Contents: Table of Contents 

Detailed Control of Optimisations

Convergence

The most important parameter that one may wish to alter controls the degree of convergence of an optimisation. The optimisation stops when the forces (elements of the gradient) are all less than a given threshold. The default theshold in 10^-4 a.u. This can be changed using the keyword GRADTOL , with format,

GRADTOL n

where n is an integer which results in the convergence threshold being set to 10^-n. The default is adequate for normal purposes - you may wish to change it if you are doing a calculation on a system with a very flat region in the potential energy surface, or if you wish to determine the position of a stationary point to very high precision.

Number of Steps

Sometimes it is useful to restrict a particular job to calculate only a small number of gradients (this is to avoid geometry optimisations which run for a very long time). This is achieved through the NUMGRAD keyword,

NUMGRAD n

This causes the geometry optimisation to procede for n steps only. There is no default value of n , i.e. by default there is no restriction.

Maximum energy and parameter changes

Two other keywords which may occasionally be of use are connected to the maximum energy change and the maximum geometry change per iteration.

The maximum energy change can be set by the keyword ,

EMAX r

where r is a real number. The default value is 0.05.

The maximum geometry change is controlled by a parameter which can be set by the keyword,

MAXSTEP r

where r is a real number. The default value is 0.2.

It is unlikely that either of these last two parameters will need altering. Generally one would only alter the maximum step size if you found the optimiser was presistantly overshooting the stationary point - a sign of this would be the energy rising when searching for a minimum. Usually the optimiser would correct itself anyway if the initial step sizes were too large as EMAX and MAXSTEP are adjusted dynamically by the program according to a 'trust radius' scheme.

Forcing use of cartesian coordinates

Another keyword which may be of occasional use is,

CARTESIAN

Normally, if one specifies the geometry in internal coordinates, then the optimisation is also carried out in internal coordinates. This keyword causes the internal coordinate definitions to be ignored, and cartesian coordinates to be used for the optimisation. This would primarily be of use if the molecular geometry were easy to set up in internal coordinates, but you suspected that the choice of internal coordinates would not be the most efficient for the optimisation. This is not a common circumstance, but some for some ring systems the geometry optimisation can be more efficient in cartesians.

    
Contents: Table of Contents