This keyword controls the functioning of the SCF procedure. Options are used to specify the desired behavior, alternate algorithms, and so on.
The default SCF procedure uses a combination of EDIIS [Kudin02] and CDIIS, with no damping or Fermi broadening. In Gaussian 09, SCF=Tight is the default.
The SCF=QC option is often helpful with difficult conversion cases. For difficult-to-converge ROHF wavefunctions, where QC cannot be used, add Use=L506 to the route section.
See reference [Schlegel91a] for a discussion of SCF convergence and stability.
ALGORITHM SELECTION OPTIONS
DIIS calls for and NoDIIS prohibits use of Pulay’s Direct Inversion in the Iterative Subspace (DIIS) extrapolation method [Pulay82].
Use only CDIIS. CDIIS implies Damp as well.
Requests temperature broadening during early iterations [Rabuck99], combined with CDIIS and damping. NoFermi suppresses Fermi broadening and is the default. By default, Fermi also implies Damp and also includes level shifting.
Turn on dynamic damping of early SCF iterations. NoDamp is the default. However, damping is enabled if SCF=Fermi or SCF=CDIIS is requested. Note that damping and EDIIS do not work well together.
Allow dynamic damping for up to N SCF iterations (the default is 10).
Calls for the use of a quadratically convergent SCF procedure [Bacskay81]. By default this involves linear searches when far from convergence and Newton-Raphson steps when close (unless the energy goes up). This method is slower than regular SCF with DIIS extrapolation but is more reliable. SCF=QC is not available for restricted open shell (RO) calculations.
Add an extra SCF=QC step in case the first-order SCF has not converged.
Sets the limit on conventional SCF cycles during SCF=XQC to N.
Use pseudo-diagonalization in Link 502 whenever possible, with full diagonalization only at the early cycles, at the end, and every Nth cycle in between. PDiag is a synonym for this option. This is the default for semi-empirical methods (the default is N=30).
Forces full diagonalization in Link 502. This is the default for HF and DFT. FDiag is a synonym for this option.
Does steepest descent SCF
Does scaled steepest descent SCF.
Save k-point information at the conclusion of the SCF. NoSaveKPoint says not to save this data, and it is the default except for numerical frequency calculations for which SaveKPoint is the default.
Calls for use of the direct minimization SCF program [Seeger76]. It is usually inferior to SCF=QC and retained for backwards compatibility and as a last resort. Available only for RHF closed shell and UHF open shell calculations.
Shift orbital energies by N*0.001 (i.e., N milliHartrees); N defaults to 100. This option disables automatic archiving. N=-1 disables level shifting; NoVShift is equivalent to this setting.
Changes the maximum number of SCF cycles permitted to N; the default is 128 (or 512 for SCF=DM and SCF=QC).
Specifies that L508 (SCF=QC, SD, or SSD) should do full linear searches at each iteration. By default, a full minimization is done only if the initial microiteration caused the energy to go up.
Set the maximum rotation gradient for a Newton-Raphson step in SCF=QC to 10-N. Above this, scaled steepest descent is used; above 100 times this, steepest descent is used. The default value for N is 2.
FinalIteration performs and NoFinalIteration prevents a final non-extrapolated, non-incremental iteration after an SCF using DIIS or a direct SCF has converged. The default is NoFinalIteration.
Forces use of incremental Fock matrix formation. This is the default for direct SCF. NoIncFock prevents the use of incremental Fock matrix formation, and it is the default for conventional SCF.
For in-core calculations, saves the integrals on disk as well, to avoid recomputing them in Link 1002. Only useful for frequency jobs in conjunction with SCF=InCore. NoPass forces integrals to be recomputed during each in-core phase.
Use tight convergence in linear equation solution throughout SCF=QC. By default, the convergence criterion is tightened up as the rotation gradient is reduced.
Use even tighter convergence in the linear equation solutions (microiterations) throughout the QCSCF. This option is sometimes needed for nearly linearly-dependent cases. VTL is a synonym for VeryTightLinEq.
INTEGRAL STORAGE OPTIONS
Requests a direct SCF calculation, in which the two-electron integrals are recomputed as needed. This is the default SCF procedure in Gaussian. This is possible for all available methods, except for MCSCF second derivatives and anything using complex orbitals.
Insists that the SCF be performed storing the full integral list in memory. This is done automatically in a direct SCF calculation if sufficient memory is available. SCF=InCore is available to force in-core storage or abort the job if not enough is available. NoInCore prohibits the use of the in-core procedure, for both the SCF and CPHF.
The two-electron integrals are stored on disk and read-in each SCF iteration. NoDirect is a synonym for Conventional.
OPTIONS RELATED TO CONVERGENCE AND CUTOFFS
Sets the SCF convergence criterion to 10-N. SCF convergence requires both <10-N RMS change in the density matrix and <10-(N-2) maximum change in the density matrix. Note that the energy change is not used to test convergence; however, an SCF 10-N RMS density matrix change typically corresponds to a 10-2N change in energy in atomic units. For GVB and CASSCF calculations, SCF convergence is determined not by change in the density matrix, but rather in terms of the orbital change and energy change, respectively.
Use modest integral accuracy early in direct SCF, switching to full accuracy later on. This is the default for direct SCF, and it can be turned off via NoVarAcc. VarInt is a synonym for VarAcc, and NoVarInt is a synonym for NoVarAcc.
Use normal, tight convergence in the SCF. This is the default. Synonymous with NoSinglePoint, NoSP and TightIntegrals.
Turns off optional O(N3) steps to speed up very large calculations (>5000 basis functions).
Provides a new algorithm which is useful for difficult SCF convergence cases involving very large molecules. It does steepest descent and then scaled steepest descent as in SCF=QC, but then switches to regular SCF instead of quadratic convergence, using the quadratic algorithm only if the regular SCF fails to converge.
Sets the threshold for switching to quadratic convergence in SCF=QC and to regular SCF in SCF=YQC to 10-N. The default is 10-2.
Symmetrize the density matrix at the first iteration to match the symmetry of the molecule (“initial density symmetrize”). NoIDSymm is the default.
Symmetrize the density matrix at every SCF iteration to match the symmetry of the molecule (“density symmetrize”). NoDSymm is the default. DSymm implies IDSymm.
Requests that all orbital symmetry constraints be lifted. It is synonymous with Guess=NoSymm and Symm=NoSCF.
Retain all symmetry constraints: make the number of occupied orbitals of each symmetry type (abelian irreducible representation) match that of the initial guess. Use this option to retain a specific state of the wavefunction throughout the calculation. It is the default only for GVB calculations.
Calls for the SCF procedure to account for integral symmetry by replicating the integrals using the symmetry operations. Allows use of a short integral list even if the wavefunction does not have the full molecular symmetry. Available for L502 (the default for RHF, ROHF and UHF) and L508 (SCF=QC).
Calls for the SCF procedure to account for integral symmetry (use of the petite integral list) by symmetrizing the Fock matrices. This is the default. FSymm is a synonym for FockSymm.
Save the wavefunction on the checkpoint file every iteration, so the SCF can be restarted. This is the default for direct SCF. NoSave suppresses saving the wavefunction.
Restart the SCF from the checkpoint file. SCF=DM cannot be restarted.
SCF=Restart skips steps which are not necessary when restarting an SCF calculation, but which are necessary when reading in a guess from a calculation with a different basis set or at a different geometry. In contrast, if you want to start a new SCF using the restart information from a calculation with a different geometry and/or a different basis set, use Guess=Restart.
Last update: 17 February 2016