Maintaining ensembles during simulations.

Hello all,

I have a conceptual question regarding ensembles. Suppose i am pulling a graphene sheet with constant velocity in the vicinity of another graphene sheet.
I am trying to maintain a NVT ensemble, ideally to do this i need to equilibrate the system for every time the pulled graphene undergoes displacement. But
this is too coslty in computational resources.

Now instead i am equlibrating the system first and then slowly pulling the graphene sheet (25m/s) and i observe that the temperature is fluctuating within my
desired value. I am extracting force data from this.

Now my question is, is it conceptually correct to do this? or will i have to equlibrate for every displacement step to extract data from this?

Previously Dr. Axel said in connection to a similar problem that using Langevin in such scenarios helps in the energy NOT getting trapped and iteratively added.
Will this apply in this case too?

Thanks,
Freddie.

Hello all,

I have a conceptual question regarding ensembles. Suppose i am pulling a
graphene sheet with constant velocity in the vicinity of another graphene
sheet.
I am trying to maintain a NVT ensemble, ideally to do this i need to
equilibrate the system for every time the pulled graphene undergoes
displacement. But
this is too coslty in computational resources.

Now instead i am equlibrating the system first and then slowly pulling the
graphene sheet (25m/s) and i observe that the temperature is fluctuating
within my
desired value. I am extracting force data from this.

Now my question is, is it conceptually correct to do this? or will i have to
equlibrate for every displacement step to extract data from this?

you have to ask yourself another question, too:

how realistic is using a thermostat at all?

the nvt/npt/nph ensembles describe the behavior
of "large" *bulk* system represented by a small,
periodic sample. if you have a nanoscale object
in vacuum, what does it couple to? so is it actually
realistic to use a thermostat algorithm to remove
kinetic energy that is being added by pushing
particles around?

Previously Dr. Axel said in connection to a similar problem that using
Langevin in such scenarios helps in the energy NOT getting trapped and
iteratively added.

i cannot imagine that this was said for
a "similar problem". also, what about
langevin is iterative? it seems to me
you are confusing things.

Will this apply in this case too?

likely not. since it is not at all clear from
your statements, what the ultimate goal
of your project is, it is not possible to give
detailed advice without knowing a sufficient
amount of details. :wink:

so the ultimate answer is: it depends.

axel.

Previously Dr. Axel said in connection to a similar problem that using
Langevin in such scenarios helps in the energy NOT getting trapped and
iteratively added.

i cannot imagine that this was said for
a “similar problem”. also, what about
langevin is iterative? it seems to me
you are confusing things.

Freddie,

I was also stumped about the iterative part about the Langevin thermostat. I thought Langevin just meant to randomly select and kick a group of atoms/particles every certain number of timesteps during the run.

again, just curious in how it’s going with your graphene model.

v/r,

Daniel.

Dr. Axel,

Thanks for you reply. You are right i need to think about the scale of the problem i am trying to solve. Also the graphene sheet i am using is actually very large and in fact i have 3 such large graphene sheets in the box.

Sorry for creating a wrong impression that i meant Langevin was iterative, what i meant was that i thought Langevin was effective in removing kinetic energies that are being added to the system due to pertubation,
which would otherwise get added along every time step (carried on due to the verlet algorithm) as i am not equilibrating the system every step even though i am using “fix move” for every step.
I was referring to the KE being added every step due to approximation in my method as iterative.

Also, my problem is simple. I just try to move a graphene sheet which is surrounded by two other sheets and analyze the response, shear force and other parameters of interest. For this i am using NVT by approximating
that the system is in NVT (fingers crossed) even though as i mentioned earlier i am not equilibrating for every step.

Regards,
Freddie.

Daniel,

Thanks for your interest. As i mentioned in the earlier email, by iterative i meant the carried over KE due to improper dissipation inflicted upon by my approximation.

Thanks,
Freddie.

Dr. Axel,

[...]

Also, my problem is simple. I just try to move a graphene sheet which is
surrounded by two other sheets and analyze the response, shear force and
other parameters of interest. For this i am using NVT by approximating
that the system is in NVT (fingers crossed) even though as i mentioned

if i would be your adviser, this "fingers crossed" statement
would get you into *big* trouble. research doesn't work on
the foundation of guessing and hoping that it will work out
somehow. whenever you make an approximation, you have
to *understand* why you make it and what its impact is.

it appears to me that you have not thought through
what your simulation is supposed to represent.
are you looking to get a *free* energy profile?
or a *static* potential energy surface?

in both cases, the application of a thermostat
has to be considered *very* carefully. in the
second it would be mostly useless in the first
i can see that i might make sense to thermalize
the two outer sheets, but not necessary the
one you are pulling through.

at this point, boundary conditions also matter.
are the sheets infinite (then why do they need
to be large?) or of limited size (which would
render any statement of performing a simulation
in an NVT ensemble as wrong).

earlier i am not equilibrating for every step.

it totally escapes me, why there should be a
necessity to maintain an NVT ensemble in
the first place.

axel.

Your question really is: how long the simulation should run to be considered as “quasi-static” loading.

  1. just looking at temperature is not enough to answer this question. I’m pretty sure any thermostat in lammps can do a good job on this as long as you choose the right parameters.
  2. depending on what problem you are dealing with, the time scale for claiming “quasi-static” condition is different. In problems like dislocation nucleation, it is an issue of thermal activated process. Physically, the rate of dislocation nucleation is dependent upon the stress you applied. That being said, strain rate sensitivity is a material-dependent property.
  3. in terms of simulation, you may never reach “quasi-static” condition because of time-scale issue in MD.

AC