At the 2011 AIChE Annual Meeting, Prof. West presented work on Automatic Reaction Mechanism Generation with Group Additive Kinetics.
The key challenge in making chemical mechanism development predictive is being able to accurately estimate any possible rate coefficient k(T) even if there are no experimental data. Reaction Mechanism Generator (RMG) is an open-source software project that can build detailed kinetic models for chemical reacting systems. It uses a database of rules to propose elementary chemical reactions and to estimate the necessary thermochemical and kinetic parameters. We are modifying the algorithm used to estimate kinetic data to make the estimated reaction rates more reliable and easier to document in cases where they are estimated from sparse data.
The reaction mechanism generation software RMG estimates reaction rate expressions using rules based on the functional groups surrounding the reacting center. A reaction typically involves more than one functional group (e.g. an atom with a hydrogen ligand XH and a radical Y·), which combine to form a “supergoup” XY. When a rule for the supergroup XY is known, it can be used to predict the reaction kinetics with reasonable accuracy. However, when data are sparse and a rule for XY is not known, RMG currently averages “similar” XY supergroups. For these scenarios we are have tested a group additive method, adding separate contributions from X and Y which are derived from known XY supergroups.
The group values can be trained using existing supergroup rules or explicit reactions. The group values can be re-trained when new kinetic data are available or the definitions and hierarchy of the groups are updated. By recording the goodness of fit when the group values are trained, confidence intervals can be calculated on each reaction rate estimated using this method.
For the hydrogen abstraction family of reactions, estimates calculated in this manner are better than those estimated using the averaging scheme previously used in RMG software, and their origin is simpler to trace. This approach is now being extended to families of reactions other than hydrogen abstraction.
Links:
Read AIChE Conference Abstract
Read Extended Abstract (conference proceedings paper)
View on SlideShare.net.