§ 1. Duality
1. Topologies compatible with a duality
In this section, E and F denote two vector spaces put into duality by a bilinear form B (II, p. 40). We recall (II, p. 41) that we defined two linear mappings
characterized by the relation
(1)
for .
Definition 1
A locally convex topology on E is said to be compatible with the duality between E and F if is a bijection from F onto the dual of the locally convex space obtained by assigning the topology to E.
If there exists one such topology , the mapping is injective, that is to say, the duality between E and F is separating in F (II, p. 41).
Proposition 1
(i) The closed convex subsets in E are the same for all the locally convex topologies on E which are compatible with the duality between E and F. (ii) *The bounded subsets of E are the same for all the locally convex topologies on E which are compatible with the duality between E and F.
Let be a topology on E compatible with the duality between E and F, hence finer than . If a convex subset of E is closed for , it is the intersection of closed, real half-spaces (II, p. 38, cor. 1), hence it is closed for . This proves (i). Assertion (ii) was proved in cor. 3 of III, p. 27.
Let denote the vector space F endowed with the weak topology . Then the linear mapping maps E onto the dual of (II, p. 43, prop. 3). Let be a family of bounded subsets of . By abuse of language, the inverse image under of the -topology on is called the . It is defined by the family of semi-norms
where A runs through . In particular, when is the family of finite subsets of F, the -topology is precisely the weak topology .
Definition 2
Let E and F be two spaces in duality. The Mackey topology on E, denoted by is defined as the -topology on E, where is the family of all subsets of F whose image in (under ) is convex, balanced and compact for .
When the duality between E and F is separating in F, is injective and the topology on F is the inverse image under of the topology on . In this case, consists of all those subsets of F which are convex, balanced and compact for .
In general, if , and if we denote by the restriction of the canonical bilinear form to , then E and are put in duality by , and this duality is separating in , since by definition we have , def. 2 shows that .
Remark 1
— Let A be a compact convex subset of a Hausdorff locally convex space G, and let be the closed convex balanced envelope of A. When the field K is , the set is the closed convex envelope of ; when K is , the set is contained in the closed convex envelope of . Consequently (II, p. 14, prop. 15), is compact.
We deduce, in particular, that when the duality between E and F is separating in F, the Mackey topology is also the -topology, where is the set of all convex subsets of F which are compact for . In an analogous way we define the Mackey topology on F.
Theorem 1 (Mackey). — Let E and F be two spaces in duality; suppose that the duality is separating in F. In order that a locally convex topology on E be compatible with the duality between E and F, it is necessary and sufficient that be finer than the topology and coarser than the Mackey topology .
Identify F with its image in under . Let denote the set of all subsets of F which are convex, balanced and compact for . By definition, is the -topology on E, hence is finer than .
Lemma 1
The subspace F of E* consists of all linear forms on E which are continuous for .
Every element of F is a continuous mapping for , hence for .
Conversely, let be continuous for . There exists a neighbourhood U of 0 in E (for ), such that on U; we can assume that there exists a set such that . In other words, belongs to the bipolar of A for the duality between and E. But the topology on F is induced by ; consequently A is convex, balanced and compact for , and the theorem of bipolars (II, p. 44, th. 1) implies the equality . Therefore we have that , from which the lemma follows.
Lemma 2
Let be a locally convex topology on E such that every linear form on E which is continuous for belongs to F. Then is coarser than .
Let be the set of convex, balanced neighbourhoods of 0 for . Let be the set of polars in F of elements of . By cor. 2 of III, p. 17, we have , and by cor. 1 of prop. 7 of III, p. 19, is identical with the -topology, where is the set of polars of sets of in the dual of E. But , by hypothesis, hence every set of is contained in a set of ; and the lemma follows.
Let be a topology on E compatible with the duality between E and F. Then is coarser than by lemma 2, and evidently is finer than . Conversely, F is the dual of E for the topology (lemma 1) and for the topology (II, p. 43, prop. 3), hence also for every topology intermediate between and .
