the result depends of the number of processors?

[nany]

Hi all,

I have some problems with a calculation (the system is composed by 135 atoms):

1)I had different results if it was performed with 16 and 48 number of processors by using the same clusters and the VASP version (5.2.11): -1197,5578 and -1197,7931 eV, respectively.
In order to eliminate possible error sources, I performed this calculation in other clusters and different VASP versions, and the results (the new ones) were very similar (around -1197,78 eV) in all of them.

2)Someone suggested me that to fix, explicitly, the NPAR and NBANDS parameters, so, I added these to the previous INCAR. In this case, the result was the same using 16 and 48 processors (around -1197,55 eV). Then, it seems fixing NPAR and NBANDS parameters is the key to obtain the same results. But, the total energy (around -1197,55 eV) is not the same that I obtained in other clusters (around -1197,78 eV). There is a considerable difference between them, around 0.2 eV (-1197,55 vs -1197,78 eV).

3)Also, I considered a small system (54 atoms), but taking into account the same input files (except POSCAR file). Surprisingly, there was no any difference between the calculations obtained with 16 and 48 processors.

Could anyone tell me what is happening in my calculations? It is very bizarre.

Thanks to all

[admin]

2) please check the difference between the clusters (optimization level of the compilation, used libraries,...)
3) in this case, NBANDS most probably was the same in both cases, by chance