<<

Positivity (2021) 25:801–818 https://doi.org/10.1007/s11117-020-00787-1 Positivity

Nuclear operators on Banach function spaces

Marian Nowak1

Received: 13 March 2020 / Accepted: 11 September 2020 / Published online: 6 October 2020 © The Author(s) 2020

Abstract Let X be a and E be a perfect Banach function space over a finite measure space (,,λ)such that L∞ ⊂ E ⊂ L1.LetE denote the Köthe dual of E and τ(E, E) stand for the natural on E. It is shown that every nuclear operator T : E → X between the locally convex space (E,τ(E, E)) and a Banach space X is Bochner representable. In particular, we obtain that a linear operator T : L∞ → X between the locally convex space (L∞,τ(L∞, L1)) and a Banach space X is nuclear if and only if its representing measure mT :  → X has the Radon-Nikodym property and |mT |() =T nuc (= the nuclear of T ). As an application, it is shown that some natural kernel operators on L∞ are nuclear. Moreover, it is shown that every nuclear operator T : L∞ → X admits a factorization ϕ ϕ through some L , that is, T = S ◦ i∞, where S : L → X is a Bochner ∞ ϕ representable and and i∞ : L → L is the inclusion map.

Keywords Banach function spaces · Mackey topologies · Mixed topologies · Vector measures · Nuclear operators · Bochner representable operators · Kernel operators · Radon–Nikodym property · Orlicz spaces · Orlicz-Bochner spaces

Mathematics Subject Classification 47B38 · 47B10 · 46E30

1 Introduction and preliminaries

We assume that (X, ·X ) is a real Banach space. For terminology concerning Riesz spaces and function spaces, we refer the reader to [9,13,27]. We assume that (,,λ)is a finite measure space. Let L0 denote the correspond- ing space of λ-equivalence classes of all -measurable real functions on . Then L0 is a super Dedekind complete Riesz space, equipped with the topology To of con-

B Marian Nowak [email protected]

1 Institute of Mathematics, University of Zielona Góra, ul. Szafrana 4A, 65–516 Zielona Góra, Poland 123 802 M. Nowak vergence in measure. By S() we denote the space of all real -simple functions = n 1 ∈  s i=1 ci Ai , where the sets Ai are pairwise disjoint. 0 Let (E, ·E ) be a Banach function space, where E is an order ideal of L such ∞ 1 that L ⊂ E ⊂ L , and ·E is a Riesz norm on E.ByTE we denote the ·E -norm topology on E.ByE we denote the Köthe dual of E, that is,     E := v ∈ L0 : |u(ω) v(ω)| dλ<∞ for all u ∈ E . 

  The associated norm ·E on E is defined for v ∈ E by  

vE = sup |u(ω) v(ω)| dλ : u ∈ E, uE ≤ 1 . 

 We will assume that E is perfect, that is, E = E and uE =uE . The order ∼ ∼  continuous dual En of E separates the points of E and En can be identified with E  v → ∈ ∼ through the Riesz isomorphism E Fv En , where 

Fv(u) = u(ω)v(ω) dλ for u ∈ E and Fv=vE 

(see [13, Theorem 6.1.1]). The Mackey topology τ(E, E) is a locally convex- solid Hausdorff topology with the Lebesgue property (see [9, Corollary 82H]). Then   τ(E, E ) ⊂ TE and τ(E, E ) = TE if the norm ·E is order continuous. The most important classes of Banach function spaces are Lebesgue spaces L p (1 ≤ p ≤∞) and Orlicz spaces Lϕ (see [19]).  Now we present a characterization of (τ(E, E ), ·X )-continuous linear operators T : E → X (see [17, Proposition 2.2]).

Proposition 1.1 For a bounded linear operator T : E → X the following statements are equivalent:  (i) Tis(τ(E, E ), ·X )-continuous. (ii) T (un)X → 0 if un(ω) → 0 λ-a.e. and |un(ω)|≤|u(ω)| λ-a.e. for some u ∈ E and all n ∈ N. ∈  ( 1 ) → λ( ) → (iii) For each u E, T u An X 0 whenever An 0.

For terminology and basic facts concerning , we refer the reader to [4,6,7,22]. For a finitely additive measure m :  → X,by|m|(A) we denote the variation of m on A ∈ . A measure m :  → X is said to be λ-continuous if m(An)X → 0 whenever λ(An) → 0. Let L1(X) denote the Banach space of λ-equivalence classes of all X-valued Bochner integrable functions g defined on , equipped with the norm g1 :=  g(ω)X dλ. Recall that a λ-continuous measure m :  → X of finite variation is said to have the Radon-Nikodym property with respect to λ if there exists a function g ∈ L1(X) 123 Nuclear operators on Banach function spaces 803  ( ) = (ω) λ ∈  = λ such that m A A g d for all A . Then we write m g and a function g is called the density of m with respect to λ. Assume that m :  → X is a λ-continuous measure. Following [6, §13] for A ∈ , we put  n |m|E (A) := sup |ci |m(Ai )X , i=1  = n 1 ∈ S() where the supremum is taken for all functions s i=1 ci Ai such that Ai ⊂ A for 1 ≤ i ≤ n and sE ≤ 1. The set function |m|E will be called a E-variation of the measure m. = ∞ | | ( ) =| |( ) ∈  If, in particular, E L , then m L1 A m A for A . Let L0(X) stand for the linear space of λ-equivalence classes of all strongly - measurable functions g :  → X.Let

0 E(X) = g ∈ L (X) :g(·)X ∈ E .

Then E(X) equipped with the norm gE(X) :=  · g(·)X E is a Banach space, called a Köthe- (see [14]).

