November 2022 - petsc-users - mailman3.cels.anl.gov

Re: [petsc-users] On PCFIELDSPLIT and its implementation
by Edoardo alinovi 14 Nov '22

14 Nov '22

Hello guys, I am getting this error while using fieldsplit: [3]PETSC ERROR: --------------------- Error Message -------------------------------------------------------------- *[3]PETSC ERROR: Nonconforming object sizes[3]PETSC ERROR: Local column sizes 6132 do not add up to total number of columns 9200* [3]PETSC ERROR: See https://petsc.org/release/faq/ for trouble shooting. [3]PETSC ERROR: Petsc Development GIT revision: v3.18.1-191-g32ed6ae2ff2 GIT Date: 2022-11-08 12:22:17 -0500 [3]PETSC ERROR: flubio_coupled on a gnu named alienware by edo Wed Nov 9 08:16:29 2022 [3]PETSC ERROR: Configure options PETSC_ARCH=gnu FOPTFLAGS=-O3 COPTFLAGS=-O3 CXXOPTFLAGS=-O3 -with-debugging=no -download-fblaslapack=1 -download-superlu_dist -download-mumps -download-hypre -download-metis -download-parmetis -download-scalapack -download-ml -download-slepc -download-hpddm -download-cmake -with-mpi-dir=/home/edo/software/openmpi-4.1.1/build/ [3]PETSC ERROR: #1 MatCreateSubMatrix_MPIBAIJ_Private() at /home/edo/software/petsc/src/mat/impls/baij/mpi/mpibaij.c:1987 [3]PETSC ERROR: #2 MatCreateSubMatrix_MPIBAIJ() at /home/edo/software/petsc/src/mat/impls/baij/mpi/mpibaij.c:1911 [3]PETSC ERROR: #3 MatCreateSubMatrix() at /home/edo/software/petsc/src/mat/interface/matrix.c:8340 [3]PETSC ERROR: #4 PCSetUp_FieldSplit() at /home/edo/software/petsc/src/ksp/pc/impls/fieldsplit/fieldsplit.c:657 [3]PETSC ERROR: #5 PCSetUp() at /home/edo/software/petsc/src/ksp/pc/interface/precon.c:994 [3]PETSC ERROR: #6 KSPSetUp() at /home/edo/software/petsc/src/ksp/ksp/interface/itfunc.c:406 [3]PETSC ERROR: #7 KSPSolve_Private() at /home/edo/software/petsc/src/ksp/ksp/interface/itfunc.c:825 [3]PETSC ERROR: #8 KSPSolve() at /home/edo/software/petsc/src/ksp/ksp/interface/itfunc.c:1071 Do you have any ideas? Probably something missing in my brief implementation here: * call PCSetType(mypc, PCFIELDSPLIT, ierr) call PCFieldSplitSetBlockSize(mypc, 4-bdim, ierr) * * !2D, 3x3 block if(bdim==1) then ufields(1) = 0 ufields(2) = 1 pfields(1) = 2 call PCFieldSplitSetFields(mypc, "u", 2, ufields, ufields, ierr) call PCFieldSplitSetFields(mypc, "p", 1, pfields, pfields, ierr) ! 3D 4x4 block else ufields(1) = 0 ufields(2) = 1 ufields(3) = 2 pfields(1) = 3 call PCFieldSplitSetFields(mypc, "u", 3, ufields, ufields, ierr) call PCFieldSplitSetFields(mypc, "p", 1, pfields, pfields, ierr) endif ! Field split type ADDITIVE, MULTIPLICATIVE (default), SYMMETRIC_MULTIPLICATIVE, SPECIAL, SCHUR call PCFieldSplitSetType(mypc, PC_COMPOSITE_SCHUR, ierr)* Thanks for the help!

3 24

Use Vectors and Scalars in MatNest matrices
by Thomas Hisch 14 Nov '22

14 Nov '22

Hi! I would like to ask if it is possible to create a MatNest matrix that contains matrices, vectors and scalars. My (jacobian) matrix looks like the following [[A, B, a, b], [C, D, c, d], [e, f, g, h]] where A-D are sparse square matrices, a-d are dense column vectors, e, f are sparse row vectors and g,h are just scalars. Is there also a petsc4py example that shows how nest objects should be created and how they should be used? Best regards Thomas

