Galois Theory HT25, Cubic equations



Finding a cubic formula

How can we use Galois theory to derive a formula for the cubic (in terms of radicals)?

Say $f(x) = x^3 + px + q$ is some irreducible polynomial over $\mathbb Q$ and suppose the roots are $\alpha _ 1, \alpha _ 2, \alpha _ 3$. The goal is to have expressions for these roots in terms of $p$ and $q$ which just involve $+, -, \times, \div$ and $\sqrt[n]{\cdot}$.

Let $G := \text{Gal} _ \mathbb Q(f) = \text{Gal}(K/\mathbb Q)$ be the Galois group of this polynomial, which is the Galois group of the splitting field $K = \mathbb Q(\alpha _ 1, \alpha _ 2, \alpha _ 3)$ for $f$. We assume for simplicity that $G \cong S _ 3$.

Possible in principle

First we show that solving the polynomial by radicals is always possible in principle without actually coming up with a formula. Since $G \cong S _ 3$, we have the sequence of normal subgroups

\[G \trianglerighteq A_3 \trianglerighteq \\{e\\}\]

and so the Galois correspondence tells us we also have the sequence of subfields

\[\mathbb Q \le K^{A_3} \le K\]

By previous results ( [[Notes - Galois Theory HT25, Determinant and discriminant]]U), we actually have $K^{A _ 3} = \mathbb Q(\delta)$ where $\delta := (\alpha _ 1 - \alpha _ 2)(\alpha _ 1 - \alpha _ 3)(\alpha _ 2 - \alpha _ 3)$, and so to keep everything very explicit we could write this sequence of field extensions as

\[\mathbb Q \le \mathbb Q(\delta) \le \mathbb Q(\alpha_1, \alpha_2, \alpha_3)\]

Since $\delta$ is a symmetric function in the roots, it follows from the fundamental theorem of symmetric functions that $\delta$ can be written in terms of $p$ and $q$. So if we could somehow find a way to write the elements of $\mathbb Q(\alpha _ 1, \alpha _ 2, \alpha _ 3)$ in terms of $\delta$, we would be done.

To do this, we can use results from [[Notes - Galois Theory HT25, Kummer extensions]]U. The important result is the following:

If:

  • $K/F$ is a Galois extension
  • $[K : F] = p$ where $p$ prime
  • For some $\varepsilon \in F$ where $\varepsilon \ne 1$, we have $\varepsilon^p = 1$

Then there exists $u \in K$ such that $u^p \in F$ and $K = F(u)$.

This would almost imply

\[\mathbb Q(\alpha_1, \alpha_2, \alpha_3) = \mathbb Q(\delta, u), \text{ where} u = \sqrt[3]{\ell} \text{ for some } \ell \in \mathbb Q(\delta)\]

since $[\mathbb Q(\alpha _ 1, \alpha _ 2, \alpha _ 3) : \mathbb Q(\delta)] = 3$ (since $ \vert A _ 3 \vert = 3$) but the problem is that we can’t guarantee the last condition, namely that $\mathbb Q(\alpha _ 1, \alpha _ 2, \alpha _ 3)$ contains a non-trivial 3rd root of unity. To resolve this, we instead consider the sequence of subfields

\[\mathbb Q \le \mathbb Q(\delta, \omega) \le \mathbb Q(\alpha_1, \alpha_2, \alpha_3, \omega)\]

where $\omega$ is a 3rd root of unity. You can then carefully check that $\mathbb Q(\alpha _ 1, \alpha _ 2, \alpha _ 3, \omega) / \mathbb Q$ is still a Galois extension and that $\text{Gal}(\mathbb Q(\alpha _ 1, \alpha _ 2, \alpha _ 3, \omega) / \mathbb Q(\delta, \omega)) \cong A _ 3$ so that now the requirements of the above results are satisfied with $K = \mathbb Q(\alpha _ 1, \alpha _ 2, \alpha _ 3, \omega)$, $F = \mathbb Q(\alpha _ 1, \alpha _ 2, \alpha _ 3)$ and $\varepsilon = \omega$. Thus we can conclude

\[\mathbb Q(\alpha_1, \alpha_2, \alpha_3, \omega) = \mathbb Q(\delta, \omega, u), \text{where }u = \sqrt[3]{\ell} \text{ for some } \ell \in \mathbb Q(\delta, \omega)\]

and with this, we are done since we can now in principle find expressions for $\alpha _ 1, \alpha _ 2, \alpha _ 3$ in terms of $\delta, \omega, u$.

