AMBER Archive (2007)

Subject: Re: AMBER: pmemd: same rst, different result?

From: Robert Duke (rduke_at_email.unc.edu)
Date: Wed Dec 19 2007 - 08:12:56 CST


Okay, urban myth indeed. I went back and looked at the 8 and 9 manuals.
Tom C. does document the capability to do diffusion on wrapped trajectories
in both the amber 8 and amber 9 manual, as long as we are dealing with an
orthorhombic unit cell (what's not clear from the manual is whether the code
will work on an unwrapped truncated octahedron). On the other hand, the
manual, in the description of iwrap, points out that using iwrap can screw
up your ability to do diffusion calcs - to whit "Performing such wrapping,
however, can mess up diffusion and other calculations.". This is what I
read a while back and accepted, rather uncritically :-) We should clarify
the manual, given that there is confusion even among some of the developers
:-) Regarding Tom's comment about precision being a possible problem, while
the wrap and unwrap operations are, I guess one might say, inverse
functions, the precision issue, it would seem to me, only comes into play in
one direction. So if you take absolute coords and wrap them, you may place
atoms too close to each other in the primary unit cell, causing interesting
energetics problems. Conversely though, if you are unwinding a wrapped
trajectory, the atoms are spreading out in space, and I am skeptical that 7
decimal digits of precision on coordinates minimum (after the decimal place,
typically higher actual precision retained) will place much of a limit on
diffusion results - the probability of dropping atoms on top of each other
due to precision errors seems rather low (and you are not looking at
energetics anyway). Okay, this was an issue I had no intention of beating
to death, peripheral to what I wanted to discuss in the first place.., but
thanks to all that helped in the ultimate clarification.
Regards - Bob

----- Original Message -----
From: "Thomas Cheatham" <tec3_at_utah.edu>
To: <amber_at_scripps.edu>
Sent: Wednesday, December 19, 2007 1:39 AM
Subject: Re: AMBER: pmemd: same rst, different result?

>
>> possible to me to construct a tool that would allow you to do this, but
>> the
>> current rub is that ptraj doesn't handle it. SO the basic assertion that
>> you
>
> ...actually, ptraj can (I believe) already do this for orthorhombic... I
> have yet to generalize this to triclinic (aka truncated octahedron) but
> this should be relatively straightforward for those able to penetrate the
> code. Everything is peachy (subject to precision) if your particles move
> less than 1/2 a box distance between frames processed (i.e. Yong is
> correct)...
>
> Accuracy is a separate issue since unwrapping with finite precision (i.e.
> what is stored in the prmtop for box sizes and coordinates) is less
> accurate than what is maintained internally in sander/PMEMD. There have
> been a few reports that the unwrapping (or equivalently imaging an
> IWRAP=0) is less than adequate due to the poor box precision in the
> restrt. This is something we hope to address is future releases. In the
> meantime, the error is not so severe for diffusion (but could
> potentially be for energies or overlap). The key here is the finite
> precision of the output files; when unwrapping or wrapping these, the
> precision in the box size matters. So, my feeling is that it is better to
> do this internally (IWRAP=1) as you recommend...
>
> --tom
>
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu