On 03/16/2013 07:53 PM, Jens Lohne Eftang wrote:
> On 03/15/2013 04:09 PM, Roy Stogner wrote:
>> On Fri, 15 Mar 2013, Jens Lohne Eftang wrote:
>>
>>> 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 ...?
>> Nope.  As John said, FEMSystem doesn't have any way to tell whether
>> your problem is linear or not.  If you've got a linear problem, then
>> it's up to you to tell the nonlinear solver to avoid doing any
>> redundant work, and if you want it to be robust with any solver (or
>> with time integration) then it's up to you to define a
>> solution-dependent residual consistently.
>>
>> Usually this isn't too hard - E.g. if I was solving
>>
>> (grad u, grad v) = (f, v)
>>
>> Then my residual is (f,v)-(grad u, grad v) instead of plain (f,v).
>>
>> If you know you're never going to evaluate your problem from a
>> starting point other than u=0, then you can get away with plain (f,v)
>> as long as you turn off any nonlinear solver option that might call a
>> residual after the first step.  But usually people intend to do
>> nonlinear or transient terms eventually too, so it's easier to just
>> start with the linear terms written in a consistent way.
> Got it! Thanks guys!
>
> Jens


FWIW with FEMSystem I really like using the the numerical Jacobian 
feature when debugging. If the numerical Jacobian version works then 
there is probably a bug in the assembly code for the Jacobian.

David


------------------------------------------------------------------------------
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