Next: 34.2.9 Transition state (saddle Up: 34.2 Geometry optimization Previous: 34.2.7 Restarting a geometry

34.2.8 Approximating hessian matrix elements (HESSIAN)

HESSIAN[,key], value, param1, param2;

where key can be

MODEL
Use Lindh's Model Hessian in optimization (default).
MODEL,SCHLEGEL
Use Schlegel Model Hessian.
MODEL,VDW
Add vdW terms to Lindh's Model Hessian.
SCHLEGEL
Same as MODEL,SCHLEGEL]
VDW
Same as MODEL,VDW
NOMODEL
Don't use Model Hessian approximation to the hessian.

For minimizations, the Model Hessian provides very good approximations the Hessian matrix, improving convergence rapidly, so it was chosen as default. At present it is implemented for the first three row elements.

You may also put in individual matrix elements of the hessian: value sets starting value for hessian matrix element between parameters param1, param2. If param2 is omitted it defaults to param1 (diagonal element). If the Model Hessian is disabled, the initial hessian is diagonal, with values 1hartree*bohr**(-2) for all lengths, 1 hartree*radian**(-2) for all angles. This is usually quite reasonable except for cases such as dihedral angles. A reasonable strategy for complicated cases is to perform an optimization with a small basis set at the SCF level with PRINT,HESSIAN in order to obtain an approximate starting hessian. These values are set before processing the START record (see above). This option is obsolete if the Model Hessian is used (default unless heavy elements are present).

In transition state searches the hessian matrix is evaluated numerically (see NUMHES section 30.2.18). Alternatively, the cartesian hessian matrix evaluated in a previous frequency calculation (see FREQUENCIES section 30.4) can be used with the HSTART command (see section 30.2.19). It is also possible to use the numerical hessian or the hessian from a frequency calculation in minimizations.



Next: 34.2.9 Transition state (saddle Up: 34.2 Geometry optimization Previous: 34.2.7 Restarting a geometry

P.J. Knowles and H.-J. Werner
molpro@tc.bham.ac.uk
Jan 15, 2002