Subseries convergence in abstract duality pairs

Let E,F be sets, G an Abelian topological group and b : E×F → G. Then (E,F,G) is called an abstract triple. Let w(F,E) be the weakest toplogy on F such that the maps {b(x, ·) : x ∈ E} from F into G are continuous. A subset B ⊂ F is w(F,E) sequentially conditionally compact if every sequence {yk} ⊂ B has a subsequence {ynk} such that limk b(x, ynk) exists for every x ∈ E. It is shown that if a formal series P xj in E is subseries convergent in the sense that for every subsequence {xnj} there is an element x ∈ E such that P∞ j=1 b(xnj , y) = b(x, y) for every y ∈ F , then the series P∞ j=1 b(xnj , y) converge uniformly for y belonging to w(F,E) sequentially conditionally compact subsets of F . This result is used to establish Orlicz-Pettis Theorems in locall convex and function spaces. Applications are also given to Uniform Boundedness Principles and continuity results for bilinear mappings. 448 Min-Hyung Cho, Li Ronglu and Charles Swartz In this paper we give an abstraction of the classical notion of duality between two vector spaces and prove several versions of the Orlicz-Pettis Theorem for subseries convergent series in the abstract setting. We give a number of applications to classical versions of the Orlicz-Pettis Theorem as well as versions of the theorem in various function spaces and spaces of operators, the Uniform Boundedness Priciple and a version of the MazurOrlicz Theorem on continuity of bilinear operators. The notion of abstract triples was first developed in the mathematics department of New Mexico State University in the period of 1988-1990 when Professor Ronglu Li was a visiting scholar from China. After returning to China, Li invited Min-Hyung Cho to join in the development of the subject. The original notes were never published in full although various results have appeared in publications (see the remarks following Example 8). This paper contains a full development of the original results with some improvements and additional results which were obtained later; see in particular the Uniform Boundedness Principle and the Mazur-Orlicz results in Cases 29, 31 and 32. This paper is dedicated to the memory of Professor Ronglu Li who tragically passed away on Feb. 4, 2014. Abstract Triples. Let E,F be sets and G a Hausdorff, abelian topological group with b : E × F → G; if x ∈ E and y ∈ F , we often write b(x, y) = x · y for convenience. We refer to E,F,G as an abstract duality pair with respect to G or an abstract triple and denote this by (E,F : G). In what follows (E,F : G) will denote an abstract triple. Note that (F,E : G) is an abstract triple under the map b(y, x) = b(x, y). We give examples of abstract triples shortly. Let w(E,F ) be the weakest topology on E such that the family of maps {b(·, y) : y ∈ F} are continuous from E into G. If P gj is a formal series in G, the series is subseries convergent if the series P∞ j=1 gnj converges inG for every subsequence {nj}. If σ is an infinite subset of N , we write P j∈σ gj = P∞ j=1 gnj , where the elements of σ are arranged in a subsequence {nj}; if σ is finite, the meaning of P j∈σ gj is clear.Triples. Let E,F be sets and G a Hausdorff, abelian topological group with b : E × F → G; if x ∈ E and y ∈ F , we often write b(x, y) = x · y for convenience. We refer to E,F,G as an abstract duality pair with respect to G or an abstract triple and denote this by (E,F : G). In what follows (E,F : G) will denote an abstract triple. Note that (F,E : G) is an abstract triple under the map b(y, x) = b(x, y). We give examples of abstract triples shortly. Let w(E,F ) be the weakest topology on E such that the family of maps {b(·, y) : y ∈ F} are continuous from E into G. If P gj is a formal series in G, the series is subseries convergent if the series P∞ j=1 gnj converges inG for every subsequence {nj}. If σ is an infinite subset of N , we write P j∈σ gj = P∞ j=1 gnj , where the elements of σ are arranged in a subsequence {nj}; if σ is finite, the meaning of P j∈σ gj is clear. Definition 1. A sequence {xj} ⊂ E or a (formal) series P xj is w(E,F ) Subseries convergence in abstract duality pairs 449 subseries convergent if for every σ ⊂ N, there exists xσ ∈ E such that X

In this paper we give an abstraction of the classical notion of duality between two vector spaces and prove several versions of the Orlicz-Pettis Theorem for subseries convergent series in the abstract setting.We give a number of applications to classical versions of the Orlicz-Pettis Theorem as well as versions of the theorem in various function spaces and spaces of operators, the Uniform Boundedness Priciple and a version of the Mazur-Orlicz Theorem on continuity of bilinear operators.
