[petsc-users] SLEPc Eigenvalue problem - Zero Pivot

[Jose E. Roman]

El 14/05/2015, a las 19:34, venkatesh g escribió:

> Actually krylovschur it is giving excellent results when my B matrix is not singular. 
> 
> This time my B matrix is large with very little non-zero entries. There are entire rows and columns with zeros. 
> So other than MUMPS, which I will try now, is there another way ?
> 
> Venkatesh
> 

This line:

$ ./ex7 -f1 A1 -f2 B1 -st_type sinvert -eps_target 0.5

will factor A-sigma*B for sigma=0.5 if command line arguments are processed correctly. So even if B has many zeros, the factorization should not fail. You could try tweaking PETSc's LU factorization parameters, or use an external LU library such as MUMPS or UMFPACK, to see if that helps. This has nothing to do with the (outer) solver used for the eigenvalues.

Jose