We'd need the full error message to determine what is happening. But note that since the lower right block is all 0 any attempt to use a factorization of that inside a preconditioner will fail with zero pivot.
> On Feb 2, 2016, at 10:40 PM, keguoyi <[email protected]> wrote: > > Dear PETSc users, > > This is Guoyi Ke, a graduate student of Texas Tech University. I have a 2D > Navier Stokes problem that has block matrices: J=[F B^T; B 0]. I use > Schur complement preconditioner for block J. The code was set as: > > PCFieldSplitSetType( pc, PC_COMPOSITE_SCHUR ); > PCFieldSplitSetSchurFactType(pc, PC_FIELDSPLIT_SCHUR_FACT_LOWER); > PCFieldSplitSetSchurPre(pc,PC_FIELDSPLIT_SCHUR_PRE_SELFP,NULL); > > For both velocity block (fieldsplit: 0) and pressure block (fieldsplit: 1) , > we set the KSP as KSPPREONLY and PC as PCILU. For this case, it works fine > for us. > However, in order to use LSC we change the code as: > > PCFieldSplitSetType( pc, PC_COMPOSITE_SCHUR ); > PCFieldSplitSetSchurFactType(pc, PC_FIELDSPLIT_SCHUR_FACT_LOWER); > PCFieldSplitSetSchurPre(pc,PC_FIELDSPLIT_SCHUR_PRE_SELF,NULL); > > We keep the same setup for velocity block (fieldsplit: 0) as above. For > pressure block (fieldsplit: 1), we set the KSP as KSPGMRES and PC as PCLSU. I > got the following errors: > > [0]PETSC ERROR: --------------------- Error Message > -------------------------------------------------------------- > [0]PETSC ERROR: Object is in wrong state > [0]PETSC ERROR: Matrix is missing diagonal entry 0 > > > Any suggestion and help will be highly appreciated. Thank you so much! > > Best, > Guoyi