Possible in practice

Actually finding the expressions requires some work. Firstly, direct calculation shows that

\[\delta = \sqrt{-4p^3 - 27q^2}\]

The proof of the Kummer extension result finds $u$ as an $\omega$-eigenvector of $\sigma$ where $\langle \sigma \rangle = \text{Gal}(\mathbb Q(\alpha _ 1, \alpha _ 2, \alpha _ 3, \omega) / \mathbb Q(\delta, \omega))$. Since the group is cyclic, one eigenvector is given by

\[u = \frac{\alpha_1 + \omega \alpha_2 + \omega^2 \alpha_3}{3}\]

We could have also considered $\omega^2$ to give another eigenvector $v$ of $\sigma$ (this time with eigenvalue $\omega^2$):

\[v = \frac{\alpha_1 + \omega^2 \alpha_2 + \omega \alpha_3}{3}\]

Remember that $u^3$ and $v^3$ are in $\mathbb Q(\omega, \delta)$. By direct computation, we can show

  • $u^3 v^3 = \frac{-p^3}{27}$
  • $u^3 + v^3 = -q$

and thus $u^3$ and $v^3$ are the roots of the quadratic

\[z^2 + qz - \frac{p^3}{27} = 0\]

and so may be written in terms of $p$ and $q$, and so $u$ and $v$ can be extracted by taking $p$th roots. Finally, we may write $\alpha _ 1, \alpha _ 2, \alpha _ 3$ in terms of $u$ and $v$ like so:

  • $\alpha _ 1 = u + v$
  • $\alpha _ 2 = \omega^2 u + \omega v$
  • $\alpha _ 3 = \omega u + \omega^2 \beta$

as required.

More general results

There is a more general result in [[Notes - Galois Theory HT25, Solvability by radicals]]U which says:

If:

  • $F$ is a field of characteristic zero
  • $\alpha$ is algebraic over $F$

Then:

  • $\alpha$ lies in a radical extension of $F$ if and only if $\text{Gal} _ F(m _ {F, \alpha})$ is solvable.

The “Possible in principle” section above is just a special case.

Flashcards

A complete characterisation of the Galois group in terms of the roots

Suppose:

  • $F$ is a field
  • $\text{char}(F) \ne 2, 3$
  • $f := t^3 + pt + q \in F[t]$
  • $f$ is irreducible
  • $K$ is a splitting field for $f$

@State a theorem which completely characterises the Galois group $G = \text{Gal}(K/F)$.


Let $\Delta$ be the discriminant (the square of the product of the differences of roots). Then:

  • If $\Delta$ is a square in $F$, then $G = A _ 3$
  • If $\Delta$ is not a square in $F$, $G = S _ 3$

@Prove that if:

  • $F$ is a field
  • $\text{char}(F) \ne 2, 3$
  • $f := t^3 + pt + q \in F[t]$
  • $f$ is irreducible
  • $K$ is a splitting field for $f$
  • $\Delta$ is the discriminant of $f$ (the square of the product of the differences of roots)
  • $G = \text{Gal}(K/F)$

Then:

  • If $\Delta$ is a square in $F$, then $G = A _ 3$
  • If $\Delta$ is not a square in $F$, then $G = S _ 3$

$G$ must be either $S _ 3$ or $A _ 3$:

Let $\alpha$ be a root of $f$. Since $f$ is irreducible over $F$, $m _ {F, \alpha} = f$. Hence

\[[F(\alpha) : F] = \deg m_{F, \alpha} = \deg f = 3\]

Hence $[F(\alpha) : F] \mid [K : F]$ by the tower law and since $ \vert G \vert = [K : F]$, $ \vert G \vert $ is a multiple of three.

The only subgroups of $S _ 3$ with this property are $A _ 3$ and $S _ 3$.

$\Delta$ determines which group:

By the result that says

If:

  • $f \in F[t]$ is a polynomial of degree $n$
  • $K$ is a splitting field for $f$
  • $\{\alpha _ 1, \ldots, \alpha _ n\}$ are the roots of $f$ in $K$
  • $G = \text{Gal}(K/F)$, identified with a subgroup of $S _ n$
  • $\text{char}(F) \ne 2$
  • $f$ has no repeated roots

Then:

  • $G \le A _ n$ if and only if $\Delta$ is a square in $F$.

it follows that if $\Delta$ is a square in $F$, $G = A _ 3$, otherwise $G = S _ 3$.

Suppose

\[f(x) = (x - \alpha_1)(x - \alpha_2)(x - \alpha_3) \in \mathbb Q[x]\]

Can you give necessary and sufficient conditions for when $f$ has Galois group $C _ 3$?


It is tempting to say (by appealing to other roots) that this occurs when

\[\Delta = (\alpha_1 - \alpha_2)^2 (\alpha_1 - \alpha_3)^2(\alpha_2 - \alpha_3)^2\]

is square in $\mathbb Q$. This is almost correct, but neglects reducibility: the correct answer is that the Galois group is $C _ 3$ iff $\alpha _ 1, \alpha _ 2, \alpha _ 3 \notin \mathbb Q$ and $\Delta \in \mathbb Q^2$.

@exam~

Simplifying substitution

Suppose:

\[f(x) := ax^3 + bx^2 + cx + d\]

@State the substitution that can be used to put the cubic in a simpler form.


