[petsc-users] ghoat point in DMDATSSetIFunctionLocal

[Smith, Barry F.]

  When you create the DMDA there is a "stencil" argument that can be star or box. For the 9 pt stencil you need to specify box.

   Barry

> On Jun 28, 2018, at 5:11 PM, Sepideh Kavousi <skavou1 at lsu.edu> wrote:
> 
> Hello, 
> I am solving Cahn-Hillard equation where I have Laplacian in my equation. In order to discretize the  Laplacian, I can discretize with  5-point or 9-point stencil. 
> 5 point stencile: laplacian=(aY[j][i+1].p+aY[j][i-1].p-2.0*aY[j][i].p)/(hx2))+((aY[j+1][i].p+aY[j-1][i].p-2.0*aY[j][i].p)/hy2)
> 
> 9-point stencil:(4.0*(aY[j][i+1].p+aY[j][i-1].p+aY[j+1][i].p+aY[j-1][i].p)+1.0*(aY[j+1][i+1].p+aY[j+1][i-1].p+aY[j-1][i+1].p+aY[j-1][i-1].p)-20.0*aY[j][i].p)/(6.0*hx2);
> 
> 
> I use DMDATSSetIFunctionLocal. In the function definition, if I use 5 point stencil, everything works fine, if I use a 9-point stencil, at the points which are in common between processors the answers behave strange and wrong. while for 9-point stencil,  I use single processor everything works fine. 
> Assume I have 4 processors and the domain that belongs to each processor is:
> (0:nx , 0:ny)   process one
> (nx+1:2nx , 0:ny)   process two
> (0:nx , ny+1:2*ny)   process three
> (nx+1:2nx , ny+1:2*n)   process four
> 
> I guess the problem is when processor one tries to access point (nx+1,ny+1). I did not find any data if the definition of local vector includes the point (nx+1,ny+1) for this processor or not. I would like to see how should I solve this issue.
> 
> Regards,
> Sepideh