[gmx-users] Re: Question about Berendsen thermostat and

[Michael Shirts]

> On that topic: There is a recent paper from the Tavan group where they
> carefully discuss and investigate the effects of different
> thermostats. They also come up with a "noninvasive" thermostatting
> procedure where they couple a Berendsen thermostat to the solvent only
> such that it acts as an explicit heat bath for the solute (which then
> samples the canonical ensemble).
>
> http://dx.doi.org/10.1021%2Fct8000365

This is an excellent paper.  Thanks so much for the reference!  Though
I would note that really, it would be better to thermostat only waters
a shell or two from the protein, since the dynamics of nearby waters
will affect the dynamics of the proteins.   Though thermostatting just
the solvent is better than thermostatting everything, certainly!

>> But does the NH chain, the mother of all thermostats, indeed yield
>> correct dynamics?
> Yes, it does. Always assuming that you couple only one (NH) thermostat
> to the system. The other thermostats in the NH chain then are
> successively coupled to one another, not to the physical system. For
> the proof that NH does indeed create a canonical ensemble (see the
> original NH papers), these "thermostats of thermostats" don't really
> matter.

The question was, does NH give the correct -dynamics-, not the correct
ensemble. The answer has to be no, since the velocities of the
particles are affected in all thermostat mechanics.  Tthe paper that
you reference (the Tavan paper) says NH doesn't get the dynamics
correct.  It's still a velocity scaling method, though it's a
oscillatory one, not an exponential one.

Best,
Michael