Currently we ask `d^n r / d s^n = 0` for `n=2, 3, …` at the starting and ending times. But this is incorrect if the boundary velocities are nonzero.