Suppose I use the Cayley-Dickson construction on complex numbers to obtain the quaternions. I then have the "real" and the "imaginary" parts of the quaternions.
The argument of a complex number $x$ is then $\mathrm{atan2}(\mathrm{Im}(x), \mathrm{Re}(x))$ where $\mathrm{atan2}$ is like the regular arctangent function but takes into account the quadrant where the angle lies.
Shouldn't the argument of a quarternion $x$ then be a complex number that is something like $\mathrm{catan}\left(\frac{\mathrm{Im}(x)}{\mathrm{Re}(x)}\right)$ where $\mathrm{catan}$ extends the arctangent to the complex plane? The trouble I'm having here is that I need to take into account the proper "quadrants" or actually octants and I don't know how that'd work.
Given that the logarithm of a complex number $r \times e^{i \theta}$ is $\ln r + i \theta$ (plus all the other angles) the logarithm of a quaternion $x$ should be $\ln \left\lvert x \right\rvert + ix_\theta$. However, $\ln \left\lvert x \right\rvert$ for a quaternion is purely real (not just "real") and it has no $i$, $j$, or $k$ components. So the straightforward generalization of the argument misses out one one component of the quaternion.
Basically, to make the math nice the argument of a quaternion shouldn't be a complex number but instead a quaternion lacking a real component. I suspect that these three numbers would be the Euler angles. There should really be a name for this one particular case.