About obtaining local energy contribution and neighboring list

[saschapojot]

Dear DeepMD developers,

According to deepmd papers, such as Eq. (37) in paper https://pubs.aip.org/aip/jcp/article/159/5/054801/2904916/DeePMD-kit-v2-A-software-package-for-deep, and Fig. 1(c) in paper https://arxiv.org/abs/1707.01478, after the network is trained, for the inference, the local energy for each atom is summed up, then the summation result is the energy for the entire system.

I need to access the local energies, as well as an atom's neighboring list.

This is because I would like to use deepmd for Monte-Carlo computation. Monte-Carlo computation can always utilize local interactions with 1 atom. Then Monte-Carlo computation can be parallelized by a checkerboard update pattern. This will greatly reduce the computation time compared to molecular dynamics computation, since molecular dynamics computation needs total energy.

As long as local energies are obtained and computation is accelerated by checkerboard update, physical properties of materials (under thermodynamic equilibrium) can be computed in a short time through MCMC subroutine, thanks to statistical physics.

Could you show me how to find the places of the local energies and neighboring lists in the source code? Thanks very much!

Liu Xi