Reaction Forces in parallel FEAP

[tgross]

Hello!

In Serial FEAP (8.4) I normally use the option "REAC ALL" and "STREss ALL" in order to write all the nodal reaction forces and Element stresses and strains to the standard Output file (Oxxx).
Using parallel FEAP (8.4) (with metis; 2 processors) I used the same commands ("REAC ALL" and "STREss ALL") in the solve.XXX file to get the Nodal reaction forces, stresses and strains in the Oxxx_000X files.
The stress and strain quantities in the Oxxx_000X files are correct and similar to the results from Serial FEAP, however all the Nodal reaction forces are 0.

Can the REAC ALL command be used in parallel FEAP the same way as in Serial FEAP?
I have attached a minimal example (8 Element test for serial and parallel FEAP(2cpus)) where you can directly see the Problem. IcubeS is for Serial FEAP. Icube + solve.cube +part.cube for parallel FEAP.

Thank you very much!
Best regards,
Thomas

[FEAP_Admin]

Reactions in the parallel code are handled a bit differently since these will be part of parallel vector.  Notwithstanding
this should be easy to make work.  I'll put it on the to do list and let you know when it is available.

[tgross]

Great! Do you have an idea how long it will take to implement this? If we are talking about months I might have to look for an alternative way to compute the reaction forces.

Best
Thomas

[FEAP_Admin]

shouldn't take too long just need to wait for the semester to wind down.

[tgross]

Perfect! Thank you!

[tgross]

Interestingly, it seems that parallel FEAP 8.3 with 2 processors writes out the correct Reaction Forces to the Oxxx_000X. Where there any modifications in REAC ALL when changing from 8.3 to 8.4? Should REAC ALL give you the correct reaction forces in parallel FEAP 8.3? Are these values reliable?

Best regards,
Thomas

[Prof. S. Govindjee]

I think I have a fix for the problem.  Please replace your parfeap/pform.f (version 8.4) with the attached version.
I did some limited testing and it seems to be providing the correct information.  One annoying thing is that you have
to be careful with ghosted information.  For example, if you do plot,reac then you will see reactions on the partitioned faces
which is kind of annoying.  I'll try to tidy it up in the future but the attached routine should at least provide the correct values in the output file.  Please test it out and let me know, if there continue to be errors.

[tgross]

Dear Prof. Govindjee,

I tried your new pform.f and it works for me! I will do some additional testing and report if any problems should come up!

Thank you!
Thomas

[blackbird]

Dear Prof. Govindjee.

I am using parFEAP 8.3 and I also want to use reac-command. When compiling your pform.f in my computer there are 4 errors. It seems the variables nrvn,v_avg,v_c and v_rho dont have implicit type. Did I miss any additional file necessary to use the new pform.f?

Best regards
Christian

[FEAP_Admin]

The posted file is for version 8.4 not 8.3.

Please first verify that REAC is not giving the correct values in 8.3. 
If this is the case, then compare the posted version with yours and try to backport the changes.
The primary thing that needs to be done is under jsw.ne.14 to add calls to pparlo( ).
The main issue will be that some of they syntax and how things are computed has
changed between versions.

Upgrading your code is another option.