The notion of abstract triples was first developed in the mathematics department of New Mexico State University in the period of [1988][1989][1990] when Professor Ronglu Li was a visiting scholar from China.After returning to China, Li invited Min-Hyung Cho to join in the development of the subject.The original notes were never published in full although various results have appeared in publications (see the remarks following Example 8).This paper contains a full development of the original results with some improvements and additional results which were obtained later; see in particular the Uniform Boundedness Principle and the 31 and 32.This paper is dedicated to the memory of Professor Ronglu Li who tragically passed away on Feb. 4, 2014.

Abstract Triples.
Let E, F be sets and G a Hausdorff, abelian topological group with b : E × F → G; if x ∈ E and y ∈ F , we often write b(x, y) = x • y for convenience.We refer to E, F, G as an abstract duality pair with respect to G or an abstract triple and denote this by (E, F : G).In what follows (E, F : G) will denote an abstract triple.Note that (F, E : G) is an abstract triple under the map b(y, x) = b(x, y).
We give examples of abstract triples shortly.Let w(E, F ) be the weakest topology on E such that the family of maps {b(•, y) : y ∈ F } are continuous from E into G.If P g j is a formal series in G, the series is subseries convergent if the series P ∞ j=1 g n j converges in G for every subsequence {n j }.If σ is an infinite subset of N , we write P j∈σ g j = P ∞ j=1 g n j , where the elements of σ are arranged in a subsequence {n j }; if σ is finite, the meaning of P j∈σ g j is clear.
Definition 1.A sequence {x j } ⊂ E or a (formal) series P x j is w(E, F ) subseries convergent if for every σ ⊂ N, there exists x σ ∈ E such that X j∈σ x j • y = x σ • y for every y ∈ F .We symbolically write P j∈σ x j = x σ and say that the series P x j is w(E, F ) subseries convergent.
Note that we do not assume any algebraic structure on E, the algebraic operations are transferred to G via the map b : E × F → G; of course, if the set E has sums defined on it the meaning of P j∈σ x j is clear.We give some examples which will be employed later.
Example 2. Of course, the simplest example of an abstract triple is a pair of vector spaces E, F in duality where G is just the scalar field and the topology w(E, F ) is just the weak topology σ(E, F ) from the duality.In this case, if P x j is w(E, F ) subseries convergent, P j∈σ x j is the usual weak sum.
Example 3. Let Σ be a σ-algebra of subsets of a set S and M be a family of countably additive G valued measures on Σ. Define b : Example 4. Let E, F be abelian groups such that there exists a biadditive map b : E × F → G. Then (E, F : G) is an abstract triple; this abstract setting was utilized in [Sw2].
Example 5. Let E be a topological space and C(E, G) be the space of all continuous maps from Example 6.Let E, G be topological vector spaces and L(E, G) the space of all continuous linear operators from E into G, then (L(E, G), E : G) forms an abstract triple under the map b(T, x) = T (x); in this case the topology w(L(E, G), E) is just the strong operator topology.Also, (E, L(E, G) : G) forms an abstract triple under the map b(x, T ) = T (x).
Example 7. Let X be a Hausdorff topological vector space and E be a vector space of X valued sequences which contains the subspace c 00 (X) of all X valued sequences which are eventually 0. Assume that E has a vector topology under which it is an AK space,i.e., the coordinate projection P k which sends each sequence x = (x 1 , x 2 , ...) in E into the sequence with x k in the k th coordinate and 0 in the other coordinates is continuous and if This situation covers the case of the sequence spaces l p (X) and c 0 (X) when X is a locally convex space.
Example 8. Let X be a Hausdorff topological vector space and E be a vector space of X valued sequences which contains the subspace c 00 (X) of all X valued sequences which are eventually 0. Let Y be a topological vector space.The β-dual of E with respect to Y is defined to be Other examples will be given in the application section.There have been several similar abstractions of the duality between vector spaces which have been used to treat versions of the Orlicz-Pettis Theorem.For example, Blasco, Calabuig and Signes ( [BCS]) have considered a bilinear map b : E × F → G where E, F, G are Banach spaces and b is a bilinear operator satisfying continuity conditions.They establish a general version of the theorem for subseries convergent series and apply it to vector integration.There is a more general version given in [Sw7] where E is a vector space, F, G are locally convex spaces and b is a bilinear map and multiplier convergent series are considered.Applications to multiplier convergent series in spaces of operators are given.Another generalization is given by Chen and Li ([CL]) where E, F are vector spaces, G is a locally convex space and b is what they call a bi-quasi-homogeneous operator.They consider multiplier convergent series of quasi-homogeneous operators.Li and Wang ([LW]) have considered the case when E is a set and F is a set of G valued functions.They consider operator valued multiplier convergent series where the space of multipliers is vector valued.The case where E, F are vector spaces, G is a locally convex space and b is a bilinear map is considered in [LS3]; general versions of the Orlicz-Pettis Theorem are established and numerous applications are given.A similiar treatment is given in Chapter 4 of [Sw6], pages 73-82.Li and Cho ([LC]) have used the general abstract setting above to obtain a generalization of an Orlicz-Pettis result of Kalton; we will consider this result later.Zheng, Cui and Li ([ZCL]) have also considered abstract duality pairs in spaces with sectional operators and indicated applications to sequence spaces.By considering subseries convergent series only, we are able to treat the case of group valued series in our setup.
