Convergence behavior of current interpolation schemes

[GardevoirX]

As mentioned by @kvhuguenin in #70, the change of the interpolation scheme from the P3M one to the Lagrange should make things converge faster, while actually the error of the force become larger (relative error from 3.5e-3 to 5.0e-3). Further tests are needed to see if Lagrange does converge faster, and if there is potential bug in the implementation.

The way how to do it is not determined for now.

[kvhuguenin]

Perhaps to add to this: the error estimates we use (or will use for PME) are also coupled to the interpolation schemes. Thus, one of the next PRs we will do on implementing the error bounds for PME will probably be the ideal place to investigate this further.

@GardevoirX there is also some testing being done in one of our main reference papers, How to mesh up Ewald sums. I. A theoretical and numerical comparison of various particle mesh routines, so this might be a good starting point for setting up our own testing routine. Ideally, we should obtain very similar results to them.