I have a quick question regarding calculation of local heat flux. To provide some content, we study a solid Argon using the example input script from the link (compute heat/flux command — LAMMPS documentation). However, we scale up the model with a larger region box which is 0 40 0 40 0 40. This entire cubic model is equally divided into 1000 smaller cubes.
Would it be feasible to compute the local heat flux for each of these smaller cubic regions? If so, how can we modify the script? Any advice will be greatly appreciated. Thank you.
There is a difference between technically possible and practical.
This would mean creating an input with thousands of similar commands to track the flux between different adjacent cubes. Are you up for something like that?
Moreover, how large would each cube be (and how large the entire system)? With the given example it is already difficult to get converged results, if you have just a few atoms in each region, this is next to impossible to do.
I don’t know what you want to achieve by the “cubing”, but you need to reconsider whether this is suitable to give meaningful results. Perhaps some 10s of cubes can do as well?
Thank you for the reply. Actually, I am going to simulate a nanostructured material with a large scale (more than 10K atoms). The size cannot be further be reduced based on symmetry. The goal is to obtain spatial distribution of heat flux. Since the model is large, it will be run on a supercomputer.
I simply use the example script to raise my question. I notice that the total number of groups should be less than 32. Therefore, I would like to know how we can compute heat flux for more than 32 groups. Thank you in advance for your help.
Supercomputers (i.e. special purpose parallel computers that would rank high on the https://top500.org/ website), require much larger systems (at least 3-4 orders of magnitude) to be properly utilized. A 10k atom system would run well on a single multi-core CPU and not scale well beyond ~ 20 processes.
Regardless of groups or chunks, I don’t think you are properly appreciating how many atoms you need and how long you need to run to get converged results.