Strengths and Weaknesses of Semiempirical Methods

These remarks refer to NDDO methods like AMI and PM3. 

A major weakness of semiempirical methods is that they must be assumed to be unreliable outside molecules of the kind used for their training set (the set of molecules used to parameterize them), until shown otherwise by comparison of their predictions with experiment or with high-level ab initio (or probably DFT) calculations. Although, as Dewar and Storch pointed out [125], the reliability of ab initio calculations, too, should be checked against experiment, the situation is somewhat different for these latter, at least at the higher levels studies of exotic species, in particular, are certainly more trustworthy when done ab initio than semiempirically. Semiempirical heats of formation are subject to errors of tens of kJ mol-1, and thus heats (enthalpies) of reaction and activation could be in error by scores of kJ mol-1. AMI and PM3 underestimate steric repulsions, overestimate basicity and underestimate nucleophilicity, and can give unreasonable charges and 

In general, the accuracy of semiempirical methods, particularly in energetics, falls short of that of current routine ab initio methods (this may not have been the case when AMI was developed, in 1985 [125]). Parameters may not be available for the elements in the molecules one is interested in, and obtaining new parameters is something rarely done by people not actively engaged in developing new methods. Semiempirical errors are less systematic than ab initio, and thus harder to correct for. Clark has soberly warned that All parameterized techniques can interpolate and none can extrapolate consistently and well , thus we can expect on occasion a catastrophic failure but semiempirical methods will do what they are designed to do [11]. 

AMI and PM3 perform similarly and usually give quite good geometries, but less satisfactory heats of formation and relative energies. A modification of AMI called SAMI (semi-ab initio method 1), relatively little-used, is said to be an improvement over AMI. AMI and SAMI represent work by the group of M. J. S. Dewar. PM3 is a version of AMI, by J. J. P. Stewart, differing mainly in a more automatic approach to parameterization. Recent extensions of AMI (RM1) and PM3 (PM6) seem to represent substantial improvements and are likely to be the standard general-purpose semiempirical methods in the near future. 

(a) Weinberg S (1992) Dreams of a final theory the search for the fundamental laws of nature. Pantheon Books, New York (h) Watson A (2000) Measuring the physical constants. Science 287 1391 

---