RTBSE

References: Marek2025

Controls options for the real-time Bethe-Salpeter (RTBSE) propagation. Note that running RTBSE requires previous low-scaling GW calculation. Also note that designating this section as RTBSE run but choosing run type ENERGY leads to potential deallocation errors. More details (including description of output files) is available in the methods section of the documentation. [Edit on GitHub]

Keyword descriptions

SECTION_PARAMETERS

Type: enum
Default: TDDFT
Lone keyword: RTBSE
Usage: &RTBSE TDDFT

Description: Which method is used for the time propagation of electronic structure. By default, use the TDDFT method. Can also choose RT-BSE method, which propagates the lesser Green’s function instead of density matrix/molecular orbitals.

Valid values:

TDDFT Use TDDFT for density matrix/MO propagation.
RTBSE Use RT-BSE for Green’s function propagation

Mentions: ⭐Real-Time Bethe-Salpeter Propagation

[Edit on GitHub]

DEBUG_DISABLE_HARTREE

Type: logical
Default: F
Lone keyword: T
Usage: DEBUG_DISABLE_HARTREE

Description: Debug option for linearized RT-BSE: disables the Hartree kernel in both the static reference initialization and the propagation. The Coulomb RI setup is still built so the run stays internally consistent.

[Edit on GitHub]

DEBUG_DISABLE_SEX

Type: logical
Default: F
Lone keyword: T
Usage: DEBUG_DISABLE_SEX

Description: Debug option for linearized RT-BSE: disables the screened-exchange kernel in both the static reference initialization and the propagation.

[Edit on GitHub]

DIAGNOSE_LIOUVILLIAN_EIG

Type: logical
Default: F
Lone keyword: T
Usage: DIAGNOSE_LIOUVILLIAN_EIG

Description: Diagnostic for linearized RT-BSE: at job initialization, build the Liouvillian on the OV subspace by probing the kernel routine with canonical OV basis vectors, then diagonalize. In TDA this equals the Casida-A eigenvalue problem, giving a broadening-free, finite-time-free correctness check against bse_full.F. In ABBA it builds and diagonalizes the full coupled (A, B) Liouvillian via the Furche reduction. Output is controlled by the LIOUVILLIAN_EIG print key in the parent &REAL_TIME_PROPAGATION%&PRINT section.

[Edit on GitHub]

ENERGY_CUTOFF_EMPTY

Type: real
Default: -1.00000000E+000 [eV]
Usage: ENERGY_CUTOFF_EMPTY 5.0

Description: Energy cutoff (relative to LUMO) defining the highest virtual molecular orbital included in the active MO window of the linearized RT-BSE propagation. Only used when LINEARIZED_BSE_PROPAGATION=.TRUE.. A non-positive value disables the virtual truncation.

[Edit on GitHub]

ENERGY_CUTOFF_OCC

Type: real
Default: -1.00000000E+000 [eV]
Usage: ENERGY_CUTOFF_OCC 5.0

Description: Energy cutoff (relative to HOMO) defining the lowest occupied molecular orbital included in the active MO window of the linearized RT-BSE propagation. Only used when LINEARIZED_BSE_PROPAGATION=.TRUE.. A non-positive value disables the occupied truncation.

[Edit on GitHub]

ENFORCE_MAX_DT

Type: logical
Default: F
Lone keyword: T
Usage: ENFORCE_MAX_DT

Description: For linearized RT-BSE, recompute TIMESTEP and STEPS so the same total propagation time is covered with the largest timestep that does not exceed the estimated RK4 stability limit.

[Edit on GitHub]

KERNEL_RI

Type: enum
Default: DEFAULT
Usage: KERNEL_RI RS

Description: Select the RI framework used to evaluate the linearized RT-BSE Hartree and screened-exchange kernels (propagation and reference). DEFAULT infers from the GW flavor: RI-RS if the GW_RI_RS section was active, AO-RI otherwise. RS/AO force that framework regardless of how GW was run. The RI-RS grid (mat_phi_mu_l, mat_Z_lP) and the V_grid/W0_grid kernels are built on demand. RI-RS is implemented for linearized RT-BSE only; for full RT-BSE an explicit RS is overridden to AO with a warning.

Valid values:

DEFAULT Infer from the GW flavor (GW_RI_RS active -> RS, else AO).
RS Real-space RI grid kernels (linearized RT-BSE only).
AO AO-RI kernels.

[Edit on GitHub]

LINEARIZED_BSE_PROPAGATION

Type: logical
Default: F
Aliases: LRRTBSE
Lone keyword: T
Usage: LINEARIZED_BSE_PROPAGATION .T.

Description: Linearizes the BSE propagation

[Edit on GitHub]

RTBSE_HAMILTONIAN

Type: enum
Default: G0W0
Usage: RTBSE_HAMILTONIAN G0W0

Description: Which Hamiltonian to use as the single-particle Hamiltonian in the Green’s propagator.

Valid values:

KS Use Kohn-Sham Hamiltonian for Green’s propagation.
G0W0 Use G0W0 Hamiltonian for Green’s function propagation

[Edit on GitHub]

TDA

Type: logical
Default: F
Lone keyword: T
Usage: TDA

Description: Apply the Tamm-Dancoff approximation to the linearized RT-BSE kernel: the Hartree and screened-exchange contributions are restricted so that the OV and VO blocks of the density response remain decoupled (i.e. only A-block coupling is kept, B-block coupling is dropped). Only effective when LINEARIZED_BSE_PROPAGATION=.TRUE..

[Edit on GitHub]

TDA_SHIFT_TO_FIRST_PEAK

Type: logical
Default: F
Lone keyword: T
Usage: TDA_SHIFT_TO_FIRST_PEAK

Description: For the linearized RT-BSE TDA path, shift the active-MO single-particle diagonals by +Omega_0/2 (occupied) and -Omega_0/2 (virtual) with Omega_0 = eps_min_ai, so the lowest active OV mode oscillates at zero frequency in the rotating frame (RK4-exact for peak 1). omega_max becomes the full active OV width Delta = eps_max_ai - eps_min_ai. The rotation is undone at I/O so observables remain in the lab frame. Only effective when TDA=.TRUE.. Use with caution, additional convergence checks w.r.t. dt needed.

[Edit on GitHub]