Authors: Ray Hefferlin
Comments: 9 Pages. May appear in Int. J. Mol. Model, Vol 4, #1
Combined least-squares and neural-network forecasts for internuclear separations of main-group diatomic molecules, most with from 9 to 12 atomic valence electrons, are presented. We require that the standard-deviation bounds of the forecasts overlap each other; this requirement is met by 65 molecules, of which 46 seem not to have been studied previously. The composite errors average 0.1036Å on either side of the composite predictions. There is agreement with 33 of 41 independent test data (80.5%), and those not in agreement fall outside the composite error limits by an average of 1.83%.