[Simbelmynë]

One final post: I found the reason why one of the methods was so much faster. It was, of course, a programming error. The error forced the Poisson loop to exit prematurely. Since I ran all the cases for a very long time I still got good steady-state results, even if the transient behavior must have been very inaccurate.

So to sum up the thread I can say that the integral condition is a very good one (and easy to implement if the distribution of nodes on the boundary & boundary layer is uniform). Also, the technique by J. Holdeman seems interesting.

Now it is on to test the code versus some experimental systems.