首页 高维抛物方程

高维抛物方程

举报
开通vip

高维抛物方程高维抛物方程 目录 摘要................................................................................................................................. 2 关键字 ................................................................................................................

高维抛物方程
高维抛物方程 目录 摘要................................................................................................................................. 2 关键字 ............................................................................................................................. 2 前言................................................................................................................................. 4 ?1 数学模型 ................................................................................................................. 4 ?2 数值格式................................................................................................................... 5 ?2.1二维抛物方程五点差分,PR格式 ..................................................................... 5 ?2.2七点差分格式,Douglas格式 .......................................................................... 6 ?3 程序设计与数值试验 ............................................................................................. 8 ?3.2二维抛物方程实例 ........................................................................................... 8 ?3.2三维抛物方程实例 ......................................................................................... 10 ?4 收敛性 分析 定性数据统计分析pdf销售业绩分析模板建筑结构震害分析销售进度分析表京东商城竞争战略分析 与误差分析 ........................................................................................... 12 ?4.1.收敛性分析 .................................................................................................. 12 ?4.2. 误差分析:................................................................................................. 15 ?4.2.1二维抛物方程 ......................................................................................... 15 ?4.2.2三维抛物方程 ............................................................................................ 17 谢辞............................................................................................................................... 19 参考文献........................................................................................................................ 20 翻译............................................................................................................................... 21 原文............................................................................................................................... 21 译文............................................................................................................................... 24 高维抛物方程数值解法 .................................................................................................. 24 附录............................................................................................................................... 27 五点差分格式源码.......................................................................................................... 27 ADI法编程源码 .............................................................................................................. 29 Douglas法编程源码 ....................................................................................................... 