Definition 1.1 A bounded linear operator T : E → X is said to be Bochner repre- sentable, if there exists g ∈ E(X) such that  T (u) = u(ω)g(ω) dλ for u ∈ E. 

The concept of nuclear operators between Banach spaces in due to Ruston [21]. Grothendieck carried over the concept of nuclear operators to locally convex spaces [10,11](seealso[26, p. 289], [18], [23, Chap. 3, §7], [4, Chap. 6], [5,22]). Following [23, Chap. 3, §7] (see also [2, Chap. 4], [12, 17.3, p. 379]), we have

Definition 1.2 A linear operator T : E → X is said to be τ(E, E)-nuclear if (v )  { : ∈ N} τ( , ) there exist a sequence n in E such that the family Fvn n is E E - 1 equicontinuous, a bounded sequence (xn) in X and a sequence (αn) ∈ such that ∞  T (u) = αn u vn dλ xn for u ∈ E. (1.1)  n=1

Let  ∞ T nuc := inf |αn|vnE xnX , n=1

 1 where the infimum is taken over all sequences (vn) in E , (xn) in X and (αn) ∈ such that T admits a representation (1.1). 123 804 M. Nowak

  It is known that a τ(E, E )-nuclear operator T : E → X is (τ(E, E ), ·X )- continuous and τ(E, E)-compact, that is, T (V ) is relatively norm compact in X for some τ(E, E)-neighborhood V of 0 in E (see [23, Chap. 3, §7, Corollary 1], [12, Corollary 4, p. 379]). In this paper we study τ(E, E)-nuclear operators T : E → X. In Section 2 it is shown that every τ(E, E)-nuclear operator T : E → X is Bochner representable (see Theorem 2.3 below). In particular, we obtain that a linear operator T : L∞ → X ∞ 1 is τ(L , L )-nuclear if and only if its representing measure mT :  → X has the Radon-Nikodym property and |mT |() =T nuc (see Theorem 2.5 below). As an application, we obtain that some natural kernel operators on L∞ are τ(L∞, L1)- nuclear (see Proposition 2.9 below). In Section 3 it is shown that every τ(L∞, L1)- nuclear operator T : L∞ → X admits a factorization through some Orlicz space Lϕ, ϕ that is, T = S ◦ i∞, where S : L → X is a Bochner representable, compact operator ∞ ϕ and i∞ : L → L denotes the inclusion map (see Corollary 3.5).

2 Nuclear operators on Banach function spaces

Assume that T : E → X is a linear operator. Then the measure mT :  → X defined by

mT (A) := T (1A) for A ∈  is called a representing measure of T .  If, in particular, T is (τ(E, E ), ·X )-continuous, then using Proposition 1.1 we obtain that mT is countably additive. Since mT (A) = 0ifλ(A) = 0, by the Pettis theorem mT is λ-continuous, that is, mT λ. The following lemma will be useful.

 Lemma 2.1 Let T : E → Xbea(τ(E, E ), ·X )-continuous linear operator. If |mT |E () < ∞ and mT has the Radon-Nikodym property with respect to λ with a density g ∈ L1(X), then g ∈ E(X) and

1AgE(X) =|mT |E (A) for all A ∈ , and  T (u) = u(ω) g(ω) dλ for all u ∈ E. 

Proof First we shall show that for A ∈ ,  

|mT |E (A) = sup |s(ω)|g(ω)X dλ : s ∈ S(), sE ≤ 1 . (2.1) A 123 Nuclear operators on Banach function spaces 805   | |( ) =  (ω) λ = k 1 ∈ S() Note that mT A A g X d .Fors i=1 ci Ai ,wehave

k k   ci mT (A ∩ Ai ) = ci g(ω) dλ = s(ω) g(ω) dλ. ∩ i=1 i=1 A Ai A  = k 1 ∈ S()   ≤ We now show that for s i=1 ci Ai and s E 1, we have

k  k |ci | g(ω)X dλ = |ci ||mT |(A ∩ Ai ) ≤|mT |E (A). ∩ i=1 A Ai i=1

Indeed, let ε>0 be given. Then for each 1 ≤ i ≤ k, there exists a -partition ( ) ji ∩ Ai, j j=1 of A Ai such that

ji   ε | |( ∩ ) ≤  ( ) + . mT A Ai mT Aij X k|ci | j=1

Hence ⎛ ⎞ k k ji ⎝ ⎠ |ci ||mT |(A ∩ Ai ) ≤ ci mT (Aij)X + ε ≤|mT |E (A) + ε, i=1 i=1 j=1      k ji 1 = k 1 . because i=1 j=1 ci Aij i=1 ci Ai Then we have

k k |ci |mT (A ∩ Ai )X ≤ |ci ||mT |(A ∩ Ai ) i=1 i=1   k   k = | |  (ω) λ = | | 1 (ω)  (ω) λ ci g X d ci Ai g X d ∩ i=1 A Ai A i=1

