Re: The use of NPT or NVT ensemble to observe dynamical parameters

From: Ana Celia Vila Verde (
Date: Mon Jan 02 2017 - 04:13:20 CST

To simulate dynamical observables you want to have an ensemble that
introduces minimal external effects to the dynamics: no barostat and
ideally no thermostat. So, preferably NVE or if anything an NVT with
the weakest thermostat you can get away with (possibly even weaker than
what you have tried below).

TIP3P is known to yield much faster dynamics than the experimentally
determined values. This is typically a plus, because you get more
sampling from the same simulated nanoseconds than what you'd get with a
water model that had a D closer to the experiment. I have not read the
literature that you mentioned, about tuning the friction parameter to
get the D that matches experiments, so I don't know how sound this
approach is. If you're not convinced by the arguments in those papers,
I'd recommend either changing to a water model that has dynamics closer
to experiment (if you're interested in dynamics with characteristic
timescales <5 ps), or keeping TIP3P with a weak thermostat and then just
acknowledging the fact that it yields dynamics about twice as fast as
experiment when you discuss your results.

I hope it helps. Best,


On 02/01/17 09:27, Harish Srinivasan wrote:
> Dear all,
> A very happy new year.
> I am trying to simulate soft matter systems like micelles, vesicles
> etc. We calculate various dynamical parameters like diffussivity,
> velocity autocorrelation, incoherent inelastic scattering function
> etc. It is quite impossible to set up an NVE simulation as it becomes
> very difficult to maintain the temperature of the system.
> One can use NPT or NVT ensemble in such scenarios. But, there is an
> important point of concern about the right choice of friction
> parameters. It is well known that the friction parameter can
> indirectly affect the values of the dynamical quantities observed and
> that is why NVE is the best choice when probing dynamics of the
> system. So is it appropriate to choose a friction coefficient
> (although they might be large) such that the dynamical parameters
> match with experimental values ? A plenty of recent publications have
> followed this kind of a scheme. But some old publications also suggest
> that one should choose this parameter as small as possible so as to
> neglect its effects on dynamics.
> For example, the diffusivity of TIP3P water
> D (expt) = 2.6 x 10^{-5} cm^2/s
> D (NVE) = 5.3 x 10^{-5} cm^2/s
> D (NVT) = 5.1 x 10^{-5} cm^2/s (friction = 1 /ps)
> D (NVT) = 2.9 x 10^{-5} cm^2/s (friction = 5 /ps)
> Now, the friction = 5/ps matches well with expt. but not with its NVE
> values. So what is the right choice friction parameter here for a
> truthful simulation ?
> Thank you for your kind inputs and suggestions,
> Harish

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2018 - 23:19:59 CST