Abstract
Let X be a Banach space and E be a perfect Banach function space over a finite measure space \((\Omega ,\Sigma ,\lambda )\) such that \(L^\infty \subset E\subset L^1\). Let \(E'\) denote the Köthe dual of E and \(\tau (E,E')\) stand for the natural Mackey topology on E. It is shown that every nuclear operator \(T:E\rightarrow X\) between the locally convex space \((E,\tau (E,E'))\) and a Banach space X is Bochner representable. In particular, we obtain that a linear operator \(T:L^\infty \rightarrow X\) between the locally convex space \((L^\infty ,\tau (L^\infty ,L^1))\) and a Banach space X is nuclear if and only if its representing measure \(m_T:\Sigma \rightarrow X\) has the Radon-Nikodym property and \(|m_T|(\Omega )=\Vert T\Vert _{nuc}\) (= the nuclear norm of T). As an application, it is shown that some natural kernel operators on \(L^\infty \) are nuclear. Moreover, it is shown that every nuclear operator \(T:L^\infty \rightarrow X\) admits a factorization through some Orlicz space \(L^\varphi \), that is, \(T=S\circ i_\infty \), where \(S:L^\varphi \rightarrow X\) is a Bochner representable and compact operator and \(i_\infty :L^\infty \rightarrow L^\varphi \) is the inclusion map.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction and preliminaries
We assume that \((X,\Vert \cdot \Vert _X)\) is a real Banach space. For terminology concerning Riesz spaces and function spaces, we refer the reader to [9, 13, 27].
We assume that \((\Omega ,\Sigma ,\lambda )\) is a finite measure space. Let \(L^0\) denote the corresponding space of \(\lambda \)-equivalence classes of all \(\Sigma \)-measurable real functions on \(\Omega \). Then \(L^0\) is a super Dedekind complete Riesz space, equipped with the topology \({\mathcal {T}}_o\) of convergence in measure. By \({\mathcal {S}}(\Sigma )\) we denote the space of all real \(\Sigma \)-simple functions \(s=\sum ^n_{i=1} c_i\mathbb {1}_{A_i}\), where the sets \(A_i\in \Sigma \) are pairwise disjoint.
Let \((E,\Vert \cdot \Vert _E)\) be a Banach function space, where E is an order ideal of \(L^0\) such that \(L^\infty \subset E\subset L^1\), and \(\Vert \cdot \Vert _E\) is a Riesz norm on E. By \({\mathcal {T}}_E\) we denote the \(\Vert \cdot \Vert _E\)-norm topology on E. By \(E'\) we denote the Köthe dual of E, that is,
The associated norm \(\Vert \cdot \Vert _{E'}\) on \(E'\) is defined for \(v\in E'\) by
We will assume that E is perfect, that is, \(E=E''\) and \(\Vert u\Vert _E=\Vert u\Vert _{E''}\). The order continuous dual \(E^\sim _n\) of E separates the points of E and \(E^\sim _n\) can be identified with \(E'\) through the Riesz isomorphism \(E'\ni v\mapsto F_v\in E^\sim _n\), where
(see [13, Theorem 6.1.1]). The Mackey topology \(\tau (E,E')\) is a locally convex-solid Hausdorff topology with the Lebesgue property (see [9, Corollary 82H]). Then \(\tau (E,E')\subset {\mathcal {T}}_E\) and \(\tau (E,E')={\mathcal {T}}_E\) if the norm \(\Vert \cdot \Vert _E\) is order continuous.
The most important classes of Banach function spaces are Lebesgue spaces \(L^p\) \((1\le p\le \infty )\) and Orlicz spaces \(L^\varphi \) (see [19]).
Now we present a characterization of \((\tau (E,E'),\Vert \cdot \Vert _X)\)-continuous linear operators \(T:E\rightarrow X\) (see [17, Proposition 2.2]).
Proposition 1.1
For a bounded linear operator \(T:E\rightarrow X\) the following statements are equivalent:
-
(i)
T is \((\tau (E,E'),\Vert \cdot \Vert _X)\)-continuous.
-
(ii)
\(\Vert T(u_n)\Vert _X\rightarrow 0\) if \(\,u_n(\omega )\rightarrow 0\) \(\lambda \)-a.e. and \(|u_n(\omega )|\le |u(\omega )|\) \(\lambda \)-a.e. for some \(u\in E\) and all \(n\in {\mathbb {N}}\).
-
(iii)
For each \(u\in E\), \(\Vert T(u\mathbb {1}_{A_n})\Vert _X\rightarrow 0\) whenever \(\lambda (A_n)\rightarrow 0\).
For terminology and basic facts concerning vector measure, we refer the reader to [4, 6, 7, 22]. For a finitely additive measure \(m:\Sigma \rightarrow X\), by |m|(A) we denote the variation of m on \(A\in \Sigma \). A measure \(m:\Sigma \rightarrow X\) is said to be \(\lambda \) -continuous if \(\Vert m(A_n)\Vert _X\rightarrow 0\) whenever \(\lambda (A_n)\rightarrow 0\).
Let \(L^1(X)\) denote the Banach space of \(\lambda \)-equivalence classes of all X-valued Bochner integrable functions g defined on \(\Omega \), equipped with the norm \(\Vert g\Vert _1:=\int _\Omega \Vert g(\omega )\Vert _X d\lambda .\)
Recall that a \(\lambda \)-continuous measure \(m:\Sigma \rightarrow X\) of finite variation is said to have the Radon-Nikodym property with respect to \(\lambda \) if there exists a function \(g\in L^1(X)\) such that \(m(A)=\int _A g(\omega )\,d\lambda \) for all \(A\in \Sigma \). Then we write \(m=g\lambda \) and a function g is called the density of m with respect to \(\lambda \).
