Similarity problems and length (Q5935716)

This paper is a survey on the author's recent results on \textit{R. V. Kadison}'s conjecture in Am. J. Math. 77, 600-620 (1955; Zbl 0064.36605)]. The problem is the following: Let \(A\) be any unital \(C^*\)-algebra, and \(u: A \mapsto B(H)\) (\(H\): Hilbert) any bounded unital homomorphism; does it exists an invertible element \(T \in B(H)\) such that: \(u_T: a \mapsto T^{-1}u(a)T\) is a \(\star \)-homomorphism,\ i.e. \(u\) is similar to a \(\star\)-homomorphism? This problem has been solved in a certain number of cases but it stands still for the general situation. \textit{U. Haagerup} [Ann. Math., II. Ser. 118, 215-240 (1983; Zbl 0543.46033)], proved that \(u\) is similar to a \(\star\)-homorphism if and only if \(u\) is completely bounded (cb), which by definition means that the sequence \(( u \otimes I_n)_{n \in {\mathbb N}} \) is norm bounded where \(I_n\) is the identity map of \(M_n(\mathbb C)\), and \(\|u\|_{cb}\) \((= \{ \sup \|u \otimes I_n \|\)) is also equal to \(\inf \{\|T^{-1} \|.\|T \|, u_T\) is a \(\star\)-homomorphism\}. An extension of this last theorem has been given by \textit{V. I. Paulsen} in J. Funct. Anal. 55, 1-17 (1984; Zbl 0557.46035), when \(A\) is any unital operator algebra (i.e. any closed unital subalgebra of some \(B(\mathcal H)\)). So G. Pisier gives a generalization, namely (SP), of Kadison's problem: For any operator algebra \(A\) and any bounded homomorphism \(u: A \mapsto B(H)\) (\(H\) an arbitrary Hilbert space), is \(u\) cb? This last conjecture is false as the author gives a counter-example using the disc algebra \(A(\mathbb D)\), which is the completion of the set of numerical polynomials \(P\) for the norm \(\|P \|_\infty = \sup \{|P(z)|\;|z \in \partial(\mathbb D) \}\). G. Pisier gives a tight relation between (SP) and a notion of length for operator algebras. By theorem 5, a unital operator algebra satisfies (SP) if and only if there exist \(d \in \mathbb N,K \in \mathbb R^+\), such that \(\forall n \in \mathbb N\), \(\forall x \in M_n(A)\) (the \(n \times n\) matrices with entries in \(A\)), there exist a natural number \(N(n,x)\), and scalar matrices: \(\alpha_0 \in M_{nN}(\mathbb C)\) (rectangular matrices), \( \alpha_1,\dots,\alpha_{d-1} \in M_n(\mathbb C),\;\alpha_d \in M_{Nn}(\mathbb C)\), together with diagonal matrices \(D_1,\dots,D_d\) in \(M_N(A)\) satisfying: \[ x = \alpha_0D_1\alpha_1D_1\dots D_d\alpha_d,\tag{1} \] \[ \prod_0^d \|\alpha_i \|\prod_1^d \|D_i \|\leq K \|x \|.\tag{2} \] Moreover, If one denotes by \(l(A)\) the smallest integer \(d\) for which this holds, one has: \(l(A) = \inf \{\alpha \mid \exists C, \forall u\) unital homomorphism \(A \mapsto B(H),\) \(\|u\|_{cb} \leq C\|u\|^{\alpha }\},\) and this last extremum is actually a minimum. Then giving explicit values for their length, the author recovers the known examples for which (SP) is true. Using the length condition, Proposition 9 says that initial Kadison's conjecture is always true if and only if there exists a universal constant \(d_0\) such that any unital \(C^*\)-algebra has a length less or equal to \(d_0\). To end the survey, the author proves that this last universal condition is not true for its generalization to (non-self-adjoint) operator algebras. In fact, for any natural number \(d\) there exists an operator algebra \(A_d\) such that \(l(A_d) = d\).