We now establish several versions of the Orlicz-Pettis Theorem for abstract triples.The classical version of the Orlicz-Pettis Theorem for normed spaces asserts that a series in a normed space which is subseries convergent in the weak topology of the space is subseries convergent in the norm topology ([Or],[Pe]).The theorem has been extended to locally convex spaces and many other situations including topological groups.See [K1], [FL] for a discussion of the history of the subject.We refer to any result which asserts that a series which is subseries convergent in some weak topology is subseries convergent in a stronger topology as an Orlicz-Pettis Theorem.
Let w(F, E) be the weakest topology on F such that all of the maps {b(x, •) : x ∈ E} from F into G are continuous.A subset B ⊂ F is sequentially conditionally w(F, E) compact if every sequence {y k } ⊂ B has a subsequence {y n k } such that the sequence {x • y n k } = {b(x, y n k )} converges in G for every x ∈ E (this is terminology of Dinculeanu ([Di]).
The method of proof used in treating our versions of the Orlicz-Pettis Theorem relies on the Antosik-Mikusinski Matrix Theorem which we now state for convenience.
Theorem 9. (Antosik-Mikusinski) Let G be an abelian topological group and x ij ∈ G for i, j ∈ N. Suppose (I) lim i x ij = x j exists for each j and (II) for each increasing sequence of positive integers {m j } there is a subsequence {n j } of {m j } such that { P ∞ j=1 x in j } is Cauchy.Then lim i x ij = x j uniformly for j ∈ N. In particular, which satisfies conditions (I) and (II) of Theorem 9 is referred to as a K matrix.
Theorem 10.If the series P x j is w(E, F ) subseries convergent, then the series P j∈σ x j • y converge uniformly for y ∈ B and σ ⊂ N , where B is any sequentially conditionally w(F, E) compact subset B ⊂ F [ that is, for every closed neighborhood of 0, U , in G there exists N such that P j∈σ x j • y ∈ U whenever y ∈ B and min σ > N; a strong form of unordered convergence for the series].
Proof.If the conclusion fails to hold, there exists a closed neighborhood U of 0 such that for every k there exist σ k with min σ k > k and y k ∈ B such that Apply the condition above to k 2 to obtain There exists a subsequence {y n k } such that lim x•y n k exists for every x ∈ E.
Consider the matrix We claim that M is a K matrix.The columns of M converge and for every subsequence {r j } the subseries Therefore, M is a K matrix whose diagonal converges to 0 by the Antosik-Mikusinski Theorem.But, this contradicts (&). 2 A subset B ⊂ F is sequentially relatively w(F, E) compact if every sequence {y k } ⊂ B has a subsequence {y n k } and there exists y ∈ F such that lim k x • y n k = x • y for every x ∈ E. A sequentially relatively w(F, E) compact is obviously sequentially conditionally w(F, E) compact so the result above holds for this family of subsets of F .
The unordered convergence form of the conclusion of Theorem 10 is useful in treating the Hahn-Schur Theorem given in Case 18.
We consider the theorem for w(F, E) compact subsets.
Theorem 11.Let G be metrizable under the metric ρ.If

P
x j is w(E, F ) subseries convergent, then the series B by • y for all j.Let S be the set of partial sums of the series sequentially compact since this topology is metrizable so it follows from the previous Orlicz-Pettis Theorem that the series P x j converges uniformly on b B and, therefore, on B. 2 We will discuss applications of Theorems 10 and 11 to locally convex spaces in Case 16 below.
A result of Kalton ([K2]) asserts that if τ is a separable polar topology on E from the dual pair E, F , then any series P x j in E which is σ(E, F ) subseries convergent is τ subseries convergent.If τ is the polar topology of uniform convergence on the family A of σ(F, E) bounded subsets of F , then for every x ∈ E and A ∈ A the set {x • y : y ∈ A} is sequentially relatively compact in the scalar field.We give an abstraction of this condition and use it to give a generalization of Kalton's result to abstract triples.
