[lammps-users] Cooling of Non-equilibrium Box

Dear LAMMPS Users,

I have a box of two joined metals at the interface and the interface between the two metals is at a high temperature (2000K) and about 5nm away from the interface the temperature of the atoms is about 400K. I want to cool this box back to room temperature. Using NVT what would be the proper setup for Tstart ? Also please enlighten me if there is another way that can be used to cool such situation.

Regards

please note that the nvt integrator applies a global thermostat, i.e. it considers the temperature of the entire system and thus your tstart should be simply the original initial temperature of the entire system. fix nvt also is not a dissipative thermostat, so it will not actively increase distributing the kinetic energy across the thermostatted group of atoms.

please note that cooling with a thermostat for this kind of system in this kind of way is in no way going to resemble the physical process. thermostat algorithms are designed to model a small, periodic system as if it was a very large system in contact with a heat bath in such a way that there will be no measurable gradients and that the sampling of phase space is approximating the NVT statistical mechanical ensemble (at constant!! temperature).

if you want to correctly model heat exchange at a surface you may apply the thermostat only to the atoms at that surface and then use fix nve for the rest, which will require some experimentation to achieve the desired cooling rate.

axel.

there is not a simple “do this not that” type of answer to your question. there are many ways to remove kinetic energy from a system.

so what you also need to explain is:

  • what exactly do you need the cooled down system for?
  • what, if anything, do you want to learn from the cooling down simulation?
  • is this connected to some kind of experiment, i.e. should it resemble some physical process? if yes, do you need to achieve some specific cooling rate?

you need to keep in mind that it is effectively impossible to model macroscopic cooling processes at the length and time scales of atomic level simulations unless they are extremely fast.

axel.

To add to my previous reply. I was thinking to add a sort of “heat reservoir” or “heat sink”. A layer of atom that is always kept at 300K so that it cools the sim box in NVE. However I tried that and the temperature doesnt seem to change. If you please could assist me to achieve such method if it is even possible ?

if done correctly (it will probably better to have more than one layer and use a dissipative thermostat on them, not fix nvt), it should work. the rate at which it will cool down will depend on the thermal conductivity of your setup and the time constant applied to the thermostat (i.e. how quickly it will remove excess kinetic energy).

I agree with you that the NVT of two regions is no where near physical and thermostats are not suitable. Regarding the global temp. Calculation, i think the fix modify can use a temp compute specifed to that region. But the double or triple NVT ensemble in the same box is not good idea.

i don’t understand what you are getting at here. from your description just using a single fix nvt for the entire system seems the simplest choice and as physically meaning, if not more, than any other method you mention.
the fact that you initially have differently heated zones does not mean that you need to keep them to be a different temperatures. so a global thermostat applied to all atom would be a most straightforward procedure to slowly cool your system. that said 10k/ps or 20k/ps is not exactly slow on a macroscopic scale. i would consider a “slow” cooling something that takes seconds or minutes.

axel.

Thank you for your reply.

1- I need the cooled down system to study the crystal structure at room temperature.

2- Same as (1). I dont want to study what happens during the cooling. Just after it.

3- it is a welding exp. … I heat the interface to make a weld joint. I’m aware of the difference between the cooling macroscopic rates and microscale rates. I want to perform cooling similar to air cooling I.e: Slow cooling. 10 or 20K/ps. The cooling rate correlation is not major issue for me. The issue is having two different regions at completely different temperatures and I need to bring the system down to equilibrium 300K.

The problem I’m facing is that I cannot specify Tstart … I thought of dividing the simulation box in two NVT regions with different Tstart but same Tstop. Therefore, one region will start to cool from 2000K to 300K slowly … say at 10K/ps and the other region at 400K will cool down to 300K and stay there for the rest of the duration of the run. But I’m not sure if that’s physically right or close.

I tried to look for NEMD simulations but most of the studies do not need to cool the NE system.

Your help is appreciated. Thank you

To add to my previous reply. I was thinking to add a sort of “heat reservoir” or “heat sink”. A layer of atom that is always kept at 300K so that it cools the sim box in NVE. However I tried that and the temperature doesnt seem to change. If you please could assist me to achieve such method if it is even possible ?

I agree with you that the NVT of two regions is no where near physical and thermostats are not suitable. Regarding the global temp. Calculation, i think the fix modify can use a temp compute specifed to that region. But the double or triple NVT ensemble in the same box is not good idea.

Is it possible to use a two layers at the sides and rescale them at 300K to act as a heat sink and cool the system ?

I meant to say that 10 or 20 k/ps is slow on the microscale and it can be correlated to the minutes or seconds cooling time in the macroscale.

Regarding the single thermostat NVT method, I just find it very crude modeling when taking the temperature average of the whole box and the box contains very high temperature gradients. The dissipative thermostats sound a great idea !

Thank you for your time and help !

I meant to say that 10 or 20 k/ps is slow on the microscale and it can be correlated to the minutes or seconds cooling time in the macroscale.

the cooling rate in K/ps is the same on any scale. you seem to be discounting that temperature is an intensive property.

Regarding the single thermostat NVT method, I just find it very crude modeling when taking the temperature average of the whole box and the box contains very high temperature gradients.

it is not “crude” it is “gentle”. the heat bath coupling with a nose-hoover thermostat is rather indirect through impacting the forces. and - obviously - fast atoms will be impacted more than slow atoms and that is exactly what you seem to be after and without - unphysically - forcing a specific cooling gradient on different parts of your system, but allowing for the temperature to dissipate inside the system.

The dissipative thermostats sound a great idea !

now that is what i would call “crude” and “brutal”

axel.