[petsc-users] DIVERGED_NANORINF when using HYPRE/BoomerAMG
Andrea Iob
andrea_iob at hotmail.com
Wed Jul 1 10:14:02 CDT 2020
Yes, I did use "-pc_gamg_agg_nsmooths 0".
Andrea
________________________________
From: Mark Adams <mfadams at lbl.gov>
Sent: Wednesday, July 1, 2020 5:04 PM
To: Andrea Iob <andrea_iob at hotmail.com>
Cc: Matthew Knepley <knepley at gmail.com>; petsc-users at mcs.anl.gov <petsc-users at mcs.anl.gov>
Subject: Re: [petsc-users] DIVERGED_NANORINF when using HYPRE/BoomerAMG
did you use: -pc_gamg_agg_nsmooths 0
As your problems become more advection dominated you will probably want to use this. The default is 1.
and you can do this from the command line:
-mg_levels_ksp_type fgmres
-mg_levels_pc_type ilu
On Wed, Jul 1, 2020 at 9:22 AM Andrea Iob <andrea_iob at hotmail.com<mailto:andrea_iob at hotmail.com>> wrote:
Thanks for the answers.
The matrix contains both the advection and the diffusion terms of the Navier-Stokes equations. The case is subsonic (M=0.5), Reynolds number is in the order of 5000 so I expect the hyperbolic features to be the dominant ones. I understand this is not the type of problems AMG is designed for, but I was hoping to still be able to solve the system. This is the first time I use BoomerAMG and I'm not familar with all the options, however all the tests I've done gave me INFs/NANs.
You suggestion to use GAMG with custom smoothers seems to do the trick: after setting the GAMG smoother of each level to FGMRES/ILU, the solver converges in 93 iterations (overall it seems faster than ILU). Great! I've set the smoothers like this:
for (int i = 0; i < nLevels; ++i) {
KSP levelSmoother;
PCMGGetSmoother(preconditioner, i, &levelSmoother);
PC levelSmootherPreconditioner;
KSPGetPC(levelSmoother, &levelSmootherPreconditioner);
PCSetType(levelSmootherPreconditioner, PCILU);
KSPSetType(levelSmoother, KSPFGMRES);
}
Thank you very much!
Best regards,
Andrea
________________________________
From: Mark Adams <mfadams at lbl.gov<mailto:mfadams at lbl.gov>>
Sent: Wednesday, July 1, 2020 2:15 PM
To: Matthew Knepley <knepley at gmail.com<mailto:knepley at gmail.com>>
Cc: Andrea Iob <andrea_iob at hotmail.com<mailto:andrea_iob at hotmail.com>>; petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov> <petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov>>
Subject: Re: [petsc-users] DIVERGED_NANORINF when using HYPRE/BoomerAMG
As Matt said AMG does not work for everything out-of-the-box.
Hypre is pretty robust and I am surprised that every option gives INFs/NANs. But if it solves your stretched Lapacian then it's probably hooked up correctly.
You might try a non-adjoint Navier-Stokes problem, if you can.
You could try 'gamg' with the fgmres/ilu solver that works, as your smoother (use -ksp_view to check that you get everything set correctly). Use fgmres as the outer Krylov method and use -pc_gamg_agg_nsmooths 0.
Mark
On Tue, Jun 30, 2020 at 9:16 AM Matthew Knepley <knepley at gmail.com<mailto:knepley at gmail.com>> wrote:
On Tue, Jun 30, 2020 at 7:59 AM Andrea Iob <andrea_iob at hotmail.com<mailto:andrea_iob at hotmail.com>> wrote:
Hi,
I'm trying to solve a linear system using HYPRE/BoomerAMG as preconditioner. The system comes from a two-dimensional adjoint Navier-Stokes problem.
Do you expect the system you are solving to be elliptic? BoomerAMG is designed for elliptic systems, and can easily fail if applied
to a more general system, say one with zeros on the diagonal.
Thanks,
Matt
The mesh is structured (6400 cells) and there are 24 DoFs for each cell (the matrix has a total of 153600 rows). The condition number of the matrix should be in the order of 1e9. Using ILU preconditioner and FGMRES, the system is correctly solved (it takes 800 iterations to reach a residual of 1e-12). However, replacing ILU with HYPRE/BoomerAMG, the solver stops right after the first iteration (DIVERGED_NANORINF). I've tried different BoomerAMG options and different solvers (including Richardson to use BoomerAMG without a Krylov method), but I always get the same DIVERGED_NANORINF error.
Using the same HYPRE/BoomerAMG + FGMERS setup on another problem (Laplacian on a highly stretched grid), the solver reaches convergence very quickly without any problems.
