Problems with pressure using NVP-NVT

Dear Axel,

Dear Axel,

First of all, thank you so much for your answer.

1 - Changing Kspace to smaller values, do I guarantee that the pressure obtained in the NPT will be maintained in the NVT?

​there is no guarantee for nothing. most of the time, when things don’t work out as expected, there is more than one cause. a loose kspace convergence will result in a larger error than a tight convergence. this error​ is much larger on the pressure than on the forces. hence, when doing NPT simulations it is more important to have a tight kspace convergence than for NVT simulations.

2 - About shrinking the box (box dimensions in Angstroms) in NPT:

Lx Ly Lz
Start 50 50 120
End 49.31 49.31 118.35

NVT -

49.31 49.31 118.34

3 - I observed that the pressure (atm) is not equal in all directions:

> Pxx Pyy Pzz Pxy Pxz Pyz
NPT 83,77 85,78 163,23 1,94 -5,98
NVT 23,67 28,55 114,22 -2,13 1,25

​so - as i suspected - you should not do isotrop​ic box deforms. your system is differently compressible in different directions. it might be suitable to couple x and y, but not z.

4 - I leave an empty space between the water and the oil to vary the number of molecules of surfactants in this space. And I apply the NPT to narrow down that space and make the densities more realistic. So, instead to use 50x50x120A should I use 49x49x118 A?

​starting with a little extra space to avoid overlaps is good practice. ​but compressing a system with fix npt can be very time consuming, so it would be better to start equilibration with a combination of fix nve, fix langevin, and fix deform (and that one only adjusted to remove the extra space added in each direction), then switch to fix npt with anisotropic cell adjustments and record a time window averaging of the box dimensions with a suitably sized window, so you can better estimate the final box dimensions required.

5 - I am sure this questioning is quite simple for you, but I am starting now with simulations and my mentor knows nothing about it. Then the little that I have learned was reading a lot and through a friend who is also not very experienced. I thought the problem could be solved using some knowledge of Lammps in the script, so I asked the question here.

i have expressed the following sentiments on this mailing list many times. ​starting MD simulations​ without competent supervision is a bad idea. you seem to be doing better in this than most people i have come across, yet i would still strongly advise that you have your mentor start a collaboration with a research group that does have the suitable “residual knowledge”, since there is much more practical knowledge to doing efficient and reliable MD simulations than what you can learn from text books and publications (people typically only write about what works and not about what did not work. without the knowledge of the latter, you are doomed to repeat many mistakes that experienced groups have learned to avoid, especially on more subtle issues). this kind of knowledge is passed on from person to person and is best obtained by (temporarily) working under the supervision of suitably experienced tutors.

​axel.​