Viscosity Data Treatment – Seeking Suggestions

LAMMPS
1

Hi everyone,

I have a quick question regarding my viscosity data. The graph I’m getting is slightly different from the A. Michels 1957 experimental data. Does anyone have suggestions on what I could improve to better align my results with the experimental data?

Thanks in advance for any insights!
Lucas A. P.

P.S. I am running MD simulations using the EPM2 potential.

2

Hello,
This is not a LAMMPS-related issue; you may have better luck on a forum dedicated to physics or materials science.
Simon

1 Like
3

How should that be possible since you provide next to no information that an advice can be based on?
Your post essentially says: “I did some stuff and didn’t get the result I wanted. What should I change?”
How can you expect any kind of meaningful response to such a post?

Specifically:

  • you don’t say which material you are investigating?
  • you don’t say how and how well you equilibrated your system and how well your data is converged. Have you estimated the error of your data?
  • what is the error of the data you are comparing to and how large is your deviation in comparison?
  • are there any other experimental data sets to compare to?
  • you should make an assessment how well the model (styles and parameters) you are using is capable to represent the data you are comparing to?
  • you don’t say if there are other publications that can reproduce the data in question better than your results.
4

Hi,

I realize my previous post was lacking important details, and I’d like to clarify and provide more context for my issue.

I am investigating the viscosity of CO2 using molecular dynamics (MD) simulations with both the EPM2 and TraPPE-small potentials. The simulations are being run at temperatures ranging from 228 K to 500 K and pressures from 1 atm to 150 atm. I am using real units with a timestep of 1.0 fs. For the viscosity calculation, I am using a correlation length of 6000 steps, a sample interval of 4 steps, and the dump interval is 24,000 steps.

In my setup, I am using a Lennard-Jones interaction cutoff distance of 10.0 Å for the EPM2 potential and 15.0 Å for the TraPPE-small potential. Coulombic interactions are handled using the PPPM method for long-range electrostatics, and tail corrections are applied for energy and pressure due to the truncation of long-range interactions.

I’m comparing my results to the experimental data presented in A. Michels et al. (1957), The Viscosity of Carbon Dioxide Between 0°C and 75°C and at Pressures up to 2000 Atmospheres (Physica XXII). The data shows discrepancies at certain pressures and temperatures, especially near the critical region. My simulations compute the viscosity using the Green-Kubo method based on stress tensor correlations. However, I’ve noticed some deviations from the experimental data, and I’m seeking suggestions on how to refine my setup to improve the agreement with the experimental values.

Are there any specific adjustments to the Lennard-Jones or Coulombic interaction parameters, or other force field optimizations, that might help improve accuracy? Additionally, if anyone knows of other publications that have successfully reproduced these results using similar models, I would greatly appreciate the references.

Thanks in advance for any insights!
Lucas A. P.

5

There are no absolute numbers for this that will lead to converged results. You will have to determine this for yourself. Please also note that the convergence has to be confirmed for the entire temperature range. If you raise the temperature, convergence may be faster or slower depending on the properties of the material under investigation.

It is well know that systems near the critical point show large pressure and density fluctuations. This can be seen macroscopically as the system becomes opaque (I saw this myself during one of the exercises/experiments in advanced Physical Chemistry lab in college). That in turn means, that you will have to deal with massive finite size effects and convergence issues, probably beyond the point of what is reasonably possible with atomic scale simulations. I have done simulations of supercritical water a long time ago and had to deal with those issues as well. We then chose to run simulations far away from the critical point.

Any change to the parameters would be a new model and would have to be completed validated. All models are compromises and represent different properties differently well.

It seems like you expect that there is some “magic bullet” type of setting that will make your results match your expectations. This doesn’t exist. Same as experimental data, the results from a simulation are the results, assuming that you have confirmed that issues due to setup and data sampling have been minimized and there are well documented methods to quantify those as well.

Doing a thorough search of the published literature to be aware of the state of the field is part of every research project. You are not likely to find somebody that can provide that to you, since that would require that somebody would be working in the same area.

6

Thanks for the help! I’ll look into the suggestions. Again, thanks for the support!

Best,
Lucas A. P.