Eric Lindgren

> You don't need me first make dq/d r_ij available to you? Yes I do, but I just started to play around with the other terms. It's just for me...

> ok, then perhpas this is not ready for a PR. Or you might need to resolve conflicts later after my PR. Yes this is not ready to actually be...

> I will make the PR soon, and might also change something about memory usage . You followed my previous NEP2/NEP3 style to use a lot of `local memory` in...

@brucefan1983 I've made a rough first implementation of the more complicated forces for NEP4. I spent quite a lot of time trying to clean up the expressions, but I did...

> You can also remove the definition and calling of `find_q_scaler` and `apply_q_scaler`. Currently you have scaled the descriptor but have not scaled the relevant derivatives. > > From my...

> > There are two things that I'm unsure about. 1) I'm not sure that I'm, accessing dq_dr correctly in `apply_gnn_compute_messages` and 2) I'm unsure of how to include `find_force_angular`...

@brucefan1983 I fixed NEP4 to use the neighbor information in dq_dr. Next step would be to check the forces using finite differences. I checked your code in `validate_force` in the...

> So you need to implement NEP4 in the GPUMD part too. You need to pay attention to some subtle differences between the GPUMD and the NEP parts! Alright, I'll...

@brucefan1983 I have begun working on implementing NEP4 in `force/nep4`. I have a few questions. 1. I don't want to copy the code for NEP4 from `main_nep` to `force`, since...

> The implementations of NEP in main_nep/ and force/ are not identical. When you get into the details, you will notice some differences. So we need to duplicate some code....