Note that if B ⊂ F is sequentially conditionally w(F, E) compact, then B is sequentially conditionally compact at each x ∈ E. Under a separability assumption, the converse holds.
Theorem 13.Let G be sequentially complete.Let F be a family of subsets of F such that each member of F is sequentially conditionally compact at each x ∈ E and let τ be the topology on E of uniform convergence on the members of F. If (E, τ ) is separable, then each member of F is sequentially conditionally w(F, E) compact.

Proof.
Let D = {d k : k ∈ N} be τ dense in E. Let B ∈ F and {y k } ⊂ B. Since B is sequentially conditionally compact at each x ∈ E, by the diagonalization procedure {y k } has a subsequence {y n k } such that the sequence {d • Corollary 14.If the conditions of Theorem 13 hold and Proof.The result is immediate from Theorems 13 and 10. 2 The separability assumption in Theorem 13 is important.
Example 15.The series

Applications
We begin with the Orlicz-Pettis Theorem for locally convex spaces.
Case 16.Let E be a Hausdorff locally convex space with dual E 0 .Suppose

P
x j is subseries convergent with respect to σ(E, E 0 ).We may assume that E is separable by replacing E with the span of {x j }, if necessary.Let γ(E, E 0 ) (λ(E, E 0 ); τ (E, E 0 )) be the topology of uniform convergence on the sequentially conditionally σ(E 0 , E) compact subsets of E 0 (σ(E 0 , E) compact subsets; convex σ(E 0 , E) compact subsets).It follows from Theorem 10 that P x j is γ(E, E 0 ) subseries convergent.Also, when E is separable the σ(E 0 , E) topology on any σ(E 0 , E) compact subset is metrizable ([Wi]9.5.3) so any σ(E 0 , E) compact subset is sequentially σ(E 0 , E) compact and it follows from Theorem 10 that P x j is subseries convergent with respect to λ(E, E 0 ) and, therefore, subseries convergent with respect to τ (E, E 0 ), the Mackey topology.
The locally convex version of the Orlicz-Pettis Theorem for the Mackey topology is due to McArthur ([Mc]); the version for the topology λ(E, E 0 ) is due to Bennett and Kalton ([BK]) and the version for γ(E, E 0 ) is due to Dierolf ([Die]).Kalton's version of the Orlicz-Pettis Theorem will be considered in Case 25.
Next, we consider a theorem of Nikodym for group valued measures.
Case 17. (Nikodym Convergence Theorem) Let Σ be a σ algebra of subsets of a set S and let µ n : Σ → G be a countably additive measure for n ∈ N and set M = {µ n }.Suppose lim n µ n (A) = µ(A) exists for every A ∈ Σ.The Nikodym Theorem asserts that (i) µ is countable additive and (ii) {µ n } is uniformly countably additive.We observe that (ii) follows from Example 3 and Theorem 10.Let {A j } be a pairwise disjoint sequence from Σ. Then the series P A j is w(Σ, M) subseries convergent by the countable additivity of the µ n .Also, the sequence {µ n } is sequentially conditionally w(Σ, {µ n }) compact by hypothesis.Hence, from Theorem 10 it follows that the series P ∞ j=1 µ n (A j ) converge uniformly for n ∈ N. But, this is just (ii).Condition (i) follows from (ii).
For the case of the Nikodym Theorem for groups see [AS1].Using a result of Drewnowski, the Nikodym Convergence Theorem can be extened to strongly additive set functions.An additive set function µ : Σ → G is strongly additive if µ(A j ) → 0 for every disjoint sequence {A j } ⊂ Σ; see [Sw5] 2.3.5 for details.
We note in passing that a version of the Nikodym Boundedness Theorem can be obtained from the Nikodym Convergence Theorem.Suppose M is a family of countably additive scalar valued set functions defined on Σ which is pointwise bounded on Σ.The Nikodym Boundedness Theorem asserts that M is uniformly bounded on Σ ( [DU]).To show that M is uniformly bounded on Σ it suffices to show that {µ n (A n )} is bounded for every {µ n } ⊂ M and every pairwise disjoint sequence {A n } ⊂ Σ (see [Sw5]4.7.1; we develop a more general result later in Theorem 34).
n µ n (A) → 0 for every A ∈ Σ so the Nikodym Convergence Theorem implies the series n µ n (A j ) converge uniformly for n ∈ N. In particular, 1 n µ n (A n ) → 0 as desired.The version of the theorem for locally convex valued measures follows from the scalar version and the Uniform Boundedness Theorem.We consider a more general version of the theorem in Corollary 36.
