I congratulate the Casino developers for neural networks implementing in the code.

[Vladimir_Konjkov]

Please accept my sincere congratulations on this epochal event!
Could you also provide an example config file and a link to updated documentation?


Vladimir.

[Vladimir_Konjkov]

The Jastrow factor redistributes the electron density from where the probability of two electrons meeting is higher to where it is lower, so a term is needed that will redistribute it back. Such a term (chi-term) is usually included and depends on the distance of the electron to the nucleus, since such a distance correlates with the magnitude of the electron density, but in the case of a chemical bond, the electron density is also concentrated along the bonds. In the article "Framework for constructing generic Jastrow correlation factors" P. López Ríos1, P. Seth, N. D. Drummond, and R. J. Needs. DOI: https://doi.org/10.1103/PhysRevE.86.036703

Screenshot_20250429_113445.png (46.93 KiB) Viewed 1205 times

N(1,2) term improves Jastrow further, because it is cylindrically symmetrical around the chemical bond. Is it possible to force the neural network to take into account the distribution of electron density along chemical bonds, which is not taken into account by conventional terms?

Vladimir

[Neil Drummond]

Thanks, but it's still very much a work in progress...

[Neil Drummond]

At present the NNJAS term mixes up u-like and chi-like terms. But it is not too difficult to add new features to the NNJAS term, to see what works.

Best wishes,

Neil.