Assume that \(m:\Sigma \rightarrow X\) is a \(\lambda \)-continuous measure. Following [6, § 13] for \(A\in \Sigma \), we put
where the supremum is taken for all functions \(s=\sum ^n_{i=1} c_i\mathbb {1}_{A_i}\in \ {\mathcal {S}}(\Sigma )\) such that \(A_i\subset A\) for \(1\le i\le n\) and \(\Vert s\Vert _E\le 1\). The set function \(|m|_{E'}\) will be called a \(E'\) -variation of the measure m.
If, in particular, \(E=L^\infty \), then \(|m|_{L^1}(A)=|m|(A)\) for \(A\in \Sigma \).
Let \(L^0(X)\) stand for the linear space of \(\lambda \)-equivalence classes of all strongly \(\Sigma \)-measurable functions \(g:\Omega \rightarrow X\). Let
Then E(X) equipped with the norm \(\Vert g\Vert _{E(X)}:=\Vert \cdot \Vert g(\cdot )\Vert _X\Vert _{E}\) is a Banach space, called a Köthe-Bochner space (see [14]).
Definition 1.1
A bounded linear operator \(T:E\rightarrow X\) is said to be Bochner representable, if there exists \(g\in E'(X)\) such that
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] (see also [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 \(T:E\rightarrow X\) is said to be \(\tau (E,E')\) -nuclear if there exist a sequence \((v_n)\) in \(E'\) such that the family \(\{F_{v_n}:n\in {\mathbb {N}}\}\) is \(\tau (E,E')\)-equicontinuous, a bounded sequence \((x_n)\) in X and a sequence \((\alpha _n)\in \ell ^1\) such that
Let
where the infimum is taken over all sequences \((v_n)\) in \(E'\), \((x_n)\) in X and \((\alpha _n)\in \ell ^1\) such that T admits a representation (1.1).
It is known that a \(\tau (E,E')\)-nuclear operator \(T:E\rightarrow X\) is \((\tau (E,E'),\Vert \cdot \Vert _X)\)-continuous and \(\tau (E,E')\)-compact, that is, T(V) is relatively norm compact in X for some \(\tau (E,E')\)-neighborhood V of 0 in E (see [23, Chap. 3, § 7, Corollary 1], [12, Corollary 4, p. 379]).
In this paper we study \(\tau (E,E')\)-nuclear operators \(T:E\rightarrow X\). In Section 2 it is shown that every \(\tau (E,E')\)-nuclear operator \(T:E\rightarrow X\) is Bochner representable (see Theorem 2.3 below). In particular, we obtain that a linear operator \(T:L^\infty \rightarrow X\) is \(\tau (L^\infty ,L^1)\)-nuclear if and only if its representing measure \(m_T:\Sigma \rightarrow X\) has the Radon-Nikodym property and \(|m_T|(\Omega )=\Vert T\Vert _{nuc}\) (see Theorem 2.5 below). As an application, we obtain that some natural kernel operators on \(L^\infty \) are \(\tau (L^\infty ,L^1)\)-nuclear (see Proposition 2.9 below). In Section 3 it is shown that every \(\tau (L^\infty ,L^1)\)-nuclear operator \(T:L^\infty \rightarrow X\) admits a factorization through some Orlicz space \(L^\varphi \), that is, \(T=S\circ i_\infty \), where \(S:L^\varphi \rightarrow X\) is a Bochner representable, compact operator and \(i_\infty :L^\infty \rightarrow L^\varphi \) denotes the inclusion map (see Corollary 3.5).
2 Nuclear operators on Banach function spaces
Assume that \(T:E\rightarrow X\) is a linear operator. Then the measure \(m_T:\Sigma \rightarrow X\) defined by
is called a representing measure of T.
If, in particular, T is \((\tau (E,E'),\Vert \cdot \Vert _X)\)-continuous, then using Proposition 1.1 we obtain that \(m_T\) is countably additive. Since \(m_T(A)=0\) if \(\lambda (A)=0\), by the Pettis theorem \(m_T\) is \(\lambda \)-continuous, that is, \(m_T\ll \lambda \).
The following lemma will be useful.
Lemma 2.1
Let \(T:E\rightarrow X\) be a \((\tau (E,E'),\Vert \cdot \Vert _X)\)-continuous linear operator. If \(|m_T|_{E'}(\Omega )<\infty \) and \(m_T\) has the Radon-Nikodym property with respect to \(\lambda \) with a density \(g\in L^1(X)\), then \(g\in E'(X)\) and
and
Proof
First we shall show that for \(A\in \Sigma \),
Note that \(|m_T|(A)=\int _A\Vert g(\omega )\Vert _X\,d\lambda \). For \(s=\sum ^k_{i=1}c_i\mathbb {1}_{A_i} \in {\mathcal {S}}(\Sigma )\), we have
We now show that for \(s=\sum ^k_{i=1} c_i\mathbb {1}_{A_i}\in {\mathcal {S}}(\Sigma )\) and \(\Vert s\Vert _E\le 1\), we have
Indeed, let \(\varepsilon >0\) be given. Then for each \(1\le i\le k\), there exists a \(\Sigma \)-partition \((A_{i,j})^{j_i}_{j=1}\) of \(A\cap A_i\) such that
Hence
because \(\sum ^k_{i=1}\big (\sum ^{j_i}_{j=1}c_i\, \mathbb {1}_{A_{ij}}\big )=\sum ^k_{i=1} c_i\,\mathbb {1}_{A_i}.\)
Then we have
Taking supremum on the left side, we get
We shall now show that \(g\in E'(X)\), that is, \(\Vert g(\cdot )\Vert _X\in E'\).
