We optimize the lattice and test several distances

We divide the system in such a way that there are two equivalent, and hence neutral regions.

Here are the distances (Angstrom) as obtained with a QM and an MM method
  distance         qm         mm    err(mm)
       B-H      1.182      1.181     -0.001
       N-H      1.007      1.041      0.034
       B-N      1.431      1.498      0.067

Of course the force field results do not exactly match the QM results, the error displayed in the last column

Now we try the hybrid engine, can we improve the bonds in the QM region?

We start from the geometry calculated with the (cheap) forcefield

In this table we show the errors in bond lengths (in the QM region) of the hybrid method with respect to the QM method

      embedding         capping          energy             B-H             N-H             B-N
     mechanical           fixed       -6.845015           0.004          -0.007          -0.049
     mechanical      fractional       -6.746367           0.013          -0.006          -0.034
  electrostatic           fixed       -6.748753           0.002          -0.004          -0.040
  electrostatic      fractional       -6.652196           0.010          -0.003          -0.024

Here are some observations
     * the B-H distance is a bit worse than with a plain forcefield, especially with fractional capping
     * the N-H distance is much better than with the plain forcefield 
     * the B-N distance is a bit better than with the plain forcefield, now too short. Fractional capping works best.
     * Electrostatic embedding is doing slightly better than mechanical embedding, the biggest improvement is on the B-N bond