LJ parameter optimization

Hello Dr. Gale,

I am trying to fit a LJ potential with coulomb charges for Li-Al-O system. While doing so in GULP, I am facing the following problems,

  1. I assume the default setting of Lennard 12 6 is “atomab” type. When I use the atomab type setting for Li-Al-O system after converting epsilon and sigma to A and B by using 4epsilonsigma^12 and 4epsilonsigma^6 respectively, the system stabilizes but drastically changes values of A and B. Whereas, when I use the “epsilon” setting and put in the corresponding values of epsilon and sigma directly , the system shows that the cell parameter has fallen below allowable limit. I checked my conversions of A,B and epsilon, sigma and they look correct. So my question is why am I getting different results by using A, B or epsilon and sigma. I have to use epsilon and sigma setting because, if I use A and B then it is not possible to obtain the epsilon and sigma from A_ij and B_ij when i is not equal to J. There are 2 equations (A and B) but 4 variables (sigma_i sigma_j, epsilon_i and epsilon_j). Also after one of two iterations of optimization with epsilon setting, some epsilon values become negative. Is this possible realistically?

2.I have taken the LJ parameters for Li-Li, Al-Al and O-O from 3 different papers and I have used the averaging rules given in GULP manual to obtain the Li-AL, Al-O and Li-O interactions. Now, while optimizing would you recommend me to vary only Li-Al, Li-O and Al-O interactions? or should I vary all the 12 parameters (2(epsilon and sigma) X 6 interactions)? Surprisingly when I vary all 12 parameters, the Al-Al and O-O interaction is being changed drastically. Since the like pair interactions (Li-Li, AL-Al, O-O) are well established in literature I was wondering if I can keep them fixed. I am using the data from the stiffness matrix of LiAlO2 for the fitting. I currently have 12 observables and 12 variables. But is it better to have lesser number of variables and more observables? I understand this may not be a direct question from GULP but I just wanted to take your advice from your vast experience for doing this the right way.

Apologies for quick replies but it’s busy end of year time:

  1. A and B should be able to be mapped to epsilon and sigma & so if things don’t work this problem just means an error in the maths. Remember that there are 2 forms of epsilon and sigma depending on whether epilson is the minimum or the point of crossing the axis. You may be using the wrong definition and that would mess it up.
  2. See earlier comments to your previous post about cation-cation potentials. As for what to fit, this is really a question for who ever is leading the project scientifically to address. Otherwise it’s better to use literature models from someone who is experienced in fitting and the model has been tested already.