30 高维抛物方程有限差分方法 谢伟伟 山东大学数学学院 摘要 抛物型方程是偏微分方程中基本方程之一(在自然科学的众多领域中,许多现象是用抛物型方程或者方程组描述的例如热传导以及其它扩散现象、化学反应、粒子的运输等等(另外在一些问 快递公司问题件快递公司问题件货款处理关于圆的周长面积重点题型关于解方程组的题及答案关于南海问题 的数值处理中也经常出现抛物型偏微分方程(在现代数值方法中,最早为人们所注意且理论分析完善的是有限差分法,因此抛物型偏微分方程的有限差分方法一直是人们关心的焦点.本文主要介绍二维、三维抛物方程几种典型的差分格式并分别于解析解作比较,作为理论的参考。文章第一节介绍抛物方程二维,三维数学模型,第二节介绍两种问题的数值格式,第三节程序设计与数值试验最后得出结论。 关键字:抛物型方程 二维问题 三维问题 有限差分方法 Finite difference method of high_dimensional Parabolic equation Xie,Weiwei (Major in Information & Computational Sciences, and Directed by Professor Aijie, Chen) Abstract: Parabolic equation is one of the basic equations of partial differential equations. In many areas of natural science, many phenomena is described with parabolic equation, such as heat conduction and other diffusion phenomenon, chemical reaction, particle transport etc. Also, parabolic equations often appear in the numerical treatment of some problems. The earliest method which has been the focus of concern of people is finite difference method, thus parabolic partial differential equation of finite difference methods has been the focus of concern. This paper mainly introduces the 2D, 3D parabolic equations of several typical difference schemes. Article introduced two-dimensional parabolic equation in the first quarter, three-dimensional mathematical model, introduced two problems in the second quarter of numerical format, the third quarter and the numerical test program design conclusion. Key words: parabolic equations two-dimensional problem three- dimensional problem finite difference method 前言 在研究热传导过程,气体扩散现象和电磁场的传播等问题时,常常遇到抛物 ,u型偏微分方程:,其中d是定义在上的复函数,系dcuaufin,,,,,,,(),,,t 数c,a 和 f 依赖时间 t . ?1 数学模型 在二维情况下,其模型方程为 22,,,,uuu,,,,,,afxytoxylt()(,,),,,0,,22,,,txy,,,uxyoxy(,,)(,),, , ,uytulyt(0,,)(,,)0,,,, uxtuxlt(,0,)(,,)0,,,,, 其中有参数. af,,, 在三维情况下,其模型问题为 222,,,,uuuu,,,,,,,,afxyztxyztT()(,,,), (0,,1,0)222,,,,txyz ufyztufyztyztT,,,,,,(,,),(,,),(0,1,0)xx,,0112 ugxztugyztxztT,,,,,,(,,),(,,),(0,1,0) yy,,0112 uyztuyztxytT,,,,,,(,,),(,,),(0,1,0),,zz,,0112 u,(xyzxyz,,),(0,,1),,,t,0 afffgg,,,,,,,,,,其中有参数. 121,212 ?2 数值格式 ?2.1 二维抛物方程五点差分,PR格式 对上述二维抛物方程, 建立一个关于空间步长和时间步长的函数,这样就可 ,以把初始区域划分为一个网格图。取空间步长h=l/M, 时间步长>0,做两族平 行于坐标轴的网线:x=x(j)=jh,y=y(k)=kh,j,k=1,2„„„„„„M,将区域 20,,,xylM分割成个小矩形. (一)五点差分格式: nnnnnnn,,,,,,111111uuuuuuu,,,,,4jkjkjkjkjkjkjk,,1,1,,1,1,,,,,,,af,jk,2,h 1 u,,,jkjk,, nnnnuuuu,,,,0,1,1,,1,1kMkjjM,, 2rah,,/ 关于同志近三年现实表现材料材料类招标技术评分表图表与交易pdf视力表打印pdf用图表说话 pdf 示网比,则将上式改为便于计算的形式,使第n层值在等式右以 边,第n+1层在等式左边,则得 nnnnnn,,,,,,111111ufuuuurru,,,,,,,,*()*(14)., jkjkjkjkjkjkjk,,1,1,,1,1,,,,, (二) PR格式 它是Peaceman和Rachford于1955年提出的,他们把第n层到n+1层得计算 uu分成两步:第n到n+1/2层,对用向后差分逼近,对用向前差分逼近,yyxx uu然后第第n+1/2到n+1层,对用向前差分逼近,对用向后差分逼近,yyxx 得道PR格式: 1nn,2,1/2n,12nuujkjk2,,(),xyu2,ujk,jk/2,h nn,,11/2,n,1/2121n,uujkjk2,,(),xyu2ujk,jk,/2,h t,其中j,k=1,2„„„M,n=0 1,2„„„„,上标n+1/2表示t==(n+1/2) 取1n,2 1,nn2uu值。假定第n层得已求的,则由(1)知求的 ,这只需按行求一些具有jkjk n,1u三对角系数矩阵即可。再有(2)求出这只需按列求一些三对角矩阵的方程jk 组,这些是很容易求的。其边值问题是非其次的,在边界上u=,则过度,(,,)xyt1,n2层=1/2(+). ,(,,)xyt,(,,)xytijnijn,1ujk ?2.2七点差分格式,Douglas格式 对上述三维抛物方程,建立一个关于空间步长和时间步长的函数,这样就可以 ,把初始区域划分为一个网格图。取空间步长h=l/M, 时间步长>0,做三族平行 于坐标轴的网线:x=x(i)=ih,y=y(j)=jh,z=z(k)=kh, i,j,k=1,2„„„„„„M,将区 30,,,,xyzl域分割成个小正方体。 M (一) 七点差分格式 nnnnn,,,,1111uuuuu,,,2ijkijkijkijkijk,,,,1,,,,1,,,,,,a2h, nnn,,,111uuu,,2ijkijkijk,1,,,,1,,,,,a2h nnn,,,111uuu,,2ijkijkijk,,1,,,,1,,,,af,ijk,,2h njknnjkn,,,,ufuf,,,,1,,11,,2jkMjk, nikn,,nikn,,ugu,,,g,ikiMk,1,1,1,,2 nijnnijn,,,,uu,,,,,,ijijM,,11,,12, u,,,ijkijk,,,, 其中i,j,k=1,2„„M-1,n=0,1„„„„N-1将上式改为 nnnnnnn,,,,,,,1111111(16)(),,,,,,,ruruuuuuuijkijkijkijkijkijkijk,,,1,,1,,,1,,11,,1,,,,,,,, n ,,uf,ijkijk,,,, 差分方程转化为矩阵方程,并分别对x,y,z轴进行计算,其系数矩阵分别是三对角矩阵且严格对角占优,故总是可解. (二) Douglas格式 由于PR格式不可推广到三维的情形,J.Douglas提出的一种在三维上交替方向隐格式,他和PR格式一样,在误差阶,计算量和稳定性方面令人满意。 