Self-duality for the Haagerup tensor product and Hilbert space factorizations (Q1178603)

An operator space is defined to be a vector space \(V\) with a system of norms, for each matrix space \(M_ n(V)\subset B(H^{\otimes n})\), satisfying \(\| v+w\|=\max\{\| v\|,\| w\|\}\), \(\|\alpha v \beta\| \leq \|\alpha\| \| v\| \|\beta\|\), where \(v\in M_ m(V)\), \(w\in M_ n(W)\), \(\alpha\in M_{p,m}\), \(\beta\in M_{m,p}\). Defining \(\varphi_ n:M_ n(V)\to M_ n(W)\) by \(\varphi_ n\{[v_{i,j}]\}=[\varphi(v_{i,j})]\), a linear map \(\varphi: V\to W\) is called a complete isometry if all the \(\varphi_ n\) are isometries and is called completely bounded if \(\|\varphi\|_{cb}=\sup_ n\varphi_ n<+\infty\). One may identify completely isometric operator spaces. An operator space is known to be completely isometric to a sub vector-space of \(B(H)\), the space of all bounded linear operators on some Hilbert space \(H\) [\textit{Z. J. Ruan}, J. Funct. Anal. 76, No. 1, 217-230 (1988; Zbl 0646.46055)]; all vector spaces are assumed to be embedded in \(C^*\)-algebras. The spaces \(M_{m,n}(V)\) of rectangular matrices, and in particular column and row matrices, over \(V\) are given operator norms, uniquely, by embedding them in square matrices. For operator spaces, analogously as for Banach spaces, there are projective \((\widehat\otimes)\) and spatial \((\check\otimes)\) tensor products preserving, respectively, quotients and completely isometric injections. The system of Haagerup operator norms for \(V\otimes W\) is defined for \(u\in M_ n(V\otimes W)\) by \[ \| u\|_ h=\inf\{\| v\| \| w\|:\;u=v\odot w,\;v\in M_{n,p}(V),\;w\in M_{p,n}(W)\}. \] For the Haagerup system, there is a natural dual operator norm structure on \(V^*\otimes W^*\), the tensor product of the Banach duals, and the authors show that the Haagerup norm is self-dual, i.e. the dual of the Haagerup operator norm of \(V\otimes W\) is the Haagerup dual operator norm on \(V^*\otimes W^*\). Denote by \(H_ c\) the column operator space corresponding to \(H\simeq B(C,H)\) regarding the vectors of \(B(C,H)\) as column matrices. When \(H_ c\) is a Hilbert space the authors call it a Hilbert column space. The authors show that the identity map \(B(H,K)\to CB(H_ c,K_ c)\) (the operator space of completely bounded operators), is a complete isometry and, using the identification \(M_{m,n}(V\otimes W)=M_{m,1}(V)\otimes M_{1,n}(W)\) they find a complete isometry of the Haagerup tensor product \(V{\overset {h}\otimes} H_ c\) with \(V\widehat\otimes H_ c\). By symmetry, \(H_ c{\overset {h}\otimes}V\simeq H_ c\check\otimes V\), and so on. Denote by \(\Gamma_ 2(V,W)\) the normed linear space of linear mappings \(\varphi: V\to W\) which have completely bounded factorizations \(\varphi=\sigma\cdot\tau\) through Hilbert column spaces \(H_ c\), i.e. \(\sigma: V\to H_ c\), \(\tau: H_ c\to W\), where the norm \(\gamma_ 2\) is defined by \(\gamma_ 2(\varphi)=\inf\{\|\sigma\|_{cb}\), \(\|\tau\|_{cb}\) over all possible factorizations\(\}\). Let \(B: W\otimes V\to\mathbb{C}\) be defined, on linear maps \(\psi: V\to W^*\), by \(\langle B(\psi),w\otimes v\rangle=\langle\psi(v),w\rangle\). The authors show that \(B\) determines a complete isometry \(\Gamma_ 2(V,W^*)\simeq(W{\overset {h}\otimes}V)^*\). The authors then try to associate \(\Gamma_ 2\) with the space of 2-summing linear maps of Banach spaces.