1、一、高斯点一、高斯点定义:高斯公式定义:高斯公式机械求积公式机械求积公式 nkkkbaxfAdxxf0)()(含有含有2n+2个待定参数个待定参数 ),.,1,0(,nkAxkk 若适当选择这些参数使求积公式具有尽量高次若适当选择这些参数使求积公式具有尽量高次(2n+1次次?!)代数精度代数精度,则这类公式称为,则这类公式称为高斯公式高斯公式。(4.1)定义:定义:高斯公式的求积节点称为高斯公式的求积节点称为高斯点。高斯点。请回顾请回顾:以前学过的梯形公式、辛甫生公式、柯以前学过的梯形公式、辛甫生公式、柯特斯公式、中矩形公式是高斯公式吗?特斯公式、中矩形公式是高斯公式吗?除中矩形公式外都不是!
2、除中矩形公式外都不是!注:注:机械型高斯求积公式一定是插值求积公式。机械型高斯求积公式一定是插值求积公式。举例举例求求 a,b上的两点高斯公式。上的两点高斯公式。解解 设两点高斯公式为设两点高斯公式为)()()(1100 xfAxfAdxxfba )(41)(31)(2144331130033221120022110010abdxxxAxAabdxxxAxAabxdxxAxAabdxAAbabababa这是关于四个未知这是关于四个未知数的非线性方程组,数的非线性方程组,是否有解?一般难是否有解?一般难于求解于求解要求其代数精度最高,四个未知数,可列出要求其代数精度最高,四个未知数,可列出4个方
3、程个方程:高斯点具有以下性质:高斯点具有以下性质:定理定理插值型求积公式插值型求积公式(4.1)成为成为Gauss求积公式求积公式的的充要条件充要条件:),.,1,0(nkxk 求积节点求积节点 为为n+1次正交多项式的零点次正交多项式的零点。如何求高斯公式如何求高斯公式?nkkkbaxfAdxxf0)()(正交多项式概述:正交多项式概述:1()()0bnnaP xx dx首先证明首先证明对于任给节点对于任给节点 x0,x1,xn,均存在某,均存在某个次数为个次数为2n+2的多项式的多项式f(x),机械型求积公式机械型求积公式不能不能精精确成立,即其最高代数精度不能达到确成立,即其最高代数精度
4、不能达到2n+2。如取:。如取:21()()nf xx证明证明2121000()()()0()bbnaannknkkkkf x dxxdxAxA f x则有:则有:),.,1,0(nkxk 设求积节点设求积节点 为为n+1次正交多项式次正交多项式n+1+1(x)的零点的零点。101()()()()即有:nnxxxxxxx现证充分性。即现证充分性。即n(x)P(x)与正交求积公式是高斯型。求积公式是高斯型。证明证明现对于任意给定的次数不超过现对于任意给定的次数不超过2n+1的多项式的多项式f(x),用用 除除 f(x),记商为,记商为P(x),余式为,余式为Q(x),)(x 即即)()()()(
5、xQxxPxf 2n+1n+1 nn由已知条件,由已知条件,(x)与与P(x)正交,故得正交,故得 babadxxQdxxf)()(由于所给求积公式由于所给求积公式(4.1)是插值型的,它至少具是插值型的,它至少具有有n次代数精度,故对次代数精度,故对Q(x)能准确成立:能准确成立:nkkkbaxQAdxxQ0)()(再注意到再注意到(xk)=0,知,知Q(xk)=f(xk),从而有,从而有 nkkkbaxfAdxxQ0)()(综之得:综之得:nkkkbaxfAdxxf0)()(这说明公式对这说明公式对一切次数不超一切次数不超过过2n+1的多项的多项式准确成立,式准确成立,综之说明综之说明xk
6、是是高斯点。高斯点。再证必要性,即再证必要性,即若是高斯求积公式若是高斯求积公式(x)P(x)与正交设设P(x)是任意次数不超过是任意次数不超过 n 的多项式,则的多项式,则P(x)(x)的次数不超过的次数不超过2n+1,因此应准确,因此应准确成立成立 nkkkkbaxxPAdxxxP0)()()()(但但),.,1,0(0)(nkxk 故故 .正交正交与与)()(xPx 求积节点构造的求积节点构造的注:注:1、总可通过、总可通过施密特正交化施密特正交化求出求出a,b上与所有次数上与所有次数不超过不超过n的多项式都正交的多项式的多项式都正交的多项式n+1(x)。2、命题:、命题:n次正交多项式
7、次正交多项式 有有n个单零点。个单零点。解:解:设设P0(x)=C,1(x)=x x0。由于。由于0)()(1110 dxxxP 即即0)(110 dxxxC展开,得展开,得00 x则一个点的高斯公式为则一个点的高斯公式为1120f(x)dxf()中矩形公式中矩形公式例例.求求-1,1上与次数为上与次数为0的多项式正交的多的多项式正交的多项式项式1(x)=?二、高斯二、高斯勒让得公式勒让得公式若若a,b=-1,1,其上的高斯公式为,其上的高斯公式为 110)()(nkkkxfAdxxf称为称为高斯高斯-勒让得公式勒让得公式。-1,1上的正交多项式称为上的正交多项式称为勒让得多项式勒让得多项式,
8、勒让得多项式勒让得多项式Pn+1(x)的零点就是高斯点。的零点就是高斯点。几个几个Legandre 多项式:多项式:若取若取P1(x)=x 的零点的零点x0=0 作求积节点构造公式作求积节点构造公式:110)0()(fAdxxf令它对令它对 f(x)=1准确成立,即可定出准确成立,即可定出A0=2.从而得到一点高斯公式:从而得到一点高斯公式:11)0(2)(fdxxf中矩形公式中矩形公式 1110)31()31()(fAfAdxxf令它对令它对 f(x)=1,x 准确成立,即可定出准确成立,即可定出A0,A1可得两点高斯可得两点高斯勒让得公式为勒让得公式为若取若取 的零点的零点 作求积节点构作
