Modelling methods

The method is the way in which an energy is obtained to correspond
to a particular geometry

Older modelling programs may support only one method, but modern ones usually
give the user a choice

The other parts of the program, e.g. setting up molecules, the energy minimisation
routine for geometry optimisation, and display facilities and geometry
measurement, can be the same irrespectively of the method used

There are two broad divisions of methods:

molecular mechanics force field methods

treat the molecule as a system of mechanical springs

completely ignore the electrons

electronic methods

calculate the distribution and energy of the electrons using MO theory

combine this with nucleusnucleus repulsion energy

The electronic methods subdivide into:

semiempirical methods

consider only the valence shell electrons, with some big simplifications

use empirically preset parameters for each element to produce results
comparable with experiment

ab initio methods

consider all the electrons without using parameters from experimental data

Computing time required by program:
Force fields < semiempirical << ab initio

Usually, quality of results:
Ab initio > semiempirical > force fields

Often, people do geometry optimisation (which requires repeated calculations
of energy) using a faster method, then do a 'single point calculation'
using a slower, better method on the geometry they have found

Single point calculations can be used to model properties such as:

total energy

electron distribution

orbital energies

To get electronic properties, e.g. uv/vis spectrum or NMR shieldings, you
must at least finish up with one of the electronic methods

For mediumsized organic molecules, people normally use forcefield methods
to get a starting geometry for each conformer of interest, and do conformational
searching

It is then possible to do singlepoint semiempirical calculations on each
conformer

It may be reasonable to reoptimise the geometry of each conformer at a
semiempirical level

On the UCS's most powerful general service unix computer aidan, one conformer
of methylcyclohexane took 14 seconds to reoptimise at the semiempirical
pm3 level

The same molecule on the same computer took 5.5 minutes to reoptimise at
a modest but useful ab initio level

The amount of effort by the user is the same for each method: you
just leave the computer to calculate for longer, for the slower methods.
However, there are limits to the length of job you are allowed to run on
shared service computers, as well as the practicality of getting your research
done in time

If you are going to do any of this practically in real research, you should
read Hehre's 'Practical Strategies for Electronic Structure Calculations',
which tells you how far you need to go to get models which are good enough
for particular purposes

The three main classes of methods are dealt with in separate notes: