The utility of higher order derivatives in constructing molecular potential energy surfaces by interpolation

In this paper we evaluate the use of higher order derivatives in the construction of an interpolated potential energy surface for the OH+H2→H2O+H reaction. The surface construction involves interpolating between local Taylor expansions about a set of known data points. We examine the use of first, second, third, and fourth order Taylor expansions in the interpolation scheme. The convergence of the various interpolated surfaces is evaluated in terms of the probability of reaction. We conclude that first order Taylor expansions (and by implication zeroth order expansions) are not suitable for constructing potential energy surfaces for reactive systems. We also conclude that it is inefficient to use fourth order derivatives. The factors differentiating between second and third order Taylor expansions are less clear. Although third order surfaces require substantially fewer data points to converge than second order surfaces, this faster convergence does not offset the large cost incurred in calculating numerical third derivatives. We therefore conclude that, without an efficient means for calculating analytic third derivatives, second order derivatives provide the most cost-effective means of constructing a global potential energy surface by interpolation.