Bug reporting: phonon parallelization in Si

[sitangshu]

Dear Sir,

There is a bug in the ground state script of Si during phonon dynamical matrix and dispersion calculation. The script runs fine when only one processor is used. But for more than one, the log shows error and stops.
:
iter # 3 total cpu time : 10.8 secs av.it.: 10.0
thresh= 3.525E-05 alpha_mix = 1.000 |ddv_scf|^2 = 3.252E-10
kpoint 1 ibnd 5 solve_e: root not converged NaN
kpoint 1 ibnd 5 solve_e: root not converged NaN
kpoint 1 ibnd 5 solve_e: root not converged NaN
.
.
.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Error in routine broyden (1):
factorization
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

Can you please correct this...

With regards,

Sitangshu

[amolina]

Dear Sitangshu,

I think the QE input file shouldn't change if you use more processors. It could be that you set more processors than k-points and this is the reason of the problem? How many processors are you using?

Alejandro

[sitangshu]

Dear Alejandro,

Thanks for your reply. You are correct. This is happening because of the change in the number of processors while calculating relax & scf, and when calculating phonons.
I found that when I am running a relax file with 3 processors, initially it runs with all three of them, however after a while, I get this message (logs are attached):

CASE: enthalpy_new > enthalpy_old
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Error in routine bfgs (1):
step_old is NOT normalized
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
stopping ...

However, the code does not stops and the relax computation continues, but now with only one processor and ends without any error.

While doing for phonons, since I am using all three of the processors, the discontinuity occurs, which makes me understand that the relax/scf/nscf/bands parallelization may be corrected first.
Can you help me fixing the relax parallelization ?

With regards,
Sitangshu

[sitangshu]

Dear Alejandro,

Its look like am I have an issue with openmpi installation?
I am using mpirun.mpich command. Using only mpirun, the run doest not execute in my system.

Regards,
Sitangsh

[sitangshu]

Dear Sir,

I see now that the parallelization issue is now resolved!
Celebrating Happy Open MPI day!

Regards,
Sitangshu