[gmx-users] stable trajectory?

[Ulrich Omasits]

Hi,
I'm simulating a protein-ligand system in a 20A-SPC waterbox using the 
gmx-ff and PME.
My problem:
The ligand should be high-affinity but in two out of five 
one-nanosecond-simulations the ligand detaches the protein!
In one simulation the ligand seems to be stable (RMSD & RMSF look nice) 
but then the same simulation performed again (only starting velocities 
differ) delivers totally different results (ligand even detaches!).
How can this happen!?! What can I do?
Should I change the forcefield, the water-type, the box-type, the 
PME-parameters,...?
Could these changes affect the trajectory such drastically?

Thanks
Ulrich