Compactness survives taking adjoints
A compact operator sends a bounded set into a relatively norm-compact set. On Hilbert spaces, orthogonal projections can approximate its range by finite-dimensional spaces. General Banach spaces need not have such projections. The compact-adjoint theorem instead treats dual functionals as a uniformly equicontinuous family on the compact image of the unit ball.
Let be real or complex Banach spaces and a bounded linear operator. The adjoint is given by . Compactness always means norm-relative compactness of the image of the unit ball. We use the metric Arzelà–Ascoli theorem and the canonical isometric embeddings and .
Theorem 2.5.1
(Schauder's theorem for compact adjoints)
The operator is compact if and only if its Banach adjoint is compact. No reflexivity hypothesis is needed.
From the operator to its adjoint
Proof
(Restrict dual functionals to one compact set)
Suppose is compact and let , where is the closed unit ball. This is a compact metric space. For each in the unit ball of , restrict to . These restrictions are uniformly bounded because is bounded, and uniformly equicontinuous because
Arzelà–Ascoli therefore gives a uniformly convergent subsequence from any sequence of such restrictions. For , continuity on and the definition of the adjoint give the exact identity
Consequently the corresponding adjoint images form a Cauchy subsequence in and converge there, since is complete. This proves that maps the unit ball to a relatively compact set.
Back from the adjoint without reflexivity
Proof
(Use the bidual and the canonical embeddings)
Assume instead that is compact. Applying the just-proved direction to shows that is compact. The embeddings satisfy
For each fixed , both sides evaluated at any give . Given a bounded sequence in , the isometry makes bounded. Compactness of and (1) give a subsequence for which converges in . The range is closed because is an isometry and is complete. Its limit is therefore for some , and isometry gives in . Every bounded sequence has such an image subsequence, so is compact.
Neither canonical embedding is assumed onto. In particular, the proof does not identify the whole bidual with the original space.
Proof X-ray
Compactness is used before finite-dimensional structure
The first direction applies Arzelà–Ascoli to restrictions of dual functionals on the compact range set. The second direction applies that result again and uses only the closed isometric image of in its bidual. No finite-rank approximation property of arbitrary Banach spaces is assumed.
Worked example 2.5.2
(A compact diagonal map on a nonreflexive space)
Let be . Its finite-coordinate truncations have finite rank and converge to in operator norm, with error at most . Thus is compact by the norm-limit theorem for finite-rank operators. Under the identification with pairing (also in the complex case), its adjoint is . Its truncations have the same operator-norm error bound, illustrating the theorem directly.
Here , and is the same diagonal map on , compact by finite-rank truncation. The canonical embedding is not onto, giving a concrete nonreflexive setting for the theorem.
Two further examples test the theorem at its extremes: a degenerate finite-rank case, and a pair of operators neither of which is compact.
Worked example 2.5.3
(A rank-one operator and its adjoint)
Fix and , and put . Then , and the range lies in , so is a bounded finite-rank operator and hence compact. For and ,
Thus has range in and is compact as well. By Hahn–Banach, , so . In the edge case or , both operators are zero.
Worked example 2.5.4
(The backward shift on and its adjoint)
Let be . For ,
so is the forward shift on . Neither operator is compact. The bounded sequence in has images with for . The bounded sequence in has images with for . Neither image sequence has a Cauchy subsequence, consistent with the equivalence in Schauder's theorem.
The diagonal example also shows why the second direction of the proof had to be written with embeddings.
Common mistake
(Identifying the bidual with the original space)
It is tempting to say that compact gives compact, and that is just . For that is false: acts on , which is strictly larger than . The proof instead uses only and the fact that is closed in , which hold without reflexivity.
Weak-to-norm convergence and its limits
Definition 2.5.5
(Completely continuous operator)
A bounded linear operator is completely continuous if weakly in implies in the norm of .
Proposition 2.5.6
(Compact operators improve weak convergence to norm convergence)
Let be compact. Then is completely continuous. Moreover, if in the weak-star topology , then in the norm of .
Proof
(Norm-convergent subsequences with a forced limit)
Let in . The weakly convergent sequence is bounded by uniform boundedness. Compactness gives norm-convergent subsequences of its image, and every such limit is because is weak-to-weak continuous. If the full image did not converge in norm, a subsequence bounded away from would have a norm-convergent subsubsequence, whose limit is again , a contradiction.
Now let weak-star. The sequence of dual functionals is norm bounded by uniform boundedness, using completeness of . By Schauder's theorem, is compact, so its images have norm-convergent subsequences, and their limits are uniquely determined by evaluation on each :
Again a subsequence contradiction gives convergence of the full sequence. The original weak-star limit is tested on , not on all of .
Common mistake
(Treating weak-star convergence as weak convergence)
In the unit vectors satisfy weak-star, since for every . They do not converge weakly in : the functional given by takes the value on every . The second statement of the proposition needs only the weaker weak-star hypothesis, which is why it is proved separately rather than read off from complete continuity of .
The converse of the first statement is where the two notions part company.
Counterexample mode
Weak-to-norm sequential continuity alone need not mean compactness
The identity on sends every weakly convergent sequence to a norm-convergent sequence by the Schur property, proved in an earlier note. Yet it is not compact: its unit ball contains the standard unit vectors, whose pairwise distances are two. Thus the weak-to-norm sequence property alone does not characterize compact operators on arbitrary Banach spaces.
If the domain is reflexive, the converse does hold. Every bounded sequence has a weakly convergent subsequence by reflexive weak compactness and Eberlein–Šmulian; the weak-to-norm property then gives a norm-convergent image subsequence. Reflexivity is used in this argument, not in Schauder's theorem itself.
Practice with solutions
Quick check
(For the compact diagonal map , , let . Show that weak-star and compute .)
Evaluate on an arbitrary .
Solution · Answer
For every , , so in . The adjoint gives , whose norm is . This is an explicit instance of weak-star-to-norm convergence.
Quick check
(Take the identity on and its adjoint on . Does every weak-star null sequence in converge to zero in norm under the adjoint?)
Reuse the sequence .
Solution · Answer
No. The same are weak-star null in , but the adjoint is the identity and . This is consistent with the proposition, because the identity on is not compact: its unit vectors have no norm-convergent subsequence.
Quick check
(For and , evaluate both sides of on .)
Unwind the definitions of , and .
Solution · Answer
By definition, . The right side also has value . Equality on every proves the equation.