From: Dr. Anselm Horn <>
Date: Thu, 14 Jan 2021 13:55:59 +0100

Dear Syeda,

On 01/14/2021 12:14 PM, Syeda Amna Arshi wrote:
> Hello,
> I did the same simulation for three times with identical conditions but the
> resultant rmsd plots are quite different from each other. Is this normal??
> if yes is there any way to plot the average rmsd of these three simulations
> or we should take only one of them (but on what basis should we select the
> same)?

I'd like to take a step back for my comment:

MD simulations are in principle deterministic in the setup, i.e. the
same system with the same defined initial conditions should yield the
same trajectory (e.g. in a single-core or single-GPU simulation).
However: MD simulations may run differently altough one used "the same
input" or "the same setup":
- if you don't start with the same velocities for each particle but use
just the same temperature (> 0 K), only your velocity distribution is
the same, but the actual initial velocity may differ for the individual
atom, and thus the simulation evolves differently;
- if you use ASCII files for coordinate and velocity input, there may be
small round-off errors that sum up during the simulation and cause a
- if you distribute the simulation over several compute entities (e.g.
running parallel via MPI), you can have small round-off effects during
the distribution process, which itself is (to the best of my knowledge)
not deterministic, and thus you also end up with a different trajectory.
Thus, this is normal and expected behaviour.

On the other hand, if you intentionally run several trajectories of your
system with different initial setups (different temperature, different
ion distribution, different initial conformation), you obtain different
simulations, of course. This is often done to obtain a certain
statistical validation.

The easiest way to see that the systems evolved differently, is the use
of the RMSD, as you did: Identical simulations yield identical RMSD
plots. But you may also want to have a look at other properties of the
system, which are of interest for you.

Thus, the *interpretation* is the up to you (as Carlos already indicated):
Is the range of the property variation of expected size or not?
Is there any inherent problem (non-equilibrium, non-convergence, ...) in
your simulations? As Carlos pointed out, differing properties of
interest may indicate an underlying problem.

Having said this, it is difficult for us to decide, whether your 'quite
different rmsd plots' are problematic, since you are of course the
specialist for your system and also did the simulation setup.



AMBER mailing list
Received on Thu Jan 14 2021 - 05:00:02 PST
Custom Search