Suppose we have a vector field over real numbers $\mathbb R^2$. In additon to vector field proporties define inner product $(x,y) = x_1\cdot y_1 + x_2\cdot y_2$, where $x_1,x_2,y_1,y_2$ are real numbers.This structure is called $2$-dimensional Euclidean Space. ( call it $\text{E-2}$ ).

I want to define complex number field from $\text{E-2}$. I have to satify $5$ addition axiom, $5$ multiplication axiom and a distribution axiom. The problem is although addition axioms are trivial, for multiplication I have to define; $xy = (ac-bd,ad+bc)$ where $x = ( a,b )$, $y = (c,d)$ as an extra axiom.

My questions are:

  1. Is there any natural way to define complex numbers without such additional axioms, by using inner product ?
  2. By defining complex numbers from $\text{E-2}$, there is no need to use inner product between two numbers, this bothers me as well, am I missing something ?
  • $\begingroup$ You mean (trivial) vector bundle, not vector field. $\endgroup$
    – WillO
    Aug 18, 2015 at 12:49
  • 2
    $\begingroup$ In the Euclidean plane there is no preferred direction, but in the complex plane the real line is special (it's invariant under complex conjugation). So you can't express complex multiplication (in a coordinate-free way) using the inner product only. $\endgroup$ Aug 18, 2015 at 13:45
  • 1
    $\begingroup$ You should probably make a distinction between your inner product notation and your ordered pair notation. $\endgroup$ Aug 20, 2015 at 22:31
  • 2
    $\begingroup$ You may be interested in this. See pages 7 and 8 for a "twisting" approach to constructing division algebras from $\Bbb R$. See page 11 for the Clifford Algebra/inner product way to construct division algebras from $\Bbb R$. $\endgroup$
    – pjs36
    Aug 21, 2015 at 0:08

1 Answer 1


I will denote the (standard) inner product by $\langle\cdot,\cdot\rangle$ and consider the elements of $\Bbb R^2$ as column vectors. Put $$A=\begin{bmatrix}1 & 0\\0 & -1\end{bmatrix}$$ and $$B=\begin{bmatrix}0 & -1\\1 & 0\end{bmatrix}.$$

We define an operation $\otimes$ on $\Bbb R^2$ by $$\vec w\otimes\vec z:=\langle\vec w,A\vec z\rangle\vec e_1+ \langle\vec w,BA\vec z\rangle\vec e_2,$$ where $$\vec e_1=\begin{bmatrix}1\\0\end{bmatrix}\text{ and }\vec e_2=\begin{bmatrix}0\\1\end{bmatrix}.$$

It's rather contrived, but it does the job. Note that left-multiplication by $A$ is the operation of complex conjugation, that $\vec e_1$ is the $\otimes$-identity, and that left-multiplication by $B$ is the same as $\otimes$-multiplication by $\vec e_2,$ which serves as our "$i$".

Let me see if I can shed some light on the connection between inner product and complex multiplication (and how I developed the peculiar operation above).

First, I observed that, if there was some function $f:\Bbb C^2\to\Bbb R$ satisfying $$f(a+bi,c+di)=ac+bd$$ for all $a,b,c,d\in\Bbb R$ (that is, if $f$ acts like our inner product), then $$w\overline z=(w_1+iw_2)(z_1-iz_2)=w_1z_1+w_2z_2+i(w_2z_1-w_1z_2)=f(w,z)+i(w_2z_1-w_1z_2)$$ and similarly $$\overline wz=f(w,z)+i(w_1z_2-w_2z_1),$$ whence we find that $$f(w,z)=\frac12\left(w\overline z+\overline wz\right).$$ Readily, then, we have that $$f(w,\overline z)=\frac12\left(wz+\overline w\overline z\right),$$ and (a bit less obviously) that $$f(w,i\overline z)=\frac12\left(-iwz+i\overline w\overline z\right)=\frac1{2i}\left(wz-\overline w\overline z\right).$$ Therefore we have that $$f(w,\overline z)\cdot 1+f(w,i\overline z)\cdot i=wz.\tag{$\star$}$$ All that remains, then, is to decide what the equivalents to $1$ and $i$ would be in $\Bbb R^2$ (which was fairly natural), and how to express conjugation and multiplication by $i$ as linear transformations on $\Bbb R^2$ (not too tricky).

Added: We have some freedom in the representatives we can choose, but we can't just pick any two vectors. Let's call our vector representatives of $1$ and $i$ by the names $\vec 1$ and $\vec i,$ respectively. Let's first find some necessary conditions for them.

First of all, note that we obviously need $\vec 1$ and $\vec i$ to be linearly independent. Assume for the moment that there exist $2\times 2$ real matrices $C$ and $R$ corresponding to complex conjugation and multiplication by $i$ (the latter of which rotates the plane). In particular, this means that the following hold:

  • $C\vec 1=\vec 1$
  • $R\vec 1=\vec i$

By our work in the above section--in particular, by translating $(\star)$ into the desired terms--we find that we need $$\langle\vec w,C\vec z\rangle\vec 1+\langle\vec w,RC\vec z\rangle\vec i=\vec w\otimes\vec z\tag{$\heartsuit$}$$ for all $\vec w,\vec z\in\Bbb R^2.$

Now, since $\vec 1$ needs to be our $\otimes$-identity, then in particular, $$\vec 1=\vec1\otimes\vec1=\langle\vec 1,C\vec 1\rangle\vec1+\langle\vec1,RC\vec1\rangle\vec i=\langle\vec1,\vec1\rangle\vec1+\langle\vec1,R\vec1\rangle\vec i=\langle\vec1,\vec1\rangle\vec1+\langle\vec1,\vec i\rangle\vec i,$$ whence linear independence shows us that $\langle\vec 1,\vec1\rangle=1$ and $\langle\vec1,\vec i\rangle=0.$ Another property we need is that $\vec1=-\vec i\otimes\vec i,$ from which we can further deduce that $\langle\vec i,\vec i\rangle=1.$ Therefore, we require that $\left\{\vec1,\vec i\right\}$ be an orthonormal basis for $\Bbb R^2.$

From this, we can conclude that $$\vec1=\begin{bmatrix}\cos\theta\\\sin\theta\end{bmatrix}$$ for some $\theta\in\Bbb R.$ Now, having chosen such a vector, we have two choices for $\vec i.$ (Can you see why, and what they are?) This might seem a bit alarming, but many constructions of $\Bbb C$ run into precisely this same problem, as discussed for example here and here. Regardless, it turns out that we must have either $R=B$ or $R=-B$ (with $B$ as given earlier), which depends on $\vec1$ and on which of the two options we choose for $\vec i.$

One final observation I will make (omitting the proof) is that we require $$C=\begin{bmatrix}\cos2\theta & \sin2\theta\\\sin2\theta & -\cos2\theta\end{bmatrix}.$$ In the case that $\theta$ is an integer multiple of $2\pi$--that is, that $\vec1=\vec e_1$--we have $C=A.$

I leave it to you to verify that the necessary conditions derived above are also sufficient to make $\otimes$ behave on $\Bbb R^2$ exactly the way we want complex multiplication to work. Observe, though, that there are uncountably-many different ways that we could define $\otimes,$ depending on our choice of $\theta$ and subsequent choice of $\vec i.$ The definition I gave above is about as close to canonical as you can get.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy