Let $g\in SO(3)$ and $p,q\in\mathbb{R}^3$. I wondered whether it is true that $$g(p\times q)=gp\times gq$$

I am not sure how to prove this. I guess I will use at some point that the last row $g_3$ of $g$ can be obtained by $g_3=g_1\times g_2$.

But I assume there is an easier prove than writing everything out.


You may use the scalar triple product formula $r \cdot (p\times q)=\det(r,p,q)$ to prove that $$ gr \cdot (gp\times gq)=gr \cdot g(p\times q)\tag{1} $$ ($=\det(r,p,q)$) for any vector $r$. Since $g$ is invertible, if $(1)$ holds for every vector $r$, we must have $gp\times gq=g(p\times q)$.

  • $\begingroup$ I know that this proof should fail if $g \in O(3)$, but it's not clear to me where it does. $\endgroup$ – Michael Seifert Nov 25 '17 at 3:15
  • 3
    $\begingroup$ @MichaelSeifert If $g\in O(3)$, you have $\det(gr,gp,gq)=\det(g)\det(r,p,q)$. Therefore the scalar triple product is only preserved if $\det(g)=1$. Otherwise you flip a sign in the determinant and the cross product. $\endgroup$ – Joonas Ilmavirta Nov 25 '17 at 5:00

Two ways of seeing it more directly:

  • The cross product of two vectors can be expressed in terms of the norms of the vectors and the angle between them, and those properties are preserved by rotations. (Update: As mephistolotl mentioned in the comments, you also need the fact that rotations preserve the orientation. Otherwise, you'll get a vector of correct length, but different direction.)
  • Numerically, you would need to show $\epsilon_{ijk} O_{jl} O_{km} =O_{ip}\epsilon_{plm}$, which follows from orthogonality and the fact that $\det O=1$.
  • 6
    $\begingroup$ f(x,y,z) = (y,x,z) preserves lengths and angles, but does not preserve the cross product. The map must also preserve orientation. $\endgroup$ – mephistolotl Nov 24 '17 at 16:34
  • $\begingroup$ @mephistolotl: You're right, I was a bit sloppy there - I'll upate the answer. $\endgroup$ – Toffomat Nov 27 '17 at 9:52

Since both expressions $g(p\times q)$ and $g(p)\times g(q)$ are linear in each of the variables $p,q$, it suffices to check the equality for the nine cases $p,q=e_1,e_2,e_3$, i.e. when $p,q$ are basis vectors. This method is quite often employed when dealing with equality of multilinear functions.


Let $v,w\in\mathbb{R}^3$. Suppose that $v=(x_1,x_2,x_3)$, that $w=(y_1,y_2,y_3)$ and that $u=v\times w=(z_1,z_2,z_3)$. Then $v\times w$ is the only vector $u\in\mathbb{R}^3$ such that

  1. $u$ is orthogonal to both $v$ and $w$;
  2. $\|u\|=\|v\|.\|w\|.\sin\theta$, where $\theta$ is the angle between $u$ and $v$;
  3. $\begin{vmatrix}x_1&y_1&z_1\\x_2&y_2&z_2\\x_3&y_3&z_3\end{vmatrix}\geqslant0$.

Let $g\in SO(3,\mathbb{R})$. Then

  1. $g(u)$ is orthogonal to both $g(v)$ and $g(w)$ (because $g\in SO(3,\mathbb{R})$);
  2. $\bigl\|g(u)\bigr\|=\bigl\|g(v)\bigr\|.\bigl\|g(w)\bigr\|.\sin\theta$ (again, because $g\in SO(3,\mathbb{R})$);
  3. if $g(v)=(x_1',x_2',x_3')$, $g(w)=(y_1',y_2',y_3')$, and $g(u)=(z_1',z_2',z_3')$, then$$\begin{vmatrix}x_1'&y_1'&z_1'\\x_2'&y_2'&z_2'\\x_3'&y_3'&z_3'\end{vmatrix}=\det g\begin{vmatrix}x_1&y_1&z_1\\x_2&y_2&z_2\\x_3&y_3&z_3\end{vmatrix}=\begin{vmatrix}x_1&y_1&z_1\\x_2&y_2&z_2\\x_3&y_3&z_3\end{vmatrix}\geqslant0.$$

Therefore, $g(u)=g(v)\times g(w)$.


If one introduces the cross product in a proper abstract way, this property will be immediate from the definition. First consider any $n$-dimensional real vector space, then each basis has an associated volume form ($n$-linear alternating form), obtained by taking the determinant after expressing the $n$ argument vectors in coordinates with respect to the basis. If the vector space is equipped with an inner product so that it becomes an Euclidean vector space$~E$, then the volume forms of the orthonormal bases for the inner product take two opposite values; choosing one of these volume forms defines the structure of an oriented Euclidean vector space. The automorphism group $O(E)$ of $E$ has a subgroup $SO(E)$ of index$~2$ that fixes the volume forms, and is the automorphism group of the oriented Euclidean vector space.

Now fixing $n-1$ vectors $p_1,\ldots,p_{n-1}$ in an $n$-dimensional oriented Euclidean vector space, there is a linear form $f:v\mapsto\operatorname{Vol}(p_1,\ldots,p_{n-1},v)$, and therefore a unique vector $w$ such that $f(v)=w\cdot v$; by definition, this $w$ is the external product of $p_1,\ldots,p_{n-1}$. For $n=3$, this gives the cross product in an oriented $3$-dimensional Euclidean vector space. Since only the oriented Euclidean structure used in the definition, the external product is automatically compatible with the action of$~SO(E)$; since the (choice of) volume form was actually used in the definition, it is not compatible with the action of$~O(E)$.


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.