Are there any known common parametrizations of SL(3,R)? I know that it is easy to obtain a local parametrization by just exponentiating generators from the Lie algebra, but I do not know if they are global or not. How can I know if a parametrization obtained by exponentiating Lie algebra generators is a global parametrization?

As a second part of this question, SL(2,R) can be easily parametrized with three variables $t,\phi, \chi$ as $$\left[\begin{array}{cc}\cosh\chi\cos t+\sinh\chi\sin\phi&\sinh\chi\cos\phi+\cosh\chi\sin t\\\sinh\chi\cos\phi-\cosh\chi\sin t&\cosh\chi\cos t-\sinh\chi\sin\phi\end{array}\right]$$

Are there similar parametrizations for SL(3,R), that have reasonably simple expressions for the elements of the matrix? If I try to exponentiate the Lie algebra elements the resulting matrix elements are a mess.

  • $\begingroup$ Perhaps the Cartan decomposition en.wikipedia.org/wiki/Cartan_decomposition would be along the lines you want? $\endgroup$ – Yemon Choi Feb 3 '17 at 13:47
  • 5
    $\begingroup$ It's not clear how strictly you mean the word 'parametrization': Your 'parametrization' of $\mathrm{SL}(2,\mathbb{R})$ is degenerate along the plane $\chi=0$ in $\chi\phi t$-space, as this whole plane is mapped to the circle $\mathrm{SO}(2)\subset\mathrm{SL}(2,\mathbb{R})$. $\endgroup$ – Robert Bryant Feb 3 '17 at 15:50
  • 1
    $\begingroup$ Indeed it's diffeomorphic to $\mathbf{R}^5\times SO(3)$ and you won't get a homeomorphic parameterization by $\mathbf{R}^8$. On the other hand surjective parametrizations by 8 parameters exist, and probably also injective parameterizations by 8 parameters of dense Zariski open subsets (of course I at least mean continuous, probably analytic, and polynomial at least for the surjective parameterization). $\endgroup$ – YCor Feb 3 '17 at 23:29
  • $\begingroup$ How do you know it is diffeomorphic to R^5 x SO(3)? Do mean globally or locally? $\endgroup$ – user2133437 Jun 26 '17 at 9:30

By singular value decomposition, every matrix in $SL_3$ factors as $UDV$ with $U$ and $V$ in $SO(3)$ and $D$ of the form $\mathrm{diag}(x,y,1/(xy))$ (or $\mathrm{diag}(e^u,e^v,e^{-u-v})$ if you prefer). Any matrix in $SO(3)$ can be factored as $$\begin{bmatrix} \cos \alpha & - \sin \alpha & 0 \\ \sin \alpha & \cos \alpha & 0 \\ 0&0&1 \\ \end{bmatrix} \begin{bmatrix} \cos \beta & 0 & \sin \beta \\ 0&1&0 \\ -\sin \beta & 0 &\cos \beta \\ \end{bmatrix} \begin{bmatrix} 1&0&0 \\ 0 & \cos \gamma & -\sin \gamma \\ 0 & \sin \gamma &\cos \gamma \\ \end{bmatrix}$$ (Euler angle factorization). Is that the sort of formula you want?

  • $\begingroup$ Thanks! Just want to point out that you have a typo in your last matrix (a gamma that became a beta) $\endgroup$ – user2133437 May 11 '17 at 9:44
  • $\begingroup$ @DavidSpeyer: It seems that the determinant of your last matrix is not 1 (it is $\cos^2\gamma +\sin^2\beta\ $). $\endgroup$ – Mikhail Borovoi May 11 '17 at 13:36
  • $\begingroup$ Looks like Ben McKay fixed it before I got there. Thanks to both of you! $\endgroup$ – David E Speyer May 11 '17 at 13:46

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.