RESPONSE_SOLVER
References: VandeVondele2012
Specifies the parameters of the linear scaling solver routines [Edit on GitHub]
Keywords
Keyword descriptions
- EPS: real = 1.00000000E-012
Usage: EPS 1.e-10
Target accuracy for the convergence of the conjugate gradient. [Edit on GitHub]
- EPS_FILTER: real = 1.00000000E-010
Usage: EPS_FILTER 1.0E-10
Threshold used for filtering matrix operations. [Edit on GitHub]
- EPS_LANCZOS: real = 1.00000000E-003
Usage: EPS_LANCZOS 1.0E-4
Threshold used for lanczos estimates. [Edit on GitHub]
- MATRIX_CLUSTER_TYPE: enum = ATOMIC
Usage: MATRIX_CLUSTER_TYPE MOLECULAR
Valid values:
ATOMIC
Using atomic blocksMOLECULAR
Using molecular blocks.
Specify how atomic blocks should be clustered in the used matrices, in order to improve flop rate, and possibly speedup the matrix multiply. Note that the atomic s_preconditioner can not be used. Furthermore, since screening is on matrix blocks, slightly more accurate results can be expected with molecular. [Edit on GitHub]
- MAX_ITER: integer = 50
Usage: MAX_ITER 200
Maximum number of conjugate gradient iteration to be performed for one optimization. [Edit on GitHub]
- MAX_ITER_LANCZOS: integer = 128
Usage: MAX_ITER_LANCZOS 128
Maximum number of lanczos iterations. [Edit on GitHub]
- METHOD: enum = AO_ORTHO
Usage: METHOD SOLVER
Valid values:
MO_SOLVER
Solver based on MO (vector) formalismAO_ORTHO
Solver based on density matrix formalism
Algorithm used to solve response equation. Both solver are conjugate gradient based, but use either a vector (MO-coefficient) or density matrix formalism in the orthonormal AO-basis to obtain response density [Edit on GitHub]
- PRECONDITIONER: enum = MULTI_LEVEL
Usage: PRECONDITIONER FULL_ALL
Valid values:
FULL_ALL
Most effective state selective preconditioner based on diagonalization, requires the ENERGY_GAP parameter to be an underestimate of the HOMO-LUMO gap. This preconditioner is recommended for almost all systems, except very large systems where make_preconditioner would dominate the total computational cost.FULL_SINGLE_INVERSE
Based on H-eS cholesky inversion, similar to FULL_SINGLE in preconditioning efficiency but cheaper to construct, might be somewhat less robust. Recommended for large systems.FULL_SINGLE
Based on H-eS diagonalisation, not as good as FULL_ALL, but somewhat cheaper to apply.FULL_KINETIC
Cholesky inversion of S and T, fast construction, robust, and relatively good, use for very large systems.FULL_S_INVERSE
Cholesky inversion of S, not as good as FULL_KINETIC, yet equally expensive.MULTI_LEVEL
Based on same CG as AO-solver itself, but uses cheaper linear transformationNONE
skip preconditioning
Type of preconditioner to be used with MO conjugate gradient solver. They differ in effectiveness, cost of construction, cost of application. Properly preconditioned minimization can be orders of magnitude faster than doing nothing. Only multi-level conjugate gradient preconditioner (MULTI_LEVEL) available for AO response solver (AO_ORTHO). [Edit on GitHub]
- RESTART: logical = F
Lone keyword:
T
Usage: RESTART
Restart the response calculation if the restart file exists [Edit on GitHub]
- RESTART_EVERY: integer = 50
Usage: RESTART_EVERY 50
Restart the conjugate gradient after the specified number of iterations. [Edit on GitHub]
- S_INVERSION: enum = SIGN_SQRT
Usage: S_PRECONDITIONER MOLECULAR
Valid values:
SIGN_SQRT
Using the inverse sqrt as obtained from sign function iterations.HOTELLING
Using the Hotellign iteration.
Method used to compute the inverse of S. [Edit on GitHub]
- S_PRECONDITIONER: enum = ATOMIC
Usage: S_PRECONDITIONER MOLECULAR
Valid values:
NONE
No preconditionerATOMIC
Using atomic blocksMOLECULAR
Using molecular sub-blocks. Recommended if molecules are defined and not too large.
Preconditions S with some appropriate form. [Edit on GitHub]
- S_SQRT_METHOD: enum = NEWTONSCHULZ
Usage: S_SQRT_METHOD NEWTONSCHULZ
Valid values:
NEWTONSCHULZ
Using a Newton-Schulz-like iterationPROOT
Using the p-th root method.
Method for the caclulation of the sqrt of S. [Edit on GitHub]
- S_SQRT_ORDER: integer = 3
Aliases: SIGN_SQRT_ORDER
Usage: S_SQRT_ORDER 3
Order of the iteration method for the calculation of the sqrt of S. [Edit on GitHub]