qrunch.chemistry.subspace_diagonalization.efficient_slater_condon

Slater-Condon subspace diagonalizer backed by the Rust EfficientSlaterCondonCalculator.

Classes

EfficientSlaterCondonSubspaceDiagonalizer

Subspace diagonalizer built on the Rust-accelerated Slater-Condon CI matrix.

class EfficientSlaterCondonSubspaceDiagonalizer

Bases: SubspaceDiagonalizer

Subspace diagonalizer built on the Rust-accelerated Slater-Condon CI matrix.

Delegates to the full CI-matrix machinery from EfficientSlaterCondonCalculator.

The most recent lowest_eigenpair() result is cached, so consumers sharing one diagonalizer – an adaptive determinant-selection strategy and the reweighter steering it, say – pay for a single diagonalization of a given subspace instead of one each.

__init__(integrals: RestrictedElectronicStructureIntegrals | UnrestrictedElectronicStructureIntegrals) → None

Initialize the diagonalizer.

Parameters:: integrals (RestrictedElectronicStructureIntegrals | UnrestrictedElectronicStructureIntegrals) – The electronic structure integrals.
Return type:: None

compute_overlap_with_neighboring_states(bitstring: int, *, paired_electron_approximation: bool, return_max_n_overlaps: int = 1000) → dict[int, float]

Compute the Hamiltonian matrix element between a determinant and every state it couples to.

Parameters:

bitstring (int) – The determinant to expand around, as a bitstring integer.
paired_electron_approximation (bool) – Whether to include only paired electron excitations.
return_max_n_overlaps (int) – Cap on the returned overlaps; the largest ones survive the cap.

Return type:

dict[int, float]

diagonal_elements(bitstrings_as_integers: list[int]) → ndarray[tuple[Any, ...], dtype[float64]]

Compute the diagonal CI matrix elements \(\bra{k}\hat{H}\ket{k}\) of the given determinants.

Parameters:: bitstrings_as_integers (list[int]) – Determinants to evaluate, as bitstring integers.
Return type:: ndarray[tuple[Any, …], dtype[float64]]

lowest_eigenpair(bitstrings_as_integers: list[int]) → tuple[float, ndarray[tuple[Any, ...], dtype[float64]]]

Compute the lowest eigenvalue and its eigenvector of the CI matrix projected onto the determinants.

The CI matrix is built dense, so a direct LAPACK solve restricted to the lowest eigenpair beats an iterative one, which would need many dense matrix-vector products to converge on the extreme end of the spectrum.

Only the most recent subspace is cached: subspaces grow monotonically over a run, so an older one is never asked for again and keeping it would only tie up memory.

Parameters:: bitstrings_as_integers (list[int]) – Determinants spanning the subspace, as bitstring integers.
Return type:: tuple[float, ndarray[tuple[Any, …], dtype[float64]]]