From: Marcos Sotomayor (sotomayo_at_ks.uiuc.edu)
Date: Wed Mar 22 2006 - 15:53:16 CST
As Sterling mentioned, the firsttimstep parameter is just a label, only
useful for SMD simulations (as far as I know).
Note that if you are using Langevin dynamics (are you?), energies and
coordinates will never be the same since *random* forces are applied
(unless you use the same seed). If you are running in parallel, you may
also have roundoff errors, but that should be minor, noticeable only after
many time steps....
On Wed, 22 Mar 2006, Sterling Paramore wrote:
> The firsttimestep parameter is totally irrelevant for most things done in
> NAMD and does not affect subsequent calculation (SMD would be an exception).
> It's just a way to label the timesteps as you see appropriate. Usually,
> you'll want it to refer to the timestep of the restart file. In your
> example, you might want to set it to 89000, as your new trajectory is
> supposed to be a continuation of the equilibration.
> You're probably getting different energies and coordinates because NAMD
> trajectories over multiple processors are not reproducible. As I understand
> it, this is because with mutliple processors, numbers are added up in
> different orders depending on what processor gets what part of the
> simulation. This changes the roundoff error between trajectories which very
> quickly leads to divergence.
> Also, catdcd does not read the timesteps, only the frames.
> Himanshu Khandelia wrote:
>> I have been totally baffled with the firsttimestep parameter, and its
>> meaning. I ran an equiliration run for 89000 steps. I then started a
>> dynamics run, as a continuation of equilibration, varying the
>> firsttimestep parameter as follows:
>> Case 1: firsttimestep 0
>> Case 2: firsttimestep 89000
>> Case 3: firstimestep 89020 (20 was the stepspercycle parameter)
>> Case 4: I did not define the firsttimestep parameter
>> In EACH of the above cases, I get different energies and coordinates.
>> Which one is the correct value to use, and why ? Also, in each case,
>> "catdcd" merges the dcd files from equilibration and dynamics without any
>> HELP !
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:41:46 CST