9、求积节点构造公式造公式)13(21)(22 xxP31 11)31()31()(ffdxxf注:更高阶的公式见书注:更高阶的公式见书p122。请思考请思考:高斯高斯勒让得公式的求积区间是勒让得公式的求积区间是-1,1,那么对,那么对于任意求积区间于任意求积区间a,b如何办?如何办?解解作变换作变换22batabx 可以化到区间可以化到区间-1,1上,这时上,这时dtbatabfabdxxfba)22(2)(11 三、带权的高斯公式三、带权的高斯公式(更一般的表现形式)(更一般的表现形式)有时需要求如下带权的积分:有时需要求如下带权的积分:ba(x)f(x)dx称上述称上述(x)0是权函数。是权
10、函数。定义:定义:若求积公式若求积公式 nkkkbaxfAdxxfx0)()()(具有具有2n+1次代数精度,则称这类公式为次代数精度,则称这类公式为带权带权的高斯公式的高斯公式.高斯点高斯点我们类似的可有:我们类似的可有:定理定理),.,1,0(nkxk 是高斯点的是高斯点的充要条件:充要条件:).()()(10nxxxxxxx 是区间是区间a,b上上带权带权(x)正交的多项式。正交的多项式。若若a,b=-1,1,权函数为,权函数为211)(xx 所建立的高斯公式所建立的高斯公式 nkkkxfAxxf0112)(1)(切比雪夫切比雪夫高斯公式高斯公式称为称为切比雪夫切比雪夫高斯公式。高斯公式
11、。xk是切比雪夫多项式的零点。是切比雪夫多项式的零点。4.7.4 Gauss-Chebyshelv quadrature formula.)()(wherein.)()()()(of zeros by the dconstructe is formula quadratic-Gauss),(by denoted,polynomiallar perpendicu called is ,11)(function t with weigh-1,1on defined,polynomiallar perpendicu The111112dxxLxAxfAdxxfxxTChebyshevxTChebys
12、hevxxiiniiinn:(monic).1 is term leading oft coefficien thearccoscos21)(:as problem phys.&math.solving in early discoveredare spolynomial of expressionExplicit 1xnxTChebyshelvnn Remark 1 three term recurrence formula v.s.Schmidt orthogonolization;Remark 2 Tn are perpendicular polynomials;),1().(polyn
13、omial theof zeros are 212cos:see easy to isIt formula.in ,of onsconstructi thesee llNext wenkxTnkxChebyshevGaussAxnkkkndxnxTxdxxTxxxTxdxxLxdxxLxAkknknknkkk01111112coscos)(1cos)()()()()()()()(nknkfndxxfx1112212cos)(11At last,well state the error estimation of the Gauss-Chebyshelv formula without the
14、proof:According to the error estimation of the Gauss-Type formula,we have:)()!2(2)()!2()();()2(12112)2(nnnnnfndxxnfGfR :gives formula then performed,is 44ation transforma first,At.sin :compute toformula point -2 UsingExample.20LegendreGausstxSolutionxdxILegendreGauss99848.04157735.0sin14157735.0sin1
15、441sin411 dttIConsult the table in p122.97732.33 :hence,23,0,23 :,212cos and 3for 11:formula toAccording :Solutionerror.estimate and 1 compute toformula point-three Using2 Example230232,3,11112112eeeIxi.e.nkxnendxexIChebyshevGaussdxxeIChebyshevGaussknkxxxk构造高斯公式的一般方法:构造高斯公式的一般方法:1、构造正交多项式,继而求其零点,再、构
16、造正交多项式,继而求其零点,再按插值求积公式获得高斯公式;按插值求积公式获得高斯公式;2、待定系数法、待定系数法此外,还可涉及到无穷区间上的广义积分等。此外,还可涉及到无穷区间上的广义积分等。例如:例如:0 xef(x)dx-拉盖尔拉盖尔-高斯积分高斯积分举例举例要构造下列形式的高斯公式要构造下列形式的高斯公式)()()(110010 xfAxfAxfx 解解则其代数精度应为则其代数精度应为311212 n即即01001 1222001 1333001 12 32 52 72 9babababaAAxdx/A xA xx xdx/A xA xxxdx/A xA xxxdx/求解求解?!定理(稳
