Lattice trace operators (Q472536)

A compact operator \(T\) on a separable Hilbert space \(\mathcal{H}\) is a trace class operator if \[ \|T\|_1 = \sum\limits_{j=1}^{\infty}\lambda_j(T) < \infty, \] where \(\lambda_j(T)\), \(j \in \mathbb{N}\) are the singular numbers of \(T\). Equivalently, \(T\) is a trace class operator if \(\sum\limits_{j=1}^{\infty}|\langle Th_j,h_j\rangle| < \infty\) for any orthonormal basis \(\{h_j\}\) of \(\mathcal{H}\) and the trace \(\operatorname{tr}(T)\) of \(T\) is defined as \[ \operatorname{tr}(T)=\sum\limits_{j=1}^{\infty}\langle Th_j,h_j\rangle. \] Let \({\mathcal{T}}_1(\mathcal{H})\) denote the collection of trace class operators on \(\mathcal{H}\). Then \({\mathcal{T}}_1(\mathcal{H})\) is an operator ideal of \(\mathcal{H}\) and also a Banach space with norm \(\|\cdot\|_1\). For a \(\sigma\)-finite measure space \((\Sigma,{\mathcal{B}},\mu)\), denote by \({\mathcal{L}}^o(\mu)\) the space of all \(\mu\)-equivalence classes of Borel measurable scalar functions equipped with the topology of convergence in \(\mu\)-measure over sets of finite measure and vector operations are defined pointwise \(\mu\)-almost everywhere. A~Banach space \(X\) that is a subspace of \({\mathcal{L}}^o(\mu)\) is a Banach function space based on \((\sum,{\mathcal{B}},\mu)\) if {\parindent=8mm \begin{itemize}\item[(i)] \(X\) is an order ideal of \({\mathcal{L}}^o(\mu)\), i.e., if \(g \in X, f \in {\mathcal{L}}^o(\mu)\) and \(|f| \leq |g|\) \(\mu\)-a.e., then \(f \in X\) and \item[(ii)] if \(f,g \in X\), and \(|f| \leq |g|\) \(\mu\)-a.e., then \(||f||_X \leq ||g||_X\). \end{itemize}} It is necessarily Dedekind complete (i.e, every bounded set has a sup and inf). A~continuous operator \(T : X \rightarrow X\) is said to be positive if \(T : X_+ \to X_+\), where \(X_+\) is the set of all \(f \in X\) with \(f \geq 0\) \( \mu\)-a.e. Let \({\mathcal{L}}_+(X)\) denote the set of all positive continuous linear operators on \(X\). Then \(T\) is said to be regular if the real and imaginary parts of \(T\) can be expressed as a difference of positive operators. The modulus \(|T|\) of a regular operator \(T\) is defined as \(|T|f = \sup\limits_{|g| \leq f}|Tg|\), \(f \in X_+\). The collection of all regular operators is written as \({\mathcal{L}}_r(X)\), which is a Banach lattice under the norm \(T\mapsto\| | T | \|\), \(T \in {\mathcal{L}}_r(X)\). \(T : X \rightarrow X\) has an integral kernel \(k\) if \(k : \sum\times \sum \rightarrow \mathbb{C}\) is a Borel-measurable function such that \(T=T_k\), which is defined as \[ (T_kf)(x) = \int\limits_{\sum}k(x,y)f(y)\,d\mu(y) \text{ for } \mu \text{-almost all } x \in \Sigma,\;f \in X. \] \(T\) is called an absolute integral operator if it has an integral kernel \(k\) for which \(T_{|k|}\) is a bounded linear operator on \({\mathcal{L}}^2(\mu)\). In [\textit{P. Meyer-Nieberg}, Banach lattices. Berlin etc.: Springer-Verlag (1991; Zbl 0743.46015)], it was shown that \(|T_k|=T_{|k|}\). For \(T \in {\mathcal{L}}(X)\) having an integral kernel \(k=\sum\limits_{j=1}^nf_j\chi_{A_j}\), that is, \(X\)-valued simple functions with \(\mu(A_j) < \infty\), it is natural to view \[ \int\limits_{\sum}\langle T,dm\rangle=\sum_{j=1}^n\int_{A_j}f_j\,d\mu=\int_{\sum}k(x,x)\,d\mu(x) \] as a bilinear integral. The author applies martingale regularization (and the martingale convergence theorem) to extend the above bilinear integral to regular operators and the class \({\mathcal{C}}_1(X)\) of absolute integral operators with \(\int\limits_{\sum}\langle |T|,dm\rangle < \infty\). The author's first result asserts that \({\mathcal{C}}_1(X)\) is a lattice ideal in \({\mathcal{L}}_r(X)\) and it is a Dedekind complete Banach Lattice with the norm \[ T \rightarrow \|T\|+\int_{\sum}\langle |T|,dm\rangle,\quad T \in {\mathcal{C}}_1(X). \] Subsequently, the author gives a sufficient condition for traceability of absolute integral operators. For positive kernel operators \(T_j : X\rightarrow X\), \(j=1,2\), he proves the following trace property: \( \int\limits_{\sum}\langle T_1T_2,dm\rangle =\int\limits_{\sum}\langle T_2T_1,dm\rangle\), which holds for extended real numbers. The author also proves a bilinear version of the Fubini-Tonelli theorem.