This has been confusing for a while, so I'll provide my argument in the hope that is it correct.

Suppose $ Q $ is an orthogonal matrix ($ Q^{T} Q = I $). It is known that orthogonal matrices preserve the dot product, as $ (Q\mathbf{x}) \cdot (Q\mathbf{y}) = (Q\mathbf{x})^{T} (Q \mathbf{y}) = \mathbf{x}^{T} Q^{T} Q \mathbf{y} = \mathbf{x}^{T} \mathbf{y} = \mathbf{x} \cdot \mathbf{y} $ ).

Thus mapping from one orthonormal basis $$ [ \mathbf{e}_{1},\mathbf{e}_{2},\mathbf{e}_{3}] $$ with $ \mathbf{e}_{i} \cdot \mathbf{e}_{j} = \delta_{ij} $ gives another $$ [R \mathbf{e}_{1}, R \mathbf{e}_{2}, R \mathbf{e}_{3}] $$ with $ R\mathbf{e}_{i} \cdot R \mathbf{e}_{j} = \delta_{ij} $, as dot product is preserved.

Morever, this means the lengths of all the vectors and the angles between them are preserved.

Alright, nearly there. So why does $ | Q | = 1 $ mean it must be a rotation? Suppose $ \mathbf{e}_{1},\mathbf{e}_{2},\mathbf{e}_{3} $ form a right-handed set, ie. they are orientated such that

$$ \mathbf{e}_{1} \cdot (\mathbf{e}_{2} \times \mathbf{e}_{3}) = 1 $$

(We know this is one as being orthonormal, they are unit vectors after all)

Knowing the relation between the determinant and the scalar triple product for a 3 x 3 matrix, , if $ | R | = 1 $ it follows that

$$ R \mathbf{e}_{1} \cdot ( R \mathbf{e}_{2} \times R \mathbf{e}_{3} ) = 1 $$

So it is clear that these vectors are orientated in the same way, hence this transformation must be a rotation. (A similar argument can show $|R| = -1$ gives a reflection).

Does that all make sense? Have I missed anything? I'm interested in seeing a nicer argument if people have got one. Also, it's tricky to see how this generalises to $n$ dimensions; shedding some light there would be appreciated.


I have looked at this question, A proof that an orthogonal matrix with a determinant 1 is a rotation matrix but as you can see the argument is pretty bare... so this is not a duplicate.

  • $\begingroup$ Possible duplicate of A proof that an orthogonal matrix with a determinant 1 is a rotation matrix $\endgroup$ – Ravi Jul 17 '17 at 21:58
  • $\begingroup$ What is your definition of “rotation?” What distinguishes it from other linear transformations? $\endgroup$ – amd Jul 17 '17 at 22:25
  • $\begingroup$ @amd a rotation fixes only one point (namely, the origin) whereas reflections leave an entire plane fixed (in 3 dimensions) $\endgroup$ – Christopher Turnbull Jul 18 '17 at 15:18
  • $\begingroup$ There is no 3D rotation that fixes only a single point. $\endgroup$ – amd Jul 18 '17 at 17:28
  • $\begingroup$ sorry, *at least one point. sure, that corresponds to a whole axis in 3d $\endgroup$ – Christopher Turnbull Jul 18 '17 at 20:55

A (real) $3 \times 3$ orthogonal matrix $Q$ with determinant $1$ has a characteristic polynomial $P(\lambda) = \det(Q - \lambda I)$ that is a cubic with real coefficients. The coefficient of $\lambda^3$ is $-1$, while the constant coefficient is $\det(Q) = 1$, therefore there must be a positive eigenvalue, and (because $Q$ preserves lengths) this positive eigenvalue can only be $1$. Thus there is some nonzero vector $w$ that $Q$ leaves fixed. Since $Q$ preserves the dot product, it maps the plane $V$ orthogonal to $w$ to itself. The restriction of $Q$ to $V$ corresponds to $2 \times 2$ orthogonal matrix with determinant $1$. From analysis of $2 \times 2$ orthogonal matrices with determinant $1$, we find that $Q$ acts as a rotation on $V$. Thus $Q$ is a rotation of $3$-dimensional space around an axis in the direction of $w$.

  • $\begingroup$ Interesting approach, would this generalize easily to higher dimensions? $\endgroup$ – Christopher Turnbull Jul 17 '17 at 22:13

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.