johanjoensson / impurityModel / 33924358038

91%

Build Type

push

github

Committed by johanjoensson

Commit Message

Name orbital groups by their key, not by a guessed shell

The `l=2,a` style label was a guess: `_block_shell_tag` counted the
spin-orbitals in a group and inverted n = 2(2l+1) to recover a shell. That is
only meaningful when a group is a whole spin-doubled shell, which is true in
the spectra path (model.py keys impurity_orbitals by l) and false by
construction in the selfenergy and double-counting paths, where
group_orbitals_by_blocks makes each group a single symmetry manifold.

On the same NiO d shell the two keyings therefore disagreed:

  key = l, whole 3d shell:      ['l=2,a', 'l=2,b']     <- eg / t2g, correct
  key = manifold index:         ['mixed,a', 'l=1,a']   <- t2g's 6 spin-orbitals
                                                          are also a p shell's

Drop the inference. A block equivalence class is named after the
impurity_orbitals group that contains it -- the same key the occupation search,
the basis-generation windows and the frozen-shell notes already use -- so one
legend now covers the whole run. A group the block structure splits further
keeps its key and gains a letter (`group 2.a`); a class spanning several groups
is `groups 1,2`; a class with an orbital outside every group falls back to a
letter, since there is then no group to name it after.

The `N(...)` columns of the thermal and per-eigenstate tables and the screening
diagnostics' channel names all follow, and print_impurity_orbital_groups no
longer needs to carry the key alongside a separate label.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012WjuzYMi6j8ciZbztSwfjP

Coverage Stats

69 of 69 new or added lines in 3 files covered. (100.0%)

105 existing lines in 8 files now uncovered.

31281 of 34547 relevant lines covered (90.55%)

0.91 hits per line