In one of Ruelle`s papers "Rotation Numbers for Flows and Diffeomorphisms" Ruelle has the following calculation which I do not understand completely. Assume you have two invertible $2 \times 2$ matrices $A$ and $B$ with polar decompositions $A = U(\theta(A))|A|$ and $B = U(\theta(B))|B|$ where $U(\theta)$ is the planar rotation matrix by $\theta$ and $|B|=\sqrt(BB^T)$ etc. Then he says that $$ |\theta(AB)-\theta(A)-\theta(B)| \leq \pi $$

I don`t quite understand how he gets this result without a constant depending on norms of A and B. One can start by saying $$ AB = U(\theta(AB))|AB| = U(\theta(A))|A| U(\theta(B))|B| $$ $$ = U(\theta(A)+\theta(B))U(-\theta(B))|A|U(\theta(B))|B| $$ so that $$ U(\theta(AB)-\theta(A)-\theta(B)) = U(-\theta(B))|A|U(\theta(B))|B||AB|^{-1}. $$ Somewhere in the paper he gives as a hint $|\theta(PQ)| \leq \pi$ if $P$ and $Q$ are positive but I cant see how to use it.


Let $P=U(-\theta(B))|A|U(\theta(B))=U(\theta(B))^tAU(\theta(B))$ and $Q=|B|$ then


Hence, $PQ=U(-\theta(A)-\theta(B))U(\theta(AB))|AB|=U(\theta(AB)-\theta(A)-\theta(B))|AB|=\theta(PQ)|PQ|$.

By the uniqueness of the polar decomposition of an invertible matrix, we get $\theta(PQ)=\theta(AB)-\theta(A)-\theta(B)$.

Since $P$ and $Q$ are positive definite then $|\theta(PQ)|<\pi$ and the result follows.

  • $\begingroup$ I suggest using multiple lines for systems of equalities. Thie last line of math reads completely wrong on mobile because mobile doesn't play well with long equations. $\endgroup$ – Cameron Williams Jan 23 '17 at 12:23
  • $\begingroup$ Thanks, turned out to be simpler than I imagined just one more polar decomposition on top of the hint he gives :) Not sure why the question itself got so much positive feedback though. Maybe they liked the paper :p $\endgroup$ – Sina Jan 23 '17 at 22:45
  • $\begingroup$ @Sina You are welcome $\endgroup$ – Daniel Jan 23 '17 at 23:34
  • $\begingroup$ @Daniel, A follow up question: Let $g_1,g_2$ be the orthogonal axis scaled by $|B|$ and $e_1,e_2$ be the one scaled by $|A|$. Let $h_1,h_2$ the $e_1,e_2$ rotated by $-\theta(B)$. Then is it true that $\theta(PQ)$ is the smallest angle between the frames $g_1,g_2$ and $h_1,h_2$. This is because $P$ simply acts as scaling on $h_1,h_2$ and $Q$ in $g_1,g_2$. $\endgroup$ – Sina Jan 24 '17 at 2:46
  • $\begingroup$ I now think this is not true but if you have any comments its welcome. $\endgroup$ – Sina Jan 24 '17 at 2:52

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.