SCF input parameters

Convergence threshold

    THRESH tol

    The convergence criterion is on the maximum value of the gradient
    vector.  The default value is 10^-6.


Maximum number of iterations

    MAXITER n

    default is 5 ... usually only 3 are needed.

    e.g.

        maxiter 10


Input/output MO vectors

    VECTORS [input] <atomic|hcore|filename> [output <filename>]

     output vectors default to either the same filename as the
     input vectors if they were read from a file or to the
     filename movecs (which is searched for by the application).

     E.g.

     These two are equivalent and implement the default

         vectors atomic output movecs

         vectors input atomic output movecs

     This gets input vectors from one file and outputs final
     vectors back to the same file

         vectors h2o.movecs

     This gets input vectors from one file and outputs final
     vectors to another file

        vectors input h2o.guess output h2o.final



SCF level shifting

    LEVEL [pcg initial [tol final]] [nr initial [tol final]]

    Both the 1-electron preconditioned conjgugate gradient (PCG) and
    the Newton-Raphson (NR) use level shifting to stabilize convergence
    (by making the preconditioning matices more positive definite and
    diagonally dominate).  For both you can specify an initial level
    shift, a gradient value at which to change shifts, and a final
    level shift.  The defaults

       pcg initial = 10.0
       pcg change shift at max. gradient = 0.5
       pcg final = 0.0
       nr  initial = 10.0
       nr  change shift at max. gradient = 0.5
       nr  final = 0.0

    are almost always acceptable because the program will increase
    the level shift if convergence problems are encountered with the
    preconditioning.  However, some cases do require increased level
    shifts to speed convergence

    e.g.

      This enables the default options

      level nr 0 0 0 pcg 10 .5 0


Controlling use of newton-raphson quadratically convergent algorithm

    NR tol

    The QC algorithm (use of exact hessian preconditioner) is enabled
    by default when the maximum element of the orbital gradient falls
    below 0.1.  This value can be changed using this directive

    e.g.

       This enables the default

       NR 0.1

       This disables the QCSCF entirely

       NR 0.0


Atomic guess shift (subject to change)

     SET "scf:atom shift" real value

     The atomic guess routines form a density matrix (D) from a superposition
     of the atomic density matrices.  Vectors are obtained from diagonalizing
     the matrix

        Fock(D) - shift * SDS

     If the density matrix was idempotent then this is equivalent to 
     standard level shifting.  The default value of shift is 0.0 which seems
     best for 'normal' molecules.  Values of 1-2 can help with transition metals.
     There is no special input for this parameter so U have to use the generic
     set directive.