[gmx-users] Proper ensemble for lateral diffusion coefficient

[Gschwend Grégoire]

Hello Everyone,


I am running simmulations in order to study the lateral diffusion coefficient of small molecules (eg. pyrene carboxylic acid) adsorbed at liquid interfaces. The system is made of a liquid phase in contact with various organic solvents, and tens of molecules are adsorbed at the interface (I am using PBC). I was wondering if working in the NPT ensemble could introduce errors in the simulations because of the inerent fluctuations of the interface size. Thus, what could be the proper way to carry out these simulations? First a short NPT equilibration followed by a longer NVT? Or just an NVT, or even NVE as recommended in Frenkel&Smit ?


Also, as I am using PBC, my system has two interfaces. Should I adsorb the molecules at both interfaces or just one? Does it matter?


Finally, I have noticed that the MSD of the adsorbed molecules was always significantly fluctuating and decreasing at the end of the simulations, no matter what the time window is in gmx msd. For instance, in a 200 ns simulation the MSD is decreasing in the last ~20 ns, but os rather straight before. But if I reduce the time window to 0 - 150 ns (using however the same trajectory), the MSD is again decreasing in the last ~20 ns. I read this post in the mailing list that somehow adressed the same question:

https://mailman-1.sys.kth.se/pipermail/gromacs.org_gmx-users/2010-July/052513.html

but, in my case, varying -trestart from 10 to 1000 had no influence on the shape of the MSD. At large -trestart values the curve was less smooth, but  displayed however the exact same shape. Could these fluctuations of the MSD be due to possible errors introduced by working in the NPT ensemble?


Thank you for your help and suggestions,


Regards,


Greg