Indeed, let \(u\in E\). Then there exists a sequence \((s_n)\) in \({\mathcal {S}}(\Sigma )\) such that \(0\le s_n(\omega )\uparrow |u(\omega )|\) \(\lambda \)-a.e. (see [13, Corollary I.6]). Choose \(c>0\) such that \(\Vert cu\Vert _E\le 1\). Then by the Fatou lemma,
and this means that \(\Vert g(\cdot )\Vert _X\in E'\), that is, \(g\in E'(X)\). Moreover, for \(A\in \Sigma \) and \(u\in E\) with \(\Vert u\Vert _E\le 1\), using (2.1) we get
and it follows that \(\Vert \mathbb {1}_A g\Vert _{E'(X)}\le |m_T|_{E'}(A)\). In view of (2.1) we get \(|m_T|_{E'}(A)\le \Vert \mathbb {1}_A g\Vert _{E'(X)}\). Hence \(|m_T|_{E'}(A)=\Vert \mathbb {1}_A g\Vert _{E'(X)}\).
Note that for \(s=\sum ^k_{i=1} c_i\mathbb {1}_{A_i}\in {\mathcal {S}}(\Sigma )\), we have
Let \(u\in E\) be given. Then there exists a sequence \((s_n)\) in \({\mathcal {S}}(\Sigma )\) such that \(|s_n(\omega )-u(\omega )|\rightarrow 0\) \(\lambda \)-a.e. and \(|s_n(\omega )|\le |u(\omega )|\) \(\lambda \)-a.e. for all \(n\in {\mathbb {N}}\). Then \(|s_n(\omega )-u(\omega )|\) \(\Vert g(\omega )\Vert _X\le 2|u(\omega )|\) \(\Vert g(\omega )\Vert _X\) \(\lambda \)-a.e., where \(u\,\Vert g(\cdot )\Vert _X\in L^1\). Hence by the Lebesgue dominated convergence theorem,
On the other hand, in view of Proposition 1.1 we have
Hence \(T(u)=\int _\Omega u(\omega ) g(\omega )\,d\lambda \). \(\square \)
As a consequence of Lemma 2.1, we have
Proposition 2.2
Assume that \(T:E\rightarrow X\) is a Bochner representable operator, that is, there exists \(g\in E'(X)\) such that
Then the following statements hold:
-
(i)
T is \((\tau (E,E'),\Vert \cdot \Vert _X)\)-continuous.
-
(ii)
For every \(A\in \Sigma \), \(|m_T|_{E'}(A)=\Vert \mathbb {1}_A g\Vert _{E'(X)}.\)
Proof
-
(i)
Assume that \(u_n(\omega )\rightarrow 0\) \(\lambda \)-a.e. and \(|u_n(\omega )|\le |u(\omega )|\) \(\lambda \)-a.e. for some \(u\in E\) and all \(n\in {\mathbb {N}}\). Since for \(n\in {\mathbb {N}}\),
$$\begin{aligned} \Vert T(u_n)\Vert _X\le \int _\Omega |u_n(\omega )|\ \Vert g(\omega )\Vert _X\,d\lambda , \end{aligned}$$where \(u\,\Vert g(\cdot )\Vert _X\in L^1\), by the Lebesgue dominated convergence theorem, we get \(\Vert T(u_n)\Vert _X\rightarrow 0\). Hence in view of Proposition 1.1T is \((\tau (E,E'),\Vert \cdot \Vert _X)\)-continuous.
-
(ii)
Assume that \(s=\sum ^k_{i=1} c_i\mathbb {1}_{A_i}\in {\mathcal {S}}(\Sigma )\) and \(\Vert s\Vert _E\le 1\). Then
$$\begin{aligned} \begin{array}{rl} \displaystyle \sum ^k_{i=1}|c_i|\ \Vert m_T(A_i)\Vert _X &{} \displaystyle =\sum ^k_{i=1}|c_i|\int _{A_i} \Vert g(\omega )\Vert _X\,d\lambda =\int _\Omega |s(\omega )|\ \Vert g(\omega )\Vert _X\,d\lambda \\ &{} \displaystyle \le \Vert s\Vert _E\ \Vert g\Vert _{E'(X)}\le \Vert g\Vert _{E'(X)}. \end{array} \end{aligned}$$
Hence \(|m_T|_{E'}(\Omega )\le \Vert g\Vert _{E'(X)}\). Using (i) and Lemma 2.1, we get \(|m_T|_{e'}(A)=\Vert \mathbb {1}_A g\Vert _{E'(X)}\) for all \(A\in \Sigma \). \(\square \)
The following result shows a relationship between \(\tau (E,E')\)-nuclear operators and Bochner representable operators \(T:E\rightarrow X\).
Theorem 2.3
Assume that \(T:E\rightarrow X\) is a \(\tau (E,E')\)-nuclear operator. Then T is Bochner representable and \(|m_T|_{E'}(\Omega )\le \Vert T\Vert _{nuc}\).
