AMBER Archive (2008)
Subject: Re: AMBER: restarting molecular dynamics
From: Germain Vallverdu (germain.vallverdu_at_lcp.u-psud.fr)
Date: Wed May 28 2008 - 09:25:23 CDT
> It is VERy unclear to me what you mean by "a simulation of about 10
> ps step by step with a restart every 5fs". Do you mean that you will
> stop every 5 fs, and restart again for another 5 fs and so on until
> you reach 10 ps ? Why would you do that ?
Yes I would like exactly do that. I would like to do a simulation with
variable charges. Then every 5fs I stop the run, compute (or not) new
charges and restart the run again.
> I also do not understand what you mean by "compare the one step
> run with the dashed run".
I have done one run with a length of 10 ps and one "dashed" run ( where
I stop and restart every 5 fs). For these tow runs I plot the
trajectories of a coordinate of interest and look for differences
between them. I see that the curves are similar during the first ps and
then are different. Nevertheless the mean and the standard deviations
> Regardless of that, you are basically right about what sander does
> when it restarts:
> "> I suppose that :
> > - when sander start it computes the PME grids
> > - if I use algorithm control of the temperature like Langevin or
> > Andersen perhaps there is a repeat of the random number chain
> because of
> > the seed is the same for every run.
> > - the accuracy of the coordinate and velocities is better in the
> > than in restrt files. Then this might lead to difference between
> > trajectories."
> but you are using ntt=1 (Berendsen thermostat) so thgere are no random
> number sequences for the temperature control.
Yes that is exactly why I chose this one.
> Indeed, the precision of the restart file is different than the
> internal precision during calculations, and different from the
> dynamics coordinates in the mdcrd files.
Then do you think this is the only origins of differences between the
> So, please clarify what you want to do and we might be able to help you.
> Germain Vallverdu wrote:
>> Hello Amber
>> I would like to know exactly what append when I restart a molecular
>> dynamics with sander.
>> here is an example of my input file, I use AMBER 9 on a Linux system.
>> This is a simulation of an organic molecule in a TIP3P water box.
>> nstlim=50000, dt=0.001,
>> irest=1, ntx=5,
>> ntpr=100, ntwr=100, ntwx=0,
>> tempi=300.0, temp0=300.0, ntt=1,
>> ntb=2, pres0=1.0, ntp=1, taup=2.0,
>> ntc=2, ntf=2,
>> My problem is the following. I would like to do a simulation of about
>> 10 ps step by step with a restart every 5fs. But if I compare the one
>> step run with the dashed run, trajectory are not similar after few
>> steps. Then I would like to know what exactly append when Sander starts.
>> I suppose that :
>> - when sander start it computes the PME grids
>> - if I use algorithm control of the temperature like Langevin or
>> Andersen perhaps there is a repeat of the random number chain because
>> of the seed is the same for every run.
>> - the accuracy of the coordinate and velocities is better in the
>> program than in restrt files. Then this might lead to difference
>> between trajectories.
>> Nevertheless do you think that something like that is possible to do
>> and does it exist other process at the beginning of sander ?
>> Best wishes
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)