[gmx-users] What is the current wisdom on energy leak?

[ms]

Dear Gromacs users,

I am a little new fellow in this community, and many things still I
don't know.

I am trying to parametrize a coarse-grained FF for polymer simulations,
without solvent. A collegue of mine advised me, just to be sure, to
check for possible energy leak artefacts. To do that he told me that I
should run a simulation with no thermal coupling and check the energy
conservation. He told me that in such case, in a well-behaved system,
"dE/E<10-4, 10-5" (where dE is the RMSD?).

By playing with that indeed I found that, as expected, using pme-switch
and shift for Coulombd and VdW respectively helps a lot (with respect to
cut-offs), but even increasing LINCS, decreasing neighbour search time
etc. my ratio doesn't go below 2E-3. I wonder however if there are also
possibly poor minimization effects in that -it seems that it minimizes
correctly, but...

My problem however is that apart from that informal advice I can find
very little practical on the subject. The only clear thing I can find on
the net is a Van der Spoel students tutorial which says:

https://extras.csc.fi/chem/courses/gmx2004/exercises/index.html#mozTocId961213

"Conservation of energy can be monitored by checking the ratio of the
fluctuations in Total energy and the Kinetic Energy. For a well behaved
system the ratio R:
    R = Δ Etot/Δ Ekin
should be less than 0.05."

but it doesn't specify, as far as I can see, the conditions in which
this is a good ratio (solvent, no solvent, which thermostat, etc.).

Has anyone advices for bibliography or some general practical wisdom on
this problem?

Thanks a lot,
Massimo