Joint matricial range and joint congruence matricial range of operators (Q782504): Difference between revisions

Given \({\mathbf A}=(A_1, \dots, A_m)\), where \(A_1, \dots, A_m\in {\mathbb R}_{n\times n}\), the real joint \((p,q)\)-matricial range of \({\mathbf A}\) is the set \begin{align*} \Lambda_{p,q}^{\mathbb R}({\mathbf A})=&\{ (B_1, \dots, B_m)\in ({\mathbb R}_{q\times q})^m: X^tA_jX = I_p\otimes B_j, \\ &\qquad \mbox{ for some } X\in{\mathbb R}_{n\times pq}, X^tX=I_{pq}\}, \end{align*} where \(X^t\) is the transpose of \(X\). When \(p=q=1\), this is reduced to the joint numerical range of \({\mathbf A}\): \[ W^{\mathbb R}({\mathbf A}) =\{(x^tA_1x, \dots, x^tA_mx): x\in{\mathbb R}^n, x^tx=1\}. \] One can decompose \(A_j = S_j +G_j\), where \(S_j\) is real symmetric and \(G_j\) is real skew-symmetric for \(j = 1, \dots , m\). Since \(v^tG_j v=0\) for all real vectors \(v\in {\mathbb R}^n\), we have \(W^{\mathbb R}(A_1, \dots , A_m) = W^{\mathbb R}(S_1,\dots, S_m)\). \textit{L. Brickman} [Proc. Am. Math. Soc. 12, 61--66 (1961; Zbl 0104.01204)] showed that \(W^{\mathbb R}(S_1, S_2)\) is convex for symmetric \(S_1, S_2\) when \(n\ge 3\). However, \(W^{\mathbb R}(S_1, S_2, S_3)\) may fail to be convex. Note that \(\Lambda_{p,q}^{\mathbb R}({\mathbf A})\) has the same structure as \(\Lambda_{p,q}^{\mathbb R}(S_1, \dots , S_m, G_1, \dots , G_m)\). In this paper, it is shown that the set \(\Lambda_{p,q}^{\mathbb R}({\mathbf A})\) is non-empty and star-shaped for sufficiently large \(n\) and lower bounds are given on the dimension \(n\) for which \(\Lambda_{p,q}^{\mathbb R}({\mathbf A})\) is non-empty and star-shaped. Similar results are obtained for the joint congruence \((p, q)\)-matricial range \begin{align*} \Lambda_{p,q}^{\mathbb C}({\mathbf A})&=\{ (B_1, \dots, B_m)\in ({\mathbb C}_{q\times q})^m: X^tA_jX = I_p\otimes B_j, \\ &\qquad \mbox{ for some } X\in{\mathbb C}_{n\times pq}, X^tX=I_{pq}\}, \end{align*} where \(A_1, \dots, A_m\in {\mathbb C}_{n\times n}\). It is the complex analogy of \(\Lambda_{p,q}^{\mathbb R}({\mathbf A})\). When \(p = q = m = 1\), this is reduced to the congruence numerical range. Some results are extended to real and complex Hilbert space bounded linear operators. In particular, it is proved that the joint essential \((p, q)\)-matricial range of \(\mathbf A\) is compact, convex, and non-empty. Some generalized numerical ranges in the context of Lie groups and algebras can be found in [\textit{C.-K. Li} and \textit{T.-Y. Tam}, Can. J. Math. 52, No. 1, 141--171 (2000; Zbl 0978.15026); \textit{D. Ž. Đoković} and \textit{T.-Y. Tam}, Can. Math. Bull. 46, No. 3, 332--343 (2003; Zbl 1047.22013); \textit{T.-Y. Tam} and \textit{W.-S. Cheung}, Pac. J. Math. 238, No. 2, 387--398 (2008; Zbl 1197.22001); \textit{W.-S. Cheung} and \textit{T.-Y. Tam}, Can. Math. Bull. 54, No. 1, 44--55 (2011; Zbl 1213.22001)] and they are somewhat related to some discussions in the paper.