ) the‰e吖interp。 讥t。f f(x)。n the element。f丁一[ 1,z ], hen E( )一厂( )一 (z)一 1(z—z)(x--x ̄) ・ 卜 Letz 一( i+ 卜1)/2,then E( )一E( +d )一号(( +如)一 卜-)(( c+d )一z ) 一 专((zc—z卜-)+dz)((zc—z +d 一 [( 一z卜。)( 一 )+(( 一 卜 )+( 一 ))如+( )。]/一E( c)+专(d )。 '‘6 ( )一EG( +dx)一I f"(x)dx I—I d;I, where d;一J 0 . (7) + ( )d .F。r a given error tolerance r,Ec( )一I d;I r,let 一/( )妇,namely ;( )一r ( )dz,when I d;I—I; 一; I—I r ( )dx I—I/(z )一 ( )I—r, construct optimal meshes based on the maximal criterion. (8) element丁一[ 卜1, f j has the maxima1 length am。ng a11 element meeting the error t。1erance.we can + Theorem: Let f and户斗1 are the linear interpolation on element丁f and T.-1 respectively'set hi— zf— 1,then we have /cz 一南‰ Proof +0(^。). P i: f(x )一厂( . 件l 二 兰 斗 厂( 斗 )一厂( )+^斗。/( )+ ^ -厂( r)+ ^ -厂( ), ( 卜。)一厂( )一h 厂 ( )+ ^;。 ( ;)一 。厂( ), /( 一 /( 一 /( )一 hifl(x,) 一扣- ( 一扣-。厂( ), 一专h ( 一 ( )・ ( 件1+h i)^件1 + 。h(^件l+ )^ l + where ∈( f, 斗1), ∈( 卜1,Xi).Coupling above four equation'we have 专厂( ( 一 )一 ( 厂( ) +o(h。). 一 hip 斗 hi+1+hf 3 Moving mesh method based on derivative error ‘ In Sect.2,we have constructed an optimal mesh for controlling the derivative error. It is the ba— sis of grid adaptation algorithm.Roughly speaking,for a given ,we will adapt OUr grids in such a way that(8)will be better and better satisfied.In this section,we want to use this idea to gene at the mesh to so1ve PDEs with standard finite element methods. We consider the one dimension partial differential equation 维普资讯 http://www.cqvip.com
4 Natural Science Journal of Xiangtan University 2008 fLM一厂 l“ g in n on n, (9) where L is a partial differential operator.Let n=a,6],“^(z)一∑“ 乒 (z)is the linear finite ele— f=0 ment solution of the PDE on meshnN一{a=xo<x1<x2……<xN=b).“^(z)is a piecewise linear function.Since taking piecewise second derivatives to piecewise linear functions will given no approxi— mation to hessian function,special Postprocessing techniques need to be used to obtain reasonable Hessian function approximation from linear finite element.One most popular technique is patch recov— cry technique proposed by Zikienwicz—Zhu[ . Let 0"h z一“ ^(z),where“ ^(z)is a distributional derivative,i.e.,piecewise derivative.Let zf and主斗1 be the centers of the element J =[z卜1,xf]and J升1=[z。,z汁1]respectively.We construct a polynomial (z)一a1+a2x such that F(a。,口2)一∑( (z,)一 (z,)) =min, (10) J f,升1 and fu nctio N (z)一∑j& (z) ( ), (11) i=0 where (z)is the basis function corresponding to the interior node xi.Then (z)is the recovery continuous derivative of“^(z). Based on the minimum error criterion,an optimal mesh for derivative error with a prescribed nutuber of elements is the mesh satisfying 1(z )一 (z )1—1 (z )一 (z )1一…一1 (z )一 (z ・)1. (12) (8)provides a choice of monitor function of moving mesh methods,i.e.M(z)一 (z).The grid Doints xi adj ust in such a way that the same change in the solution“ occurs over each grid interval (z 1,z。).But the disadvantage is the fact that,if“ 0,then hl—xi—Xi,-1 。。.An alternative choice forMis to takeM(z)一1+l“ (z)l,in this case,(12)is changed to f -(z斗1)一 (z )f+h斗1一constant,h升1一x斗1一zi, i一0,1,2,…,N一1. (13) In Dractica1 computation,the problem(9)is solved iteratively.We construct the following itera— tion algorithm: SteD 1 Initialize mesh:the initial mesh{z ∞f z;∞= /N,i一0,1,2,…,N}is uniform. SteD 2 For k一0,1,…,assuming that the mesh{z }is given,compute the linear finite element solution{“ }on{z }. SteD 3 RecOvery the derivative of“ (z)by ZZ superconvergent patch recovery methods,such that N+I Z =l 。 的(z ’)一 D(zH‘的)I+^ ,L 一 ∑L 娃 ]. 