Langevin thermostat

Dear All

I have a problem using the Langevin thermostat. It is supposed to fix the temperature but it freezes the system! I thought the problem might come from inappropriate values of the arguments. So I checked lower "Tstart"s, higher "Tstop"s, and wider “damp” parameters, however, the system is still frozen.

The simulation is set in the LJ unit and the parameters are:

Could you please help me with this?

It is probably metastable.
Are you trying to simulate melting?

Are you sure that a temperature of 0.15 does not correspond to a solid phase for the LJ parameters (epsilon, sigma, pressure) your choose?

Hi, thank you for your reply.
I am trying to simulate and study glass transition and T=0.15 is below the transition temperature.

Hi Simon Gravelle

Yes. it is supposed to simulate a glass phase since the glass transition is expected to occur at about T=0.3.
My system contains a mixture of particles of two types that interact with each other via a smoothed LJ12-6 potential. The parameters are set as follows:

Thank you.

Do you have any technical description from a publication that you are following?
Simulating phase transitions in atomic scale molecular dynamics is notoriously difficult.
Phase transitions are activated processes and you have to deal with significant finite size effects causing metastable states and thus a very large hysteresis. In almost all cases you need to use a different approach than what people would do in an experiment.

I think so far there is no indication that there is an issue with LAMMPS itself, but if there is, it is not with the fix command, but elsewhere in your input. A discussion of how to correctly model phase transitions in general, is off topic for any of the LAMMPS categories. There is the “Science Talk” category for general questions about science/reserach.

thank you for your consideration.

I am going to simulate a system almost similar to the system described in the simulation method in the article:

but when I use the “nvt” ensemble, the time series of the forces acting on the particles does not change smoothly although it is expected to do (as shown in the supplementary of the article). I have been studying the role of every parameter value and command. Now It seems to me that it might be because of the ensemble. So, I am going to use the Langevin instead of the nvt. I guess it can help me to find out what causes the problem.

This may be the best clue you’ve given so far, since you haven’t provided an input deck to check your problem explicitly. Note that fix nvt is an integrator with a built-in thermostat, whereas fix langevin is only a thermostat. It is usually used in conjunction with fix nve, an integrator without a thermostat. Is that what you’re doing? Or are you only using fix langevin?

1 Like

Hi @Fatima,

I am a bit puzzled by what you write:

The time series of the forces acting on which particles? The only plot corresponding to your description in the SI (a time series, but note that it is actually plotted against U\times{}t which is a depth) represents the forces acting from the surrounding bath to a probe particle that is not thermostated (it goes through the bath at a constant velocity) and that interacts with the particles in the bath through a different potential interaction. This was never mentioned in your previous messages and I am not sure this is what something you are doing right now. Plus, the continuity of the time series is independent on the glassy state of the medium (one of the curves is in the liquid state).

This raises a few more questions from me, that are actually quite close to one another:

  • The langevin thermostat as implemented in LAMMPS acts on the forces of the particles. It would be normal that the time series of the forces on each particles is not continuous since random forces are added each timesteps. What did you expect to compute and why?
  • You mention that your system gets frozen but the paper clearly states that the glass transition temperature T_k is 0.3. If you simulated your system below this temperature, would you expect diffusion? As far as I know, glassy systems are particles frozen in a disordered metastable state. What do you mean by frozen? How different is it from a glassy system in your definition?
  • As @akohlmey mentioned, this does not look like a problem in using LAMMPS but more in defining your model. What do you expect to see from your model and simulations? How would you caracterise the glass phase of your system? From my point of view, it might be beneficial for you to spend some time figuring out the relevant informations in the paper you use as a reference and thinking about the output you expect from your simulation.

Note that I made several guesses here, that is due to the fact that it is hard to understand what you are doing and what you want to achieve from the inputs you provided so far… If you have a hard time figuring out what you want to achieve, you might find more help in a dedicated section. If you know what you want to achieve or compute and would like help in using LAMMPS to do so, then we might need a better description of your problem and your input script to provide relevant insights. Note that the paper you provided is very technical and advanced so it is not unlikely that you need time to understand everything that is going on there.


dear Michael Jacobs
Thank you for your reply lot, and thank you for mentioning this point. I had applied it in a wrong way. Now it works well and the system is not frozen anymore.

Dear Germain

I appreciate a lot for your comprehensive and detailed response. Thank you for your kind and precious suggestion.
I am sorry if my explanation was brief and I made the problem difficult to understand.
My system is similar to the system described in the mentioned article: a binary mixture of particles interacting with Kube-Anderson potential and a probe particle moving with a constant velocity and interacting with the particles of the bath via an expanded LJ potential.
As you have expressed the probe particle is not thermostated, however, applying different thermostats can influence the force acting on the probe particle. The variable which I am trying to reproduce now because the behavior of the autocorrelation function of this time series is used for characterizing the state of the system.

About the continuity of the time series, yes you are right, both liquid and glassy phases show continuous time series in the reference article. but my results show discontinued time series which means something is missing in my simulation. I guess it might be related to the thermostat.
(By freeze I mean the particles are quite fixed. Applying the hint that Michael Jacobs implied my system is not frozen anymore but the discontinuity problem still exists).
I welcome your professional suggestions and comments.

Dear @Fatima

I’d like to comment two points on the answer you provided to both @Michael_Jacobs and me:

Your system was apparently not integrated at all. This is different than a “frozen” system in which particles should move but not diffuse. This is why I asked you what you meant by “frozen”. “Quite fixed” is not very informative. I would suggest getting informed opinion on your new set of simulation to be sure everything is running as expected.

I disagree with your conclusion. The time series you referred to is the force applied from the particles to the probe along its trajectory. It should show continuity if measured at sufficiently short intervals (which I suppose you do, but cannot check) in a condensed phase, fluid or glass. If “applying a different thermostat” leads to discontinuities, and since your previous simulation were not running as expected, I can see two main reasons for that (but keep in mind there might be others): you are either applying a langevin thermostat to the probe particle or not computing the correct quantity. I would also suggest being sure that your systems are in a proper glassy state at the correct density, but this is contained in my previous remark.