[gmx-users] Periodic Images - clarification

[Justin A. Lemkul]

Kavyashree M wrote:
> 
> Sir,
>  I would like to thank you for patiently replying for my repeatedly 
> asked question.
> 
> 
>     For most stable, well-behaved proteins, setting a suitable box size
>     at the outset of the simulation is sufficient to avoid spurious PBC
>     interactions.  In your case, there are several possibilities: (1)
>     the protein is not well-behaved, (2) you didn't set the box you
>     think you did, (3) the .mdp settings are wrong and lead to
>     instability, or (4) your pressure coupling settings cause the box to
>     shrink unreasonably.
> 
>  
> 1. protein is not well behaved - This point I dont know how to quantify.

It's not necessarily something you can quantify, it's more of a qualitative 
measure in many cases and comes from anticipating what the system may do.  Not 
all proteins are stably folded.  Some may unfold, others may have multiple 
domains that will move along hinge regions, causing the protein to expand its 
size, etc.  The initial box size assumes that large changes in the structure 
will not occur.  Sometimes this assumption is not good.

> 2. Box dimensions - I repeated from the model, editconf gave the same 
> box dimensions which I had used earlier but did not  
>     repeat the NVT. and the distance -d between wall of box and protein 
> atom was kept as 1.0nm while the max cut of used was
>    1.4nm.
> 3. I am attaching the mdp file.
> 4. I checked the whole trajectory for change in box size with the output 
> of g_energy. But did not find and abrupt deviations
>  

OK.

> 
>     If you want to use the first 26 ns only, I suppose these data are
>     legitimate, although then several questions arise.  Why did you run
>     100 ns in the first place?  Presumably you felt that you needed such
>     a simulation length to address whatever question you're asking, so
>     is 26 ns legitimate, or is it simply convenient because you don't
>     want to run the simulation again?  Also, why trust these results
>     when you know that just a short time later these dynamics produced
>     flawed information?  The PBC violation may not have simply happened
>     suddenly; maybe it was a product of some long-term motion in the
>     system that was continually trending towards disaster.
> 
>  
> I did not anticipate such a violation would as it did not happen in 
> other cases. so I did not check the minimum image violation
> while running the simulation but caculated after 100ns. I agree it was 
> my stupidity. Because of time constraints and system unavailability now 
> I might not be able to run another simulation. But I will be running it 
> later with corrected parameters for sure.
> 
> I agree that it is producing flawed results. But My point was if at all 
> it was caused only due to the box dimension being smaller
> and not due to any wrong parameters used why is that 26ns wrong. 
> Probably if I had selected a bigger box size may be that
> loop would have continued to move without minimum image violation.
>  
> 
>     The biggest question is, if you run the simulation again (which you
>     should, but only after answering the four points above and the
>     following), how do you know the same thing won't happen again?
>      You've been asking related questions for weeks and I still do not
>     know if you have followed my repeated advice to watch the trajectory
>     with a PBC unit cell enabled in your favorite visualization program
>     and, in concert with the identified problematic atoms in the
>     g_mindist output, identify where and why the minimum image violation
>     occurred.  Doing so should take minutes and you should immediately
>     see what went wrong, which would be valuable information for
>     avoiding such behavior in the future.  If you have done this, you've
>     posted no evidence of your findings and thus just wasted weeks
>     posting the same (or tangentially related) questions with no answer,
>     time that could have been spent running a proper simulation to
>     recover what you lost.
> 
> 
> If I am running again I would increase the box size and run. I did what 
> you had suggested.  I visualized that part of trajectory
> in VMD (which I am not very comfortable with ) and could see a loop 
> movement coming closer to it periodic image. but
> unfortunately because of my lack of know-how I was unable to measure the 
> distance between them in VMD. I could only visualixe the loop movement 
> but I am unable to produce and concrete outputs for my observation.
>  

It seems you have identified the source of the problem then.  If you have a 
large, unpredictable loop region, then you need to account for the fact that it 
might do funny things throughout the simulation.  In this case, maybe your 26 ns 
is useful, but my point is that if you thought 100 ns was needed to answer your 
question of interest, then you still probably need to do a new simulation.

Also realize that the results of just a single simulation is usually not 
sufficient to derive converged quantities.  What if your one simulation is the 
outlier in the sample set?  You have no way to know if you've done just one 
simulation.

-Justin

-- 
========================================

Justin A. Lemkul
Ph.D. Candidate
ICTAS Doctoral Scholar
MILES-IGERT Trainee
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin

========================================