|
| const Eigen::MatrixXd & | operator[] (Index i) const |
| Eigen::MatrixXd & | operator[] (Index i) |
| Index | auxsize () const |
| Index | get_mmin () const |
| Index | get_mmax () const |
| Index | get_nmin () const |
| Index | get_nmax () const |
| Index | msize () const |
| Index | nsize () const |
| void | Initialize (Index basissize, Index mmin, Index mmax, Index nmin, Index nmax) |
| void | Fill (const AOBasis &auxbasis, const AOBasis &dftbasis, const Eigen::MatrixXd &dft_orbitals) |
| void | Fill (const AOBasis &auxbasis, const AOBasis &dftbasis, const Eigen::MatrixXd &dft_orbitals, const Eigen::MatrixXd &inv_sqrt) |
| void | Rebuild () |
| void | MultiplyRightWithAuxMatrix (const Eigen::MatrixXd &matrix) |
| void | MultiplyRightWithAuxMatrixLeading (const Eigen::MatrixXd &U, Index full_slices, Index lead) |
| | Rotate the auxiliary index of only the part of the tensor a consumer reads: all rows of slices [0, full_slices) and the first lead rows of slices [full_slices, lead).
|
| bool | PartiallyRotated () const |
| | Whether only a leading part was rotated since the last Fill.
|
| const Eigen::MatrixXd & | InvSqrt () const |
| const Eigen::MatrixXd & | AuxFrame () const |
| void | RestoreAuxFrame (const Eigen::MatrixXd &frame) |
| Eigen::MatrixXd | ToCurrentAuxFrame (const Eigen::MatrixXd &K) const |
| bool | AuxFrameIsOrthogonal () const |
| void | DressAuxIndex (const Eigen::MatrixXd &S) |
| | Dress the auxiliary index with S: M_fill -> M_fill S.
|
| void | UndressAuxIndex () |
| bool | Dressed () const |
| const Eigen::MatrixXd & | Dressing () const |
| | S of the current dressing (empty if bare).
|
| void | setAdditionalResponse (std::vector< ResponsePoint > points) |
| bool | HasAdditionalResponse () const |
| Eigen::MatrixXd | AdditionalResponse (bool imaginary, double frequency) const |
| | F^T Y F at this frequency, in the frame operator[] is in now.
|
| void | setAdditionalPoles (Eigen::VectorXd energies, Eigen::MatrixXd vectors) |
| | A response given as poles, in the form of transitions: Y(i w) = sum_k 4 W_k/(W_k^2 + w^2) c_k c_k^T, with energies W_k and vectors c_k (columns, fill-time frame). The RPA treats each pole as one more transition with energy W_k and integrals c_k: in epsilon at any frequency, and in the two-particle Hamiltonian of Sigma_Exact, where the poles couple to the transitions of M through the Coulomb interaction. E.g. the inactive region of a projection-based embedding, fitted to its response at two frequencies (InactiveScreening::FitPoles).
|
| Index | AdditionalPoleCount () const |
| const Eigen::VectorXd & | AdditionalPoleEnergies () const |
| Eigen::MatrixXd | AdditionalPoleVectors () const |
| | The pole vectors in the frame operator[] is in now: F^T c_k.
|
| void | setExcludedHoles (Index k) |
| | The first k occupied slices (levels rpamin .. rpamin+k-1) are left out of the RPA response sums, but stay in the integrals, so the self-energy still sums over them in G. For a projection-based embedding whose inactive occupied orbitals were put in front of the active ones (InactiveScreening, complete G): the inactive region then screens only through its additional response or poles. Survives Fill. Their slices are not stored (only their rows in the other slices, which G needs): set it before Fill. Nothing may read those slices.
|
| Index | ExcludedHoles () const |
| void | RotateOrbitals (const Eigen::MatrixXd &U, Index first, Index last) |
| | Change the orbital basis of levels [first, last] for QSGW: phi_i = sum_k U_ki psi_k, both orbital indices, all slices.
|
| | TCMatrix ()=default |
| virtual | ~TCMatrix ()=default |
| | TCMatrix (const TCMatrix &)=default |
| | TCMatrix (TCMatrix &&)=default |
| TCMatrix & | operator= (const TCMatrix &)=default |
| TCMatrix & | operator= (TCMatrix &&)=default |
| Index | Removedfunctions () const |
Definition at line 146 of file threecenter.h.
| void votca::xtp::TCMatrix_gwbse::DressAuxIndex |
( |
const Eigen::MatrixXd & | S | ) |
|
Dress the auxiliary index with S: M_fill -> M_fill S.
