PAOFLOW.spectrum.sparse_bands#
sparse_bands.py — Sparse on-the-fly Lanczos band structure for EDTB models.
Bypasses dense H(R) storage. For each k-point the Hamiltonian H(k) is
assembled as a sparse CSR matrix from a precomputed bond list, then
scipy.sparse.linalg.eigsh extracts the requested eigenvalues via
implicitly-restarted Lanczos.
Usage#
from sparse_bands import SparseEDTB ham = SparseEDTB(model_dict) result = ham.compute_bands(“K-G-M-K’”, high_sym_pts, nk=100, n_eigs=50)
Classes#
Sparse EDTB Hamiltonian for large supercells. |
Module Contents#
- class PAOFLOW.spectrum.sparse_bands.SparseEDTB(model_dict, verbose=True)[source]#
Sparse EDTB Hamiltonian for large supercells.
- Parameters:
- eigvals_at_k(k_frac, n_eigs=50, sigma=None, **kwargs)[source]#
Compute selected eigenvalues at one k-point.
- Parameters:
k_frac (array_like, shape (3,))
n_eigs (int) – Number of eigenvalues to compute.
sigma (float or None) – Shift-invert target. If provided, the eigenvalues nearest sigma are returned (fast for interior eigenvalues); otherwise the smallest eigenvalues are returned.
**kwargs – Extra arguments passed to scipy.sparse.linalg.eigsh.
- Returns:
eigenvalues
- Return type:
ndarray, shape (n_eigs,) sorted ascending
- eigh_at_k(k_frac, n_eigs=50, sigma=None, **kwargs)[source]#
Compute selected eigenvalues and eigenvectors at one k-point.
Same solver as
eigvals_at_k()but also returns the eigenvectors, needed to build spectral weights when unfolding.- Parameters:
k_frac (array_like, shape (3,))
n_eigs (int) – Number of eigenpairs to compute.
sigma (float or None) – Shift-invert target (see
eigvals_at_k()).**kwargs – Extra arguments passed to scipy.sparse.linalg.eigsh.
- Returns:
evals (ndarray, shape (n_eigs,) sorted ascending)
evecs (ndarray, shape (nawf, n_eigs) columns match
evals)
- compute_bands(band_path, high_sym_points, nk=100, n_eigs=50, sigma=None, outputdir=None, return_eigenvectors=False, n_workers=1, backend='loky', **kwargs)[source]#
Compute band structure (and optionally eigenvectors) along a k-path.
- Parameters:
band_path (str) – Path specification, e.g. “K-G-M-K’”.
high_sym_points (dict) – {label: [k1, k2, k3]} in fractional coordinates.
nk (int) – Total number of k-points along the path.
n_eigs (int) – Number of eigenvalues at each k-point.
sigma (float or None) – Shift-invert target energy (eV).
outputdir (str or None) – If set, write bands_0.dat and kpath_points.txt.
n_workers (int) – Number of parallel workers for k-point loop. 1 = serial (default). -1 = all available cores. Requires joblib; falls back to serial if unavailable. Capped at this process’ core budget (its share of the node under
mpirun; seePAOFLOW.utils.parallel_resources).backend (str) – joblib backend for the k-point loop. Default ‘loky’ (separate processes). ARPACK’s shift-invert solve runs a Python-level reverse-communication loop that holds the GIL, so a ‘threading’ backend does not parallelize it — use processes for real speedup.
**kwargs – Extra arguments for eigsh (e.g. tol, maxiter).
- Returns:
result – ‘eigenvalues’ : ndarray (nk, n_eigs) ‘k_dist’ : ndarray (nk,) ‘tick_pos’ : list ‘tick_labels’ : list ‘bands_file’ : str or None
- Return type: