R: Re: Re: temperature in NVE ensembles

I tried to modify in the fix nvt the damp parameter, but I see no particular

When plotting istantenous temperature (given as output by lammps) versus time
during nvt integration, I see that the frequency of oscillation of temperature
depends upon the damp parameter but the amplitude of oscillations is
independent. Is this correct?

I am not a frequent user of nosè hoover integration. Maybe when I switch off
nvt integration and shitch on nve, the new temperature will try to oscillate
around the istantenous value of the temperature at which I switched off nosè

Maybe as you say these oscillations in nvt are a finite size effect, since I'm
using quite small systems (hundreds of atoms), but I think there is a brute
force way to overcome the problem, that is starting NVE integrations from
different configurations that have the same fixed total energy and equilibrate
using an NVE ensemble. I don't know if this is a good and feasible idea in
lammps and which is the best way to implement it.

Again, thanks a lot for the help!

----Messaggio originale----
Da: [email protected]
Data: 13/08/2013 15.27
A: "[email protected]"<[email protected]>
Cc: "LAMMPS Users Mailing List"<[email protected]>
Ogg: Re: Re: [lammps-users] temperature in NVE ensembles

----Messaggio originale----
Da: [email protected]
Data: 13/08/2013 15.02
A: "[email protected]"<[email protected]>
Cc: "LAMMPS Users Mailing List"<[email protected]>
Ogg: Re: [lammps-users] temperature in NVE ensembles

Hi everybody,

      I've got a basic question I think. I'd like to create several


NVE ensembles in a simple system (say a crystal sylicon, or LJ Argon).
Temperature in equilibrium is not a well defined quantity in such



oscillates around a mean value T , that depends in a non trivial way on


total energy, and that we call the 'temperature' of the system.

The point is that I'd like to run different NVE simulations fixing a


this mean value of the temperature. Which is the best efficient way in


to achieve that ? (equilibrating with an NVT thermostat and than switch



gets me close to the right temperature but not close enough)

what is close enough? if you run your simulation long enough with a
proper thermalization until it is well equilibrated, then it should
maintain that (average) temperature. if not, then you need to
investigate your settings, how long you ran the MD, how well you
sampled phase space, how well it is equilibrated, how well you
conserve energy. when running with a thermostat reduce the intensity
of the thermal coupling, you should transition from NVT to NVE

First of all, thanks for your reply.

I'll try to check this. What I observed till now is that if I try to
equilibrate using NVT at a certain temperature T and after equilibration I
switch to NVE, temperature will start oscillating aroung a value close to T


not exactly T (say a 10% difference). Maybe I did something wrong I'll try


check once again the equilibration...

I didn't understand your last sentence:

"when running with a thermostat reduce the intensity
of the thermal coupling, you should transition from NVT to NVE

if you increase the coupling constant (or characteristic time) of the
nose-hoover, it will couple less. essentially, you control with it the
characteristic frequency that you couple to. if this is very low, the
exchange of kinetic energy between the nose-hoover chains and your
system will be very slow and at some point there should not be a
significant difference.

Maybe it sounds strange to work in a NVE ensemble and to fix the mean


The temperature of the NVE simulation should depend only on the value
of E at the end of the NVT run. It does not matter how you thermostat,
you will never come closer to the desired temperature than deltaT ~
sqrt(1/N). During an NVT run, the total energy fluctuates about an
average value that is determined by the set temperature. The
fluctuations are proportional to sqrt(heat capacity). When you switch
from NVT to NVE, you are taking a sample from that energy
distribution, not the average of the distribution.

To do what you want, you should follow Axel's original suggestion and
generate a master plot of <T>_NVE versus E.

also, regarding

"because I'd like to study different replicas of the same system."

If you want to generate *equivalent* independent replicas, it is
sufficient that they have the same value of some state variable, in
this case either E or T. There is no strong reason to favor one or the
other and choosing E eliminates your problem. If it ain't broke, don't
fix it, especially if you don't know how to.