We next consider a version of the Hahn-Schur Theorem for group valued series (see [AS1]; [Sw6]8.1).
Case 18. (Hahn-Schur Theorem) Let P j x ij be a subseries convergent series in G for every i ∈ N and suppose lim i P j∈σ x ij exists for every σ ⊂ N. Set x j = lim i x ij .The subseries version of the Hahn-Schur Theorem then assets that (i) P x j is subseries convergent, (ii) the series P j∈σ x ij converge uniformly for i ∈ N, σ ⊂ N, and (iii) lim i P j∈σ x ij = P j∈σ x j uniformly for σ ⊂ N. We show that (ii) follows directly from Theorem 10.Let E be the power set of N , define f i : E → G by f i (σ) = P j∈σ x ij and set F = {f i : i ∈ N}.Then (E, F : G) is an abstract triple and the (formal) series P j j is w(E, F ) subseries convergent with P j f i (j) = P j x ij .By hypothesis, F is sequentially conditionally w(F, E) compact so from Theorem 10 , it follows that (ii) holds.Conditions (i) and (iii) follow immediately from (ii).
The usual scalar version of the theorem can be obtained easily from Case 18 (see [AS1]8.2; [Sw5]8.1).
We now consider subseries convergence in the space of continuous linear operators.
Case 19.Let E, G be topological vector spaces with G metrizable and consider the abstract triple (L(E, G), E : G) as in Example 6. Suppose that P T j is a series in L(E, G) which is subseries convergent in the strong operator topology.Note that any subset B ⊂ E which is compact in E is w(E, L(E, G)) compact so the series P T j is subseries convergent in L c (E, G), the topology of uniform convergence on compact subsets of E, by Theorem 11.
We consider subseries convergence with respect to pointwise convergence in spaces of continuous functions.
Case 20.Let Ω be a sequentially compact topological space and let SC(Ω, G) be the space of all sequentially continuous functions from Ω into G.Suppose P f j is a series in SC(Ω, G) which is subseries convergent in the topology of pointwise convergence on G.We claim the series P f j (t) is subseries convergent uniformly for t ∈ Ω.To see this consider the abstract triple (SC(Ω, G), Ω : G) under the map (f, t) = f (t).The series P f j is subseries convergent with respect to w(SC(Ω, G), Ω) and the set Ω is w(Ω, SC(Ω, G)) sequentially compact since Ω is sequentially compact so the claim follows from Theorem 10.
Case 21.Let Ω be a topological space, G be metrizable and C(Ω, G) the space of continuous functions from Ω to G. Suppose P f j is a series in C(Ω, G) which is subseries convergent in the topology of pointwise convergence on Ω.We claim that the series is subseries convergent in the topology of uniform convergence on compact subsets of Ω.To see this consider the abstract triple (C(Ω, G), Ω : G) under the map (f, t) → f (t).The series G), Ω) subseries convergent and any compact subset of Ω is w(Ω, C(Ω, G)) compact so the claim follows from Theorem 11.
Theorems of this type relative to pointwise convergent series in spaces of continuous functions were established in [Th] and [Sw2].
Case 22.Let X be a topological vector space and assume that E is a vector space of X valued sequences which contains the space c 00 (X) of all X valued sequences which are eventually 0. If z ∈ X and k ∈ N, let e k ⊗ z be the sequence with z in the k th coordinate and 0 in the other coordinates.
Assume that E has a vector topology under which the coordinate mappings {x k } → e k ⊗ x k form E into E are continuous (i.e., E is a K-space) and for every x = {x k } ∈ E we have x = P ∞ k=1 e k ⊗ x k (i.e., E is an AKspace).Suppose that P x j is a series in E which is subseries convergent with respect to the topology of coordinatewise convergence.We claim that the series P x j is subseries convergent in the original topology of E. To see this consider the following abstract triple.Define P k : E → E by P k x = P k i=1 e i ⊗x i so P k x → x in E for every x ∈ E by the AK assumption.Set F = {P k : k ∈ N} and note that (E, F : E) is an abstract triple under the map (x, P k ) → P k x and that the series P x j is subseries convergent with respect to w(E, F ).The set F is sequentially conditionally w(F, E) compact by the AK hypothesis so it follows from Theorem 10 that the series P ∞ j=1 P k x j is subseries convergent uniformly for k ∈ N. To establish the claim let U be a closed neighborhood of o in E. Set s n = P n j=1 x j and s = P ∞ j=1 x j , where this is the coordinate sum of the series.Since lim n P k s n = P k s uniformly for k ∈ N, there exists N such that P k s n −P k s ∈ U for n ≥ N, k ∈ N. Fixing n and letting k → ∞ gives s n − s ∈ U for n ≥ N .Since the same argument can be applied to any subsequence, the claim follows.This result applies to such sequence spaces as l p (X) ,1 < p < ∞, and c 0 (X).
