Re: spontaneous system explosion after 2 ns of normal run

From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Thu Oct 15 2009 - 11:15:23 CDT

On Thu, 2009-10-15 at 11:22 -0400, Roman Petrenko wrote:
> Guys, thanks for replies,
>
> 1) yes i wrap all
>
> 2) why 2 fs is an aggressive timestep. i thought the larger the system
> the more applicable is this timestep, because we don't care about high
> frequency motion.

first off, the physics of the system and the integrator don't
care about what you care about. the larger the system, the larger
the possible fluctuations (including velocities) within the system.

what i suspect is that you may just accidentally have had some
very large force(s) on some atom(s) and that then the integrator
broke down. since you increased the additional force due to the
external field by a factor of 10, that may have become a significant
contribution and just enough to bring the system over the top.

of course, this is just a speculation based on my experiences
with simulations on driving electrolytes through nanopores
with (strong) external fields some 10+ years back. i would
assume that a membrane protein system is much more sensitive
than the crude models that we were able to run in the (good?)
old times. ;-)

if you don't care about fast motions, you could fiddle with the
settings of the langevin thermostat, i.e. add more friction
to make fast movements slower and stabilize a larger time step.

> 3) center of mass drifts a _little bit_; is it that bad taking into
> account that box expands eventually to 5000 A and center of mass moved
> mere 170 A; here is center of mass of the system from complete
> trajectory:

as soon as the system expands significantly your results are bogus,
so it is useless to discuss the 170A (which would be horrible), but
the rest looks not too bad. when working with external fields, the
COM drift is always an iffy issue and has to be monitored carefully,
since there is always a chance that you become a victim of the
"flying icecube syndrom", i.e. the total system starts to drift,
and then due to the impact of the termostat your system effectively
cools down. this is, obviously, correlated with the strength of the
external field. in a "normal" system without an external field and
if it is reasonably well equilibrated system, the langevin thermostat
in NAMD should take care that the drift, once removed, does not
build up again.

[...]

> 4) I wonder why people who did DNA translocation don't reply. Am i the
> only one who set up simulation system incorrectly? :)

i don't think you are the only one and i believe this is a general
problem. i've been complaining about this on various mailing lists
and many times. my estimate is that on every person that asks a
question on a mailing lists, there are at least 100 people that
are subscribed and should have posted the same question and did
dare to, and for people answering the ratio is even worse.

i am still trying to make sense of this scenario. i feel a lot of
it is due to how the way we do research has been changing of the
last 10-15 years. it is much easier to get started to do simulations
with packages like NAMD and all the documentation and tutorial
material around, so people feel more often that they are not expert
enough to answer, but also there is the problem of people that
tend to rather post a (too) trivial question instead of sitting
for five minutes and thinking about it. so if this kind of question
gets "shot down", it seems to intimidate those, that actually do
have good questions and _should_ ask.

personally, i feel a strong obligation to help people with what
i know, when people from the same community have helped me before
(my boss repeatedly called me "a good samaritan" because of that),
but then i also get the impression (and have been told so by some
people) that (too) many people view answering (trivial) questions
on mailing lists a waste of time (as they feel there is nothing in
it for them). in my personal experience, there is a lot to be gained
from sharing your knowledge, as it helps confirming your understanding
of the principles and one always has the option to start with
answering simple questions and then move on. i learned a lot about
many tools and methods just through this, _much_ more than i would
have ever learned by just "lurking". it is a pity that there are
no simple ways to reward people when they more actively participate
in forums like this... (i know that intel for example has a point
system, where people that answer often and correct in their support
forums, get a higher ranking and - depending on the ranking - some
benefits, but then again, that is a much larger community and they
have loads of cash).

cheers,
   axel.

-- 
Dr. Axel Kohlmeyer  akohlmey_at_gmail.com 
Institute for Computational Molecular Science
College of Science and Technology
Temple University, Philadelphia PA, USA.

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:51:34 CST