> @sunqm if I may: we have been recently discussing a few issues reported by the TREXIO users. I can summarize them as follows:

1. Failures to produce TREXIO files from PySCF when using Dunning cc-pvXz basis sets due to the contraction scheme being used, reported here by @scemama , @NastaMauger and @kousuke-nakano . @scemama proposed a solution here which would basically "decontract"/expand the basis set. Do you think it is feasible? Are there any internal PySCF objects which hold the necessary pieces?
2. As mentioned by @NastaMauger here: the current TREXIO interface implemented in pyscf-forge PR Trexio interface #60 does not produce some details about the atomic orbitals (we are only missing shell and their normalization arrays from the ao group in TREXIO format), which prevents importing of the PySCF-produced TREXIO files in some other codes. Would it be possible to add this information to the TREXIO file produced by the _mol_to_trexio function?

1. Is it an issue the mixing of segment contraction and general contraction in pyscf? Does basis sets like ANO work using this the interface? The cc-pv*Z basis in pyscf was obtained as an "optimized" format, which contains from basis set exchange. The zeros in the contraction coefficients are squeezed, leading to , e.g. multiple S-shell blocks, P-shell blocks. Does it work if I aggregate all S shells into one block, and P into one block, so on so forth?

2. Sure. I'm not clear how to build the index and normalization. An example could explain easily I guess. What do these variables look like for STO-3G of Carbon?

Originally posted by @sunqm in #80 (comment)