Acknowledgements

which requires about 7–10 eV per molecule. This "waste of energy" explains the lower energy efficiency of CO2 splitting in a DBD. On the other hand, in a MW and GA plasma, vibrational excitation of CO2 is dominant, and VV relaxation gradually populates the higher vibrational levels (so-called ladder climbing). This is the most energy-efficient way of CO2 dissociation, as it requires only 5.5 eV per molecule,

We also presented the important reaction pathways in CO2/CH4, CH4/O2, CO2/ H2 and CO2/H2O mixtures, as well as for the effect of N2 addition to a CO2 plasma. In a DBD plasma, the conversion is always initiated by electron impact dissociation, creating radicals that react further into value-added compounds. The main products formed are syngas (CO/H2), but higher hydrocarbons and oxygenates are also formed in limited amounts. However, the selective production of these targeted compounds is not yet possible, due to the high reactivity of the plasma. Therefore, a catalyst must be inserted in the plasma. Our models reveal that CO2/CH4 and CH4/ O2 mixtures exhibit totally different chemical reactions, resulting in different products. A CO2/H2 mixture does not produce many higher hydrocarbons and oxygenates, and the CO2 conversion is very limited, due to the lack of CH2 (and CH3) radical formation. Indeed, the CH2 radicals are the main collision partners of CO2 in the CO2/CH4 mixture. Furthermore, adding H2O to a CO2 DBD plasma yields a drop in CO2 conversion, and also the H2O conversion is limited, and virtually no oxygenated hydrocarbons are formed, which could also be explained from the chemical reaction paths. The insights obtained by the model might be useful to provide possible solutions. The last example of 0D chemical kinetic modeling was given for a CO2/N2 plasma, where it was shown that also NOx compounds are produced, which might give several environmental problems. Again, the model can explain their formation, which is useful to provide possible solutions on how to avoid this

Although 0D models can give useful information on the plasma chemistry, they

In the future work, we intend to implement the more complex CO2 chemistry (either pure or mixed with other gases) in such fluid models, to obtain a more comprehensive picture of CO2 conversion in a real plasma reactor geometry. As this is quite challenging in terms of computation time, reduced chemistry sets must be developed for CO2 and its gas mixtures. When modeling CO2 conversion in a MW or GA plasma, the vibrational kinetics must be accounted for. To avoid the need of describing all individual levels, we have developed a level-lumping strategy [39], which enables to group the vibrational levels of the asymmetric stretch mode of CO2 into a number of groups. This reduces the calculation time, so that it can be implemented in 2D models [86]. We believe that a combination of 0D chemical kinetic models (to obtain detailed insight in the entire plasma chemistry and to develop reduced chemistry sets, identifying the main species and chemical reactions) and 2D/3D fluid models (for a detailed understanding of the reactor design)

is the most promising approach to make further progress in this field.

cannot really account for details in the plasma reactor configuration and thus predict how modifications to the reactor design might lead to improved CO2 conversion. For this purpose, 2D or 3D fluid models of specific reactor designs are needed. Developing such fluid models for a detailed plasma chemistry, however, leads to excessive calculation times. Therefore, these models are up to now mainly developed for simpler chemistry, in argon or helium. We have shown here examples for a packed bed DBD reactor and a GAP. These models allow to elucidate why certain reactor designs give beneficial results and to pinpoint the limitations and finally how improvements in the reactor designs might yield a better CO2 conver-

that is, exactly the C=O bond energy.

Plasma Chemistry and Gas Conversion

NOx formation.

sion and energy efficiency.

26

We would like to thank R. Aerts, A. Berthelot, C. De Bie, T. Kozák, and K. Van Laer for sharing their simulation results.
