vibeqc.SelectedCIOptions

class vibeqc.SelectedCIOptions(target_size=100, conv_tol_energy=1e-06, max_iter=50, verbose=0, random_seed=42, pt2_threshold=1e-06, selection_growth_factor=2.0, max_det_per_iter=5000, do_pt2_correction=True, spin_restricted=False, use_davidson=True, davidson_threshold=50, davidson_max_iter=100, davidson_conv_tol=1e-08, significant_coeff=0.01, spin_complete=True, select_eps=0.0)

Bases: SolverOptions

Options for the Selected-CI solver.

Parameters:

• target_size (int)

• conv_tol_energy (float)

• max_iter (int)

• verbose (int)

• random_seed (int)

• pt2_threshold (float)

• selection_growth_factor (float)

• max_det_per_iter (int)

• do_pt2_correction (bool)

• spin_restricted (bool)

• use_davidson (bool)

• davidson_threshold (int)

• davidson_max_iter (int)

• davidson_conv_tol (float)

• significant_coeff (float)

• spin_complete (bool)

• select_eps (float)

target_size

Maximum number of determinants in the variational space.

Type:

int

max_iter

Maximum selection + diagonalization cycles.

Type:

int

conv_tol_energy

Convergence threshold on total energy change (Hartree).

Type:

float

pt2_threshold

Perturbative threshold for selecting new determinants. Lower = more determinants selected per iteration.

Type:

float

selection_growth_factor

Max ratio by which the determinant space can grow each iteration.

Type:

float

max_det_per_iter

Hard cap on new determinants per iteration.

Type:

int

do_pt2_correction

Compute final PT2 energy correction after convergence.

Type:

bool

spin_restricted

Default False (issue #107): the solver walks SpinDet = (alpha_occ, beta_occ) determinants, the full S_z sector of the active space, so a converged run is the CAS/full-CI answer. True selects the closed-shell (seniority-zero, DOCI) basis, in which each Det is a tuple of doubly-occupied spatial orbitals excited by whole pairs, coupled through the pair-move integral g_{iiaa} (issue #639). That basis is a closed subspace: it cannot represent an open-shell configuration, so on a system whose exact state has seniority > 0 it converges to the DOCI energy, not the CAS limit; a restricted run whose candidate pool empties short of target_size reports converged=False with stop_reason="space_exhausted" and warns.

Type:

bool

use_davidson

Use Davidson iterative diagonalization for ndet > davidson_threshold.

Type:

bool

davidson_threshold

Switch to Davidson when ndet exceeds this count.

Type:

int

davidson_max_iter

Maximum Davidson iterations.

Type:

int

davidson_conv_tol

Davidson residual convergence tolerance.

Type:

float

__init__(target_size=100, conv_tol_energy=1e-06, max_iter=50, verbose=0, random_seed=42, pt2_threshold=1e-06, selection_growth_factor=2.0, max_det_per_iter=5000, do_pt2_correction=True, spin_restricted=False, use_davidson=True, davidson_threshold=50, davidson_max_iter=100, davidson_conv_tol=1e-08, significant_coeff=0.01, spin_complete=True, select_eps=0.0)

Parameters:

• target_size (int)

• conv_tol_energy (float)

• max_iter (int)

• verbose (int)

• random_seed (int)

• pt2_threshold (float)

• selection_growth_factor (float)

• max_det_per_iter (int)

• do_pt2_correction (bool)

• spin_restricted (bool)

• use_davidson (bool)

• davidson_threshold (int)

• davidson_max_iter (int)

• davidson_conv_tol (float)

• significant_coeff (float)

• spin_complete (bool)

• select_eps (float)

Return type:

None

Methods

__init__([target_size, conv_tol_energy, ...])

Attributes

conv_tol_energy	Energy convergence threshold (Hartree).
davidson_conv_tol
davidson_max_iter
davidson_threshold
do_pt2_correction
max_det_per_iter
max_iter	Maximum number of macro-iterations.
pt2_threshold
random_seed	Random seed for reproducibility.
select_eps	during candidate generation, drop the contribution of generator determinant I to candidate D when |H_DI| * max_k |c_I^k| < select_eps.
selection_growth_factor
significant_coeff	Coefficient magnitude above which a variational determinant contributes to candidate generation (max over roots).
spin_complete	For an M_s = 0 active space, close the selected set under the alpha/beta swap so the truncated roots stay (near-)spin-pure.
spin_restricted
target_size	Target number of determinants / bond-dimension / constraint set size.
use_davidson
verbose	0 = silent, 1 = per-iteration, 2 = debug.

pt2_threshold: float = 1e-06

selection_growth_factor: float = 2.0

max_det_per_iter: int = 5000

do_pt2_correction: bool = True

spin_restricted: bool = False

use_davidson: bool = True

davidson_threshold: int = 50

davidson_max_iter: int = 100

davidson_conv_tol: float = 1e-08

significant_coeff: float = 0.01

Coefficient magnitude above which a variational determinant contributes to candidate generation (max over roots).

spin_complete: bool = True

For an M_s = 0 active space, close the selected set under the alpha/beta swap so the truncated roots stay (near-)spin-pure.

select_eps: float = 0.0

during candidate generation, drop the contribution of generator determinant I to candidate D when |H_DI| * max_k |c_I^k| < select_eps. 0 (default) keeps exact CIPSI numerators. The C++ backend walks presorted |H|-ordered double-excitation lists under this threshold, so enumeration stops at the first sub-threshold entry instead of visiting every virtual pair; the Python kernel applies the same predicate by brute force (identical candidate sets, the oracle for the walk).

Type:

Heat-bath prefilter (Holmes-Tubman-Umrigar 2016)