Usage: PETScOptions for mumps?

1

Hi,
I am trying to modify the preconditioner object associated with the Krylov subspace problem by following,

eps = SLEPc.EPS()
st = eps.getST()
st.setType(st.Type.SINVERT)
ksp = st.getKSP()
ksp.setType(ksp.Type.GCR)
pc = ksp.getPC()
pc.setType(pc.Type.LU)
pc.setFactorSolverPackage('mumps')

I get the following error.

AttributeError: 'petsc4py.PETSc.PC' object has no attribute 'setFactorSolverPackage'

Any suggestions would be appreciated!

2

Based on the petsc4py API here, do you maybe want setFactorSolverType? (Googling that and setFactorSolverPackage, it looks like there may have been an API change at some point.)

2 Likes
3

Thank you that was very helpful.

My code is very similar :

# Solver
E = slp.EPS(); E.create()
E.setOperators(-A,B) # Solve dM/dq*x=sigma*N*x (A*x+sigma*B*x=0)
E.setWhichEigenpairs(E.Which.LARGEST_REAL)
E.setDimensions(k) # Find k eigenvalues only
E.setTolerances(ae,ncv) # Set absolute tolerance and number of iterations
E.setProblemType(slp.EPS.ProblemType.PGNHEP) # Specify that A is no hermitian, but M is semi-positive
# Spectral transform
ST = E.getST()
# Krylov subspace
KSP = ST.getKSP()
KSP.setType('preonly')
# Preconditioner
PC =  KSP.getPC()
PC.setType('lu')
PC.setFactorSolverType('mumps')

Notice that I am trying to get the eigenvalues wiith the largest real part, not close to a target.

I understand that the petsc default is Krylov-Schur with GMRES inside every subspace, I tried above to move to Krylov-Schur with MUMPS. In in my case A is non Hermitian, B diagonal semi-positive.

The code above produces :

Traceback (most recent call last):
  File "xxx", line XX, in <module>
    yo.Eigenvalues(a,b)
  File "code.py", line 385, in Eigenvalues
    E.solve()
  File "SLEPc/EPS.pyx", line 1233, in slepc4py.SLEPc.EPS.solve
setFactorSolverPackagepetsc4py.PETSc.Error: error code 76
[0] EPSSolve() at /usr/local/slepc/src/eps/interface/epssolve.c:136
[0] EPSSetUp() at /usr/local/slepc/src/eps/interface/epssetup.c:350
[0] STSetUp() at /usr/local/slepc/src/sys/classes/st/interface/stsolve.c:582
[0] STSetUp_Shift() at /usr/local/slepc/src/sys/classes/st/impls/shift/shift.c:107
[0] KSPSetUp() at /usr/local/petsc/src/ksp/ksp/interface/itfunc.c:408
[0] PCSetUp() at /usr/local/petsc/src/ksp/pc/interface/precon.c:1016
[0] PCSetUp_LU() at /usr/local/petsc/src/ksp/pc/impls/factor/lu/lu.c:133
[0] MatLUFactorNumeric() at /usr/local/petsc/src/mat/interface/matrix.c:3186
[0] MatFactorNumeric_MUMPSpetsc4py.PETSc.Error: error code 76
[0] EPSSolve() at /usr/local/slepc/src/eps/interface/epssolve.c:136
[0] EPSSetUp() at /usr/local/slepc/src/eps/interface/epssetup.c:350
[0] STSetUp() at /usr/local/slepc/src/sys/classes/st/interface/stsolve.c:582
[0] STSetUp_Shift() at /usr/local/slepc/src/sys/classes/st/impls/shift/shift.c:107
[0] KSPSetUp() at /usr/local/petsc/src/ksp/ksp/interface/itfunc.c:408
[0] PCSetUp() at /usr/local/petsc/src/ksp/pc/interface/precon.c:1016
[0] PCSetUp_LU() at /usr/local/petsc/src/ksp/pc/impls/factor/lu/lu.c:133
[0] MatLUFactorNumeric() at /usr/local/petsc/src/mat/interface/matrix.c:3186
[0] MatFactorNumeric_MUMPS() at /usr/local/petsc/src/mat/impls/aij/mpi/mumps/mumps.c:1686
[0] Error in external library
[0] Error reported by MUMPS in numerical factorization phase: INFOG(1)=-9, INFO(2)=4008881() at /usr/local/petsc/src/mat/impls/aij/mpi/mumps/mumps.c:1686
[0] Error in external library
[0] Error reported by MUMPS in numerical factorization phase: INFOG(1)=-9, INFO(2)=4008881

I found here a decoding of this error stating :

image
image1460×129 20.2 KB

I have no idea how to fix this…

I am using dolfinx with a docker installation, activated in complex mode - I believe slepc4py came with it.

4

Quick update : I’ve taken inspiration from p101 of this book and added at the beginning of my script :

import petsc4py
petsc4py.init(['-mat_mumps_icntl_14 100'])

This has absolutely no effect.

5

This not really a fix (sorry) but I had a similar problem and using PC.setFactorSolverType('superlu_dist') did not lead to any failure.