sums in parFEAP 8.5.2i

[arktik]

sums command in parFEAP is not working correctly. For testing, I used a unit cube discretized with 1000 8-node brick elements, applied 0.01 tensile displacement to obtain homogeneous stress and strain. Material model is elasto-plastic with nonlinear isotropic hardening. I solve the nproc partitioned input files as

Code: [Select]

mpirun -n nproc $FEAPHOME8_5/parfeap/feap -ksp_type cg -pc_type jacobiwhere solve.filename contains.

Code: [Select]

batch
tplot,,1
petsc on
tol iter 1.d-7
prop,,1
dt,,1d-2
loop,,100
 time
  loop,,10
   tang,,1
  next
pview,time
next
end
sums,3,3,1.0,1.d-4
With paraview, I can see that stress/strain results in cube are matching for all nproc cases
However, when I collect all nproc *.sums files and post-process it (with matlab) to sum up the reactions in each time step, something goes wrong. Stress vs strain for each nproc case are attached. Can you please suggest what may be wrong? Are ghost nodes messing up the total reaction?

[Prof. R.L. Taylor]

My guess is that the ghost element reactions are included.  This is a bug.

[arktik]

Thank you Prof. Taylor for your reply. Any suggestion on fixing this?

[Prof. R.L. Taylor]

I will try to work on it, but am away from Berkeley for the summer and have very limited access to something to use for parfeap.

Basically, in pform, I think omitting the region 21 elements may do it (ix(nen1-1,n) = 21)- but only for dfl = .true. and there could be another check.  --- only modify the pform.f in the ./parfeap directory.

[Prof. R.L. Taylor]

One thing I noticed in your input file is the single file with both a TPLOt and a PROPld command.  These commands require inputs after the END matching the BATCh.  I would suggest splitting things in to multiple batch and check that all is really input correctly.

[arktik]

Thanks for your suggestion. Putting PROPld and TPLOt as separate BATCh commands did not help. In pform subroutine in /parfeap directory, I am not very sure where and how to omit the reactions of "region 21" elements from assembly. Therefore, I will wait for the solution to this bug.

As an alternative strategy, is it possible to output the nodal reactions with pview so that one can post process it in paraview?