[petsc-users] MKL Pardiso returning NAN values

[Barry Smith]

  We don't by default, for example, check the norm of the solution returned by an external direct solver due to the added expense. 

  We do check the error condition returned by MKL Pardiso with 

  PetscCheck(mat_mkl_pardiso->err >= 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "Error reported by MKL PARDISO: err=%d. Please check manual", mat_mkl_pardiso->err);

 except (I don't know why) in  MatDestroy_MKL_PARDISO() 

So are you saying that MKL_Padiso is returning inf/nan in the solution but a return code less than or equal to zero?  Is this expected behavior of MKL Pardiso?



> On Jul 10, 2024, at 5:01 PM, Chris Hewson <chris at resfrac.com> wrote:
> 
> This Message Is From an External Sender
> This message came from outside your organization.
> Hi There,
> 
> We have a matrix that is singular and trying to solve it. We first use an iterative solve with KSPBCGS, the solution vector is nan values and the converged reason from PETSc of KSP_DIVERGED_NANORINF, that's great and what I would expect.
> 
> Sometimes in our program we redo a failed solve using the MKL Pardiso solver, when the same matrix and vectors get put into that solver which is a KSPPREONLY, I get KSP_CONVERGED_ITS as a converged reason and solution vector with nan values in it. 
> 
> Stepping through the PETSc calls, I see that the external call to Pardiso doesn't return an error for this, so not really the fault of PETSc, but curious if y'all have seen this before or a solution/workaround to this?
> 
> Chris Hewson
> Senior Reservoir Simulation Engineer
> ResFrac
> +1.587.575.9792

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20240711/53563698/attachment-0001.html>