[petsc-users] Spectrum Slicing with MATMPISBAIJ

[Jose E. Roman]

If you look at section 3.4.5 (Spectrum slicing) of the SLEPc users manual, you will see it is CHOLESKY that you need, not LU, so both AIJ and SBAIJ should work.

PETSc's LU and Cholesky are sequential, i.e., they can only be used in sequential runs, or as a local preconditioner in parallel runs. If you plan to run your code with several MPI processes, you should configure PETSc with an external package such as MUMPS (those with the X in the Parallel column). The Complex column indicates if the factorization is available in complex scalar builds.

Jose
 

> El 31 jul 2025, a las 12:07, Fischer, Franz via petsc-users <petsc-users at mcs.anl.gov> escribió:
> 
> Dear all,
> 
> I am dealing with large, sparse symmetric (and hermitian) matrices (N ~ 1e5) and I am using slepc in order to solve them.
> For now, I was using the matrix type MATMPISBAIJ to fill my matrix in parallel and I was solving the EVP for the EPS_SMALLEST_REAL eigenvalues just fine.
> 
> Now I am dealing with spectra, where I am not desired in knowing the small real part of the eigenvalue spectrum, but rather in some energy interval - thus, making me want to use spectrum slicing.
> 
> For that I have browsed your documentation and now I am a little concerned on what to do.
> I saw that my matrix type is not supported for LU decomposition, so I have converted my matrix to type MATMPIAIJ.
> There is this table here (https://urldefense.us/v3/__https://petsc.org/release/overview/linear_solve_table/__;!!G_uCfscf7eWS!fe4ysln5TfzUVOsGUnRA53e7CXOZAr8euf_lpIw-8QBTEN4ZFAk_iwlgW4f9EM_Ymn_tCdWH1roT2mD1PnyqDu8G$ ), where I found which solvers are available for which matrix type.
> 
> What I can not fully understand are the columns Parallel and Complex. What does it mean to have an X in there, does it mean it is possible or not?
> 
> Do I need to compile Petsc with external LU-solver packages or not?
> 
> Thanks in advance for your reply!
> 
> Best,
> Franz
> 
> ---------------------------------------------------------
> MSc. Franz Fischer
> Universität Hamburg
> HARBOR, Geb. 610
> Luruper Chaussee 149
> D-22761 Hamburg
> ---------------------------------------------------------
>