Semi-empirical parameterization...

 Something for folks to think about...
 The various SE parameter sets were all (to the best of my knowledge)
 developed on VAXen about 10-ish years ago (6-ish for PM3?).  Since
 that time, there's been a 2-4 orders-of-magnitude increase in
 CPU horsepower, and there has to be an improvement in the quality and/or
 amount of experimental data to parameterize against.
 Doesn't this suggest that it's a worthwhile idea to develop an AM2/PM4
 set (with or without d-orbitals) that could (a) use the extra
 stuff mentioned above and (b) specifically examine the known problem
 areas of the various methods?  Judging from solvation work such
 as AMSOL, parameter sets can be generated in a CPU-day or a CPU-week
 on the high-end stuff available now.  Several/Many sets of params can
 be generated very quickly, using whatever scientific or
 empirical constraints one wishes to impose...
 This doesn't even take "novel" machines like SIMD/MIMD/MPP boxes into
 account - these might make the process even faster.
 Folks will probably say that "we should not invalidate past results".
 True, but if Allinger's allowed three different mechanics FF's (not
 counting the modifications to each that have appeared over time), I think
 folks will readily accept a new SE parameter set if it's properly done and
 disseminated.
 This way, we'll be able to get rid of any problem areas that are artifacts
 of the limited capabilities and/or datasets used in the earlier
 work.  It will also let folks include LARGE small molecules as opposed to
 the SMALL small molecules used in the past.
 Whatcha think?
 Joe Leonard
 jle -8 at 8- world.std.com