36.2 Automatic geometry optimization (OPTG)

The OPTG command is used to perform automatic geometry optimizations for all kinds of wavefunctions. The coordinates to be optimized can be chosen using the COORD directive (see section optgeo:coord). Various optimization methods can be selected as described in section 32.2.4. MOLPRO allows minimization (i.e. search for equilibrium geometries), transition state optimization (i.e. search for saddle points on energy surfaces), and reaction path following. The standard algorithms are based on the rational function approach and the geometry DIIS approach. Also available is the quadratic steepest descent following method of Sun and Ruedenberg (see J. Sun and K. Ruedenberg, J. Chem. Phys. 99, 5257 (1993)). This method is often advantageous in Transition State searches. For a detailed discussion of the various minimization algorithms see (see F. Eckert, P. Pulay and H.-J. Werner, J. Comp. Chem 18, 1473 (1997)).

The OPTG must directly follow the input for the wavefunction used in the geometry optimization. It will call FORCE, OPT, INT, and, as needed, HF, RHF, MCSCF, CI, CCSD etc. For each of these programs, the input file is automatically repositioned to the last corresponding input before the OPTG card; so any input for RHF, MCSCF, CI, CCSD etc. can be used and will be correctly processed. It is essential, however, that the most recently optimized orbitals are used in the wavefunction for which the geometry is optimized. Any input needed for OPTG must directly follow the OPTG card. The gradients are computed analytically for HF, DFT, MP2, QCISD, or MCSCF wavefunctions; otherwise the gradients are computed by finite differences (see OPTG, NUMERICAL). Davidson corrected energies or excited state energies can be optimized using the VARIABLE and STATE subdirective.

Various options, in particular convergence criteria, can be specified on the OPTG command:

OPTG,key1=value, key2=value,......

where key can be

MAXIT

to set the maximum number of optimization cycles. The default is 50.

GRAD

sets the required accuracy of the optimized gradient. The default is $3 \cdot 10^{-4}$.

STEP

to set the convergence threshold for the geometry optimization step; if $value \ge 1$, the threshold is set to $10^{-value}$. The default is $3 \cdot 10^{-4}$.

ENERGY

sets the required accuracy of the optimized energy. The default is $1 \cdot 10^{-6}$.

GAUSSIAN

Use Gaussian convergency criteria.

SRMS

sets (for Gaussian convergency criterion) the required accuracy of the RMS of the optimization step. The default is $0.0012$.

GRMS

sets (for Gaussian convergency criterion) the required accuracy of the RMS of the gradient. The default is $3 \cdot 10^{-4}$.

BAKER

Use Baker's convergency criteria (see J. Baker, J. Comp. Chem. 14,1085 (1993)).

NUMERICAL

Force the use of numerical gradients.

The standard MOLPRO convergency criterion requires the maximum component of the gradient to be less then $3 \cdot 10^{-4}$ [a.u.] and the maximum energy change to be less than $1 \cdot 10^{-6}$ [H] or the maximum component of the gradient to be less then $3 \cdot 10^{-4}$ [a.u.] and the maximum component of the step to be less then $3 \cdot 10^{-4}$ [a.u.].

It is also possible to use the convergency criterion of the Gaussian program package. It is somewhat weaker than the MOLPRO criterion and requires the maximum component of the gradient to be less then $4.5 \cdot 10^{-4}$ [a.u.] and the root mean square (RMS) of the gradient to be less then $3 \cdot 10^{-4}$ [a.u.] as well as the maximum component of the optimization step to be less then $0.0018$ [a.u.] and the RMS of the optimization step to be less then $0.0012$ [a.u.].

The defaults for the convergence parameters can also be changed by using a global GTHRESH directive, i.e.

GTHRESH, OPTSTEP=step, OPTGRAD=grad, ENERGY=energy;