DMFT+DFT Compilation Failed

[franck ngassam]

Dear colleague;

We've been trying to compile your DMFTwDFT code and run for over 4 months now. We have compiled it by all possible methods. the Non-Selfconsistent calculation (Niter = 1) works very well. But the self-consistent full load calculation DFT + DMFT (Niter> 2) does not work. 

Are there people here who have compiled the latest version of the code on the site and run it with the ful self-consistent option (Niter> 1)? if so can you give me details of your supercomputer configurations i installed it with mpif90, mpic ++, mpicc with Intel20 + openmpi-4.0.i20

https://github.com/DMFTwDFT-project/DMFTwDFT

 

We will really appreciate your help. 

 

Franck O.

[Benny Wah]

Hi Franck,

Do you have error files?

Benny

--
You received this message because you are subscribed to the Google Groups "DMFTwDFT" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dmftwdft+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/dmftwdft/d2be530d-4d5c-42dc-a384-ca03539d1029n%40googlegroups.com.

[franck ngassam]

I have compiled the code with openmpi (mpif90, mpic ++, mpicc) vasp5.4.4 + wannier90v2 and we have the impression that it is the parallelization that is causing the problem. Because it only works if we remove in front of each executable the "para_com" in the RUNDMFT.py file as you will see in the attached file. otherwise, the vasp calculation is not correct in root / DMFT which produces bad fermi energy which in turn is used to generate the (false) energy window and therefore the wannier trash calculation

[Benny Wah]

Franck,

I am confused, do you assign para_com.dat as "mpirun -np ##"  and then the impurity solver? 

You can complete the run, it is just that parallel run is faulty?

To view this discussion on the web visit https://groups.google.com/d/msgid/dmftwdft/a6e11958-fd16-4efd-bda4-48567c446aa4n%40googlegroups.com.

[franck ngassam]

I have checked again the energy of fermi in the OUTCAR for the "de-parallelized" case of which I have although the calculation is finished correctly, the energy of fermi is around 21 eV which seems to me quite large and therefore not the true one. I think more that the problem is the parallelization but a bug in the code (for the installation with openmpi)