4

I want to refer to a mathematics book that explains the n-dimensional rotation matrix or rotation transformation.

Wikipedia concentrates most on 2D or 3D. There are things that one can say definition here and there, but I think it is not a good idea to use the definition there. Actually they don't seem to be definitions.

Strang's "Linear Algebra", Barret O'neill's "Elementary Differential Geometry" deal only with 2D or 3D cases. I think physicist are more interested in the general case, due to the theory of relativity. I found one explanation in "Geometrical Methods of Mathematical Physics" by Bernard Schutz. But I think it doesn't define the rotation matrix.

Artin's "Geometric Alegebra" defines the rotation group as an isometry $\sigma:V\to V$ such that $\det\sigma=1$. But the language there is so abstract that I can't catch any of them.

Can anyone give a reference that defines rotation transformation on $\mathbb R^n$ and state as a property that $A$ is a rotation matrix if and only if $A\in SO(n)$?

This is the end of the question and the below is what I wanted to do. I wanted to prove that if $A\in SO(n)$, then $A$ is a rotation about a line through the origin in $\mathbb R^n$. So I need to define the rotation transformation(or matrix) in $n$ dimensional Euclidean space.

Asaf Karagila
  • 405,794
govin
  • 355

1 Answers1

10

The definition Artin gives is probably the most transparent and standard:

  1. It is a linear isometry, meaning that it preserves distances and lengths as given by a bilinear form (You may want to restrict yourself to the dot product in $\mathbb R^n$).
  2. It has determinant $1$, meaning that it doesn't change the orientation of the space by reflecting it ($\mathbb R$-linear isometries must have determinant $\pm 1$, so we are excluding half with this condition.)

I think you should spend time trying to understand this rather than discarding it as "too abstract." It's really quite concrete.

But I think it doesn't define the rotation matrix.

To split hairs for a second, the most important nature of a rotation is that it is a transformation, not just a matrix. A matrix is just a particular way to represent a transformation. That's why the definition above emphasizes the qualities that make it a rotation: it does not mess with distances, and it does not mess with orientation. (Out of infinitely many choices of bases it can have infinitely many different matrix representations.)

I wanted to prove that if $A\in SO(n)$, then $A$ is a rotation about a line through the origin in $\mathbb R^n$.

That isn't even true in $\mathbb R^2$, as I'm sure you can see.

It happens to be true for rotations in $\mathbb R^n$ for odd $n$ due to the fundamental theorem of algebra, because it says each rotation (like in Artin's definition) has an eigenvector. But even then I doubt it's what you wanted: some rotations in $\mathbb R^n$ preserve much more than just one line through the origin.

And for even $n$'s, you still might not preserve any line: take, for example

$\begin{bmatrix}0&-1&0&0 \\ 1&0&0&0 \\ 0&0&0&-1\\ 0&0&1&0\end{bmatrix} $ as an $\mathbb R$ linear transformation.

rschwieb
  • 160,592
  • Thanks. Actually, I postulated that the above statement holds for odd $n\ge3$. In my argument, I used the fact that $A\in SO(n)$ has 1 as an eigenvalue, which followed from $\det(-B)=-\det B$. – govin May 26 '20 at 15:13
  • 1
    @shyzealot08 Yes, that much is true. The eigenspace for $1$ could have many more dimensions than $1$, though, if you've accounted for that. – rschwieb May 26 '20 at 15:15
  • (About the original question) If the definition of rotation matrix were an isometry whose determinant(regardless of the basis of $V$) is one, then it seems to be the same as the definition of being an element of $SO(n)$, provided $SO(n)$ is a collection of orthogonal matrices whose determinant is $1$. So the statement I wanted to show becomes ridiculous, since it is "if p, then p". So, now I'm not sure that I can state anything that is meaningful. – govin May 26 '20 at 15:20
  • @shyzealot08 It's slightly different, in that you have fixed a basis to produce the matrices. – rschwieb May 26 '20 at 15:21
  • So, can I define the rotation transformation as follow? "It is a linear isometry. If I fix a basis for $V$ or ($\mathbb R^n$), then the corresponding matrix has determinant 1." – govin May 26 '20 at 15:30
  • @shyzealot08 You can... the determinant will be $1$ no matter what basis you pick of course... – rschwieb May 26 '20 at 15:32
  • I see. I think that's because $A=PBP^{-1}$. – govin May 26 '20 at 15:39