Operator-space Grothendieck inequalities for noncommutative \(L_p\)-spaces (Q817954): Difference between revisions

\textit{G.\,Pisier} and \textit{D.\,Shlyakhtenko} [Invent.\ Math.\ 150, No.\,1, 185--217 (2002; Zbl 1033.46044)] obtained the following operator space version of the classical Grothendieck inequality: if \(E,F \subset B(H)\) are exact operator spaces, and \(u:E\times F\to \mathbb{C}\) is a bilinear form, then \(u\) is jointly completely bounded if and only if there exist a constant \(K\) and states \(f_i, g_i\, (i=1,2)\) on \(B(H)\) such that \[ | u(a,b)| \leq K[(f_1(aa^*)g_1(b^*b))^{1/2} +(f_2(a^*a)g_2(bb^*))^{1/2}], \quad a\in E,\;b\in F. \] The author presents several \(L^p\) versions of this inequality. The starting point is the following little Grothendieck inequality of \textit{B.\,Maurey} [Astérisque 11, 1--163 (1974; Zbl 0278.46028)]: if \(u:L_p(\Omega,\mu)\to H\) is a bounded map \((2<p\leq\infty)\), then there is a positive linear functional \(f\in (L_{p/2}(\Omega,\mu))^*\) of norm one such that \[ \| u(a)\| \leq K\| u\| [f(| a| ^2)]^{1/2},\quad a \in L_p(\Omega,\mu), \] where \(K\) is a universal constant. This result implies the following inequality for a bilinear form \(u:L_p(\Omega,\mu)\times L_q(\Omega,\mu)\to\mathbb{C}\) with \(2<p,q\leq\infty\), for some positive linear functionals \(f\in (L_{p/2}(\Omega,\mu))^*\), \(g\in (L_{q/2}(\Omega,\mu))^*\): \[ | u(a,b)| \leq [f(| a| ^2)g(| b| ^2)]^{1/2},\quad a\in L_p(\Omega,\mu),\,b\in L_q(\Omega,\mu). \] The following deep result from the paper generalizes the above results to operator spaces: Let \(M\) be a von Neumann algebra, and let \(2<p<\infty\). Let \(E,F\subset L_p(M)\) be two closed subspaces, and let \(u: E\times F\to \mathbb{C}\) be a bilinear form. Then \(u\) is jointly completely bounded if and only if there exist positive \(f_1, f_2, g_1, g_2 \in L_{p/2}(M)^*\) such that \[ | u(a,b)| \leq f_1 (a^* a) g_1(b b^*) + f_2 (a a^*) g_2(b^* b), \quad a \in E,\;b \in F. \] From this, one can easily deduce the operator space version of a little Grothendieck inequality, generalizing the semifinite extension of the inequality due to \textit{F.\,Lust--Piquard} [Isr.\ J.\ Math.\ 79, No.\,2--3, 331--365 (1992; Zbl 0789.47019)]. The author proves a stronger result: Let \(E\subset L_p(M)\) be a subspace with \(2 < p < \infty\), and let \(H\) be a Hilbert space. Let \(\theta \in [0,1]\), \(1/q = (1-\theta)/p + \theta/p^\prime\), and let \(u:E \to H_q^c\) be a linear map. Then \(u\) is completely bounded if and only if there exist positive \(f, g \in L_{p/2}(M)^*\) such that \[ \| u(a)\| \leq (f (a^* a))^{(1-\theta)/2} (g(a a^*))^{\theta/2}, \quad a \in E. \] For both results, other equivalent conditions are also given. An interesting ingredient of the proofs is the introduction of an \(L^p\) version of the Haagerup tensor product.