[petsc-users] conditioning of snes with dmcomposite & grid sequencing

[Gideon Simpson]

Ok, I have been able to recast so that the components are properly separated and grid sequencing appears to work.  
-gideon

> On Sep 12, 2015, at 5:50 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> 
>> 
>> On Sep 12, 2015, at 4:38 PM, Gideon Simpson <gideon.simpson at gmail.com> wrote:
>> 
>> 
>>> On Sep 12, 2015, at 4:43 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
>>> 
>>>   I do not understand. You have the real part of the solution at the origin of u0 stored in the "redundant" part of the DMComposite? What about the on the DMDA mesh? Wouldn't the real part also be there and with the same value? Thus when you interpolate with the solution on the DMDA you would still get the correct values "near" the origin? Hence there is no reason to "involve" the u0 stored in the redundant part in the interpolation?
>>> 
>>> Barry
>> 
> 
>   Oh, you are doing a 1d problem! 
> 
>   Why put that x0 in a different place then the "interior points"? I would never do that because you are right the whole interpolation business is messed up.  Can you reorganize things so that x0 is just included in the DMDA? If not, why not and we'll fix them. It would be insanely cumbersome and totally unnecessary IMHO to build a complicated construct to handle the interpolation with the x0 in a different place. Jeepers, of course the interpolation won't work.
> 
>  Barry
> 
> 
> 
>> Suppose my coarse mesh is x1, x2, x3,…, and x0 = 0, corresponds to the origin.  At x0=0, my constraints are that u’(0) = 0, and v(0) = 0.  So u0, the value at the origin, is an unknown that needs to be solved for.  Now, when I do grid sequencing, I ought to be adding a point between x0 and x1, but I don’t, because x0 isn’t part of the DMDA.  

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20150914/4a101c5d/attachment.html>