= |s(ω)|g(ω)X dλ ≤|mT |E (A). A

Taking supremum on the left side, we get  

|mT |E (A) = sup |s(ω)|g(ω)X dλ : s ∈ S(), sE ≤ 1 . A

  We shall now show that g ∈ E (X), that is, g(·)X ∈ E . Indeed, let u ∈ E. Then there exists a sequence (sn) in S() such that 0 ≤ sn(ω) ↑ |u(ω)| λ-a.e. (see [13, Corollary I.6]). Choose c > 0 such that cuE ≤ 1. Then by the Fatou lemma, 123 806 M. Nowak  

c |u(ω)|g(ω)X dλ ≤ sup csn(ω)g(ω)X dλ ≤|mT |E (),  n 

  and this means that g(·)X ∈ E , that is, g ∈ E (X). Moreover, for A ∈  and u ∈ E with uE ≤ 1, using (2.1) we get  

|u(ω)|1A(ω) g(ω)X dλ ≤ sup sn(ω)1A(ω) g(ω)X dλ ≤|mT |E (A)  n  and it follows that 1AgE(X) ≤|mT |E (A).Inviewof(2.1) we get |mT |E (A) ≤ 1   | |  ( ) =1   Ag E (X). Hence mT E A Ag E (X). = k 1 ∈ S() Note that for s i=1 ci Ai ,wehave

k  T (s) = ci mT (Ai ) = s(ω) g(ω) dλ.  i=1

Let u ∈ E be given. Then there exists a sequence (sn) in S() such that |sn(ω) − u(ω)|→0 λ-a.e. and |sn(ω)|≤|u(ω)| λ-a.e. for all n ∈ N. Then |sn(ω) − u(ω)| 1 g(ω)X ≤ 2|u(ω)|g(ω)X λ-a.e., where u g(·)X ∈ L . Hence by the Lebesgue dominated convergence theorem,           sn(ω) g(ω) dλ − u(ω) g(ω) dλ  ≤ |sn(ω) − u(ω)|g(ω)X dλ → 0.   X  On the other hand, in view of Proposition 1.1 we have

T (sn) − T (u)X =T (sn − u)X → 0.  Hence T (u) =  u(ω)g(ω) dλ.  As a consequence of Lemma 2.1,wehave Proposition 2.2 Assume that T : E → X is a Bochner representable operator, that is, there exists g ∈ E(X) such that  T (u) = u(ω) g(ω) dλ forallu∈ E. 

Then the following statements hold:  (i) T is (τ(E, E ), ·X )-continuous. (ii) For every A ∈ , |mT |E (A) =1AgE(X).

Proof (i) Assume that un(ω) → 0 λ-a.e. and |un(ω)|≤|u(ω)| λ-a.e. for some u ∈ E and all n ∈ N. Since for n ∈ N, 

T (un)X ≤ |un(ω)|g(ω)X dλ,  123 Nuclear operators on Banach function spaces 807

1 where u g(·)X ∈ L , by the Lebesgue dominated convergence theorem, we  get T (un)X → 0. Hence in view of Proposition 1.1 T is (τ(E, E ), ·X )- continuous.  = k 1 ∈ S()   ≤ (ii) Assume that s i=1 ci Ai and s E 1. Then

k k   |ci |mT (Ai )X = |ci | g(ω)X dλ = |s(ω)|g(ω)X dλ  i=1 i=1 Ai ≤sE gE(X) ≤gE(X).

Hence |mT |E () ≤gE(X). Using (i) and Lemma 2.1, we get |mT |e (A) = 1AgE(X) for all A ∈ .  The following result shows a relationship between τ(E, E)-nuclear operators and Bochner representable operators T : E → X. Theorem 2.3 Assume that T : E → Xisaτ(E, E)-nuclear operator. Then T is Bochner representable and |mT |E () ≤T nuc.  Proof Let ε>0 be given. Then there exists a bounded sequence (vn) in E , a bounded 1 sequence (xn) in X and a sequence (αn) ∈ such that ∞  T (u) = αn u vn dλ xn for u ∈ E  n=1 and ∞ |αn|vnE xnX ≤T nuc + ε. (2.2) n=1

Hence we have ∞  mT (A) = αn vn dλ xn. (2.3) n=1 A  | |  () ≤  = k 1 ∈ S() We shall now show that mT E T nuc. Indeed, let s i=1 ci Ai with sE ≤ 1. Then using (2.2) we get

k k ∞  |ci |mT (Ai )X = ci αn vn(ω) dλ xnX = = = Ai i 1 i 1 n1  ∞ k  ≤ |αn| |ci | |vn(ω)| dλ xnX n=1 i=1 Ai ∞  = |αn|xnX |s(ω)||vn(ω)| dλ  n=1 123 808 M. Nowak

