2

I was studying Galois theory when some questions came to my mind.

The bibliography for my course indicates Bosch's "Algebra" as a source and suggests to prove "a variant of Proposition 12 when only one of the extensions is required to be Galois". I cannot ask for explanations currently so I ask here.

Proposition 12

Not having much expertise in the field (pardon the pun), my questions are the following:

  1. Which of the two extensions can be not restricted to be Galois?
  2. Does this lead to some changes, given that not all the extensions above are Galois (so one should use Aut instead of Gal)?
  3. Where do I find a proof of that variant?
  4. Is there some patent reason why dropping that hypothesis is so crucial?

These may be silly questions but for a beginner they are not.

  • 1
    Somehow this question by Pete L. Clark sprang to mind, but that runs a bit deeper. – Jyrki Lahtonen Aug 10 '24 at 14:48
  • 1
    Anyway, if $E'/K$ is Galois then so are $E'/(E\cap E')$ and $E\cdot E'/E$. After all, $E'$ is the splitting field of some separable polynomial $f(x)$ over $K$, and continues to have that role even if we view $f$ as a polynomial with coefficients in $E\cap E'$. I think that $E\cdot E'$ is then the splitting field of $f(x)$ over $E$. This answers your first question, and hopefully keeps you occupied while we search for related material (likely exists on the site). – Jyrki Lahtonen Aug 10 '24 at 14:53
  • Superficially, it seems the point of this theorem is the two isomorphisms. Dropping the Galois requirements for the extensions $E$ and $E'$ over $K$ probably just makes either 1-1 or onto fail. I would try to just prove the theorem and see where you need to insert this condition. – I Zuka I Aug 19 '24 at 06:13
  • @IZukaI The point is that the proof by Bosch does assume both the hypotheses and I can’t figure out another. – Amanda Wealth Aug 19 '24 at 06:50

1 Answers1

2

Let's start with a useful proposition from a different text:

Proposition 21. (p.529 Dummit and Foote's "Abstract Algebra"):
Let $K_1$ and $K_2$ be two finite extensions of a field $F$ contained in $K$. Then: $$[K_1K_2:F]\leq [K_1:F][K_2:F]$$ with equality iff an $F$-basis for one of the fields remains linearly independent over the other field. If $\alpha_1,...,\alpha_n$ and $\beta_1,...,\beta_m$ are bases for $K_1$ and $K_2$ over $F$, respectively, then the elements $\{\alpha_i\beta_j\}_{i,j}$ span $K_1K_2$ over $F$.


Immediately, we see that if $E/K$ and $E'/K$ are finite extensions, so is $E\cdot E'/K$. $\square$

Moreover, this gives us the form of the composite extension: $$E\cdot E'/K = K(\{\alpha_i\beta_j\}_{i,j}),$$ where the $\alpha_i$'s and $\beta_j$'s are the respective algebraic elements over $K$ from $E$ and $E'$.

By definition (p.562), $K/F$ is Galois when $|Aut(K/F)| = [K:F]$.

In general, for a composite field: $$|Aut(E/K)|\cdot|Aut(E'/K)|\leq|Aut(E\cdot E'/K)|$$ $\color{orange}{\text{[Pf: Exercise]}}$. Also, in general: $$|Aut(K/F)|\leq [K:F]$$ $\color{orange}{\text{[Pf: Exercise]}}$. Combining these results, we get in the event that $E/K$ and $E'/K$ are Galois: $$|Aut(E/K)||Aut(E'/K)| \leq |Aut(E\cdot E'/K)|\leq [E\cdot E':K]\leq [E:K][E':K] = |Aut(E/K)||Aut(E'/K)|.$$ This squeezes together $|Aut(E\cdot E'/K)| = [E\cdot E':K]$, making the composite Galois. $\square$


Part (i): Show the homomorphism: $$\varphi:Gal(E\cdot E'/E) \to Gal(E'/E\cap E');$$ $$\sigma \mapsto \sigma|_{E'}$$ is an isomorphism.


Proof:
$\underline{\text{1-1:}}$ Suppose $\sigma,\tau\in Gal(E\cdot E'/E)$ are such that $\varphi(\sigma) = \varphi(\tau)$. Then: $$\sigma|_{E'} \equiv \tau|_{E'}.$$ As these fix $E$, we know $\sigma|_E \equiv Id_E \equiv \tau|_E$. Lastly, since they are automorphisms they agree over linear combinations (constructing the composite field). So $\sigma = \tau$.

$\underline{\text{Onto:}}$ Take $\sigma \in Gal(E'/E\cap E')$, then trivially (and linearly) extend to $\widetilde{\sigma}\in Gal(E\cdot E'/E)$ by: $$\widetilde{\sigma}|_E \equiv Id_E\text{ }\text{ }\text{ and }\text{ }\text{ }\widetilde{\sigma}|_{E'}\equiv \sigma|_{E'}.$$ Then we have existence of $\widetilde{\sigma}$ such that $\varphi(\widetilde{\sigma}) = \widetilde{\sigma}|_{E'} \equiv \sigma|_{E'} = \sigma$. $\square$


Part (ii): Show the homomorphism: $$\psi:Gal(E\cdot E'/K)\to Gal(E/K)\times Gal(E'/K);$$ $$\sigma \mapsto (\sigma|_E,\sigma|_{E'})$$ is injective. In addition, if $E\cap E' = K$, then $\psi$ is surjective and hence an isomorphism.


Proof:
$\underline{\text{1-1:}}$ Just because, suppose $\sigma \in ker(\psi)$, then $\psi(\sigma) = (\sigma|_E,\sigma|_{E'}) = (Id|_E,Id|_{E'})$. Because $\sigma$ is an automorphism on $E\cdot E'/K$, we can separate the $K$-linear combinations, where all images (including $K$) are fixed. Hence $\sigma = Id_{E\cdot E'/K}$. In other words, $ker(\psi)$ is trivial.

$\underline{\text{Onto:}}$ Suppose $(\sigma,\tau)\in Gal(E/K)\times Gal(E'/K)$. Define $\widetilde{\sigma}\in Gal(E\cdot E'/K)$ by: $$\widetilde{\sigma}: e\cdot e' \mapsto \sigma(e)\tau(e')$$ and extend $K$-linearly. We get $\psi(\widetilde{\sigma}) = (\widetilde{\sigma}|_E,\widetilde{\sigma}|_{E'}) = (\sigma,\tau).$ This seems to hold regardless of what $K$ is. $\blacksquare$



In closing, we really only required the extensions $E/K$ and $E'/K$ to be Galois to make the composite field Galois (see the second to last equality around the second '$\square$'). To me, this whole thing could be glossed over by just writing $Aut$...

I Zuka I
  • 1,238