A weaker way to observe convergence
Norm convergence controls every continuous linear measurement at once. In an infinite-dimensional space this can be too restrictive: a sequence may move into new directions, keeping its norm, while each fixed measurement tends to zero. Weak convergence records this latter behavior. Its usefulness comes with a warning: changing the measurement with the index is not allowed in the definition.
Definition 6.1.1
(Weak convergence and the weak topology)
Let be a real or complex normed space with continuous dual . A sequence converges weakly to , written , if
The weak topology on is the coarsest topology making every continuous; its convergent sequences are exactly those just defined. A set is weakly compact when it is compact in this topology.
The functional is fixed before the limit is taken. Norm convergence implies weak convergence because . The converse generally fails. Weak limits are unique: if all functionals vanish on the difference of two proposed limits, Hahn–Banach implies that difference is zero.
For a Hilbert space , Riesz representation says that exactly when for every fixed , using the convention that the inner product is linear in its first argument. Weak convergence is relative to the stated space and norm; one cannot replace its dual with an arbitrary smaller family of tests without proof.
Worked example 6.1.2
(In finite dimensions weak and norm convergence coincide)
Let be finite-dimensional with basis , and let be the th coordinate of . Each is linear, and linear functionals on a finite-dimensional normed space are continuous, so . If , then for every , and
So the distinction drawn in this unit is invisible in or ; every example separating the two notions must use infinitely many directions.
Moving directions and boundedness
Worked example 6.1.3
(Unit vectors in converge weakly to zero)
Let be the th standard unit vector. Every has , because . So for every fixed , and , although for every and for . A fixed measurement eventually sees little of a vector moving into new coordinates.
Common mistake
(Letting the test functional depend on the index)
The functional gives value one at , and one may be tempted to conclude that does not tend weakly to zero. But that functional changes with ; it is not a single fixed test. Weak convergence asks, for each fixed , about the scalar sequence . Requiring instead is, by Hahn–Banach, exactly norm convergence.
Proposition 6.1.4
(Weakly convergent sequences are bounded)
Every weakly convergent sequence in a normed space is bounded, and its limit satisfies
Proof
(Uniform boundedness on the dual)
For each , define by . Since converges weakly, for every fixed the scalar sequence is bounded. The dual is Banach, even if is not, so the uniform boundedness principle applied to gives . Hahn–Banach gives , proving boundedness.
For the second claim, every with satisfies . Taking the supremum over the dual unit ball and using Hahn–Banach again yields the inequality. It remains valid if the liminf is infinite.
Proof X-ray
Why boundedness is not just pointwise boundedness
The hypothesis gives only pointwise information: for each , the numbers are bounded. The key move is to regard the vectors as functionals on the dual space. That reverses the roles of vector and test, and puts the problem in the setting of the uniform boundedness principle, whose hypothesis is completeness of the domain. Here the domain is , which is always complete; so the argument works even when itself is incomplete.
Hahn–Banach enters twice, both times through the norming identity : once to identify with , and once to turn the bound on each into a bound on . The inequality is only a liminf because each is passed to the limit separately before the supremum over is taken; the order of the two operations cannot be reversed.
This is weak lower semicontinuity of the norm along sequences, not continuity: the example has limit norm zero while every term has norm one. It follows that a closed norm ball is weakly sequentially closed, since implies for every center .
Common mistake
(Expecting the norm to pass to the weak limit)
A norm-closed set need not be weakly sequentially closed: the unit sphere of contains every but not their weak limit zero. Only the inequality survives, and it can be strict. Statements such as “the weak limit of unit vectors is a unit vector” are false in every infinite-dimensional Hilbert space.
Proposition 6.1.5
(Weak convergence plus convergent norms in a Hilbert space)
In a Hilbert space, if and , then .
Proof
(Expand the square)
Weak convergence against the fixed vector gives , so
The norm assumption is essential, as shows.
A bounded sequence in a separable Hilbert space
Theorem 6.1.6
(Weak subsequences in a separable Hilbert space)
In a separable Hilbert space, every bounded sequence has a weakly convergent subsequence.
Proof
(Diagonal extraction on coordinates)
Let be a finite or countable orthonormal basis and suppose . In the finite-dimensional case, ordinary bounded-sequence compactness gives a norm-convergent subsequence, which is also weakly convergent. In the countably infinite case, successively extract subsequences on which the first coordinate converges, then the second, and so on; take the diagonal subsequence . For every , . Bessel's inequality gives
Consequently , and the orthonormal-series theorem gives . For a finite linear combination of basis vectors, . For arbitrary , approximate it in norm by such a finite combination . Cauchy–Schwarz bounds the remaining term by , independently of . Hence .
This argument uses a countable basis to perform the diagonal extraction and completeness to construct . It does not imply that the original bounded sequence converges, nor that its subsequence converges in norm. For example, no subsequence of the orthonormal sequence converges in norm, since distinct terms are distance apart.
Reflexivity and the compactness theorems used
Definition 6.1.7
(Canonical embedding and reflexivity)
The canonical map is defined by ; it is an isometric embedding by Hahn–Banach. A normed space is reflexive if it is Banach and is onto.
