Why an equation is not yet an existence proof
A first-variation calculation says that a sufficiently regular minimizer must satisfy an equation. It does not guarantee that a minimizer exists. For instance, on has infimum zero but never attains it. A different danger is that a minimizing sequence converges to an object outside the chosen class. The direct method addresses both dangers: choose a complete admissible space, obtain a convergent subsequence in a useful topology, and show that the objective cannot jump upward when passing to the limit.
Definition 7.1.1
(Weak convergence)
A sequence in a real Hilbert space converges weakly to , written , if for every fixed .
Strong convergence means and implies weak convergence by Cauchy–Schwarz, since . The converse is false: the standard unit vectors in converge weakly to zero while .
Definition 7.1.2
(Weak lower semicontinuity and coercivity)
Let and .
- is weakly lower semicontinuous if implies .
- is coercive on if for every sequence in with .
- is sequentially weakly closed if and imply .
These are different requirements: coercivity prevents escape to infinity; lower semicontinuity controls the value at a limit; weak closedness keeps the limit admissible.
The compactness argument
Theorem 7.1.3
(Existence by the direct method)
Let be a real Hilbert space and be nonempty and sequentially weakly closed. Suppose is weakly lower semicontinuous, coercive on , finite at some point of , and bounded below on . Then attains its infimum on .
Proof
(Minimizing sequence, weak subsequence, lower semicontinuity)
Let ; it is finite because is bounded below on and finite at a point of . Choose with . Coercivity makes bounded: otherwise a subsequence would have norms tending to infinity, so its values would tend to , while they converge to the finite number . The weak sequential compactness theorem for bounded sequences in a Hilbert space supplies a subsequence . Since is sequentially weakly closed, . Lower semicontinuity now yields
By definition of , as well. Hence .
Proof X-ray
Track what each hypothesis does
The minimizing sequence controls objective values, not the functions themselves; the proof is a chain that converts one kind of control into the next. Boundedness below and one finite value make a real number, so "values converge to " is meaningful. Coercivity is then used once, by contradiction, to turn value control into a norm bound. The only non-elementary input is the compactness theorem, and it is used in the weak topology: bounded sets in an infinite-dimensional Hilbert space are not norm compact, so no strongly convergent subsequence is available. Because the limit is only weak, the two remaining hypotheses must be weak ones: weak closedness of keeps admissible, and weak lower semicontinuity gives the inequality in the one direction needed. Upper bounds on come from the limit; the lower bound is free because . A formal stationary equation supplies none of these steps.
The hypotheses are checked in practice with two standard facts.
Lemma 7.1.4
(Norms and linear functionals under weak limits)
In a real Hilbert space, is weakly lower semicontinuous, and every continuous linear functional is weakly continuous.
Proof
(Expand the square around the weak limit)
If , then
and the middle term tends to zero by weak convergence while the last is nonnegative, so . A continuous linear functional has the form by the Riesz representation theorem, so it converges along by the definition of weak convergence.
Worked example 7.1.5
(A weak limit can lose norm)
In , for each fixed , because . Hence . Yet for every , while . The inequality of the lemma is therefore strict here, . Lower semicontinuity is all one can expect: the square norm is not weakly continuous, and the convergence is not strong because .
An endpoint problem with a source
Let : the completion of smooth compactly supported real functions under the norm . Its members have zero endpoint traces. The one-dimensional Poincaré inequality makes this norm control the function as well as its weak derivative. Fix and consider
By Cauchy–Schwarz and Poincaré, . Thus , proving coercivity and a lower bound. The first term is weakly lower semicontinuous and the second weakly continuous by the lemma. The existence theorem, with , supplies a minimizer.
Proposition 7.1.6
(Weak Euler equation and uniqueness)
A function minimizes if and only if
The minimizer is unique, and its energy is .
Proof
(Differentiate along lines, then compute the exact gap)
If minimizes , then for any the real quadratic has a minimum at , so its linear coefficient vanishes. Conversely, the weak equation gives, for every , the exact gap
so is a minimizer, and equality forces , so the minimizer is unique. Taking in the weak equation gives , hence .
If additional regularity gives , integration by parts identifies the weak equation with almost everywhere and zero endpoint traces. In this one-dimensional example that regularity follows from the weak equation when ; the existence proof itself did not assume a classical second derivative.
Worked example 7.1.7
(Constant source )
The function has zero endpoints and . Integration by parts against smooth compactly supported test functions, followed by density in , verifies the weak equation, so the exact gap proves it is the unique minimizer. Its energy is
As a check, and , so . Merely solving the differential equation would not by itself establish the global minimum; the gap does.
Worked example 7.1.8
(Linear source )
Try . Then and , so ; as before, the weak equation holds and is the unique minimizer. The energy formula gives
Directly: , so , and , giving .
Counterexample mode
Strong closedness alone is not enough
One might hope that norm closedness of suffices. In , the set is norm closed: distinct points are distance apart, so a norm-convergent sequence in is eventually constant. But , so is not weakly closed. The functional is continuous, weakly lower semicontinuous by the lemma, and coercive, since the linear part is bounded by . It has , so its infimum on is and is not attained. Weak closedness in the existence theorem cannot simply be replaced by norm closedness.
Common mistake
(Solving the Euler equation and calling it a minimizer)
A solution of the Euler equation is only a candidate. On , has no stationary point and no minimizer, and in general a stationary point can be a saddle. In the endpoint problem, the exact gap formula is what upgrades a solution of the weak equation to the unique global minimizer.
Common mistake
(Bounded below is not coercive)
is bounded below by on , but a minimizing sequence escapes to infinity. Coercivity, not a lower bound, is what keeps minimizing sequences bounded.
Practice and solutions
Quick check
(Why does in , and why is the convergence not strong?)
Test against a fixed .
Solution · Answer
For each , , since the terms of a square-summable sequence tend to zero. Yet for all , so there is no strong convergence.
Quick check
(For the endpoint problem with , identify the unique minimizer without solving an ODE.)
Look at the sign of the energy.
Solution · Answer
The energy is , which is nonnegative and zero only at because is a norm. So is the unique minimizer.
Quick check
(Let on a real Hilbert space. Find its minimizer and minimum value.)
Complete the square.
Solution · Answer
, so is uniquely minimized at , with minimum .
Quick check
(Give a coercive function on that fails to attain its infimum because it is not lower semicontinuous.)
Change the value of at one point.
Solution · Answer
Take and for . It is coercive and has infimum zero, approached as , but no point attains it. Lower semicontinuity fails at zero: .