Confined polymer melts experience considerable drift.

Dear Lammps users:

Im performing MD simulations on confined PE slabs by graphite walls and im experiencing some drift related problems. I would like to have your advise on whether the “problem” i experience is artifact related or if its normal.

A quick review of my system

*My system consists for PE chains with 250 atoms per chain.
*PE is modeled using the TraPPE united atom potentials and the long range interactions are evaluated using the eward summation method.

*The dimensions of the box are large enough (4 times the chains radius of gyration).
*Used the NVT integrator
*Boundary condition p p f
*Temperature: 450 K
*The system was relaxed for a long time while constraining the films momentum

Ive tried to model the graphite walls using many different approaches:
*Mansfield-theodorou potential for infinite graphite.
*Hermite interpolation method.
*Real graphite atoms: tried to use rigit bilayer graphite and flexible bilayer graphite using the lcbop potential. In this case i constrained the momentum of the walls using the fix momentuum command at those groups (up and down walls)

Problem
The following image that displays a top down snapshots of my system every 0.4 ns.

image.png

(dropbox link to the image in case you cant view it)
https://www.dropbox.com/s/97qwaqj18qrnvt2/PE_Drift.gif?dl=0

As you can see my system experiences considerable drift, hence i would like your opinion whether this is normal or this is an artifact related effect.

It should be noted that if the wall is smooth (eg. using the fix wall/lj commands) there is no drift (though thats to be expected since the interaction of the wall corrugations with the polymer generate that drift.)

Here are some checks i did (and failed!) in order to isolate the problem.
*NVE integrator
*Only short range interactions (LJ/CUT)
*I tried disabling the angle, and dihedral potentials.
*I tried using p p p boundary conditions with a single layer graphite (without any constrains) at one face of the box. The graphene sheet was drifting as usual.
*I tried increasing the polymer/graphite interaction by increasing the values of the lennard jones well (ε). Here when the interaction is 7 times larger the atoms sticked to the walls and no drift was observed.

  • Additionally i checked whether the momentum is conserved by dumping the atom velocities when using flexible graphene without any restrictions. It did. However when using rigit graphite the momentum wouldnt conserve though that normal i suppose.

A possible solution
I know that a possible solution would be to artificially conserve the momentum of the polymer melt. Indeed that works. However i would like to perform couette flow on that system (or systems will shorter PE chains since here the slip would be very very large) using a profile biased thermostat hence, i would like to avoid artificially conserving the momentum.
I did some test runs with couette flow on hexadecane and it seems that by artificially conserving the momenta the flow is suppressed even though the operation is symmetrical.

Im sorry for this long post though i tried describe the effect in detail.

Thanks a lot for any replies,
All best wishes!

A. Sgouros

Dear Lammps users:

Im performing MD simulations on confined PE slabs by graphite walls and im experiencing some drift related problems. I would like to have your advise on whether the "problem" i experience is artifact related or if its normal.

A quick review of my system
*My system consists for PE chains with 250 atoms per chain.
*PE is modeled using the TraPPE united atom potentials and the long range interactions are evaluated using the eward summation method.
*The dimensions of the box are large enough (4 times the chains radius of gyration).
*Used the NVT integrator
*Boundary condition p p f
*Temperature: 450 K
*The system was relaxed for a long time while constraining the films momentum

Ive tried to model the graphite walls using many different approaches:
*Mansfield-theodorou potential for infinite graphite.
*Hermite interpolation method.
*Real graphite atoms: tried to use rigit bilayer graphite and flexible bilayer graphite using the lcbop potential. In this case i constrained the momentum of the walls using the fix momentuum command at those groups (up and down walls)

Problem
The following image that displays a top down snapshots of my system every 0.4 ns.

(dropbox link to the image in case you cant view it)
https://www.dropbox.com/s/97qwaqj18qrnvt2/PE_Drift.gif?dl=0

As you can see my system experiences considerable drift, hence i would like your opinion whether this is normal or this is an artifact related effect.

It should be noted that if the wall is smooth (eg. using the fix wall/lj commands) there is no drift (though thats to be expected since the interaction of the wall corrugations with the polymer generate that drift.)

Here are some checks i did (and failed!) in order to isolate the problem.
*NVE integrator
*Only short range interactions (LJ/CUT)
*I tried disabling the angle, and dihedral potentials.
*I tried using p p p boundary conditions with a single layer graphite (without any constrains) at one face of the box. The graphene sheet was drifting as usual.
*I tried increasing the polymer/graphite interaction by increasing the values of the lennard jones well (ε). Here when the interaction is 7 times larger the atoms sticked to the walls and no drift was observed.
* Additionally i checked whether the momentum is conserved by dumping the atom velocities when using flexible graphene without any restrictions. It did. However when using rigit graphite the momentum wouldnt conserve though that normal i suppose.

what i am missing here is a test checking for the impact of the timestep.

also, i'd like to state that what your image shows is not what i would
describe as a drift, but more as a creeping, somewhat randomized
motion, not unlike a particle in an implicit solvent.

axel.

Dear Axel, lammps users:

Thanks lot for your reply!

Regarding the checks on the timestep ive performed such simulations as well, using a timestep that was 10 times smaller (0.1 fs). However the “drift” persists.

What troubles me, is the fact the whole system wanders around randomly with a diffusion coefficient that is much larger than the diffusion coefficient of a single chain relative to the systems center of mass.

I have calculated the diffusion coefficient of the single chains while conserving the momentum of the melt and i found it to be in excellent agreement with other theoretical and experimental works from the literature.

Also i should mention that the snapshots on my previews mail where taken in the case of rigit walls. If the walls (bilayer graphene) are flexible without any constrains the displacement of the melt is smaller and the walls are displaced in the opposite direction hence conserving the momentum of the system.

I guess the effect could be attributed on the weak lj interaction between polyethylene (pe) and graphite that results on an extremely slippery interface:

εpe = 0.0934, σpe = 3.93, εpe-graph = 0.071279, σpe-graph = 3.675,

in real units.

Finally i did some test simulations with couette flow on hexadecane and i got some very interesting results.
The shear rate was the one used by J. Non-Newtonian Fluid Mech., 77 (1998) 53–78.

When i set εpe-graph equal to 0.5 the response seems “normal”, and the velocity profile was the expected one (atoms near the walls had the same speed with the walls).

However, when i runned the exact same script while reducing εpe-graph equal to 0.07 (which is the “correct” one) i got some very counter intuitive results…
The velocity of the atoms near the walls points on the opposite direction than the velocity of the walls!!

You may take a look at the attached trajectory files if you want (dropbox links), that can be viewed on vmd since they have the .lammpstrj format. Ive also attached a script as well.

https://www.dropbox.com/s/x9erw3fvv43cfyc/e0.5_vel0.0005.lammpstrj?dl=1

https://www.dropbox.com/s/5jpdaiqgucbfern/e0.07_vel0.0005.lammpstrj?dl=1

Again, i would like greatly thank you for replying,
A. Sgouros

input.in (2.37 KB)