The trace formula in Banach spaces (Q476508)

Lidskii's Theorem from 1959, which was already known to \textit{A.~Grothendieck} [Produits tensoriels topologiques et espaces nucléaires. Mem.\ Am.\ Math.\ Soc. 16 (1955; Zbl 0064.35501); Chap.~II, p.~13], says that the trace of every nuclear operator on a (complex) Hilbert space coincides with the sum of its eigenvalues (counted according to their multiplicities). In the same work, Grothendieck proved the following facts. (1) All nuclear operators on a Banach space \(X\) admit a well-defined trace if and only if \(X\) has the approximation property. (2) The eigenvalue sequence \((\lambda_n(T))\) of any nuclear operator \(T\) on a Banach space is square summable. The authors refer to a Banach space \(X\) (with the approximation property) as a Lidskii space if the formula \({\mathrm{ trace}} (T) = \sum_{n=1}^\infty \lambda_n(T)\) holds for all nuclear operators on \(X\) for which \((\lambda_n(T))\) is absolutely summable. Their main result says that \(X\) is a Lidskii space whenever \(\liminf_{n\to\infty}\Gamma_n(X) < \infty\). Here the \(\Gamma_n(X)\) are the Grothendieck numbers of a Banach space \(X\) defined by \[ \Gamma_n(X):= \sup \bigg\{ \big| \det\big(\langle{x_h^{\hphantom\ast},{x_k^\ast}\rangle}\big) \big|^{1/n} : {x_1^{\hphantom\ast},\dots,x_n^{\hphantom\ast} \in B_X},\, {\;x_1^\ast,\dots,x_n^\ast \in B_{X^\ast}} \bigg\}, \] where \(B_X\) and \(B_{X^\ast}\) denote the unit balls of \(X\) and \(X^\ast\), respectively; see [\textit{A.~Pietsch}, History of Banach Spaces and Linear Operators. Birkhäuser, Boston (2007; Zbl 1121.46002); p.~324] for further references. Direct \(\ell_2\)-sums \(\big(\sum_{k=1}^\infty \bigoplus \ell_{p_k}^{n_k} \big)_2\) such that \(\delta_k := |p_k-2| \searrow 0\), \(n_{k+1} \geq 2 n_k\), and \(n_k^{\delta_{k+1}} \leq c\) are Lidskii spaces, since they satisfy the condition \(\liminf_{n\to\infty}\Gamma_n(X) < \infty\). Note that Lidskii spaces are very close to Hilbert spaces. On the other hand, if \(\lim_{k\to\infty} \frac {\log n_{k+1}}{\log n_k} =\infty\), then \(\big(\sum_{k=1}^\infty \bigoplus \ell_{p_k}^{n_k} \big)_2\) fails to be isomorphic to a Hilbert space.