These tools will no longer be maintained as of December 31, 2024. Archived website can be found here. PubMed4Hh GitHub repository can be found here. Contact NLM Customer Service if you have questions.


Pubmed for Handhelds

PUBMED FOR HANDHELDS

Search MEDLINE/PubMed

  • Title: Molecular dynamics investigations of ozone on an ab initio potential energy surface with the utilization of pattern-recognition neural network for accurate determination of product formation.
    Author: Le HM, Dinh TS, Le HV.
    Journal: J Phys Chem A; 2011 Oct 13; 115(40):10862-70. PubMed ID: 21888438.
    Abstract:
    The singlet-triplet transformation and molecular dissociation of ozone (O(3)) gas is investigated by performing quasi-classical molecular dynamics (MD) simulations on an ab initio potential energy surface (PES) with visible and near-infrared excitations. MP4(SDQ) level of theory with the 6-311g(2d,2p) basis set is executed for three different electronic spin states (singlet, triplet, and quintet). In order to simplify the potential energy function, an approximation is adopted by ignoring the spin-orbit coupling and allowing the molecule to switch favorably and instantaneously to the spin state that is more energetically stable (lowest in energy among the three spin states). This assumption has previously been utilized to study the SiO(2) system as reported by Agrawal et al. (J. Chem. Phys. 2006, 124 (13), 134306). The use of such assumption in this study probably makes the upper limits of computed rate coefficients the true rate coefficients. The global PES for ozone is constructed by fitting 5906 ab initio data points using a 60-neuron two-layer feed-forward neural network. The mean-absolute error and root-mean-squared error of this fit are 0.0446 eV (1.03 kcal/mol) and 0.0756 eV (1.74 kcal/mol), respectively, which reveal very good fitting accuracy. The parameter coefficients of the global PES are reported in this paper. In order to identify the spin state with high confidence, we propose the use of a pattern-recognition neural network, which is trained to predict the spin state of a given configuration (with a prediction accuracy being 95.6% on a set of testing data points). To enhance the prediction effectiveness, a buffer series of five points are validated to confirm the spin state during the MD process to gain better confidence. Quasi-classical MD simulations from 1.2 to 2.4 eV of total internal energy (including zero-point energy) result in rate coefficients of singlet-triplet transformation in the range of 0.027 ps(-1) to 1.21 ps(-1). Also, we find very low dissociation probability up to 2.4 eV of internal energy during the investigating period (5 ps), which suggests that dissociation does not occur directly from the singlet ground-state, but it involves the excited triplet-state as an intermediate step and requires more reaction time to occur.
    [Abstract] [Full Text] [Related] [New Search]