Transient explicit bug in parallel

[arktik]

Dear FEAP Admin,

There seems to be a problem in the parallel implementation of transient explicit (Newmark). I have demonstrated the problem with the attached MWE. Serial trans,expl solution of the feap's library element and my user element works alright and yield the expected results. Parallel solution gives wrong results!

Furthermore, parallel solution does not accept MASS,LUMP and FORM,ACCE. It throws the following error
*ERROR* DSOL: No lumped mass matrix available

Note: Parallel solution with implicit transient algorithm trans,newm yield correct results (except slow convergence problem posted at http://feap.berkeley.edu/forum/index.php?topic=2451.0)

Your suggestions on how to solve this problem are highly appreciated.

EDIT: Tested feap version is 8.6.1h

[Prof. S. Govindjee]

I've fixed a bug in the parallel lumped mass.  Can you give your test problem a check again.

[arktik]

Can you please tell me where can I find the fix?

Are these the files quoted below?

It is going to take me a few more days probably to get the next bug-fix release out.  In the meantime, you can try these files to see if they fix the problem for you. pform.f and pamcr7.F go in parfeap.  pmacr1.f goes in program.

[Prof. S. Govindjee]

You will need to download  the latest release from the download site.  The holder of the license you are using should be able to do this for you.
The quoted files are for a different bug fix.  The lump mass bug is in uasblem.F.  Best is to get the complete update.

[arktik]

I have tested the new update 8.6.1j for parallel solution of transient explicit. Unfortunately, the problem still exists. I get the same results as reported above.

Using mass lump also throws the same error.

[Prof. S. Govindjee]

please post the error you are now getting from MASS,LUMP.  Best is if you can post a test case that only uses FEAP's built-in features.

[arktik]

I have attached the test case again (slightly simplified from the first MWE above). Actually MASS,LUMP alone does not throw an error. However, together with FORM,ACCE the following error is displayed

Code: [Select]

*ERROR* DSOL: No lumped mass matrix available
The parallel solution in any case does not yield the expected solution. The attached MWE is a simple transient test case of homogeneous strain over a cube with resulting stress=1800. The serial explicit and implicit as well as parallel implicit solutions yield the exact results.

[Prof. S. Govindjee]

Good that MASS,LUMP is now ok.  I will have a look at the explicit issues.

[Alan]

Dear arktik,

I encountered the same problem with you, could you give me some hints about how to fix this problem, I will very appreciate with your help.

Thanks in advance.

[Alan]

Dear Prof. S. Govindjee,

I have a question for this code:

Code: [Select]

VecSetValue( Mdiag, i, p(n), ADD_VALUES, ierr)
I know that the parallel feap fills Mdiag with p(n), but now, I want to print out Mdiag vector for every processor, how can I do that? Because as I test, the Mdiag is a Vec type which

is not a standard fortran type and is more like a user defined type, when I use the open(...),write(...) same as what I always do for fortran programs, but the Mdiag can't be

printed out successfully. I need your help really.

Thanks in advance.

[Prof. S. Govindjee]

Yes, Mdiag is of type Vec which is a PETSc type.  You can print this out using VecView.

I believe, if you type  PETSC,VIEW at the Macro prompt and then form the mass, then it shoul print the mass out to a file, mass.m.  You should then issue the command PETSC,NOVIEW to turn off the writing.  I have not tried this in a while but the coding looks to be in place in parfeap/parform.F which is called out of program/formfe.f.

The viewer type is set to PETSC_VIEWER_ASCII_MATLAB in parform.F; you can change it if you want a different format.

[arktik]

Dear arktik,

I encountered the same problem with you, could you give me some hints about how to fix this problem, I will very appreciate with your help.

Thanks in advance.

Alan, the problem with transient explicit solution in parallel is still unresolved. I think a solution is currently a work in progress (see Prof. Govindjee's reply above). In the mean time, I can suggest to use implicit solver (if applicable to your problem) with parallel solution. I refer to version 8.6.1j.

[Prof. S. Govindjee]

The problem with parallel explicit solutions has been fixed and will be available in version 8.7, which will be released early next year.