PARAM is a program for optimizing parameters for semiempirical methods. It is controlled by a single data set that specifies the type of operation (usually parameter optimization or a survey to determine errors), the conditions for the operation (AM1, PM6, etc., or a new method), and the set of data-files to be used.
Given a set of reference data, the program PARAM can optimize parameters for elements for use in semiempirical methods. Historically, this operation has been extremely difficult. As Dewar reported, the effort involved in developing the early methods such as MNDO and AM1 was enormous. Most of the time was spent optimizing parameters and eliminating artifacts. Problems of the type reported by Dewar have now been solved, and with the new PARAM, methods such as MNDO and AM1 can be developed in a matter of minutes or, at most, hours.
The program PARAM is the central program used for method development. There are several utility programs (see PROGRAMS) that are useful in manipulating the large files and large number of files. Before using the program PARAM, first become familiar with these utilities.
PARAM uses several types of files. For convenience, these should be stored in different folders, so each folder contains only one type of file. The main types are normal data sets, data sets representing atomic states, data sets for exotic systems, data sets for PARAM, and data sets for parameters. This last set does not need to exist - if it does not exist, PARAM will use parameters from the default method. PARAM can handle 15,000 reference data in 10,000 reference data files and 4,000 parameters.
If, as is inevitable, things go wrong, there are extensive diagnostics available for finding out what happened. These diagnostics can also indicate what to do, but of course they cannot take any action on their own. Options also exist to identify faulty data, and, depending on the keywords used, either ignore, i.e., exclude, the faulty data or stop the calculation, so the user can take appropriate action.
Although parameter optimization is now fully automatic, a whole new set of problems must be addressed. Among these are:
(A) Large data bases are needed in order to adequately represent
all important chemical phenomena.
(B) Efficient methods to manipulate these data sets are essential.
(C) Techniques to determine absolute and relative accuracy of various methods are needed.
(D) Migration from method development to production quality programs such as MOPAC must be facile.
PARAM has only been used a few times thus far, first, for the development of PM3, to "fill in the blanks" in MNDO, AM1, and PM3, and for the development of PM6. Of these, the development of PM6 could be considered as the most complete.
Causes of error in Semiempirical methods