Corollary
Let p be a semi-norm on E. The following conditions are equivalent :
(i) p is continuous for the topology ; (ii) every linear form f on E, such that , comes from an element of F.
(i) (ii) : if p is continuous for , every linear form f on E such that is continuous for , hence comes from an element of F by lemma 1.
(ii) (i) : let be the topology on E defined by the semi-norm p. If condition (ii) is satisfied, the linear forms on E which are continuous for belong to F. By lemma 2 is coarser than , hence p is continuous for .
Remark 2
Let K be a convex subset of F which is compact for the weak topology and a positive measure on K. Put
for all . It is immediate that p is a semi-norm. Moreover, for every , the relation « for all » implies that . This proves that the semi-norm p on E is continuous for the Mackey topology . \
Example
Let G be a locally convex space and G' its dual. On G', the weak topology and the topology of convex compact convergence (III, p. 14) are compatible with the duality between G' and G. In general, the strong topology and the topology of compact convergence on G' are not compatible with the duality between G' and G. Recall however that when G is Hausdorff and quasi-complete, the topology of compact convergence on G' coincides with that of convex compact convergence (III, p. 8), hence is compatible with the duality between G' and G.
Definition 3
Let E and F be two vector spaces in duality, and the family of subsets of F which are bounded for . Then the -topology on F is denoted by .
Similarly, we define the topology on F. It can be seen easily that the topology is identical with , and we can reduce to the case when the duality between E and F is separating in F.
Remarks. — 3) Let denote the space E endowed with the topology . The barrels (III, p. 24) in are the subsets of E which are convex, balanced closed and absorbent for . These are none other than the polars of the subsets of F which are convex, balanced and bounded for . Consequently, the family of all barrels in is a fundamental system of neighbourhoods of 0 for the topology in E. In other words, a semi-norm on E is continuous for if and only if it is lower semi-continuous for (cf. III, p. 24, prop. 1).
4) Let be a topology on E compatible with the duality between E and F. By prop. 1, (ii) of IV, p. 1, the topology on F is none other than the strong topology on F, when F is identified with the dual of E (with the topology ).
5) The topology on E is finer than . It is not, in general compatible with the duality between E and F (cf. however § 2). In particular, a subset of E which is bounded for is not necessarily bounded for .
2. Mackey topology and weakened topology on a locally convex space
Let E be a locally convex space and E' its dual. We put E and E' in duality by means of the canonical bilinear form on . This duality is separating in E'. We shall consider three topologies on E compatible with the duality between E and E' :
a) the given topology on E, which we shall call the initial topology, whenever any confusion is likely to arise;
b) the topology , called the weakened topology on E;
c) the topology , called the Mackey topology on E.
The initial topology is finer than the weakened topology and coarser than the Mackey topology; moreover, these three topologies can be distinct (IV, p. 49, exerc. 8).
By prop. 1 of IV, p. 1, these three topologies have the same closed convex sets, the same barrels, the same bounded sets and the same adapted bornologies. In particular :
Proposition 2
Let E be a locally convex space, and let A be a convex subset of E (for example, a vector subspace of E). The closure of A is the same for the initial topology and for the weakened topology of E.
Remark 1
For a family of elements of E to be total (resp. topologically independent) for the initial topology, it is necessary and sufficient that it is so for the weakened topology; this follows from prop. 2. Hence we can apply the criteria of II, p. 43.
Remark 2
Let and be two locally convex topologies on E, compatible with the duality between E and , being finer than . Then every neighbourhood of 0 for , which is convex and closed for is closed for by prop. 1 of IV, p. 1. Consequently (GT, II, § 3, No. 3, corollary) every subset of E which is complete for is so for also.
In particular, every subset of E which is complete for the weakened topology is complete for the initial topology, and every subset of E complete for the initial topology is so for the Mackey topology. If E is quasi-complete for the weakened topology, it is so for every topology compatible with the duality between E and . If it is quasi-complete for the initial topology, it is so for the Mackey topology.
