Re: [AMBER] Particles moving outside of simulation box for pmemd.cuda.mpi

From: Ross Walker <ross.rosswalker.co.uk>
Date: Tue, 3 Apr 2012 16:14:19 -0700

Hi Adam,

 

To be honest it should work but I actually think your box is being specified
incorrectly, if the origin is not in the right place this will cause you
issues as will the wrapping it the box dimensions and angles are not
correctly defined. I also believe that AMBER expects the 'familiar'
coordinate frame for triclinic boxes and so this is likely your problem.

 

I would try imaging this through ptraj first to see if you can get it
correct - that is looking like a traditional truncated octahedron which is
what I assume you want the box to be yes?

 

I would also work entirely with the CPU code until you are confident your
input coordinates are being processed correctly and only then move to the
GPU code.

 

All the best

Ross

 

From: Adam Jion [mailto:adamjion.yahoo.com]
Sent: Monday, April 02, 2012 11:41 PM
To: Ross Walker
Subject: Re: [AMBER] Particles moving outside of simulation box for
pmemd.cuda.mpi

 

Hi Ross,

 

Attached is a picture of the input coordinates with it's periodic box.

Some of the particles are outside the box even before the start of
simulation.

These particles are just images of what is supposed to be in the primary
box.

Can Amber handle such input coordinates? Or must all input coordinates be
inside the box?

Gromacs, for example, accept input coordinates that are outside the box and
treat them as images of particles inside the box. Can I assume that Amber
does the same thing?

 

Regards,

Adam

 

ps. Unable to send you all the input files because they are really large.

 

  _____

From: Ross Walker <ross.rosswalker.co.uk>
To: 'Adam Jion' <adamjion.yahoo.com>
Sent: Tuesday, April 3, 2012 12:18 AM
Subject: RE: [AMBER] Particles moving outside of simulation box for
pmemd.cuda.mpi

 

Hi Adam,

 

You said that your box was Triclinic but your before image looks like
Orthorhombic to me. If indeed you have your angles wrong or some other
setting incorrect it is perfectly reasonable that you'd see things going
haywire. We'll need to see all your input files to figure out what is going
on here. Also have you tried this with the CPU code? Please do, it is no
good asserting that it might be a bug in the GPU MPI code unless you can
show that it runs fine in serial GPU and MPI CPU and that it is just with
MPI GPU that the behavior is different.

 

Either way something looks fishy to me and I suspect a problem with the
setup of the simulation.

 

All the best

Ross

 

From: Adam Jion [mailto:adamjion.yahoo.com]
Sent: Sunday, April 01, 2012 4:37 AM
To: AMBER Mailing List; ross.rosswalker.co.uk
Subject: Re: [AMBER] Particles moving outside of simulation box for
pmemd.cuda.mpi

 

Hi Ross,

 

Attached are the vmd images before and after the simulation.

Despite the iwrap=1 command, the particles (whole molecules) still moved
outside the box.

I had the same problem when using Gromacs. The problem then was that the
GPU-version of Gromacs used OpenMM which unboxed the trajectories (although
Gromacs had a post-simulation option to box-up the trajectories).

So could it be that the MPI code for Amber-GPU did the same thing?

 

Regards,

Adam

 

ps. Amber-GPU is about 3 times fater than Gromacs-GPU :-)

 

 

  _____

From: Ross Walker <ross.rosswalker.co.uk>
To: 'Adam Jion' <adamjion.yahoo.com>; 'AMBER Mailing List'
<amber.ambermd.org>
Sent: Saturday, March 31, 2012 1:34 AM
Subject: Re: [AMBER] Particles moving outside of simulation box for
pmemd.cuda.mpi


Hi Adam,

> I used the 'iwrap = 1' function in my input file but still got
> particles moving out of the box when I looked at the .rsrt file.
> My simulation box is a triclinic cell.
>
> Is this an inevitable feature of pmemd.cuda.mpi?
> Can I get the particles to stay within the simulation box without post-
> simulation modifications?
>
> Or is there something wrong with the implemenation in my input file
> (see below).

We will need to see your input files in order to investigate this further.
In particular specifically what you mean by particles not staying within the
simulation box. Both PMEMD and Sander by default track the coordinates of
the original molecule, I.e. as it moves out of the central box it is imaged
internally (in that fractional coordinates are used) but the coordinates
that get tracked are that of the original molecule. Trajectory files can be
reimaged later in ptraj to recenter things etc. Setting iwrap will cause the
trajectory file coordinates to be imaged as a 'molecule' moves out of the
central box. The key here is molecule, AMBER only images based on molecules,
not atoms, so you can get individual hydrogen's in water appearing to poke
out of the edge of a box. You can also get whole chunks of your protein
appearing to stick out of the box. This is purely a visual artifact though
and can be 'fixed' by post processing with ptraj.

That said there could be an issue with the way your box is defined that is
leading to problems but we'll need a lot more information to be able to
track that down. One thing you should consider trying if you believe this is
a bug with the CUDA code is to run with the cpu version of pmemd (or sander)
and see if the behavior is different.

All the best
Ross

/\
\/
|\oss Walker

---------------------------------------------------------
| Assistant Research Professor |
| San Diego Supercomputer Center |
| Adjunct Assistant Professor |
| Dept. of Chemistry and Biochemistry |
| University of California San Diego |
| NVIDIA Fellow |
| http://www.rosswalker.co.uk | http://www.wmd-lab.org/ |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
---------------------------------------------------------

Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.





_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber

 

 

_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Apr 03 2012 - 16:30:03 PDT
Custom Search