Predicting observables on different potential energy surfaces using feature sensitivity analysis: Application to the collinear H+H2 exchange reaction

Johnny Chang, Nancy J. Brown, Michael D'Mello, Robert E. Wyatt, Herschel Albert Rabitz

Research output: Contribution to journalArticlepeer-review

8 Scopus citations

Abstract

Two sensitivity-analysis techniques are used to show how one can predict observables on new or perturbed potential energy surfaces (PES) without doing any additional dynamics calculations on the new PESs. Both techniques require the computation of the observable (O) and its sensitivity to variations in the potential (δO/δV) on just one surface. The first approach uses a simple first order expansion of the observable with respect to δV. The second approach uses a nonlinear least-squares fit of particular features in the observable, and then uses the same functional form to predict the observable on a different PES but with a modified set of fitting parameters. The new fitting parameters are related to the old via a functional Taylor expansion. Both approaches work well when variations in the potential are small. When the potential difference is large, the second approach gives reasonable results even in cases where the first approach is giving spurious predictions. These ideas are tested for the collinear H+H2 reaction within the framework of quantum reactive scattering. The log-derivative version of the Kohn variational principle is used for the scattering calculations.

Original languageEnglish (US)
Pages (from-to)6240-6248
Number of pages9
JournalThe Journal of chemical physics
Volume97
Issue number9
DOIs
StatePublished - Jan 1 1992

All Science Journal Classification (ASJC) codes

  • Physics and Astronomy(all)
  • Physical and Theoretical Chemistry

Fingerprint Dive into the research topics of 'Predicting observables on different potential energy surfaces using feature sensitivity analysis: Application to the collinear H+H<sub>2</sub> exchange reaction'. Together they form a unique fingerprint.

Cite this