The detailed chemical analysis performed for these experiments allows for a further comprehensive test of the published reaction scheme by the comparison of experimental results and computer simulations. The general agreement between the simulated and experimental selectivities and behaviour of the main products, propene oxide, acetaldehyde, formaldehyde carbon monoxide and carbon dioxide gives some confidence in the bulk of the model as well as highlighting topics that require further study. The epoxide selectivities and their increase on adding acetaldehyde are both well predicted as are the selectivities for many of the other products. The further oxidation of acetaldehyde is overpredicted in the simulations, suggesting that peroxy radical attack is less significant than considered in the model.

In metal systems, there is evidence that carbon oxides are also formed by heterogeneous reactions (Evzerikhin and Artsis, 1968), increasing in importance with increasing temperature. This may explain why on adding acetaldehyde in the flow system with the corresponding reduction in operating temperature, there is a slight decrease in the yields of carbon oxides. This behaviour is not predicted by the model, which only includes homogeneous gas phase reactions.

The predicted rate of reaction for the CSTR experiments was too low, so to obtain the experimental oxygen conversion the gas temperature in the calculations was increased by 10-37K. The low predicted overall rate of reaction for the CSTR experiments is not understood and could either be associated with key reactions (eg. peroxide decomposition) being highly pressure-dependent between 0.9 and 55 bar, or due to local heating of parcels of gas in the vicinity of the walls, which had temperatures up to 40K higher than the bulk of the gas.

A noticeable problem with the model is that the yields of total peroxides predicted by the model (ca. 20% of converted propene) are much greater than found experimentally (0.3-3%). The reason for this is currently under investigation, but two possibilities are either that there is a problem with the technique for determining the peroxides, or that there are inappropriate rates or reactions involving peroxides in the model. It is unlikely that the peroxides in the reaction vessels are either decomposing or isomerising heterogeneously on the walls of the reactor, as inclusion of this type of reaction in the model leads to a negligibly slow overall rate of reaction. It is most likely that peroxides are over-predicted because rates of hydrogen atom abstraction by peroxy radicals (which have generally not been measured) are too fast. However, at this stage, no attempt has been made to adjust or "force" these rate constants to improve the comparison between experiment and theory, because other reactions (eg. peroxy radical epoxidation, homogeneous peroxide decomposition etc.) also effect the predicted peroxide concentration and have rate constants that are uncertain.