Prove that if $\mathbf{x},\mathbf{y},\mathbf{z} \in \mathbb{R}^3$ are coplanar vectors and $\mathbf{N}$ is a unit normal vector to the plane then $$(\mathbf{x}\times\mathbf{y} \cdot \mathbf{N})\ \mathbf{z} + (\mathbf{y}\times\mathbf{z} \cdot \mathbf{N})\ \mathbf{x} + (\mathbf{z}\times\mathbf{x} \cdot \mathbf{N})\ \mathbf{y}=\mathbf{0}.$$

This is an elementary identity involving cross products which is used in the proof of the Gauss-Bonnet Theorem and whose proof was left as an exercise. I've tried it unsuccessfully. Initially I tried writing $\mathbf{N}=\frac{\mathbf{x}\times\mathbf{y}}{\| \mathbf{x}\times\mathbf{y}\|}=\frac{\mathbf{y}\times\mathbf{z}}{\| \mathbf{y}\times\mathbf{z}\|}=\frac{\mathbf{z}\times\mathbf{x}}{\| \mathbf{z}\times\mathbf{x}\|}$ and substituting into the equation to get $\| \mathbf{x}\times\mathbf{y}\|z +\| \mathbf{y}\times\mathbf{z}\|\mathbf{x}+\| \mathbf{z}\times\mathbf{x}\|\mathbf{y}=\mathbf{0}$ but then I realised these terms are only correct up to $\pm$ signs. You could write the norms in terms of sines of angles and divide by norms to get unit vectors with coefficients $\sin\theta,\sin\psi,\sin(\theta+\psi)$ (or $2\pi -(\theta+\psi)$ I suppose) but I don't know what to do from there, especially when the terms are only correct up to sign. Any hints how to prove this identity? Perhaps there is a clever trick to it but I can't see it. Edit: Maybe writing $\mathbf{z}=\lambda\mathbf{x}+\mu\mathbf{y}$ will help.

  • 1
    $\begingroup$ What does $x \times y \cdot N$ mean? Dot product $(x \times y) \cdot N$ ? $\endgroup$ – Widawensen Mar 20 '19 at 12:51
  • 4
    $\begingroup$ @Widawensen Yes, what else could it mean? $\endgroup$ – Marc van Leeuwen Mar 20 '19 at 12:53
  • 1
    $\begingroup$ @MarcvanLeeuwen: that could mean a badly written problem. That happens here sometimes. $\endgroup$ – Taladris Mar 21 '19 at 2:54

Here's an observation: If $Q$ is a rotation matrix, then $$ (Qx) \times (Qy) = Q(x \times y) $$

You have to prove that, of course, but it's not too tough. Similarly, $$ (Qx) \cdot (Qy) = x \cdot y $$ and, for a scalar $\alpha$, we have $$ Q (\alpha x) = \alpha (Q x) $$

Now suppose that for some vector $v$, we have $$ (\mathbf{x}\times\mathbf{y} \cdot \mathbf{N})\ \mathbf{z} + (\mathbf{y}\times\mathbf{z} \cdot \mathbf{N})\ \mathbf{x} + (\mathbf{z}\times\mathbf{x} \cdot \mathbf{N})\ \mathbf{y}=\mathbf{v}. $$

Key idea 1: You can apply the rules above to show that for any rotation matrix $Q$, you can apply $Q$ to all the elements on the left to get $Qv$.

Key idea 2: You can choose $Q$ so that it takes $N$ to the vector $(0,0,1)$, and puts $x, y,$ and $z$ into the plane consisting of vectors of the form $(a, b, 0)$. And in that plane, it's easy to see that you get $0$, so $Qv = 0$. Hence $v = 0$, and you're done.

In short: by a change of basis, you can assume that $N$ is the vector $(0,0,1)$ and that the other vectors all lie in the $(a, b, 0)$ plane, and things get easy.

