On 03/15/2013 03:49 PM, Roy Stogner wrote:
>
> On Fri, 15 Mar 2013, Jens Lohne Eftang wrote:
>
>> What do you mean by that exactly. My Jacobian is A and the residual 
>> is Ax-b. The linear solver seems to successfully solve x=A\b so I 
>> don't see how it's possible that something is inconsistent 
>> (independent of whether my assembly loops are actually correct ... )
>
> Sure it's possible.
>
> A common example: your residual is r(x):=Ax-b; if you instead just
> always return "b", then (assuming your A and b are correct and you
> don't set any initial guess) the first residual evaluation
> r(0)=A*0-b=b will be correct, the first solve will be correct, you'll
> get approximately x=inv(A)*b after this step.. but then when the
> solver asks for a nonlinear residual to test, you'll be handing it "b"
> instead of "A*x-b"="0", and it'll think you haven't converged on the
> nonlinear problem at all.  So it will try a line search, but that
> search will never find a residual reduction no matter how many
> intermediate values of "x" it tries.
Hmm, ok, and it does seem like that's what happens as my residual 
doesn't decrease.

But when the residual is computed how can I hand it just "b," ? I 
thought it would calculate it as A*x-b based on my A and b ...?

Jens

------------------------------------------------------------------------------
Everyone hates slow websites. So do we.
Make your web apps faster with AppDynamics
Download AppDynamics Lite for free today:
http://p.sf.net/sfu/appdyn_d2d_mar
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to