[gmx-users] anomalous free energy dgdl.xvg values every nstlist steps while using a twin-range cutoff

[chris.neale at utoronto.ca]

Thanks Berk, 4.0.4 does show the expected behaviour for me with the  
twin-range cutoff. Since I obtained 4.0.4 prior to mentioning this  
issue, I gather that you gave the free energy code a good looking-over  
recently. I really appreciate it.

I have seen only one previous paper that utilized a twin-range cutoff  
for annihilation simulations, but that was done via gromos in 2003  
(Lei and Smith, Biophys. J., 3513-3520, 2003) and since this part of  
the free-energy code in gromacs is done in C, not fortran, I expect  
that gromos annihilation simulations utilizing a twin-range cutoff are  
not necessarily subject to any of these issues.

Thanks again,
Chris.

-- original message --

Hi,

I just fixed this in 4.0.4.
The forces are always correct, dgdl in 4.0.3 was only correct at  
nstlist steps.

Currently 4.0.4 is only available through ftp.
We still need to add it (and the release notes) to the download page.

Berk

> Date: Tue, 17 Feb 2009 17:50:17 -0500
> From: chris.neale at utoronto.ca
> To: gmx-users at gromacs.org
> Subject: [gmx-users] anomalous free energy dgdl.xvg values every  
> nstlist steps while using a twin-range cutoff Hello,
>
> I believe that the free-energy code dgdl contribution from the  
> twin-range cutoff is being calculated incorrectly in gromacs 4.0.3  
> (and probably other versions as well).
>
> Specifically, I notice that the dgdl values spike at nstlist  
> intervals. This can be seen directly from my dgdl.xvg and also from  
> running g_analyze -ac.
>
> I suspect that while the forces from the LJ interactions that are in  
> the longer range of the twin-range are added every step while the  
> dgdl value is modified to have nstlist times these contributions  
> every nstlist step.
>
> My belief that the forces are truley added every step comes from  
> section 4.6.3 of the gromacs 4 manual:
>
> "In the neighbor list all interaction pairs that fall within rlist  
> are stored. Furthermore, the interactions between
> pairs that do not fall within rlist but do fall within  
> max(rcoulomb,rvdw) are computed during NS, and the forces
> and energy are stored separately, and added to short-range forces at  
> every time step between successive NS."
>
> Perhaps the long-range LJ component forces are actually applied to  
> the particles every nstlist steps as a force multipled by nstlist,  
> as is done in NAMD, but I am currently under the impression that  
> this is not the case.
>
> I am happy to provide more details and files if necessary, but  
> hopefully this information is sufficient for a skilled coder to take  
> a look and determine if this is indeed the case.
>
> Thank you,
> Chris.
  -- snip --