Hello,
I’m fitting a potential for the cross LJ terms between the interaction of a gas molecules and a carbon structure.
I’ve obtained the interaction energies by performing DFT single point calculations moving the molecule towards the structure.
I successfully obtained the LJ parameters fitting only to energies and weighting the configurations with Boltzmann weights.
The interaction energies curves are pretty close to my DFT data, and I’m starting my fit with forces with these parameters.
However, when I perform the fit including the forces, the parameters go crazy (negative epsilons, sigmas going to zero etc…), even if I try to fit just epsilons, or just sigmas, or just a few parameters at a time.
I’m suspicious that I might be doing something wrong in the input of the atomic forces, or that the forces are not being weighted.
The way that I’m passing the forces to GULP is in the observables section, by adding a “gradients” section with the force on each atom.
I’m using the flags “fit conv” to perform the fit, as I don’t want the lattice parameters to be optimized.
I have something like:
fit kcal conv
cell
50.00000000 50.00000000 22.10400000 90.00000000 90.00000000 90.00000000
fractional 1
C1 core 0.135599999 0.999299999 0.556731813 0.0000000 1.0000 0.0000
C1 core 0.132780000 0.027500000 0.556731813 0.0000000 1.0000 0.0000
…
C1 core 0.134779999 0.985120000 0.501176257 0.0000000 1.0000 0.0000
C2 core 0.000000000 0.000000000 0.000000000 0.0000000 1.0000 0.0000
O2 core 0.000000000 0.011499999 0.954985522 0.0000000 1.0000 0.0000
O2 core 0.000000000 0.988500000 0.045014477 0.0000000 1.0000 0.0000
observables
gradients eV/Angs
1 0.133382 0.051353 -0.003565
2 0.105167 0.085753 -0.003581
3 0.085530 0.116774 -0.003956
…
energy
-0.04897332 1.760000
endthen the file follows with several configurations, until
lenn epsilon zero 12 6
C1 core C2 core 0.41327703E-02 3.3085125 0.000 12.000 1 1
lenn epsilon zero 12 6
C1 core O2 core 0.10017456E-01 2.9127077 0.000 12.000 1 1
Any ideas on what I might be doing wrong?
Thanks in advance!
Henrique