Hello,
On Mon, 2012-03-26 at 17:05 +0000, Aronica, Pietro wrote:
> Hello,
> my bonds are specialised bonds which are supposed to mimic normal bonds. The atom types involved are artificial, and by using a forcemod file I'm able to give them wanted masses, charges and force constant for bonds, angles and dihedrals. In some cases, these parameters are identical to what they would be in a normal CT-CT bond, to make an example. In others, they are modified so that the bond length is shorter. Formally, it's still a CT-CT bond except its equilibrium bond length has been manually changed to something shorter.
> I wasn't very surprised when it failed to simulate at shorter distances, but my main question is whether I can control forces that crazy besides from changing the time step. I hoped that playing with different masses, charges and force constants could do it, but I would be willing to look at pmemd.CUDA code.
> Thanks again
> Pietro Aronica
> Imperial College
The equilibrium bond distances should have little effect on the
stability of your simulation, I would think. You don't have to worry
about VDW and EEL clashes between bonded pairs (since those interactions
are excluded from bond/angle partners).
It is definitely important, though, to do a minimization to relieve any
bad contacts and carefully heat your system. When you do a
minimization, sander/pmemd will print out the maximum gradients, which
could give you a clue about where the stresses on your system are
(visualization [visualisation?]) also helps.
Assuming that your starting geometry isn't very different from your
equilibrium values (which would cause large forces and integration
errors), the main thing that will impact your simulation is the
stiffness of the bond (the bond force constant). Stiff bonds have high
vibrational frequencies, which, unSHAKEn, can easily cause integration
errors.
Note that forces are unrelated to the time step (but the larger the
forces, the smaller the time step should be to ensure no integration
errors).
I will also echo Ross's advice that you try sander or pmemd first to
make sure it survives those tests first, before pushing on to
pmemd.cuda.
HTH,
Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Mar 26 2012 - 13:30:02 PDT
