Notes - Galois Theory HT25, Computing the Galois group


Why isn’t the Galois group just the permutations of the roots?

Suppose $f \in \mathbb Q[t]$ is an irreducible polynomial with roots $\alpha _ 1, \ldots, \alpha _ n$. The Galois group of $f$ is then

\[G = \text{Gal}(\mathbb Q(\alpha_1, \ldots, \alpha_n) / \mathbb Q) := \{\mathbb Q\text{-linear automorphisms of }\mathbb Q(\alpha_1, \ldots, \alpha_n)\}\]

It’s clear that $G$ can be identified with a subgroup of $\text{Sym}(\alpha _ 1, \ldots, \alpha _ n)$, since if $\sigma \in G$ is a $\mathbb Q$-linear automorphism, then

\[f(\sigma(\alpha_i)) = \sigma(f(\alpha_i)) = 0\]

and so $\sigma(\alpha _ i)$ is also a root of $f$.

But why isn’t the Galois group always exactly $\text{Sym}(\alpha _ 1, \ldots, \alpha _ n)$? This is because there might be non-trivial relationships between the roots that have to be preserved by any $\mathbb Q$-linear automorphism of $\mathbb Q(\alpha _ 1, \ldots, \alpha _ n)$.

For example, consider

\[f(t) := t^3 - 3t + 1\]

You can show that the roots of $f$ are:

\[\begin{aligned} \alpha_1 &= 2\cos(2\pi/9) \\ \alpha_2 &= 2\cos(4\pi / 9) \\ \alpha_3 &= 2\cos(8\pi / 9) \end{aligned}\]

None of these roots are in $\mathbb Q$ (otherwise $f$ wouldn’t be irreducible). Trigonometric identities give non-trivial relationship between roots. For example, since $\cos(4\pi / 9) = 2\cos^2(2\pi / 9) - 1$, it follows that

\[\alpha_2 = \alpha_1^2 - 2\]

and by a similar identity

\[\alpha_3 = \alpha_2^2 - 2\]

The important part is that any element $\sigma \in G$ must preserve these identities:

\[\begin{aligned} \alpha_2 = \alpha_1^2 - 2 &\iff \sigma(\alpha_2) = \sigma(\alpha_1^2 - 2) \\\\ &\iff \sigma(\alpha_2) = \sigma(\alpha_1)^2 - 2 \end{aligned}\]

For example, the permutation $(13)$ cannot be in $G$, since permuting the second identity gives:

\[\alpha_1 \ne \alpha_2^2 - 2\]

It turns out that the only permutations which work are those in $A _ 3$, and so $G = A _ 3$, not $S _ 3$.




Related posts