Proof
Let \(\varepsilon >0\) be given. Then there exists a bounded sequence \((v_n)\) in \(E'\), a bounded sequence \((x_n)\) in X and a sequence \((\alpha _n)\in \ell ^1\) such that
and
Hence we have
We shall now show that \(|m_T|_{E'}(\Omega )\le \Vert T\Vert _{nuc}\). Indeed, let \(s=\sum ^k_{i=1}c_i\mathbb {1}_{A_i}\in {\mathcal {S}}(\Sigma )\) with \(\Vert s\Vert _E\le 1\). Then using (2.2) we get
Hence we get
For \(n\in {\mathbb {N}}\), let \(g_n:=\sum ^n_{i=1}\alpha _i v_i\otimes x_i\). Then for \(n,k\in {\mathbb {N}}\) with \(n>k\), we have
This follows that \((g_n)\) is a Cauchy sequence in the Banach space \(L^1(X)\), so there exists \(g\in L^1(X)\) such that \(\Vert g_n-g\Vert _1\rightarrow 0\). One can easily show that
Hence for \(A\in \Sigma \), we have
Then in view of (2.3) for \(A\in \Sigma \), we get
Using Lemma 2.1 and (2.4) we see that \(g\in E'(X)\) with \(\Vert g\Vert _{E'(X)}=|m_T|_{E'}(\Omega )\le \Vert T\Vert _{nuc}\) and \(T(u)=\int _\Omega u(\omega )\, g(\omega )\,d\lambda \) for all \(u\in E\). \(\square \)
Now we shall study \(\tau (L^\infty ,L^1)\)-nuclear operators \(T:L^\infty \rightarrow X\).
Making use of the Dunford–Pettis theorem (see [3, Theorem, p. 93]) we have
Theorem 2.4
For a subset H of \(L^1\) the following statements are equivalent:
-
(i)
H is relatively weakly compact.
-
(ii)
\(\sup _{v\in H}\Vert v\Vert _1<\infty \) and H is uniformly integrable.
-
(iii)
\(\{F_v:v\in H\}\) is \(\tau (L^\infty ,L^1)\)-equicontinuous.
Note that in view of Theorem 2.4 and Definition 1.2, a linear operator T : \(L^\infty \rightarrow X\) is \(\tau (L^\infty ,L^1)\) -nuclear if there exist a bounded uniformly integrable sequence \((v_n)\) in \(L^1\), a bounded sequence \((x_n)\) in X and a sequence \((\alpha _n)\in \ell ^1\) such that
and then
where the infimum is taken over all sequences \((v_n)\) in \(L^1\), \((x_n)\) in X and \((\alpha _n)\in \ell ^1\) such that T admits a representation (2.5).
Now we can state our main result.
Theorem 2.5
For a linear operator \(T:L^\infty \rightarrow X\) the following statements are equivalent:
-
(i)
\(m_T\) has the Radon-Nikodym property with respect to \(\lambda \).
-
(ii)
T is Bochner representable.
-
(iii)
T is a \(\tau (L^\infty ,L^1)\)-nuclear operator.
In this case \(|m_T|(\Omega )=\Vert T\Vert _{nuc}\).
Proof
(i)\(\Rightarrow \)(ii) Assume that (i) holds with the density \(g\in L^1(X)\). Note that for \(s\in {\mathcal {S}}(\Sigma )\), we have
Let \(u\in L^\infty \). Choose a sequence \((s_n)\) in \({\mathcal {S}}(\Sigma )\) such that \(\Vert u-s_n\Vert _\infty \rightarrow 0\). Then we have
(ii)\(\Rightarrow \)(i) This is obvious.
(ii)\(\Rightarrow \)(iii) Assume that (ii) holds, that is, there exists \(g\in L^1(X)\) such that
Let \(L^1\hat{\otimes } X\) denote the projective tensor product of \(L^1\) and X, equipped with the norm \(\pi \) defined for \(w\in L^1\hat{\otimes } X\), by
where the infimum is taken over all sequences \((v_n)\) in \(L^1\), \((x_n)\) in X with \(\lim \Vert v_n\Vert _1=0=\lim \Vert x_n\Vert _X\) and \((\alpha _n)\in \ell ^1\) such that \(w=\sum ^\infty _{n=1}\alpha _n v_n\otimes \,x_n\) (see [22, Proposition 2.8, pp. 21–22]). It is known that \(L^1\hat{\otimes }\, X\) is isometrically isomorphic to \(L^1(X)\) through the isometry J, where
(see [4, Example 10, p. 228], [22, Example 2.19, p. 29], [5, Theorem 1.1.10, p. 14]). Let \(\varepsilon >0\) be given. Then there exist sequence \((v_n)\) in \(L^1\) and \((x_n)\) in X with \(\lim _n\Vert v_n\Vert _1=0=\lim \Vert x_n\Vert _X\) and \((\alpha _n)\in \ell ^1\) such that
and
Hence
Note that
Hence
Since \(\lim \Vert v_n\Vert _1=0\), the set \(\{v_n:n\in {\mathbb {N}}\}\) is uniformly integrable and it follows that T is \(\tau (L^\infty ,L^1)\)-nuclear. In view of (2.6) we get
(iii)\(\Rightarrow \)(ii) Assume that (iii) holds. Then by Theorem 2.3T is Bochner representable and
Thus (i)\(\Leftrightarrow \)(ii)\(\Leftrightarrow \)(iii) hold and using (2.7) and (2.8), we get \(|m_T|(\Omega )=\Vert T\Vert _{nuc}\). \(\square \)
Let \(ba_\lambda (\Sigma )\) denote the Banach space of all bounded finitely additive real measures \(\mu \) on \(\Sigma \) such that \(\mu (A)=0\) if \(\lambda (A)=0\), equipped with norm \(\Vert \mu \Vert :=|\mu |(\Omega )\). The Banach dual \((L^\infty )^*\) of \(L^\infty \) can be identified with \(ba_\lambda (\Sigma )\) through the integration mapping \(ba_\lambda (\Sigma )\ni \mu \mapsto F_\mu \in (L^\infty )^*\), where \(F_\mu (u)=\int _\Omega u d\mu \) for all \(u\in L^\infty \) and \(|\mu |(\Omega )=\Vert F_\mu \Vert \).
