[gmx-users] energy drift in nve with conservative parameters

[Michael Shirts]

Energy is not conserved in NVE if the potential is not continuous and
differentiable.  If you use an abrupt VdW cutoff, the potential is obviously
not continuous at the cutoff.  I would suggest using a switched VdW
function that tapers smoothly to zero at the cutoff distance.  I think
technically, the second derivative of the potential needs to be continuous at
the boundary (somebody correct me if I am wrong), but in practice having the
first derivative continuous at the boundary seems to be sufficient.

Also, even if your potential is differentiable, if rlist = rvdw, you get
particles that drift into the non-zero potential region between neighborlist
updates, and you get lack of energy conservation.  This will be a lesser
effect, but it will still be noticeable.  I would suggest something
like:

rlist			 = 1.0
vdw-type                 = Switch
rvdw-switch              = 0.8
rvdw                     = 0.9

I think that the current version of GROMACS doesn't allow rvdw < rlist, but
this is an artifact of the use of rlist being used as an option for where
short-range / long-range cutoffs are, and can be changed in the one line of
code (Erik, do you remember where that line is?) where it's checked for in
grompp without any ill effects.  It will be fixed in the next gromacs version,
as I understand.

Note that in this case, the VdW energy will not be treated appropriately, as
the long range correction is only for an abrupt cutoff.  I've given Erik code
which computes the long range correction correctly for arbitrary switched
potentials, so that should be in the next version of gromacs.

I believe PME might cause some energy conservation problems as the particles
are treated differently as they cross the boundary, but I haven't tested that
yet (On my list of things to do for the week!)  The effect will almost
certainly be less than for the above problems.

Hopefully this is helpful  - anyone out there that has additional information
on energy conservation, let me know . . .

Cheers,
Michael