34.2.3 Selecting the optimization method (METHOD)

METHOD,key;

key defines the optimization method.

For minimization the following options are valid for key:

RF

Rational Function method (default).

AH

Augmented Hessian method. This is similar to RF algorithm but uses a more sophisticated step restriction algorithm.

DIIS

Pulay's Geometry DIIS method. As an an additional option you may add the number of geometries to be used in GDIIS interpolation (default 5) and the interpolation type (i.e. the subspace in which the GDIIS interpolation is made.
METHOD,DIIS, number, type
type may be GRAD interpolation using the gradients (default), working good for rigid molecules, STEP interpolation using Quasi-Newton steps which could be advantageous in dealing with very floppy molecules, ENER interpolation using energies, which is an intermediate between the above two.

QSD

Quadratic steepest descent method of Sun and Ruedenberg.

SRMIN

Old version of QSD.

For transition state searches (invoked with the ROOT option, see section 30.2.9) key can be

RF

Rational Function method (default).

DIIS

Pulay's Geometry DIIS method (see above).

QSD

Quadratic Steepest Descent Transition State search using the image hessian method (see J. Sun and K. Ruedenberg, J. Chem. Phys. 101, 2157 (1994)) The use of this option is recommended for transition state searches - especially in complicated cases. The optimization step is checked and the hessian is recalculated when approaching a troublesome region of the PES. Thus this method is somewhat safer (and often faster) in reaching convergence than the RF or DIIS method. The hessian recalculation safeguard may be turned off using the METHOD,QSD,NOHESS input card.

SRTRANS

Old version of QSD.

For reaction path following the input key is

QSDPATH

Quadratic Steepest Descent reaction path following. This methods determines reaction paths (intrinsic reaction coordinates, IRCs) by following the exact steepest descent lines of subsequent quadratic approximations to the potential energy surface. The hessian matrix is calculated numerically at the first optimization step and subsequently updated by Powell or BFGS update. If a given arc length of the steepest descent lines is exceeded, the hessian is recalculated numerically (see OPTION section 30.2.20). For details see J. Sun and K. Ruedenberg, J. Chem. Phys. 99, 5269 (1993) It is also possible to recalculate the hessian after each m steps using the NUMHES,m command (see section 30.2.18). If the hessian matrix is recalculated in every optimization step (NUMHES,$1$) a algorithm different to the one with updated hessians is used, which is very accurate. Using the PRINT,OPT card, this algorithm prints in every optimization step a reaction path point r which is different from the point where the energy and the gradient is calculated but closer to the real reaction path (for further details of the algorithm see J. Sun and K. Ruedenberg, J. Chem. Phys. 99, 5257 (1993)). For further input options of the QSD reaction path following see OPTION section 30.2.20.

SRSTEEP

Old Version of QSDPATH.