We next consider a result of Stiles for spaces with a Schauder basis.
Case 23. (Stiles) Let E be a topological vector space with a Schauder basis {b k } and coordinate functionals {f k } .For each k let P k : E → E be the projection x j is subseries convergent with respect to σ(E, {f j }), we claim that the series is subseries convergent in the original topology of E. To see this, set F = {P j : j ∈ N} and consider the abstract triple (E, F : E) under the map (x, P j ) → P j x.Then P x j is w(E, F ) subseries convergent and F is sequentially conditionally w(F, E) compact since P k x → x so the series P ∞ j=1 P k x j is subseries convergent uniformly for k ∈ N by Theorem 10.As in Case 22 this establishes the claim.
The result in Case 23 can be generalized somewhat.Assume that E is a topological vector space and there exist a sequence of linear operators {P k } such that for each x ∈ E we have x = P ∞ k=1 P k x with convergence in E. When the {P k } are continuous, then {P k } is called a Schauder decomposition ( [LT]).Then the proof in Case 22 shows that if a series is subseries convergent in w(E, {P k }), then the series is subseries convergent in the topology of E.
We next consider a result of Tweddle.
Case 24.(Tweddle) Let E, F be a pair of vector spaces in duality.Let E be the family of all σ(E, F ) subseries convergent series in E and let E # be all linear functionals x 0 on E such that for all {x j } ∈ E, where P ∞ j=1 x j is the σ(E, F ) sum of the series.Then E, E # form a dual pair and each {x j } ∈ E is σ(E, E # ) subseries convergent.It follows from Case 16 that every {x j } ∈ E is subseries convergent in the Mackey topology τ (E, E # ) of uniform convergence on convex σ(E, E # ) compact subsets of E # .This is the Tweddle topology of E and Tweddle has shown that this is the strongest locally convex topology on E which has the same σ(E, F ) subseries convergent series ( [Tw]).
The topology of Tweddle can also be extended to our abstract setting.Let (E, F : G) be an abstract triple and let E be the family of all w(E, F ) subseries convergent series.Let E # be all functions f : E → G such that f ( We indicate applications of Theorem 13.First, we consider a result of Kalton ([K2]).
Case 25. (Kalton) Let E, F be a dual pair of vector spaces and τ a polar topology on E from this duality which is separable.If τ is the polar topology of uniform convergence on the family A of σ(F, E) bounded subsets of F , then every subset A of A is sequentially conditionally compact at each x ∈ E so by Theorem 13 any series in E which is σ(E, F ) subseries convergent is τ subseries convergent.
Next, we consider the space of compact operators.
Case 26.Let E, G be Hausdorff topological vector spaces and K(E, G) the space of all continuous linear operators which carry bounded subsets of E into sequentially conditionally compact subsets of G.If E, G are Banach spaces K(E, G) is the space of compact operators.Then (K(E, G), E : G) is an abstract triple and if B is the family of bounded subsets of E each B ∈ B is sequentially conditionally compact at each T ∈ K(E, G).The topology w(K(E, G), E) is just the strong operator topology.Let K b (E, G) be the topology of uniform convergence on the members of B; if E, G are normed spaces this is just the uniform operator topology of K(E, G).From Theorem 13 it follows that if F(E, G) is any separable subspace of K b (E, G), then any series in F(E, G) which is subseries convergent in the strong operator topology is subseries convergent in K b (E, G).
We indicate situations where Case 26 is applicable.Let E, G be Banach spaces and let F(E, G) be the space of operators with finite dimensional range.Thus, every operator LT]1.e.4 or 1.e.5) so Case 26 also applies in this case to K b (E, G).
Another situation where Case 26 applies is given as follows.Let E be a metrizable nuclear space and G be separable.Then any continuous linear operator from Also, if E is dual nuclear (i.e., the strong dual of E is nuclear) and G is nuclear, then L b (E, G) is nuclear ( [Pi] 5.5.1) and , therefore, separable so Case 26 applies.
