Hi Clemens,

As Charles pointed out in his note on 7/25,

> The script doesn't appear to be particularly robust as it seems to  
> blow
> up during dynamics unless I reduce the time stepsize to 0.001 ps.  
> Use of
> the variable stepsize feature of the IVM would probably fix this,
> though.

The ATMCHK error you're seeing is caused by a solvent atom moving too  
far from the center of the coordinate system,
so that its position has a coordinate greater than 9999 A.

I glanced through the scripts, and the only energy term holding the  
waters in place that I noticed is the electric field.  Thus,
if a particular water molecule gets going fast enough, there'll be  
nothing to keep it bound near the protein.  It'll go flying off
into space, eventually triggering the ATMCHK error.

Individual atoms can get heated up quite a lot by integration errors  
during molecular dynamics.  If, for example, two atoms
move from being close, but not overlapping, to overlapping each other  
significantly in the course of one dynamics step,
then the system has instantaneously gained a whole bunch of potential  
energy.  Those two atoms in our example will
spring away from each other, having heated up significantly.

So how can you avoid this kind of behavior?  You have two choices:

1.  Change the timestep settings in Jens's h20refinenew2.xplor (just  
search through for 'time').  He has them all
set at 5 fs, which is rather agressive for Cartesian-space dynamics.   
Dropping them all to 1 or 2 fs should eliminate
the problem, as Charles pointed out.  But remember to increase the  
number of dynamics steps in each case to keep
the total time constant.  This will, unfortunately, increase the CPU  
time per structure.

2.  Switch the scripts from using the Langevin dynamics engine to  
using Charles's internal variable dynamics engine.
Turn on its variable-stepsize setting.  This will cause the dynamics  
engine to take short timesteps in situations where
large steps could inject a lot of energy into the system (as per our  
example above), but allow the dynamics integrator
to take long timesteps in areas where it can get away with it.  This  
should eliminate the ATMCHK error without costing
too much CPU time.

Hope this helps.

--JK

On Aug 14, 2007, at 10:16 PM, Clemens C. Heikaus wrote:

> Hello,
>
> I have a very similar problem like Martin. I tried to run water  
> refinement with XPLOR-NIH 2.16.0 and used the example files from  
> ftp://ftp.pasteur.fr/pub/BIS/linge/ to see whether it is running  
> correctly but get the following error message after approx. 2  
> minutes in which it runs just fine. For my real case (155 residues  
> plus organic non-bound ligand) I have tried water refinement with  
> the recoord shell scripts and it was the same error message.
>
>  %ATMCHK-ERR: unknown coordinates for atom "WAT1-386 -TIP3-H1  "
>  %ATMCHK-ERR: Unknown coordinates
>  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>  BOMLEV=    0 reached.  Program execution will be terminated.
>  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>  Subroutine DIE called . Terminating
>
> Do you have any ideas what the problem might be? Help would be very  
> much appreciated.
> Clemens
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
http://dcb.cit.nih.gov/pipermail/xplor-nih/attachments/20070814/8f9fd2df/attachment.html
 

Reply via email to