2

What I mean by hypercomplex numbers are quaternions, numbers written as $a+bi+cj+dk$ where $i²=j²=k²=-1$

The subtility, is that $ij=k, ji=-k, jk=i, kj=i, ki=j, ik=-j$ The multiplication isn't commutative. Fortunately, with $e$, we can easily compute a real number to the power of any quaternion, but how should we define something like $i^j$?

We could use exponential rules such as $i^j=e^{ln(i)j}=e^{\frac{\pi}{2}i×j}$, but the multiplication isn't commutative. So what should we do? $\frac{\pi}{2}ij$ or $j\frac{\pi}{2}i$.

If we do the first option, we have $i^j=k$, but if we do the 2nd we get $i^j=-k$. The results suspiciously resembles the multiplication, and actually it does.

So I think that, with quaternion units only, we should define $a^b$ as $ab$, but I'm not sure it's a good approach. And how should we extend this in order to define any quaternion exponentiation ?

  • 1
    You probably want $(a^b)^c = a^{bc}$ so $a^b = e^{\ln(a)b}$ would be the way to go. – Nicholas Todoroff May 15 '23 at 17:04
  • 1
    Short answer: there is no solution that will satisfy everything you want it to. You have to make some choices about what properties you really want. – Captain Lama May 15 '23 at 17:06
  • That would make sense with my def since $i^j=k, (i^j)^k=k^k=i^{jk}=i^i$, and that is a real number ($e^-\frac{\pi}{2}≈0.2$) – Pierre Carlier May 15 '23 at 17:06
  • Yeah that's the main problem – Pierre Carlier May 15 '23 at 17:08
  • In short exponentiation is much harder than most people make it out to be (see my answer here https://math.stackexchange.com/questions/4495918/when-does-xyz-xyz/4495968#4495968 for some of the problems) and is only really meaningfully defined when you raise a number to a rational power. I don't know about the specifics of quaternion exponentiation but I suspect that non-commutativity will make it highly nontrivial. The real question is why you want to define exponentiation for quaternions, that will determine what properties are important and whether it is possible. – Fishbane May 15 '23 at 19:11
  • @NicholasTodoroff It is worth pointing out that due to the non-commutativity of the quaternions it is not possible to define a surjective "exponential function" (where that is taken to mean a homomorphism from the aditive structure to the multiplactive structure). This means that it isn't clear what you mean by $\ln (x)$ and $e^x$ in the context of quaternions, I assume you mean the analytic definition of $\exp$ and $\ln$ extended to the quaternions, but it isn't immediately clear that these would have desirable properties. – Fishbane May 15 '23 at 20:19
  • @Fishbane I wasn't saying anything beyond pointing out OP might desire that particular exponential rule. That aside, it is actually very common to define $\exp$ on quaternions using its Taylor series, in which case $\exp$ maps onto all non-zero quaternions. This definition of $\exp$ is very practical and well-studied since it is intimately related to 3D rotations. – Nicholas Todoroff May 16 '23 at 03:02
  • To be fair, I just really like to extend operations to strange mathematical objects. – Pierre Carlier May 16 '23 at 07:08
  • @Fishbane: there is a standard notion of exponential in the theory of Lie algebras and Lie groups that generalises what goes on in the commutative case. The exponential maps the Lie algebra given by the tangent space $T$ of a Lie group $G$ at $e$ onto the connected component of of $G$ containing $e$. It restricts to a homomorphism from the additive structure on any line through the origin in $T$ to a one-dimensional subgroup of $G$. It has been widely studied and has many good properties. – Rob Arthan May 17 '23 at 20:49
  • @RobArthan The problem I have is that there is no obvious reason to use the exponential maps from Lie groups in this context. Exponential maps are related to differential equations, this question seems to be about raising one number to the power of another and the two are not really very related when you are looking at quaternions (or almost all lie groups). Also it isn't clear how to interpret the product of $\ln(a)$ and $b$ due to one being in the tangent space and one being a group element (yes they could both be interpreted as matrices but there is no reason for that to be meaningful). – Fishbane May 17 '23 at 23:43
  • @Fishbane: My comment is about the question in the title of the post. We can naturally identify the tangent space of the multiplicative group $\Bbb{H} \setminus {0}$ at $1$ with the additive group $\Bbb{H}$. Under this identification, the standard exponential from Lie group theory, gives us $e^0 = 1$ and $e^{x+q} = e^xe^q$ whenever $x$ is real and that is the kind of law the OP should surely be looking for. The algebra tells us that $\ln(q)$ has to be multivalued, just as it is in $\Bbb{C}$, so we can't expect good algebraic identitities involving $\ln$. – Rob Arthan May 18 '23 at 19:53
  • ... and, of course, $i^i$ is not uniquely defined in $\Bbb{C}$, so why would we expect $i^j$ to be uniquely defined in $\Bbb{H}$. – Rob Arthan May 18 '23 at 20:00
  • ... and the exponential maps in the theory of Lie algebras and Lie groups are motivated by geometric considerations not differential equations. – Rob Arthan May 18 '23 at 22:36

