I have a couple of questions regarding skew-Hermitian matrices over finite fields. A matrix $A$ over $\mathbb{F}_{q^{2}}$ is skew-Hermitian if $A + A^{*} = 0$, where $A^{*}$ is the conjugate transpose with respect to the involutory field automorphism $x \mapsto x^{q}$.

  1. Is there a restriction on the rank of a skew-Hermitian matrix, as in the case of skew-symmetric matrices? Is this affected by whether the characteristic is $2$, or whether we add the requirement that the diagonal is zero?
  2. If $C$ and $D$ are skew-Hermitian matrices with the same rank, can we say there exists a matrix $A$ such that $A C A^{*} = D$?
  • 1
    $\begingroup$ It will certainly depend on whether the polynomial $x^2 - 1$ has a root, or at least how one chooses this "involutory automorphism" $\endgroup$ Oct 22 '15 at 15:37
  • $\begingroup$ Characteristic two does require exceptional handling, because a diagonal matrix with zeros and ones only, viewed as having entries in $\Bbb{F}_4$, is skew-Hermitian according to this definition. And its rank can be anything. $\endgroup$ Oct 22 '15 at 17:33
  • 1
    $\begingroup$ Duh. Even in odd characteristic any diagonal matrix with entries from the kernel of the relative trace $x\mapsto x+x^q$ is skew-Hermitian. So all ranks occur without he assumption that the diagonal should be zero. Thinking... $\endgroup$ Oct 22 '15 at 17:39
  • $\begingroup$ @JyrkiLahtonen That's a good way to think about it. I had a nagging worry that they should have even rank as for skew-symmetric/alternating matrices, but I see now that is clearly not the case. $\endgroup$ Oct 22 '15 at 21:15

Okay, I think I have managed to solve these questions.

We can have a skew-Hermitian matrix of any rank (thanks Jyrki); we just take an element $\alpha$ such that the trace $T(\alpha) = \alpha+\alpha^{q} = 0$. Then if we want an $n \times n$ skew-Hermitian matrix with rank $k$ we just take $$C = \alpha \begin{bmatrix} I_{k} & 0\\0&0 \end{bmatrix}.$$
If the characteristic of the field is $2$, we can take $\alpha = 1$.

I'm going to actually cop out and ignore my question about what happens when the diagonal is required to be zero. I mentioned this because in dealing with bilinear forms, this is naturally implied with a skew-symmetric matrix in odd characteristic, but needs to be explicitly stated in even characteristic for the bilinear form to define a symplectic geometry. Nothing like this comes up in the skew-symmetric situation.

I will approach the second problem by first considering Hermitian matrices. Let $C$ be an $n \times n$ Hermitian matrix with entries in $\mathbb{F}_{q^{2}}$ having rank $k$ (WLOG assume $k \geq 1$). I want to show that there is an invertible matrix $A$ such that $$ ACA^{*} = \begin{bmatrix} I_{k} & 0\\0 & 0 \end{bmatrix}$$ (block matrix representation where the bottom right block is $(n-k)\times (n-k)$). The key here (and the motivation for the question) is to think of $C$ as representing a Hermitian bilinear form $B(u,v) = u C v^{*}$, then the matrix $A C A^{*}$ represents this bilinear form with respect to a different basis (basis vectors being the rows of $A$). The collection of vectors $u$ such that $u C v^{*} = 0$ for all $v \in V = \mathbb{F}_{q^{2}}^{n}$ is a a subspace called the radical, so call this $W$. It is clear that $W = \mathrm{null}(C)$ so $\dim(W) = n-k$; therefore we will say that $\{w_{1}, \ldots, w_{n-k}\}$ is a basis for $W$. Then $V = V_{0} \oplus W$ for some subspace $V_{0} \leq V$ for which the Hermitian form restricted to $V_{0}$ is nondegenerate (and $\dim(V_{0}) = k$).

Now our bilinear form is nondegenerate on $V_{0}$, so we can find a vector $v \in V_{0}$ with $vCv^{*} = c \neq 0$. It can also be seen that $c = c^{q}$ (because $(vCv^{*})^{*} = vCv^{*}$), therefore $c \in \mathbb{F}_{q}$. The norm map $N:\mathbb{F}_{q^{2}} \to \mathbb{F}_{q}$ is surjective, so there exists an $\omega \in \mathbb{F}_{q^{2}}$ with $N(\omega) = \omega^{q+1} = c$, and so putting $v_{1} = \frac{1}{\omega}v$ we have $v_{1}Cv_{1}^{*} = 1$, and $V_{0} = \langle v \rangle \oplus \langle v \rangle^{\perp}$. Using induction on $\dim(V_{0})=k$ lets us find a basis $\{v_{1}, \ldots, v_{k}\}$ with $v_{i}Cv_{i}^{*} = 1$ for all $i$, and $v_{i} C v_{j}^{*} = 0$ for all $i \neq j$. Now finally we can put $$A = \begin{bmatrix} \phantom{\ldots \ldots} v_{1} \phantom{\ldots \ldots} \\ \vdots \\ v_{k} \\ w_{1} \\ \vdots \\ w_{n-k} \end{bmatrix}$$ and we have $$ACA^{*} = \begin{bmatrix} I_{k} & 0 \\ 0 & 0 \end{bmatrix}.$$

We will now prove the analogous result for skew-Hermitian matrices. Take $\alpha \in \mathbb{F}_{q^{2}}$ such that $T(\alpha) = \alpha^{q}+\alpha = 0$. WLOG we can assume that $N(\alpha) = \alpha^{q+1} = 1$, because otherwise if $N(\alpha) = c$ then we can replace $\alpha$ with $\frac{\alpha}{\sqrt{c}}$. Let $C$ be skew-Hermitian so $C^{*} = -C$. We want to show that there exists a matrix $A$ such that $$ACA^{*} = \alpha \begin{bmatrix} I_{k} & 0\\0&0\end{bmatrix},$$ where $k$ is the rank of $C$. Now $(\alpha^{q} C)^{*} = \alpha C^{*} = \alpha^{q} C$, that is, $\alpha^{q} C$ is Hermitian, so there exists a matrix $A$ such that $$A (\alpha^{q} C) A^{*} = \alpha^{q} A C A^{*} = \begin{bmatrix} I_{k} & 0 \\ 0 & 0 \end{bmatrix}.$$ Multiplying both sides by $\alpha$ we have $$\alpha^{q+1} A C A^{*} = A C A^{*} = \alpha \begin{bmatrix} I_{k} & 0 \\ 0 & 0 \end{bmatrix}.$$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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