∞ ≤ |αn|xnX vnE ≤T nuc + ε. n=1

Hence we get

|mT |E () ≤T nuc. (2.4)  ∈ N := n α v ⊗ , ∈ N > For n ,letgn i=1 i i xi . Then for n k with n k,wehave        n  gn(ω)− gk (ω)X dλ =  αi vi (ω)xi  dλ      i=k+1  X  n n  ≤ |αi ||vi (ω)|xi X dλ = |αi |xi X |vi | dλ   i=k+1 i=k+1 n n ≤ |αi |xi X 1E vi E ≤ sup x j X sup v j E 1E |αi |. ∈N ∈N i=k+1 j j i=k+1

1 This follows that (gn) is a Cauchy sequence in the Banach space L (X), so there 1 exists g ∈ L (X) such that gn − g1 → 0. One can easily show that    v dλ x = v(ω) xdλ for v ∈ E , x ∈ X, A ∈ . A A

Hence for A ∈ ,wehave          n   n       αi vi dλ xi − gdλ  =  αi vi (ω) xi dλ − g(ω) dλ  A A A A i=1   X  i=1 X  n    ≤  αi vi (ω) xi − g(ω) dλ = gn(ω) − g(ω)X dλ → 0. A i=1 X A

Then in view of (2.3)forA ∈ , we get 

mT (A) = g(ω) dλ. A

 Using Lemma 2.1and (2.4) we see that g ∈ E (X) with gE(X) =|mT |E () ≤ T nuc and T (u) =  u(ω) g(ω) dλ for all u ∈ E. 

Now we shall study τ(L∞, L1)-nuclear operators T : L∞ → X. Making use of the Dunford–Pettis theorem (see [3, Theorem, p. 93]) we have

Theorem 2.4 For a subset H of L1 the following statements are equivalent: (i) H is relatively weakly compact. v < ∞ (ii) supv∈H 1 and H is uniformly integrable. ∞ (iii) {Fv : v ∈ H} is τ(L , L1)-equicontinuous. 123 Nuclear operators on Banach function spaces 809

Note that in view of Theorem 2.4 and Definition 1.2, a linear operator T : L∞ → X ∞ 1 is τ(L , L )-nuclear if there exist a bounded uniformly integrable sequence (vn) in 1 1 L , a bounded sequence (xn) in X and a sequence (αn) ∈ such that ∞  ∞ T (u) = αn u vn dλ xn for all u ∈ L (2.5)  n=1 and then  ∞ T nuc = inf |αn|vn1 xnX , n=1

1 1 where the infimum is taken over all sequences (vn) in L , (xn) in X and (αn) ∈ such that T admits a representation (2.5). Now we can state our main result. Theorem 2.5 For a linear operator T : L∞ → X the following statements are equiv- alent:

(i) mT has the Radon-Nikodym property with respect to λ. (ii) T is Bochner representable. (iii) Tisaτ(L∞, L1)-nuclear operator.

In this case |mT |() =T nuc. Proof (i)⇒(ii) Assume that (i) holds with the density g ∈ L1(X). Note that for s ∈ S(),wehave  

sdmT = s(ω) g(ω) dλ.  

∞ Let u ∈ L . Choose a sequence (sn) in S() such that u − sn∞ → 0. Then we have    

T (u) = udmT = lim sn dmT = lim sn(ω) g(ω) dλ = u(ω) g(ω) dλ.    

(ii)⇒(i) This is obvious. (ii)⇒(iii) Assume that (ii) holds, that is, there exists g ∈ L1(X) such that  ∞ T (u) = u(ω) g(ω) dλ for all u ∈ L . 

Let L1⊗ˆ X denote the projective of L1 and X, equipped with the norm π defined for w ∈ L1⊗ˆ X,by  ∞ π(w) := inf |αn|vn1 xnX , n=1 123 810 M. Nowak

(v ) 1 ( ) v  = where the infimum is taken over all sequences n in L , xn in X with lim n 1 =   (α ) ∈ 1 w = ∞ α v ⊗ 0 lim xn X and n such that n=1 n n xn (see [22, Proposition 2.8, pp. 21–22]). It is known that L1⊗ˆ X is isometrically isomorphic to L1(X) through the isometry J, where

J(v ⊗ x) := v(·) x for v ∈ L1, x ∈ X

(see [4, Example 10, p. 228], [22, Example 2.19, p. 29], [5, Theorem 1.1.10, p. 14]). Let 1 ε>0 be given. Then there exist sequence (vn) in L and (xn) in X with limn vn1 = 1 0 = lim xnX and (αn) ∈ such that

∞   −1 1 ˆ J (g) = αn vn ⊗ xn in L ⊗ X,π n=1 and ∞ −1 |αn|vn1 xnX ≤ π(J (g)) + ε =g1 + ε. (2.6) n=1

