Deforming epoxy network in different ensambles

Hello dear all,
I try to deform an epoxy network in the x direction at 300K. My system consists of ~60k atoms and I use the DreidingX6 force field.

In the literature I have found many ways in terms of ensamble to do it like deforming in the x direction with fix deform and keeping the other two dimentions fixed (fix nvt) or applying atmoshperic pressure in the other two directions (fix # all npt temp 298 298 100 y 1.0 1.0 1000 z 1.0 1.0 1000 couple none) or also applying 0 pressure in the other two dimentions (fix # all npt temp 298 298 100 y 0 0 1000 z 0 0 1000 couple none)

All these aproaches can be found in the literature and all give different results. I want to ask what is the most proper way to deform such a system.

Thank you in advance,

The way a sample is deformed depends on what you want to do (for example, a stress-strain curve or a plastic deformation to match the box vectors of two different samples).

There is not much difference between these two fixes:

fix 1 all npt temp 298 298 100 y 1.0 1.0 1000 z 1.0 1.0 1000 couple none
fix 1 all npt temp 298 298 100 y 0 0 1000 z 0 0 1000 couple none

as the variance on the average pressure will likely be much bigger than the target pressure.

@hothello Thanks for your kind response!

I try to apply high deformations up until failure point (~ 200% strain). I want to plot the stress - strain curve. I have seen works that use 1 atm , others that use 0 atm (which hase some difference on the transverse deformation like ly and lz due to the applied pressure) and others use NVT (so that ly and lz remain fixed). The one with NVT has really different results due to the constant lengths on the y and z directions. My question is which is the correct way…

From memory, the mechanical properties should be computed in the small deformation regime (<10% of equilibrium values).
You should first equilibrate the system in the NPT ensemble and then apply the deformation with 0 pressure in the y and z directions. I don’t have a reference to suggest, but I encourage you to verify this protocol in a textbook and share the citation here.

This is what I am trying to do. Compute fracture properties as this ref suggests:

I have well equilibrated my structure in NPT @ 1atm and 300K for 100ns. My question is on what ensemble will the deformation ocure. In the above ref. the ly and lz are held fixed while for example in this lammps tutorial (Simulation of Deformation Behavior in Amorphous Polymer - LAMMPS Tube) have pressure with the command

fix   		   2 all npt temp 298 298 $(100*dt) y 0.0 0.0 $(1000*dt) z 0.0 0.0 $(1000*dt) couple none

The method suggested on LAMMPS Tube is analogous to the one I remembered. The only difference is the use of instant variables to compute the coupling constants.

In the paper you cited, the authors are studying cracks and it (may) makes sense to use an NVT integrator. In your case, I would allow the relaxation along the other axes. I found a citation for you, from: Shaofan Li, Jun Li - Introduction to Computational Nanomechanics - Multiscale and Statistical Simulations (2023, Cambridge University Press):

Uniaxial compressive strains are applied on the optimized C-S-H structure along the c-axis, while allowing full structure relaxation of the other five strain components and requiring the residual stresses after relaxation are less than 0.1 GPa. At each compression deformation step, a small increment (a 0.02 compressive strain) is applied sequentially to the relaxed structure in the previous step.

If your sample is isotropic and the simulation box is orthorhombic, then just relax the y and z components. If the cell is triclinic, you may also want to include the off-diagonal components.

I want to study cracks too. So can you please explain your opinion on why NVT may be a better option?

Hey, I am not your tutor, nor do I have any interest in studying crack propagation. I have provided a reference to a textbook on multiscale simulations: do your homework as this is no longer a LAMMPS-related issue.

1 Like

On what concerns this question relative to NPT, I think this paper by Germain (@Germain) et al can give you good insights also: ^^

1 Like

If I am not mistaking, @Iakovos_Delasoudas already got in touch with me with regard to both NVT stress fluctuations and computing elastic properties of their network. I am a bit busy at the moment so gave a brief bit (IMHO) sufficient answer by email.

There is also a large litterature on how to strain polymer/epoxy systems and compute cracks propagation, elastic properties and mechanical properties using MD. That’s where you should be looking at. For example, digging the work of Doros Theodorou’s group might be of benefit on this kind of topic.

1 Like