  • $\begingroup$ Ah so it is a clever approach with orthogonal matrices. I'm familiar with those identities. But I don't see how $Qv=0$. I can't see how we apply anything other than the third identity $Q(\alpha x)=\alpha (Qx)$. $\endgroup$ – AlephNull Mar 20 '19 at 11:55
  • $\begingroup$ Oh I see, you're talking about the elements, not the terms. I understand the solution now. $\endgroup$ – AlephNull Mar 20 '19 at 13:42
  • 2
    $\begingroup$ By the way, there's a general principle at work here, much loved by physicists, but worth remembering even if that's not your domain: "find the right coordinate system for your problem." (The math version is mostly "choose the right basis/generating set/...") $\endgroup$ – John Hughes Mar 20 '19 at 16:51

If what is required is only to prove the validity of the given identity, there is another approach. Observe that if $x$ and $y$ are linearly dependent, i.e. for some $c$, $x=c y$ or $y=c x$, then the identity holds trivially because $w\times v =-(v\times w)$ and $v \times v=0$ for all $v,w$. Thus, we may assume $x$ and $y$ are linearly independent and hence $z$ is a linear combination of $x$ and $y$, that is, $z=ax+by$ for some $a,b$. Now, since the given identity is linear in each variable and it holds for both $z=x$ and $z=y$, it is also true for $z=ax+by$. This proves the identity. It can be also noted that $N$ being perpendicular to the plane containing $x,y,z$ plays no role in this proof.

  • 2
    $\begingroup$ very nice solution! $\endgroup$ – John Hughes Mar 20 '19 at 16:54
  • $\begingroup$ Indeed, this is very elegant. So my last remark had some significance! $\endgroup$ – AlephNull Mar 20 '19 at 17:06
  • $\begingroup$ Thank you both :-) $\endgroup$ – Song Mar 21 '19 at 7:07

Writing $x=a\hat{i}+b\hat{j},\,y=c\hat{i}+d\hat{j},\,z=e\hat{i}+f\hat{j},\,N=N\hat{k}$ reduces the sum to $$N((ad-bc)(e\hat{i}+f\hat{j})+(cf-de)(a\hat{i}+b\hat{j})+(be-af)(c\hat{i}+d\hat{j})).$$The $\hat{i}$ coefficient is $N(ade-bce+acf-ade+bce-acf)=0$. The $\hat{j}$ coefficient can be handled similarly.

  • $\begingroup$ I accepted a different answer but +1 because I appreciate that this finishes off the solution in that answer. $\endgroup$ – AlephNull Mar 20 '19 at 13:45

Since $\bf x, \bf y, \bf z$ are coplanar, they are linearly dependent. Since the result to be proved is symmetric in $\bf x, \bf y, \bf z$, withouht loss of generality we can write $\bf z = \lambda \bf x + \mu \bf y$ for some scalars $\lambda, \mu$.

Now, $$\begin{align} & (\bf y \times \bf z \cdot \bf N)\; \bf x \\ =\ & (\bf y \times (\lambda \bf x + \mu \bf y) \cdot \bf N)\; \bf x \\ =\ & (\bf y \times \lambda \bf x \cdot \bf N)\; \bf x \\ =\ & (\bf y \times \bf x \cdot \bf N)\, (\lambda \bf x) \end{align}$$ and similarly $$\begin{align} & (\bf z \times \bf x \cdot \bf N)\; \bf y \\ =\ & (\bf y \times \bf x \cdot \bf N)\, (\mu \bf y) \end{align}$$ So $$\begin{align} & (\bf y \times \bf z \cdot \bf N)\; \bf x + (\bf z \times \bf x \cdot \bf N)\; \bf y \\ =\ & (\bf y \times \bf x \cdot \bf N)\, (\lambda \bf x + \mu \bf y)\\ =\ & -(\bf x \times \bf y \cdot \bf N)\;z \end{align} $$ and the result follows.


If you know a little about the exterior algebra we can see this almost immediately, and in a way that generalizes substantially.

Pick any plane $\Pi$ containing ${\bf x}, {\bf y}, {\bf z}$. The map on $\Pi$ defined by $$({\bf a}, {\bf b}, {\bf c}) \mapsto [({\bf a} \times {\bf b}) \cdot {\bf N}] {\bf c} + [({\bf b} \times {\bf c}) \cdot {\bf N}] {\bf a} + [({\bf c} \times {\bf a}) \cdot {\bf N}] {\bf b}$$ is visibly trilinear and totally skew in its arguments, so it is a (vector-valued) $3$-form on a $2$-dimensional vector space and hence is the zero map.

NB this argument doesn't use any properties of $\bf N$.


