[gmx-developers] Question on reasons for temperature control

[Shirts, Michael (mrs5pt)]

I'm adding in the velocity verlet/NVT/NPT documentation, and
I've got a philosophical question about this paragraph in the manual:

"For several reasons (drift during equilibration, drift as a result of
force truncation and integration errors, heating due to external or
frictional forces), it is necessary to control the temperature of the
system. {\gromacs} can use either
the {\em weak coupling} scheme of Berendsen~\cite{Berendsen84} or
the extended ensemble Nos{\'e}-Hoover scheme~\cite{Nose84,Hoover85}."

I think this could be a bit misleading.  The main reason for using an NVT
ensemble instead of an NVE ensemble is that one is interested in
thermodynamic quantities.   It's not really to make up for the fact that
there might be errors in the integration.   While such errors may be
unavoidable, and a reasonable cost to pay to get significant speedup in the
results, using a thermostat to cover is really more of a bandaid that may do
more to fix the symptoms (increasing simulation temperature) than actually
fix the problem (errors in the physical dynamics).  Siphoning off energy
that is being generated by simulation errors gives rise to a steady state
system that is not technically in equilibrium. In most cases of interest,
statistical error is greater than any error induced by this, but it's always
a second best option to running a more accurate integration.

I'd propose a change to something like this:

"While direct use of molecular dynamics gives rise to the NVE (constant
number, constant volume, constant energy ensemble), most quantities that we
wish to calculate are from a constant temperature (NVT) ensemble. {\gromacs}
can use either the {\em weak coupling} scheme of
Berendsen~\cite{Berendsen84}, the extended ensemble Nos{\'e}-Hoover
scheme~\cite{Nose84,Hoover85}, or the velocity rescaling
scheme~\cite{Bussi2007a}.

There are several other several reasons why it might be necessary to control
the temperature of the system (drift during equilibration, drift as a result
of force truncation and integration errors, heating due to external or
frictional forces), but this is not entirely correct to do from a
thermodynamic standpoint, and caution must be taking in interpreting the
results."

~~~~~~~~~~~~
Michael Shirts
Assistant Professor
Department of Chemical Engineering
University of Virginia
michael.shirts at virginia.edu
(434)-243-1821