5
$\begingroup$

The well-known matrix determinant lemma states that for an invertible square matrix $A$ and column vectors $u,v$ one has $$ \det(A + uv^T) = \det(A)(1 + v^T A^{-1} u). $$ Is there any analogous formula in case we are adding to $A$ a matrix which is not rank-one, but still has some special structure? The case I have in mind is when $A,B$ are square matrices, $I$ is the identity matrix, $v$ is a vector and we are dealing with tensor (or Kronecker, if you wish) products: $$ \det(A \otimes I + I \otimes B + (v^T v) \otimes I) = ? $$ Eventually I'd like to work with more components of the tensor product, but for simplicity let me just state the question for two.

$\endgroup$
1
  • $\begingroup$ In $v^Tv$, the $v$ is a row vector, but in $uv^T$ both $u,v$ are column vectors. $\endgroup$ Oct 3, 2022 at 13:30

1 Answer 1

4
$\begingroup$

If $A,B$ are square matrices, then $A\otimes I+I\otimes B$ is commonly known as the Kronecker sum of $A$ and $B$. If $A$ has eigenvalues $\mu_1,\dots,\mu_m$ and $B$ has eigenvalues $\nu_1,\dots,\nu_n$, then $A\otimes I-I\otimes B$ has eigenvalues $\mu_i-\nu_j$ for $1\leq i\leq m,1\leq j\leq n$. Therefore, $\det(A\otimes I-I\otimes B)=\text{res}(\chi(A),\chi(B))$ (which denotes the resultant of the characteristic polynomial of $A$ and $B$).

Therefore (Peter Taylor originally mentioned that we can do something like this), by the matrix determinant lemma, we have

$$\text{Det}(A\otimes I+I\otimes B+pq^T\otimes I+I\otimes uv^T)$$

$$=\text{res}\big(\chi(A+pq^T),\chi(B+uv^T))=\text{res}(\det(xI-A-pq^T),\det(xI-B-uv^T)\big)$$

$$=\text{res}\big(\det(xI-A)\cdot(1-q^T(xI-A)^{-1}p),\det(xI-B)\cdot(1-v^T(xI-B)^{-1}u)\big)$$

$$=\text{res}\big(\chi(A)\cdot(1-q^T(xI-A)^{-1}p),\chi(B)\cdot(1-v^T(xI-B)^{-1}u)\big).$$

$\endgroup$
2
  • 1
    $\begingroup$ The characteristic polynomial is a determinant, so the matrix determinant lemma gives us $\chi(A + uv^T) = (1 + v^T (A - xI)^{-1} u) \chi(A)$. Should the last paragraph be interpreted as wanting a stronger result than this? $\endgroup$ Oct 3, 2022 at 14:45
  • $\begingroup$ @PeterTaylor. That is true. I will reword my answer to take that into consideration. $\endgroup$ Oct 3, 2022 at 14:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.