1
$\begingroup$

Given a quaternion of the form,

$$q= a + bi + cj + dk$$

Which is the norm of $q$?

(1) $\sqrt{a^2+b^2+c^2+d^2}$

(2) $a^2+b^2+c^2+d^2$

This page from MathWorks says (1) but another page says (2). Wikipedia says (1). My lecture slides say:

$$|q|^2=a^2+b^2+c^2+d^2$$

but it is not clear whether $|q|$ is the norm or $|q|^2$ is the norm. It is not clear whether $||^2$ is just the L2 norm notation or it is the power of 2 notation.

I am trying to figure this out because I am trying to normalize and compute inverses of quaternions. Both of these operations depend on the norm.

$\endgroup$
  • 1
    $\begingroup$ Conventions vary. But which one of these does normalise $q$? How do you express the inverse in terms of each? You can easily work this out yourself. $\endgroup$ – ancientmathematician Apr 4 at 6:37
  • 2
    $\begingroup$ You wouldn't use $|q|^2$ as a norm, for if $c\in\mathbb{R}$ we would have $|cq|^2=|c|^2|q|^2$ were $|c|$ is just absolute value. We want to have absolute scalability, so $|q|^2$ refers to the norm squared. $\endgroup$ – Melody Apr 4 at 6:38
  • $\begingroup$ Agree with @ancientmathematician. Conventions vary. In pure algebra (ignoring all real analysis), the norm of $q$ would be the determinant of left multiplication by $q$ which comes out as $N(q)=(a^2+b^2+c^2+d^2)^2$. A degree four polynomial on the coefficients reflecting the fact that the quaternions form a 4-dimensional space. But, when studying division algebras, this would be replaced with the reduced norm $nr(q)=a^2+b^2+c^2+d^2$. Of course, if your context dictates that the norm should have something to do with the length of a vector, then you would take the square root one more time. $\endgroup$ – Jyrki Lahtonen Apr 4 at 6:49
  • $\begingroup$ From the point of view calculating inverses I would use the reduced norm, for then we have $$q^{-1}=\frac{\overline{q}}{nr(q)}.$$ If you want to scale them to unit quaternions, then you normalize as $q\mapsto q/\sqrt{a^2+b^2+c^2+d^2}$ (handy when using quaternions to represent rotations as in 3D-graphics programming). But, really, the choice is very much context dependent. Pick one, and stick to it. $\endgroup$ – Jyrki Lahtonen Apr 4 at 6:50
  • $\begingroup$ The reason for this variety is that norm is one of the more overloaded terms in math. Here there's a bit of a clash between various uses. As long as the readers are aware of the different possibilities, and thus know to check the local convention, we'll do fine :-) $\endgroup$ – Jyrki Lahtonen Apr 4 at 6:58
0
$\begingroup$

From a practitioner point of view, you have two cases:

  1. Your quaternion is not normalized, i.e., $Q Q^* \neq 1$, then the rotation is:

$$v' = Q v Q^{-1}$$

Where $Q^{-1} = Q^* / \|Q\|^2$ and $\|Q\|^2 = a^2 + b^2 + c^2 + d^2$.

As you see the nornalization term must be squared since Q appears twice in the quaternion ("sandwitch") product.

  1. Your quaternion is unit quaternion, i.e., $Q Q^* = 1$, then the rotation is just:

$$v' = Q v Q^*$$

Now, a unit quaternion is just $Q / \|Q\|$ where $\|Q\| = \sqrt { a^2 + b^2 + c^2 + d^2}$.

$\endgroup$

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.