Approximation of operators in dual spaces by adjoint operators (Q393020): Difference between revisions

Let \(X\) and \(Y\) be Banach spaces, and let \(L(X,Y)\) denote the Banach space of bounded linear operators from \(X\) to \(Y\). This paper was inspired by a discussion on \textit{Math Overflow} in April 2010 [cf.\ {\url{http://mathoverflow.net/questions/22799}}] on the problem of approximating operators \(U\in L(Y^*, X^*)\) by adjoint operators \(S^*\), \(S\in L(X,Y)\). The author presents an overview on related questions and results, obtained mainly by himself already in the 1980s and 1990s. He also develops them further using the following concept. Let \(n\in\mathbb{N}\) and \(C\geq 1\). Say that \(X\) has the \((C,n)\)-metric approximation property if, for every \(\varepsilon >0\) and every \(n\)-dimensional subspace \(F\) of \(X\), there exists a finite-rank operator \(T\in L(X,X)\) such that \(\| T\|\leq C\) and \[ \|(T-Id)|_F\|<\varepsilon . \] Assume now that \(X\) has the \((C,n)\)-metric approximation property. The main theorem (cf. Theorem 3.1) asserts that for every operator \(U\in L(Y^*, X^*)\) there exists a number \(\gamma_n(U)>0\) such that \[ | \sum^n_{j=1}\langle U y^*_j, x_j\rangle\lambda_j|\leq \gamma_n(U)\sup\{|\sum^n_{j=1}\langle y^*_j, Sx_j\rangle\lambda_j| : S\in L(X,Y), \| S\|\leq 1\} \] for all \(x_j\in X\), \(\| x_j\|\leq 1\), and \(y^*_j\in Y^*\), \(\| y^*_j\|\leq 1\), and all \(\lambda_j\geq 0\) with \(\sum^n_{j=1}\lambda_j=1\). Moreover, \[ \| U\|\leq \inf\gamma_n(U)\leq C\| U\|, \] where the infimum is taken over all choices of \(x_j, y^*_j\), and \(\lambda_j\) as above. An interesting immediate corollary is that \[ \| U\|=\inf\gamma_n(U) \] for all \(n\in\mathbb{N}\), whenever \(X\) has the classical metric approximation property (which is the same as the \((1,n)\)-metric approximation property for all \(n\in\mathbb{N}\)). Based on his earlier work [Proc. Est. Acad. Sci., Phys. Math. 45, No. 2--3, 226--233 (1996; Zbl 0865.47011)], the author shows that the estimates in the main theorem cannot be improved. Also, other interesting relevant examples and counter-examples are presented. Reviewer's remark: The reader should be wary of mal-formulation of results. This paper was translated from the Russian journal ``Problems in Mathematical Analysis''. The author kindly informed the reviewer that the editors of the latter journal had modified his originally correct text without asking his permission.