How to improve my fitting for Buckingham parameters

Hi everyone, I am new in GULP, and I was trying to fit Buckingham parameters of Cs-Cl by reproducing the crystal structure of CsCl and Cs3ClO, but the results were bad, the final lattice structures were very different from the initial structures. The following is my input file, and note that I use the partial effective charge that O is -0.8e, Cl is -0.4e, Cs is +0.4e, and the initial parameters for Cs-Cl are set casually. Could anyone tell me how to improve my fitting? Thanks!

fit relax conp optimise properties compare phono distance

name CsCl
4.11 4.11 4.11 90 90 90

fractional 2
Cs core 0.000000 0.000000 0.000000
Cl core 0.500000 0.500000 0.500000

name Cs3ClO
9.443 4.452 16.32 90 90 90
Cl1  -0.5782 -0.75 0.6724
Cs2  -0.2897 -0.25 0.7098
Cs3  -0.3288 0.25 0.4981 
O1 -0.1659 -0.25 0.5616
Cs1 0.0278 0.25 0.5996

Cs core 0.4
O core -0.8
Cl core -0.4

Cl core Cs core 6800.5797000000 0.5000220000 0.0000000000 0.0000000000 &
 12.0000000000 1 1 0 
Cs core O core 19256.42519 0.60003 0.0000000000 0.0000000000 &
 12.0000000000 0 0 0 
O core O core 3000.4300000000 0.2500000000 0.0000000000 0.0000000000 &
 12.0000000000 0 0 0 

There are a few things you can do to make fitting easier & better, so here are some suggestions:

  1. Normally I’d recommend starting from an existing model since there are already models for CsCl and almost certainly Cs2O & so fitting would hardly be necessary. Of course they would have formal charges, which makes it easier.
  2. Related to the above formal charges would work better since it will be hard to get good results for CsCl with such low charges. You would be able to fit the lattice parameter since it’s such a simple structure, but the bulk modulus would probably be too soft.
  3. Rather than fitting both structures together, if you are creating new potentials then it’s better to fit CsCl first and then assume you can transfer the potentials to your other structure so that you have less to fit. After that you can always perform a refinement of both together.
  4. Always try to include properties (e.g. bulk modulus, elastic constants, etc) since this will help give more sensible and transferable results.
  5. Your Cs-Cl/O potentials have very odd rho values - typically they are close to 0.3 for most systems. If you have no properties in the fit then you can’t really fit A and rho since you don’t have any curvature info, so just fit A.
  6. With partial charges as low as the ones you are trying to use you may need to generate extra binding through C6 terms. This may be unphysical, but if you can’t increase the charges there may not be a choice.
    Hope that helps.

Thank you for your kindly suggestions, they are of great help to me!