At the bottom of this mail, you'll find the log of a run that stops with DIVERGED_NANORINF. Do you spot something in my setup that may explain why the solver is not converging?
Thanks. Best regards,
Andrea
-----------------------------------------
Assembly started...
Reading matrix file...
- Number of rows (N) = 153600
- Number of non-zero entries (NNZ) = 18247680
Assembly completed.
Time elapsed 83.3231s
Set up started...
Set up completed.
Time elapsed 6.62441s
Solution started...
0 KSP unpreconditioned resid norm 7.736650641501e-01 true resid norm 7.736650641501e-01 ||r(i)||/||b|| 1.000000000000e+00
0 KSP Residual norm 7.736650641501e-01 % max 1.000000000000e+00 min 1.000000000000e+00 max/min 1.000000000000e+00
[0]PETSC ERROR: --------------------- Error Message --------------------------------------------------------------
[0]PETSC ERROR:
[0]PETSC ERROR: KSPSolve has not converged due to Nan or Inf inner product
[0]PETSC ERROR: See http://www.mcs.anl.gov/petsc/documentation/faq.html for trouble shooting.
[0]PETSC ERROR: Petsc Release Version 3.10.3, unknown
[0]PETSC ERROR: bitpit_system_solver on a named xxx by xxx Tue Jun 30 09:42:09 2020
[0]PETSC ERROR: Configure options PETSC_ARCH=arch-linux2-c-opt-gcc7-hypre --with-blaslapack-dir=/opt/lapack/3.8.0-gcc.7.4.0/lib64 --with-debugging=0 COPTFLAGS=-O3 CXXOPTFLAGS=-O3 FOPTFLAGS=-O3 --with-valgrind=1 --with-valgrind-dir=/opt/valgrind/3.14.0/ --prefix=/opt/petsc/3.10.3_gcc-7.4.0 --with-mpi=1 --download-hypre
[0]PETSC ERROR: #1 KSPGMRESClassicalGramSchmidtOrthogonalization() line 67 in /root/InstallSources/petsc/src/ksp/ksp/impls/gmres/borthog2.c
[0]PETSC ERROR: #2 KSPFGMRESCycle() line 175 in /root/InstallSources/petsc/src/ksp/ksp/impls/gmres/fgmres/fgmres.c
[0]PETSC ERROR: #3 KSPSolve_FGMRES() line 291 in /root/InstallSources/petsc/src/ksp/ksp/impls/gmres/fgmres/fgmres.c
[0]PETSC ERROR: #4 KSPSolve() line 780 in /root/InstallSources/petsc/src/ksp/ksp/interface/itfunc.c
KSP Object: 1 MPI processes
type: fgmres
restart=45, using Classical (unmodified) Gram-Schmidt Orthogonalization with no iterative refinement
happy breakdown tolerance 1e-30
maximum iterations=10000, nonzero initial guess
tolerances: relative=1e-12, absolute=1e-50, divergence=10000.
right preconditioning
using UNPRECONDITIONED norm type for convergence test
PC Object: 1 MPI processes
type: hypre
HYPRE BoomerAMG preconditioning
Cycle type V
Maximum number of levels 25
Maximum number of iterations PER hypre call 1
Convergence tolerance PER hypre call 0.
Threshold for strong coupling 0.7
Interpolation truncation factor 0.3
Interpolation: max elements per row 2
Number of levels of aggressive coarsening 4
Number of paths for aggressive coarsening 5
Maximum row sums 0.9
Sweeps down 1
Sweeps up 1
Sweeps on coarse 1
Relax down sequential-Gauss-Seidel
Relax up sequential-Gauss-Seidel
Relax on coarse Gaussian-elimination
Relax weight (all) 1.
Outer relax weight (all) 1.
Using CF-relaxation
Smooth type Euclid
Smooth num levels 25
Euclid ILU(k) levels 0
Euclid ILU(k) drop tolerance 0.
Euclid ILU use Block-Jacobi? 1
Measure type local
Coarsen type HMIS
Interpolation type ext+i
linear system matrix = precond matrix:
Mat Object: Mat_0x1e88040_0 1 MPI processes
type: seqaij
rows=153600, cols=153600, bs=24
total: nonzeros=18247680, allocated nonzeros=18247680
total number of mallocs used during MatSetValues calls =0
using I-node routines: found 32000 nodes, limit used is 5
Solution completed.
Time elapsed 6.93336s
Solution information...