Hence   ∞ ∞   1 g = J αnvn ⊗ xn = αn vn(·) xn in L (X), ·1 . n=1 n=1

Note that      n     u(ω) g(ω) dλ − αi u(ω) vi (ω) dλ xi      i=1  X  n    ≤ |u(ω)| g(ω) − αi vi (ω) xi  dλ    i=1  X    n   n      ≤u∞ g(ω) − αi vi (ω) xi  dλ =u∞ · g − αi vi ⊗ xi   i=1 X i=1 1

Hence ∞  ∞ T (u) = αn u vn dλ xn for u ∈ L .  n=1

Since lim vn1 = 0, the set {vn : n ∈ N} is uniformly integrable and it follows that T is τ(L∞, L1)-nuclear. In view of (2.6) we get

T nuc ≤g1 =|mT |(). (2.7) 123 Nuclear operators on Banach function spaces 811

(iii)⇒(ii) Assume that (iii) holds. Then by Theorem 2.3 T is Bochner representable and

|mT |() =g1 ≤T nuc. (2.8)

Thus (i)⇔(ii)⇔(iii) hold and using (2.7) and (2.8), we get |mT |() =T nuc. 

Let baλ() denote the Banach space of all bounded finitely additive real measures μ on  such that μ(A) = 0ifλ(A) = 0, equipped with norm μ:=|μ|(). ∞ ∗ ∞ The Banach dual (L ) of L can be identified with baλ() through the integration ∞ ∗ ∞ mapping baλ() μ → Fμ ∈ (L ) , where Fμ(u) =  udμ for all u ∈ L and |μ|() =Fμ. Recall (see [26, p. 279]) that a linear operator T : L∞ → X is said to be nuclear if there exist a bounded sequence (μn) in baλ(), a bounded sequence (xn) in X and 1 a sequence (αn) ∈ such that

∞  ∞ T (u) = αn udμn xn for all u ∈ L .  n=1

∞ ∞ 1 It is well known that a bounded linear operator T : L → X is (τ(L , L ), ·X )- continuous if and only if its representing measure mT :  → X is λ-continuous (see [17, Proposition 3.1]). Hence due to Swartz (see [24, Theorem 1 and Theorem 3]) we have

∞ ∞ 1 Theorem 2.6 Assume that T : L → Xisa(τ(L , L ), ·X )-continuous linear operator. Then the following statements are equivalent: (i) T is nuclear. (ii) mT has the Radon-Nikodym property with respect to λ.

Combining Theorem 2.5 and Theorem 2.6, we get

Corollary 2.7 For a linear operator T : L∞ → X the following statements are equiv- alent: (i) Tisτ(L∞, L1)-nuclear. ∞ 1 (ii) Tis(τ(L , L ), ·X )-continuous and nuclear.

As an application of Theorem 2.5 we show that some natural kernel operators on L∞ are τ(L∞, L1)-nuclear. Assume that K1 and K2 are compact Hausdorff spaces and k(·, ·) ∈ C(K2 × K1). Let Bo be the σ -algebra of Borel sets in K1 and λ : Bo →[0, ∞) be a countably additive measure. We will need the following lemma.

∞ Lemma 2.8 For every u ∈ L (λ), the mapping u : K1 s → u(s) k(·, s) ∈ C(K2) is continuous. 123 812 M. Nowak

Proof Let s0 ∈ K1 and ε>0 be given. Then for every t ∈ K2 there exist a neighbor- hood Vt of t and a neighborhood Wt of s0 such that ε | k(z, s) − k(t, s0)|≤ for all z ∈ Vt , s ∈ Wt . u∞  ,..., ∈ = n :=  Hence there exist t1 tn K2 such that K2 i=1 Vti . Let us put W n ∈ ≤ ≤ ∈ ∈ i=1 Wti .Fort K2, choose i0 with 1 i0 n such that t Vti . Then for s W, | ( , ) − ( , |≤ ε 0 we have k t s k t s0 u∞ . Hence

 u(s) − u(s0)∞ = sup |k(t, s) − k(t, s0)|u∞ ≤ ε. t∈K2

This means that u is continuous. 

Note that the Banach space C(K1, C(K2)) can be embedded in the Banach space 1 L (C(K2)) such that with each function from C(K1, C(K2)) is associated its λ- 1 equivalence class in L (C(K2)). ∞ In view of Lemma 2.8 we can define a kernel operator Tk : L (λ) → C(K2) by  ∞ Tk(u) := u(s) k(·, s) dλ(s) for all u ∈ L (λ). K1

∗ For t ∈ K2,letδt (w) := w(t) for all w ∈ C(K2). Then δt ∈ C(K2) and according to the Hille’s theorem (see [DU, Theorem 6, p. 47]), we have  ∞ Tk(u)(t) = δt (Tk(u)) = u(s) k(t, s) dλ(s) for all u ∈ L (λ), t ∈ K2. K1

Then  ( ) = (1 ) = (·, ) λ( ) ∈ B , mTk A Tk A k s d s for all A o A

1 where the mapping K1 s → k(·, s) ∈ C(K2) belongs to L (C(K2)). Hence for A ∈ Bo,   | |( ) =  (·, ) λ( ) = | ( , )| λ( ). mTk A k s ∞ d s sup k t s d s A A t∈K2

