Statistical Analysis of Accuracy of Methods in MOPAC

This statistical analysis allows the accuracy of the various methods in MOPAC to be compared.  It is impossible to do an exact comparison of the various methods, because the sets of elements in each method are different, but the results presented here are a good-faith attempt to provide unbiased information. 

The color code for errors compares all methods to PM7.  If the AUE for a given property for a given method is less than that for PM7, the number is in blue; it it is greater, the number is in red.

Where the data-sets for a given property for a given method are identical to that for PM7, the number of data in the data-set is in black, otherwise it is in red.

The early methods (MNDO and AM1) did not use d-orbitals, so results for hypervalent phosphorus and sulfur are very bad.  This is reflected in the sets "H C O S" and "H C O P"

In going from PM6 to PM7 there is a large increase in error for heats of formation, geometries, etc. for molecules.  This is because the molecule set (4419 species) does not include a large number of diatomic interactions that occur in chemistry.  These interactions are present in solids.  For solids, PM6 does poorly, but PM7 gives good heats of formation and geometries. 

Relative to PM7, PM6 also does poorly with intermolecular interactions, but variants of PM6 (PM6-DH+ and PM6-DH2) do much better.

Set	Property
All normal molecules	ΔHf -Per Element	Bond lengths	Bond angles	Dipole	Ionization Potential
Sets of normal molecules	ΔHf -Per Set	Bond lengths	Bond angles	Dipole	Ionization Potential
All solids	ΔHf and Geometries per Element			Graph of Densities
Solids by element	Periodic Table			Solid Elements
Weak interactions S22	Intermolecular Energy
Weak interactions S66	Intermolecular Energy
Weak interactions L7	Intermolecular Energy