LJ parameters fitting failed

Dear Users,

I want to fit the LJ potential parameters of MOF with the experimental data as the observables.

The potential parameters related to Cu will be fitted under 1 atm pressure in my crystals. In the first stage, I fit the epsilon and sigma between Cu and O.
fit.gin (12.8 KB)

the observables are the volume and bulk modulus.

when I set the epsilon as the variables (the output is:
fit20.got (152.5 KB)
), although the output says: ‘Fit completed successfully’, the sum of squares shows ‘*********’, BTW, what does a string of asterisk mean? means ‘none’? or ‘can not be calculated’?

secondly, I tried to fit sigma separately (
fit9.got (152.1 KB)
), it gives: **** No lower sum of squares could be found ****

also, I tried to fit other types of pair parameters separately, but I got the same failed results.

I am thinking about why it cannot fit just one parameter, the possible reasons could be:

  1. I checked example2.got, 4, 9, 14, 22, 48 and 71, the final and initial parameters are close to each other, except for some in examples 22, 48 and 71, which means the starting parameters are just a bit distorted from the correct ones. So in my case, the starting force field parameters may be too far away from the reasonable ones, so the GULP try very hard but cannot find the matched ones. if this is the reason, should I use more accurate ways ( the quantum mechanical approach) to optimize the parameters?

  2. maybe the observables are not comprehensive, I need to add more benchmarks to help the GULP. moreover, the experimental data as the observables may not be the best way, because it’s hard to reproduce the same data every time and difficult to ensure the physical correctness. can the weighting factors be used to make the fitting valid? or it’s better to use the quantum mechanical data.

  3. I didn’t add charges in all my fittings, which means the gulp only fits my parameters with the non-bonded interaction but the coulomb interaction is omitted, could this affect the fitting process?

please let me know if these possible reasons make sense or not, and if none of them works, what could be the reason?

Thank you so much.

wang

Dear Wang,
I’d strongly suggest taking a look at the Phil. Mag. B reference on fitting with GULP as an introduction to the theory & this will help a lot with what you want to do. A key point from this article is that the properties are only meaningful for an optimised structure & so there is no point trying to fit them for a complex structure without optimisation (except in the unlikely situation where all the forces fit to be zero at the experimental structure). Bottom line is you need to use “relax” fitting as described in this paper. If you get *********** in the output it means the sum of squares is so large it’s off scale and can’t be printed. Usually it’s a pointer that your initial model is not good enough, or in this case you’re using the wrong type of fitting such that the observables aren’t sensible.
Regards,
Julian