ZBL and hybrid/overlay

Hello,

I am using a hybrid/overlay potential (pace + table) for my simulation. I want to simulate irradiation in the system so I figured out I should use ZBL for short-range.
In order to smoothly link the ZBL and the pace+table, I can’t find how to do it. Indeed, I understand how to do it in the case of a pairwise potential where I could plot energy as a function of distance for my potential and for ZBL and then just use a smooth function and tabulate it. But with my potential that depends on the atomic environment I don’t know how to do. Moreover, I tried to visualize the energy as a function of distance for two atoms by using the command pair_write but it is also not working.
Any suggestion on how to link ZBL and pace+table smoothly? And how to visualize energy as a function of distance for two atoms only?

Hi, I can show how I solved the problem of getting info on energy(distance) function:

variable dx equal "0.05"
create_atoms 1 single 0 0 0 
create_atoms 2 single ${dx} 0 0 

group move type 2

# atom mass and potentials

timestep 1

fix write all print 1 "$(time*v_dx+v_dx) $(pe)" file zbl.txt screen yes
fix move move move linear ${dx} 0 0

run 60
1 Like

I am confused why you would need to use ZBL with a machine learning derived potential. Shouldn’t that one have a more realistic physical behavior for close contacts than ZBL does?
This is assuming that it was trained with a sufficient number of such close contact configurations, of course.

Moreover, ML-PACE is a manybody potential, so you cannot just “fix it” by splining ZBL into a interaction between two isolated atoms. The interaction between those two atoms would be different if there are other atoms present and thus your table would be wrong as well.

Thanks for the answer. I was wondering that since from reading the ML-PACE paper for carbon, I think they don’t have data points below 1 Angstrom. Then, I’m not sure how to assess how ‘good’ the potential behaves at very close distance.

We often use SNAP with ZBL to avoid the “black hole” effect, i.e. clustering when a close contact or overlap occurs, since the ML potential may not be trained on strong overlaps. You can see an example here: lammps/potentials/W_2940_2017_2.snap at 16b19c71c1b7f8e1bad54d26039bd3098528d24e · lammps/lammps · GitHub

2 Likes