Error (L2) : 32.8359
Error (Linf) : 1.93278
************************************************************************************************************************
*** WIDEN YOUR WINDOW TO 120 CHARACTERS. Use 'enscript -r -fCourier9' to print this document ***
************************************************************************************************************************
---------------------------------------------- PETSc Performance Summary: ----------------------------------------------
bitpit_system_solver on a named xxx with 1 processor, by xxx Tue Jun 30 09:43:46 2020
Using Petsc Release Version 3.10.3, unknown
Max Max/Min Avg Total
Time (sec): 9.718e+01 1.000 9.718e+01
Objects: 5.300e+01 1.000 5.300e+01
Flop: 1.107e+08 1.000 1.107e+08 1.107e+08
Flop/sec: 1.139e+06 1.000 1.139e+06 1.139e+06
MPI Messages: 0.000e+00 0.000 0.000e+00 0.000e+00
MPI Message Lengths: 0.000e+00 0.000 0.000e+00 0.000e+00
MPI Reductions: 0.000e+00 0.000
Flop counting convention: 1 flop = 1 real number operation of type (multiply/divide/add/subtract)
e.g., VecAXPY() for real vectors of length N --> 2N flop
and VecAXPY() for complex vectors of length N --> 8N flop
Summary of Stages: ----- Time ------ ----- Flop ------ --- Messages --- -- Message Lengths -- -- Reductions --
Avg %Total Avg %Total Count %Total Avg %Total Count %Total
0: Main Stage: 9.7178e+01 100.0% 1.1071e+08 100.0% 0.000e+00 0.0% 0.000e+00 0.0% 0.000e+00 0.0%
------------------------------------------------------------------------------------------------------------------------
See the 'Profiling' chapter of the users' manual for details on interpreting output.
Phase summary info:
Count: number of times phase was executed
Time and Flop: Max - maximum over all processors
Ratio - ratio of maximum to minimum over all processors
Mess: number of messages sent
AvgLen: average message length (bytes)
Reduct: number of global reductions
Global: entire computation
Stage: stages of a computation. Set stages with PetscLogStagePush() and PetscLogStagePop().
%T - percent time in this phase %F - percent flop in this phase
%M - percent messages in this phase %L - percent message lengths in this phase
%R - percent reductions in this phase
Total Mflop/s: 10e-6 * (sum of flop over all processors)/(max time over all processors)
------------------------------------------------------------------------------------------------------------------------
Event Count Time (sec) Flop --- Global --- --- Stage ---- Total
Max Ratio Max Ratio Max Ratio Mess AvgLen Reduct %T %F %M %L %R %T %F %M %L %R Mflop/s
------------------------------------------------------------------------------------------------------------------------
--- Event Stage 0: Main Stage
BuildTwoSidedF 2 1.0 1.8471e-04 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 0 0 0 0 0 0
MatMult 3 1.0 6.8688e-02 1.0 1.09e+08 1.0 0.0e+00 0.0e+00 0.0e+00 0 98 0 0 0 0 98 0 0 0 1587
MatConvert 2 1.0 3.8534e-01 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 0 0 0 0 0 0
MatAssemblyBegin 3 1.0 2.8112e-04 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 0 0 0 0 0 0
MatAssemblyEnd 3 1.0 7.8563e-02 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 0 0 0 0 0 0
MatGetRowIJ 2 1.0 3.6100e-06 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 0 0 0 0 0 0
MatView 2 1.0 5.4618e+01 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 56 0 0 0 0 56 0 0 0 0 0
VecView 2 1.0 7.6981e-01 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 1 0 0 0 0 1 0 0 0 0 0
VecMDot 1 1.0 1.7635e-04 1.0 3.07e+05 1.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 0 0 0 0 0 1742
VecNorm 3 1.0 5.4832e-04 1.0 9.22e+05 1.0 0.0e+00 0.0e+00 0.0e+00 0 1 0 0 0 0 1 0 0 0 1681
VecScale 1 1.0 1.5216e-04 1.0 1.54e+05 1.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 0 0 0 0 0 1009
VecCopy 1 1.0 1.5406e-04 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 0 0 0 0 0 0
VecSet 21 1.0 7.1106e-03 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 0 0 0 0 0 0
VecAYPX 1 1.0 1.9940e-04 1.0 1.54e+05 1.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 0 0 0 0 0 770
VecWAXPY 1 1.0 7.8802e-04 1.0 1.54e+05 1.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 0 0 0 0 0 195
KSPSetUp 2 1.0 6.8995e-03 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 0 0 0 0 0 0
PCSetUp 2 1.0 1.3190e+01 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 14 0 0 0 0 14 0 0 0 0 0
PCApply 1 1.0 2.2728e-01 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00 0 0 0 0 0 0 0 0 0 0 0
------------------------------------------------------------------------------------------------------------------------
Memory usage is given in bytes:
Object Type Creations Destructions Memory Descendants' Mem.
Reports information only for process 0.
--- Event Stage 0: Main Stage
Matrix 3 1 2872 0.