Remark 3
Suppose E is Hausdorff (for the initial topology). Let A be a subset of E which is closed and bounded for , hence also for every topology compatible with the duality between E and . Since A is precompact for (III, p. 3, Remark 5), assuming that A is compact for is equivalent to A being complete for .
Therefore, on account of remark 2, we see that :
Proposition 3
Suppose E is Hausdorff, and its dual. Every subset of E which is precompact for the initial topology and compact for , is compact for the initial topology.
Remark 4
The topology (IV, p. 4, def. 3) is finer than the Mackey topology. If is distinct from , it is not compatible with the duality between E and . The space E is barrelled if and only if the initial topology is equal to (III, p. 24).
Proposition 4
Let E be a locally convex space. The Mackey topology on E is identical with the initial topology in each of the following cases :
a) E is barrelled; b) E is bornological; c) E is metrizable.
We note first that the Mackey topology on E is identical with the initial topology if and only if every convex subset of which is compact for , is equi-continuous. This is certainly the case if E is barrelled (III, p. 24, corollary).
Suppose E is bornological; let V be a convex and balanced neighbourhood of 0 in E for the topology . Let B be a subset of E, bounded for the initial topology. Since B is bounded for the Mackey topology, V absorbs B, and since E is bornological, V is a neighbourhood of 0 for the initial topology.
In case c), the space E is bornological (III, p. 12, prop. 2).
3. Transpose of a continuous linear mapping
In this section, and denote two locally convex spaces, with respective duals and .
Let be a linear mapping from into . For to be continuous when and are assigned the weakened topologies, it is necessary and sufficient that belongs to for all ; this is the case if is continuous. Then the linear mapping from into is called the transpose of and is denoted by .
Proposition 5
Let be a continuous linear mapping from into .
(i) If and are Hausdorff then is injective if and only if the image of is dense in for the weak topology .
(ii) For to be injective it is necessary and sufficient that the image of is dense in .
A vector subspace of is dense for the initial topology if and only if it is dense for the weakened topology (IV, p. 4, prop. 2). Prop. 5 then follows from II, p. 47, cor. 2.
Proposition 6
Let be a linear mapping from into which is continuous for the weakened topologies. For , let be a family of bounded subsets of . In order that is a continuous mapping from into , it is necessary and sufficient that, for every set , there exist sets in and a real number such that is contained in the closed convex balanced envelope of \footnote{In other words, is contained in the smallest adapted bornology containing (III, p. 3).}.
This is an immediate consequence of prop. 2 of III, p. 15.
Corollary
Let be a continuous linear mapping from into . Then is continuous when the duals are assigned the following topologies:* a) the weak topologies ; b) the strong topologies ; c) the Mackey topologies ; d) the topologies of precompact convergence. Moreover, if is Hausdorff, is continuous when the duals are assigned: e) the topologies of compact convergence (resp. compact convex).
The only point which requires a proof is the case c), when the topologies of and are not necessarily Hausdorff. Then for every linear form , is a linear form on ; hence there is a linear mapping which is continuous for the topologies and and is such that , where is the canonical mapping from into (). Consequently, if is a subset of such that is convex, balanced and compact for then is convex, balanced and compact for since the topologies and are Hausdorff.
Proposition 7
— Let be a linear mapping. We assume that is continuous for the weakened topologies of and .
(i) The mapping is continuous if and are assigned their Mackey topologies.
(ii) If is bornological or barrelled, then is continuous for the initial topologies of and .
(iii) In order that be continuous for the initial topologies of and , it is necessary and sufficient that the image under of every equicontinuous subset of be equicontinuous in .
The hypothesis implies that is continuous for the weak topologies and (II, p. 46, corollary) hence the image under of a convex, balanced and compact subset for is convex, balanced and compact for , the topologies and being Hausdorff. Therefore, assertion (i) follows from GT, X, § 1, No. 4, prop. 3, b). Assertion (ii) is a consequence of (i) : for, if is bornological or barrelled, its initial topology is the Mackey topology, and the Mackey topology of is finer than the initial topology of . Finally, the initial topology of is that of uniform convergence on equicontinuous subsets of (III, p. 19, cor. 1 of prop. 7). This proves (iii).
