[gmx-developers] Question about Conserved Energy value in NVT simulations with new thermostat.

[Mike Colvin]

Hi All,
    First, thanks to the developers for quickly incorporating Bussi, et 
al.'s new velocity rescaling thermostat. The thermostat works great, but 
I have a question about the "Conserved En" term reported, which so far 
as I could glean from the source is the same as the Bussi's effective 
energy (H hat) defined in eq. 15 of his paper.  However, for some NVT 
test simulations that seem otherwise very well behaved, the Conserved 
Energy changes steadily over the simulation which I wouldn't expect for 
simulations with sufficiently small time steps. For a liquid argon test 
case (T=303.9, dens=1.485 gm/cm3) this value drops exactly linearly in 
time with a slope that is exactly proportional to the timestep (tested 
at 2fs, 1fs, and .5 fs).  For an SPC/E water (at STP) this value drifts 
upward during the simulation with a slope that is roughly proportional 
to the square root of the timestep (ts=.25, .5, 1, and 2fs).  Anyway, I 
was curious if anyone else had noticed this phenomena and whether I'm 
misinterpreting what this term is.  (And please let me know whether this 
or the gmx-user list is the best place for a question like this.)   
Thanks much!

--Mike Colvin
University of California, Merced