On Thu, Dec 01, 2005, Ilyas Yildirim wrote:
>
>
> 1. I am thinking that sander is reading the improper torsional
parameters
> given in frcmod file wrong.
Did you apply bugfix.53 to your Amber 8 runs? If not, this would explain
differences between Amber 8 and Amber 9 for your system. You should use
rdparm (or similar) to verify exactly which dihedrals are being used.
>
> 2. My second thought is that the SHAKE algorithm has a bug in it. But
> still this does not explain why I am ending up with snap01 and snap02
> structures (SHAKE is off).
There is a potential problem in using SHAKE with TI if the ideal bond
lengths
at equivalent positions are different in the two prmtop files. This would
mean that the coordinates are not really the same on the two cpus. We
need
to establish some checks for this. The sledgehammer approach is to not
allow
SHAKE to be on, but there is probably a better way (i.e. check for the
identity of the SHAKE lists on the two nodes.)
...dac
Received on Wed Apr 05 2006 - 23:49:47 PDT
