How to use per-atom quantity as chunck ID

LAMMPS LAMMPS General Discussion
1

Dear Lammps users

I want to treat per atom as per chunk, so I used below command to do it.

variable position atom x
compute chunk_x2 water chunk/atom v_position
compute myChunk2 water dipole/chunk chunk_x2
fix ave_T_P all ave/time 1 10 10 c_myChunk2[*] file current2.profile mode vector

However, the output only has 30 rows rather than the number of raws equal to the atom number. I do not understand why. How to to use per-atom quantity as chunck ID?

Thanks for the help.

fan

2

You need something integer, v_position is not.

What you do looks like 1-d binning. Why not use compute chunk/atom bin/1d?

3

According to the manual, chunk/atom bin/1d will divide the atoms into 1d bins, which means the number of chunk should be same as the number of bins, not the number of atoms. Or there is a trick I don’t know?

4

Using 1 chunk per atom makes no sense. You need to better explain what you are trying to do.

5

I am trying to calculate the dielectric constant of bulk water. According to this post replied by you as well
https://matsci.org/t/can-anyone-help-me-understand-why-my-dielectric-constant-calculation-is-not-leading-to-the-correct-result/33032

I am testing the results of computing the total dipole moment by summing over the dipole moments of the individual water molecules and over individual charges. For the latter, I have to use each atom as a chunk and calculate the dipole moment for each chunk, followed by sum them up. This is why need to teat per atom as a chunk.

6

Dielectric constant can be tricky to calculate, it can be safer to use postprocessing open source code that have been carefully tested, such as:

  • MDAnalysis for bulk analysis,
  • MAICoS for spatially dependent analysis (planar, cylinder, and spherical) – see also these explanations and guide to avoid falling in the common pitfalls associated with dielectric constant calculation.
7

That is needlessly complicated and makes no sense. Technically speaking, there is no dipole moment to be computed from a single point change. So that kind of calculation cannot work.

Instead the (total like any) dipole moment is easily computed from:

\vec{M} = \sum q_i\cdot\vec{r}_i

That is the sum of the charges times the unwrapped(!) positions.
For that LAMMPS already has a compute: compute dipole command — LAMMPS documentation

When I did the research in mentioned in the post that you are referring to, I wrote my own analysis tool, since none of the available tools were 100% accurate, which you would not easily notice, since people at the time didn’t do trajectories as long as I did, but that was required due to nature of the Ewald summation implementations. It is also extremely important that the correct fluctuation formula is applied. This is all extremely subtle business. I have spelled most of it out in my PhD thesis. I am happy to send you a copy, but it was written in German. :frowning:

Update: I am uploading some talk slides from a talk I gave a few years later, that summarizes the results (in English!) and also has key references included. I strongly advise to read up on this to properly understand what you are doing. Dielectric properties are highly difficult. I also strongly advise to get and study the text book “Theory of Simple Liquids” by Hansen, McDonald for better insight. I would not have figured this all out without that book. It saved my bacon many times.
talk-trieste2004.pdf (1.1 MB)

8

Hi

Thanks for the useful information and disscussion.