@atharvchagi This might be a good first issue to get your feet wet. If you look at e3nn/e3nn-jax, there's a lot of complains about not supporting lmax>=10 spherical_harmonics. I think the machinery should help us resolve that issue since we can precompute, store it as .npz and then call it again. Similar story with cg coefficients as well
TODOS
I know this is pretty vague. If you are interested can provide more details :)