[petsc-users] Matrix missing diagonal entry

Barry Smith bsmith at mcs.anl.gov
Tue Oct 14 11:24:22 CDT 2014


  Should be ok.

On Oct 14, 2014, at 10:00 AM, Florian Lindner <mailinglists at xgm.de> wrote:

> Am Dienstag, 7. Oktober 2014, 08:34:55 schrieb Barry Smith:
>> 
>>   PETSc has this “extra” requirement that you provide values for the entire diagonal of the matrix, even those locations with zero. So make sure that you use MatSetValues() to put a zero on each diagonal entry that has zero.
> 
> Thanks for that information. I came up with this scheme:
> 
>  _matCLU.assemble(MAT_FLUSH_ASSEMBLY);
>  petsc::Vector zeros(_matCLU);   // construct a vector like the given matrix
>  MatDiagonalSet(_matCLU.matrix, zeros.vector, ADD_VALUES);
>  _matCLU.assemble(MAT_FINAL_ASSEMBLY);
> 
> (contains a bit of my petsc helper functions but should be clear I guess)
> 
> To set all unset entries to zero. Do you see any problem with that?
> 
> Best Regards,
> Florian
> 
>> 
>>  Also from below you are using ICC that is incomplete cholesky factorization. If you want to use a direct solver you should select -pc_type cholesky
>> 
>> 
>>   Barry
>> 
>> On Oct 7, 2014, at 8:14 AM, Florian Lindner <mailinglists at xgm.de> wrote:
>> 
>>> Hello,
>>> 
>>> when I try to KSPSolve my matrix petsc prints out that a diagonal entry is missing:
>>> 
>>> KSPSetOperators(solver, _matCLU.matrix, _matCLU.matrix );
>>> KSPSolve(solver, vin.vector, vp.vector);
>>> 
>>> And petsc is perfectly right about that specific entry being zero. So this may be more of a mathematical problem then pure petsc.
>>> 
>>> The algorithm that I try to reimplement does the following:
>>> 
>>> lu(_matrixCLU, _pivotsCLU);  // Compute LU decomposition
>>> // Left out: Permutate in according to pivotsCLU
>>> forwardSubstitution(_matrixCLU, in, y); // CLU^-1 * in = y  (lower triangle of CLU)
>>> backSubstitution(_matrixCLU, y, p);     // CLU^-1 * y = p (upper triangle of CLU)
>>> multiply(_matrixA, p, out );            // out = A * p (not relevant here)
>>> 
>>> I want to replace to lu, forward- and backwardSubstituation by using a petsc KSP solver. Before doing any of these operations _matrixCLU and _matCLU are equal:
>>> 
>>> =============== CLU Before LU =======
>>> ==== Dynamic Matrix of size 7 x 7 ====
>>> 1.0000 0.3679 0.1353 0.3679 1.0000 0.0000 0.0000
>>> 0.3679 1.0000 0.3679 0.1353 1.0000 1.0000 0.0000
>>> 0.1353 0.3679 1.0000 0.3679 1.0000 1.0000 1.0000
>>> 0.3679 0.1353 0.3679 1.0000 1.0000 0.0000 1.0000
>>> 1.0000 1.0000 1.0000 1.0000 0.0000 0.0000 0.0000
>>> 0.0000 1.0000 1.0000 0.0000 0.0000 0.0000 0.0000
>>> 0.0000 0.0000 1.0000 1.0000 0.0000 0.0000 0.0000
>>> 
>>> Mat Object:CLU 1 MPI processes
>>> type: seqsbaij
>>> row 0: (0, 1)  (1, 0.367879)  (2, 0.135335)  (3, 0.367879)  (4, 1)  (5, 0)  (6, 0) 
>>> row 1: (1, 1)  (2, 0.367879)  (3, 0.135335)  (4, 1)  (5, 1)  (6, 0) 
>>> row 2: (2, 1)  (3, 0.367879)  (4, 1)  (5, 1)  (6, 1) 
>>> row 3: (3, 1)  (4, 1)  (5, 0)  (6, 1) 
>>> row 4:
>>> row 5:
>>> row 6:
>>> 
>>> LU decomposition of matrixCLU works, no rank deficiency (no 0 on the main diagonal after LU and I tested for full rank of matrixCLU). Petsc complains with this matrix:
>>> 
>>> [0]PETSC ERROR: --------------------- Error Message --------------------------------------------------------------
>>> [0]PETSC ERROR: Object is in wrong state
>>> [0]PETSC ERROR: Matrix is missing diagonal entry 4
>>> [0]PETSC ERROR: See http://www.mcs.anl.gov/petsc/documentation/faq.html for trouble shooting.
>>> [0]PETSC ERROR: Petsc Release Version 3.5.2, unknown 
>>> [0]PETSC ERROR: ../build/debug/binprecice on a arch-linux2-c-debug named helium by lindnefn Mon Oct  6 16:26:49 2014
>>> [0]PETSC ERROR: Configure options 
>>> [0]PETSC ERROR: #1 MatICCFactorSymbolic_SeqSBAIJ() line 2430 in /data2/scratch/lindner/petsc/src/mat/impls/sbaij/seq/sbaijfact2.c
>>> [0]PETSC ERROR: #2 MatICCFactorSymbolic() line 6266 in /data2/scratch/lindner/petsc/src/mat/interface/matrix.c
>>> [0]PETSC ERROR: #3 PCSetup_ICC() line 20 in /data2/scratch/lindner/petsc/src/ksp/pc/impls/factor/icc/icc.c
>>> [0]PETSC ERROR: #4 PCSetUp() line 902 in /data2/scratch/lindner/petsc/src/ksp/pc/interface/precon.c
>>> [0]PETSC ERROR: #5 KSPSetUp() line 305 in /data2/scratch/lindner/petsc/src/ksp/ksp/interface/itfunc.c
>>> [0]PETSC ERROR: #6 KSPSolve() line 417 in /data2/scratch/lindner/petsc/src/ksp/ksp/interface/itfunc.c
>>> [0]PETSC ERROR: #7 map() line 696 in src/mapping/PetRadialBasisFctMapping.hpp
>>> numerical equality test failed: 0.00000000000000000000e+00 instead of 1.00000000000000000000e+00
>>> file: src/mapping/tests/PetRadialBasisFctMappingTest.cpp       line: 196
>>> statement: value==1.0
>>> difference:  1.00000000000000000000e+00
>>> 
>>> 
>>> What's indeed perfectly right. But what is the best way to cope with that?
>>> 
>>> Thanks,
>>> Florian
>> 



More information about the petsc-users mailing list