I am trying to conduct a coarse-grained agglomeration simulation using, for example, the LJ potential, with 16000 beads in the system. However, as the simulation proceeded, I observed that many particles were ‘frozen’ (with a velocity of ~1e-16 m/s), which is unphysical. The issue proceeds when I decrease the cutoff distance or use other types of potentials (e.g., tabulated and Buckingham potentials). The ‘frozen’ particles become less if the magnitude of the potential well decreases but a relatively deep potential well is required in my case.
I wonder if this is a software issue or can we do the coarse-grained simulation in lammps only using pair_style colloid?
I have attached the input file, data file, dump and veldump files from my simulation. Any help would be appreciated.
You are using “fix temp/rescale” and expect meaningful results?
Does your system include some implicit solvent?
Have a look at the log file. How many average neighbors do you have?
Since your masses, diameters, cutoffs and so on are all very unusual compared to the
default settings assuming an atomic or molecular system, you need to make some adjustments there, e.g. for the neighbor list skin, timestep etc. For example, given the particle mass your system should be capable of running easily with a 100fs timestep, probably even 200fs or 300fs or more.
Which means that for the most part you have a system behaving like a dilute gas where items just keep moving with their set velocity until they occasionally get within the cutoff of another item and have some interaction with them.
The thermostat will try to maintain the total temperature but has very little chance to do so: fix nvt modifies the forces, but when atoms have no interactions and thus no forces, it can do very little. fix temp/rescale (and similarly fix temp/berendsen) will rescale the velocities, but that will rescale faster atoms more and thus at best increase the difference between fast and slow atoms, especially when you have a close encounter between two objects and thus may have some kind of “slingshot” move that makes the one object much faster at the expense of everything else.
It is difficult to discuss this further without knowing what you are trying to model, but what you get out can be very much rationalized from the geometry and settings that you are using. So either you are incorrectly expecting a condensed fluid like behavior, or your model is bogus and not correctly representing what you want to model.
Thanks, Axel, for the detailed explanations. What I’m trying to model is the coagulation of nanoscale particles in vacuum at 500 K with a solid volume fraction of 0.001. Eventually, agglomerates will form by physical bonding of the particles due to the interparticle interactions.
I found that if I use fix nve alone, the particle frozen issue disappears but the temperature keeps increasing a lot. As long as I make the temperature remain constant (e.g., using fix nvt, fix temp/csvr, fix nvk), many particles will be “frozen” during the simulation. Does this mean that the kinetic energy causes some particles to move at a higher velocity but some have almost zero velocity to maintain the temperature constant at 500 K?
These are all questions about your model and not about LAMMPS.
As I already mentioned, you have a very dilute system, where particles rarely couple to each other. You have to ask yourself: what sense does a thermostat make, which is supposed to represent the coupling to a larger system of the same material beyond the box?
What you are observing makes perfect sense for the kind of system you have with the kind of parameters you are using. Whether this is meaningful is not a LAMMPS problem, but something you have to figure out. LAMMPS will execute whatever settings and models you will feed it. It will follow the physics that has been programmed into it. I don’t see any indication that there is a bug in LAMMPS itself.
That makes it a science problem, and thus off-topic for this category. Whether your model is a meaningful model with the parameters and settings you chose is a topic for a discussion with those that do care about the science of your research, like your adviser or supervisor or colleagues and collaborators.
Do remember that molecular dynamics models atoms as if they do not have internal degrees of freedom. Their temperature is purely a function of translational and rotational velocity.
If your interparticle interaction is attractive and agglomerates are stable, then the agglomeration process must convert significant amounts of potential energy into thermal energy, and doing so for a system of structureless particles (which effectively have no heat capacity) would indeed result in all kinds of trouble. I would imagine that in your physical system this energy instead thermalizes into the internal heat capacity of your nanoparticles.
This is still not an exact guide to how to do your systems – but hopefully it gives you some hints.