[petsc-dev] New DMDA boundary type?

[Aron Ahmadia]

Lisandro wrote this. Not sure if it is checked in. 

Sent from my iPhone

On Jun 8, 2012, at 5:18 PM, Richard Katz <Richard.Katz at earth.ox.ac.uk> wrote:

> Hi all,
> 
> Writing to ask that you again consider this feature request.  
> 
> Thanks,
> Rich
> 
> 
> On Jan 8, 2012 2:43 PM, "Richard Katz" <richard.katz at earth.ox.ac.uk> wrote:
>> Dear all,
>> 
>> I wonder if I can interest you in a new DMDA boundary type... that I'd probably need for you to implement... and it might be slightly painful.
>> 
>> The idea is a periodic boundary with a 180 degree topological "twist." (i.e. Mobius strip in 2D).  I'll call it DMDA_BOUNDARY_PERIODIC_TWIST here for sake of argument.
>> 
>> If I have a 1D DMDA, this is meaningless.
>> 
>> If I have a 2D array with DMDA_BOUNDARY_PERIODIC_TWIST as my periodic type in the y-direction, then the row of ghost points at the y-limits of the grid would be as for DMDA_BOUNDARY_PERIODIC, but with the order of values in the x-direction reversed.
>> 
>> In 3D this would be more complicated, because a PERIODIC_TWIST boundary condition in one direction could involve twists in either/both of the other directions.
>> 
>> I'd be thrilled with just the 2D version.  What do you think?
>> 
>> Cheers,
>> Rich
>> 
>> 
>> ps - my motivation: consider a 2D fluid undergoing simple shear as a background strain-rate field.  I am solving for a perturbation to this background field. I would like the perturbation to be periodic in all directions.  Because the perturbation field is advected by the background field, this can only work with Mobius-strip periodicity in the direction normal to the shear direction.
> 
> 
> 
> On 23 May 2012, at 15:27, Jed Brown wrote:
> 
>> We're basically in code freeze for the 3.3 release now, but I will get back to this. Sorry for dragging on.
>> 
>