Recall (see [26, p. 279]) that a linear operator \(T:L^\infty \rightarrow X\) is said to be nuclear if there exist a bounded sequence \((\mu _n)\) in \(ba_\lambda (\Sigma )\), a bounded sequence \((x_n)\) in X and a sequence \((\alpha _n)\in \ell ^1\) such that
It is well known that a bounded linear operator \(T:L^\infty \rightarrow X\) is \((\tau (L^\infty ,L^1),\) \(\Vert \cdot \Vert _X)\)-continuous if and only if its representing measure \(m_T:\Sigma \rightarrow X\) is \(\lambda \)-continuous (see [17, Proposition 3.1]). Hence due to Swartz (see [24, Theorem 1 and Theorem 3]) we have
Theorem 2.6
Assume that \(T:L^\infty \rightarrow X\) is a \((\tau (L^\infty ,L^1),\Vert \cdot \Vert _X)\)-continuous linear operator. Then the following statements are equivalent:
-
(i)
T is nuclear.
-
(ii)
\(m_T\) has the Radon-Nikodym property with respect to \(\lambda \).
Combining Theorem 2.5 and Theorem 2.6, we get
Corollary 2.7
For a linear operator \(T:L^\infty \rightarrow X\) the following statements are equivalent:
-
(i)
T is \(\tau (L^\infty ,L^1)\)-nuclear.
-
(ii)
T is \((\tau (L^\infty ,L^1),\Vert \cdot \Vert _X)\)-continuous and nuclear.
As an application of Theorem 2.5 we show that some natural kernel operators on \(L^\infty \) are \(\tau (L^\infty ,L^1)\)-nuclear.
Assume that \(K_1\) and \(K_2\) are compact Hausdorff spaces and \(k(\cdot ,\cdot )\in C(K_2\times K_1)\). Let \({\mathcal {B}}o\) be the \(\sigma \)-algebra of Borel sets in \(K_1\) and \(\lambda :{\mathcal {B}}o\rightarrow [0,\infty )\) be a countably additive measure.
We will need the following lemma.
Lemma 2.8
For every \(u\in L^\infty (\lambda )\), the mapping \(\Psi _u:K_1\ni s\mapsto u(s)\,k(\cdot ,s)\in C(K_2)\) is continuous.
Proof
Let \(s_0\in K_1\) and \(\varepsilon >0\) be given. Then for every \(t\in K_2\) there exist a neighborhood \(V_t\) of t and a neighborhood \(W_t\) of \(s_0\) such that
Hence there exist \(t_1,\dots ,t_n\in K_2\) such that \(K_2=\bigcup ^n_{i=1} V_{t_i}\). Let us put \(W:=\bigcap ^n_{i=1} W_{t_i}\). For \(t\in K_2\), choose \(i_0\) with \(1\le i_0\le n\) such that \(t\in V_{t_{i_0}}\). Then for \(s\in W\), we have \(|k(t,s)-k(t,s_0|\le \frac{\varepsilon }{\Vert u\Vert _\infty }\). Hence
This means that \(\Psi _u\) is continuous. \(\square \)
Note that the Banach space \(C(K_1,C(K_2))\) can be embedded in the Banach space \(L^1(C(K_2))\) such that with each function from \(C(K_1,C(K_2))\) is associated its \(\lambda \)-equivalence class in \(L^1(C(K_2))\).
In view of Lemma 2.8 we can define a kernel operator \(T_k:L^\infty (\lambda )\rightarrow C(K_2)\) by
For \(t\in K_2\), let \(\delta _t(w):=w(t)\) for all \(w\in C(K_2)\). Then \(\delta _t\in C(K_2)^*\) and according to the Hille’s theorem (see [DU, Theorem 6, p. 47]), we have
Then
where the mapping \(K_1\ni s\mapsto k(\cdot ,s)\in C(K_2)\) belongs to \(L^1(C(K_2))\).
Hence for \(A\in {\mathcal {B}}o\),
Hence as a consequence of Theorem 2.5 we have
Proposition 2.9
The kernel operator \(T_k:L^\infty (\lambda )\rightarrow C(K_2)\) is \(\tau (L^\infty (\lambda ),L^1(\lambda ))\)-nuclear and
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 \(\varphi :[0,\infty )\rightarrow [0,\infty )\) that vanishes only at 0 and \(\varphi (t)/t\) \(\rightarrow 0\) as \(t\rightarrow 0\) and \(\varphi (t)/t\rightarrow \infty \) as \(t\rightarrow \infty \). By \(\varphi ^*\) we denote the complementary function of \(\varphi \) in the sense of Young, that is, \(\varphi ^*(t)=\sup \{ts-\varphi (s):s\ge 0\}\) for \(t\ge 0\). Note that \(\varphi ^{**}=\varphi \).
