self-diffusion coefficient and periodic boundary conditions

hello everyone, I am trying to determine the self-diffusion coefficient of water but end up with values far above the experimental value of D = (2.42 ± 0.08) × 10^−9 m^2 /s. Your advice is highly appreciated. I have followed the following steps;

1. Used amorphous cell module to construct a water box with 1200 molecules.

2. geometry optimized using forcite, COMPASS III forcefield, smart algorithm (also tried steepest descent), 2000 iterations, summation method for non-bond interaction; VDW-atom-based, electrostatic-Ewalds.

3. Relaxed the cell; annealing-5cycles (also tried 10 cycles), NVE. Followed by dynamics run under NVT ensemble, time step-1fs, steps-50000 (50ps). Production run with NVE, for 50ps.

4. forcite>analysis>mean square displacement.

am not sure if periodic boundary conditions are automatically applied but the molecules ended up out of the box during the dynamics. The diffusion coefficient was ranging from 4.366x10^-9m^2/s and above in all the trials I attempted. 

I then decided to constrain the oxygen atoms of the water molecules at the walls of the box and repeated the runs. Diffusion coefficient dropped to 1.055x10^-9 m^2/s.

Ps; I am a beginner with materials studio. Thank you, in advance for your help.

Answers 5
FH
FH 2021-06-10 Best Answer

Hi, 
if your cell only contains water, the result should indeed be the self-diffusion coefficient of water only. In that case, please make sure that the R^2 correlation coefficient from your linear fit (not to be confused with the actual mean-square displacement) is sufficiently high and the simulations are sufficiently long to be in the long-term limit. Second, the molecules 'leaving the cell' is a visual crutch to see how far an actual molecule went. If you want to see the periodic images all in the same cell, you might want to change the lattice display style to in-cell and back to default. 
Felix

FH
FH 2021-06-09

Hi, few questions to get you started. 
(1) The diffusion coefficient calculation via MSD only works in the long-term limit, e.g. when is proportional to time. This means that the MSD plot should only be fitted in the upper half, with the lower half being ignored. Materials Studio 2020 or later does this automatically where possible, and also prints the correlation coefficient to help you decide how good the fit is. 
(2) Are you interested in the diffusion of the entire system, or only a part of it? If it's the latter please make sure you are restricting your analysis to the correct set of atoms. There is a tutorial that discusses how to do that in detail, at Tutorials | Forcite Plus | Calculating the diffusivity of a gas in a polymer.
Felix

PA
PA 2021-06-09

Hey Felix, am using materials studio 2020 and followed the tutorial you have mentioned to get the above results. What I didn't do, however, was create sets of atoms. After the production run, I simply went to forcite>analysis>MSD. Wouldn't the resulting diffusion coefficient be self-diffusion coefficient of water? Also how can I ascertain that PBC were maintained? The molecules were leaving the box.

PA
PA 2021-06-13

hello, for the periodic boundary conditions to apply, should i create various surrounding boxes or is it assumed that the PBC are applied even when i use only one cubic box?

CM
CM 2021-06-22

The PBCs should be applied no matter which display style is used.