3 3

Report Bug TaoALMM class
by Stephan Köhler 11 Nov '22

11 Nov '22

Dear PETSc/Tao team, it seems to be that there is a bug in the TaoALMM class: In the methods TaoALMMSubsolverObjective_Private and TaoALMMSubsolverObjectiveAndGradient_Private the vector where the function value for the augmented Lagrangian is evaluate is copied into the current solution, see, e.g., https://petsc.org/release/src/tao/constrained/impls/almm/almm.c.html line 672 or 682. This causes subsolver routine to not converge if the line search for the subsolver rejects the step length 1. for some update. In detail: Suppose the current iterate is xk and the current update is dxk. The line search evaluates the augmented Lagrangian now at (xk + dxk). This causes that the value (xk + dxk) is copied in the current solution. If the point (xk + dxk) is rejected, the line search should try the point (xk + alpha * dxk), where alpha < 1. But due to the copying, what happens is that the point ((xk + dxk) + alpha * dxk) is evaluated, see, e.g., https://petsc.org/release/src/tao/linesearch/impls/armijo/armijo.c.html line 191. Best regards Stephan Köhler -- Stephan Köhler TU Bergakademie Freiberg Institut für numerische Mathematik und Optimierung Akademiestraße 6 09599 Freiberg Gebäudeteil Mittelbau, Zimmer 2.07 Telefon: +49 (0)3731 39-3173 (Büro)

2 8

Use of LSC preconditioning
by Alexander Lindsay 11 Nov '22

11 Nov '22

Under what conditions can I use LSC preconditioning for field split problems with Schur? Let's imagine that all I've done is called SNESetJacobian with global A and P and provided the index sets for 0 and 1. Based off of the documentation on the man page https://petsc.org/release/docs/manualpages/PC/PCLSC/ it seems like I'd have to do something more programmatically, e.g. PetscObjectCompose((PetscObject)Sp,"LSC_L",(PetscObject)L); PetscObjectCompose((PetscObject)Sp,"LSC_Lp",(PetscObject)Lp); If I try to naively get LSC preconditioning from the command line without doing something like the above, then I get SUBPC_ERROR. Is this to be expected? 0 Nonlinear |R| = 4.164062e-02 0 Linear |R| = 4.164062e-02 Linear solve did not converge due to DIVERGED_PC_FAILED iterations 0 PC failed due to SUBPC_ERROR

2 2

Re: [petsc-users] On PCFIELDSPLIT and its implementation
by Edoardo alinovi 11 Nov '22

11 Nov '22

Hi guys, anyone with any suggestion to make this thing work? 🥺🙏 Il Ven 11 Nov 2022, 10:24 Edoardo alinovi <edoardo.alinovi(a)gmail.com> ha scritto: > Hi Barry, > > FYI, in test.F90 I noted that "ui" starts for 1 and not from 0. I fixed it > but the situation does not change much. I attached the new file in this > email. > > Thanks for the support, I am really walking in the dark! > > Il giorno gio 10 nov 2022 alle ore 23:28 Edoardo alinovi < > edoardo.alinovi(a)gmail.com> ha scritto: > >> Ah I see you have already added the missing interfaces for fortran >> enthusiasts :) So you likely do not need Matt's hack! >> >> [image: image.png] >> >

1 0

Get solution and rhs in the ts monitor
by Tang, Qi 10 Nov '22

10 Nov '22

Hi, How could I get rhs and solution in a ksp solve of ts? I am testing a linear problem (TS_Linear) using a bdf integrator. I tried to get the operator, rhs, and solution in the ts monitor through TSGetKSP and KSPGet***. But r = Ax-b is much larger than the ksp norm. I know the solver works fine. Did I misunderstand something about how TS works here? Perhaps one of the vectors is changed after the ksp solve? If so, is there a simple way to get rhs and solution that ksp of ts solved? Thanks, Qi

3 4

Local columns of A10 do not equal local rows of A00
by Alexander Lindsay 09 Nov '22

09 Nov '22

I have errors on quite a few (but not all) processes of the like [1]PETSC ERROR: --------------------- Error Message -------------------------------------------------------------- [1]PETSC ERROR: Nonconforming object sizes [1]PETSC ERROR: Local columns of A10 4137 do not equal local rows of A00 4129 when performing field splits. We (MOOSE) have some code for identifying the index sets for each split. However, the code was written by some authors who are no longer with us. Normally I would chase this down in a debugger, but this error only seems to crop up for pretty complex and large meshes. If anyone has an idea for what we might be doing wrong, that might help me chase this down faster. I guess intuitively I'm pretty perplexed that we could get ourselves into this pickle as it almost appears that we have two different local dof index counts for a given block (0 in this case). More background, if helpful, can be found in https://github.com/idaholab/moose/issues/22359 as well as https://github.com/idaholab/moose/discussions/22468. I should note that we are currently running with 3.16.6 as our PETSc submodule hash (we are talking about updating to 3.18 soon).

