Hello, I am new to GULP and molecular dynamics in general, but I have been working on creating a VDOS plot for amorphous silicon.
For context, I ran an MD simulation of 2744 a-Si atoms in LAMMPS, from which I exported their coordinates and pasted into a .gin file. I’ve been trying to output raw frequencies to create a plot in MATLAB, but I’ve been having a hard time finding the correct potential numbers. I either get a matrix of all zeros or a shape that looks almost correct but has the wrong area and frequency range. I would appreciate any guidance.
I’m unable to attach the file since I am a new member, but this is essentially what my code looks like:
Hi Jared,
I’m afraid without the coordinates I can’t see what you’re seeing. With the partial coordinates, none of the atoms are within the cutoff of each other and so I get zero for everything (which is correct). If I add 2 dummy atoms within cutoff then everything works.
One thing to note is that you don’t need to specify “inter” on the potential input line since you don’t have any molecules in the system to be “intermolecular” with.
Regards,
Julian
I am working with Jared on the same project. We keep getting negative/imaginary frequencies with the script Jared originally posted. Because we are new users, we cannot upload the .gin file, but here is a google drive link that contains the file. We’ve made a lot of progress in our work, but are stuck on this last part.
I’ve just visualised your structure and it looks like your amorphous nanoparticle is about 1/3 of the size of the cell. Is this intended to be a nanoparticle or a periodic amorphous material? If I run a quick optimisation with conjugate gradients (due to the number of atoms) and setting “spatial” as a keyword and “rcspat 3.8” to invoke domain decomposition to make things faster, then the initial energy of ~4200 eV drops to more sensible negative values below -350 eV after about 40 steps. This means your initial structure is horribly unstable and would explain why the frequency spectra is a mess. If you optimise to the nearest local minimum before computing the phonons then you have a chance that all frequencies will be real.
Thank you for the feedback! We had the incorrect dimensions for our cell, and it is supposed to be a periodic amorphous material. When we re-ran it with the proper stillinger-webber optentials and cell size, we now have 113 negative frequencies out of 8232 total frequencies. I am trying to relax the structure in GULP, but I am having some issues. I am not familiar with the optimization features on GULP. What is rscpat 3.8, and how would I implement it conjugate gradient to lower the initial energy? Thank you again for your help, we really appreciate it.
If you every need to find out about options for GULP then the help.txt that comes with the code (or the web version) should hopefully explain them a bit. In the case of “rcspatial”, this controls the domain size for domain decomposition when using this algorithm (which is faster when unit cells are large compared with the potential cutoff, which is the case for your system).
No need to implement conjugate gradients since it’s already implemented. Just add the “conjugate” keyword. This (or “unit”) will certainly be better for the initial relaxation for large systems as it avoids the cost of matrix inversion for 2nd derivatives. However, later on you may need 2nd derivatives and even “rfo” to get a disordered system to have zero imaginary phonons.