Talk:Preconditioned conjugate gradient method

They're mostly duplicate - seems no need for separate page for the PCG method.

Anyone? Bueller? Thr PCG page is redundant since both methods are described (and better) by [conjugate gradient] and [preconditioner].

I would prefer to merge this page with "conjugate gradient method".
 * Absolutely. This is conjugate gradient method.  Not a preconditioner.  Let's keep like things together, instead of merging unlike by related things together.  128.114.60.100 01:02, 3 September 2007 (UTC)

This page appears to be in error in that (P^-1 A) will not in general be symmetric, which is one of the requirements for CG's application. (Is there something about CG I am not aware of that would allow CG to converge without symmetry here?) Preconditioning can be approached by letting P^-1=M^T M, which is possible since P is symmetric positive definite. Then, the system (M A M^T) (M^-T x) = (M b) can be solved using CG. Finally, the resulting CG iteration can be rearranged so that only M^T M is actually used, so that M itself needn't be computed.

This page has unsupported claims (about needing a symmetric preconditioner) and is partly wrong (see above comment). Can we delete it already? I tried a couple of months ago but the moderators jumped on me and reverted all my changes. Please can someone suggest a way to proceed. 68.101.160.213 (talk) 20:15, 19 April 2008 (UTC)


 * Will change this page into a #REDIRECT once finish checking if there is still something worth moving into Conjugate gradient method.&mdash;Kxx (talk) 16:34, 24 March 2009 (UTC)


 * Done.&mdash;Kxx (talk) 19:30, 24 March 2009 (UTC)