3
$\begingroup$

Given two $n$ dimensional positive definite matrices $A', B'$, is there a matrix $O \in SO(n)$ such that $A=O A', B=O B'$ and $$ \frac{A_{ij}}{\sqrt{A_{ii}A_{jj}}} = \frac{B_{ij}}{\sqrt{B_{ii}B_{jj}}}, i,j \in \{1,2,\ldots,n\}.~~~~~~~~ (1) $$

The above algebra question is from an geometry question: Given two affine coordinate system $\{a'_1, a'_2, \ldots, a'_n\}$ and $\{b'_1, b'_2, \ldots, b'_n\}$ in $n$ dimensional space, is there a matrix $O \in SO(n)$ such that $a_i = \sum_j O_{ij} a'_j$ and $b_i = \sum_j O_{ij} b'_j$ and the angle between each pair $a_i, a_j$ and $b_i, b_j$ is same.

Remark 1: The answer is yes for $n=2$. Intermediate value theorem can do this work. That is, choose $O=(\cos(x),\sin(x);-\sin(x),\cos(x))$ and define \begin{equation} f(x) = angle(a_1 \rightarrow a_2) - angle(b_1 \rightarrow b_2), \end{equation} then $f(0) = -f(\pi/2)$. Thus there must be a point $c \in (0, \pi/2)$ such that $f(c) = 0$, that is $angle(a_1 \rightarrow a_2) = angle(b_1 \rightarrow b_2)$. But this method may be hardly to solve the general $n$ dimensional case.

Remark 2: There are $n(n-1)/2$ independent parameters in $O$ and the number of constraints in the equation $(1)$ is also $n(n-1)/2$. It seems that the answer for the general $n$ deimensional case is also yes.

Remark 3: The question in the language of algebra may be more general than the one in the language of geometry, because $$ angle(a_1, a_2) + angle(a_1, a_3) \geq angle(a_2, a_3), $$ while there is no corresponding limits in the algebraic one.

Remark 4: The answer for the general $n$ dimensional case is yes if $A'B' = B'A'$ because they can be diagonlized at the same time.

I will also be appreciate if you revise this post or add some proper tags.

$\endgroup$
5
  • $\begingroup$ Either I am completely stupid today or for any such $O$ one has angle$(a_i\to a_j)$=angle$(a_i'\to a_j')$ and angle$(b_i\to b_j)$=angle$(b_i'\to b_j')$. Then, if angle$(a_i'\to a_j')$=angle$(b_i'\to b_j')$ for every $i$ and $j$, any such $O$ will do whereas if the equality fails for at least some pair, no such $O$ exists. $\endgroup$ Commented Apr 2, 2014 at 5:59
  • $\begingroup$ @მამუკაჯიბლაძე Thanks for your comment! Note that $a'_i, b'_i$ may be un-normalized. $\endgroup$ Commented Apr 2, 2014 at 8:05
  • $\begingroup$ @EdenHarder Sorry seems I indeed am stupid today. $SO$ transformations preserve angles between any vectors, either normalized or not, don't they?? $\endgroup$ Commented Apr 2, 2014 at 8:20
  • $\begingroup$ @მამუკაჯიბლაძე I give an example for $n=2$ in a Mathematica file(click the link please) $\endgroup$ Commented Apr 2, 2014 at 11:15
  • $\begingroup$ @EdenHarder Thank you very much for your effort! I think I understand my blunder now. $\endgroup$ Commented Apr 2, 2014 at 20:34

0

You must log in to answer this question.

Browse other questions tagged .