Finally, If E, G are normed spaces and E 0 , G are separable, then F(E, G) is separable under the nuclear norm ν on F(E, G) (see [Pi]3.1) so the space of nuclear operators N (E, G) is separable under the nuclear norm ν ( [Pi]3.1.4)and Case 26 applies.Similar remarks apply to the space of Hilbert-Schmidt (absolutely summing) operators on Hilbert spaces ( [Pi]2.5).
Some of the results of this case were announced without proofs in [LC].
We consider another result related to compact operators and a family of operators introduced by A. Mohsen.Let X, Y, Z be normed spaces and let W * (Y 0 , Z) be the space of all sequentially weak*-k•k continuous linear operators from Y 0 into Z (these operators were introduced and studied by Mohsen ([Mo]) and were shown to be bounded).
Case 27.Let B(Z) denote the unit ball of any normed space Z. Assume B(Y 0 ) is weak* sequentially compact and that the series P U j is subseries convergent in the strong operator topology of W * (Y 0 , Z).Consider the abstract triple (W * (Y 0 , Z), B(Y 0 ) : Z) under the map (U, y 0 ) → Uy 0 .Then the series P U j is w(W * (Y 0 , Z), B(Y 0 )) subseries convergent and B(Y 0 ) is w(B(Y 0 ), W * (Y 0 , Z)) sequentially compact since if {y 0 j } ⊂ B(Y 0 ), then there is a subsequence {y 0 n j } which is weak* convergent to some y 0 ∈ B(Y 0 ) and °°°U y 0 n j − Uy 0 °°°→ 0 by the definition of W * (Y 0 , Z).Hence, by Theorem 10 the series P ∞ j=1 U j y 0 converge uniformly for y 0 ∈ B(Y 0 ) and similarly for any subseries.That is, the series P U j is subseries convergent in norm.
As a special case of Case 27, we can obtain a result of Kalton ([K3]).We say a normed space Z has the Diestel-Faires property (DF property) if any weak* subseries convergent series P z j in Z 0 is norm subseries convergent ( [DF]; Diestel and Faires have characterized the Banach spaces with DF as the spaces Z whose dual does not contain a copy of l ∞ ).
Case 28.(Kalton) Let P T j be subseries convergent in the weak operator topology of K(X, Y ) and assume X has the DF property.Since each T j has separable range we may assume that Y is separable by replacing Y with the union of the ranges of the T j , if necessary.For each z 0 ∈ Y 0 the series P T 0 j z 0 is weak* subseries convergent in X 0 and by the DF property is norm subseries convergent.Let K 0 (X, Y ) be {T 0 ∈ K(Y 0 , X 0 ) : T ∈ K(X, Y )} and consider the abstract triple (K 0 (X, Y ), B(Y 0 ); X 0 ).The series P T 0 j is w(K 0 (X, Y ), B(Y 0 )) subseries convergent by the observation above and the ball of Y 0 is weak* sequentially compact since Y is separable.Also, K 0 (X, Y ) ⊂ W * (Y 0 , X 0 ) ([DS]VI.5.6) so Case 27 implies that the series P T 0 j is norm subseries convergent and, hence, the series We show that a version of the Uniform Boundedness Principle can be obtained from the Orlicz-Pettis Theorem.
Case 29.(Uniform Boundedness Principle) Let G be a locally convex space and let E be a sequentially complete locally convex space.Let Γ be a subset of L(E, G) which is pointwise bounded on E. We claim that Γ is uniformly bounded on bounded subsets of E; a version of the Uniform Boundedness Principle (see [Sw5] 4.2 for this version; if E is metrizable, Γ is equicontinuous which is a more familiar conclusion of the Uniform Boundedness Principle).Suppose there exists a bounded subset B of E such that Γ(B) is not bounded.Then there exists a continuous seminorm p on G such that Since {x k } is bounded, the series P k x k /2 k is absolutely convergent and, therefore, subseries convergent in E by the sequential completeness hypothesis.Set F = {T k /2 k : k ∈ N} and consider the abstract triple (E, F : G) under the map (x, T ) → T x.Since the series P k x k /2 k is subseries convergent in E, the series is w(E, F ) subseries convergent.For each x ∈ E the sequence {T k x} is bounded by hypothesis so T k x/2 k → 0 which implies that that the sequence {T k /2 k } is sequentially relativelyw(F, E) compact; i.e., F is sequentially relatively w(F, E) compact.The Orlicz-Pettis Theorem 10 implies that the series This implies a result similar to that of Mazur and Orlicz.Namely, if E is braked and sequentially complete, then b is jointly sequentially continuous.
