error arised from parallel calculation

[lanjh]

Dear Professor,
I want to consult you a question. I have just completed the compilation of parallel version of vasp. But I found the parallel computation brought wrong results. For example, the different cpu numbers will genera te different energy which is far from the right result. In addition, when I run the example of Si, in serial computation only 2 iteration times are carried out. But in parallel computation the more cpu numbers will lead to more iteration times. This is so ridiculous.
Can you tell me how to repair this error?
My library is as below,
GotoBLAS-1.15.tar.gz
mpich.1.2.5.2
vasp.4.lib
My version is vasp.4.6

Thanks again.

[graeme]

This was cross-posted to the vasp forum (which is more appropriate) and adressed by the admins there.