QUADRATIC STEEPEST DESCENT ON POTENTIAL-ENERGY SURFACES .1. BASIC FORMALISM AND QUANTITATIVE ASSESSMENT

A novel second-order algorithm is formulated for determining steepest-descent lines on potential energy surfaces. The reaction path is deduced from successive exact steepest-descent lines of local quadratic approximations to the surface. At each step, a distinction is made between three points: the center for the local quadratic Taylor expansion of the surface, the junction of the two adjacent local steepest-descent line approximations, and the predicted approximation to the true steepest-descent line. This flexibility returns a more efficient yield from the calculated information and increases the accuracy of the local quadratic approximations by almost an order of magnitude. In addition, the step size is varied with the curvature and, if desired, can be readjusted by a trust region assessment. Applications to the Gonzalez-Schlegel and the Muller-Brown surfaces show the method to compare favorably with existing methods. Several measures are given for assessing the accuracy achieved without knowledge of the exact steepest-descent line. The optimal evaluation of the predicted gradient and curvature for dynamical applications is discussed.