[petsc-users] Question about KSPSolve

[John Yawney]

Hi Matt,

Thank you for the response. I've rerun the test using 2 processors and in
serial and got similar results. I turned off my implicit viscosity terms
though and it appears as though the pressure solve produces very little
error on its own. I'll have to check over my FD laplacian matrices used for
the viscous terms and see if I can spot any errors in their definitions.

Thanks again,
John


On Fri, Aug 2, 2013 at 4:28 PM, Matthew Knepley <knepley at gmail.com> wrote:

> On Sat, Aug 3, 2013 at 1:27 AM, John Yawney <jyawney123 at gmail.com> wrote:
>
>> Good Afternoon Everyone,
>>
>> I'm using PETSc to solve some linear systems in my ocean model. Currently
>> I'm using the KSPSolve environment with 4 MPI processors. I've established
>> the matrix A and the RHS b and confirmed that everything looks correct
>> using VecView and MatView. Here are some code snippets that show the basic
>> steps I took.
>>
>> -------------------------------------------------------------------------
>> *Assembling A:*
>> MatConvert(matLaplacianX, MATSAME, MAT_INITIAL_MATRIX, &matLaplacian);
>> MatAssemblyBegin(matLaplacian, MAT_FINAL_ASSEMBLY);
>> MatAssemblyEnd(matLaplacian, MAT_FINAL_ASSEMBLY);
>>
>> MatAXPY(matLaplacian, 1.0, matLaplacianY, DIFFERENT_NONZERO_PATTERN);
>> MatAssemblyBegin(matLaplacian, MAT_FINAL_ASSEMBLY);
>> MatAssemblyEnd(matLaplacian, MAT_FINAL_ASSEMBLY);
>>
>> MatAXPY(matLaplacian, 1.0, matLaplacianZ, DIFFERENT_NONZERO_PATTERN);
>> MatAssemblyBegin(matLaplacian, MAT_FINAL_ASSEMBLY);
>> MatAssemblyEnd(matLaplacian, MAT_FINAL_ASSEMBLY);
>>
>> *Defining KSP environment:*
>> KSPCreate(MPI_COMM_WORLD, &m_inksp);
>> KSPSetOperators(m_inksp, matLaplacian, matLaplacian,
>> DIFFERENT_NONZERO_PATTERN);
>> KSPSetType(m_inksp, KSPGMRES);
>> KSPSetInitialGuessNonzero(m_inksp,PETSC_TRUE);
>> KSPSetFromOptions(m_inksp);
>> KSPSetUp(m_inksp);
>>
>> *Defining RHS vector:*
>> VecCreateMPI(MPI_COMM_WORLD, nLocalElements, nGlobalElements, &m_vecRHS);
>>
>> *Solving the linear system:*
>> VecAssemblyBegin(m_vecRHS);
>> VecAssemblyEnd(m_vecRHS);
>> KSPSolve(m_inksp, m_vecRHS, m_vecPressure);
>> -------------------------------------------------------------------------
>>
>> If I modify my problem to consider a 2D (x-z) domain with flat bottom
>> topography and I set the initial velocity fields to 0 and a constant
>> density of 1025 throughout, then if I compute a number of time steps I get
>> computational artifacts at the beginning and end locations of each block. I
>> should also mention I'm only splitting up the domain into sub-blocks in the
>> x direction currently. After about 10 time steps, the min density is off by
>> about 1E-8 but only at these locations. I've attached a figure to
>> demonstrate the errors.
>>
>> Are there ways for me to remove these errors? Should I be looking at the
>> DM manual pages?
>>
>
> The above looks correct, so I assume there is a problem with the
> definition of the system. I would try putting
> in an exact solution, or comparing a serial and parallel run.
>
>   Thanks,
>
>      Matt
>
>
>> Thanks for any help and suggestions.
>>
>> All the best,
>> John
>>
>>
>>
>
>
> --
> What most experimenters take for granted before they begin their
> experiments is infinitely more interesting than any results to which their
> experiments lead.
> -- Norbert Wiener
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20130803/a0ed3647/attachment.html>