A typical engineering-oriented course in ordinary differential equations focuses on solving initial value problems (IVP): first by elementary methods, then with power series (if nobody updated the syllabus since 1950s), then with the Laplace transform. One also considers the stability of equilibrium solutions and draws various diagrams to show how other solutions flow around equilibria. And then the semester is over.
And so, the boundary value problems (BVP) are likely to go unnoticed. Let’s compare:
- IVP answers the question: if I aim at this angle, who will be shot?
- BVP answers the question: how should I aim to shoot this guy?
Hm, which question is more practical?
- IVP has a unique solution, as long as the functions involved are not too bad. This was observed already by the Cheshire Cat: you’re sure to get somewhere if you only walk long enough.
- BVP can easily have no solutions, more than one, or infinitely many. The Scholarpedia article on BVP makes this fact intuitive with the example of the motion of a projectile: there may be one, two, or zero ways to hit the target, depending on how far it is.
The stability is also quite different for IVP and BVP. For example,
- the IVP , , has a unique solution , which is unstable
- the BVP , , has a unique solution , which is stable
The instability of IVP means that a small perturbation of the initial values such as , , will produce a solution that moves away from . Indeed, and the difference grows exponentially in time. One can also see this instability without solving the equation. Just convert it into a 2×2 system of 1st order ODE and find that the Jacobian matrix has eigenvalues . The positive eigenvalue makes the equilibrium unstable.
But if we perturb the boundary values, for example , , the new solution is . The plot is below: the important this is that , and this would be true even if we considered the BVP on the interval instead of .
Same happens if the boundary derivatives are prescribed: for example, and :
Notice how the perturbed solution tries to approach the unperturbed in the middle of the plot.
Where does this BVP stability come from? Turns out that the equation is the Euler-Lagrange equation for the energy functional . Although the Wikipedia page explains why, let’s verify this directly. First, perform the variation by replacing with , where is some small “bump” function that vanishes near the endpoints. We get
Expand, take the first derivative with respect to , then equate it to zero at (if is a critical point, the derivative must be zero.) One notices that all this stuff amounts to keeping only the terms in which appears to first power:
In the term we want to move the derivative from onto . That is, integrate by parts:
Now factors out, and since it was arbitrary, the Euler-Lagrange equation is , which is what we had. This explains why the solution was slightly bent toward the axis: it tried to minimize the integral while keeping the first derivative small, as to not have a large term .
The boundary conditions determine a space of all functions that satisfy them. The energy functional is a strictly convex function from to real numbers. Therefore, it has only one critical point on which is the global minimum of energy, and therefore a stable equilibrium of whatever object is described by the energy .