Vector 32 20 17234800 0.
Index Set 4 4 3200 0.
IS L to G Mapping 2 0 0 0.
Vec Scatter 2 0 0 0.
Viewer 6 4 3392 0.
Krylov Solver 2 1 61676 0.
Preconditioner 2 1 1432 0.
========================================================================================================================
Average time to get PetscTime(): 3.36e-08
#PETSc Option Table entries:
-ksp_converged_reason
-ksp_error_if_not_converged
-ksp_monitor_singular_value
-ksp_monitor_true_residual
-log_view
-pc_hypre_boomeramg_agg_nl 4
-pc_hypre_boomeramg_agg_num_paths 5
-pc_hypre_boomeramg_coarsen_type HMIS
-pc_hypre_boomeramg_eu_bj true
-pc_hypre_boomeramg_interp_type ext+i
-pc_hypre_boomeramg_P_max 2
-pc_hypre_boomeramg_relax_type_all sequential-Gauss-Seidel
-pc_hypre_boomeramg_smooth_type Euclid
-pc_hypre_boomeramg_strong_threshold 0.7
-pc_hypre_boomeramg_truncfactor 0.3
#End of PETSc Option Table entries
Compiled without FORTRAN kernels
Compiled with full precision matrices (default)
sizeof(short) 2 sizeof(int) 4 sizeof(long) 8 sizeof(void*) 8 sizeof(PetscScalar) 8 sizeof(PetscInt) 4
Configure options: PETSC_ARCH=arch-linux2-c-opt-gcc7-hypre --with-blaslapack-dir=/opt/lapack/3.8.0-gcc.7.4.0/lib64 --with-debugging=0 COPTFLAGS=-O3 CXXOPTFLAGS=-O3 FOPTFLAGS=-O3 --with-valgrind=1 --with-valgrind-dir=/opt/valgrind/3.14.0/ --prefix=/opt/petsc/3.10.3_gcc-7.4.0 --with-mpi=1 --download-hypre
-----------------------------------------
Libraries compiled on 2020-06-26 12:35:46 on xxx
Machine characteristics: Linux-3.10.0-1127.8.2.el7.x86_64-x86_64-with-centos-7.6.1810-Core
Using PETSc directory: /opt/petsc/3.10.3_gcc-7.4.0
Using PETSc arch:
-----------------------------------------
Using C compiler: mpicc -fPIC -Wall -Wwrite-strings -Wno-strict-aliasing -Wno-unknown-pragmas -fstack-protector -fvisibility=hidden -O3
Using Fortran compiler: mpif90 -fPIC -Wall -ffree-line-length-0 -Wno-unused-dummy-argument -O3
-----------------------------------------
Using include paths: -I/opt/petsc/3.10.3_gcc-7.4.0/include -I/opt/valgrind/3.14.0/include
-----------------------------------------
Using C linker: mpicc
Using Fortran linker: mpif90
Using libraries: -Wl,-rpath,/opt/petsc/3.10.3_gcc-7.4.0/lib -L/opt/petsc/3.10.3_gcc-7.4.0/lib -lpetsc -Wl,-rpath,/opt/petsc/3.10.3_gcc-7.4.0/lib -L/opt/petsc/3.10.3_gcc-7.4.0/lib -Wl,-rpath,/opt/lapack/3.8.0-gcc.7.4.0/lib64 -L/opt/lapack/3.8.0-gcc.7.4.0/lib64 -Wl,-rpath,/opt/openmpi4.0.0-gcc7.4.0_slurm/lib -L/opt/openmpi4.0.0-gcc7.4.0_slurm/lib -Wl,-rpath,/opt/gnu/7.4/lib/gcc/x86_64-pc-linux-gnu/7.4.0 -L/opt/gnu/7.4/lib/gcc/x86_64-pc-linux-gnu/7.4.0 -Wl,-rpath,/opt/gnu/7.4/lib/gcc -L/opt/gnu/7.4/lib/gcc -Wl,-rpath,/opt/gnu/7.4/lib64 -L/opt/gnu/7.4/lib64 -Wl,-rpath,/opt/gnu/7.4/lib -L/opt/gnu/7.4/lib -lHYPRE -llapack -lblas -lm -lX11 -lstdc++ -ldl -lmpi_usempif08 -lmpi_usempi_ignore_tkr -lmpi_mpifh -lmpi -lgfortran -lm -lgfortran -lm -lgcc_s -lquadmath -lpthread -lstdc++ -ldl
-----------------------------------------
--
What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
-- Norbert Wiener
https://www.cse.buffalo.edu/~knepley/<http://www.cse.buffalo.edu/~knepley/>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20200701/2b235118/attachment-0001.html>
More information about the petsc-users
mailing list