# 37 to eliminate reveals the orthogonality condition r

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

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: , then we may write the above expression as (r( ) )T p( ) = (r( ) )T r( ) = 0 1 ::: : (9.3.9b) 36 Solution Techniques for Elliptic Problems Let us next expand (9.3.6) while using (9.3.1) and (9.3.2b) to show that E (x( +1) ) = E (x( ) ) ; (r( ) )T p( ) + (1=2) 2 (p( ) )T Ap( ) : Using (9.3.7) and (9.3.9b) () T ()2 1 () T ()2 E (x( +1) ) = E (x( ) ) ; 2 (r( ) )T p (] ) = E (x( ) ) ; 1 (r( ) )T r (] ) : (9.3.10) 2 (p ) Ap (p ) Ap Having (9.3.10), we see that the error E (x( +1) ) is decreased most rapidly when (p( ) )T Ap( ) is minimal. This will be our criterion for determining . Using (9.3.5c), we have (p( ) )T Ap( ) = (r( ) + ;1 p( ;1) )T A(r( ) + ;1p( ;1) ): Minimizing with respect to ;1 gives (r( ))T Ap( ;1) ;1 = ; (p( ;1) )T Ap( ;1) or reindexing r( +1) T Ap( ) = ; ((p( ) )) Ap( ) T = 0 1 ::: : (9.3.11) Using (9.3.5c) (p( +1) )T Ap( ) = (r( +1) )T Ap( ) + (p( ) )T Ap( ) which, upon use of (9.3.11), reveals (p( +1) )T Ap( ) = 0 = 0 1 ::: : (9.3.12a) Thus, the search directions are orthogonal with respect to a strain energy inner product. We usually call this a conjugacy condition and say that the search directions are conjugate. Using (9.3.12a) with (9.3.5c), we have (p( ) )T Ap( ) = (r( ) )T Ap( ) + ;1 (p( ;1) )T Ap( ) = (r( ) )T Ap( ) : Combining this result with (9.3.8) yields (r( +1) )T r( ) = (r( ) )T r( ) ; (p( ) )T Ar( ) = (r( ) )T r( ) ; (p( ) )T Ap( ) : 9.3. Conjugate Gradient Methods 37 Using (9.3.7) and (9.3.9b), we nd the orthogonality relation (r( +1) )T r( ) = 0: (9.3.12b) Equation (9.3.11) can be put in a slightly simpler form by using (9.3.8) and (9.3.12b) to obtain (r( +1) )T r( +1) = (r( +1) )T r( ) ; (r( +1) )T Ap( ) = ; (r( +1) )T Ap( ) : Using this with (9.3.11) and (9.3.7) ( +1) )T ( +1) ( +1) T ( +1) = 1 (r ( ) T r ( ) = (r ( )) T r ( ) : (p ) r (p ) Ap Finally, using (9.3.9b) ( +1) )T r( +1) = (r (r( ) )T r( ) : We summarize our ndings as a theorem. (9.3.12c) Theorem 9.3.1. The residuals and search directions of the conjugate gradient method satisfy (r( ) )T...
View Full Document

## This document was uploaded on 03/16/2014 for the course CSCI 6840 at Rensselaer Polytechnic Institute.

Ask a homework question - tutors are online