Let $t = x + \frac{b}{3a}$. Then the equation becomes

\[f(t) = t^3 + pt + q\]

for some $p$, $q$ (really this is a specific case of being able to make a substitution that eliminates the next lower-degree term).

Resolvent quadratic

Suppose:

  • $\text{char}(F) \ne 3$
  • $f = t^3 + pt + q \in F[t]$ is an irreducible cubic with roots $\alpha _ 1, \alpha _ 2, \alpha _ 3$.
  • $u := \frac{\alpha _ 1 + \omega \alpha _ 2 + \omega^2 \alpha _ 3}{3}$
  • $v := \frac{\alpha _ 1 + \omega^2 \alpha _ 2 + \omega \alpha _ 3}{3}$

@Prove that:

  • $uv = -\frac{p}{3}$
  • $u^3 v^3 = -\frac{p^3}{27}$
  • $u^3 + v^3 = -q$
  • $u^3$ and $v^3$ are the roots of a quadratic polynomial with coefficients in $F$
  • $\alpha _ 1$ can be written in terms of $u$ and $v$

Why this is relevant:

  • This is basically a derivation of the cubic formula; if $\alpha _ 1$ can be written in terms of $u$ and $v$ and $u^3$ and $v^3$ are the roots of a quadratic polynomial with coefficients in $F$, then you can calculate $\alpha _ 1$ by first solving the “resolvent quadratic” to derive $u^3$ and $v^3$, then plugging $u$ and $v$ into the expression for $\alpha _ 1$.
  • This seems a lot more like random algebra than Galois theory. However, Galois theory is the reason we knew what to set $u$ and $v$ to, these are eigenvectors of $K(\omega)$ where $K$ is a splitting field of $f$ and $\omega$ is a cube root of unity.

Equating coefficients in $t^3 + pt + q = (t - \alpha _ 1)(t - \alpha _ 2)(t - \alpha _ 3)$, we have that

  • $\alpha _ 1 + \alpha _ 2 + \alpha _ 3 = 0$
  • $\alpha _ 1 \alpha _ 2 + \alpha _ 2 \alpha _ 3 + \alpha _ 3 \alpha _ 1 = p$
  • $\alpha _ 1 \alpha _ 2 \alpha _ 3 = -q$

Using the fact that $\omega + \omega^2 = -1$, it follows that

\[\begin{aligned} 9uv &= (\alpha_1 + \omega \alpha_2 + \omega^2 \alpha_3)(\alpha_1 + \omega^2 \alpha_2 + \omega \alpha_3) \\\\ &= \alpha_1^2 + \omega \alpha_1 \alpha_2 + \omega^2 \alpha_1 \alpha_3 + \omega^2 \alpha_1 \alpha_2 + \alpha_2^2 + \omega \alpha_2 \alpha_3 + \omega \alpha_2 \alpha_3 + \omega \alpha_1 \alpha_3 + \omega^2 \alpha_2 \alpha_3 + \alpha_3^2 \\\\ &= \alpha_1^2 + \alpha_2^2 + \omega_3^2 - \alpha_1 \alpha_2 - \alpha_2 \alpha_3 - \alpha_3 \alpha_1 \\\\ &= (\alpha_1 + \alpha_2 + \alpha_3)^2 - 3(\alpha_1 \alpha_2 + \alpha_2 \alpha_3 + \alpha_3 \alpha_1) \\\\ &= -3p \end{aligned}\]

Since $\text{char}(F) \ne 3$, it follows $uv = - \frac p 3$.

(b) follows from (a).

For (c), note that

\[u^3 + v^3 = (u + v) (\omega u + \omega^2 v) (\omega^2 u + \omega v)\]

Then

\[\begin{aligned} 3(u + v) &= 2\alpha_1 + (\omega + \omega^2) \alpha_2 + (\omega^2 + \omega) \alpha_3 = 3\alpha_1 \\\\ 3(\omega u + \omega^2 v) &= (\omega + \omega^2)\alpha_1 + (\omega^2 + \omega)\alpha_2 + 2\alpha_3 = 3\alpha_3 \\\\ 3(\omega^2 u + \omega v) &= (\omega^2 + \omega) \alpha_1 + 2\alpha_2 + 9\omega + \omega^2) \alpha_3 = 3\alpha_2 \end{aligned}\]

For (d), this follows from (b, c), since they are the roots of

\[z^2 + qz - \frac{p^3}{27} = 0\]

Finally, since

\[\begin{aligned} u + v &= \frac{\alpha_1 + \omega \alpha_2 + \omega^2 \alpha_3 + \alpha_1 + \omega^2 \alpha_2 + \omega \alpha_3 }{3} \\\\ &= \frac{2 \alpha_1 - \alpha_2 - \alpha_3}{3} \\\\ &= \alpha_1 \end{aligned}\]

and $uv = -\frac p 3$, it follows that

\[\alpha_1 = u - \frac{p}{3u}\]



Related posts