Right now if you want to have the results publishable you should probably use PM6, which is freely available in MOPAC and superior to both PM3 and AM1.

There might be, however, instance when AM1 or PM3 would work better. If you're unsure, so some benchmarks of the systems you're interested in and see which one reproduces the desired experimental features best.

Hi, I hope you find it somewhat useful

http://www.ccl.net/chemistry/resources/messages/1994/04/12.012-dir/

Right now if you want to have the results publishable you should probably use PM6, which is freely available in MOPAC and superior to both PM3 and AM1.

There might be, however, instance when AM1 or PM3 would work better. If you're unsure, so some benchmarks of the systems you're interested in and see which one reproduces the desired experimental features best.

Agree with above answers. PM6 is more general (has been parameterized against many atoms, and many kind of molecule) yet more accurate too. And as Bartosz said, sometimes there will be an exception case where the result is no better than AM1 or PM3, but this is very less likely. I've read somewhere about PM6 where it can generate the right connectivity when AM1 or PM3 couldn't. I also read a table that shows the average deviation of energy compared to experimental one, and PM6 (generally) could give much less error than AM1 or PM3.

Hope that help,

Radif

Thank you very much for your answers...

The thing is I have a bunch of AM1 and PM3 calculated energy values for the same series of molecules. And in some cases the energy obtained from the AM1 calculation varies quite a bit from those obtained from PM3.. 

Why could that be?

As we know, semiempiric method is derived from the ab initio and simplified (reducing the costly equation part) to make it faster, but the result will deviate and need to be parameterized (A.K.A. adding some constant number to the formula) with experimental or highly accurate ab initio calculation, a process called curve fitting. So there are two major reason why there are differences in AM1 versus PM3 result, that is the equation used, and the atom/molecule used in parameterization process. AM1 & PM3 uses different molecule subset in the parameterization (eg. one of them using more aromatic molecule, or one of them using more anorganic molecule).

In a nutshell, this is simply the matter of transferability, it is the matter of how similar the molecule you're trying to calculate to the subset of molecule used to parameterize the semiempiric method.

Thank you very much...

Your answers were very useful, and helped me alot...

Mr./Miss/Mrs. Albar,

The formalism is the same, but the parameters are different.

Please find in this topic the following refs:

[1] M. Dewar, W. Thiel, Ground States of Molecules. 38. The MNDO Method.

Approximations and Parameters, J. Am. Chem. Soc. 99 (1977) 4899 - 4907

[2]M. Dewar, E. Zoebisch, E. Healy, J. Stewart, AM1: A New General Purpose Quantum Mechanical

Molecular Model, J. Am. Chem. Soc. 107 (1985) 3902 - 3909

[3] J. Stewart, Optimization of Parameters for Semiempirical Methods

I. Method, J. Comput. Chem. 10 (1989) 209-220

In addition you can find their application in vibrational spectroscopy and the correlations obtained for few examples (attachment).