[gmx-users] nve energy conservation

Mark Abraham mark.j.abraham at gmail.com
Fri Mar 16 21:51:18 CET 2018


On Fri, Mar 16, 2018 at 7:42 PM Jo <jojo412202 at gmail.com> wrote:

> Thank you for your reply!
> I am attempting to conserve energy in an NVE run of 1000 SPCE water - I
> have tried a number of different verlet-buffer-tolerances (0.001 to 5e-5),
> sometimes the run output file suggests a specific verlet buffer-tolerance.
> However I am still experiencing ~600 kJ/mol shift per ns.

You can't be getting that over that whole range. If you're doing NVE with
tolerance 5e-5 then the drift is negative (from SETTLE, e.g. I just
observed -0.000069 kJ/mol/ps/atom on 1728 tip3p waters, GROMACS 2018 mixed
precision on a GPU, with SETTLE at 300K and 2fs timestep with PME and other
settings at defaults).

I have tried
> double precision which does not seem to make a difference.  I also use a
> potential modifier (potential-shift) to ensure the potential reaches 0.

You can never have an unshifted potential with the Verlet scheme.

> I
> have tried turning off the charges (to see if the ewald parameters are the
> source of the problem), but the same massive energy shift occurs - which
> leads me to believe the ewald parameters are not the source of the problem.
> I am using 'settle' to constrain the water molecule.  I am suspicious that
> this could be the cause of the energy shift. Although, looking at some of
> the previous posts on gromacs email forums, it appears that 'settle' has
> not been a problem for energy conservation previously.
> I have also tried different versions of Gromacs (5.1.4 and 2018), but the
> energy shift still occurs.
> Can you recommend any other parameters to change?  Or any other strategies
> to go about conserving energy for NVE?

First, given that you can't have zero drift in a numerical simulation with
a finite time step (and particularly not with constraints), have you
decided what level you regard as acceptable? For what application? Have you
considered the thoughts at https://www.biorxiv.org/node/23099.full?


More information about the gromacs.org_gmx-users mailing list