Corollary
— Suppose is a normed space. Let be a linear mapping from into . The following properties are equivalent :
a) is continuous;
b) is continuous for the weakened topologies;
c) the image of the unit ball in under is bounded in ;
d) for every sequence of points of tending to 0 for the initial topology, the sequence is bounded for the weakened topology of .
Since is bornological the equivalence of a) and b) follows from prop. 7; that of a) and c) is immediate. The equivalence of a) and d) follows from prop. 1 of IV, p. 1, and from prop. 1 of III, p. 11.
Proposition 8
— (i) Let be a normed space, with dual . For every , we have
(ii) Let and be two normed spaces and a continuous linear mapping from into . We have
Let . For every such that , we have
By Hahn-Banach theorem (II, p. 23, cor. 2), there exists an element in such that and . This proves (i).
Let us now prove (ii). By formula (3) and the definition of the transpose, we have
Remark 1
Formula (3) is a particular case of (4), corresponding to the linear mapping from K into E.
Remark 2
Put for . The above proof shows that B is a continuous bilinear form on , with norm (GT, X, § 3, No. 2) equal to .
Corollary
— *Let be a normed space satisfying the first axiom of countability. There exists a countable subset D of such that we have
for all .
Let B' be the unit ball of the dual of E with the weak topology assigned to it. Then B' is a compact metrizable space (III, p. 19, cor. 2); hence there exists a countable dense subset D' in B'. Put . Let ; the mapping from B' into K is continuous, therefore
Formula (5) now follows from (3).
4. Dual of a quotient space and of a subspace
Throughout this section, E denotes a locally convex space, M a vector subspace of E, and the orthogonal of M in the dual of E. Let be the canonical mapping from E onto ; then is injective, with image , hence defines a vector space isomorphism (not topological)
Similarly, let be the canonical injection from M into E. Then is surjective (II, p. 24, prop. 2); its kernel is equal to , and we get a vector space isomorphism (not topological)
Proposition 9
— (i) For a subset A of to be equicontinuous, it is necessary and sufficient that is an equicontinuous subset of .
(ii) Let be a set of bounded subsets of , and the set of the images of subsets in . Then is an isomorphism from onto , where is assigned the topology induced by that of .
(iii) Suppose is a normed space then is an isometry from the normed space onto the normed subspace of .
Let be a subset of and . Put
Let and let be a continuous linear form on . For every , we have on if and only if on ; hence (ii).
Finally we prove (iii). Let be in . An element in has norm if and only if it is the image under of an element of norm in . Hence
Proposition 10
— (i) For a subset of to be equicontinuous, it is necessary and sufficient that it is the image under of an equicontinuous subset of .
(ii) Suppose is closed in . Let be a covering of consisting of bounded subsets and let be the set of subsets of of the form for in . The bijective linear mapping from onto is continuous. It is a homeomorphism if is a directed set for the relation and consists of closed convex and compact sets for .
(iii) If is assumed to be normed, then is an isometry from onto .
The image under of an equicontinuous subset of is an equicontinuous subset of (IV, p. 7, prop. 7). Conversely, let be an equicontinuous subset of . The topology of is defined by the set of restrictions to of the continuous semi-norms on . Hence there exists a continuous semi-norm on such that for all and for all . Let be the set of all linear forms on such that and whose restriction to belongs to . The set is equicontinuous in ; by Hahn-Banach theorem (II, p. 23, cor. 1), we have , hence (i) follows.
We now prove (ii). By prop. 6 of IV, p. 6, the linear mapping from into is continuous, and defines, by passing to the quotient, a continuous linear mapping from onto . Let be the topology on obtained by transferring that of by ; this is finer than the -topology.