11,21/322nnn,n,n,,,(())()uuu,,,3xijkijkyzijk,,,,,,uu,,ijkijk,,,,2,,af,ijk,,2h,,2n,2,1121/322nnnn,n,3n(())((())uuuuu,,,,3,,,,xijkijkyijkijkzijk,,,,,,,,,,uu,,ijkijk,,,,22,,af,ijk,,2h,,2,n,1112nnnnn,,1/32213nn,1((,,,,,))(())(())uuuuuu,,,xijkijkyijkijkzijkijk,,,,,,,,,,,,uu,,ijkijk,,,,222,a,,fijk,,2h,,,, 对上式整理后,可化为下列用于实际计算的形式: nn,,1/31/3uu,r1ijkijk,,,,2222n()(),Iu,,,,,,,,xxyzijk,,22h, nn,,2/31/32uu,n,1ijkijk,,,,2n3,,(),uu,yijkijk,,,,2 h, 2 nn,,12/3uu,1ijkijk,,,,21nn,,,().uuzijkijk,,,,2,h, 2 ?3 程序设计与数值试验 ?3.2二维抛物方程实例 用五点差分格式和ADI法求解二维问题的初边值问题: ,u,24(),(,)(0,1)(0,1),0,,,,,,,uuxyGtxxyy,t (0,,)(1,,)0,01,0,uytuytyt,,,,, (,,)(,1,)0,01,0,uxotuxtxt,,,,,yy ,,(,,0)sinsin.uxyxy, 2,uxyt,,sincosexp(),,该问题精确解为 . 8 x,yt=jh(j=0,,1„„„„,J),=kh(k=0,1„„„„K),=n(n=0,1„„„„N),设jkn nu差分解为,则边值条件为 jk nnuuk,,,0,0,1………K0kJk nnnn,u,u,0,1,,,uuj………J011jjjKjK 0usincos,,,xy初值条件为. jkjk ,,取空间步长h=,时间步长,网比r=/h^2=,用五点差分格式和ADI法分别计 算到时间层t=1. 五点差分格式: nnnnn,,,,,11111ufuuuur,,,,,,*()*,jkjkjkjkjkjk,,1,1,,1,1,,,, n,1,,(14),rujk, nnuuk,,,0,0,1………K0kJk . nnnnu,u,0,1,,,uuj………J,011jjjKjK 0usincos,,,xy jkjk PR格式: 1nn,2,n,1/212nuujkjk2,,(),xyu2u,jk,jk,/2h 11/2nn,,,1/2n,121n,uujkjk2,,(),xyuu2jk,jk,/2,h nnuuk,,,0,0,1………K0kJk ,nnnnu,u,0,1,,,uuj………J011jjjKjK 0usincos,,,xy jkjk 其编程源码见附录 ?3.2三维抛物方程实例 用七点差分格式和Douglas法求解三维抛物方程问题,为此考虑如下初边值问题: 222,,,,uuuu ,,,,,,,,(0,,1,0),xyztT222,,,,txyz ,,tt33ueyzueyzyztT,,,,,,,,,sin(),sin(1),(0,1,0),xx,,01 ,,tt33uexzuexzxztT,,,,,,,,,sin(),sin(1),(0,1,0),yy,,01 ,,tt33,,tT),ueyxuexyxy,,,,,,,sin(),sin(1),(0,1,0zz,,01 uxyzxyz,,,,,sin(),(0,,1),t,0 ,3tu(x,y,z,t)=esin()xyz,,该问题精确解: . y设=ih(i=0,,1„„„„,I), =jh(j=0,1„„„„J),=kh(k=0,1„„„„K)xzjiktn,nu=n(n=0,1„„„„N),差分解为,则边值条件为: ijk nnnn,,33,,uejkhuejkh,,,,,sin(),sin(1), 1jkIjk nnnn,,33,, ueikhueikh,,,,,sin(),sin(1),ikiJk1 nnnn,,33,,ueijhueijh,,,,,sin(),sin(1), ijijK1 1uijkh,,,sin().初值条件: ijk 其七点差分格式: nnnnn,,,,1111uuuuu,,,2ijkijkijkijkijk,,,,1,,,,1,,,,,,a2,h nnn,,,111 uuu,,2ijkijkijk,1,,,,1,,,,,a2h nnn,,,111uuu,,2ijkijkijk,,1,,,,1,,,,af,ijk,,2h 边值条件为: nnnn,,33,,uejkhuejkh,,,,,sin(),sin(1), 1jkIjk nnnn,,33,, ueikhueikh,,,,,sin(),sin(1),ikiJk1 nnnn,,33,,ueijhueijh,,,,,sin(),sin(1), ijijK1 1uijkh,,,sin().初值条件: ijk Douglas格式: nn,,1/31/3uu,r1ijkijk,,,,2222nIu,,,,()(),,,,,xxyzijk,,22h, nn,,2/31/32n,uu,1ijkijk,,,,2n3,,(),uu,yijkijk,,,,2h, 2 nn,,12/3uu,1ijkijk,,,,21nn,,,(),uuzijkijk,,,,2,h, 2 边值条件为: nnnn,,33,,uejkhuejkh,,,,,sin(),sin(1), 1jkIjk nnnn,,33,,ueikhueikh,,,,,sin(),sin(1), ikiJk1 nnnn,,33,,ueijhueijh,,,,,sin(),sin(1), ijijK1 1uijkh,,,sin().初值条件: ijk 其编程源码见附录 ?4 收敛性分析与误差分析 ?4.1.收敛性分析 考虑二维热传导方程的边值问题 22,,,,uuu,,,,,,afxytoxylt()(,,),,,0,,22,,,txy,,,uxyoxy(,,)(,),, , ,uytulyt(0,,)(,,)0,,,, uxtuxlt(,0,)(,,)0,,,,,考虑ADI格式 1nn,2,n,1/212nuujkjk2 (4.1.1) ,,(),xyu2,ujk,jk,/2h nn,,11/2,n,1/2121n,uujkjk2 (4.1.2) x(),,,yu,u2jk,jk,/2h 现在对其估计截断误差的阶。将(4.1.1)和(4.1.2)相加、相减,依次得 nn,11uu,n,21jkjk221nn,2,, (), (4.1.3) ,,,uuuxjkyjkjk22/2hh, 和 1n,,nnnn,,121242()(),uuuuu,,,,,jkjkjkyjkjk2 (4.1.4) h 1,n2u消去过渡项,则 jk nnnn,,112uuuu,,11,jkjkjkjk2222()(),I,,, (4.1.5) ,,,,xyxy4242hh, nxyt,xyt,u以表示精确解在节点()的值u(),利用Taylor展式,易得 jkjkn,jkn, n,1nn,1n2,ujk11,uu,jkjkjk222222 (4.1.6) ()()(),IuOh,,,,,,,,,,xyxy4242hh, 22与(4.1.5)比较,可见截断误差的阶为。 