[gmx-users] Energy conservation issue

[Berk Hess]

>From: Yujie Wu <yujie.wu at hec.utah.edu>
>Reply-To: wu at hec.utah.edu,Discussion list for GROMACS users 
><gmx-users at gromacs.org>
>To: gmx-users at gromacs.org
>Subject: [gmx-users] Energy conservation issue
>Date: Thu, 27 Oct 2005 00:08:03 -0600
>
>
>I am having an energy conservation problem with my system, and I am not
>sure exactly what causes the problem. Previously, I thought it could be
>due to the single-precision. But I just noticed the recent GROMACS paper
>published on JCC, saying:
>
>(quote)"By taking care of summation order and sometimes using
>intermediate double precision variables it has been possible to use
>single precision by default while still conserving energy perfectly in
>microcanonical ensemble runs."
>
>This is good (if true), but it worries me because I am afraid that I did
>something stupid that caused the energy drifting. To confirm the problem,
>I ran an NVE simulation on a 216-SPC-water system (version 3.1.4, single
>precision, serial). The energy conservation is bad (-32 kJ/mol/ns). For
>the same system and equivalent simulation conditions, DL_POLY (which
>uses double precision) can give really almost perfect conservation. I
>still cannot see what is wrong with my GROMACS run, but I could be
>easily blind to my own mistakes. So I post my mdout.mdp file below, and
>I would appreciate it if anyone please points out the problem. Many
>thanks in advance.

I would say that the LJ cut-off is the largest source of errors.
I don't know how DL_POLY handles this.
But setting
vdw-type = shift
rvdw = 0.85
should help a lot.

BTW Why are you using such an old Gromacs version?

Berk.