|
votca 2026-dev
|
Static screening by the inactive region of a PbE calculation. More...
#include <inactivescreening.h>
Classes | |
| struct | PoleFit |
| Poles in the form of transitions that reproduce a response at zero and at one imaginary frequency: Y(i w) = sum_k 4 W_k/(W_k^2 + w^2) c_k c_k^T, exactly at w = 0 and w = w_i on the range of Y(0). With Y(0) = G G^T (eigenvectors above tolerance times the largest eigenvalue) and G^+ Y(i w_i) G^+T = Q diag(l) Q^T, l_k = W_k^2/(W_k^2 + w_i^2) and c_k = sqrt(W_k / 4) (G Q)_k. For a genuine sum of transitions every l lies in (0, 1); for a difference of two responses (exact mode) it may not, and such modes keep only their static response (a pole at 1e3 Ha; counted). More... | |
Static Public Member Functions | |
| static InactiveSpace | BuildSpace (const Orbitals &full, Index ncore=0) |
| The inactive space from the full reference orbitals and the inactive density stored by DFTEngine::EvaluateActiveRegion. | |
| static Eigen::MatrixXd | Response (const AOBasis &auxbasis, const AOBasis &dftbasis, const Eigen::MatrixXd &occupied, const Eigen::VectorXd &occupied_energies, const Eigen::MatrixXd &virtuals, const Eigen::VectorXd &virtual_energies, double chunk_bytes=4e9, double imaginary_frequency=0.0) |
| X = sum_ia 4/(e_a - e_i) M_ia M_ia^T over the given occupied and virtual orbitals, in the frame TCMatrix_gwbse::Fill stores its integrals in for auxbasis. | |
| static std::vector< Eigen::MatrixXd > | Responses (const AOBasis &auxbasis, const AOBasis &dftbasis, const Eigen::MatrixXd &occupied, const Eigen::VectorXd &occupied_energies, const Eigen::MatrixXd &virtuals, const Eigen::VectorXd &virtual_energies, double chunk_bytes, const std::vector< double > &imaginary_frequencies, const Eigen::MatrixXd &inv_sqrt=Eigen::MatrixXd()) |
| static Eigen::MatrixXd | Response (const AOBasis &auxbasis, const AOBasis &dftbasis, const InactiveSpace &space, double chunk_bytes=4e9, double imaginary_frequency=0.0) |
| static InactiveScreeningInput | Prepare (const Orbitals &full, InactiveScreeningInput::Mode mode, bool complete_g=false, double radius=0.0) |
| The input for GWBSE, from the full reference orbitals (before they are replaced by the embedded ones). Mode None returns an empty input. Throws for a truncated basis. | |
| static InactiveSpace | NearSpace (const Orbitals &full, double radius, Index &n_far) |
| Block mode with a spatial cutoff: the inactive LMOs whose charge centroid lies within radius (bohr) of the centroid of an active LMO, semicanonical among themselves (the Fock operator of the full calculation diagonalized in their span). The response of the farther inactive orbitals is left out; their coupling to the near ones enters at the same order as the active-inactive coupling of block mode. Needs the LMOs of the localization step. | |
| static tools::EigenSystem | CompleteGOrbitals (const tools::EigenSystem &embedded, Index n_active_occupied, const InactiveScreeningInput &input) |
| The MOs GW-BSE works with for complete G: the inactive occupied orbitals (semicanonical, KS energies), then the active occupied and the virtual embedded MOs; the inactive levels the projector shifted to the top are left out. All of them are occupied for the self-energy and Vxc (full density); the RPA leaves out the first n_inactive holes. | |
| static Eigen::MatrixXd | Kernel (const AOBasis &auxbasis, const AOBasis &dftbasis, const InactiveScreeningInput &input, const tools::EigenSystem &embedded, Index homo, double chunk_bytes, double imaginary_frequency=0.0, Index first_active=0) |
| Y, the response of the inactive region: X_B (block) or X_full - X_A (exact), static or at an imaginary frequency. embedded holds the embedded MOs as GW uses them (active occupied levels 0..homo); the virtuals of X_A are the input's number of virtuals above the HOMO, which leaves out the inactive levels the projector has shifted to the top. | |
| static std::vector< Eigen::MatrixXd > | Kernels (const AOBasis &auxbasis, const AOBasis &dftbasis, const InactiveScreeningInput &input, const TCMatrix_gwbse &active, const Eigen::VectorXd &rpa_energies, Index homo, Index rpamin, Index rpamax, double chunk_bytes, const std::vector< double > &imaginary_frequencies, double virtual_cutoff=0.0) |
| Y at several imaginary frequencies, as GWBSE uses it: one integral pass for the full (exact) or inactive (block) occupied orbitals, with the full calculation's virtuals inside the RPA range (homo+1..rpamax; the inactive levels the projector shifted above them are left out). In exact mode X_A comes from active, the integrals GW holds (rpa_energies for levels rpamin..rpamax, excluded holes left out as in the RPA), so the active integrals are not built a second time. The virtual_cutoff > 0 (Ha above the LUMO) keeps only the virtuals up to it, in both terms of exact mode. The static point is the imaginary frequency eta, as RPA::calculate_epsilon_r at 0 (identical weights). | |
| static PoleFit | FitPoles (const Eigen::MatrixXd &Y0, const Eigen::MatrixXd &Yi, double w_i, double tolerance=1e-10) |
| static Eigen::MatrixXd | ReactionField (const Eigen::MatrixXd &Y, const Eigen::MatrixXd &R_env=Eigen::MatrixXd()) |
| The reaction field R with 1 + R = ((1 + R_env)^-1 + Y)^-1; without an environment (R_env empty) R = (1 + Y)^-1 - 1. | |
Static screening by the inactive region of a PbE calculation.