The corresponding Orlicz space \(L^\varphi \) is an ideal of \(L^0\) defined by
and equipped with the topology \({\mathcal {T}}_\varphi \), defined by two equivalent norms:
called the Luxemburg norm and the Orlicz norm. Then we have:
\(\Vert u\Vert _\varphi \le 1\) if and only if \(\int _\Omega \varphi (|u(\omega )|)\,d\lambda \le 1\),
\(\Vert u_n\Vert _\varphi \rightarrow 0\) if and only if \(\int _\Omega \varphi (\alpha |u_n(\omega )|)\,d\lambda \rightarrow 0\) for every \(\alpha >0\),
\((L^\varphi )'=L^{\varphi ^*}.\)
The Orlicz-Bochner space \(L^\varphi (X)\) is defined by
For \(g\in L^\varphi (X)\), let
Then
Assume that \(\varphi \) is Young function. If \(m:\Sigma \rightarrow X\) is a \(\lambda \)-continuous measure, we write \(|m|_{\varphi ^*}\) instead of \(|m|_{(L^\varphi )'}\).
Lemma 3.1
Assume that \(m:\Sigma \rightarrow X\) is a \(\lambda \)-continuous measure and \(\varphi \) is Young function. Then for \(A\in \Sigma \), we have
Proof
Let \(A\in \Sigma \) and \(\varepsilon >0\) be given. Then there is a \(\Sigma \)-partition \((A_i)^n_{i=1}\) of A such that
It is known that \(\Vert \mathbb {1}_A\Vert _\varphi =\left( \varphi ^{-1}\left( \frac{1}{\lambda (A)} \right) \right) ^{-1}\) (see [RR, Chap. 3.4, Corollary 7, p. 79]). Hence
and
It follows that \(\varphi ^{-1}\left( \frac{1}{\lambda (A)}\right) |m|(A)\le |m|_{\varphi ^*}(A)\), so
\(\square \)
By \(\gamma [{\mathcal {T}}_\varphi ,{\mathcal {T}}_0\big |_{L^\varphi }]\) (in brief, \(\gamma _\varphi \)) we denote the natural mixed topology on \(L^\varphi \) (in the sense of Wiweger), that is, \(\gamma _\varphi \) is the finest linear topology that agrees with \({\mathcal {T}}_0\) on \({\mathcal {T}}_\varphi \)-bounded sets in \(L^\varphi \) (see [1, 25] for more details). Then \(\gamma _\varphi \) is a locally convex-solid Hausdorff topology and \(\gamma _\varphi \) and \({\mathcal {T}}_\varphi \) have the same bounded sets. This means that \((L^\varphi ,\gamma _\varphi )\) is a generalized DF-space (see [20]) and it follows that \((L^\varphi ,\gamma _\varphi )\) is quasinormable (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 \(S:L^\varphi \rightarrow X\) the following statements are equivalent:
-
(i)
S is \((\gamma _\varphi ,\Vert \cdot \Vert _X)\)-continuous and compact.
-
(ii)
S is \(\gamma _\varphi \)-compact, that is, there exists a \(\gamma _\varphi \)-neighborhood V of \(\,0\) in \(L^\varphi \) such that T(V) is relatively norm compact in X.
We say that a Young function \(\varphi \) increases essentially more rapidly than another \(\psi \) (in symbols, \(\psi \ll \varphi \)) if for arbitrary \(c>0\), \(\psi (ct)/\varphi (t)\rightarrow 0\) as \(t\rightarrow 0\) and \(t\rightarrow \infty \). Note that \(L^\varphi \subset E^\psi \) if \(\psi \ll \varphi \).
The following result will be useful (see [16, Theorem 2.1]).
Theorem 3.3
Let \(\varphi \) be a Young function. Then
-
(i)
\(\gamma _\varphi \) is generated by the family of norms \(\{\Vert \cdot \Vert _\psi \big |_{L^\varphi }:\psi \ll \varphi \}\).
-
(ii)
\((L^\varphi ,\gamma _\varphi )^*=\{F_v:v\in E^{\varphi ^*}\}\), where \(F_v(u)=\int _\Omega u(\omega )\, v(\omega )\,d\lambda \) for all \(u\in L^\varphi .\)
According to [15, Corollary 1.6] we have the following identity:
Now we can state the main result in this section.
Theorem 3.4
Assume that a measure \(m:\Sigma \rightarrow X\) has the Radon-Nikodym property with the density \(g\in L^1(X)\). Then there exists a Young function \(\varphi \) such that \(g\in E^{\varphi ^*}(X)\) and the following statements hold:
-
(i)
The operator \(S_g:L^\varphi \rightarrow X\) defined by
$$\begin{aligned} S_g(u):=\int _\Omega u(\omega )\, g(\omega )\, d\lambda \ \ \text{ for } \text{ all } \ u\in L^\varphi \end{aligned}$$is \((\gamma _\varphi ,\Vert \cdot \Vert _X)\)-continuous and compact.
-
(ii)
\(S_g\) is \(\gamma _\varphi \)-compact, that is, there exists a Young function \(\psi \) with \(\psi \ll \varphi \) such that \(\{\int _\Omega u(\omega )g(\omega )\,d\lambda : u\in L^\varphi ,\Vert u\Vert _\psi \le 1\}\) is a relatively norm compact subset of X.
