The triangulant
Determinants, permanents, traces, other special matrix functions (15A15) Eigenvalues, singular values, and eigenvectors (15A18) Diagonalization, Jordan forms (15A20) Canonical forms, reductions, classification (15A21) Commutativity of matrices (15A27) Miscellaneous inequalities involving matrices (15A45) Vector and tensor algebra, theory of invariants (15A72)
The main topic of the article is a generalization of a well-known result concerning the relationship between certain notions of commutativity of two matrices and their simultaneous diagonalizability or triangularizability. The authors focus in particular on matrices that share nontrivially intersecting invariant subspaces of complementary dimensions. To this end, they introduce a novel invariant -- in first and higher order forms -- aptly named the \textit{triangulant}.\N\NFor two \(n \times n\) matrices \(A\) and \(B\) over a field \(K\), let \(M(A, B)\) be an \(n^2 \times n^2\) block matrix consisting of \(n\times n\) blocks, where the \((i,j)\)-th block is given by \(A^{j-1} B^{i-1}\) for \(i,j=1,\ldots,n\). The \textit{first triangulant} of \(A\) and \(B\) is defined as \(T(A,B) = \det M(A,B)\).\N\NIn the case of \(2 \times 2\) matrices \(A\) and \(B\) over an algebraically closed field, the authors prove (Corollary 2.8) that \(A\) and \(B\) share a common eigenvector -- a simple instance of nontrivially intersecting invariant subspaces of complementary dimensions -- if and only if \(T(A,B) = 0\). They also show (Proposition 2.7) that in this case \(T(A,B)=\det[A,B]\), where \([A,B]=AB-BA\) is the commutator of matrices \(A\) and \(B\).\N\NThe first triangulant also plays a role in describing the open problem of determining the maximal number of mutually unbiased bases in a given dimension \(n\). Specifically, for any two \(n \times n\) unitary matrices \(A\) and \(B\) the following inequality holds (Proposition 2.9): \[ |T(A,B)| \leq n^{n^2/2}, \] with equality if and only if \(A\) and \(B\) have no multiple eigenvalues, \(A^n\) and \(B^n\) are scalar multiples of the identity, and the orthonormal eigenbases of \(A\) and \(B\) are mutually unbiased. Hence, the previously mentioned open problem is equivalent to finding the maximal number of \(n \times n\) unitary matrices such that their pairwise triangulants have absolute value \(n^{n^2/2}\).\N\NThe definition of the \textit{higher order triangulant} (Definition 3.4) is more intricate. For \(0 \leq k \leq n\), the \(k\)-th triangulant of \(n\times n\) matrices \(A\) and \(B\) is \N\[\NT_k(A,B)=\frac{T(A_k,B_k)}{G_k(A)G_k(B)},\N\]\Nwhere \(A_k\) is a linear transformation acting on the \(\binom{n}{k}\)-dimensional vector space \(\bigwedge^k K^n\) of exterior tensors via the Leibniz rule: \N\[\NA_k(v_1\wedge \cdots \wedge v_k)=\sum_{i=1}^k v_1 \wedge \cdots \wedge Av_i \wedge \cdots \wedge v_k.\N\]\NThis defines \(A_k\) as a \(\binom{n}{k} \times \binom{n}{k}\) matrix. The factors \(G_k(A)\) and \(G_k(B)\) are scalar invariants defined in Section 3 of the paper.\N\NIt is worth mentioning that \(T_k(A,B)\) is a bihomogeneous polynomial with integer coefficients and degree \(\binom{n}{2}\binom{n}{k}\binom{n-2}{k-1}\) in each of both sets of \(n^2\) variables. The triangulants satisfy the following relations: \(T_1(A,B)=T(A,B)\), \(T_{n-k}(A,B) = T_k(B,A)\), and \(T_0(A,B)=T_n(A,B)=1\).\N\NThe main result of the paper (Theorem 3.11) asserts that for \(n\times n\) matrices \(A\) and \(B\) with entries from an algebraically closed field \(K\) the \(k\)-th triangulant vanishes, \(T_k(A,B)=0\), if and only if \(B\) has \(k\)-dimensional invariant subspace that nontrivially intersects a \(k\)-codimensional invariant subspace of \(A\).\N\NThe article is well written, free of typographical errors, and contains detailed proofs that are accessible to readers with a background in linear algebra.
This page was built for publication: The triangulant
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7020895)
