1j n hence absolute errors in the one and two norms

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: to the largest absolute ˜ error, x − x ∞ = max |xj − xj |. ˜ ˜ 1 ≤j ≤n For the one and two norms we have max |xj − xj | ≤ x − x ˜ ˜ 1 1≤j ≤n ≤ n max |xj − xj | ˜ 1 ≤j ≤n and max |xj − xj | ≤ x − x ˜ ˜ 1≤j ≤n 2 ≤ √ n max |xj − xj |. ˜ 1≤j ≤n Hence absolute errors in the one and two norms can overestimate the worst componentwise error by a factor that depends on the vector length n. Unfortunately, normwise relative errors give much less information about componentwise relative errors. Example. Let x be an approximation to a vector x where ˜ x= 1 , ǫ The normwise relative error 0 < ǫ ≪ 1, x −x ∞ ˜ x∞ x= ˜ 1 . 0 = ǫ is small. However, the componentwise −˜ relative error in the second component, |x2x2x2 | = 1, shows that x2 is a totally ˜ || inaccurate approximation to x2 in the relative sense. The preceding example illustrates that a normwise relative error can be small, even if individual vector elements have a large relative error. In the infinity norm, for example, the normwise relative error only bounds the relative Downloaded 01/17/14 to 143.215.200.123. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php 32 2. Sensitivity, Errors, and Norms error corresponding to a component of x with the largest magnitude. To see this, let |xk | = x ∞ . Then ˜ max1≤j ≤n |xj − xj | |xk − xk | ˜ x −x ∞ ˜ ≥ . = x∞ |xk | |xk | For the normwise relative errors in the one and two norms we incur additional factors that depend on the vector length n, x −x ˜ x1 1 ≥ ˜ 1 |xk − xk | , n |xk | x −x ˜ x2 2 ˜ 1 |xk − xk | . ≥√ n |xk | Therefore, normwise relative errors give no information about relative errors in components of smaller magnitude. If relative errors in individual vector components are important, then do not use normwise errors. Remark 2.11. When measuring the normwise relative error of an approximation ˜ ˜ x to x , the question is which error to measure, x −x or x −x ? If x ≈ x , ˜ ˜ x x ˜ then the two...
View Full Document

This note was uploaded on 01/30/2014 for the course MATH 1502 taught by Professor Mcclain during the Spring '07 term at Georgia Institute of Technology.

Ask a homework question - tutors are online