6
$\begingroup$

Let $A$ be an $m\times n$ matrix with integer entries. Let $d_i(A)$ be the greatest common divisor of all $i\times i$ minors of $A$, and define $d_0(A)=1$. Whenever $i\leq j$, one has that $d_i(A)$ divides $d_j(A)$, and this can easily be seen using determinental identities: given a $j\times j$ submatrix of $A$, perform successive Laplace expansions to express its determinant as a linear combination of $i\times i$ minors. Let $\alpha_i(A)=d_i(A)/d_{i-1}(A)$ be the invariant factors of $A$. From the existence of the Smith normal form of $A$, one has that $\alpha_i(A)$ divides $\alpha_j(A)$ whenever $i\leq j$, and it follows from this that $d_i(A)d_{j-1}(A)$ divides $d_{i-1}(A)d_j(A)$. I am interested in ways to show this latter result that do not appeal to the Smith normal form of $A$. Specifically:

Can one use determinental identities to show directly that $d_i(A)d_{j-1}(A)$ divides $d_{i-1}(A)d_j(A)$ whenever $i\leq j$?

For example, if one could express the product of an arbitrary $(i-1)\times(i-1)$ minor and an arbitrary $j\times j$ minor as a linear combination of products of $i\times i$ minors and $(j-1)\times(j-1)$ minors, the result would follow. Can this be done?

$\endgroup$
1
  • 1
    $\begingroup$ The question makes me think of algebras with straightening laws. Algebras generated by minors (Plücker coordinates!) often have combinatorial multiplication rules that might give something like you want. To get started, maybe the example on page 253 of this is helpful: msri.org/people/staff/de/papers/pdfs/1980-001.pdf $\endgroup$ Commented Feb 28, 2022 at 10:56

0

You must log in to answer this question.