1
$\begingroup$

Is there a way to construct a real skew-symmetric matrix given a single purely imaginary eigenvalue and corresponding eigenvector? The other eigenvalues and eigenvectors can be chosen at will. Thanks

$\endgroup$
1
$\begingroup$

I don't think so. Consider $0\in\mathbb R$ is a skew-symmetric matrix, but it always has eigenvalue $0$. Furthermore it is easy to show that every skew-symmetric matrix $A\in\mathbb R^{2\times 2}$ with $$ A = \begin{bmatrix}0 & a\\ -a & 0\end{bmatrix}$$ for $a\in\mathbb R$ has always the eigenvalues $ia$ and $-ia$ with eigenvectors $$ \begin{bmatrix}-i\\1\end{bmatrix},\qquad \begin{bmatrix}i\\1\end{bmatrix}.$$ This means you cannot construct a skew-symmetric $2\times 2$ matrix to any given eigenvector, for example $$\begin{bmatrix}1\\1\end{bmatrix}.$$

$\endgroup$
1
$\begingroup$

Let me assume that you want to construct a skew-symmetric matrix $A \in M_{2n}(\mathbb{R})$ with eigenvalues $a_1, -a_1 \ldots, a_n, -a_n$ for pure imaginary, non-zero and distinct $a_i$. First, as Wauzl observed, you can't specify the eigenvectors for $A$ arbitrary. If $v_i$ is an eigenvector of $A$ corresponding to the eigenvalue $a_i$, then $\overline{v_i}$ must be an eigenvector of $A$ corresponding to the eigenvalue $-a_i$. In particular, since $a_i \neq -a_i$, the vectors $v_i$ and $\overline{v_i}$ must be linearly independent. In fact, since $A$ is normal (considered as a matrix over the complex numbers), the eigenvalues $(v_1, \overline{v_1}, \ldots, v_n, \overline{v_n})$ must be an orthonormal basis of $\mathbb{C}^{2n}$ (with respect to the standard Hermitian inner product).

In the other direction, if you are given a list $(v_1, \ldots, v_n)$ of complex vectors such that $\mathcal{B} = (v_1, \overline{v_1}, \ldots, v_n, \overline{v_n})$ is an orthonormal ordered basis for $\mathbb{C}^{2n}$, you can construct a unitary matrix $U \in M_{2n}(\mathbb{C})$ with the vectors of $\mathcal{B}$ as columns (in the same order as the are given in $\mathbb{B}$). Choose any pure imaginary list of numbers $(a_1, \ldots, a_n)$ and let $D = \mathrm{diag}(a_1, \ldots, a_n)$. Then $UDU^{*}$ will be a real, skew-symmetric matrix with eigenvalues $(a_1, -a_1, \ldots, a_n, -a_n)$ and corresponding eigenvectors $(v_1, \overline{v_1}, \ldots, v_n, \overline{v_n})$.

You can easily modify this argument to construct odd-dimensional skew-symmetric matrices.

$\endgroup$

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.