In First Course in Noncommutative rings of T.Y.Lam (p.210), the author stated that "It is known that for each $n$, there exists a $\mathbb{Q}$-division algebra $A_n$ of dimension $p_n^2$, with $Z(A_n)=\mathbb{Q}$." Here $p_n$ is a prime number and $Z(A_n)$ is the center of $A_n$.

How to construct $A_n$?



The recipe is the following:

  • Locate an extension field $K$ such that $[K:\Bbb{Q}]=p=p_n$, and $K/\Bbb{Q}$ is Galois with a cyclic Galois group generated by an automorphism $\sigma$. This latter requirement is automatic as a group of prime order is necessarily cyclic, but if you want non-prime index then you need to ascertain this also.
  • Locate an element $\gamma\in\Bbb{Q}$ such that $\gamma\neq N(z)$ for all $z\in K$. Here $N:K\to\Bbb{Q}$ is the norm map: $$N(z)=\prod_{i=0}^{p-1}\sigma^i(z).$$
  • Given all this let $A_p$ be the $p$-dimensional vector space over $K$ with basis $1,u,u^2,\ldots, u^{p-1}$. Here $u$ is an auxiliary generator. $A_p$ becomes a $\Bbb{Q}$-algebra when we define a multiplication by the rules $u^p=\gamma$, and $ux=\sigma(x)u$ for all $x\in K$.
  • The relation $ux=\sigma(x)u$ really forces the center of $A_p$ to be $\Bbb{Q}$ only. Among other things it implies that if an element $x\in K$ commutes with $u$, we need $\sigma(x)=x$, i.e. $x\in\Bbb{Q}$.
  • The requirement $\gamma\neq N(x)$ for all $x\in K$ turns out be a sufficient criterion for $A_p$ not to have zero divisors, hence proving that $A_p$ is a division algebra. This fact (and the previous bullet) are not at all obvious, see e.g. Jacobson's Basic Algebra II, chapter 8, for more details.

A few remarks:

  • You can find the required field $K$ as a subfield of a suitable cyclotomic extension. In fact, by Kronecker-Weber, this is the only way to find such a $K$.
  • Finding a $\gamma$ takes a bit of algebraic number theory. There are many choices that work (I like Lord Shark the Unknown's suggestion of using an inert prime). Two choices $\gamma$ and $\gamma'$ yield isomorphic division algebras, iff $\gamma/\gamma'=N(x)$ for some $x\in K$.
  • You can realize $A_p$ as matrices of the form $$\left(\begin{array}{ccccc} x_0&\sigma(x_{p-1})&\sigma^2(x_{p-2})&\cdots&\sigma^{p-1}(x_1)\\ \gamma x_1&\sigma(x_0)&\sigma^{2}(x_{p-1})&\cdots&\sigma^{p-1}(x_2)\\ \gamma x_2&\gamma\sigma(x_1)&\sigma^2(x_0)&\cdots&\sigma^{p-1}(x_3)\\ \vdots&\vdots&\vdots&\ddots&\vdots\\ \gamma x_{p-1}&\gamma\sigma(x_{p-2})&\gamma\sigma^2(x_{p-2})&\cdots&\sigma^{p-1}(x_0) \end{array}\right).$$ Here $x_0,x_1,\ldots,x_{p-1}$ are arbitrary elements of $K$, $\gamma$ is the prescribed non-norm element and appears as a factor in all the entries below the diagonal. The powers of $\sigma$ increase column-by-column from left to right, and the variables $x_i,i=0,1,\ldots,p-1$, appear in a circulant way.
  • If you want a non-prime index $n$, then the extra requirement is that you need to check that the lowest power $\gamma^\ell$, $\ell>0$, that is a norm of some element occurs only at $\ell=n$.
  • $\begingroup$ Thanks for your answer. Can you give me some ways to find $\gamma$? If $\gamma$ is an inert prime, then how to prove it is not contained in $N(K^*)$ ? $\endgroup$ – Huỳnh Việt Khánh Apr 2 '18 at 9:22
  • $\begingroup$ @HuỳnhViệtKhánh If $\gamma=q$ is an inert prime, then the ideal of the integers $\mathcal{O}_K$ of the field $K$ it generates remains a prime ideal, call it $\mathfrak{q}$. If the norm of an element $x\in K$ were equal to $q$, then the fractional ideal $\mathcal{O}_Kx$ must have $\mathfrak{q}$ as factor (with positive multiplicity). But, the same then holds for all the conjugates of $x$, so the norm will be divisible by $\mathfrak{q}^p$, i.e. with $q^p$. Here $p>1$ so the norm cannot be equal to $q$. $\endgroup$ – Jyrki Lahtonen Apr 2 '18 at 9:35

Let $p$ be a prime. Then there is a prime $q\equiv1\pmod p$, and so the $q$-th cyclotomic field will have a subfield $K$ of degree $p$ over $\Bbb Q$ with cyclic Galois group generated by $\sigma$. There will be a prime $\ell$ which remains inert in $K$.

One now constructs $A$ as a crossed product of $K$. It is a free left $K$ module $A=K\oplus Ku\oplus Ku^2\cdots \oplus Ku^{p-1}$ where we define multiplication by $u^p=\ell$ and $u\alpha=\sigma(\alpha)u$ for $\alpha\in K$.

This is a central simple algebra by the theory of crossed product. Tensoring with $\Bbb Q_\ell$ will give a division algebra over $\Bbb Q_\ell$, so $A$ must be a division algebra.


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.