The test of a 2D model used periodic boundary conditions

[Yulin Wan]

Dear all,

  When I impose the periodic boundary conditions on my model, the loading seems not to be applied. So I build a simple model to test, but it still doesn't work. These are my input files, the question I discussed is a 2D model, the third degree of freedom is the phase field variable.
  I am looking forward to your help.

Sincerely,
Yulin Wan

[Prof. R.L. Taylor]

What version of code are you using?  ver8.6?

One error in IZ1L4 is "elink" belongs after mesh END command.

Have you tested a smaller problem in the serial code  before doing parallel?

I cannot test since you use a user element.

[Prof. R.L. Taylor]

I attach a set of files that develop a set of parallel files that can run -- without your user elements so just mechanical.  But test everything to get it working for this.  The files are
1. Itest, INodetest which is run using the single solution command "OUTM".  This creates Itest.rev which is a FLAT file as required.  Look carefully at the file Itest to see where all the commands are located and the proper sizes -- of the coordinates.  This file is executed with the serial version.

2. Run Itest.rev  -- I had to edit to add the commands to partition the mesh.  This file is run with the parallel version of the code and will create two files Itest_0001 & Itest_0002.

3 Run the parallel program with the input files specified as Itest_0001 -- solve with whatever loading you want.  You may have to modify Itest to specify the proportional load you want, only one is allowed.

N.B. You must use version 8.6, earlier versions do not do links.

[Yulin Wan]

Thanks for your advice, my version is 8.5. So I can't use the periodic boundary conditions in my user element, am I right?

[Prof. S. Govindjee]

Periodic BC work in the serial version of FEAP8.5.  Starting with FEAP8.6 periodic boundary conditions are also available in parallel.  They are not available in parallel 8.5.

[epn]

Dear All,

I have a similar problem with FEAP 8.6. I have installed the parallel version using PETSC 3.17.4 and OpenMPI 4.1.2. It works very well with the problems where there is no periodic boundary conditions.

However, when I try to run the same problem with periodic boundary conditions, it gives the following error:

\[3]PETSC ERROR: #642 MatSetValues() at ~/petsc_3_17_4/src/mat/interface/matrix.c:1343
\[3]PETSC ERROR: #645 MatSetValues_MPIAIJ() at ~/petsc_3_17_4/src/mat/impls/aij/mpi/mpiaij.c:506
\[0]PETSC ERROR: #954 MatAssemblyEnd_MPIAIJ() at ~/petsc_3_17_4/src/mat/impls/aij/mpi/mpiaij.c:692

I have tested it for several meshes and BVPs, all give the same error. I am attaching a test problem, which works with a serial FEAP, but not with parallel.

Thank you in advance for you help!

[Prof. R.L. Taylor]

Did you try by using the "flat" file you created using "OUTM"?

[epn]

Dear Prof. Taylor,

thank you very much for the prompt response and your help.

 I tried the flat file created using "OUTM" in serial feap and it works. I also realized that "OUTM" command does not include "PERIodic HILL" boundary conditions in the flat file "*.rev". I needed to add this boundary condition manually.

Unfortunately, in parallel feap, it does not work. It gives the same errors.

Do you think that it could be associated with the PETSC version or how I compiled the parallel feap? I did not used the PETSC version mentioned in the manual because it was not getting installed completely in my system.

[Prof. R.L. Taylor]

I apologize.  While we did add "link" with the update to 8.6 the "periodic" part was not added.  Thus, at this time only the serial version works.  Possibly, if you have the Intel compiler you also have access to "pardiso".  Using a sparse solver may allow you to solve the problem then.  Or you can use "superlu" or "umfpack" -- you will find the necessary "umacro" modules at the web site for all these.

[epn]

Dear Prof. Taylor,

thank you for your response.

I might have generated a confusion with my previous comment. Although the "periodic" boundary conditions were not in the "flat" file, but it is included in the input files for different domain by feap automatically.

At the moment, I would like to use the "link" command with parallel feap, which appears to be the main problem for me. In my applications, I will not use "periodic hill" from feap and will have a user defined loading, element and material routines.

I was wondering if there is any working example with only linking command, so that I can test my version.

Thank you for your help  and sorry for any confusion.

[Prof. S. Govindjee]

I'll look for the test files we used in developing the parallel implementation and post some of them, but it may take a day or two, sorry about the delay.

[epn]

Dear Prof. Govindjee,

thank you very much!

[epn]

Dear Prof. Govindjee,

I would like to ask if you had time to take a look at if there is a possible input file with Link command.  Thanks.

[Prof. S. Govindjee]

Sorry for the delay.  I just got around to looking for the files.  I was not able to find them, they may be on an older computer.  Notwithstanding, I tried to create some test problems and they failed for me.  I will need to perform a bit more detective work to figure out what the problem is.

[ahmedashraf]

What version of code are you using?