This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Solutions Homework 3 Numerical Linear Algebra 1 Fall 2010 Problem 3.1 Problem 3.1.6 Golub and Van Loan p. 93 Solution: This is a standard incremental algorithm based on a partitioning of the matrices involved. The algorithm is O ( n 2 ) over all when we start with k = n 1, i.e., S c and T c are initialized as 1 × 1 matrices and getting the initial x c and w c is therefore O (1). If we can show that producing x + and w + requires O ( n k ) computations then we have total work ∑ n 1 k =1 O ( n k ) = O ( n 2 ) as desired. The identities are easily verified through simple algebra by multiplying out the given partitioned linear system: σ u T S c τ v T T c γ x c λγ λx c = β b c σ u T S c τγ + v T x c T c x c λγ λx c = β b c σ u T S c τγ + v T x c w c λγ λx c = β b c From this consider only the first equation to deduce γ = β σv T x c u T w c στ λ as desired. An algorithm follows easily. Given w c and x c they can each be extended by one component to x + and w + • η = v T x c • μ = u T w c • γ = ( β σv T x c u T w c ) / ( στ λ ) • ω = τγ + η and we have w + = ω w c and x + = γ x c Note that the computation of η and μ require 2( n k ) + O (1) each and everything else in the basic step is O (1). We therefore have that x + and w + can be produced in O ( n k ) as desired. 1 Problem 3.2 Recall that any unit lower triangular matrix L ∈ < n × n can be written in factored form as L = M 1 M 2 ··· M n 1 (1) where M i = I + l i e T i is an elementary unit lower triangular matrix (column form). Given the ordering of the elementary matrices, this factorization did not require any computation. Consider a simpler elementary unit lower triangular matrix (element form) that differs from the identity in one offdiagonal element in the strict lower triangular part, i.e., E ij = I + λ ij e i e T j where i 6 = j . 3.2.a . Show that computing the product of two element form elementary matrices is simply superposition of the elements into the product given by E ij E rs = I + λ ij e i e T j + λ rs e r e T s whenever j 6 = r . 3.2.b . Show that if j 6 = r and i 6 = s then computing E ij E rs with requires no compu tation and E ij E rs = E rs E ij i.e., the matrices commute. 3.2.c . Write a column form elementary matrix M i in terms of element form elementary matrices. Does the order of the E ji matter in this product? 3.2.d . Show how it follows that the factorization of (1) is easily expressed in terms of element form elementary matrices. 3.2.e . Show that the expression from part (3.2.d) can be rearranged to form L = R 2 . . . R n where R i = I + e i r T i is an elementary unit lower triangular matrix in row form....
View
Full
Document
This note was uploaded on 07/21/2011 for the course MAD 5932 taught by Professor Gallivan during the Fall '06 term at FSU.
 Fall '06
 gallivan

Click to edit the document details