Repository navigation
[vasp] run DMFT on the irreducible BZ with symmetrization - #17
Merged
Merged
Conversation
Collaborator
|
Hi @the-hampel, now that we have the IBZ symmetry operations are they in the same format as the Wien2k IBZ symmetry ops? Then the DFTtools and ModEST symmetrizers can consume the same data and apply the symmetrization to k-summed quantities. |
Member
Author
|
Hi @harrisonlabollita , |
Collaborator
|
That answers my question! Thank you |
Member
Author
|
Note after seeing all these comments on Wien2k with SOC: I did not think or test this with SOC. I am not sure if I have to be careful with anything in that case. TBD next week as well. |
Add an IBZ mode to the VASP converter so DMFT can run on the irreducible Brillouin zone (with space-group symmetrization) instead of unfolding to the full k-grid. On a dense NiO mesh this is ~10x fewer k-points and ~9x faster Gloc for an identical result. New module triqs_dftkit.vasp.symmetry: - real cubic-harmonic rotation matrices D^l(R) for l=1,2 (tensor method, basis orders matching VASP: d=[xy,yz,z2,xz,x2-y2], p=[y,z,x]) - read VASP symmetry from vaspout.h5; convert the reciprocal-fractional symops to Cartesian and close them into the full point group (VASP only stores one coset representative per k-point, which is incomplete on coarse/special meshes) - atom permutations under the space group (with translation detection) - build the dft_symmcorr_input payload (mat/perm/time_inv) consumed by triqs_modest / dft_tools to restore the full-BZ average converter.convert_dft_input gains use_ibz (auto-detect from vaspout.h5), vasp_h5 and plo_cfg arguments: in IBZ mode it slices the data to the irreducible k-points, sets bz_weights to the IBZ multiplicities and symm_op=1, and writes the real symmetry matrices. Without symmetry data (text mode) it falls back to the full grid and suggests the h5 interface. Tests: converter IBZ tests for SrVO3 (t2g, cubic) and AFM NiO (full d + O p, non-cubic, two equivalent Ni); ignore generated test artifacts.
… subspaces Two fixes for the IBZ symmetrization path: 1. Source the per-shell TRANSFORM from the PLOVASP output instead of re-parsing plo.cfg. PLOVASP now serializes each shell's transform matrices (real harmonics -> correlated orbitals) into the .pg header (completing the long-standing TODO), and the converter reads them back per ion. This guarantees the matrices used to build the correlated-shell symmetry operations are exactly the ones used to build the projectors, and removes the fragile re-parse (which silently mishandled TRANSFILE, complex transforms, and fell back to a wrong identity slice). The plo_cfg argument to convert_dft_input is now unused (kept for back-compat) and read_transforms_from_plo_cfg is removed. 2. Reject correlated orbital sets that are not symmetry-closed. Q = T D(R) T^dag is a representation only if the orbital subspace (row span of T) is invariant under every point-group operation. build_symmcorr now checks this per shell (gauge-free, via the subspace projector) and raises with a clear message otherwise. Full shells and complete irreps (t2g, eg) pass; an arbitrary LOCPROJ subset that splits a multiplet (e.g. dxy,dz2,dx2-y2) is caught instead of silently producing a wrong full-BZ average. Tests: existing converter IBZ refs (SrVO3 t2g, AFM NiO full d + O p) reproduce unchanged via the header transform; non-IBZ converter tests unaffected.
Complements the subspace-invariance guard. A partial LOCPROJ selection (fewer than 2l+1 real harmonics) without a TRANSFORM leaves the absent orbitals zero-filled in proj_mat; with an identity transform the invariance check passes (identity is trivially invariant) but symmetrization on the IBZ would leak the kept orbitals into the dead partners. Detect zero-weight correlated orbitals before building the symmetry operations and raise a clear error pointing at the full-shell / complete-irrep / full-grid alternatives. Existing IBZ tests (SrVO3 t2g, AFM NiO full d + O p) have no zero-weight orbitals and are unaffected.
