<div dir="ltr"><div><div>Note, I suspect the zero pivot is coming from a coarse grid. I don't know why no-inode fixed it. N.B., this might not be deterministic.<br><br></div>If you run with -info and grep on 'GAMG' you will see the block sizes. They should be 3 on the fine grid and the 6 on the coarse grids if you set everything up correctly.<br><br></div>Mark<br></div><div class="gmail_extra"><br><div class="gmail_quote">On Wed, Nov 11, 2015 at 1:36 PM, David Knezevic <span dir="ltr"><<a href="mailto:david.knezevic@akselos.com" target="_blank">david.knezevic@akselos.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div><div class="h5">On Wed, Nov 11, 2015 at 12:57 PM, David Knezevic <span dir="ltr"><<a href="mailto:david.knezevic@akselos.com" target="_blank">david.knezevic@akselos.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div><div><div class="gmail_extra"><div class="gmail_quote">On Wed, Nov 11, 2015 at 12:24 PM, David Knezevic <span dir="ltr"><<a href="mailto:david.knezevic@akselos.com" target="_blank">david.knezevic@akselos.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div><div><div class="gmail_quote">On Tue, Nov 10, 2015 at 10:28 PM, David Knezevic <span dir="ltr"><<a href="mailto:david.knezevic@akselos.com" target="_blank">david.knezevic@akselos.com</a>></span> wrote:<br></div></div></div><div class="gmail_quote"><div><div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div><div>On Tue, Nov 10, 2015 at 10:24 PM, Matthew Knepley <span dir="ltr"><<a href="mailto:knepley@gmail.com" target="_blank">knepley@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div><div>On Tue, Nov 10, 2015 at 9:21 PM, David Knezevic <span dir="ltr"><<a href="mailto:david.knezevic@akselos.com" target="_blank">david.knezevic@akselos.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Tue, Nov 10, 2015 at 10:00 PM, Matthew Knepley <span dir="ltr"><<a href="mailto:knepley@gmail.com" target="_blank">knepley@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div><div>On Tue, Nov 10, 2015 at 8:39 PM, David Knezevic <span dir="ltr"><<a href="mailto:david.knezevic@akselos.com" target="_blank">david.knezevic@akselos.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div>I'm looking into using GAMG, so I wanted to start with a simple 3D elasticity problem. When I first tried this, I got the following "zero pivot" error:</div><div><br></div><div>-----------------------------------------------------------------------</div><div><br></div><div><div>[0]PETSC ERROR: Zero pivot in LU factorization: <a href="http://www.mcs.anl.gov/petsc/documentation/faq.html#zeropivot" target="_blank">http://www.mcs.anl.gov/petsc/documentation/faq.html#zeropivot</a></div><div>[0]PETSC ERROR: Zero pivot, row 3</div><div>[0]PETSC ERROR: See <a href="http://www.mcs.anl.gov/petsc/documentation/faq.html" target="_blank">http://www.mcs.anl.gov/petsc/documentation/faq.html</a> for trouble shooting.</div><div>[0]PETSC ERROR: Petsc Release Version 3.6.1, Jul, 22, 2015 </div><div>[0]PETSC ERROR: /home/dknez/akselos-dev/scrbe/build/bin/fe_solver-opt_real on a arch-linux2-c-opt named david-Lenovo by dknez Tue Nov 10 21:26:39 2015</div><div>[0]PETSC ERROR: Configure options --with-shared-libraries=1 --with-debugging=0 --download-suitesparse --download-parmetis --download-blacs --with-blas-lapack-dir=/opt/intel/system_studio_2015.2.050/mkl --CXXFLAGS=-Wl,--no-as-needed --download-scalapack --download-mumps --download-metis --download-superlu_dist --prefix=/home/dknez/software/libmesh_install/opt_real/petsc --download-hypre --download-ml</div><div>[0]PETSC ERROR: #1 PetscKernel_A_gets_inverse_A_5() line 48 in /home/dknez/software/petsc-3.6.1/src/mat/impls/baij/seq/dgefa5.c</div><div>[0]PETSC ERROR: #2 MatSOR_SeqAIJ_Inode() line 2808 in /home/dknez/software/petsc-3.6.1/src/mat/impls/aij/seq/inode.c</div><div>[0]PETSC ERROR: #3 MatSOR() line 3697 in /home/dknez/software/petsc-3.6.1/src/mat/interface/matrix.c</div><div>[0]PETSC ERROR: #4 PCApply_SOR() line 37 in /home/dknez/software/petsc-3.6.1/src/ksp/pc/impls/sor/sor.c</div><div>[0]PETSC ERROR: #5 PCApply() line 482 in /home/dknez/software/petsc-3.6.1/src/ksp/pc/interface/precon.c</div><div>[0]PETSC ERROR: #6 KSP_PCApply() line 242 in /home/dknez/software/petsc-3.6.1/include/petsc/private/kspimpl.h</div><div>[0]PETSC ERROR: #7 KSPInitialResidual() line 63 in /home/dknez/software/petsc-3.6.1/src/ksp/ksp/interface/itres.c</div><div>[0]PETSC ERROR: #8 KSPSolve_GMRES() line 235 in /home/dknez/software/petsc-3.6.1/src/ksp/ksp/impls/gmres/gmres.c</div><div>[0]PETSC ERROR: #9 KSPSolve() line 604 in /home/dknez/software/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c</div><div>[0]PETSC ERROR: #10 KSPSolve_Chebyshev() line 381 in /home/dknez/software/petsc-3.6.1/src/ksp/ksp/impls/cheby/cheby.c</div><div>[0]PETSC ERROR: #11 KSPSolve() line 604 in /home/dknez/software/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c</div><div>[0]PETSC ERROR: #12 PCMGMCycle_Private() line 19 in /home/dknez/software/petsc-3.6.1/src/ksp/pc/impls/mg/mg.c</div><div>[0]PETSC ERROR: #13 PCMGMCycle_Private() line 48 in /home/dknez/software/petsc-3.6.1/src/ksp/pc/impls/mg/mg.c</div><div>[0]PETSC ERROR: #14 PCApply_MG() line 338 in /home/dknez/software/petsc-3.6.1/src/ksp/pc/impls/mg/mg.c</div><div>[0]PETSC ERROR: #15 PCApply() line 482 in /home/dknez/software/petsc-3.6.1/src/ksp/pc/interface/precon.c</div><div>[0]PETSC ERROR: #16 KSP_PCApply() line 242 in /home/dknez/software/petsc-3.6.1/include/petsc/private/kspimpl.h</div><div>[0]PETSC ERROR: #17 KSPSolve_CG() line 139 in /home/dknez/software/petsc-3.6.1/src/ksp/ksp/impls/cg/cg.c</div><div>[0]PETSC ERROR: #18 KSPSolve() line 604 in /home/dknez/software/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c</div></div><div><br></div><div>-----------------------------------------------------------------------<br></div><div><br></div><div>I saw that there was a thread about this in September (subject: "gamg and zero pivots"), and that the fix is to use "<span style="color:rgb(0,0,0);white-space:pre-wrap">-mg_levels_pc_type jacobi." When I do that, the solve succeeds (I pasted the -ksp_view at the end of this email).</span></div><div><span style="color:rgb(0,0,0);white-space:pre-wrap"><br></span></div><div><span style="color:rgb(0,0,0);white-space:pre-wrap">So I have two questions about this:</span></div><div><span style="color:rgb(0,0,0);white-space:pre-wrap"><br></span></div><div>1. Is it surprising that I hit this issue for a 3D elasticity problem? Note that matrix assembly was done in libMesh, I can look into the structure of the assembled matrix more carefully, if needed. Also, note that I can solve this problem with direct solvers just fine.</div></div></blockquote><div><br></div></div></div><div>Yes, this seems like a bug, but it could be some strange BC thing I do not understand.</div></div></div></div></blockquote><div><br></div><div><br></div><div>OK, I can look into the matrix in more detail. I agree that it should have a non-zero diagonal, so I'll have a look at what's happening with that.</div><div><br></div><div><br></div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div>Naively, the elastic element matrix has a nonzero diagonal. I see that you are doing LU<br></div><div>of size 5. That seems strange for 3D elasticity. Am I missing something? I would expect</div><div>block size 3.</div></div></div></div></blockquote><div><br></div><div><br></div><div>I'm not sure what is causing the LU of size 5. Is there a setting to control that?<br></div><div><br></div><div>Regarding the block size: I set the vector and matrix block size to 3 via VecSetBlockSize and MatSetBlockSize. I also used MatNullSpaceCreateRigidBody on a vector with block size of 3, and set the matrix's near nullspace using that.</div></div></div></div></blockquote><div><br></div></div></div><div>Can you run this same example with -mat_no_inode? I think it may be a strange blocking that is causing this.</div></div></div></div></blockquote><div><br></div><div><br></div></div></div><div>That works. The -ksp_view output is below.</div></div></div></div></blockquote><div><br></div><div><br></div></div></div><div>I just wanted to follow up on this. I had a more careful look at the matrix, and confirmed that there are no zero entries on the diagonal (as expected for elasticity). The matrix is from one of libMesh's example problems: a simple cantilever model using HEX8 elements.</div><div><br></div><div>Do you have any further thoughts about what might cause the "strange blocking" that you referred to? If there's something non-standard that libMesh is doing with the blocks, I'd be interested to look into that. I can send over the matrix if that would be helpful.</div><div><br></div><div>Thanks,<br>David</div><div><br></div></div></div></div>
</blockquote></div><br></div></div></div><div class="gmail_extra">P.S. I was previously calling VecSetBlockSize and MatSetBlockSize to set the block size to 3. When I don't do that, I no longer need to call -mat_no_inodes. I've pasted the -ksp_view output below. Does it look like that's working OK?</div></div></blockquote><div><br></div><div><br></div></div></div><div>Sorry for the multiple messages, but I think I found the issue. libMesh internally sets the block size to 1 earlier on (in PetscMatrix::init()). I guess it'll work fine if I get it to set the block size to 3 instead, so I'll look into that. (libMesh has an enable-blocked-storage configure option that should take care of this automatically.)<span class="HOEnZb"><font color="#888888"><br></font></span></div><span class="HOEnZb"><font color="#888888"><div><br></div><div>David</div><div><br></div><div><br></div></font></span></div></div></div>
</blockquote></div><br></div>