Suppose now that is a directed set for and consists of closed, convex, balanced and compact sets for . To show that is a homeomorphism, i.e. that is coarser than the -topology on , it is enough to prove that is compatible with the duality between and and that every equicontinuous set in (considered as the dual of with ) is contained in the homothetic of a set belonging to . Since is finer than the -topology and is a covering of , the linear form on is continuous for for every . Let be a linear form on which is continuous for ; then is a continuous linear form on . The -topology on is coarser than the Mackey topology ; for, the mapping is continuous for the topologies and , and since the latter is Hausdorff, the image under of a set which is compact for is compact for . By lemma 1 of IV, p. 3, there exists such that for all . In particular, for all , and since is closed in , we have (II, p. 45, cor. 2); and finally, for all . This proves that is compatible with the duality between and .
Now let be a subset of equicontinuous for the topology on . By the definition of , and in view of the hypothesis that is directed, this means that there exists a set containing 0 and such that the upper bound of the numbers for and , is finite (III, p. 19, prop. 7). Since is closed in , the theorem of bipolars (II, p. 44, th. 1) shows that we have ; this completes the proof of (ii).
We shall now prove (iii). Let . We shall prove the formula
By prop. 8, (ii) of IV, p. 7, we have , and so , and
By Hahn-Banach theorem (II, p. 23, cor. 3), there exists a linear form on which extends and is of the same norm; hence we get the inequality opposite to (7), since .
Remark
— We know (II, p. 48, prop. 7, (ii)) that is a topological vector space isomorphism from onto (weak duals). For the topology of compact convex convergence, prop. 10 shows that is an isomorphism from onto when is Hausdorff and closed in . For the strong topologies, is a continuous mapping from onto ; it is an isomorphism if is a Banach space \* or if is semi-reflexive and is closed in (IV, p. 15) *, but this is not always so if is a Fréchet space (IV, p. 58, exerc. 5, c)).
Proposition 11
— (i) The weakened topology on is the quotient of that on ; the weakened topology on is induced by that of .
(ii) The Mackey topology on E/M is the quotient of that on E; the Mackey topology on M is finer than the topology induced by τ(E, E').
Assertion (i) follows from prop. 7 of II, p. 48.
The canonical injection is continuous for the weakened topologies, hence for the Mackey topologies and (IV, p. 7, prop. 7). Similarly, the canonical projection is continuous for the Mackey topologies. We see immediately that the quotient topology on obtained from is compatible with the duality between and , hence is coarser than the Mackey topology on , by Mackey's theorem (IV, p. 2, th. 1). This proves (ii).
5. Dual of a direct sum and of a product
For every , let be a pair of vector spaces, set in duality by a bilinear form . We put and , and we identify each with a subspace of . We put and in duality by means of the bilinear form
(the family has finite support).
We recall (II, p. 50, prop. 8) that the weak topology is the product of the weak topologies .
Lemma 3
— (i) *For every , let be a family of subsets of , which is bounded for ; put . Then the -topology on is the product of the -topologies on the .
(ii) For every , let be an adapted bornology on the space endowed with the weak topology , none equal to . Let be the family of subsets of such that for all . Then the -topology on is the direct sum of the -topologies on the .
Let be the product of the -topologies. The sets of the form
where is finite and for all , form a fundamental system of neighbourhoods of 0 for . We have , hence is identical with the -topology. This proves (i).
We assign the -topology to and the -topology to each . For every subset of , we have , hence the injection from into is continuous. Let be a semi-norm on ; we assume that the restriction of to is continuous for all . Then we can find non-empty subsets such that we have
Put ; then . For in , we have
Proposition 12
The topology is the direct sum of the topologies . The topology is the product of the topologies .
We shall apply lemma 3 taking for the family of all subsets of which are bounded for and for the family of all subsets of which are bounded for .
By cor. 2 of III, p. 4, is the family of all subsets of which are bounded for the product topology of the , which is identical with . Hence our assertion on follows.
