Grid refining is standard procedure in any numerical study because you should investigate the effect of discretisation on your results. Results are only interesting when they do converge upon grid refinement toward a limit. 

Of course the grid refinement depends also on the scales you are trying to resolve. When you carry out a very fine grid simulation you could be able to describe effects such as viscous fingering (instability of the compositional transition between CO2 and crude). If you are using a coarser grid in order to obtain a model for an entire reservoir, you should somehow include a model that takes the macroscopic effects of such instabilities into account. Hence grid refinement depends on the scales you should resolve. This depends on the model you are using.

Regards,

Mico Hirschberg

Selecting the coarser grid blocks (lower number of cells) in the compositional modeling through CO2 EOR flooding cause the larger effect of numerical dispersion. The presence of numerical dispersion has the effect of smearing out the recovery curves and make the determination of a break point more difficult / inaccurate. Hence, the task of determining the MMP becomes a balance between acceptable accuracy and acceptable CPU time consumption.  Further, as explained by Zick and Stalkup, the MMP can always be found by compositional simulation, though fine grids and repeated simulations at multiple pressures and certainly a significant computation times will be required.

Numerical dispersion and simulating accurate flood frontal movement. It is used not only in CO2 injection, but also in water injection and gas injection (flood frontal movement). 

Good luck.