-
(iii)
\(|m|_{\varphi ^*}(\Omega )<\infty \) and \(|m|_{\varphi ^*}(A)=\Vert \mathbb {1}_A g\Vert ^0_{\varphi ^*}\) for every \(A\in \Sigma \).
-
(iv)
\(|m|_{\varphi ^*}\) is \(\lambda \)-continuous, that is, \(|m|_{\varphi ^*}(A_n)\rightarrow 0\) if \(\lambda (A_n)\rightarrow 0\).
Proof
In view of (3.1) there exists a Young function \(\psi \) such that \(g\in E^\psi (X)\), that is, \(\Vert g(\cdot )\Vert _X\in E^\psi \). Hence \(|u| \ \Vert g(\cdot )\Vert _X\in L^1\) for every \(u\in L^{\psi ^*}\). Let us put \(\varphi =\psi ^*\). Then \(\varphi ^*=\psi ^{**}=\psi \) and \((L^\varphi )'=L^{\varphi ^*}=L^\psi \) and \(L^{\psi ^*}=L^\varphi \), and so \(g\in E^{\varphi ^*}(X)\). Then for \(u\in L^\varphi \), we have
where \(\Vert g(\cdot )\Vert _X\in E^{\varphi ^*}\). The inequality shows that \(S_g\) is \((|\sigma |(L^\varphi ,E^{\varphi ^*}),\Vert \cdot \Vert _X)\)-continuous, where \(|\sigma |\) denotes the absolute weak topology. Since \((|\sigma |(L^\varphi ,E^{\varphi ^*})\) is the coarsest locally convex-solid topology on \(L^\varphi \) with dual \(E^{\varphi ^*}\), and by Theorem 3.3(ii) \(\gamma _\varphi \) is such a topology, it follows that \(S_g\) is \((\gamma _\varphi ,\Vert \cdot \Vert _X)\)-continuous.
To show that \(S_g\) is compact, choose a sequence \((f_n)\) of X-valued simple functions on \(\Omega \) such that \(\Vert f_n(\omega )-g(\omega )\Vert _X\rightarrow 0\) and \(\Vert f_n(\omega )\Vert _X\le \Vert g(\omega )\Vert _X\) \(\lambda \)-a.e. and for all \(n\in {\mathbb {N}}\) (see [7, Theorem 6, p. 4]). Hence for every \(\alpha >0\), we have that \(\varphi ^*(\alpha \Vert f_n(\omega )-g(\omega )\Vert _X)\rightarrow 0\) \(\lambda \)-a.e. and \(\varphi ^*(\alpha \Vert f_n(\omega )-g(\omega )\Vert _X) \le \varphi ^*(2\alpha \Vert g(\omega )\Vert _X)\) \(\lambda \)-a.e. and for all \(n\in {\mathbb {N}}\), where \(\int _\Omega \varphi ^*(2\alpha \Vert g(\omega )\Vert _X)\,d\lambda <\infty \). By the Lebesgue dominated convergence theorem, \(\int _\Omega \varphi ^*(\alpha \Vert f_n(\omega )-g(\omega )\Vert _X)\,d\lambda \rightarrow 0\) and this means that \(\Vert f_n-g\Vert ^0_{\varphi ^*}\rightarrow 0\).
For each \(n\in {\mathbb {N}}\), let \(S_n:L^\varphi \rightarrow X\) be a linear operator defined by
Note that the range of each \(S_n\) is contained in the span of finite set of values of \(f_n\). Therefore \(S_n\) is compact and by the Hölder’s inequality for each \(u\in L^\varphi \), we have
It follows that \(\Vert S_n-S_g\Vert \rightarrow 0\), so \(S_g\) is a compact operator.
(ii) In view of (i) and Proposition 3.2\(S_g\) is \(\gamma _\varphi \)-compact. Using Theorem 3.3 we obtain that (ii) holds.
(iii) Let \(s=\sum ^k_{i=1} c_i\mathbb {1}_{A_i}\in {\mathcal {S}}(\Sigma )\) and \(\Vert s\Vert _\varphi \le 1\). Then
and hence \(|m|_{\varphi ^*}(\Omega )\le \Vert g\Vert ^0_{\varphi ^*}\). Note that \(m=m_{S_g}\). Hence using Lemma 2.1 we have that \(|m_{\varphi ^*}|(A)=\Vert \mathbb {1}_A g\Vert ^0_{\varphi ^*}\) for \(A\in \Sigma \).
(iv) This follows from (iii) because \(g\in E^{\varphi ^*}(X)\). \(\square \)
Let \(i_\infty :L^\infty \rightarrow L^\varphi \) denotes the inclusion map. Note that \(i_\infty \) is \((\sigma (L^\infty ,L^1),\) \(\sigma (L^\varphi ,L^{\varphi ^*}))\)-continuous, and it follows that \(i_\infty \) is \((\tau (L^\infty ,L^1),\tau (L^\varphi ,L^{\varphi ^*}))\)-continuous (see [8, Theorem 8.6.1]). Since \(\gamma _\varphi \subset \tau (L^\varphi ,L^{\varphi ^*})\), we obtain that \(i_\infty \) is \((\tau (L^\infty ,L^1),\gamma _\varphi )\)-continuous.
As a consequence of Theorem 3.4 and Theorem 2.5, we can show that every \(\tau (L^\infty ,L^1)\)-nuclear operator \(T:L^\infty \rightarrow X\) admits a factorization through some Orlicz space \(L^\varphi \).
