Different DOS result with MPI and sequential run
Posted: Sat Jan 10, 2015 9:17 am
Hi, I discover a peculiarity with DOS calculation in PAW method when I run the code in parallel.
For example I follow the PAW 1 tutorial part 4 (PAW contributions to the Density of States (DOS)). It is stated that for the used PAW dataset the smooth PW contribution and the PS on-site contribution are close.
But when I run the code in parallel this is not the case:
- the smooth PW contribution dominates
- the all-electron and PS on-site contributions cancel out each other appromixately
Furthermore the results varies when I tried using mpi with different numbers of cores.
I notice the correct results as shown in the tutorial is obtained when I run the code sequentially.
Is this the probem of the PAW DOS parallelization or my MPI configuration itself?
Thanks.
For example I follow the PAW 1 tutorial part 4 (PAW contributions to the Density of States (DOS)). It is stated that for the used PAW dataset the smooth PW contribution and the PS on-site contribution are close.
But when I run the code in parallel this is not the case:
- the smooth PW contribution dominates
- the all-electron and PS on-site contributions cancel out each other appromixately
Furthermore the results varies when I tried using mpi with different numbers of cores.
I notice the correct results as shown in the tutorial is obtained when I run the code sequentially.
Is this the probem of the PAW DOS parallelization or my MPI configuration itself?
Thanks.