Which convergence should we ask for?
A bounded sequence in an infinite-dimensional normed space need not have a norm-convergent subsequence. Yet in a dual space, pointwise convergence on the original space is often enough to pass to a limit.
Definition 6.2.1
(Weak-star convergence)
Let be a real or complex normed space. A sequence in its continuous dual converges weak-star to , written , when
The tests here are vectors of . Weak convergence in uses every member of as a test; it can therefore be stronger. We identify with the functional on , but in general not every element of arises this way.
Common mistake
(Treating weak-star convergence as weak convergence in the dual)
Weak convergence in implies weak-star convergence, but not conversely unless every element of is an evaluation. In , with acting by on , the unit vectors satisfy for every , so . They do not converge weakly in : the sum functional, given by , takes value at every and value at the only possible limit . Always name the test space.
The following compactness theorem is stated here without proof.
Theorem 6.2.2
(Banach–Alaoglu theorem)
For every normed space , the closed unit ball of is compact in the weak-star topology.
Compactness of a general topological space does not by itself guarantee that every sequence has a convergent subsequence. The useful sequential statement below needs an additional hypothesis: separability of the predual.
A sequential compactness theorem
Theorem 6.2.3
(Weak-star sequential compactness for a separable predual)
Let be a separable normed space and let satisfy . There are a subsequence and with such that for every .
Proof
(Diagonal extraction on a dense sequence, then a three-term estimate)
Choose a countable dense set in . The scalar sequence is bounded by , hence has a convergent subsequence. From it choose a further subsequence on which converges, and continue. Taking the diagonal subsequence gives convergence at every : after its th term it lies in the subsequence selected at stage .
For arbitrary , choose close to . The uniform operator norm bound implies
First choose so the first term is small, then take large. Thus is Cauchy in the scalar field and has a limit; call it . Passing to limits in linearity shows that is linear. Also , so is continuous and . This is precisely weak-star convergence.
Proof X-ray
Density works only because the norms are uniformly bounded
Separability is used once, to make the diagonal argument countable: one extraction per point of the dense sequence. Completeness of the scalar field, not of , is what turns the Cauchy sequences into limits.
The bound is used three times: to make each scalar sequence bounded, so that Bolzano–Weierstrass applies; to make the first term of the estimate independent of the subsequence indices, which transports the Cauchy property from a nearby dense-set point to any chosen ; and to show that the limit is continuous with . The middle use is the non-obvious one. Without it, convergence on a dense set controls nothing elsewhere. On , let ; then for every finitely supported , a dense set, but at with and all other coordinates zero, while for other , so does not converge.
A concrete dual sequence
The dual of is identified with by . The theorem applies, because is separable.
Worked example 6.2.4
(Coordinate functionals on )
Let be the sequence with a in position and zeros elsewhere. For every ,
because the terms of an absolutely summable series tend to zero. Thus converges weak-star to zero in . Its norm remains one, so this is not norm convergence. In this example the full sequence converges; the theorem promises only a subsequence for an arbitrary bounded sequence.
Worked example 6.2.5
(Partial-sum indicators converge to the constant one)
In the same dual, let have its first entries equal to one. For , , and these partial sums converge to . Hence converges weak-star to the constant-one sequence . Nevertheless has entries from position on, so for every . The weak-star limit has the same norm as the terms here, unlike the previous example, where the norm dropped from one to zero.
Worked example 6.2.6
(Separability cannot be dropped)
Take , which is not separable, and let . These are norm-one functionals. Given any subsequence , choose with and all other coordinates zero. Then does not converge. Thus this bounded sequence has no weak-star convergent subsequence, even though the dual unit ball is weak-star compact by Banach–Alaoglu. The difference is between compactness, defined by open covers, and sequential compactness.
The norm bound in the theorem is a premise, not a consequence of pointwise behavior, as the next example shows.
Counterexample mode
Pointwise boundedness need not bound the norms on an incomplete domain
Claim to test: if continuous functionals converge pointwise on a normed space, their norms are bounded. Let , the finite sequences with the norm, and define . Every is continuous, with : , with equality at the th unit vector. For each fixed , once exceeds the support of . Nevertheless there is no uniform bound on the operator norms. The repair is completeness of the domain: on a Banach space the uniform boundedness principle gives a norm bound from pointwise boundedness, but is not Banach.
This example concerns pointwise boundedness versus bounded norms; it does not contradict the theorem, whose uniform norm bound is an explicit premise.
Practice and solutions
Quick check
(Show that weak convergence in implies weak-star convergence.)
Each defines an element of .
Solution · Answer
For each , evaluation is a continuous linear functional on ; weak convergence tests against , so .
Quick check
(For the coordinate functionals on , compute and the weak-star limit.)
Test at the th unit vector.
Solution · Answer
We have and equality at the th unit vector, so . Since for every , the weak-star limit is zero.
Quick check
(Why does the diagonal proof not claim convergence of the entire original sequence?)
Look for a bounded sequence with two different subsequential limits.
Solution · Answer
At each stage we discard terms to ensure scalar convergence, and the diagonal retains only selected indices. For nonzero , the bounded sequence has even terms equal to and odd terms equal to , two different weak-star limits, so the full sequence need not converge.
Quick check
(Let be separable and bounded in . If every weak-star convergent subsequence has limit zero, prove that the whole sequence converges weak-star to zero.)
Argue by contradiction and apply the theorem to a bad subsequence.
Solution · Answer
If convergence failed, some , and subsequence would satisfy throughout. That subsequence is still bounded by , so the theorem supplies a weak-star convergent further subsequence; its limit is zero by hypothesis, so its values at tend to , contradicting .