Corollary 3.5
Assume that \(T:L^\infty \rightarrow X\) is a \(\tau (L^\infty ,L^1)\)-nuclear operator. Then there exists a Young function \(\varphi \) such that \(T=S\circ i_\infty \), where
-
(i)
\(S:L^\varphi \rightarrow X\) is a Bochner representable and \(\gamma _\varphi \)-compact linear operator.
-
(ii)
\(|m_S|_{\varphi ^*}(A_n)\rightarrow 0\) if \(\lambda (A_n)\rightarrow 0\).
Proof
-
(i)
In view of Theorem 2.5 the representing measure \(m_T:\Sigma \rightarrow X\) has the Radon-Nikodym property with respect to \(\lambda \), that is, \(m_T=g\lambda \), where \(g\in L^1(X)\). Hence according to Theorem 3.4 there exists a Young function \(\varphi \) such that \(g\in E^{\varphi ^*}(X)\) and an operator \(S:L^\varphi \rightarrow X\) defined by
$$\begin{aligned} S(u):=\int _\Omega u(\omega )\,g(\omega )\,d\lambda \ \text{ for } \text{ all } \ u\in L^\varphi , \end{aligned}$$is \(\gamma _\varphi \)-compact. Note that for \(u\in L^\infty \), \(T(u)=\int _\Omega u\,dm_T=\int _\Omega u(\omega )g(\omega )\,d\lambda =S(u)\).
-
(ii)
Since \(m_S(A)=\int _A g(\omega )\,d\lambda \) for all \(A\in \Sigma \), using Theorem 3.4 we have that \(|m_S|_{\varphi ^*}(A_n)\rightarrow 0\) if \(\lambda (A_n)\rightarrow 0\).
\(\square \)
References
Cooper, J.B.: The strict topology and spaces with mixed topologies. Proc. Amer. Math. Soc. 30(3), 583–592 (1971)
Cristescu, R.: Topological Vector Spaces. Noordhoff Int. Publ, Leyden (1977)
Diestel, J.: Sequences and Series in Banach Spaces. Graduate Texts in Math, vol. 92. Springer-Verlag, New York (1984)
Diestel, J., Uhl, J.J.: Vector Measures, Math. Surveys, vol. 15. American Mathematical Society, Providence, RI (1977)
Diestel, J., Fourie, J., Swart, J.: The Metric Theory of Tensor Products. Grothendieck’s resume revisited. American Mathematical Society, Providence (2008)
Dinculeanu, N.: Vector Measures. Pergamon Press, New York (1967)
Dinculeanu, N.: Vector Integration and Stochastic Integration in Banach Spaces. A Wiley-Interscience Publication, New York (2000)
Edwards, R.E.: Functional Analysis. Theory and Applications. Holt, Rinehart and Winston, New York (1965)
Fremlin, D.H.: Topological Riesz Spaces and Measure Theory. Cambridge University Press, London-New York (1974)
Grothendieck, A.: Sur les espaces (F) et (DF). Summa Brasil. Math. 3, 57–123 (1954)
Grothendieck, A.: Produits tensoriels topologiques et espaces nucléaires. Mem, vol. 16. American Mathematical Society, Providence (1955)
Jarchow, H.: Locally Convex Spaces. Teubner, Stuttgart (1981)
Kantorovitch, L.V., Akilov, G.P.: Functional Analysis. Pergamon Press, Oxford-Elmsford, New York (1982)
Lin, P.-K.: Köthe-Bochner Function Spaces. Birkhaüser, Boston (2003)
Nowak, M.: Some equalities among Orlicz spaces II. Bull. Pol. Acad. Sci. Math. 34(11–12), 675–687 (1986)
Nowak, M.: Compactness of Bochner representable operators on Orlicz spaces. Positivity 13, 193–199 (2009)
Nowak, M.: Absolutely continuous operators on function spaces and vector measures. Positivity 17, 525–533 (2013)
Pietsch, A.: Nuclear Locally Convex Spaces, Ergebnisse der Mathematik und ihrer Grenzebiete, vol. 66. Springer, Berlin (1972)
Rao, M.M., Ren, Z.D.: Theory of Orlicz Spaces. Marcel Dekker, New York (1991)
Ruess, W.: [Weakly] compact operators and DF-spaces. Pacific J. Math. 98, 419–441 (1982)
Ruston, A.F.: On the Fredholm theory of integral equations for operators belonging to the trace class for a general Banach space. Proc. London Math. Soc. 55(2), 109–124 (1951)
Ryan, R.: Introduction to the Tensor Products of Banach Spaces. Springer Monographs in Math. Springer-Verlag, London (2002)
Schaefer, H.H.: Topological Vector Spaces. Springer, New York, Heidelberg, Berlin (1971)
Swartz, C.: An operator characterization of vector measures which have Radon-Nikodym derivatives. Math. Ann. 202, 77–84 (1973)
Wiweger, A.: Linear spaces with mixed topology. Studia Math. 20, 47–68 (1961)
Yosida, K.: Functional Analysis, 4th edn. Springer-Verlag, Berlin, Heidelberg, New York (1974)
Zaanen, A.C.: Riesz Spaces II. North-Holland Publ. Comp, Amsterdam (1983)
Acknowledgements
The author wishes to express his thanks to the referee for remarks and suggestions leading to an improvement of the paper.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
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/.
About this article
Cite this article
Nowak, M. Nuclear operators on Banach function spaces. Positivity 25, 801–818 (2021). https://doi.org/10.1007/s11117-020-00787-1
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11117-020-00787-1
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