We endow with the topology which is the direct sum of the topologies . Then the dual of consists of the linear forms where runs through (II, p. 30, prop. 6). By prop. 1 of IV, p. 1, the topologies and have the same bounded sets. Assume first that that the topologies are Hausdorff. By prop. 5 of III, p. 5, these sets are contained in a subset of the form with finite and bounded in (for ) for all . Since is contained in the convex envelope of , where , we can apply lemma 3, to prove the assertion on in this case.
For the general case, let be the intersection of all neighbourhoods of 0 for , and let ; then is the topological direct sum of the (II, p. 31, prop. 8); we deduce from this that every subset of which is bounded for is contained in a set of the form with finite and bounded in for all (III, p. 2, Remark 3); since the polar of this set in is the same as that of , the result follows as above.
Proposition 13
The Mackey topology is the direct sum of the Mackey topologies . The topology is the product of the topologies .
The assertion on follows from lemma 3 (ii) and the following property : for a closed, convex and balanced subset of to be compact for , it is necessary and sufficient that its projection on each is compact for .
To prove the assertion on , assume first that the topologies are Hausdorff, it is enough (lemma 3 (i)) to prove that every subset of which is convex, balanced and compact for is contained in a set of the form where is finite and where is convex, balanced and compact for . But such a subset is bounded for . By the proof of prop. 12, there exists a finite subset of such that , and it is enough to take for the projection of on .
In the general case, with the same notations as in the proof of prop. 12, we have and (IV, p. 2) and since is the topological direct sum of the , we have reduced to the preceding case.
Q.E.D.
For the remainder of this paragraph, we assume that is a family of locally convex spaces. Let denote the topological direct sum of the and , their product. We define a linear mapping , said to be canonical, by
(where denotes the dual of , and that of ).
Proposition 14
(i) *The mapping is an isomorphism from the strong (resp. weak) dual of onto the product of the strong (resp. weak) duals of the :
(ii) For a subset of to be equicontinuous, it is necessary and sufficient that the projection of onto be equicontinuous for all .
(iii) The Mackey topology is the direct sum of the Mackey topologies .
(iv) The topology is the direct sum of the topologies .
That is bijective follows immediately from the definition of a topological direct sum (II, p. 30, prop. 6). Assertion (i) then follows from prop. 12 of IV, p. 12, for the strong topologies, and from prop. 8 of II, p. 50, for the weak topologies. Similarly (iii) follows from prop. 13 (IV, p. 12) and (iv) from prop. 12 (IV, p. 12).
To prove (ii), let be a subset of . Put
let denote the restriction of to , whence
where denotes the projection of on . For to be equicontinuous, it is necessary and sufficient that is finite (that is, that each is finite) and continuous. In view of the characterization of continuous semi-norms on a topological direct sum (II, p. 27, prop. 5), this is the same as assuming that each is continuous, or in fact, that each set is equicontinuous.
Let be the linear mapping, said to be canonical, from into the dual of , defined by the formula
for in and in .
Proposition 15
— (i) *The map is an isomorphism from the topological direct sum of the strong duals of the onto the strong dual of .
(ii) For a subset of to be equicontinuous, it is necessary and sufficient that it is contained in a finite sum , where is finite and where is equicontinuous in for all .
(iii) The Mackey topology is the product of the topologies .
(iv) The topology is the product of the topologies .
It is immediate that is injective. A fundamental system of neighbourhoods of 0 in consists of sets of the form , where is finite and is a neighbourhood of 0 in for in . The polar of in is equal to .
This proves the surjectivity of and also assertion (ii).
Assertions (i) and (iv) follow from prop. 12 (IV, p. 12) and (iii) from prop. 13 (IV, p. 12).
Corollary
— *Every product of barrelled spaces is barrelled.
A locally convex space is barrelled if and only if the initial topology is identical with (IV, p. 4, Remark 3). Hence it is enough to apply prop. 15 (iv).
Exercises
See the exercises for § 1.