17、定性)定理(稳定性)高斯求积公式的求积系数Ak0.证明:事实上2200()().nbki kikailxdxAlxA这表明高斯求积法是稳定的。关于关于积分余项积分余项和和收敛性收敛性有:有:积分余项:积分余项:2221()22!nbnnafRfxx dxn收敛性:收敛性:设设f(x)Ca,b,则有:则有:0lim()()nbkkankA f xf xx dx4.1 Numerical Differentiationposed.-ill error;roundnemesis,oldthenote,)()()()(lim)(:)(function a of derivative theof pla
18、ce at thestart willWe000000hxfhxfhxfhxfxfxfenoughsmallhhHowever,(i)There is no error estimation;(ii)Are there any other numerical methods for ND?How to construct them&what about error?To answer these questions,we observe first:01011100010(i)The elementary formula above can also be obtained by consid
19、ering interpolation with,:()()()()(),2!()()()()()2()()(),thus2!()xxxa bxxxxf xP xff xf x2 xxhfxfhxxxxD ffx100100()()(),2()()so:().2f xf xhfhf xf xhfxMhError Bound011(1)0(ii)To obtain approximation formulas using multi-points,suppose,are(1)distinct points in some and(),Lagrangeinterpolating formula t
20、ells us:()()()()(1)!nnnnkkkx xxnIfCIff xf xL xn1(1)10(1)0.Differentiating this expression gives:()()()(),so:(1)!()()()()().(1)!called(1)point formula.nnnnkkxkjknkjnjkkjjkfxf xLxDfnxxfxf xLxfxnn120201010210120122The most common used formulas are those involving 3 and 5evaluation points.Such as for 3-
21、point formula:22()()()()()()()2()(jjjjxxxxxxfxf xf xxxxxxxxxxxxf xx2(3)00211()().)()6jjkkkjfxxxxx).(2)()()(:is formulasimplest Then the000fhhxfhxfxf02(3)00002(3)000When the nodes are equally spaced,:,3-point formula become especially useful:1()3()4()(2)()23or :1()()()().26ji.e.xxjhhfxf xf xhf xhfhhf
22、xf xhf xhfhCalled forward difference¢ral difference formula.There are also backward difference formulas.00004(5)0000004(5)01()25()48()36(2)1216(+3)3(4)().51()(2)8()8()12(2)()30.3-point formula is more sensitive(unstable)than 5-point fxf xf xhf xhhhf xhf xhffxf xhf xhf xhhhf xhfNote1one with resp
23、ect to the exaggrated factor of by view of denumerators.The formula involved points in the nearerneigbour of the node will have smaller truncated error.hNote 2.Five-point formula below can be obtained similarly:It then be called compact form.).(12)()(2)(1)(:gives)(and)(cancels toequations twoadding,
24、)(241)(61)()()()()4(2000200041)4(3020000fhhxfxfhxfhxfxfxfhfhxfhxfhxfxfhxf For higher order derivatives,it can also be obtained by interpolation like to the 1st order derivative using more points.Alternately,we can obtain the formulas which are algebraically tedious by Taylors expansion such as:Cf.th
