7.10. CP-SCF Options¶
The coupled perturbed self-consistent field (CP-SCF) equations have to be solved in many cases, such as when second derivative properties (e.g. vibrational frequencies, polarizability, NMR shielding, indirect spin-spin coupling, hyperfine coupling, g-tensor) or the MP2 relaxed density (in this case they are referred to as Z-vector equations) are calculated. They are a set of linear equations generally expressed as
where \(\mathbf U^x\) is the vector of solutions for perturbation \(x\), the right-hand side (RHS) matrix \(\mathbf B^x\) is perturbation-specific and the left-hand side (LHS) matrix \(\mathbf A\) is perturbation-independent and contains, among other terms, the two-electron repulsion integrals \(\left(ij\vert ab\right)\) and \(\left(ia\vert jb\right)\). The equations are solved iteratively and the LHS is reassembled at every step, while the RHS does not change. The generation and transformation of the two-electron integrals are therefore the most time-consuming parts of the CP-SCF solution.
The ORCA module which solves these equations accepts several options given below with their default values:
%method
Z_Solver Pople # (default) Use the Pople algorithm to solve the equations
DIIS # Use the DIIS algorithm
CG # Use the conjugate gradient algorithm
Z_Tol 1e-3 # Convergence tolerance for the residual norm.
# Default is 1e-5 for VeryTightOpt
# and varies from 3e-3 to 3e-6 from LooseSCF to ExtremeSCF
Z_MaxIter 128 # Maximum number of iterations
Z_MaxDIIS 12 # Maximum number of DIIS vectors
Z_Shift 0.3 # Level shift for DIIS
Z_GridXC 1 # XC angular grid used for the LHS
Z_IntAccXC 3.467 # XC radial grid accuracy used for the LHS
Z_GridX 1 # COSX angular grid used for the LHS
Z_IntAccX 3.067 # COSX radial grid accuracy used for the LHS
Z_GridX_RHS 2 # COSX grid used for the RHS of MP2 Z-vector eqs (see below)
Z_COSX_Alg 0 # (default) choose the best COSX algorithm automatically
1 # better prescreening, more efficient for few densities
2 # uses more memory, more efficient for many densities
end
Since ORCA 6, the same settings are used for all electric response
property calculations as well as for CIS/TD-DFT gradients and relaxed densities.
For convenience, the keywords in the %elprop
input block are still
available but they modify the same internal variables as those in %method
.
For magnetic response properties, the solver and convergence tolerance are
set separately in %eprnmr
, because the convergence behavior of the magnetic
response CP-SCF equations is sometimes different.
%elprop
Solver # Alias, see: %method Z_Solver
Tol # Alias, see: %method Z_Tol
MaxIter # Alias, see: %method Z_MaxIter
MaxDIIS # Alias, see: %method Z_MaxDIIS
LevelShift # Alias, see: %method Z_Shift
end
%eprnmr
Solver # Solver for magnetic response, see options at: %method Z_Solver
Tol # Convergence tolerance for magnetic response
MaxIter # Alias, see: %method Z_MaxIter
MaxDIIS # Alias, see: %method Z_MaxDIIS
LevelShift # Alias, see: %method Z_Shift
end
The keywords Z_GridX
and Z_IntAccX
are applicable if the
RIJCOSX approximation is chosen for the
treatment of two-electron integrals. They determine the angular and
radial COSX integration grids, as discussed in
section Changing TD-DFT, CP-SCF and Hessian grids. Analogously, the keywords
Z_Grid
and Z_IntAcc
determine the integration grid for DFT XC
functionals.
Integrals on the RHS are evaluated differently for different
perturbations - refer to
sections Using the RI Approximation for Hartree-Fock and Hybrid DFT (RIJCOSX),
EPR and NMR properties,
RIJCOSX-RI-MP2 Gradients, and
MP2 and RI-MP2 Second Derivatives and
RI-MP2 and Double-Hybrid DFT Response Properties for SCF-level gradients,
EPR/NMR calculations with GIAOs, MP2 gradients, and MP2 second
derivatives, respectively. For MP2 Z-vector equations, the RIJCOSX
Fock-response terms in the RHS are evaluated with the COSX grid
specified by Z_GridX_RHS
. Note that it is used differently to
Z_GridX
: instead, it selects one of the three grids used in the SCF
(see
Sections Using the RI Approximation for Hartree-Fock and Hybrid DFT (RIJCOSX),
COSX Grid and Convergence Issues, and
Details on the numerical integration grids for details) and it is not recommended
to change the default value of 2.
If the RIJONX or RIJK approximation is used in the SCF, the same is also employed in the CP-SCF. Note, however, that the RI-K approximation is not efficient for these terms.