Numerical radii for tensor products of operators (Q2445446)

For an operator \(A\) acting on a Hilbert space \(H\), let \(\sigma(A)\) and \(W(A)\) denote, respectively, the spectrum and the numerical range of \(A\) (that is, the set of values \((Ax,x)\) when \(x\) varies over the unit sphere of \(H\)). Let also \(w(A)\) and \(\rho(A)\) stand for the numerical radius and the spectral radius of \(A\): \[ w(A)=\sup\{ |z|: z\in W(A)\}, \quad \rho(A)=\sup\{ |z|: z\in \sigma(A)\}. \] As was shown in [\textit{E. S. W. Shiu}, Tohoku Math. J. (2) 30, 257--262 (1978; Zbl 0383.47004)], the numerical range of the tensor product \(A\otimes B\) contains the product of the numerical ranges \[ W(A)\cdot W(B)=\{ z_1z_2: z_1\in W(A),\quad z_2\in W(B)\}, \] and if one of the operators \(A,B\) is normal, then the closure \(\overline{W(A\otimes B)}\) coincides with the convex hull of the product \(\overline{W(A)}\cdot\overline{W(B)}\). From these results it is of course clear that \(w(A\otimes B)\geq w(A)w(B)\), and that the equality holds if \(A\) or \(B\) is normal. The main results of the paper under review are as follows: (i) If \(w(A\otimes B)=w(A)w(B)\), then \(w(A)=\rho(A)\) or \(w(B)=\rho(B)\). The converse is not true. (ii) The above stated Shiu's description of \(\overline{W(A\otimes B)}\) (and thus the formula for \(w(A\otimes B)\) holds even if one of the operators \(A,B\) is just hyponormal (that is, its self-commutator is positive semi-definite).