2 10

TSBEULER vs TSPSEUDO
by Francesc Levrero-Florencio 08 Nov '22

08 Nov '22

Hi PETSc people, We are running highly nonlinear quasi-static (steady-state) mechanical finite element problems with PETSc, currently using TSBEULER and the basic time adapt scheme. What we do in order to tackle these nonlinear problems is to parametrize the applied loads with the time in the TS and apply them incrementally. While this usually works well, we have seen instances in which the adaptor would reject the time step according to the calculated truncation errors, even if the SNES converges in a small number of iterations. Another issue that we have recently observed is that in a sequence of converged time steps the adaptor decides to start cutting the time step to smaller and smaller values using the low clip default value of TSAdaptGetClip (again because the truncation errors are high enough). What can we do in order to avoid these issues? The first one is avoided by using TSAdaptSetAlwaysAccept, but the latter remains. We have tried setting the low clip value to its maximum accepted value of 1, but then the time increment does not increase even if the SNES always converges in 3 or 4 iterations. Maybe a solution is to increase the tolerances of the TSAdapt? Another potential solution we have recently tried in order to tackle these issues is using TSPSEUDO (and deparametrizing the applied loads), but generally find that it takes a much longer time to reach an acceptable solution compared with TSBEULER. We have mostly used the default KSPONLY option, but we'd like to explore TSPSEUDO with NEWTONLS. A first question would be: what happens if the SNES fails to converge, does the solution get updated somehow in the corresponding time step? We have performed a few tests with TSPSEUDO and NEWTONLS, setting the maximum number of SNES iterations to a relatively low number (e.g. 5), and then always setting the SNES as converged in the poststage function, and found that it performs reasonably well, at least better than with the default KSPONLY (does this make any sense?). Thanks a lot! Regards, Francesc.

2 3

On PCFIELDSPLIT and its implementation
by Edoardo alinovi 08 Nov '22

08 Nov '22

Hello Guys, Thanks to your suggestions on the block matrices, my fully coupled solver is proceeding very well! I am now about to take advantage of the block structure of the matrix using PCFIELDSPLIT. I have learned a bit from the user manual and followed with interest this discussion in the mailing list: https://lists.mcs.anl.gov/pipermail/petsc-users/2015-February/024154.html which is actually the exact same situation I am in, so I guess most of the command line options will be copy and paste. I would like however to code them in fortran, as I usually provide some default implementation alongside the command line options. While coding some of the options I got an error here in PCFieldSplitSetFields() which looks to be undefined. I am importing petscksp, do I need to import something else maybe? Thank you!

2 1

Re: [petsc-users] [petsc-maint] Issues linking petsc header files and lib from FORTRAN codes
by Barry Smith 08 Nov '22

08 Nov '22

Please send your attempted makefile and we'll see if we can get it working. I am not sure if we can organize the include files as Fortran compiler include files easily. We've always used the preprocessor approach. The Intel compiler docs indicate the procedure for finding the Fortran compiler include files https://www.intel.com/content/www/us/en/develop/documentation/fortran-compi… is the same as for the preprocessor include files so I don't understand how the using the Fortran compiler include file approach would make the makefiles any simpler for users? Barry > On Nov 3, 2022, at 8:58 PM, Jianbo Long <longtuteng249(a)gmail.com> wrote: > > Hello, > > I'm struggling to make my FORTRAN code work with petsc as I cannot link the required header files (e.g., petscksp.h) and compiled library files to my FORTRAN code. > > Compiling petsc was not a problem. However, even with the fortran examples (see those on https://petsc.org/main/docs/manual/fortran/) and the guide on using petsc in c++ and fortran codes (see Section "Writing C/C++ or Fortran Applications" at https://petsc.org/main/docs/manual/getting_started/) I still cannot make my FORTRAN code work. > > The Fortran test code is exactly the example code ex83f.F90 (see attached files). Aftering following the 2nd method in the Guide (see the picture below), I still get errors: > > petsc/finclude/petscksp.h: No such file or directory > > Even if I set up the path of the header file correctly in my own makefile without using environment variables, I still can only find the file "petscksp.h" for my code. Of course, the trouble is that all other headers files required by KSP are recursively included in this petscksp.h file, and I have no way to link them together for my Fortran code. > > So, here are my questions: > 1) in the Guide, how exactly are we supposed to set up the environment variables PETSC_DIR and PETSC_ARCH ? More details and examples would be extremely helpful ! > 2) Is there a way to get rid of the preprocessor statement > #include <petsc/finclude/petscvec.h> > when using c++/Fortran codes ? > > For example, when using MUMPS package in a Fortran code, we can simply use compiler 'include', rather than a preprocessor, to extract all required variables for the user's codes : > INCLUDE 'zmumps_struc.h' > where the header file zmumps_struc.h is already provided in the package. Similarly, I think it's much more portable and easier when using petsc in other codes if we can make it work to use petsc. > > (Note: similar issues were discussed before, see https://lists.mcs.anl.gov/mailman/htdig/petsc-users/2019-January/037499.html. Unfortunately, I have no clue about the solution archived there ...) > > Any thoughts and solutions would be much appreciated ! > > Thanks, > Jianbo Long > > <image.png> > <ex83f.F90>

4 10