AMBER Archive (2003)

Subject: Re: results of MD comparison

From: Carlos Simmerling (carlos_at_csb.sunysb.edu)
Date: Wed Feb 26 2003 - 15:00:20 CST

how long does it take for the energies to
differ? Run a simulation on 1 and 4 cpus,
and write output each step (ntpr=1). Compare
the files and let us know. It can be that they
differ right away (a problem) or not until
hundreds of steps (maybe not a problem).
After some time the calculations are not the same
because the operations are done in a different
order and the small differences cause changes in
the trajectories. However, they may both be valid
trajectories, just as you would obtain different trajectories
if you ran with a different initial random number seed
for the velocities.
Carlos Simmerling

----- Original Message -----
From: "Xiaoqin Huang" <xiaoqin_at_csit.fsu.edu>
To: "amber-heimdal.compchem.ucsf.edu" <amber_at_heimdal.compchem.ucsf.edu>
Sent: Wednesday, February 26, 2003 10:08 AM
Subject: results of MD comparison

> Hello,
> I have a question about the parallel mode of amber, that is:
> when I compare the MD results (just 20 ps) of both the parallel mode (4
CPUs)
> and the single running, the RMSD data is different greatly, and also the
> total energy, the potential and kinetic energy of the whole system. it
seems
> these two runnings are on different directions, and the RMSD of parallel
mode
> is more fluctuating than the single one's.
> The solvated systems are the same (about 15000 atoms), and with the same
mdin
> conditions. how can I explain these results?, which one should I select?
> My computing environment is the PC cluster with Linux operation system and
> typical MPI.
> do you have any suggestions? thank you very much!
>
>
> Xiaoqin Huang
> School of Computation Science and Information Technology
> Florida State University
> Tallahassee, FL32306
> U. S. A.
> Tel: 850-644-5792(O)
> Fax: 850-644-0098
> E-mail: xiaoqin_at_csit.fsu.edu
>
>
>
>