Oh(),, 五点差分格式: nnnnnnn,,,,,,111111uuuuuuu,,,,,4jkjkjkjkjkjkjk,,1,1,,1,1,,,,, (4.1.6) ,a,2,h 2同上面分析,由taylor展式知其误差为 Oh(),, 2Ru()Ru()我们对网格上引进一个泛数:=,=. humaxuhh,ij,ij2C1,1,,,ijN……ij,,1N很明显,想要差分算子在某种泛数意义下逼近微分算子,为此我们定义以下概念: 定义1.1 设是某一充分光滑的函数类,是由截断误差Ru(),h 所定义的网格函数,若对任何,恒有 u,,RuLuxLu()(),,hhihi lim()0Ru, (4.1.7) h,0h 则说差分算子逼近微分算子,而称(4.1.7)为相容条件。 LLh 由(4.1.6)可知,差分算子逼近微分算子,且逼近的阶是: 2Ruh()(),,,, hC 定义1.2 称差分解收敛到边值问题的解u,如果当h充分小后,(4.1.1),uh .(4.1.2)的解存在,且按某一范数有 uh lim0uu,, (4.1.8) h,0h 这里把u看成是I上的网格函数。 h 引进误差 euxu,,() (4.1.9) iii exe(),则误差函数满足下列差分方程: hii LeRu,(),,hiiiN,,1,2,...,1 。 (4.1.10) ,ee,,0,0N, Ru()于是收敛性及收敛速度的估计问题,就归结到右端(截断误差)估计误差i 函数的问题。这和差分方程的稳定性有关。 eh 定义1.3 称差分方程,关于右端稳定,如LvfiN,,,(1,2,...,1)vv,,0hii0N果存在与网格及右端无关的正常数M和,使 ffxf(()),Ihhhhii0 vMf, ,当 (4.1.11) 0,,hhhh0R f.其中是右端的某一范数,它可以和相同,也可以不同,fhhR 。 vxviN(),1,2,...,1,,,hii 不等式(4.1.11)表明,解连续依赖右端,即右端变化小时解的变化也vfhh 121212小,实际上,设是差分方程对应右端的解,则满足uu,ff,vuu,,hhhhhhh 12,由,(4.1.11) Lvffvv,,,,,0hiiiN0 1212 uuMff,,,hhiiR 1212若,则。 lim0ff,,lim0uu,,iihhR 由(4.1.11)推出,与(4.1.1)(4.1.2)相应的齐方程()只f,,,0,0,,i能有平凡解,从而非齐次方程对任何边值及右端有唯一解。 uiN,,,0(1,2,...,1)i 将不等式(4.1.11)用到误差方程(4.1.10),则 eMRu,() (4.1.12) hhR .e,0h,0若解u充分光滑,关于范数满足相容条件,则当时,,Lhh 从而差分解收敛到边值问题的解,且和截断误差有相同的收敛阶。 .定理1.1 若边值问题的解u充分光滑,差分方程按照满足相容条件,且R .Ru()关于右端稳定,则差分解u按照收敛到边值问题的解,且有和相同的hhR收敛阶。 ?4.2. 误差分析: ?4.2.1二维抛物方程 用五点差分格式和ADI法求解二维问题的初边值问题: 22,,,,uuu1,,,,,(),,,0,oxylt,22,,,txy16,,uxyoxy(,,)sinsin,,,,, ,uytulyt(0,,)(,,)0,,,, uxtuxlt(,0,)(,,)0.,,,, 2,该方程精确解:. sinsinexp()xy,,,8 x,yt=jh(j=0,,1„„„„,J),=kh(k=0,1„„„„K),=n(n=0,1„„„„N),设jkn nu差分解为,则边值条件为 jk nnuuk,,,0,0,1………K0kJk nnnn,u,u,0,1,,,uuj………J011jjjKjK ouxy,sinsin,,初值条件为. . jkjk 以 PR方法为例: ,,取空间步长h=1/40,时间步长=0.1,网比r=/h^2=1/16000, ADI法计算到时间 层t=1. Matlab运行结果如下: 数值解结果如图4.1 精确解结果如图4.2 图4.1 图4.2 ,由,分别代入各个数值可得误差阶 eMh,ii 固定t0为0.01时,分别取N=40,80,160时的误差,列表如下: ,N 収敛 L 阶 20 5.8833e-004 40 1.49 1.8191e-005 80 1.7649e-006 1.84 表4.1 另固定h=1/40 ,N 収敛 L 阶 10 2.6626e-004 20 0.76 2.2116e-005 40 0.91 4.1977e-006 表4.2 由此知道时间误差阶为1,步长误差阶为2. ?4.2.2三维抛物方程 222,,,,uuuu ,,,,,,,,(0,,1,0),xyztT222,,,,txyz ,,tt33ueyzueyzyztT,,,,,,,,,sin(),sin(1),(0,1,0),xx,,01 ,,tt33sin(),sin(1),(0,1,0),uexzuexzxztT,,,,,,,,,yy,,01 ,,tt33,,tT),sin(),sin(1),(0,1,0ueyxuexyxy,,,,,,,zz,,01 uxyzxyz,,,,,sin(),(0,,1).t,0 ,3tu(x,y,z,t)=esin().xyz,,其精确解: yxz设=ih(i=0,,1„„„„,I), =jh(j=0,1„„„„J),=kh(k=0,1„„„„K)jik tn,nu=n(n=0,1„„„„N),差分解为,则边值条件为: ijk nnnn,,33,,uejkhuejkh,,,,,sin(),sin(1), 1jkIjk nnnn,,33,, ueikhueikh,,,,,sin(),sin(1),ikiJk1 nnnn,,33,,ueijhueijh,,,,,sin(),sin(1), ijijK1 1uijkh,,,sin()初值条件: ijk 以 Douglas方法为例: ,,取空间步长h=1/40,时间步长=0.1,网比r=/h^2=1/16000, ADI法计算到时间 层t=1. 其程序源码见附录 ,由,分别代入各个数值可得误差阶 eMh,ii 固定t0为0. 1时,分别取N=5,10,20时的误差,列表如下: ,N 収敛 L 阶 5 0.0498 10 1.51 0.0234 20 0.0098 1.63 表4.1 另固定h=1/10 ,N 収敛L 阶 5 0.1021 10 0.73 0.0498 20 0.0123 0.85 由此知道时间误差阶为1,步长误差阶为2. 谢辞 当代科学计算已经渗透到极其广泛的专业领域,形成了许多新的边缘学科,如计算物理学、计算力学、计算生物学等,而在这其中,计算数学则是联系他们的纽带。在四年的学习过程中,我学习了计算数学的一些基础课程。在论文的选题,以及写作过程中,程爱杰老师都给予了我很大的帮助,在此向程老师以及相关帮助过我的同学致谢,作为一名即将毕业的学生,同时在此也深深感谢学院四年来对我的栽培。 参考文献 [1]黄明游,刘播,徐涛.数值计算方法. 北京:科学出版社,2005 [2]李瑞遐,何志庆编著,微分方程数值解法. 上海,华东理工大学出版社 2005 [3]李荣华,刘播编著,微分方程数值解法. 北京,高等教育出版社 [4]胡键明,汤怀民编著,微分方程数值解法. 北京,科学出版社 [5]Richard L. Burden and J.Douglas Faires 著,冯烟利,朱海燕译,数值 分析. 北京 高等教育出版社 [6]王正林,刘明 编著 精通MATLAB(升级版). 北京,电子工业出版社 [7]郭本瑜.1998.偏微分方程的有限差分法. 北京,科学出版社 [8]张守慧,抛物型方程的几种可并行的有限差分方法, 博士论文. 山东大学计 算所 2009 [9]赵才地,高维抛物方程Cauchy问题的构造性解法,温州,师范学院学报(自然 科学版),第24卷第5期,2003年10月 [10]王正林,刘明 编著 精通MATLAB7. 北京,电子工业出版社,2006 [11]张志刚等MATLAB与数学实验. 北京:中国铁道出版社 [12]陆金甫,关冶.偏微分方程数值解法(第二版). 北京:清华大学出版社,2004 [13]马云峰,方钟波. 椭圆形方程的有效迭代法. 