CFOUR | Main / Additional Keywords For CASSCF Calculations

CAS_INPUT: specifies how the input for a CASSCF calculation is provided. Options:

a) ORBITALS (=0, default): a list of orbital numbers, as printed after the SCF solution, is provided as a list of integers separated by spaces.

b) OCCUPATION (=1): the number of internal (i.e., doubly occupied) orbitals is provided for each irrep as a list of integers separated by one space and, on a new line, the list of active orbitals is provided for each irrep, as a list of integers separated by one space

CAS_GUESS: specifies what kind of guess to use for the CASSCF wavefunction.

a) CANONICAL (=0, default): use the Hartree-Fock wavefunction as a guess. For UHF orbitals, this corresponds to using alpha orbitals only.

b) CASCI (=1): starting from the Hartree-Fock orbitals, compute a CAS-CI wavefunction and use it as a guess. This might help convergence.

c) MODCASCI (=2): experimental use.

d) CORECASCI (=3): computes a CASCI guess using the core orbitals.

CAS_PRINT=XY: printing flag. Y=1 prints information on the microiterations convergence, X=1,2,3 prints (more and more detailed) timing informations at each iteration.
CAS_ITMAX=N: Max number of macroiterations. Default is 50, which should be plenty.
CAS_MITMAX=N: Max number of microiterations. Default is 100
CAS_MMAX=M: largest allowed subspace size for the microiterations, which require a Davidson-like diagonalization
CAS_GECCO: (ON/OFF) specifies whether to set up the files for a subsequent internally contracted MRCC calculation using the GeCCo code by A. Koehn and M. Hanhauer. Default: OFF (=0)

Advanced options
The following keywords control the algorithm used to solve the microiterations and some internal parameters and should be used with caution

QC_RTRUST=X: X/1000 specifies the starting trust radius for the adaptive restricted-step second order optimization procedure. Default: 700
CAS_MICRO: specifies what preconditioner to use for the microiterations. The options are

a) ILU/JACOBI (=0, default): use the incomplete LU decomposition (ILU) to precondition the Davidson iterations for the MO part. If the Jacobi-Davidson algorithm has been chosen, use the standard diagonal preconditioner for the first iterations and the ILU one to precondition the GMRES solution to the Jacobi-Davidson linear equations

b) JACOBI (=1): use the standard Davidson diagonalization, using a diagonal (i.e., Jacobi) preconditioner

c) ILU (=2): use the incomplete LU decomposition of the MO rotation Hessian to precondition the orbital updates in the microiterations. This is cheap and effective and is used as a default

d) EIG(=3): Explicitly diagonalize the MO rotation Hessian. This option is mainly for debugging purposes, but might be used with caution for tough cases.

CAS_JACDAV: (ON/OFF) specifies whether the Jacobi-Davidson (JD) scheme should be used instead than Davidson for the microiterations. Either the ILU (default) or EIG (with CAS_MICRO=EIG) preconditioner is used to solve the linear equations. JD is an expensive, yet very powerful method to find one or few eigenvectors of a matrix. It can be used for very difficult cases.
CAS_MICMIN=N: the minimum (loosest) convergence criterion to be used for the microiterations is set to 10^-N. Default: 2
CAS_MICMAX=N: the maximum (tightest) convergence criterion to be used for the microiterations is set to 10^-N. Default: 10