Hence as a consequence of Theorem 2.5 we have

∞ ∞ 1 Proposition 2.9 The kernel operator Tk : L (λ) → C(K2) is τ(L (λ), L (λ))- nuclear and 

Tknuc = sup |k(t, s)| dλ(s). K1 t∈K2 123 Nuclear operators on Banach function spaces 813

3 Application of the theory of Orlicz spaces to vector measures

First we recall terminology and basic facts concerning Orlicz spaces and Orlicz- Bochner spaces (see [19] for more details). By a Young function we mean here a convex continuous mapping ϕ :[0, ∞) →[0, ∞) that vanishes only at 0 and ϕ(t)/t → 0ast → 0 and ϕ(t)/t →∞as t →∞.Byϕ∗ we denote the complementary function of ϕ in the sense of Young, that is, ϕ∗(t) = sup{ts − ϕ(s) : s ≥ 0} for t ≥ 0. Note that ϕ∗∗ = ϕ. The corresponding Orlicz space Lϕ is an ideal of L0 defined by    ϕ L := u ∈ L0 : ϕ(α|u (ω)| ) dλ<∞ for some α>0 ,  and equipped with the topology Tϕ, defined by two equivalent norms:    uϕ := inf α>0 : ϕ(|u(ω)| /α)dλ ≤ 1 ,      0 ϕ∗ ∗ uϕ := sup |u(ω) v(ω)| dλ : v ∈ L , ϕ (|v(ω)|) dλ ≤ 1 ,  

called the Luxemburg norm and the Orlicz norm. Then we have: uϕ ≤ 1 if and only if ϕ(|u(ω)|) dλ ≤ 1, unϕ → 0 if and only if  ϕ(α|un(ω)|) dλ → 0 for every α>0,    ϕ Eϕ := u ∈ L : ϕ(α | u(ω)|) dλ<∞ for all α>0    = ∈ ϕ :1  → λ( ) → , u L An u ϕ 0if An 0

∗ (Lϕ) = Lϕ . The Orlicz-Bochner space Lϕ(X) is defined by    ϕ 0 L (X) := g ∈ L (X) : ϕ(α  g(ω)X ) dλ<∞ for some α>0    0 ϕ = g ∈ L (X) :g(·)X ∈ L .

For g ∈ Lϕ(X),let

0 0 gϕ :=  g(·)X ϕ and gϕ :=  g(·)X ϕ.

Then

ϕ 0 ϕ E (X) := g ∈ L (X) :g(·)X ∈ E .

Assume that ϕ is Young function. If m :  → X is a λ-continuous measure, we write |m|ϕ∗ instead of |m|(Lϕ ) . 123 814 M. Nowak

Lemma 3.1 Assume that m :  → Xisaλ-continuous measure and ϕ is Young function. Then for A ∈ , we have

−1 −1 1 |m|(A) ≤ ϕ |m|ϕ∗ (A). λ(A)

Proof ∈  ε>  ( )n Let A and 0 be given. Then there is a -partition Ai i=1 of A such that

n −  1 1 |m|(A) ≤ m(A ) + ϕ−1 ε. i X λ(A) i=1    −1 1  = ϕ−1 1 It is known that A ϕ λ(A) (see [RR, Chap. 3.4, Corollary 7, p. 79]). Hence      n     ϕ−1 1 1  = ϕ−1 1 1  =  Ai   A 1 λ(A) ϕ λ(A) ϕ i=1 and

n 1  1 ϕ−1 |m|(A) ≤ ϕ−1 m(A ) + ε. λ(A) λ(A) i X i=1   − ϕ 1 1 | |( ) ≤| | ∗ ( ) It follows that λ(A) m A m ϕ A ,so

−1 −1 1 |m|(A) ≤ ϕ |m|ϕ∗ (A). λ(A)    ϕ γ [Tϕ, T ] γϕ By 0 Lϕ (in brief, ) we denote the natural mixed topology on L (in the sense of Wiweger), that is, γϕ is the finest linear topology that agrees with T0 ϕ on Tϕ-bounded sets in L (see [1,25] for more details). Then γϕ is a locally convex- solid Hausdorff topology and γϕ and Tϕ have the same bounded sets. This means that ϕ ϕ (L ,γϕ) is a generalized DF-space (see [20]) and it follows that (L ,γϕ) is quasi- normable (see [20, p. 422]). Hence as a consequence of the Grothendieck’s classical result (see [20, p. 429]), we have

Proposition 3.2 For a linear operator S : Lϕ → X the following statements are equivalent:

(i) S is (γϕ, ·X )-continuous and compact. ϕ (ii) S is γϕ-compact, that is, there exists a γϕ-neighborhood V of 0 in L such that T (V ) is relatively norm compact in X. 123 Nuclear operators on Banach function spaces 815

We say that a Young function ϕ increases essentially more rapidly than another ψ (in symbols, ψ ϕ) if for arbitrary c > 0, ψ(ct)/ϕ(t) → 0ast → 0 and t →∞. Note that Lϕ ⊂ Eψ if ψ ϕ. The following result will be useful (see [16, Theorem 2.1]).

