I am currently simulating the lubrication behavior of water molecules between two frictional surfaces. I have encountered an issue with the kspace_style pppm, which requires periodic boundary conditions in all three directions (xyz). However, I do not expect (and indeed, it should not happen) any periodic forces acting on the z-direction.
Is there a way to handle this situation where I do not want periodic forces in the z-direction? Using the kspace_modify slab nozforce option disables all kspace forces in the z-direction, including those within the water molecule group, which is not desirable.
Any suggestions or insights on how to appropriately handle this situation would be greatly appreciated.
Thank you very much for your response. I believe you are referring to setting “kspace_modify slab volfactor,” but I may have some confusion about the description of this command in the manual. I would like to know if all atoms in a simulation box are divided into a single or multiple “atom slabs”? If it is the latter case, assuming the atoms are divided into 1 to n layers, are the inter-slab interactions between the m-th layer in the second period and all layers in the first period eliminated, or only the interactions between the m-th layer in the second period and the m-th layer in the first period are removed? And are the removed “dipole inter-slab interactions” only the main component of the inter-slab interactions? In other words, when volfactor is set to the recommended value of 3, is the periodic component of the kspace force in the z-direction eliminated to a negligible extent, but not completely?
I understand that the documentation provides references to relevant literature, but unfortunately, I currently do not have access to them. Therefore, I would be extremely grateful if you could provide a brief explanation or any additional insights to help me better understand the concept.
Thank you very much in advance for your assistance. I truly appreciate your support!
The system is treated as a slab. This is quite evident from the documentation.
This is some conjecture that has no support in any form from the documentation.
How can you be expected to do research, if you do not have access to literature?
That said, the slab correction to lattice sums (or their grid based equivalents) is a standard technique used in molecular computer simulations for more than 20 years. You cannot tell me that you have not access to any relevant literature about lattice sums and poisson solvers etc.
The method used in LAMMPS (as it is also explained in the documentation) does the following.
it enlarges the box by the “volfactor” multiplier in z direction (this the minimum requirement for the poisson solver mentioned below)
it then computes the lattice sum as usual with full 3d (lattice sums require that).
it also computes the net dipole of your system
then then solves the poisson equation for dipole-dipole interactions between periodic images of the system. This will not converge properly unless the vacuum between two periodic images is not double the space than the extension of the system in z-direction
it uses the solution from the above step to cancel the dipole-dipole interactions in z-direction
Since proper kspace handling already requires a system with a net zero charge, you cannot have charge-dipole interactions between periodic images. Dipole-dipole is removed, so only dipole-quadrupole and higher terms are present between periodic images. So you are going from 1/r for plain Coulomb to 1/r^4 as the decay of the remaining interactions between the periodic images. This is very, very small, especially since the distance is also extended.
If you want to absolutely have no interactions between periodic images you either have to switch to use a cutoff coulomb (or some damped cutoff coulomb like coul/wolf or coul/dsf with a very long cutoff and thus bad performance) or use multi-level summation (MSM), which is also either very slow or not very accurate.
If you would survey the relevant literature of simulations of systems with a slab geometry you will see that most use a regular (3d) kspace solver (ewald, pme, spme, pppm) with extended box dimensions and some do this even without applying a poisson solver to remove the dipole-dipole interactions.
If you have doubts about the accuracy or validity, I suggest you make tests. Altogether, I suspect you are worrying about a fly in the soup while ignoring the rat that is also swimming in it (i.e. the systematic errors and approximations of the force field as such and, e.g. using a cutoff on all LJ interactions).