How to get a porous model for simulating thermal conductivity

At first, I got it by deleting atoms, such as “delete_atom random” command or I delete some atoms in the spherical region.
But when I want to get a high porosity model, for example, 90% or more, the model became strange and there is no reasonable temperature gradient in the output file, did it mean that my method was wrong?
I used NEMD to simulated thermal conductivity

Please be a bit more detailed next time.
But, as you were increasing the porosity it seems to me that the volume of the simulation box stayed the same? For a small number of atoms I doubt you will achieve what you want with NEMD, since NEMD with respect to thermal conductivity is essentially based on continuum description as opposed to Green-Kubo.

Thank you for your answer, so for a small number of atoms, what method should I choose except NEMD?

For starters, what is the number of atoms we are talking about?

I would expect that you cannot compare results with the same volume but different number of atoms. At some point, you will have finite size effects that will taint your results.
I would thus recommend to grow the volume of your simulation box as you reduce the density.

Secondly, because the transfer of kinetic energy is becoming increasingly difficult with higher porosity, you will also need to check for convergence. It is quite possible that simulations with increased porosity will require much longer trajectories to converge.
This would correlate with common sense since material used for insulation, say styropor, is characterized by its porosity which leads to the limited thermal conductivity. At some point, I would guess that the transfer of kinetic energy is reduced to that of the gas enclosed in the pores. That is the point where your model would break down.

Of course, this is all assuming that your structure will not collapse at low density. So checking the structural integrity through visualizing the trajectory is paramount.

Thank you so much, I will try as you suggest·

What I’m doing now is the initial structure of 3000 atoms, which means that if the porosity is 90%, it should be about 300 atoms, do you mean the problem is that I haven’t changed the volume of the box? But I have two more questions:

  1. If I change the volume of the box, will these holes not change, and can it represent that it is a structure with 90% pores in its initial structure

  2. In my opinion, the holes generated by deleting atoms should be similar to vacuum, and the actual situation should be filled with air in these holes, but the literature I read is also generated by this method. What’s the problem with my cognition

The fundamental problem with non-equilibrium molecular dynamics is that macroscopic gradients are very small on an MD length scale. Consider a 10K temperature difference across 1mm (very large by macroscopic standards) – if you calculate the equivalent gradient across only 10nm and try to impose it in a simulation, the resulting observable thermal flux will be statistically zero.

So NEMD intrinsically uses incredibly large gradients and the onus is on you, the simulator, to convince your reviewers that your results are really applicable to real materials and not just computer numbers. That’s why everyone here is worried on your behalf over box size and so on. In particular you need to show your reviewers that your result is not just an artifact of simulating very few particles in an effectively crystalline periodic repeat. If you get different results for a 10nm box, a 20nm box and a 30nm box, then neither of these results are really applicable to a macroscopic material of mm or cm length.

In any case, have you tried visualising your configurations? When you are deleting 90% of atoms I would expect that, at least sometimes, you end up with disconnected fragments that simply can’t transfer heat or energy to each other.

Thank you for your patience. The problem I’m facing now is that the porosity of the material I need to simulate is very high. After deleting too many atoms, I will indeed encounter the disconnected fragments you said, which leads to the failure of thermal conductivity simulation. How can I solve it, such as building a larger box and more atoms?

You may find the replicate function useful to make bigger versions of your current box.

Beyond that I can only recommend the usual scientific methods: search the literature for someone else who has done something similar and gotten convincing results, and see what you can learn from them. I imagine whatever physical process creates 90% voids in your real-life material does not turn it into crumbly dust, and there is some physical reason for that. If you know the physical reason, you can try to apply that to whatever method you use to generate voids; if you don’t know the physical reason, do you really know enough about the material to run a successful molecular model of it in the first place?

Highly porous materials will have to have some specific structures, e.g. they are built from a foam. You won’t be able to build a foam from randomly deleting atoms. You will also have to know the distribution of the foam bubble diameters. You would not be able to build such a construct from a few thousand atoms. As already alluded to, you may need to set up structures much larger than you expect and then run simulation for much longer than you think right now and at some point, you just have to factor in what those foam bubbles are filled with, because that may just be as important for transferring kinetic energy than the material of the foam itself. You cannot solve these issues by asking for some LAMMPS input commands.

If there was a “correct_my_simulation” command (or better yet a “write_my_paper” command), a lot of problems of LAMMPS users would be solved. Alas, those don’t exist so we have to do with doing proper research work and that means, first to study the methodology and learn from the work of others before even considering running simulations. This is obviously a step you have skipped or abbreviated and thus have to revisit.

I will work hard in these directions. Thank you very much for your suggestions

I think I understand what you mean, and I also find my shortcomings in my work. Thank you very much for your patient guidance