Theorem 3.3 Let ϕ be a Young function. Then   γϕ { · ψ : ψ ϕ} (i) is generated by the family of norms  Lϕ . ϕ ∗ ϕ∗ ϕ (ii) (L ,γϕ) ={Fv : v ∈ E }, where Fv(u) =  u(ω) v(ω) dλ for all u ∈ L . According to [15, Corollary 1.6] we have the following identity:  ϕ L1 = E , where ϕ runs over the family of all Young functions. (3.1) ϕ

Now we can state the main result in this section.

Theorem 3.4 Assume that a measure m :  → X has the Radon-Nikodym property ∗ with the density g ∈ L1(X). Then there exists a Young function ϕ such that g ∈ Eϕ (X) and the following statements hold: ϕ (i) The operator Sg : L → X defined by  ϕ Sg(u) := u(ω) g(ω) dλ for all u ∈ L 

is (γϕ, ·X )-continuous and compact. (ii) Sg is γϕ-compact, that is, there exists a Young function ψ with ψ ϕ such that ϕ {  u(ω)g(ω) dλ : u ∈ L , uψ ≤ 1} is a relatively norm compact subset of X. 0 (iii) |m|ϕ∗ () < ∞ and |m|ϕ∗ (A) =1Agϕ∗ for every A ∈ . (iv) |m|ϕ∗ is λ-continuous, that is, |m|ϕ∗ (An) → 0 if λ(An) → 0. Proof In view of (3.1) there exists a Young function ψ such that g ∈ Eψ (X), that is, ∗ g(·) ∈ Eψ . Hence |u|g(·) ∈ L1 for every u ∈ Lψ . Let us put ϕ = ψ∗. Then X X∗ ∗ ∗ ϕ∗ = ψ∗∗ = ψ and (Lϕ) = Lϕ = Lψ and Lψ = Lϕ, and so g ∈ Eϕ (X). Then for u ∈ Lϕ,wehave   ( ) ≤ | (ω)| (ω) λ = (| |), Sg u X u g X d Fg(·)X u 

ϕ∗ ϕ ϕ∗ where g(·)X ∈ E . The inequality shows that Sg is (|σ |(L , E ), ·X )- ∗ continuous, where |σ| denotes the absolute . Since (|σ |(Lϕ, Eϕ ) is the ∗ coarsest locally convex-solid topology on Lϕ with dual Eϕ , and by Theorem 3.3(ii) γϕ is such a topology, it follows that Sg is (γϕ, ·X )-continuous. To show that Sg is compact, choose a sequence ( fn) of X-valued simple functions on  such that  fn(ω) − g(ω)X → 0 and  fn(ω)X ≤g(ω)X λ-a.e. and for all ∗ n ∈ N (see [7, Theorem 6, p. 4]). Hence for every α>0, we have that ϕ (α fn(ω) − ∗ ∗ g(ω)X ) → 0 λ-a.e. and ϕ (α fn(ω)−g(ω)X ) ≤ ϕ (2αg(ω)X )λ-a.e. and for all 123 816 M. Nowak  ∈ N ϕ∗( α (ω) ) λ<∞ n , where  2 g X d . By the Lebesgue dominated convergence ∗ 0 theorem,  ϕ (α fn(ω) − g(ω)X ) dλ → 0 and this means that  fn − gϕ∗ → 0. ϕ For each n ∈ N,letSn : L → X be a linear operator defined by  ϕ Sn(u) := u(ω) fn(ω) dλ for all u ∈ L . 

Note that the range of each Sn is contained in the span of finite set of values of fn. ϕ Therefore Sn is compact and by the Hölder’s inequality for each u ∈ L ,wehave        (Sn − Sg)(u)X =  u(ω)( fn(ω) − g(ω)) dλ    X 0 ≤ |u(ω)|fn(ω) − g(ω)X dλ ≤uϕ  fn − gϕ∗ . 

It follows that Sn − Sg→0, so Sg is a compact operator. (ii) In view of (i) and Proposition 3.2 Sg is γϕ-compact. Using Theorem 3.3 we obtain that (ii) holds. = k 1 ∈ S()   ≤ (iii) Let s i=1 ci Ai and s ϕ 1. Then

k k   0 |ci |m(Ai )X = |ci | g(ω)X dλ = |s(ω)|g(ω)X dλ ≤gϕ∗  i=1 i=1 Ai

| | ∗ () ≤ 0 = and hence m ϕ g ϕ∗ . Note that m mSg . Hence using Lemma 2.1 we have 0 that |mϕ∗ |(A) =1Agϕ∗ for A ∈ . ∗ (iv) This follows from (iii) because g ∈ Eϕ (X). 

∞ ϕ ∞ Let i∞ : L → L denotes the inclusion map. Note that i∞ is (σ(L , L1), ϕ ϕ∗ ∞ ϕ ϕ∗ σ(L , L ))-continuous, and it follows that i∞ is (τ(L , L1), τ(L , L ))-continuous ϕ ϕ∗ ∞ (see [8, Theorem 8.6.1]). Since γϕ ⊂ τ(L , L ), we obtain that i∞ is (τ(L , L1), γϕ)- continuous. As a consequence of Theorem 3.4 and Theorem 2.5, we can show that every τ(L∞, L1)-nuclear operator T : L∞ → X admits a factorization through some Orlicz space Lϕ.