科技信息第13期2010: 134-135. [14]林群. 微分方程数值解基础教程-第2版. 科学出版社,2003. [15]南京大学数学系计算数学专业. 偏微分方程数值解法.计算数学讲义(四)- 第一版. 科学出版社,1979. 翻译 原文 Numerical Solution of Parabolic Equations in High Dimensions Parabolic equation is one of basic partial equations(In many science fields,many phenomena are described by parabolic equation(s),such as the process of heat conductionand diffusion,the chemical reaction etc(Among modern numerical methods,the finite difference method is the earliest and most perfect method(So the finitedifference method for solving parabolic equation is always a focal which peoples careabout(As the parallel computer comes into being and develops,some disadvantagedisappears in different means(For example,the classical explicit scheme is suit forparallel computing but it’S conditional stability(Especially for high—dimension problem,the time step is limited very severely(The classical implicit and Crank-Nicolsonscheme is absolutely stable,but they can be solved only by solving linear equations(Obviously they are not suit for parallel computing(So it’S worthy to constructing othernew difference methods which has better stability,parallelism and high-precision( In the early seventies,Miranker pointed out that organizing the traditionaldifference method in order to parallel computing is the main method when we approximatedthe partial equation by finite difference method(Between the seventies,theresearch was mainly about high—order difference scheme for different equations(But since the eighties,the situation changed because of Evans and Abdullah’S work(In the early eighties,Evans and Abdullah proposed the idea which constructedgroup explicit method by appropriate combination of different Saul’yev asymmetric scheme(The group explicit method keeps the stability of numerical computing,and has better parallelism because it can be solved explicitly(Because some terms in the truncation error of different Saul’yev scheme is equal for their absolute value and the sigh is contrast,making use of them alternating in a time layer or different layer?X?may cancel some truncation error and the calculation accuracy can be improved(And these Saul’yev scheme were implicit,but the group scheme can be solved explicitly because of appropriate combination(This is Evans-AbduUah’S Group Explicit(GE)(This work indicated that it’S possible to construct new difference methods which satisfy the above conditions(But when they extended the method to variable coefficient problem,the proving of stability is difficult(Based on this,Zhang Baolin et a1(proposed the idea which constructed the segment implicit scheme by using the Saul’yev asymmetric scheme,and set up a variety of explicit—implicit and pure implicit alternating parallel methods by making use ofalternate technology(These methods Can keep the stability and parallelism(After that,they extended it to variable coefficient problem and proved its stability by energy method(In the course of numerical experiments,they found that the result of segment or block parallel algorithm is better than the result of the method no splitting(So constructing new method by divide and conquer strategy can not only be used forparallel computing but also improve calculation accuracy-At the same time,there are many research coming into being( The numerical solution of parabolic evolution problems by Finite Elements in a domain d and by implicit time-stepping in the interval (0, T ) is used in numerous applications. ,,R There exists a sizeable and well-developed literature on the numerical analysis of discretization schemes。 