3
$\begingroup$

It is a known fact that if $X$ is a skew-symmetric matrix, then $e^X$ is an orthogonal matrix.

Is it also true the opposite, ie that any orthogonal matrix admits a representation like $e^X$ for some $X$ skew-symmetric?

If not, is there a way to parametrize the space of the orthogonal matrices? With this I mean, a function $f$ that takes some parameters $\Theta$ and returns a matrix $O$ such that:

  • $f(\Theta)=O$ is orthogonal
  • If $O$ is orthogonal, there exists a set of parameters $\Theta$ s.t. $f(\Theta)=O$?

Does the topic get easier if we focus on orthonormal matrices?

$\endgroup$
  • 1
    $\begingroup$ "Orthonormal matrix" is not a common term. "Orthogonal matrix" already means a matrix with orthonormal rows / columns. $\endgroup$ – Qiaochu Yuan Jul 24 '17 at 18:01
5
$\begingroup$

As you said, if $X$ is a skew-symmetric matrix then $e^X$ is orthogonal, and precisely $e^X$ is a special orthogonal matrix, i.e. with determinant equal to $+1$: $$\det(e^X)=e^{\operatorname{Tr}(X)}=e^0=1$$ Conversely, for $O\in SO_n(\Bbb R)$, there is an orthonormal basis in which $O$ is similar to $$S=\operatorname{diag}(I_p,R(\theta_1),\ldots,R(\theta_s))$$ where $$R(\theta_i)=\begin{pmatrix}\cos\theta_i&-\sin\theta_i\\\sin\theta_i&\cos\theta_i\end{pmatrix}$$ Moreover, we have $\exp(J_i)=R(\theta_i)$ with $$J_i=\begin{pmatrix}0&-\theta_i\\\theta_i&0\end{pmatrix}$$ so we define the skew-symmetric matrix $$A=\operatorname{diag}(0_p, J_1,\ldots,J_s)$$ and we get $e^A=S$.

$\endgroup$
  • $\begingroup$ however, this method only parametrize special orthogonal matrix; is there any similar way of parameterize orthogonal matrix with determinant -1 as well ? $\endgroup$ – ElleryL Dec 11 '18 at 19:20
5
$\begingroup$

Every orthogonal matrix with determinant $1$ has the form $\exp(X)$ with $X$ skew-symmetric. Another representation of orthogonal matrices is the Cayley parameterisation: $(I+X)(I-X)^{-1}$ is orthogonal whenever $X$ is skew-symmetric. This produces all orthogonal matrices of determinant $1$ which do not have $-1$ as an eigenvalue.

$\endgroup$
  • $\begingroup$ however, this method only parametrize special orthogonal matrix; is there any similar way of parameterize orthogonal matrix with determinant -1 as well $\endgroup$ – ElleryL Dec 11 '18 at 19:23
1
$\begingroup$

As Lord Shark of the Unknown already told, the answer is negative for $O(n,\mathbb{R})$, but it is affirmative for $SO(n,\mathbb{R})$. That the answer is negative in the case of $O(n,\mathbb{R})$ follows from the fact that it is not connected, whereas the space of all skew-symmetric matrices is (together with the fact that the exponential map is continuous).

Actually, if $K$ is a compact and connected matrix group and if$$\mathfrak{k}=\left\{X\in M(n,\mathbb{R})\,|\,(\forall t\in\mathbb{R}):\exp(tX)\in K\right\},$$then $\exp(\mathfrak{k})=K$.

$\endgroup$

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.