By the properties of the triple product ( circluar shift) we can rearrange formula:

$\ \ \ \ (\mathbf{x}\times\mathbf{y}) \cdot \mathbf{N})\ \mathbf{z} + (\mathbf{y}\times\mathbf{z}) \cdot \mathbf{N})\ \mathbf{x} + (\mathbf{z}\times\mathbf{x}) \cdot \mathbf{N})\ \mathbf{y} \\ =(\mathbf{N}\times\mathbf{x}) \cdot \mathbf{y})\ \mathbf{z} + (\mathbf{N}\times\mathbf{y}) \cdot \mathbf{z})\ \mathbf{x} + (\mathbf{N}\times\mathbf{z}) \cdot \mathbf{x})\ \mathbf{y} $

All cross product vectors $$v_1=(\mathbf{N}\times\mathbf{x}),v_2=(\mathbf{N}\times\mathbf{y}), v_3=(\mathbf{N}\times\mathbf{z})$$
lie in the plane of coplanar vectors $\mathbf{x},\mathbf{y},\mathbf{z}$ and they are vectors $\mathbf{x},\mathbf{y},\mathbf{z}$ rotated by $\pi/2$ in this plane.

So we can limit themselves to this plane and take any vectors with components $\mathbf{x}=[ x_1 \ \ x_2]^T,\mathbf{y}=[ y_1 \ \ y_2]^T,\mathbf{z} =[ z_1 \ \ z_2]^T$.

Transform them with the rotation matrix $R=\begin{bmatrix} 0 & -1 \\ 1 & 0 \end{bmatrix}$ , calculate appropriate dot products and finally check the formula with these assumed general components.

Namely we need to calculate: $$(y^TRx)z+(z^TRy)x+(x^TRz)y$$


Another approach to the problem uses a formula for triple product.

$ \mathbf{a}\cdot(\mathbf{b}\times \mathbf{c}) = \det \begin{bmatrix} a_1 & b_1 & c_1 \\ a_2 & b_2 & c_2 \\ a_3 & b_3 & c_3 \\ \end{bmatrix} $

Then consider determinant

$\begin{vmatrix} n_1 & x_1 & y_1 & z_1 \\ n_2 & x_2 & y_2 & z_2 \\ n_3 & x_3 & y_3 & z_3 \\ n_1 & x_1 & y_1 & z_1 \end{vmatrix} $

where columns consist of vectors $ \mathbf{N} ,\mathbf{x},\mathbf{y},\mathbf{z}$ components (the fourth row repeats the first one).

Of course such determinant equals to $0$.
Developing the determinant along the fourth row we obtain:

$-n_1\begin{vmatrix} x_1 & y_1 & z_1 \\ x_2 & y_2 & z_2 \\ x_3 & y_3 & z_3 \\ \end{vmatrix} +x_1\begin{vmatrix} n_1 & y_1 & z_1 \\ n_2 & y_2 & z_2 \\ n_3 & y_3 & z_3 \\ \end{vmatrix} -y_1\begin{vmatrix} n_1 & x_1 & z_1 \\ n_2 & x_2 & z_2 \\ n_3 & x_3 & z_3 \\ \end{vmatrix} +z_1\begin{vmatrix} n_1 & x_1 & y_1 \\ n_2 & x_2 & y_2 \\ n_3 & x_3 & y_3 \\ \end{vmatrix}=0$

from which the formula for the first component of the vector given in the question follows

(the first summand is equal to $0$ as the vectors $\mathbf{x},\mathbf{y},\mathbf{z}$ are collinear, the columns can be permuted (required for the third summand) if needed to give appropriate sign in expression)

Similarly the determinants

$\begin{vmatrix} n_1 & x_1 & y_1 & z_1 \\ n_2 & x_2 & y_2 & z_2 \\ n_3 & x_3 & y_3 & z_3 \\ n_2 & x_2 & y_2 & z_2 \end{vmatrix} $ and $\begin{vmatrix} n_1 & x_1 & y_1 & z_1 \\ n_2 & x_2 & y_2 & z_2 \\ n_3 & x_3 & y_3 & z_3 \\ n_3 & x_3 & y_3 & z_3 \end{vmatrix} $

give the second and the third component of the question vector, equal to $0$.


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.