Corollary 3.5 Assume that T : L∞ → Xisaτ(L∞, L1)-nuclear operator. Then there exists a Young function ϕ such that T = S ◦ i∞, where ϕ (i) S : L → X is a Bochner representable and γϕ-compact linear operator. (ii) |mS|ϕ∗ (An) → 0 if λ(An) → 0.

123 Nuclear operators on Banach function spaces 817

Proof (i) In view of Theorem 2.5 the representing measure mT :  → X has the 1 Radon-Nikodym property with respect to λ, that is, mT = gλ, where g ∈ L (X). Hence according to Theorem 3.4 there exists a Young function ϕ such that g ∈ ∗ Eϕ (X) and an operator S : Lϕ → X defined by

 ϕ S(u) := u(ω) g(ω) dλ for all u ∈ L , 

  ∞ is γϕ-compact. Note that for u ∈ L , T (u) =  udmT =  u(ω)g(ω) dλ = S(u).  ( ) = (ω) λ ∈  (ii) Since mS A A g d for all A , using Theorem 3.4 we have that |mS|ϕ∗ (An) → 0ifλ(An) → 0. 

Acknowledgements The author wishes to express his thanks to the referee for remarks and suggestions leading to an improvement of the paper.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

References

1. Cooper, J.B.: The strict topology and spaces with mixed topologies. Proc. Amer. Math. Soc. 30(3), 583–592 (1971) 2. Cristescu, R.: Topological Vector Spaces. Noordhoff Int. Publ, Leyden (1977) 3. Diestel, J.: Sequences and Series in Banach Spaces. Graduate Texts in Math, vol. 92. Springer-Verlag, New York (1984) 4. Diestel, J., Uhl, J.J.: Vector Measures, Math. Surveys, vol. 15. American Mathematical Society, Prov- idence, RI (1977) 5. Diestel, J., Fourie, J., Swart, J.: The Metric Theory of Tensor Products. Grothendieck’s resume revisited. American Mathematical Society, Providence (2008) 6. Dinculeanu, N.: Vector Measures. Pergamon Press, New York (1967) 7. Dinculeanu, N.: Vector Integration and Stochastic Integration in Banach Spaces. A Wiley-Interscience Publication, New York (2000) 8. Edwards, R.E.: . Theory and Applications. Holt, Rinehart and Winston, New York (1965) 9. Fremlin, D.H.: Topological Riesz Spaces and Measure Theory. Cambridge University Press, London- New York (1974) 10. Grothendieck, A.: Sur les espaces (F) et (DF). Summa Brasil. Math. 3, 57–123 (1954) 11. Grothendieck, A.: Produits tensoriels topologiques et espaces nucléaires. Mem, vol. 16. American Mathematical Society, Providence (1955) 12. Jarchow, H.: Locally Convex Spaces. Teubner, Stuttgart (1981) 13. Kantorovitch, L.V., Akilov, G.P.: Functional Analysis. Pergamon Press, Oxford-Elmsford, New York (1982) 14. Lin, P.-K.: Köthe-Bochner Function Spaces. Birkhaüser, Boston (2003) 123 818 M. Nowak

15. Nowak, M.: Some equalities among Orlicz spaces II. Bull. Pol. Acad. Sci. Math. 34(11–12), 675–687 (1986) 16. Nowak, M.: Compactness of Bochner representable operators on Orlicz spaces. Positivity 13, 193–199 (2009) 17. Nowak, M.: Absolutely continuous operators on function spaces and vector measures. Positivity 17, 525–533 (2013) 18. Pietsch, A.: Nuclear Locally Convex Spaces, Ergebnisse der Mathematik und ihrer Grenzebiete, vol. 66. Springer, Berlin (1972) 19. Rao, M.M., Ren, Z.D.: Theory of Orlicz Spaces. Marcel Dekker, New York (1991) 20. Ruess, W.: [Weakly] compact operators and DF-spaces. Pacific J. Math. 98, 419–441 (1982) 21. Ruston, A.F.: On the of integral equations for operators belonging to the for a general Banach space. Proc. London Math. Soc. 55(2), 109–124 (1951) 22. Ryan, R.: Introduction to the Tensor Products of Banach Spaces. Springer Monographs in Math. Springer-Verlag, London (2002) 23. Schaefer, H.H.: Topological Vector Spaces. Springer, New York, Heidelberg, Berlin (1971) 24. Swartz, C.: An operator characterization of vector measures which have Radon-Nikodym derivatives. Math. Ann. 202, 77–84 (1973) 25. Wiweger, A.: Linear spaces with mixed topology. Studia Math. 20, 47–68 (1961) 26. Yosida, K.: Functional Analysis, 4th edn. Springer-Verlag, Berlin, Heidelberg, New York (1974) 27. Zaanen, A.C.: Riesz Spaces II. North-Holland Publ. Comp, Amsterdam (1983)

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

123