Hi, all
I am simulating a C60 in a CNT, and I want to control the temperature of the C60 to be 300K. However, the translational diffusion velocity of the C60 is very high. When we calculate the temperature, the velocity used in LAMMPS is the V(atom), In fact, the more reasonable value should be V(atom)-V(averaged moving of C60). I wonder is there any thermal control method which can eliminate the influence of V(averaged moving of C60)?
Thanks
Yu
Hi, all
I am simulating a C60 in a CNT, and I want to control the temperature of the
C60 to be 300K. However, the translational diffusion velocity of the C60 is
very high. When we calculate the temperature, the velocity used in LAMMPS is
the V(atom), In fact, the more reasonable value should be V(atom)-V(averaged
moving of C60). I wonder is there any thermal control method which can
eliminate the influence of V(averaged moving of C60)?
actually, applying a thermostat under these circumstances
seems like a bad physical model to me.
remember that on the atomic level, temperature is not really
well defined. we typically treat it as if it is just the kinetic energy
divided by the number of degrees of freedom and multiplied
by a scaling factor, but on this level of resolution using nve
would be _much_ more realistic.
temperature control would mean to couple a system to a heat bath.
that is a reasonable concept, if you want to represent the thermal
fluctuations of a large dense (bulk) system with a relatively small
number of atoms, but if you have individual objects like a C60 or
a CNT, where would be the justification for that coupling??
cheers,
axel.
Thanks, Axel
What you mean NVE should be NVE without any thermal control, correct? It is a good choice, but it still need a starting configuration and velocity profile of the atoms, which is corresponding to a initial temperature value.
Besides, The goal of my current simulation is to study the effect of temperature(thermal ) on the reorientation and diffusion of the C60 under 1D confinement, thus, NVE without thermal control is not suitable. In my opinion, the thermostat is just a modification method on the velocity of atoms, regardless of the model size, it should be the same as the temp/rescale. Now, what I am using now is NVE+(temp/berendsen on CNT and C60 respectively), but I do not know whether it is OK.
2010-7-12下午11:34，Axel Kohlmeyer <akohlmey@…12…24…>撰写：
For a single C60, you can use compute temp/com to subtract out
the motion of the center-of-mass before computing its themal T.
You can also use that compute with a thermostat.
Steve
2010/7/12 <[email protected]...>:
Thanks, Axel
What you mean NVE should be NVE without any thermal control, correct? It is
a good choice, but it still need a starting configuration and velocity
profile of the atoms, which is corresponding to a initial temperature value.
everything is allowed in love, war and setting up your system.
Besides, The goal of my current simulation is to study the effect of
temperature(thermal ) on the reorientation and diffusion of the C60 under 1D
confinement, thus, NVE without thermal control is not suitable. In my
opinion, the thermostat is just a modification method on the velocity of
i disagree with this statement. a thermostat is modeling the coupling
to a larger reservoir of degrees of freedom. for your system, this
doesn't exist, at least not for the C60. i would not want to do any
thermalization
on it, as its only coupling with the "outside world" is the direct
interaction with
the CNT, and that you have represented explicitly. any thermostat would
add unphysical changes to the C60 trajectory in my opinion.
now the CNT is a different story. it could be coupled to the matrix it is
embedded in and thus thermalization could be justified. but i would make
it a very, very weak thermostat (i.e. use a very long time constant, only
large enough to remove the drift in total energy from numerical inaccuracies
during time integration).
atoms, regardless of the model size, it should be the same as the
temp/rescale. Now, what I am using now is NVE+(temp/berendsen on CNT and C60
respectively), but I do not know whether it is OK.
applying thermostats to "free" particles can have lots of undesireable
or unexpected effects on your results. you want to avoid temp/rescale
at all costs, as this gives your system an unphysical "kick" at every
rescale. generally, any method based on velocity rescaling (i.e. also
a berendsen thermostat) has the downside of emphasizing any existing
collective motion (normal mode) and thus adding unwanted bias
to your phase space sampling. taking about sampling, often a setup
with a "massive thermostat" (i.e. one thermostat per degree of freedom)
is used in these cases to improve ergodicity, but one has to be aware
that it will also affect (dampen) the magnitude of fluctuations in the
kinetic energy (so the opposite of a global rescaling temperature control).
finally, for a small system like yours, sampling is fairly easy to achieve
with a classical model, but you will have to compensate for the unavoidable
drift in total energy due to the numerical integration of the equations of
motion.
this is why short of having a massive nose-hoover thermostat available
to me the choice of a langevin thermostat, but with a very long time constant
(and thus minimal bias) and only applied to the CNT would appear to be
the best choice.
cheers,
axel.
I want to add a couple of comments to the thread. As Axel has said, it will be a bad idea to thermostat this system. I would go a bit further and say, in general, it is a bad idea to thermostat any system from which you are planning to calculate dynamical properties, eg. diffusion. Thermostats alter dynamics - period. Some more noticeably than others. Why not avoid the possible complications if you can?
If you need a trajectory with some average constant temperature, you can do this in the NVE ensemble. I do it this way - choose some starting configuration (this will give you a starting potential energy), and some starting distribution of atomic velocities (this will give you a starting kinetic energy). By doing this you have in effect chosen a starting total energy, which will ideally be conserved. Then let the thing run. Assuming good sampling, the system will fluctuate around some average temp. If it is too low or high, adjust the starting temperature accordingly, and repeat. As Axel has pointed out, there will be some energy drift and concomitant temperature drift. If you are lucky the property you are interested in will converge quickly and the drift will be negligible. Otherwise, you will need to do as he suggests and use a *very* weak thermostat.
Removing the COM motion of the C60 from the temperature is a terrible idea. Your system should have no net COM drift to start with. If you remove the COM motion of the C60 and thermostat with Nose Hoover I'm almost certain you will end up with a net translation. Maybe not.
Matt
Quoting Axel Kohlmeyer <[email protected]>:
Thanks for all your comments.
Thus, i think now there are two relative reasonable solutions for this system:
1, assign initial KE to the system and then run NVE to reach equilibrium without any temp control.
2, run the system in NVE and apply a very weak thermostat on the CNT, but not C60. Possibly, even thought, in the equilibrium, the temp of C60 (subtract the bias) may not be the same as the temp of CNT.
Well, it seems to reminder me that MD has a limitation on the model size, as the static mechanics many only apply for model, whose atoms are more than a certain value, then, what factors decide this ‘certain value’?
2010-7-14上午12:08，“Matt K. Petersen” <matt@…1669…>撰写：
Removing the COM motion of the C60 from the temperature is a terrible
idea. Your system should have no net COM drift to start with. If you
remove the COM motion of the C60 and thermostat with Nose Hoover I'm
almost certain you will end up with a net translation. Maybe not.
I think this depends on what your definition of temperature is. If you replace
60 with a million, e.g. for a nanoparticle, and ask what it means for the
nanoparticle to be at 300K, then it really means you want the COM subtracted
out. If the nanoparticle is ballistic at some large velocity, that
really doesn't
contribute to the T of the atoms in the nanoparticle. You may be right
that this will cause problems with Nose/Hoover however, particularly
if N is small, e.g. 60.
Steve