Identification of Missing Reactions in Detailed Reaction Mechanisms

Anders Broe Bendtsen, Peter Glarborg, Kim Dam-Johansen

CHEC Research Group, Dept. of Chemical Engineering, Technical University of Denmark,

Poster presented at 27th International Symposium on Combustion, Boulder, CO, USA, Aug. 3-7th 1998

The information structure of a reaction mechanism consist of three parts:

The species participating in reactions
The reactions occurring between species
The rates of reactions

Most of the focus in the construction of detailed chemical reaction mechanisms is put on the accurate measurement or estimation of reaction rate constants. However, one important fact which is often overlooked is that the absence of a reaction is equal to estimating the rate to be zero!

A reaction grid may be designed to ensure exploration of all combinations of reactants and products from a predefined set of species. A reaction grid based on the reaction rates used in GRI Mech. 2.11 is presented graphically. In this figure the color intensity symbolizes the reaction rate of reactions between two species as 1000K, e.g. white correspond to a reaction rate of 1015, while black correspond to a reaction rate of zero. The objective of this work was to investigate whether any of the reactions left out should have reaction rates different from zero. This is done by constructing an empirical QSAR model for estimation of reaction rates from their gross structure.

The concept for estimation of reaction rates²
The objective of our method for estimation of reactions was to

estimate reaction rates for reactions between as wide a range of species as possible
with a minimal computational effort (as many estimates are to be calculated)
with an accuracy sufficient for determining whether the reaction should be investigated in detail or not.

The reaction rates are related to the electronic structure of reactants and products. This structure was described as the number of bond types, the number of unpaired electrons and the number of lone-pairs, as it is exemplified in the table for O/H species. For O/H/C/N species the total number of characteristics was 27.
A reaction was described by a vector consisting of the characteristics of reactants and products, and a vector b of coefficients in a linear regression model may be calculated from the reactions in an existing mechanism using Partial Least Squares (PLS). This model can be used for prediction of reaction rates not included in the reaction mechanism.

Table showing the prediciton concept

Case Study Results³ A case study of this method was made by studying GRI Mech. 2.1's ability to predict NO sensitized oxidation of methane. The original GRI Mech. 2.1 did not predict this phenomenon. For the 279 reversible (558 irreversible) reactions in GRI-Mech 2.1 a PLS model with a passable prediction error was found. Estimates of 2138 potential reactions were included and through a sensitivity analysis 9 reactions were found to be important. Literature rates for these reactions were added where available. One added reaction was CH3+NO2=CH3O+NO as suggested by Bromly et al.⁴ This expanded mechanism predicted NO sensitized oxidation of methane well.

Conclusion
A method for expansion of reaction mechanisms was suggested, and demonstrated in a case study. An improved estimation quality of the PLS model is desired, and the some suggestions for obtaining this may be an estimation of reaction rates in the exothermal direction as well as a better characterization of the reacting species, e.g. through graph theoretical descriptors. The estimation task may also be simplified by removing reactions with radicals, since other methods exist for such estimations.

References
1) Frenklach M. et al (1996) http://www.me.berkeley.edu/gri_mech/
2) Bendtsen A.B., & P. Glarborg. (1998) Fashioning a Model: Optimization Methods in Chemical Physics pp. 55-60
3) Bendtsen A.B., P. Glarborg, & K. Dam-Johansen (1998). Chemom.Intell.Lab.Syst. 44, 357‑365
4) Bromly J.H., F.J. Barnes, S. Muris, X. You, & B.S. Haynes. Comb.Sci.Tech. 115, (1996), 259-296

Acknowledgments
This work was supported by the Danish Ministry of Energy, the Danish Gas Technology Centre and the CHEC Research Programme. The Danish Gas Technology Centre is sponsored by the national and regional Danish gas companies. The CHEC (Combustion and Harmful Emission Control) Research Programme is cofunded by a.o. the Danish Technical Research Council and the Danish energy consortia Elsam and Elkraft.

to Anders Broe Bendtsen's home page

Author Anders Broe Bendtsen, andbbdk@gmail.com

Last updated: February18th 2001