For a static environment the QM electrons interact through u = v + v_reac, which is 1 + R in the metric of M, and the screened interaction is W = [u^-1 - chi0]^-1. With S = (1 + R)^(1/2), that is exactly the ordinary RPA and correlation self-energy built from M S in place of M: every place that pairs two M's through the bare interaction pairs them through u instead.
S is in the fill-time frame, and is applied in whatever frame the integrals are in now; AuxFrame() records it, so ToCurrentAuxFrame still maps fill-time kernels correctly. Exchange-type sums that must keep the bare v take ToCurrentAuxFrame(identity) as their kernel once the frame is no longer orthogonal (see AuxFrameIsOrthogonal).
Throws if already dressed. Fill and Rebuild undress.
Definition at line 169 of file threecenter.cc.
| void votca::xtp::TCMatrix_gwbse::MultiplyRightWithAuxMatrixLeading |
( |
const Eigen::MatrixXd & | U, |
|
|
Index | full_slices, |
|
|
Index | lead ) |
Rotate the auxiliary index of only the part of the tensor a consumer reads: all rows of slices [0, full_slices) and the first lead rows of slices [full_slices, lead).
The BSE reads slices and rows up to its highest conduction level (lead) and, for the dielectric matrix, all rows of the occupied slices (full_slices). Rotating only those is exact for that use and skips the rows above the BSE window. The rest of the tensor is left in the old frame and is stale afterwards: AuxFrame() describes the rotated part, and dressing, which needs the whole tensor in one frame, throws until the next Fill or Rebuild.
Definition at line 108 of file threecenter.cc.
| void votca::xtp::TCMatrix_gwbse::RotateOrbitals |
( |
const Eigen::MatrixXd & | U, |
|
|
Index | first, |
|
|
Index | last ) |
Change the orbital basis of levels [first, last] for QSGW: phi_i = sum_k U_ki psi_k, both orbital indices, all slices.
rows n in the window of every slice: M[m].rows <- U^T M[m].rows slices in the window: M[i] <- sum_k U_ki M[k]
Levels outside the window keep their orbitals. Everything that reads the tensor (RPA, self-energy, exchange) then works in the new basis without knowing about U. The auxiliary index is not touched.
- Parameters
-
| U | Rotation matrix (window x window), columns = new orbitals |
| first | First level of the window (absolute MO index) |
| last | Last level of the window (absolute MO index) |
Definition at line 308 of file threecenter.cc.
| void votca::xtp::TCMatrix_gwbse::setAdditionalPoles |
( |
Eigen::VectorXd | energies, |
|
|
Eigen::MatrixXd | vectors ) |
|
inline |
A response given as poles, in the form of transitions: Y(i w) = sum_k 4 W_k/(W_k^2 + w^2) c_k c_k^T, with energies W_k and vectors c_k (columns, fill-time frame). The RPA treats each pole as one more transition with energy W_k and integrals c_k: in epsilon at any frequency, and in the two-particle Hamiltonian of Sigma_Exact, where the poles couple to the transitions of M through the Coulomb interaction. E.g. the inactive region of a projection-based embedding, fitted to its response at two frequencies (InactiveScreening::FitPoles).
Definition at line 323 of file threecenter.h.
| void votca::xtp::TCMatrix_gwbse::setExcludedHoles |
( |
Index | k | ) |
|
The first k occupied slices (levels rpamin .. rpamin+k-1) are left out of the RPA response sums, but stay in the integrals, so the self-energy still sums over them in G. For a projection-based embedding whose inactive occupied orbitals were put in front of the active ones (InactiveScreening, complete G): the inactive region then screens only through its additional response or poles. Survives Fill. Their slices are not stored (only their rows in the other slices, which G needs): set it before Fill. Nothing may read those slices.
Definition at line 83 of file threecenter.cc.