I have a proof that every quaternion algebra over a field $A=\left(\frac{a,b}{F}\right)$ is simple, i.e. has no nontrivial two-sided ideals, which appeals to the algebraic closure of $F$ and the elevation of ideals $I\subset A$ to ideals $\overline I\subset\left(\frac{a,b}{\overline F}\right)=M_2(\overline F)$ and uses properties of this latter space.

EDIT: Sorry, I didn't realise this notation wasn't common. $A$ is a four-dimensional algebra over a field $F$, which we assume has characteristic $\neq 2$, spanned by $\{1,i,j,k\}$ thus: $$ i^2=a\,1\quad j^2=b\,1\quad ij=-ji=k. $$ where $a,b\in F^\ast$.

However, I'd like to be able to accomplish this proof without using the algebraic closure, but I'm not sure what the most appropriate method is. Here's my proof:

Take some nonzero $I\subset A$ and examine an arbitrary element $x_0+ix_1+jx_2+kx_3=x\in I\subset A$. Then we know $$ (x_0+ix_1+jx_2+kx_3)(x_0-ix_1-jx_2-kx_3)=x_0^2-ax_1^2-bx_2^2+abx_3^2\in I $$ since $I$ is closed under right (and left) multiplication.

Now it would be great to suggest that this element is nonzero. If that were true, then we'd be done. If it's zero, then we have $$ x_0^2+abx_3^2=ax_1^2+bx_2^2. $$ Is this possible?

LAST EDIT: I found a good proof of what I wanted to do here. Thanks everyone for your help!

  • $\begingroup$ Also, how do you know that $x_0^2+x_1^2+x_2^2+x_3^2$ is nonzero? What if $x_0=x_1=x_2=x_3$ and the field has characteristic 2? $\endgroup$ – rschwieb Mar 25 '13 at 13:38
  • 1
    $\begingroup$ The approach you're taking is to try to show that nonzero elements are invertible, showing that it is a division ring, which is much stronger than being simple. While there's nothing inherently wrong with that (in fact I think it shows some good developing sense!) it is a little suspicious that it is trying to prove more than the question asked :) $\endgroup$ – rschwieb Mar 25 '13 at 13:47
  • 1
    $\begingroup$ Correct me, if I'm wrong, but when we have $\left(\dfrac{a,b}{F}\right)$ don't we have something like $x_1^2=a$, $x_2^2=b$, $x_3^2=-ab$? These are not always division algebras, but they are always simple (unless we're in characteristic two, as rschwieb points out). For many choices of $(a,b)$ we get the 2x2 matrix algebra instead (which is still simple). $\endgroup$ – Jyrki Lahtonen Mar 25 '13 at 20:07
  • $\begingroup$ Thanks for clarifying! That makes sense now, so I'll yank the original comment. $\endgroup$ – rschwieb Mar 25 '13 at 21:02
  • 1
    $\begingroup$ Hm, using that scheme, though, the product of an element with its conjugate should contain some $a$'s and $b$'s, right? $\endgroup$ – rschwieb Mar 25 '13 at 21:17

I'll retract this if whatever $A=\left(\frac{a,b}{F}\right)$ means prohibits my example :)

I decided to run fast with my "what if $F$ has characteristic two" comment fast and mention why the quaternions over a field of characteristic 2 are never a simple ring.

Since $-1=1$ in such a ring, the generators wind up commuting instead of anticommuting, so the whole thing winds up commutative. Thus the ring would be simple iff it is a field. However, as mentioned, $(1+i+j+k)\cdot(1-i-j-k)=0$ in that characteristic, and so it is clearly not a field.

Thanks for clarifying the original post. That construction is just the Clifford algebra of a two dimensional space with the symmetric bilinear form $\begin{bmatrix}a&0\\0&b\end{bmatrix}$. You will find a lot said about this when the form is nondegenerate (that is, $a,b$ are both nonzero.) If you check out the link, you can see that for real algebras, an even dimensional vector space with a nondegenerate form is isomorphic to a full matrix ring over a division ring (which is necessarily simple).

I wish I could instantly recall if the same is true for rings of characteristic other than 2. My memory is telling me something like that appears (or at least, the proof for "quaternion algebras") in Jacobson's Basic Algebra I in the section on the Hurwitz problem (7.6, I see from the googlebooks preview).

  • 1
    $\begingroup$ I should have say that we assume $F$ has characteristic $\neq2$. And I'll go up and explain my notation in the initial question... $\endgroup$ – Ian Coley Mar 25 '13 at 20:28

The algebra $A=\left(\dfrac{a,b}{F}\right)$ is simple when $ab\neq0$, and $2\neq0$ in $F$.

You haven't used the fact that an ideal must be 2-sided. Let's see how that helps.

Let $I$ be any non-zero ideal, and let $q=x_0+x_1i+x_2j+x_3k$ be an arbitrary element non-zero element of $I$. Then using the multiplication rules (and their consequence $ki=-jii=-aj, ik=iij=aj$) we see that both $$ qi=-ax_1+x_0i-ax_3j-x_2k $$ and $$ iq=-ax_1+x_0i+ax_3j+x_2k $$ are in $I$. As $2$ is invertible in $F$ we have that both $$ \frac{iq-qi}2=ax_3j+x_2k\in I $$ and $$ \frac{iq+qi}2=-ax_1+x_0i\in I. $$ Furthermore, one of them is non-zero. Because $j$ is invertible $ax_3j+x_2k=(ax_3+x_2i)j$ and $ax_3+x_2i$ generate the same ideal, so we can deduce in either case that the ideal $I$ contains a non-zero element of the form $z=y_0+y_1i$. Then also the following elements are in $I$: $$ zj=y_0j+y_1k\in I\qquad\text{and}\qquad jz=y_0j-y_1k\in I. $$ Repeating the above trick this implies that $$ \frac{zj-jz}2=y_1k\in I\qquad\text{and}\qquad\frac{zj+jz}2=y_0j\in I. $$ At least one these non-zero. But any non-zero $F$-multiple of any of the basis elements is a unit in $A$, so we can conclude that $I$ contains a unit of $A$ and is thus all of $A$.

Compare this with the exercise you have probably had earlier. The algebra of 2x2-matrices over any field has no non-zero ideals (i.e. it is simple - the same applies to matrix algebras of all sizes). That algebra does have non-trivial left (resp. right) ideals. Consider the sets of matrices where a fixed column (resp. row) is constrained to be all zeros. The upshot here is that in non-commutative algebra 2-sidedness of an ideal is a much stronger requirement. The early examples of ideals a student sees are more often than not in commutative rings, and things are simpler there.


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.