I am trying to calculate the lift coefficient for a NACA 0012 wing section. I am experimenting with different meshes and am strangely finding better results for my coarser meshes. I have am setting up a mesh with zero gradient and setting a number of cells per cord. I first ran only 10 cells per cord with no refinement for partial cells. I am running for a level 2 and 4 refinement for partial cells, and also doubling the mesh precision to 20 cells per cord, and when those are done, I will keep going to 40 and so on. Computation domain is 10c behind, and 5c behind, top and bottom.

I was expecting the 10 cells per cord mesh to be far too coarse to give good results, so I was very surprised to find that I was getting good correlation. Can anybody explain how a mesh this course is giving me good results when making my mesh more fine is making things worse. Also, I recently learned about Courant-Friedrichs-Lewy condition. Is this something that need not be satisfied within the solution that solidworks uses?

The plot below shows the results of the parametric study I ran varying angle of attack for the two shown cases. The blue CL curve was created using XFLR5 for the same foil, Re, Mach, and fluid properties.

Mesh from computational domain level

Airfoil detail

Solution at 15 deg AoA with no partial cell refinement

solution at 15 deg AoA with level 4 partial cell refinement and 1 global refinement while the solver was running