This problem came from the way the boundary conditions were imposed in our code: the value of
the wave function on boundaries was put at zero at every time step. Consequently it generated an
infinite barrier at both ends of our space interval but a discontinuity in the wave function that was of
course non-physical. We solved this problem by putting at boundaries a quickly increasing exponential
potential that allowed us to impose smoother boundary conditions and avoid discontinuities but still
confine our wave packet in a box. Fig. 3 shows the success of this method: the loss of norm is only
less than 3.10 5%.