For the solution of the linear system at each time step efficient solvers are available, e.g., based on suitable multilevel schemes. Most of these developments have been focussed on problems in dimension d ? 3. In some applications, however, the efficient numerical solution of parabolic problems in dimensionsd > 3 is necessary. We mention here only the pricing of contracts on baskets of d assets, e.g., for an index where d can be as large as 50, and the Kolmogoroff equations for diffusions in high dimensions. Here, the straightforward application of standard numerical schemes fails due to the so-called ‘curse of dimension’: the number of degrees of freedom on a tensor product Finite Element mesh dhof width h in dimension d grows like O() as h ? 0. This observation has led to the belief that parabolic problems in dimension d larger than 3 can in effect not be solved by conventional, deterministic methods.Therefore Monte Carlo methods are used where the error decreases like O(N?1/2) if one uses a work of N operations. This holds for any d ? 1, but only in a probabilistic sense. In this paper we describe a Finite Element algorithm for parabolic equations in high ,pNdimensions with an error of O() for a work of N operations. Here p is the degree of the finite elements which can be any integer ? 1. The method is based on two observations: (i)To reduce the number of degrees of freedom in high dimensions, so-called sparse tensor product Finite Element spaces are used. Their number of degrees of freedom grows like d,1,1hhlogO(as )for the full tensor product spaces. At the same time, the h,0 p1happroximation rate in for elements of degree p ? 1 and smooth functions is O(), the H(), p,1same as for full tensor product spaces., this result requires more regularity than for the H(), approximated function, and the amount of extra regularity increases with d. In the contract pricing problem mentioned above, the initial data u0 of the problem(the pay-off function) is usually not smooth However, the solution operatorE(t) of the parabolic problem is an analytic semigroup and increases the smoothness of the solution u(?, t)for t > 0. We prove that this parabolic smoothing effect suffices for optimal convergence of sparse space discretizations at T > 20 for any d, even for initial data that are just in. L(), , (ii) Even with a sparse space discretizations the number of spatial degrees of freedom NL is substantial if d is large. Reducing the number of time steps (and thus, the number of spatial problems to be solved) to pass from t = 0 to the final time T is therefore essential. Time analyticity of E(t) implies analytic time regularity of the solution u(t) for t > 0, but not uniformly in (0, T ). As was shown in [13], this allows to construct hp discontinuous Galerkin (DG) time-stepping schemes with exponential convergencein the number of spatial problems. We analyze the fully discrete method with sparse tensor product Finite Elements of degree p ? 1 and meshwidth h in space, and hp DG discretization in time. Because of the exponential convergence of the DG method in time it is sufficient to use O(|log h|) time intervals, and 2Lpolynomial degree O(|log h|) intime. We then obtain at the final time T for u(x, T ) an error ,,p,0hof O() where θ0 ? (0, 1] is relatedto the regularity of the elliptic problem in , and δ = , p/((p + 1)d ? 