I have tried everything : 1) Non-conformal mesh. 2) Disabling the on to one pairing 3) making the two edges from the beginning. It always run for some time steps (either 5-10 or even 900 depends on which solution from the above-mentioned I followed) but always gets to the same error (The node numbering changes though in every error). The even more strange incident is that this error only persists on my finer mesh (since I am trying to do a grid independence study). For the medium and coarse sized meshes, the periodic boundary condition works perfectly for the total of timesteps I want to run the simulation. 

Any further support would be much appreciated.