11 Apr 2026

Today I would like to bring up a simple result of which I was unaware until a few days ago: Sylvester's determinant theorem – which is not Sylvester's determinant identity, the one that relates the determinant of a matrix with some special submatrices –. It is more often referred to as the Weinstein-Aronszajn identity (see [1] and [2]), and overall it is very simple to state:

Given a field $\mathbf F$ and $A \in \mathscr M_{m \times n} (\mathbf F), B \in \mathscr M_{n \times m} (\mathbf F)$, then $$ \det (\mathbf 1 + AB) = \det (\mathbf 1 + BA). $$

As a remark, notice that the first $\mathbf 1$ is the $m\times m$ identity, which differs from the second, $n\times n$ identity; I will be calling both indistinguishably henceforth. This theorem is not particularly hard to prove either:

Recall first the determinant of a block-partitioned matrix; if $A, B, C, D$ are $m\times m, m\times n, n\times m$ and $n\times n$ matrices respectively with $A$ and $D$ invertible, then: $$ \det A \det (D - CA^{-1}B) = \det \begin{pmatrix} A & B\\ C & D \end{pmatrix} = \det D \det (A - B D^{-1} C). $$ Applying this result in the two different fashions above yields the desired equality: $$ \det(\mathbf 1 + AB) = \det \begin{pmatrix} \mathbf 1 & -A\\ B & \mathbf 1 \end{pmatrix} = \det(\mathbf 1 + BA). $$

Since the previous determinants look similar to the characteristic polynomials of $AB$ and $BA$, we obtain that they are equal up to a factor of $(-\lambda)^{m - n}$ where, without loss of generality, $m \geq n$. In other words, $AB$ and $BA$ have the same eigenvalues counting multiplicities except for the zeros.

Given $A \in \mathscr M_{m \times n} (\mathbf F), B \in \mathscr M_{n \times m} (\mathbf F)$, with $m \geq n$, $$ \chi_{AB} = \det (AB - \lambda \mathbf 1) = (-\lambda)^{m-n} \det (BA - \lambda \mathbf 1) = (-\lambda)^{m-n} \chi_{BA}. $$
$$ \begin{align*} \chi_{AB} &= \det (AB - \lambda \mathbf 1) = (-\lambda)^m \det \left(\mathbf 1 - \frac{1}{\lambda} AB\right)\\ &= (-\lambda)^m \det \left(\mathbf 1 - \frac{1}{\lambda} BA\right) = (-\lambda)^{m - n} \det (BA - \lambda \mathbf 1) = (-\lambda)^{m - n} \chi_{BA}. \end{align*} $$


Because the essence of mathematics is solving problems, let's apply this new tool to one:

Let $U \in \mathscr M_n (\mathbf C)$ be a unitary matrix, block-partitioned as follows: $$ U = \begin{pmatrix} U_{11} & U_{12}\\ U_{21} & U_{22} \end{pmatrix}, $$ where $U_{11}$ is $k\times k$. Prove that $|\det U_{11}| = |\det U_{22}|$.
Recall that the definition of being unitary states that $U\adj U = \adj U U = \mathbf 1$, so multiplying the two previous matrices must give $$ \begin{align*} U \adj U &= \begin{pmatrix} U_{11} & U_{12}\\ U_{21} & U_{22} \end{pmatrix} \begin{pmatrix} \adj U_{11} & \adj U_{21}\\ \adj U_{12} & \adj U_{22} \end{pmatrix} = \begin{pmatrix} U_{11} \adj U_{11} + U_{12} \adj U_{12} & U_{11} \adj U_{21} + U_{12} \adj U_{12}\\ U_{21} \adj U_{11} + U_{22} \adj U_{12} & U_{21} \adj U_{21} + U_{22} \adj U_{22} \end{pmatrix} = \begin{pmatrix} \mathbf 1 & 0\\ 0 & \mathbf 1 \end{pmatrix},\\ \adj U U &= \begin{pmatrix} \adj U_{11} & \adj U_{21}\\ \adj U_{12} & \adj U_{22} \end{pmatrix} \begin{pmatrix} U_{11} & U_{12}\\ U_{21} & U_{22} \end{pmatrix} = \begin{pmatrix} \adj U_{11} U_{11} + \adj U_{21} U_{21} & \adj U_{11} U_{12} + \adj U_{21} U_{22}\\ \adj U_{12} U_{11} + \adj U_{22} U_{21} & \adj U_{12} U_{12} + \adj U_{22} U_{22} \end{pmatrix} = \begin{pmatrix} \mathbf 1 & 0\\ 0 & \mathbf 1 \end{pmatrix}; \end{align*} $$ from this we get the following equalities: $$ \begin{align} U_{11} \adj U_{11} &= \mathbf 1 - U_{12} \adj U_{12} \implies |\det U_{11}|^2 = \det(\mathbf 1 - U_{12} \adj U_{12}),\\ \adj U_{11} U_{11} &= \mathbf 1 - \adj U_{21} U_{21} \implies |\det U_{11}|^2 = \det(\mathbf 1 - \adj U_{21} U_{21}),\\ U_{22} \adj U_{22} &= \mathbf 1 - U_{21} \adj U_{21} \implies |\det U_{22}|^2 = \det(\mathbf 1 - U_{21} \adj U_{21}),\\ \adj U_{22} U_{22} &= \mathbf 1 - \adj U_{12} U_{12} \implies |\det U_{22}|^2 = \det(\mathbf 1 - \adj U_{12} U_{12}). \end{align} $$ Finally, using Sylvester's theorem for determinants on $(1)$ and $(4)$ finishes the problem off: $$ |\det U_{11}|^2 = \det(\mathbf 1 - U_{12} \adj U_{12}) = \det(\mathbf 1 - \adj U_{12} U_{12}) = |\det U_{22}|^2 \implies |\det U_{11}| = |\det U_{22},| $$ where the last step is possible because the modulus of a complex number is always a positive real.

To conclude this post, I would like to cite "The Usefulness of Formulas in Solving Matrix Problems" (Stanescu) in Crux Mathematicorum vol. 51, no. 9, as a good source of some lesser-known determinant identities.