Re: many short runs or one long run?

From: Marcos Sotomayor (sotomayo_at_ks.uiuc.edu)
Date: Mon Jul 10 2006 - 16:03:15 CDT

Dear Jas,

As far as I can tell, running many short or one long simulation shouldn't
have an important impact on the outcome of your simulations. Perhaps the
only two things that may change are the motion of the center of mass of
your system (which is eliminated by the option COMmotion no) and the
number and size of files you will have to handle (which may become a
serious practical issue when you deal with huge systems).

Trajectories (of every single atom) will never be "true" trajectories
of the system just because of unavoidable numerical errors (in any MD), as
nicely discussed in the book "Understanding Molecular Simulation, from
algorithms to applications" (Frenkel & Smit, pages 71 to 74). Moreover,
parallelization and the use of Langevin dynamics may introduce some
"extra noise". Therefore, reproducing the exact same trajectory is
difficult, if not impossible, for long runs (even in the NVE ensemble,
see the discussion about the Lyapunov instability in the same book).

Thus, you should focus on results that are reproducible in MD
simulations, like averages or macroscopic quantities. Repeating simulations
may also give you an idea of how robust your result is, but the main
point is whether the question you are asking makes sense in the context of
MD simulations or not. For instance, if you ask whether an ion will always
go through the exact same path (within 0.1 Angstroms) when crossing a wide
pore in a membrane and under the influence of an external field, you will
likely get very frustrated. While if you ask what kind of ions can go
through that pore (positive or negative) and what is the macroscopic
current, then you will likely get very consistent and robust results from
MD simulations. In short, if your result depends on the parameters used or
is not reproducible, likely is not a result or you are asking the wrong
question.

There are multiple examples of results reproduced by several other labs
among papers published in our group (www.ks.uiuc.edu).

Hope that helps,
Marcos

On Mon, 10 Jul 2006, Ching Wong wrote:

> Hi, all.
>
>> From your experience, do you think it is better to run many shorter run
> or just one longer run (such as four 2ns runs vs. one 8ns run)?
>
> And do you get the exact same result (trajectory, RMSD, RMSF, etc) from
> a repeated run (same seed)? Is the MD repeatable at all in NAMD?
> I thought I saw it somewhere in the manual that results may differ.
> Can someone confirm that, I just want to make sure. :)
>
> If the result is not reproducible, what would you say? How can we claim
> the MD simulation experiment is still reliable???? Any ideas please?
>
> Sincerely yours,
> Jas
>
>
>
>
>
> Quote from Manual
> seed random number seed
> Acceptable Values: positive integer
> Default Value: pseudo-random value based on current UNIX clock time
> Description: Number used to seed the random number generator if
> temperature or langevin is selected. This can be used so that
> consecutive simulations produce the same results. If no value is
> specified, NAMD will choose a pseudo-random value based on the current
> UNIX clock time. The random number seed will be output during the
> simulation startup so that its value is known and can be reused for
> subsequent simulations. Note that if Langevin dynamics are used in a
> parallel simulation (i.e., a simulation using more than one processor)
> even using the same seed will not guarantee reproducible results.
>
> -------
> Jasmine
>

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:43:49 CST