1). The case that u(x, t) is smooth in x for all t > 0 corresponds to θ0 = 1. , For each DG time step we have to solve a linear system of size (r + 1) . We can NL , decouple this and obtain r + 1 linear systems of size . Each of these r + 1 linear systems NL is of the same form as for the backward Euler method, but contains complex numbers. We solve these linear systems iteratively with GMRES and a wavelet preconditioner, and show that O(|log h|) iterations are sufficient.. 译文 高维抛物方程数值解法 抛物型方程是偏微分方程中基本方程之一(在自然科学的众多领域中,许多现象是用抛物型方程或者方程组描述的,例如热传导以及其它扩散现象、化学反应、粒子的运输等等(另外在一些问题的数值处理中也经常出现抛物型偏微分方程(在现代数值方法中,最早为人们所注意且理论分析完善的是有限差分法,因此抛物型偏微分方程的有限差分方法一直是人们关心的焦点(随着并行机的问世和发展,传统的有限差分方法在不同方面暴露出各自的弱点(例如,古典显式虽然适合于并行计算,但它是条件稳定的,特别是多维问题中计算步长受到严格的限制;古典隐式和Crank-Nicolson格式是绝对稳定的,但需要求解联立方程组,不便于直接在并行机上应用(因此需要构造具有良好稳定性、并行性和计算精度的新的差分方法( 七十年代初,Miranker 指出用有限差分逼近偏微分方程时,主要是组织传统差分方法的并行实现,至于设计新算法,推动力是很小的;之后的十几年高阶差分格式方面的研究得到了发展(八十年代初,上述情况由于Evans和Abdullah的工作而发生了变化,他们设计的分组显式方法保证了数值计算的稳定性,同时由于显式求解而使该方法具有很好的并行性质(它是不同类型Saul’yev非对称格式的恰当组合(由于不同的Saul’yev格式的截断误差中某些项绝对值相等,符号相反,在同一时间层和不同时间层上连续交替使用不同的非对称格式,可带来截断误差的部分抵消,从而提高方法的计算精度(这些非对称格式都是隐格式,但由于它们之间的巧妙结合,可以显式求解,这就是Evans-Abdullah分组显式(GE)(这项工作说明了建立满足上述要求的新的差分格式是可能的(但是在将分组显式思想应用于变系数问题时,稳定性的证明遇到了困难(在此基础上,张宝琳等在中提出利用Saul’yev非对称格式构造分段隐式的思想,并恰当的使用交替技术建立了多种显一隐式和纯隐式交替并行方法,取得了稳定性和并行兼顾的研究成果(之后又将方法推广到变系数问题,并用能量法证明了方法的绝对稳定性(在数值试验中发现,分段或分块并行计算的结果一般都比原来相应的未加分 裂时的结果精确(所以通过分而治之的策略来建立新算法,不但可以用于并行,还可以提高精度(之后涌现出大量的并行差分算法的研究成果,上述的方法在并行性和稳定性方面都有其优良的表现,但是他们都存在一个共同的问题,那就是它们都是基于二阶差分格式建立的,这直接影响数值计算中空间的误差精度(近年来,研究人员开始致力于研究高阶紧致差分格式( d在区域上抛物型进化方程的问题有限元数值解法和在区间(0,T)的隐,,R 式时间步法有在许多实际应用。已经存在相当大的和成熟的数值分析的离散化 方案 气瓶 现场处置方案 .pdf气瓶 现场处置方案 .doc见习基地管理方案.doc关于群访事件的化解方案建筑工地扬尘治理专项方案下载 ,对每个时间步长的线性系统是可利用的,例如,基于适当的多级方案。大部分这些发展中存在的问题都集中在维数:d?3。 然而,在一些抛物问题应用程序的有效的数值解dimensionsd> 3是必要的。我们在这里提及的资产仅为d的定价合同,如包括篮子的地方,由于指数d大到50,在高维度对扩散可以多相湍流动力学方程。 在这里, 标准 excel标准偏差excel标准偏差函数exl标准差函数国标检验抽样标准表免费下载红头文件格式标准下载 数值方案直接应用失败原因是所谓的“曲线维数” ,当h ? dh0,维上自由度的数,在张量积有限元网格的宽度尺寸h在维数d和O()一样。这个观察结果使我们认为在抛物方程维数:d>3时不能解决传统的、不确定性的问题。因此蒙特卡罗方法是用在误差在O(N?1/2)以下,操作数是N。这一规律同样适用于任何d?1,但只在概率意义。 ,pN在本文中,我们描述一个误差在O()之下,操作数为N的高维抛物型方程的有限元算法。这里p为它可以被任何整数?1的有限元单元的程度。该方法是基于两个方面: (i)在高维方程降低自由度数量,即在有限元空间使用所谓的稀疏的张量积 d,1,1hhlogh,0在充分张量积空间中,当,他们的自由度数目像O()与此同 p1h时,在中,其近似程度在p?1和平滑函数O(),同为全部张量积空H(), 间,这个结果比需要更多的规律性,且数量近似函数增加额外的规律与d。 在合同定价问题提到的,初始数据的问题(半的盈利功能)通常是不顺利。 然而,解决经营者(t)抛物线形的问题是一种分析半群和增加的光滑解 (,. t)为t > 0。我们证明该抛物线平滑效果的一致最优强收敛的稀疏足以为空 2间离散在T > 0为任何d,甚至对初始数据,就在。. L(), (ii)即使有一个稀疏空间离散数量的空间自由度的实质性如果d的损失是巨大的。数量的减少时间步(从而,一定数量的空间还需解决的问题)通过从t = 0到最后时刻t是必要的。美国萨迪教授E的时间(t)意味着时间规律的解析解的u(t)为t > 0,但不均匀(0,t)。也是一种表现在[13],这使得构建惠普(DG)对不连续伽辽金方案与指数convergencein空间问题的数量。 , 对于每个DG的时间步长,我们得处理大小为(r + 1) 的线性系统。NL , 我们可以解耦这并取得的大小为的线性系统r+1个。每一个这些r + 1NL 个线性系统用相同的形式:后退欧拉法且包括复杂的数字。我们为解决这些线性系统,设计了一种小波,不断地用GMRES预处理,用O(|log h|)的迭代量是足够的. 附录: 五点差分格式源码: function wudian() xa=0; xb=1; ya=0; yb=1; ta=0; tb=1; N=40; N0=N+1; M=N-1; L=100; t0=0.01; t=linspace(ta,tb,101); h=(xb-xa)/N; A=zeros(M,M); x=linspace(xa,xb,N0); y=linspace(ya,yb,N0); F=zeros(M,1); E=zeros(M,1); H=zeros(M,1); K=zeros(1,N0); G=zeros(M,M); C=[H G H]; B=[K;C;K]; for i=0:N; for j=0:N; B(i+1,j+1)=sin(pi*i*h)*cos(pi*j*h); end end l=1; r=(1/16)*t0/h^2 while(l<=100) for i=2:N for j=2:N B(i,j)=r*(B(i,j-1)+B(i,j+1)+B(i-1,j)+B(i+1,j))+(1-4*r)*B(i,j) end end l=l+1 end C=[H B H]; B=[K;C;K]; surf(x,y,B); hold on; ADI法编程源码 PR程序: function