Unformatted text preview: Taylor series expansion of f=cos(x) about x=0. f '(0) =  sin(0) = 0, f (3) (0) = sin(0) = 0, f "(0) =  cos(0) = 1 f (4) (0) = cos(0) = 1, etc. So Taylor series expansion is (as given in Problem 4.10) cos( x) = 1  x 2 x 4 x 6 x8 +  + + 2! 4! 6! 8! An mfile that calculates this approximation with n terms is function apx=costaylor(x,n) %Calculates the Maclaurin series approximaton to cos(x) using the first n %terms in the expansion. apx=0; for i=0:n1 apx=apx+(1)^i*x^(2*i)/factorial(2*i); end Problem 4.10 asks us to increment n from 1 until the approximate error indicates that we have accuracy to two significant digits for x=pi/3. We start with >> a1=costaylor(pi/3,1) a1 = 1 >> a2=costaylor(pi/3,2) a2 = 0.4517 With cos(pi/3)=0.5, the true error in a1 is 100% and in a2 it is >> true2=(a20.5)/0.5*100 true2 = 9.6623 That is about 9.7%. The approximate error, though is >> aprerror=(a1a2)/a2*100 aprerror = 121.3914% With one more term we get >> a3=costaylor(pi/3,3) a3 = 0.5018 >> true3=(a30.5)/0.5*100 true3 = 0.3592 >> aprerror=(a2a3)/a3*100 aprerror = 9.9856 So the actual error is only 0.36% while the estimated is 10%. Finally with a fourth term >> a4=costaylor(pi/3,4) a4 = 0.5000 >> true4=(a40.5)/0.5*100 true4 = 0.0071 >> aprerror=(a3a4)/a4*100 aprerror = 0.3664 The true error is less than one percent of one percent, the approximate error is 0.37% so we have at least two significant digits and we stop. Total error: The difficulty of finding a good step size for differentiation is particularly acute when solving illconditioned equations. For example, Consider the system: 10001u + xv = 1000 , xu + 10000v = 1000 . For x=10,000 the determinant is almost zero. The system is ill conditioned, and it is hard to find a good step size for calculating derivatives with respect to x. One of my graduate students ran into difficulties calculating derivatives of the deformation of a car model seen below With respect to variables that define the dimensions of the car. ...
View
Full Document
 Spring '09
 RAPHAELHAFTKA
 Derivative, Taylor Series, Colin Maclaurin, approximate error

Click to edit the document details