The word refers to the canonical map, not merely to the existence of some abstract isomorphism between and .
Proposition 6.1.8
(Hilbert spaces are reflexive)
Every Hilbert space is reflexive.
Proof
(Apply Riesz twice)
To see the complex case with our convention, write . Riesz says every is a unique , and . For , the map is therefore continuous and conjugate-linear. Its complex conjugate is a continuous linear functional of , so Riesz gives some with , that is, for every . Thus . The real case is the same without conjugations.
The following deep results are stated, not proved here. The first ties reflexivity to compactness of the unit ball in the weak topology.
Theorem 6.1.9
(Reflexive-space weak compactness theorem (Kakutani))
A Banach space is reflexive if and only if its closed unit ball is weakly compact.
Compactness in the weak topology is defined through open covers, and in general it is not equivalent to a statement about sequences. For subsets of Banach spaces the next theorem shows that it is.
Theorem 6.1.10
(Eberlein–Šmulian theorem)
A subset of a Banach space is weakly compact if and only if every sequence in it has a weakly convergent subsequence whose limit lies in that subset.
For the unit-ball equivalence, see Siyan Daniel Li’s functional-analysis notes, Theorem 5.1; the general relative-compactness version of Eberlein–Šmulian is stated and proved in these functional-analysis notes. Together they give the following characterization.
Theorem 6.1.11
(Reflexive subsequence theorem)
A Banach space is reflexive if and only if every bounded sequence has a weakly convergent subsequence.
Proof
(Combine the two stated theorems)
For the forward direction, a bounded sequence lies in a multiple of the weakly compact closed unit ball; Eberlein–Šmulian supplies a weakly convergent subsequence. For the converse, apply the assumed property to sequences in the closed unit ball. Weak lower semicontinuity puts their limits in that ball, so Eberlein–Šmulian makes the ball weakly compact; the reflexive-space theorem then gives reflexivity.
These statements concern weak topology, not norm compactness. An infinite-dimensional Hilbert unit ball is weakly compact but is not norm compact. Without reflexivity, the subsequence property can fail outright.
Counterexample mode
A bounded sequence in ℓ¹ with no weak subsequence
Claim to test: in every Banach space, a bounded sequence has a weakly convergent subsequence. In , take the unit vectors , all of norm one. If , each coordinate functional gives , so . But the bounded functional , represented by the constant sequence , has for all , while . So no subsequence converges weakly. The repair is the hypothesis of the reflexive subsequence theorem; in particular, that theorem shows that is not reflexive.
Compact operators turn weak convergence into norm convergence
Proposition 6.1.12
(Compact operators upgrade weak convergence)
Let be a bounded linear operator between normed spaces that is compact: images of bounded sequences have norm-convergent subsequences in . If in , then .
Proof
(Every subsequence has a sub-subsequence with the right limit)
Weakly convergent sequences are bounded, so compactness gives convergent subsequences of their images. Also : for each , , hence . Suppose norm convergence fails. A subsequence then stays at least from ; compactness gives a further subsequence converging in norm to some . Norm convergence implies weak convergence, and uniqueness of weak limits gives , a contradiction.
Proof X-ray
Compactness supplies limits, weak convergence names them
The argument separates existence from identification. Compactness gives norm-convergent subsequences of the images, but says nothing about which vector is the limit; it needs a bounded input sequence, which is where boundedness of weakly convergent sequences enters. Weak convergence, transported to by composing each with , pins down the only candidate through uniqueness of weak limits. The proof closes with the subsequence principle, run by contradiction: if every subsequence has a further subsequence converging to the same , the whole sequence converges to .
Worked example 6.1.13
(A compact diagonal operator on the unit vectors)
In the diagonal operator is compact: its finite-rank truncations satisfy , and the norm limit of compact operators into a Banach space is compact. It sends to , and , as the proposition predicts from . Compactness strengthens convergence of this particular weakly convergent sequence after applying ; it does not make itself converge in norm. By contrast the identity, which is not compact, sends to itself, and for all .
Practice with solutions
Quick check
(In , show that for converges weakly to but not in norm.)
Test against a fixed .
Solution · Answer
For each fixed , . But . Any norm limit would also be a weak limit, which by uniqueness must be ; therefore no norm limit exists.
Quick check
(If in a Hilbert space and , prove norm convergence. Why does the conclusion fail for ?)
Expand .
Solution · Answer
; weak convergence controls the inner product with the fixed vector , and the norm assumption controls the remaining terms, so the sum tends to . For , the norms are all and do not converge to the norm of the weak limit.
Quick check
(Show that no subsequence of the unit vectors in can converge weakly, using two kinds of continuous functionals.)
Use coordinate functionals and the sum functional.
Solution · Answer
Coordinate functionals force every coordinate of a proposed limit to vanish, so the limit would be . The bounded sum functional takes value one on every unit vector and value zero at , a contradiction.
Quick check
(Let be compact on a Hilbert space and an orthonormal sequence. Prove .)
First show .
Solution · Answer
Bessel's inequality implies for each fixed , so . The compact-operator proposition gives in norm.