Repository navigation
[wien2k] pure-Python case.ctqmcout correlated-shell projectors - #9
krystophny wants to merge 2 commits into
Conversation
|
Accuracy: reproduces the released dmftproj |
5a805f7 to
97e1020
Compare
|
GPT Sol rebased this onto the corrected #8 and fixed duplicated window flags, mismatched SO spin windows, spin-specific band offsets, and NaN-blind comparisons. Unsupported paths now fail explicitly. The current full-rank fixture matches within Chris&AI |
7ceca82 to
9bc18c3
Compare
|
GPT Sol backported the remaining projector fixes into this standalone step: per-shell representation parsing, explicit unsupported-input checks, below-window spin consistency, and scale-relative Loewdin rank detection. Native-reference, tiny full-rank overlap, singular-overlap and parser regressions pass. Restacked on updated #8. Latest-head CI is running/queued. Chris&AI |
Read the almblm coefficients, assemble and locally rotate correlated projectors, apply the cubic/complex basis, and Loewdin-orthonormalize the stacked shells before writing ctqmcout. Part of TRIQS#7, following TRIQS#8. Keep the native separate window-open/valid flags. Validate matching SO spin windows and use each spin's own band offsets. Reject unsupported non-SO, band-index, cubic p/f and non-identity struct rotation paths instead of producing misleading files; subsequent port stages add them. Reuse the current full-rank CaOs2 inputs and TRIQS#8 fixture handling. Compare integer fields exactly and finite float fields at atol=1e-7, rtol=0. Add regressions for final-band-only windows and mismatched SO spins. Local validation gives max native residual 1.3e-8 and full 20x50 projector orthonormality error 3.2e-15. The old rank-deficient-window claims no longer apply to this fixture.
9bc18c3 to
6d3188c
Compare
1 participant
Part of #7, stacked on the corrected #8.
Generate
case.ctqmcoutfrom the almblm coefficients: assemble the correlated projectors, apply local symmetry and cubic/complex basis rotations, and Loewdin-orthonormalize the stacked shells. This stage supports spin-polarized SO, mode 0, whole shells and identity struct local rotations. Unsupported non-SO, band-index, fromfile and cubic p/f paths are rejected; later PRs extend support.The rebase retains the current native fixtures and #8's compressed-input handling. Window selection now keeps opened and valid windows separate, checks that SO spin windows coincide, and uses each spin's own band offsets. Empty windows fail explicitly.
The CaOs2 comparison checks integers exactly and finite floats at
atol=1e-7, rtol=0. Regressions cover final-band-only windows and mismatched spin windows. Local checks pass: the maximum native residual is1.3e-8; the generated full 20×50 projector has orthonormality error3.2e-15. The earlier rank-deficiency claim concerned the superseded narrow window, not this fixture.The SOC convention issue in dft_tools#148 is distinct from native precision loss; its non-centrosymmetric frame correction remains in #18/#19.
Chris&AI