Random point defects spike in consecutive PKA Cascade Simulation

LAMMPS LAMMPS General Discussion
1

Hello,
I am having problem with cascade simulation. I am trying to do consecutive cascade to observe cumulative damage progression in FeNiCr FCC alloy. I’m here to seek help from experts. The methodology I’m following -

  1. The system is relaxed for 100ps via NPT ensemble and a snapshot is taken of the
    relaxed system.
  2. A random atom in the system is selected as PKA and the entire cell is shifted to move the PKA to the center to avoid the damage cascade to reach the system boundary.
  3. 5keV kinetic energy is given to the PKA in a random direction.
  4. During the cascade, the boundary layers of the cell (3 atom layers) are kept in NVT ensemble (Tdamp=0.1ps) to cool down the system back to 300K at the end of each cascade and the interior of the system is kept in NVE ensemble.
  5. During the cascade adaptive timestep is used with parameters Tmin=1e-7, Tmax=0.001ps, Xmax=0.05A and Emax=125eV.
  6. Electron stopping is applied for the atoms with energy greater than 10eV.
  7. I am running each cascade for around 33ps enough to cool down the cell to 300K before the next cascade starts.
  8. After each cascade run the atoms were shifted back over the periodic boundaries to keep the cell cantered at the origin in order to allow analysis, which uses as a reference the original location of the box.
  9. A snapshot of the system is taken as dump as each cascade finishes.
  10. Then steps from 2 to 9 are again repeated and this is done 300 times for 300 consecutive cascades.

The system size is 30ax30ax30a (108000 atoms) enough to contain the cascade within the cell boundaries. ZBL potential is used via hybrid overlay with eam/alloy potential from Bonny et al.

I am using Ovito’s Wigner Seitz analysis modifier the identify the point defects in each cascades using the relaxed system snapshot of the 1st step as reference. According to literature, the defect number should increase linearly at first for 30-35 cascades with a very steep slop. Then it should increase slowly. But I am getting random defect spikes in dump files like 6, 13, 20, 26, 30, 140, 16648, 15158, 7444, 1360, 723, 1593, 222, 82, 70.

To investigate further, I have taken dumps during a cascade. Plotting the defect vs time graph for each cascades gives the expected trend (rapid increase of defects at a very short time, then it reaches to a peak value, then recombination starts and the defects starts to fall and at the end some surviving defects remain).

The problem happens when I shift the cell back to it’s original position. After shifting it back, when compared to the initial relaxed system using Wigner Seitz modifier, the defect shows a spike.

I have tried to use Berendsen temperature controlling instead of using NVT, have ran each cascades for longer times, used different dt/reset parameters, different temperature damping parameters. I even tried relaxing the system using both NVT and NPT ensemble after each cascades. All showed the same problem.

Has anyone faced this problem before? Any kind of advice from you guys would help a lot.

Thank you,
Tawseef

3

Dear colleague,

Hello! I’m writing to let you know that I’ve encountered the same issue as you did. I was wondering if you’ve managed to solve it yet.

I’ve tried using the “fix recenter INIT INIT INIT” to address this problem, but unfortunately, it didn’t work. The number of point defects in my simulation still fluctuates. However, what’s really strange is that when I conduct analyses using the Common Neighbor Analysis (CNA) and dislocation analysis methods, everything seems to be normal.

I’m really confused about this situation and would appreciate it if you could share your experience or any insights you might have regarding this problem. Looking forward to your reply.

Best regards

4

Hi,

This is rather an OVITO issue. CNA and Dislocation analyses do not use a reference structure. On the other hand WS analysis does. It works by creating a “voronoi” tesselation around the particles which are called “sites” and counts the particles present in each site, returning the occupancy. Beter read Wigner-Seitz defect analysis — OVITO User Manual 3.11.3 documentation

If at the end of your shifts the reference and the frame you are analyzed are shifted respect to each other then it is normal that many of the particles fall in the wrong site (even though there are not point defects) and therefore you get many more defects than there are. To debug that you just need to be sure that you are comparing matching cells and not pears and apples. Alternatively, if you are not interested in the defect type use CNA.

5

(post deleted by author)

6

Hi,
I generally understand what you mean. However, this phenomenon does not occur in every frame. It is fluctuating. For instance, in this cascade event, I ran 1000 loops and output one frame each time. It is highly likely that the first 20 frames are all normal, with the distribution of point defects meeting expectations. But when it comes to the 21st frame (that is, the 21st loop), the number of point defects suddenly jumps to tens of thousands, which is abnormal. In the cascade simulation, displacement occurs in each loop. According to what you said, the abnormal increase in defects should have emerged from the very first displacement, rather than suddenly showing up after multiple loops (it seems like a random event). You can check my input file
pka1.in (3.5 KB)

7

This is just a wild guess: does it make a difference if you turn off LAMMPS’ atom sorting with

atom_modify 0 1.0

?