PR() xa=0; xb=1; ya=0; yb=1; ta=0; tb=1; N=80; N0=N+1; M=N-1; t0=0.01; t=linspace(ta,tb,101); h=(xb-xa)/N; A=zeros(M,M); x=linspace(xa,xb,N0); y=linspace(ya,yb,N0); for i=1:M A(i,i)=-(1/16)*(t0/h^2+1); end for i=2:M A(i,i-1)=(1/16)*t0/(2*h^2); end for j=2:M A(j-1,j)=(1/16)*t0/(2*h^2); end F=zeros(M,1); E=zeros(M,1); H=zeros(M,1); K=zeros(1,N0); G=zeros(M,M); C=[H G H]; B=[K;C;K]; for i=0:N; for j=0:N; B(i+1,j+1)=sin(pi*i*h)*cos(pi*j*h); end end for l=1:10 for k=2:N for j=1:M F(j)=(1/16)*(-t0/(2*h^2))*B(j,k+1)+(1/16)*(t0/(h^2)-1)*B(j,k)-(1/16)*(t0/(2*h^2))*B(j,k-1); X=A\F; for j=1:M G(j,k-1)=X(j); end end end C=[H G H]; B=[K;C;K]; for j=2:N; for k=1:M E(k)=(1/16)*(-t0/(2*h^2))*B(j+1,k)+(1/16)*(t0/(h^2)-1)*B(j,k)-(1/16)*(t0/(2*h^2))*B(j-1,k); Y=A\E; for k=1:M G(j-1,k)=Y(k); end end end l=l+1; end C=[H G H]; B=[K;C;K]; surf(x,y,B); z=zeros(N0,N0); for i=1:N0 for j=1:N0 z(i,j)=sin(pi*x(i)).*cos(pi*y(i))*exp(-(pi^2)*tb/8); end end surf(x,y,z); w=z-B; an=0; for i= 1:N0 for j=1:N0 v=abs(w(i,j)) an=w(i,j)^2+an; end end max(max(v)) an Douglas程序源码 function Douglas() xa=0; xb=1; ya=0; yb=1; za=0; zb=1; ta=0; tb=1; N=40; N0=N+1; M=N-1; t0=0.01; t=linspace(ta,tb,101); h=(xb-xa)/N; r=(t0/2)*h^2; A=zeros(M,M); x=linspace(xa,xb,N0); y=linspace(ya,yb,N0); z=linspace(za,yb,N0); for i=1:M A(i,i)=-(t0/h^2+1); end for i=2:M A(i,i-1)=t0/(2*h^2); end for j=2:M A(j-1,j)=t0/(2*h^2); end F=zeros(M,1); E=zeros(M,1); D=zeros(M,1); H_0=zeros(N0,1); H_1=zeros(N0,1); K_0=zeros(N0,1); K_1=zeros(N0,1); C_0=zeros(N0,1); C_1=zeros(N0,1); G=zeros(M,M,M); B=[] for i=0:N; for j=0:N; for k=0:N B(i+1,j+1,k+1)=sin((i+j+k)*h); end end end for l=1:100 for k=2:N for j=2:N for i=2:N F(i-1)=(1-10*r)*B(i,j,k)+r*(B(i+1,j,k)+B(i-1,j,k))+2*r*(B(i,j,k-1)+B(i,j,k+1)+B(i,j-1,k)+B(i,j+1,k )+B(i+1,j,k)+B(i-1,j,k)); X=A\F; for i=1:M G(i,j-1,k-1)=X(i); end end end end for i=1:N0 for j=1:N0 H_0=exp(3*t0*(-l))*sin((i+j)*h); H_1= exp(3*t0*(-l))*sin((i+j)*h+1); K_0= exp(3*t0*(-l))*sin((i+j)*h); K_1 =exp(3*t0*(-l))*sin((i+j)*h+1); C_0 =exp(3*t0*(-l))*sin((i+j)*h); C_1 =exp(3*t0*(-l))*sin((i+j)*h+1); end end B_1=[] B_2=[] B(:,:,1)=H_0; B_1(:,:,1) =H_0; B_2(:,:,1)=H_0; B(:,:,N0)= H_1; B_1(:,:,N0)= H_1; B_2(:,:,N0)=H_1; B(1,:,:) =K_0; B_1(1,:,:) =K_0; B_2(1,:,:)=K_0; B(N0,:,:)=K_1; B_1(N0,:,:)=K_1; B_2(N0,:,:)=K_1; B(:,1,:)= C_0; B_1(:,1,:)= C_0; B_2(:,1,:)=C_0; B(:,N0,:)=C_1; B_1(:,N0,:)=C_1; B_2(:,N0,:)=C_1; B_1(2:N, 2:N, 2:N)=G; for i=2:N; for k=2:N for j=2:N E(j-1)=B_1(i-1,j,k)*r-2*r*B_1(i,j,k)+r*B_1(i+1,j,k)+r*(B(i,j-1,k)+B(i,j+1,k)+B(i-1,j,k)+B(i+1,j,k ))+2*r*(B(i,j,k-1)+B(i,j,k+1))-8*r*B(i,j,k) Y=A\E; for j=1:M G(i-1,j,k-1)=Y(j); end end end end B_2(2:N,2:N,2:N)=G for i=2:N for j=2:N for k=2:N D(k-1)=r*(B_2(i,j-1,k)+2*B_2(i,j,k)+B_2(i,j+1,k))+r*(B_1(i-1,j,k)-2*B_1(i,j,k)+ B_1(i+1,j,k))+r*(B(i-1,j,k)+ B(i+1,j,k)+B(i,j-1,k)+B(i,j+1,k)+ B(i,j,k-1)+B(i,j,k+1)-6*B(i,j,k)) Z=A\D; end for k=1:M G(i-1,j-1,k)=Z(k); end B(2:N,2:N,2:N)=G; end end end z=zeros(N0,N0,N0); for j=1:N0 for k=1:N0 z(i,j,k)=exp(-3)*sin((i+j+k)*h); end end end
本文档为【高维抛物方程】,请使用软件OFFICE或WPS软件打开。作品中的文字与图均可以修改和编辑, 图片更改请在作品中右键图片并更换,文字修改请直接点击文字进行修改,也可以新增和删除文档中的内容。
该文档来自用户分享,如有侵权行为请发邮件ishare@vip.sina.com联系网站客服,我们会及时删除。
[版权声明] 本站所有资料为用户分享产生,若发现您的权利被侵害,请联系客服邮件isharekefu@iask.cn,我们尽快处理。
本作品所展示的图片、画像、字体、音乐的版权可能需版权方额外授权,请谨慎使用。
网站提供的党政主题相关内容(国旗、国徽、党徽..)目的在于配合国家政策宣传,仅限个人学习分享使用,禁止用于任何广告和商用目的。
下载需要: 免费 已有0 人下载
最新资料
资料动态
专题动态
is_882336
暂无简介~
格式:doc
大小:127KB
软件:Word
页数:37
分类:企业经营
上传时间:2017-12-29
浏览量:59