Re: [AMBER] output nuclear force and hessian

From: John Travers <jtravers70.yahoo.com>
Date: Fri, 9 Aug 2013 20:01:38 -0700 (PDT)

Hi,

I had another strange observation for the forces computed by nmode and sander. For a molecule in solvent, I performed a microcanonical dynamics. I found the the forces on the solute molecule obtained from these two code agree reasonably well, the difference < 10%. However, the forces on solvent molecules show large disparity, often > 50%. Are these difference still resulted from the different implementation and different algorithms? Thanks!

Best 

JT


________________________________
 From: Jason Swails <jason.swails.gmail.com>
To: John Travers <jtravers70.yahoo.com>; AMBER Mailing List <amber.ambermd.org>
Sent: Wednesday, August 7, 2013 2:41 PM
Subject: Re: [AMBER] output nuclear force and hessian
 

On Wed, Aug 7, 2013 at 12:49 PM, John Travers <jtravers70.yahoo.com> wrote:

> Thanks!  I made the modifications to output nuclear cartesian forces in
> runmd.F90 and nmode.F90. For the force, the outputs from these two
> subroutines are slightly different. For example, I made an test on an
> isolated benzene molecule, the forces (for the 1st C) from runmd and nmode
> are " -0.1505921505018395E+01 -0.2234835466490895E+01
>  0.4347184130879028E+00" and " -0.1505920513395167E+01
> -0.2234834582828459E+01  0.4347184163951585E+00".
>
> Does anyone know the reason for this?
>

Computers ;).  The force field is implemented differently for nmode and
sander, meaning that the order of operations will be different and errors
will accumulate -- especially in quantities like the forces where you're
adding contributions from a bunch of different force field terms.

Also, nmode and sander compute torsion terms fundamentally differently.
While both approaches are mathematically equivalent, the finite numerical
precision consequences are more ambiguous.  These numbers appear to agree
quite well.

HTH,
Jason


-- 
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
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 Fri Aug 09 2013 - 20:30:02 PDT
Custom Search