I agree that reflecting walls are not usually good models. NpT simulations are of course more realistic, if they can be implemented. I can think of three challenges in doing so,

1. Anisotropy in the system: a simple NpT MD corresponds to isotropic pressure prescribed in all directions. When you have 1D or 2D nanostructures such as CNT or graphene in the simulation box, an anisotropic pressure tensor needs to be considered and a simple barostat might not be sufficient. There are schemes such as Parrinello-Rahman dynamics that take care of these cases and can control the pressure tensor of the system differently in various directions.

2. Inhomogeneity: Consider the case that you have a CNT under tension in the simulation box, but the whole simulation is being performed in atmospheric pressure. So, you would need two different barostats, one taking care of the constant tension in the CNT, and the other, the keeping the atmospheric pressure in the whole box. Doing so might not be so straightforward.

3. Pressure transfer medium: If you want to do NpT, your simulation box needs to contain the proper means of sustaining the external pressure. For example, if you have a nanostructure in the middle of the box, with the rest being vacuum, you cannot expect a positive pressure to develop for the barostat to compensate.

Thank you so much Mr. Sadeghi

How about frozen nanotube which is infinite in one dimension? Which one is better; 1, place the nanotube in the center of box, between 2 wall ( vaccum space before the first wall and after the second wall) without doing NPT

2, or using Parrinello-Rahman,...

I want to study the interaction of FUNCTIONALIZED nanotube with protein, what is your suggestion? How can i come up with these challenges? Which one is better?

All the mentioned methods depends on your strategies used for managing your simulations and what you are looking for from the MD simulations and what kinds of interaction analysis are your purpose. It seems like you want to evaluate the biological ability of the CNT in coupling with the protein.

Yes i want to study effect of nanotube on protein folding, alpha helix structure,... . Probably i want to get the PMF too.

What I would do for this setup, assuming that you want to do the simulation in vacuum (no water), is to use a barostat in the X direction (along the CNT axis) for zero pressure and leave the L_y and L_z (box size in the Y and Z direction) fixed. Choose the box size in the Y and Z direction large enough so that the CNT and the protein don't see each other's mirror images (Larger than 2.5 x sigma of the non-bonded interaction with the largest sigma ) and apply periodic boundary conditions in all dimension.

If you have water in the box, then apply an isotropic barostat in all directions for 1 bar.

As a general note: make sure you have built the geometry and bond connectivity of the CNT such that it is actually connected to its mirror image. If you want to make it rigid, use SHAKE constraints.

It is not CNT. It's ZnS nanotube along Z axis and the nanotube in frozen in all dimensions. Is it still need to use SHAKE constraints?

Regards