Problem: Let $\{A_r\},\{B_r\},\{C_r\}$ each be families of (determinant 1) 2x2 matrices in $SL(2,\mathbb{R})$ such that each family is continuously indexed by a parameter $0<r<1$. Is there a known formula for calculating $tr(W)$, where $W$ is $(A_rB_rC_r)^nA_r$, $(A_rB_rC_r)^nA_rB_r$, or $(A_rB_rC_r)^n$?

Progress thus far: The indexing is sufficiently complicated that simply multiplying and looking for a pattern has not been successful. Each entry of each family is a ratio of quadratics in $r$, and I do not remember the right computational linear algebra to attack this problem.

Motivation: the families of matrices represent isometries of the hyperbolic plane, and I am interested in knowing when such isometries change from elliptic to parabolic to hyperbolic

  • $\begingroup$ I've adjusted the question to say that I'm looking at matrices with determinant 1. I only care about the absolute value of the trace, but I'm looking to plot it as a function of the indexing variable $r$, so a polynomial output would be nice. $\endgroup$ – KReiser Aug 1 '12 at 2:07
  • 2
    $\begingroup$ If you can solve for the eigenvalues and eigenvectors of $A_r B_r C_r$ as a function of $r$ (this is something of a big if), then you can work over $\mathbb{C}$ and change basis so that $A_r B_r C_r$ is in Jordan normal form. Things are not so bad from here. $\endgroup$ – Qiaochu Yuan Aug 1 '12 at 2:22
  • $\begingroup$ Instead of thinking of them as families of matrices in $SL(2,\mathbb R)$, you can instead think of them as single matrices $A,B,C \in SL\big(2,{\mathbb R}(r)\big)$. $\endgroup$ – Greg Martin Aug 1 '12 at 5:53

$\def\Tr{\mathrm{Tr}}$Let $M(r) = A(r) B(r) C(r)$. Let the characteristic polynomial of $M(r)$ be $$\lambda^3 + x(r) \lambda^2 + y(r) \lambda + z(r)$$ Each of $x(r)$, $y(r)$ and $z(r)$ is a rational function, computable by a computer algebra system.

If you can solve this cubic then proceed as Qiaochu says. (And it is definitely worth putting this cubic into a computer algebra system to see whether it simplifies in some surprising way.) But solving cubics is usually painful, so here is an alternative:

The Cayley-Hamilton theorem tells you that $$M(r)^3 + x(r) M(r)^2 + y(r) M(r) + z(r)=0$$ so $$\Tr M(r)^{n+3} = - x(r) \Tr M(r)^{n+2} - y(r) \Tr M(r)^{n+1} - z(r) \Tr M(r)^n$$ So this gives a linear recursion for $\Tr M(r)^n$ which you can use to compute $\Tr M(r)^n$ pretty easily, and similarly for your other traces.


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.