This is a method used to show that a minimizer *exists* for a variational problem, though it typically provides no clues as to *what* the minimizer is. It is typically referred to as “the direct method in the calculus of variations”. See yeseterday’s post, at least the first paragraph or two, for background.

First, a point I forgot about yesterday since it is such a convention: “why minimization?” I would guess this is motivated by physics, and has been explored by Hildebrandt and Tromba in The Parsimonious Universe– roughly, that nature seeks to minimize (rather than maximize) energy spent. So much of math has been motivated by physics (often useful, as it provides an intuition as to what *should* be true), it is only natural many of our questions are posed in this manner. This parsimony can also explain why we typically make sure that the Lagrangian is nonnegative- since energy is nonnegative.

Now, the direct method, along with intuition of how we find a least area surface given a prescribed boundary, and what could go wrong in general. We denote the area of a surface *u* by *F[u]*:

1.

**Show that there is at least one candidate minimizer**. For the least area surface, we pick a point

*not*on the boundary, and draw a straight line from that point to each boundary point to make our first candidate. As a silly example of what could go wrong, consider trying to minimize the integral of

*|u|*, subject to

*u(0) = 1*and

*u(0) = 2*. Well, we just cannot satisfy that boundary condition with

*any*function.

2. **Take an infimum**. We now have a nonempty set of candidates , which is associated with a nonempty set of real numbers . Going back to the physics explanation, I’ll assume that the integrand is bounded below by zero, and invoke a property of the real numbers to say that there is a *greatest lower bound*. It will be our goal to find a surface realizing this infimum. Notice though that this step is *not* constructive, so right here we lose our chance at finding a minimizer analytically. Really the only thing that could go wrong is if our integrand is not bounded below. It would be somewhat silly to try to find the function whose integral is the “most negative”.

Once we have an infimum, *M*, we can take a sequence of candidates which converge to the infimum, . Notice that the candidate functions do not necessarily converge to anything. Yet.

**Invoke compactness.**Now, lurking under the surface has been this “space of admissible functions”. Often this is a Sobolev space, which has the property that the unit ball is “(weak-*) compact”. See this short paper (pdf link) by Frank Morgan on compactness, which he called “the most important concept in mathematics”. Compactness implies that

*every bounded sequence has a convergent subsequence*. Here “bounded” means “with respect to the norm on the space”. In the case of area minimizers, there is a “mass norm” on the admissible functions (“integer rectifiable currents”) which measures the surface area, and a famous (for some value of the word “famous”)

*compactness theorem*for these functions (defining a space of admissible functions that was physically relevant and so that such a theorem existed was, perhaps, the most difficult part of Plateau’s problem). Hence, we now have a subsequence , and a target (admissible) function

*u*so that

*and*, still.

Last thing to show is that *F[u] = M*. The two conditions we’ve established certainly nod suggestively at this last equation, but don’t give it themselves. The figure at right shows what could go wrong.

4. **Lower semicontinuity of the functional. **We need to show that *F[u] = M. *A typical approach is to prove that the functional is *lower semicontinuous*. That is to say, (fun story: when I asked my advisor to define “upper semicontinuity”, he gave me the finger. Bonus points if you can see why this is a reasonable definition). If this is true, then we have , and are done. The surface area is lower semicontinuous, which is how we complete the proof of the existence of area minimizing surfaces.

The Takagi curves from yesterday provide an example of a functional which is *not* lower semicontinuous, since , but .

One of my professors related the following mnemonic device: horizontal line with a point above it (i.e., the graph of the characteristic function of a singleton) depicts USC kicking a field goal. That was back in the days when USC ruled college football.