Identifying collective dynamical observables bearing on local features of potential surfaces

A singular value decomposition of dynamical sensitivities provides insight into the relationship between a data set and the potential which is often not evident from the sensitivities of individual observables. An illustration is treated consisting of data sets drawn from reactive transition probabilities as a function of energy for the collinear H+H-2 system. While the sensitivities of individual reactive transition probabilities to the two-dimensional potential are highly structured functions of the potential coordinates, a set of reactive transition probabilities is identified which collectively has localized sensitivity primarily to the saddle point region and secondarily to the slope along the H-3 symmetric stretch line in the outer corner tunneling region and to the width of the barrier. Information of this type garnered from a principal component sensitivity analysis can be especially valuable when attempting to use dynamics data to refine potential surfaces. (C) 1998 American Institute of Physics.