1 Answers1

1

I would define it like this: $$ \begin{align*} q_{1}^{q_{2}} &= \left( q_{1} \right)^{q_{2}}\\ q_{1}^{q_{2}} &= \left( e^{\ln\left( q_{1} \right)} \right)^{q_{2}}\\ q_{1}^{q_{2}} &= e^{\ln\left( q_{1} \right) \cdot q_{2}}\\ q_{1}^{q_{2}} &= e^{\overbrace{\left(\ln\left( \left| q_{1} \right| \right) + \operatorname{sgn}\left( q_{1} - a_{1} \right) \cdot \arccos\left( \frac{a_{1}}{\left| q_{1} \right|} \right) \right) \cdot q_{2}}^{=~ q_{3}}}\\ q_{1}^{q_{2}} &= e^{q_{3}}\\ q_{1}^{q_{2}} &= e^{\Re\left( q_{3} \right)} \cdot \left( \cos\left( \left| q_{3} - \Re\left( q_{3} \right) \right| \right) + \sin\left( \left| q_{3} - \Re\left( q_{3} \right) \right| \right) \cdot \operatorname{sgn}\left( q_{3} - \Re\left( q_{3} \right) \right) \right)\\ \end{align*} $$ where $\Re\left( q \right) = a \in \mathbb{R}$ and $\operatorname{sgn}\left( \cdot \right)$ is the Signum Function ($\operatorname{sgn}\left( q \right) = \frac{q}{\left| q \right|} \wedge q \ne 0$).

This definition would be based on a series expansion of $e^{z}$ around the point $z = 0$ ($e^{z} = \sum\limits_{k = 0}^{\infty}\left[ \frac{1}{k!} \cdot z^{k} \right]$). For a derivation of this see Exponential Function of Quaternion - Derivation.

A major problem that arises is that they are not commutative with respect to multiplication, so it makes a difference whether we left- or right-multiply $q_{2}$ in the exponent ($e^{q_{1} \cdot q_{2}} \ne e^{q_{2} \cdot q_{1}}$). Another problem is that $\left( x^{y} \right)^{z} = x^{y \cdot z}$ is not always valid, which we already know from complex numbers.

I didn't make up the formulas. They are also in the Wikipedia Article on the quaternions. I just recombined them: $$e^{q} = e^{a} \cdot \left( \cos\left( \left| q \right| \right) + \sin\left( \left| q \right| \right) \cdot \operatorname{sgn}\left( q - a \right) \right)$$ $$\ln\left( z \right) = \ln\left( \left| z \right| \right) + \arccos\left( \frac{a}{\left| q \right|} \right) \cdot \operatorname{sgn}\left( q - a \right)$$

Properties that would emerge from the definition are (where $q_{n} = a_{n} + b_{n} \cdot i + c_{n} \cdot j + d_{n} \cdot k$): $$ \begin{align*} \Re\left( q' \right) = \Re\left( q_{1}^{q_{2}} \right) &= e^{a_{3}} \cdot \cos\left( \sqrt{b_{3}^{2} + c_{3}^{2} + d_{3}^{2}} \right)\\ q' - \Re\left( q' \right) &= e^{a_{3}} \cdot \sin\left( \sqrt{b_{3}^{2} + c_{3}^{2} + d_{3}^{2}} \right) \cdot \frac{b_{3} \cdot i + c_{3} \cdot j + d_{3} \cdot k}{\sqrt{b_{3}^{2} + c_{3}^{2} + d_{3}^{2}}}\\ \end{align*} $$


But this is kind of boring, so let's take another approach: We know that $z^{a} = \operatorname{_{1}F_{0}}\left( -a;\, \cdot;\, z - 1 \right)$ holds, where $\operatorname{_{1}F_{0}}$ is a generalized hypergeometric function given by $\operatorname{_{1}F_{0}}\left( a;\, \cdot;\, z \right) = \sum\limits_{k = 0}^{\infty}\left[ \frac{\Gamma\left( a + k \right)}{\Gamma\left( a \right) \cdot k!} \cdot z^{k} \right]$ where $\Gamma\left( \cdot \right)$ is the Complete Gamma Function.

So: $$q_{1}^{q_{2}} = \sum\limits_{k = 0}^{\infty}\left[ \frac{\Gamma\left( -q_{2} + k \right)}{\Gamma\left( -q_{2} \right) \cdot k!} \cdot q_{1}^{k} \right]$$

I'll leave generalizing this for quaternion as a little exercise for you.

The Art Of Repetition
  • 2,428
  • 1
  • 7
  • 34