Hi all. Is there any explicit matrix expression for a general element of the special orthogonal group $SO(3)$? I have been searching texts and net both, but could not find it. Kindly provide any references.
-
3$\begingroup$ For every $n\in\mathbb N$ and any field $k$, there is a birational map $\mathfrak{so}_n\left(k\right)\to \mathrm{SO}_n\left(k\right)$ given by $A\mapsto \left(I_n-A\right)^{-1}\left(I_n+A\right)$. It is defined for "almost" all $A\in \mathfrak{so}_n\left(k\right)$ (namely, for those without eigenvalue $1$), is injective and is "almost" surjective (meaning that "almost" all elements of $\mathrm{SO}_n\left(k\right)$ are images under this map). Is this anything like what you are searching for? Because I don't think you can do any better. Even for $n=2$, there is nothing explicit in sin and cos! $\endgroup$– darij grinbergJul 12, 2011 at 17:43
-
4$\begingroup$ You can write down the general element in terms of sines and cosines of the Euler angles: en.wikipedia.org/wiki/Charts_on_SO(3). Also:en.wikipedia.org/wiki/Euler_angles $\endgroup$– Jim BryanJul 12, 2011 at 18:11
-
$\begingroup$ @darij: "Even for $n=2$ there is nothing explicit in sin and cos!", I don't understand your comment: what about the usual matrix $(\cos\theta$, $-\sin\theta$ ;$\sin\theta$, $\cos\theta )$ ? $\endgroup$– QfwfqJul 12, 2011 at 18:15
-
2$\begingroup$ @darij: I am pretty sure it is standard for the notation $\text{SO}(3)$ to refer to the case $k = \mathbb{R}$ only. $\endgroup$– Qiaochu YuanJul 12, 2011 at 19:04
-
2$\begingroup$ @Qiaochu: Yes, I was generalizing. What I mean by "there is nothing explicit in sin and cos" is that $\sin$ and $\cos$ are two functions whose goal IS more or less to parametrize $\mathrm{SO}_2\left(\mathbb R\right)$, so one can wonder what "explicit matrix expression" means, and if it allows $\sin$ and $\cos$, why it doesn't allow any other special function I might think up. $\endgroup$– darij grinbergJul 12, 2011 at 20:08
3 Answers
Here is the standard quaternion answer: Given $(a,b,c,d)$ such that $a^2+b^2+c^2+d^2=1$, the matrix $$\begin{pmatrix} a^2+b^2-c^2-d^2&2bc-2ad &2bd+2ac \\ 2bc+2ad &a^2-b^2+c^2-d^2&2cd-2ab \\ 2bd-2ac &2cd+2ab &a^2-b^2-c^2+d^2\\ \end{pmatrix}$$ is a rotation and every rotation matrix is of this form. Note that $(a,b,c,d)$ and $(-a, -b,-c,-d)$ give the same rotation.
There is a good way to derive the sort of thing you're looking for: use the double cover $SU(2) \to SO(3)$. $SU(2)$ is diffeomorphic to the 3-sphere $S^3 \subseteq \mathbb{C}^2$ $$ SU(2) = \left\{ \begin{pmatrix} a & -\overline{b} \\ b & \overline{a} \end{pmatrix} : |a|^2 + |b|^2 = 1 \right\} $$ Now $SU(2)$ acts on its Lie algebra $\mathfrak{su}_2$ (which is 3-dimensional) by conjugation. This action preserves the inner product $$ \langle X, Y \rangle = - \frac12 \mathrm{tr}(XY) = \frac12 \mathrm{tr}(X^*Y)$$ (which is a scalar multiple of the Killing form of $\mathfrak{su}_2$, FYI) and hence this gives a homomorphism $SU(2) \to SO(\mathfrak{su}_2) \simeq SO(3)$. (A priori this gives a map to $O(3)$, but $SU(2)$ is connected so the image lands in $SO(3)$.
Now consider the orthonormal basis for $\mathfrak{su}_2$ given by $$ e_1 = \begin{pmatrix} i & 0 \\ 0 & -i \end{pmatrix}, e_2 = \begin{pmatrix} 0 & 1 \\ -1 & 0 \end{pmatrix}, e_3= \begin{pmatrix} 0 & i \\ i & 0 \end{pmatrix}, $$ let $$ x = \begin{pmatrix} a & -\overline{b} \\ b & \overline{a}\end{pmatrix},$$ and write down the adjoint action of $x$ on $e_1,e_2,e_3$. For instance you get $$ \begin{align} xe_1x^{-1} & = \begin{pmatrix} i(|a|^2 - |b|^2) & 2ia\overline{b} \\ 2i\overline{a} b & -i(|a|^2 - |b|^2) \end{pmatrix} \\ & = (|a|^2 - |b|^2)e_1 + i(a\overline{b} - \overline{a}b)e_2 + (a \overline{b} + \overline{a}b)e_3. \end{align}$$ This gives you the first column of the matrix representation conjugation by $x$. I'll leave the others to you. But this way you can see where the formulas come from.
This gives exactly David Speyer's answer (possibly modulo some re-ordering of the basis). His four real numbers $a,b,c,d$ would correspond to my complex numbers $a,b$ via $$a_{mine} = a + i b, \quad b_{mine} = c + id.$$
To give something explicit in sine and cosine,
$$ \left( \begin{array}{ccc} \cos\theta \cos\psi & -\cos\phi \sin\psi + \sin\phi \sin\theta \cos\psi & \sin\phi \sin\psi + \cos\phi \sin\theta \cos\psi \\\ \cos\theta \sin\psi & \cos\phi \cos\psi + \sin\phi \sin\theta \sin\psi & -\sin\phi \cos\psi + \cos\phi \sin\theta \sin\psi \\\ -\sin\theta & \sin\phi \cos\theta & \cos\phi \cos\theta \end{array} \right) $$
Note that three parameters are required. In odd dimension, there is a real eigenvalue. For $SO_n$ this eigenvalue is $+1.$ So there is a fixed vector in some direction. It takes two parameters to specify this point on the unit sphere. The Lie group element is then a rotation around this point. So it takes a third parameter specifying the amount of rotation about that axis.