The little Grothendieck theorem and Khintchine inequalities for symmetric spaces of measurable operators (Q880103): Difference between revisions

The paper under review can be viewed as a common generalization of two extensions of the classical little Grothendieck theorem: for bounded linear operators defined on \(2\)-convex Banach lattices due to \textit{B. Maurey} [Appl. radonif., Géom. Espaces de Banach, Exposé XII (1974; Zbl 0295.47029)] and for bounded linear operators defined on \(C^*\)-algebras due to \textit{G. Pisier} [J. Funct. Anal. 29, 397--415 (1978; Zbl 0388.46043)]. Recall that if \(\mathcal{M}\) is a semifinite von Neumann algebra equipped with a semifinite normal faithful trace \(\tau\) and \(E\) is a rearrangement invariant Banach function on \((0, \infty)\), then \(E(\mathcal{M}, \tau)\) denotes the symmetric space of \(\tau\)-measurable operators, the (non commutative) analogue of the function space \(E\). The main result of the paper is the following: Assume that \(E\) is \(2\)-convex with constant \(1\). Let \(H\) be a Hilbert space. Then for any bounded linear map \(T: E(\mathcal{M}) \to H\), there exists a positive norm one functional \(f \in E_{(2)}(\mathcal{M})^*\) (where \( E_{(2)}\) is the 2-convexification of \(E\)) such that \[ \|T(x)\|^2 \leq K^2 \|T\|^2 f(x^*x +xx^*), \quad \forall x \in E(\mathcal{M}). \] The authors deduce from this result generalizations of non-commutative Khintchine inequalities to symmetric spaces of measurable operators (in the next statement, \((\varepsilon_k)\) denotes a Rademacher sequence on a probability space): (i) If \(E\) is \(2\)-concave with constant \(1\), then for every sequence \((x_k) \subset E(\mathcal{M})\), \[ \begin{aligned} \left( \mathbb{E}\left\|\sum_k \varepsilon_k x_k\right\|^2\right)^{1/2} &\leq \inf\left\{\left\| \left(\sum_k a_k^*a_k \right)^{1/2}\right\| , \left\| \left(\sum_k b_k b_k^* \right)^{1/2}\right\|\right\} \\ &\leq K \left( \mathbb{E}\left\|\sum_k \varepsilon_k x_k\right\|^2\right)^{1/2}, \end{aligned} \] where the infimum is taken over all decompositions \(x_k=a_k +b_k\) in \(E(\mathcal{M})\). (ii) If \(E\) is 2-convex and \(q\)-concave with constant \(1\) for some \(q<\infty\), then for every finite sequence \((x_k) \subset E(\mathcal{M})\), \[ \begin{aligned} K_q^{-1}\left( \mathbb{E}\left\|\sum_k \varepsilon_k x_k\right\|^2\right)^{1/2} &\leq \max\left\{\left\| \left(\sum_k x_k^*x_k \right)^{1/2}\right\| , \left\| \left(\sum_k x_k x_k^* \right)^{1/2}\right\|\right\} \\ &\leq \left( \mathbb{E}\left\|\sum_k \varepsilon_k x_k\right\|^2\right)^{1/2}, \end{aligned} \] where \(K_q\) depends only on \(q\) and \(K_q \leq Kq\). The authors also characterize Schur multipliers from a \(2\)-convex unitary ideal into a \(2\)-concave unitary ideals as another application of the little Grothendieck theorem stated above.