From Landau's book, here's the following extract:
I don't understand how that can be the case. Knowing the positions and velocities in a given moment allow me to calculate the new positions at a dt time afterwards, if accelerations are all zero. If not, I'll need to know the accelerations from...