25、e results obtained by the two methods.Balance between round-off&truncated error 000002(3)In evaluating ,we usally encounter roundoff bounded by,hence the total error for a 3-point formula in the approximation is:()()()()26hhhhhf(xh)e(xh)ef xhef xhefx2heehfhh26(wherein the involved boundness)the opti
26、mal choice for can be obtained by calculus;it shows the algorithem has the old nemesis!The finite element method()may help us.hM,MhFEM4.2 Richardsons Extrapolation(1927)Richardsons Extrapolation is used to generate high-accuracy results while using low-accuracy formulas.23123Suppose that for each0,w
27、e have a formula()to approximate an unknownand that itstruncation error involved in the appproximation hasthe form:()().ifhN hMMN hK hK hK hO h23123Since the formula is assumed to hold for all positive,consider the result when we replace the parameterbyor 2,then we have:2()()2222eliminating()results
28、 in2()2alsohhhhhhhhMNKKKO hO hhMNN h223223:()3()24MNhh-hKKO h332223238)2(:havellweformula,abovethein2/byreplaceweSimilarly,hKhKhNMhhThen combined with the formula of N2(h)to eliminate the h2 term,we obtain:3332228)(3)()2/()2(hKhNMhNhNhNM:Which posses higher order truncated error!.12)()2/()2()(formth
29、eofionapproximat)(anhavewe,3,2eachforthen)()(formtheinwrittenbecanif,generalIn111111jjjjjjmmjjjhNhNhNhNhOmjhOhKhNMMThe geometry explanation(For h0,the approximation should be accuracy):.12)()2(2)0(:have we thus)2(2)(22)(:get weformula,given thefrom)2(,2and)(,dataely approximat heion with tExtrapolat
30、211111111111111NhNhNNhNhKhKhKzhNhKhKhKzzNhNhKhNhKionextrapolatby Related topic:steffensens acceleration for convergent linearly iterative sequence.Numerical Differentiation Revisit-Using Extrapolation Method)(120)(6)()(120)(62)()()(:formula difference-divided centralobtain canwe,)(1201)(241)(61)(21)
31、()()(expansionTaylorthebyFirstly,)5(402)5(40200052,1)5(40)4(3020000fhxfhfhxfhhhxfhxfxfhfhxfhxfhxfhxfxfhxfhN:00004(5)0 Next,using the formula with 2 replacing,and Richardsons extrapolation will result in 5pointformula with the error of order 4:1()(2)8()8()12(2)()30hhfxf xhf xhf xhhhf xhfThe technique
32、 of Richardsons extrapolation is also used in approximating definite integrals and in determining approximate solution to differential equations in later Chapters.Summery:We have studied 3 ways to obtain numerical derivatives algorithms:(1)Lagrange Interpolating formula;(2)Taylor mean value Theorem;(3)Richardson extrapolation process.Homework(p184):5;15*作业:作业:P136P136习题习题 1111谢谢观看!谢谢观看!2020