In the auxiliary frame of the stored three-centre integrals (M = (mn|Q) T, T T^T = V^-1) the static RPA dielectric matrix is 1 + X with X = sum_ia 4/(e_a - e_i) M_ia M_ia^T. Dressing the active integrals with S = (1 + R)^(1/2), R = (1 + Y)^-1 - 1, for a positive semidefinite Y gives S (1 + S X_A S)^-1 S = (1 + X_A + Y)^-1. With Y = X_full - X_A ("exact") this is the static screened interaction of the full calculation; with Y = X_B over the semicanonical inactive orbitals ("block") it is that up to the active-inactive Fock coupling. R enters GW and BSE like the reaction field of a polarizable environment (GW::setReactionField, BSE::setReactionField), and combines with one: 1 + R = ((1 + R_env)^-1 + Y)^-1, the Dyson equation in which the inactive region and the environment polarize each other.
Definition at line 118 of file inactivescreening.h.
|
static |
The inactive space from the full reference orbitals and the inactive density stored by DFTEngine::EvaluateActiveRegion.
full must still hold the MOs of the full reference calculation (not the embedded ones). The lowest ncore semicanonical inactive orbitals are left out (frozen core).
Definition at line 38 of file inactivescreening.cc.
|
static |
The MOs GW-BSE works with for complete G: the inactive occupied orbitals (semicanonical, KS energies), then the active occupied and the virtual embedded MOs; the inactive levels the projector shifted to the top are left out. All of them are occupied for the self-energy and Vxc (full density); the RPA leaves out the first n_inactive holes.
Definition at line 275 of file inactivescreening.cc.
|
static |
Definition at line 377 of file inactivescreening.cc.
|
static |
Y, the response of the inactive region: X_B (block) or X_full - X_A (exact), static or at an imaginary frequency. embedded holds the embedded MOs as GW uses them (active occupied levels 0..homo); the virtuals of X_A are the input's number of virtuals above the HOMO, which leaves out the inactive levels the projector has shifted to the top.
Definition at line 296 of file inactivescreening.cc.
|
static |
Y at several imaginary frequencies, as GWBSE uses it: one integral pass for the full (exact) or inactive (block) occupied orbitals, with the full calculation's virtuals inside the RPA range (homo+1..rpamax; the inactive levels the projector shifted above them are left out). In exact mode X_A comes from active, the integrals GW holds (rpa_energies for levels rpamin..rpamax, excluded holes left out as in the RPA), so the active integrals are not built a second time. The virtual_cutoff > 0 (Ha above the LUMO) keeps only the virtuals up to it, in both terms of exact mode. The static point is the imaginary frequency eta, as RPA::calculate_epsilon_r at 0 (identical weights).
Definition at line 324 of file inactivescreening.cc.
|
static |
Block mode with a spatial cutoff: the inactive LMOs whose charge centroid lies within radius (bohr) of the centroid of an active LMO, semicanonical among themselves (the Fock operator of the full calculation diagonalized in their span). The response of the farther inactive orbitals is left out; their coupling to the near ones enters at the same order as the active-inactive coupling of block mode. Needs the LMOs of the localization step.
Definition at line 168 of file inactivescreening.cc.
|
static |
The input for GWBSE, from the full reference orbitals (before they are replaced by the embedded ones). Mode None returns an empty input. Throws for a truncated basis.
Definition at line 228 of file inactivescreening.cc.
|
static |
The reaction field R with 1 + R = ((1 + R_env)^-1 + Y)^-1; without an environment (R_env empty) R = (1 + Y)^-1 - 1.
Definition at line 415 of file inactivescreening.cc.
|
static |
X = sum_ia 4/(e_a - e_i) M_ia M_ia^T over the given occupied and virtual orbitals, in the frame TCMatrix_gwbse::Fill stores its integrals in for auxbasis.
The occupied orbitals are processed in chunks of at most chunk_bytes of integrals, so the full tensor is never stored. With imaginary_frequency w > 0 the weights are 4D/(D^2 + w^2), the response at i w as RPA::calculate_epsilon_i builds it.
Definition at line 134 of file inactivescreening.cc.
|
inlinestatic |
Definition at line 156 of file inactivescreening.h.
|
static |
Response at several imaginary frequencies from one integral pass. inv_sqrt: the metric V^-1/2 of auxbasis (TCMatrix_gwbse::InvSqrt), if already known; otherwise computed once for all chunks.
Definition at line 86 of file inactivescreening.cc.