Math 128A - Fall 2001 - Strain - Final

Math 128A - Fall 2001 - Strain - Final - 04/22/2002 MON...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: 04/22/2002 MON 15:38 FAX 6434330 MOFFITT LIBRARY Math 128a, Section 3 — Final Exam — December 17, 2001 Pro-p Siren}? Problem 1 Let 332/4 + 312/9 — 1 f(w,y)= x_y_1 (a) Define quadratic convergence of a sequence of vectors :17“ to a limit 33. (b) Compute the Jacobian matrix D f (3:, y). (c) Determine where D f (m,y) is invertible and compute D f (3:, y)‘1 when it exists. ((1) Write down Newton’s . method for solving f (m,y) = 0. (e) Start with 550 : (2,0)T and compute the first two approximations x1 and 3:2 generated by Newton’s method. (f) Explain why your results demonstrate quadratic convergence. Problem 2 Consider the iteration mg + 3am“ $ 1:: . 71+ 3:13?l + a (a) What is it intended to compute? (b) Given a 2 and 3:0 1, compute 3:1 and 2:2. (c) Define and determine the order of convergence of this iteration. Problem 3 Find the QR factorization of A— 3g —QQoR—3PLP 1 PR — ""‘ 123 _5l\/§2‘/m3' You don’t need to multiply together the matrices Pi. Problem 4 (a) Derive a numerical integration formula [01mm = were) + w1f(1)+ w2f(2) which is exact for polynomials of as high degree d as possible, and determine the maximal degree d. (b) Without any additional work, determine an equally accurate rule of the form fol f(x)da: = u0f(-1)+ u1f(0) + u2f(1). (c) Show that f sedan = hearse) + wrnh) + WNW!» + 0W1.) as h e 0. (d) Use (a), (b), and (c) to build a quadrature formula with error 0(hd) on an arbitrary interval [03, (7] divided into n >_ 1 subintervals of length h = (b — a) 001 04/22/2002 MON__15:38 FAX 6434330 MOFFITT LIBRARY 002 Problem 5 (a) Write down the Newton and Lagrange forms of the quadratic interpolant p(:r) to a function f at three points a, b and c. (b) Give a formula for the error p(:c) —— f (3:) if f is a nice function with all derivatives bounded. Explain why your error formula makes sense in terms of dimensions, zeroes and the derivatives which appear versus the degree of polynomial used. (c) Specialize to f (as) = R * 1/w and evaluate the coeflicients in the Newton representation of p(3:). (d) Use (c) to express p in the power form p($) : q" + qla: + (123:2. (a) How would you use the formula of (d) to derive an iterative method for finding I/R? Problem 6 Suppose A is a square invertible matrix. (a) Define the condi— tion number (b) Suppose E is a matrix the same size as A and HE” 1 a A ——-— < e < —. ( )IIAII - - 2 Show that A + E is invertible. (c) Show that “(A + E)‘1 - All“ < 26. "fl—1|! _ Extra Credit Problem 7 Given an approximate solution 3; to the linear system AI = b with a square invertible matrix A, let 7* 2 b — Ay be the residual of y. (a) Show that if y satisfies a perturbed linear system (A + E)y = b then the perturbation E must satisfy m > uru IIAH _ IIAIIIIyll' (b) Show that there is a matrix E such that (A + E)y = b with the norm of E satisfying fl§fl_ IMI llAll _ llAllllyll' (Hint: Try a rank-one matrix E = admin for some well-chosen scalar as.) (c) Define the backward relative error in an approximate solution 3; of Am = b. (d) Show that an'approximate solution y of Ax = b has backward relative - error 0(6) if and only if it has a residual r satisfying urn _ E “Anny” ‘ 0‘ )‘ Extra Credit Problem 8 Prove that any model of floating—point arithmetic which requires that the floating—point result of the multiplication my be given by the exact result correctly rounded satisfies the relative error bound lwy—fltvwll I93 * yl as long as no overflow or underflow occurs and 3: =1: 3; aé 0. $6 ...
View Full Document

This note was uploaded on 05/17/2009 for the course MATH 128A taught by Professor Rieffel during the Spring '08 term at University of California, Berkeley.

Ask a homework question - tutors are online