At the moment I’m working on SiO2/Si interfaces. I was running a 0K minimization in LAMMPS (6Dec12 version) using slab boundary conditions. I am trying to find the equilibrium offset position of the SiO2 slab with respect to the Si one, by moving the former slab along the x and y axes and minimizing.
Here’s how the potential energy surface for such a calculation I get before the minimization (initial energy at each point on the grid):
Here’s how the potential energy surface for such a calculation I get after minimizing at each point:
I am using this as a template input file for my calculations (where I just change the file names for the data and the dump file):
Here is a sample data. file (they are automatically generated by a script):
I was wondering why do I keep getting such noise in the PES, given that my steps size for a calculation is only 0.05 Angstroem, but I do get differences larger that 100eV between neighboring cells. Would you have any suggestions for some terms I could tweak in the ffield.comb file to get a smoother PES?
School of Physics
Trinity College Dublin
Did you make sure that your minimization always converged? You have very few steps in your energy criterion. Something to start with…
If I do a grep for ‘Stopping criterion’ they all seem to give “Stopping criterion = linesearch alpha is zero”. Also most of the minimizations I checked ended under 25 iterations (from the max 200 in my script) and the number of force evaluations seem to be under 500. Am I missing something?
That means they are not converging at all. I would recommend trying one step by hand, i.e. your two slabs separated a certain distance and then run minimization on it. You need to understand what the problem is coming from. Maybe you need a different minimizer, or tune a bit the minimizer options or there might be a problem with your system.
Also pick more meaningful convergence criteria. You could set the energy one to 0.0 and then chose
something more meaningful for the force. Start at much higher values and lower them as needed.
Those mega-small force tolerances do nothing but slow down your analysis. I don’t think you can even reach such values with certain reactive parametrizations of Reaxff for example.
You have very thin slabs, so I’d suggest you increase the slab thicknesses and/or fix the top and bottom most layer of atoms.