We finally consider a uniform boundedness result for bilinear mappings. Case The series P k (x k , y k )/2 k is absolutely convergent in E × F and, therefore, subseries convergent by the sequential completeness assumption.By the remarks following Case 31, each b k is jointly sequentially continuous so the series is w(E × F, {b k }) subseries convergent.Also, by the pointwise boundedness assumption the sequence b k /2 k → 0 in w({b k }, E × F ) so the sequence is relatively w({b k }, E × F ) compact.By Theorem 10 the series P ∞ j=1 b k (x j /2 j , y j )/2 k converge uniformly for k ∈ N. In particular, b k (x k , y k )/2 2k → 0 contradicting (*).This result also implies that Γ is left sequentially equihypocontinuous with respect to the bounded subsets of F .For suppose x k → 0 in E and B ⊂ F is bounded in F and In particular, this implies that Γ is sequentially equicontinuous.
A similar uniform boundedness result for bilinear maps is given in 6.3.1 of [Sw5].

Partial Sums
of integers and set τ = ∪ ∞ j=1 B k j .Since the {B j } are pairwise disjoint and finite and P x j is w(E, F ) subseries convergent, we have Hence, M is a K matrix and by the Antosik-Mikusinski Matrix Theorem the diagonal of M converges to 0. But this contradicts (#). 2 We will make a remark on the semiconvexity assumption following Corollary 37.
Corollary 36.Let E be a semiconvex space with a nontrivial dual E 0 .If P x j is σ(E, E 0 ) subseries convergent, then the set of partial sums, S, of P x j is β(E, E 0 ) bounded.In particular, this holds for locally convex spaces.
Example 15 shows that a series may be weak subseries convergent but fail to be subseries convergent in the strong topology while Corollary 36 shows that the partial sums are always bounded in the strong topology.
We now observe that the Nikodym Boundedness Theorem follows from Theorem 35.
Corollary 37. (Nikodym) Let Σ be a σ algebra of subsets of a set S, G be a semiconvex space.If M is a family of countably additive G valued set functions defined on Σ which is pointwise bounded on Σ, then M is uniformly bounded on Σ, i.e., {µ(A) : µ ∈ M, A ∈ Σ} is bounded.

Proof.
The proof of Theorem 35 shows that if {µ(A) : µ ∈ M, A ∈ Σ} is not bounded, there exist {µ k } ⊂ M and a pairwise disjoint sequence {B k } ⊂ Σ such that {µ k (B k )} is not bounded [this is a generalization of a well known result of Drewnowski ([Dr])].In the notation of Example 3 the series P B k is w(Σ, M) subseries convergent and M is pointwise bounded on S = {∪ j∈σ B j : σ ⊂ N} so M is uniformly bounded on S by Theorem 35.In particular, {µ k (B k )} is bounded. 2 Corollary 37 is applicable to the case when M is single measure so a countably additive set function with values in a semiconvex space has bounded range; this gives a generalization of 3.6.3 of [Rol].Turpin has given an example of a countably additive set function defined on a σ algebra with values in a nonlocally convex space which has unbounded range so the semiconvex assumption cannot be dropped ( [Rol]3.6.4).What conditions on the space which are necessary and sufficient for a vector measure to have bounded range seem to be unknown.
The version of the Nikodym Boundedness Theorem for semiconvex spaces is due to Contantinescu ([Co]) and Weber ([We]).

P
j∈σ x j • y converge uniformly for y ∈ B and σ ⊂ N , where B is any w(F, E) compact subset B ⊂ F .Proof.Let B be w(F, E) compact.Define an equivalence relation on B by y ∼ z iff x j • y = x j • z for every j.Let b B be the collection of equivalence classes and b y the equivalence class to which y belongs.Define a metric d on b since bounded subsets of E are relatively compact.Now E is separable ([G]II.VI.5) and E 0 b is separable ([G]II.VI.12) so L b (E, G) is separable ([G]III.II.11.b and 13.c) and Case 26 applies.
t k → ∞ with t k x k → 0. Let p be a continuous seminorm on G and set M = sup{p(b(t k x k , y) : k ∈ N, y ∈ B, b ∈ Γ}; M < ∞ by the result above.Thenp(b(x k , y)) = 1 t k p(b(t k x k , y)) ≤ M/t k → 0uniformly for b ∈ Γ.
32.Let E, F, G be locally convex spaces with E, F sequentially complete and E braked.Let Γ be a family of separately continuous bilinear mappings from E × F into G which is pointwise bounded on E × F .If A ⊂ E, B ⊂ F are bounded, we claim that Γ(A, B) is bounded, i.e., Γ is uniformly bounded on bounded subsets of E × F .If the conclusion fails to hold, there exist a continuous seminorm p on G,x k