The ABINIT Users Mailing List ( CLOSED )

[<6671011@163.com>]

Dear friends,

Recently, I found my d/dk calculation always failed with the following error:
     [cli_5]: aborting job:
     Fatal error in MPI_Comm_free: Invalid communicator, error stack:
     MPI_Comm_free(138): MPI_Comm_free(comm=0x2f2bad0) failed
     MPI_Comm_free(82).: Null communicator

At first, I guessed that it came from MPI. Then I reinstalled and upgraded the
MPI, nevertheless, nothing changed. By chance, I ran the jobs with fewer CPUs
and it finally worked. After further tests, I confirm that the d/dk 
calculation
can not work on too many CPUs, and the irreducible k-points play an important
role in this issue.

You can easily repeat this process by changing the k-point parameter in
trf1_5.in (test in tutorespfn) to ngkpt 2 2 2 and run it with more than 3 
CPUs.

However, I have not got the exact relationship between the irreducible 
k-points
and the maximum CPUs that can be used. I appreciate any kind of comment.

Sincerely,
Guangfu Luo


PS: There is no such problem in the ordinary SCF energy calculation; I use
Intel mpi (mpich2).