Derivatives of first atom in fit

GULP
1

I´ve been trying to fit some cross Lennard-Jones parameters between a molecule and a rigid framework using GULP.

I successfully fitted a potential with reasonable accuracy, but I´ve noticed that some forces are zero within the force field while not zero in my observables (DFT calculations).
Here is a picture to illustrate:

scatter_force
scatter_force1734×1294 166 KB

In the process of trying to understand this behavior, I have noticed that for some reason the first line of the gradients section is not considered in the fitting.
You can see this by looking at the output when the observables are printed:

         81        Energy           -0.403059      349.1410               81
         82        Energy           -0.289607        4.3360               82
         83        Derivative        0.001900        0.0000    2 x         1
         84        Derivative        0.001950        0.0000    2 y         1
         85        Derivative       -0.001238        0.0000    2 z         1

I have double-checked by printing in the setcfg.F90 file the lines where the fgrad is assigned to fobs.
Is there a particular reason for this?

Thank you in advance!

2

I think I’d need to see your input file in full to see exactly what’s happening here, but I think I can explain why the derivatives on one atom might go missing (or be set to zero). This is because one atom always has to be fixed due to translational invariance otherwise the Hessian matrix can’t be inverted during optimisation. Another way of looking at it is that the information gained from fitting the forces on 1 atom in the system is redundant since the sum of the forces on all atoms is constrained to be zero (assuming there is no real space integration grid in your DFT calculation that is not atom-centred and therefore gives rise to noise in the forces).
Anyway, if you can post your input (or send to the GULP help email) then it will be easier to check out what is happening for your case in full.