I know a number of ways of solving this basic DE:
$\ddot{u} = -u$
Besides the fact that the solution is obvious, one can do:
$\ddot{u} = \frac{d\dot{u}}{dt} = \frac{d\dot{u}}{du}\frac{du}{dt} = \frac{d\dot{u}}{du}\dot{u} = -u$
and then bring the du in the denominator to the right and integrate both sides to get:
$\dot{u}^2 = - u^2$ or $\dot{u} = \pm i u$
Then if the solution still wasn't obvious to us, we could separate variables again to get u(t) = e$^{\pm it}$
Now my question is, why is it not possible to use separation of variables more directly, that is, what is wrong with the following:
$\ddot{u} = \frac{d}{dt}\frac{du}{dt} = - u$
and bring the dt over to the right and u to the left to get:
$\frac{d}{dt}\frac{du}{u} = - dt$
Which can be integrated on both sides. The problem is that this doesn't seem to work, you get a wrong solution to the DE (u(t) = e$^{-t^2/2}$), even if you keep careful track of the limits of integration. But I can see no problem with it; the solution we know is well-behaved, doesn't have any singularities and is differentiable. What goes wrong?