0
$\begingroup$

In Vignéras' book Arithmétique des algèbres de quaternions, a quaternion algebra $A$ over a field $K$ is defined as a $4$-dim central algebra for which there exists a separable $2$-dim (necessarily commutative, so étale) algebra $L/K$ and $u \in A^\times$ such that conjugation by $u$ induces the nontrivial automorphism $\sigma$ of $L$: $$u l u^{-1} = \sigma(l) \quad \forall l \in L$$

With the data from the definition, the text works out the multiplication of two elements, and one sees from this computation that the nontrivial automorphism $\sigma$ of $L$ extends uniquely to an anti-automorphism of $A$ that sends $u$ to $-u$. We call it the conjugation on $A$.

Question. Is there a conceptual way to understand why $\sigma$ extends uniquely to an anti-automorphism that sends $u$ to $-u$?

$\endgroup$
1
$\begingroup$

Note that, by conjugating by $u$, it is equivalent to ask for an anti-automorphism of $A$ that leaves $L$ invariant and sends $u$ to $-u$. Writing $\theta := u^2 \in K$ (it lies in the center of $A$), this is very similar to the problem of constructing homomorphisms between algebraic field extensions:

Fact. When $A/K$ and $B/K$ are field extensions and $u \in A$ and $v \in B$ are generators with the same minimal polynomial over $K$, then there exist an isomorphism of $K$-algebras $A \to B$ that sends $u$ to $v$.

What we are led to here is

Fact. When $A, B$ are $4$-dim algebras over $K$, and $L$ is a $2$-dim separable $K$-algebra embedded in $A$ and $B$, and $u \in A^\times$ and $v \in A^\times$ are such that:

  1. $u$ and $v$ generate $A$ resp. $B$ together with $L$
  2. $u^2 = v^2 = \theta \in K$
  3. conjugation by $u$ and $v$ leaves $L$ invariant, and induces the same $K$-automorphism of $L$

then there exists an $L$-linear isomorphism of $K$-algebras $A \to B$ that sends $u$ to $v$.

Proof. In the commutative case above, one uses the universal property of the polynomial ring $K[X]$ (free commutative algebra) to write $A$ and $B$ as quotients of $K[X]$.

In this associative case, we proceed similarly: Let $\sigma$ be that automorphism of $L$. Let $K \langle L, X \rangle$ be the free associative algebra on the elements of $L$ and another variable $X$, and let $I$ be the ideal generated by all algebraic relations in $L$, as well as $Xl - \sigma(l) X$ for all $l \in L$ and $X^2 - \theta$. Inspecting degrees, it is not hard to see that $K \langle L, X \rangle / I$ has dimension $4$, so that the natural maps $K \langle L, X \rangle / I \to A, B$ sending $X$ to $u, v$, are $L$-linear $K$-algebra isomorphisms. $\square$

Note how every step in the proof has its commutative counterpart.

Back to the problem, we can apply this to $(A, u)$ and $(A^{\text{op}}, -u)$ to obtain the required anti-automorphism.

$\endgroup$

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.