…e it DFTTools' Symmetry finds the image of a correlated shell by comparing the whole orbit dict, 'sort' included, while the VASP converter gives every site its own sort. Any operation that exchanges correlated atoms (the two O in the NiO test) therefore made SumkDFT fail with "list.index(x): x not in list". Shells mapped onto each other by the space group now share one sort in dft_symmcorr_input/orbits. ModEST does not look at the sort, which is why its IBZ test did not see this. Further changes to the symmetry construction: - refuse, instead of assuming the identity, when the atom permutation of an operation cannot be determined - refuse when a symmetry image of a correlated shell is not itself a correlated shell (e.g. only one of two equivalent atoms projected) - refuse spin-orbit coupling, where the operations would have to act on the spinors as well - check that the first n_k_ibz points of the full grid are the IBZ points, which the slicing in the converter relies on Converter: - drop the unused plo_cfg argument of convert_dft_input() - look up vaspout.h5 next to the basename by default, as the KPOINTS_OPT conversion already does, instead of in the working directory - in automatic mode (use_ibz=None) fall back to the full grid with a note when the projector set cannot be symmetrized, so that setups that worked on the full grid keep working; use_ibz=True still raises - the VASP Driver passes use_ibz through for CSC runs Tests: the IBZ test now also checks the physics. The DFT occupation matrices on the IBZ, symmetrized the way DFTTools does it, must agree with the symmetrized full-grid ones (1e-8) and with the raw full-grid ones to the symmetry of the VASP projectors (1e-10 SrVO3, 1e-4 NiO). The NiO reference changes only in dft_symmcorr_input/orbits[3]/sort. The SrVO3 full-grid regression test now asks for use_ibz=False, since the default would otherwise pick up its vaspout.h5 and switch to the IBZ.
…e known to work The automatic IBZ mode is on by default, so it must switch itself on only for correlated shells the symmetry operations are correct for: - rotation matrices exist for l = 0, 1, 2 only (l = 0 is new and trivial); f shells are refused up front with a clear message instead of failing deep inside real_harmonic_rotation - the TRANSFORM of every correlated shell must have orthonormal rows, otherwise T D T^dag is not unitary - the symmetry operations are checked on the data themselves: the full grid is still at hand in the converter, and the symmetrized IBZ occupation and local Hamiltonian of the correlated shells must reproduce the full-grid ones within ibz_tol (default 1e-3). Correct operations leave only the symmetry error of the VASP projectors (1e-12 SrVO3, 7e-5 NiO); swapping z^2 and x^2-y^2 in the d matrices, which passes every structural check, gives 6e-3 (occupation) and 3e-2 eV (local levels) As before, use_ibz=None falls back to the full grid with a note when any check fails, and use_ibz=True raises. Tests: D^l is orthogonal and a representation; f shells, non-orthonormal transforms and SO are refused; projecting only one of two equivalent O and wrong d matrices (mocked) both fall back in auto mode and raise with use_ibz=True.
the-hampel
force-pushed
the
vasp_ibz
branch
from
September 24, 2026 15:41
24a4c7d to
2b6272b
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Add an IBZ mode to the VASP converter so DMFT can run on the irreducible Brillouin zone (with space-group symmetrization) instead of unfolding to the full k-grid. On a dense NiO mesh this is ~10x fewer k-points and ~9x faster Gloc for an identical result.
New module triqs_dftkit.vasp.symmetry:
converter.convert_dft_input gains use_ibz (auto-detect from vaspout.h5), vasp_h5 and plo_cfg arguments: in IBZ mode it slices the data to the irreducible k-points, sets bz_weights to the IBZ multiplicities and symm_op=1, and writes the real symmetry matrices. Without symmetry data (text mode) it falls back to the full grid and suggests the h5 interface.
Tests: converter IBZ tests for SrVO3 (t2g, cubic) and AFM NiO (full d + O p, non-cubic, two equivalent Ni); ignore generated test artifacts.