Crouzeix's conjecture for classes of matrices (Q6565844)

Crouzeix's conjecture (referred to as ``Conjecture 1.1'' in the paper) is an open problem in matrix analysis. It was initiated by \textit{M. Crouzeix} in 2004 [Integral Equations Oper. Theory 48, No. 4, 461--477 (2004; Zbl 1069.47004)]). The problem is as follows:\N\N{Crouzeix's Conjecture.} Let \(m, n \in \mathbb{N}\), \(A \in \mathbb{M}_n(\mathbb{C})\) and \(p(\mathbf{w}) : = \sum_{i=0}^m a_i {\mathbf{w}}^i \in \mathbb{C}[\mathbf{w}]\) be an arbitrary complex one-variable polynomial. Then\N\[\N\Vert p(A)\Vert_2 \leq C_{W(A)}\sup_{z \in W(A)}\vert p(z)\vert = C_{W(A)}\sup_{\Vert x \Vert_2 = 1}\vert p(x^\ast Ax)\vert\,,\N\]\Nwhere \(C_{W(A)} = 2\)\,.\N\NHere, \(W(A) : = \{x^\ast Ax : x \in \mathbb{C}^n, \Vert x \Vert_2 = 1\}\) denotes the numerical range of the matrix \(A\) and \(p(A) : = \sum_{i=0}^m a_i A^i\) (polynomial functional calculus). \(\Vert A \Vert_2\) denotes the standard operator norm of \(A\), induced by the Euclidean norm in the Hilbert space \(l_2^n \equiv (\mathbb{C}^n, \Vert \cdot \Vert_2)\) (coinciding with the largest singular value of the matrix \(A\)), i.e., \(\Vert A \Vert_2 \equiv \Vert A \Vert_{2,2}:= \sup_{\Vert x \Vert_2 = 1}\Vert Ax\Vert_2 = (\lambda_{\text{max}}(A^\ast A))^{1/2} \equiv \sigma_{\text{max}}(A)\).\N\NIn [J. Funct. Anal. 244, No. 2, 668--690 (2007; Zbl 1116.47004)], \textit{M. Crouzeix} showed that \(C_{W(A)} = 11.08\). Thirteen years later, the universal constant \(C_{W(A)}\) was strongly improved to \(C_{W(A)} \leq 1 + \sqrt{2} = \exp(\sinh^{-1}(1)) \approx 2.414\) (see [\textit{M. Crouzeix} and \textit{C. Palencia}, SIAM J. Matrix Anal. Appl. 38, No. 2, 649--655 (2017; Zbl 1368.47006)]).\N\NThe present paper is completely built on a very important partial positive result, also contained in the {M. Crouzeix}'s contribution of 2004 [loc. cit.]. It states that Crouzeix's conjecture is satisfied for all matrices \(A \in \mathbb{M}_2(\mathbb{C})\) (see Theorem 4.1 in [loc. cit.]). Further examples of suitable matrices are revealed in detail in [\textit{K. Li}, Crouzeix's conjecture and beyond for special classes of matrices, Ph.D. thesis, University of Washington, ProQuest LLC, Ann Arbor, MI (2021), \url{https://digital.lib.washington.edu/researchworks/bitstream/handle/1773/47918/Li_washington_0250E_23475.pdf}], cited as reference [25].\N\NA quite remarkable result is Proposition 2.4 and its implication. Proposition 2.4 states that Crouzeix's conjecture is satisfied for any rank-one matrix. Since for any \(n \in \mathbb{N}, \{e_i e_j^\top : i, j \in [n]\}\) is a basis of \(\mathbb{M}_n(\mathbb{C})\), consisting of rank-one matrices, the authors correctly recognise that Crouzeix's conjecture would be true if for any \(n \in \mathbb{N}\) the set of complex \(n \times n\) matrices which satisfy Crouzeix's conjecture were closed under addition (Question 2.5). Consequently, an affirmative answer to Question 2.5 is equivalent to Crouzeix's conjecture being true.\N\NThe main part of Section 3 (the key section of the paper) is built on Proposition 2.1, Theorem 3.1 and Corollary 3.3. Theorem 3.6 and Corollary 3.7 reflect two applications of these results. The authors claim that Section 3 offers a novel link between the concept of cyclicity and Crouzeix's conjecture, induced by extremal functions and so called ``extremal vectors''. Extremal functions and their relation to finite Blaschke products were introduced by {M. Crouzeix} in his 2004 paper [loc. cit., Theorem 2.1] and further studied in large detail in [K. Li, loc. cit.].\N\NRecall that for any \((y, A) \in \mathbb{C}^n \times \mathbb{M}_n(\mathbb{C})\), \(\mathcal{Z}(y; A) : = [\{A^i y : i \in \mathbb{N}_0\}]\) (i.e., the linear hull of \(\{A^i y : i \in \mathbb{N}_0\}\)) is known as the \(A\)-cyclic subspace generated by \(y\). If \(\mathbb{C}^n = \mathcal{Z}(y; A)\), then \(y\) is called a cyclic vector for \(A\). \(\mathcal{Z}(y; A)\) is the smallest \(A\)-invariant subspace of \(\mathbb{C}^n\) which contains \(y\); i.e., \(\mathcal{Z}(y; A)\) is the smallest subspace \(\mathcal{V}\) of \(\mathbb{C}^n\), such that \(y \in \mathcal{V}\) and \(A(\mathcal{V}) \subseteq \mathcal{V}\).\N\NUnfortunately, the very short proof of Theorem 3.1-(b) does not reveal the place where the assumption of the existence of an \(A\)-cyclic subspace generated by an extremal vector \(x \in \mathbb{C}^n\) actually is implemented. In particular, I cannot recognise whether already the existence of a suitably constructed \(A\)-invariant subspace would suffice. In this regard, the reader should also note Question 3.11 and Question 3.12 on page 289 of the paper. Moreover, in my opinion, it would support the reader's understanding if Example 3.10 was placed directly below Theorem 3.1; especially since Example 3.10 also clearly reflects that an extremal vector actually exists.\N\NThe paper concludes with Section 4, which discusses the conjecture for symmetric matrices \(A = A^\top = \overline{A}^{\,\ast} \in \mathbb{M}_n(\mathbb{C})\) and sheds light on the role truncated Toeplitz operators (TTOs) play when studying numerical ranges of symmetric matrices (see, e.g., Question 4.1).\N\NIn conclusion, I would like to make the following two remarks:\N\begin{enumerate}\N\item[(i)] Although the paper contains very interesting results, the seemingly decisive role of cyclicity and extremal vectors is quite a mystery. The related proofs should therefore be elaborated in detail;\N\item[(ii)] A comprehensive recent survey of some major developments over the past 100 years in the study of the numerical ranges of bounded linear operators on a complex Hilbert space -- including Crouzeix's conjecture - is provided by \textit{P. Y. Wu} and \textit{H.-L. Gau} [in: Matrix and operator equations and applications. Cham: Springer. 413--439 (2023; Zbl 1546.47023)].\N\end{enumerate}