4.3. Lessons. In this section there are two lessons:
(1) Before solving a problem of interest, always test a model by first solving a problem
whose solution is mathematically tractible.
(2) The two discretization methods that we’ve discussed throughout this paper are commonly
used for simple numerical integration of ODEs. In this case, the problem is
well-posed if the number of equations matches the number of variables; i.e., there
are no degrees of freedom. One can show that a problem is well-posed with respect
to midpoint discretization if and only if it is well-posed with respect to trapezoidal
discretization. However, as we saw in this example, for control problems, i.e. problems
where there are degrees of freedom, the midpoint discretization will sometimes
have fewer degrees of freedom than the trapezoidal discretization. It seems
that the midpoint discretization has the “correct” number and that the trapezoidal
discretization has too many.