I'm still a rookie for ABINIT and I'm now working on band structure calculation of graphitic carbon nitride.
When I moved on to the step of the LDA band structure, I tried to use the parallel version to accelerate my calculation. Unfortunately I got a strange result compared with the one with the sequential version.
Here are the pictures of the two results.
- normal LDA band structure of g-C3N4
- the same input file but calculated with parallel version
You can find the discrepancy in the above picture. In the result of mpi version, the eigenvalues at Gamma point seems to be at unexpected places, while the ones at other k-points are almost the same with the first result. What may be the reason for this problem? If I can get the correct result with parallel version, my work will finish faster.
Thanks a lot for any advice.
For more detail information, I list them below. If you need more, just tell me.
Code: Select all
OS info: openSUSE 11.3 (x86_64)
Kernel: 2.6.34.7-0.7-default
ABINIT version: 6.4.2 (ps: I also test with the newest 6.6.1, the problem still exists.)
compiler: gfortran --version
GNU Fortran (SUSE Linux) 4.5.0 20100604
mpi connector: openmpi-1.4.3 & mpich2-1.3.1 (Both of them have the problem.)
Btw, I also attach the input file here.
Thanks & best regards,
Agnxy
