johanjoensson / rspt2spectra / 31096220512

79%

Build Type

push

github

Committed by johanjoensson

Commit Message

Declare spin_ordering: down_first as a structural guarantee, not a measurement

build_h0 now writes spin_ordering: "down_first" unconditionally, and
write_h0_file adds "spin_ordering" to required_features whenever the
caller declares it, so an older reader that doesn't understand the key
must refuse rather than silently mismatch it.

The convention is not detected per file -- it follows from how RSPt lays
out its arrays. green_trunk_interface.F90's lda_mlmsatomicqn assigns
qn(2, :offset) = -1 (2*ms, spin down) to the first packed block and
qn(2, offset+1:) = 1 (spin up) to the second, for both the local
Hamiltonian and the rotation matrices RSPt prints; generate_rspt_T_matrix
already reproduces that block order, so no permutation is ever needed.
This happens to already match c2i's own convention (s=0 first, s=0=down),
which is why no checked workload ever needed the spin blocks swapped.

The originally planned empirical check (compare a sign to a reported
moment on a spin-split workload) turned out not to be executable: every
stored RSPt workload's printed Local hamiltonian and hybridization
function are spin-degenerate between the two blocks, even for magnetic
materials like FCC Ni -- confirming and generalizing the earlier BCC_Fe
finding rather than contradicting it. The source-level argument does not
depend on such a workload existing.

Also fixes a real bug this surfaced: the golden_h0_v1_index.h0 fixture
(shared with impurityModel, hash-pinned) already had a "spin_ordering"
header key but never listed it in required_features -- exactly the
declared-but-dead state that made the key silently ignorable everywhere.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Coverage Stats

3 of 3 new or added lines in 1 file covered. (100.0%)

129 existing lines in 4 files now uncovered.

1888 of 2359 relevant lines covered (80.03%)

3.2 hits per line