一1 Step 4 Test mesh:Let Co be a user—chosen constant with oC>1' if max l /L C0/N hole true。then go to step 6,otherwise,continue to step 5. Step 5 C-enerate the new mesh by(13).Our new mesh is then defined to be{ ㈣}・Return to step 2・ Step 6 Set z 一z; and ui*一H; then stop. 4 Numerical experiment We consider the following partial differential equation(PDE): f—f“ (z)一“ (z)=0, z∈(0,1), I“(0)一0,“(1)一1, (14) which has the exact solution“(z)一(1一e- 。)/(1一e ). 维普资讯 http://www.cqvip.com No.2 YANG Yin et al Moving Mesh Method For Minimizing Derivative Error 5 For f《1 the solution has a boundary layer of thickness 0(0 near the boundary X一0,it is advanta— geous to use a mesh that concentrates nodes in the boundary layer.We compute(1 4)use the algo— rithm in section 3 and define E ll一 )一 ^(zf)l 1/2 一lOg ( ), N N )一 (z )I 1/2 ll一 ∑(I 。(z f=1 ,rate—log2 N where u(x )i∑ s the value of the exact solution (z)at X , 。(z )is the value of 。(z)at Xf and (zf) / is the value of (z)at X . / z Tab.1 ErrorforM(z)一1+l (z)l Tab.2 Derivative error for M(z)=1+l (z) An other choice ofMis arc—length function,i.e.M(z)一、/r干而,see[2,3,6,8,91.Tab.3 and Tab.4 is the numerical results using the same iterative algorithm. Tab.3 Error for M(z)一~/『干 厂 The following figures illustrate the pointwise error and derivative error using M一1+I (z)I and M(z)一 干而reSpectively.Comparing the tables and figureS,they show that pointwise e卜 ror and derivative error using the monitor function M一1+I (z)I are smaller than using the monitor function M(z)一、/r干而. 维普资讯 http://www.cqvip.com
Natural Science Journal of Xiangtan University 2008 Fig.1 When f=0.01 and N=256.the pointwise error(1eft)and derivative error(right)using M( )=1+I ( )I andM( )=v,f 7 Referenees — = ∞ g口 [13 AZEVEDO E F D.Optimal triagular mesh generation by coordinate transformation[J].SIAM J SCI Stat Comput,1991,13(4)l 755—786. [2] CHEN Y.Uniform ointwipse cOnvergence for a singularly perturbed problem using arc-length equidistribution[J].Journa1 of Com— putational and Applid Mathematiecs,2003,159l 25—34. [3]CHEN Y.Uniform convergence analysis of finite difference approximations for singular perturbation problems 0n an adapted grid [J].Advances in Computational Mathematics,2006,24;197—212. [4]HUANG W Z,REN Y.Russell,Moving mesh partial differential equations(MMPDES)based on the equidistribution principle [J].SIAM J NumerAnal,1994,31:709—730. [53 NATALIA Kopteva.Maximum Norm A Posterior/Error Estimates for a one-dimensional convection—diffusion problem[刀.SIAM J Numer Ana1,2001,39(2):423—441. [63 NATALIA Kopteva,MARTIN Stynes.A robust adaptive method for a quasi—linear one-dimensional convection—diffusion problem. [J].SIAM J Numer Anal,2001,39(4):1 446—1 467. [73 MILLER K,N MILLER R.Moving finite element I[33.SIAM J Numer Anal。1981,(6)l1 019—1 032. [83 QIU Y,SLOAN D M,TANG T.Numerical solution of a singularly perturbed two-point boundary value problem using equidis— tributionlanalysis of convergence[J].J Comput Appl Math,2000,116l 121~143. [93 LINSS T.Uniforming pointwise convergence of finite different schemes using grid requidistribution[J].SIAM J Numer Anal, 2001,39l 423—441. [10]TANG H Z,TANG T,ZHANG P W.An adaptive mesh redistribution methods for non—linear Hamilton—Jacobi equations in tWO- and three-dimensions[J].J Comput Phys,2003,188:543—572. [113 ZIENKIEWICZ O c,ZHU J Z.The superconvergent Patch Recovery and a Posteriori Error Estimates.Partl:The Recovery Technique[J].J Numer Methods Engrg,1992,33:1 331—1 364. 责任编辑:龙顺潮
因篇幅问题不能全部显示,请点此查看更多更全内容