13
$\begingroup$

Let $E/K$ be an elliptic curve over a field $K$, with discriminant $\Delta$. Then the polynomial $x^3-\Delta$ has a root (and hence all roots since Galois) in $K(E[3])$; this can be shown laboriously through solving the 3-division polynomial (a quartic).

Is there a nicer/more intuitive way of seeing this and can you please provide a proper reference for either the above method or whatever you suggest?

$\endgroup$
  • $\begingroup$ This is a very good question, $x^3$ has a real root exactly on the real number line "if & only if" it is a cube of any constructible number $\endgroup$ – Bassam Karzeddin Dec 30 '15 at 11:54
  • $\begingroup$ "All roots since Galois" holds only if $K$ contains the cube roots of unity. (Presumably you don't allow characteristic $3$, in which case $E[3]$ behaves very differently.) $\endgroup$ – Noam D. Elkies Jan 10 '16 at 4:45
8
+50
$\begingroup$

One way to think about this is via the modular curves parametrizing elliptic curves $E$ with either $E[3]$ or $\Delta^{1/3}$ rational. Note that $\Delta^{1/3}$ is rational iff $j^{1/3}$ is rational, because $j = E_4^3 / \Delta$.

Assume for simplicity that $K$ contains the cube roots of unity (because $K(E[3])$ contains them in any case thanks to the Weil pairing). The $E[3]$ curve is the modular curve $X(3)$, with a map $X(3) \to X(1)$ that forgets the $3$-torsion structure, and is Galois with group ${\rm PSL}_2({\bf Z}/3{\bf Z}) \cong A_4$.

Now $A_4$ has a normal subgroup, the "Klein $4$-group" $V_4$ consisting of the identity and the three double transpositions. (The $V$ stands for German "Vierergruppe".) So $X(3) / V_4$ is a Galois cover, call it $X'(3)$, of $X(1)$ with group $A_4 / V_4 \cong {\bf Z} / 3 {\bf Z}$, i.e. a 3:1 cyclic cover. Once $K$ contains cube roots of unity, Kummer theory says that any 3:1 cyclic cover is obtained by adjoining a cube root to the function field; here this function field is $K(j)$, so the function field $K(X'(3))$ is $K(j,F^{1/3})$ where $F$ is some rational function of $j$ that's not already a cube.

The punchline is that we can take $F=j$. Since the function field of $X'(3)$ is contained in the function field of $X(3)$, this explains why $\Delta^{1/3}$ is a rational function of the coordinates of $E[3]$.

The fact that $K(X'(3)) = K(j^{1/3})$ can almost be recovered from the ramification structure of the map $X(3) \to X(1)$. It is ramified only above $j=\infty$, $j=0$, and $j=1728$, with cycle structures $(3,1)$, $(3,1)$, $(2,2)$ respectively. Hence the cover $X'(3) \to X(1)$ is ramified only above $j=0$ and $j=\infty$, so $K(X'(3))$ must be $K((cj)^{1/3})$ for some constant $c \in K^*$. The fact that we can use $c=1$ takes a bit more work, but once we know that the cover has this form it's enough to just try some convenient $E$ with $j(E)\neq 0$ to complete the proof.

$\endgroup$

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.