I am trying to simulate a system consisting of Silica, water, and a carbon chain molecule. However, every parameter set I use gives me segmentation fault, which I assume is bad parameters (not trained for this system). Specifically, water stability on the silica surface is not there. I even tried combining different parameter sets (silica/water) + (silica+carbon), with no luck.
I appreciate if you know of any parameter set that can simulate this without segmentation faults.
Yes and no. There should not be any segmentation faults directly cause by bad parameters. They usually indicate a bug in the code.
However, ReaxFF is a particular animal and has some unusual quirks. The most prominent is a memory management that is not as robust as it is for other force fields. But there are remedies like using the KOKKOS package version of ReaxFF. Using bad parameters can accelerate this, but usually the common cause for persistent segmentation faults are a general bad choice of simulation settings or some other issues like a LAMMPS executable created by a broken compiler or an outdated LAMMPS version with known bugs.
But this is all hypothetical unless you provide us with more information and the means to reproduce the segmentation fault(s).
Thank you for your response. I made a change and it is working perfectly.
This was very helpful. For whatever reason, I observed these two behaviors which I think might be helpful to other who see the same issues:
1- when I changed the boundary conditions (from shrinked to periodic) it just suddenly started working without error. I understand that for a surface, this would remove the edge effects but I have no idea why it would create a segmentation fault?
2- when I introduced 1 nm gap from the bottom of the silica surface (to avoid interactions between the z-direction images in periodic conditions, the minimization got stuck! when I reduced that to 3 Angstroms it worked smoothly.
Again, I have no idea why these issues are caused in ReaxFF, but it might be help for someone.
It creates a segmentation fault because the domain decomposition changes massively. This is a peculiarity of the memory management of ReaxFF (and discussed here a lot) that its memory management is not suitable for cases where the environment changes a lot and the ‘guesstimated’ memory allocation block become insufficient. As discussed many times, the workaround for this is to compile and use the KOKKOS version of ReaxFF (which - contrary to common believe - does not require a GPU, but can also be compiled for serial and multi-thread only). Also, your initial box is probably far too large for the extent of your atoms, if you would adjust the box dimensions accordingly, you also would likely avoid the segfault.
Again, the secret probably lies in the domain decomposition and the cutoff of ReaxFF. But why use periodic boundaries, why not fixed ‘f’ or shrinkwrapped with minimum ‘m’ boundaries which both should avoid interactions across the z-direction boundary without the catastrophic collapse effect of ‘s’ boundaries when the box is too large.
It is worrisome that you are trying to give people advice without having understood what the root cause of the problem is. How do you know that your choice are addressing the problem for real and are not just a coincidence?
It is also problematic to do this only at this level of vague descriptions of the situation. That offers a lot of room for misinterpretation. It is usually much better to discuss specific (test) examples and those will also allow to give confirmed explanations instead of having to speculate. Please always keep in mind that the quality of responses is crucially dependent on the quality of the inquiry and yours have now repeatedly been of particularly low quality.